You are on page 1of 610

Foams and Emulsions

NATO ASI Series


Advanced Science Institute Series

A Series presenting the results of activities sponsored by the NATO Science Committee,
which aims at the dissemination of advanced scientific and technological knowledge,
with a view to strengthening links between scientific communities.

The Series is published by an international board of publishers in conjunction with the NATO
Scientific Affairs Division

A Life Sciences Plenum Publishing Corporation


B Physics London and New York

C Mathematical and Physical Sciences Kluwer Academic Publishers


D Behavioural and Social Sciences Dordrecht, Boston and London
E Applied Sciences

F Computer and Systems Sciences Springer-Verlag


G Ecological Sciences Berlin, Heidelberg, New York, London,
H Cell Biology Paris and Tokyo
I Global Environment Change

PARTNERSHIP SUB-SERIES
1. Disarmament Technologies Kluwer Academic Publishers
2. Environment Springer-Verlag I Kluwer Academic Publishers
3. High Technology Kluwer Academic Publishers
4. Science and Technology Polley Kluwer Academic Publishers
5. Computer Networking Kluwer Academic Publishers

The Partnership Sub-Series incorporates activities undertaken in collaboration with NATO's


Cooperation Partners, the countries of the CIS and Central and Eastern Europe, in Priority Areas of
concern to those countries.

NATO.PCO.DATA BASE
The electronic index to the NATO ASI Series provides full bibliographical references (with keywords
and/or abstracts) to about 50,000 contributions from international scientists published in all sections of
the NATO ASI Series. Access to the NATO-PCO-DATA BASE is possible via a CD-ROM "NATO Science
and Technology Disk'' with user-friendly retrieval software in English, French, and German (©WTV
GmbH and DATAWARE Technologies, Inc. 1989). The CD-ROM contains the AGARD Aerospace Data-
base.

The CD-ROM can be ordered through any member of the Board of Publishers or through NATO-PCO,
Overijse, Belgium.

Series E: Applied Sciences -Vol. 355


Foams and Emulsions
edited by

J.F.Sadoc
Universite de Paris-Sud,
Laboratoire de Physique des Solides,
Orsay, France
and

N. Rivier
Universite Louis Pasteur,
lnstitut de Physique,
Strasbourg, France

Springer-Science+Business Media, B.V.


Proceedings of the NATO Advanced Study Institute on
Foams and Emulsions, Emulsions and Cellular Materials
Cargese, Corsica
12-24 May, 1997

A C.i.P. Catalogue record for this book is available from the Library of Congress.

ISBN 978-90-481-5180-6 ISBN 978-94-015-9157-7 (eBook)


DOI 10.1007/978-94-015-9157-7

Printed on acid-free paper

AII Rights Reserved


© 1999 Springer Science+Business Media Dordrecht
Originally published byKluwer Academic Publishers in 1999
Softcover reprint of the hardcover Ist edition 1999
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical, including photo-
copying, recording or by any information storage and retrieval system, without written
permission from the copyright owner.
TABLE OF CONTENTS

(Introduction and conclusion of each chapter are not listed)

Foreword xiii

I. Surface energy and surface rheology


Relation to foam properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
D. Langevin
Defmition of surface properties 1
Surface energy 1
Surface rheology 5
Foaming and foam stability 9
Foaming and dynamic surface tension 9
Foam stability 10

II. Foams and emulsions: Their stability and breakdown by solid


particles and liquid droplets. The colloid chemistry of a dog's
breakfast ...................................................... 21
R. Aveyard, B.P. Binks, J.H Clint and P.D.l. Fletcher
Foam and emulsion stability 22
Particles and droplets at fluid/fluid interfaces 28
Bridging of liquid films by particles and droplets 34
Effects of particles and oil drops on foam stability 37
Effects of solid particles on emulsion stability 40
Summ~ 42

III. An introduction to forces and structure in individual foam and


emulsion films .................................................. 45
Vance Bergeron
Disjoining pressure 47
Disjoining pressure models 52
Experimental measurements 66

IV. Structure of foam films containing additional polyelectrolytes .......... 73


R. von Klitzing, A. Espert, A. Colin and D. Langevin
Experimental section 74
Results 74
Discussion and conclusions 74
vi

V. Drainage of foam films ......................................... 83


Roumen Tsekov
The classical theory 85
New theory 88

VI. Foam evolution in two dimensions.


A particular limit of domain growth ............................. 91
Joel Stavans
Cellular structures: The foam as paradigm 94
Ostwald ripening 99

VII. Statistical thermodynamics of foam ............................. 105


Nicolas Rivier
Two-dimensional foams 106
Structural equations of state (Lewis, Aboav-Weaire and Peshkin) 107
Disorder control. Lemaitre equation of state Ill
Lemaitre's law as the virial equation of state for foams I 13
Universality ofLemaitre's law 115
Foaming 117
Thermodynamics of solid foams 120

VIII. Polygonal networks resulting from dewetting ..................... 127


U. Thiele, M Mertig, W. Pompe and H Wendrock

IX. Two-dimensional magnetic liquid froth . . . . . . . . . . . . . . . . . . . . . . . . . . 137


F. Elias, C. Flament, J.-C. Bacri and F. Graner
Two-dimensional instabilities in magnetic fluids 13 7
Formation of a magnetic fluid froth 140
Topological caracterization 141
Artificial coarsening: Statistical study 141
The role of time: Quasi Von Neumann behaviour 144
Topological correlations 146
Two-dimensional liquid magnetic froths: A model for studying
the topology of two-dimensional cellular structures 147

X. Cellular structures in metallurgy ............................... 151


Y.J.M Brechet and D. Weygand
Grain structure in metals 152
Plasticity and ductile fracture 161
Mechanical behaviour of metallic foams 167
vii

XI. The compression of closed-cell polymer foams . . . . . . . . . . . . . . . . . . . . 175


N.J. Mills and HX Zhu
Material responses 176
Element mechanics 178
Foam microstructures 180
Foam structural model 181
Young's modulus and Poisson's ratio 185
Yielding 187
Discussion 190

XII. Hard cellular materials in the human body: Properties and


productions of foamed polymers for bone replacement ............. 193
C.S. Pereira, M E. Gomes, R.L. Reis and A.M Cunha
Introduction -Bone properties 193
- Biomaterials 196
-Foams for implantation 197
Experimental - Materials and methods 198
Results and discussion - Morphology 199, 203
-Chemistry 201
- Mechanical properties 202
- Degradation 204

XIII. Rheology and glassy dynamics offoams ......................... 207


ME. Cates and P. Sollich
Basic rheology 208
The elastic modulus 210
Dissipation in foams 218
Soft glassy rheology 224

XIV. Surfactants and stress conditions at fluid interfaces ................ 237


Kathleen J. Stebe and Charles D. Eggleton
Basic equations 238
Stagnant interfaces 242
Surfactant effects on strongly deforming interfaces 244
Remobilizing surfactants: Controlling stresses 249

XV. Foam micromechanics. Structure and rheology of foams, emulsions,


and ceUular solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Andrew M Kraynik, Michael K Neilsen, Douglas A. Reinelt and
William E. Warren
Theoretical approach 260
viii

Two dimensions 261


Three dimensions 265

XVI. The structure and geometry of foams ............................ 287


D. Weaire, R. Phelan and G. Verbist
Introduction: The ideal foam model 287
Laplace law and curvature 288
Thin films 289
Vertices and Plateau borders in the ideal foam model 290
Gas and liquid pressures 291
Inversion and the decoration theorem 292
The hexagonal honeycomb in 2D 293
Simulations 293
The Surface Evolver 295
Application to wet foams 298

XVII. Rheology and drainage of liquid foams .......................... 303


G. Verbist, D. Weaire and R. Phelan
Foams: Solid or liquid? 303
Elastic moduli 304
Bulk modulus of a foam 304
Shearing a disordered foam 305
Foam drainage 305
Some solutions and experimental verifications 308

XVIII. Electrical and thermal transport in foams ........................ 315


R. Phelan, G. Verbist and D. Weaire
Electrical conductance 315
Calculation of nonlinearities 317
Measuring the conductivity (and hence the liquid fraction) 318
Thermal conductivity 318

XIX. Decontamination of nuclear components through the use of foams ... 323
G. Boissonnet, M Faury and B. Fournel
Industrial application of the foam process 324
Structure and stability of decontamination foam 326
Foam rheology 331
General conclusion and prospects 334
ix

XX. Foams in porous media ....................................... 335


William R. Rossen
The nature of foam in porous media 335
Foam mobility at fixed bubble size 338
Processes that change bubble size 34I
Simplifying paradigm for foam in porous media 344
Other considerations in application of foam 344

XXI. Application of the Voronoi tessellations to the study of flow


of granular materials 349
J Lemaitre, L. Samson, P. Richard, L. Oger and N.N. Medvedev
Bidimensional results 350
Tridimensional results 355

XXII. Determination of real three-dimensional foam structure using


optical tomography ........................................... 359
C. Monnereau and M Vignes-Adler
Experimental 36 I
Foam reconstruction 364
Wet/dry foam 366
A coarsening foam 372

XXIII. The geometry of bubbles and foams ............................. 379


John M Sullivan
Variational problems for surfaces 380
Forces in constant mean curvature (CMC) surfaces 383
Bubble clusters and singularities in soap films 385
Combinatorics of foams 388
Bounds on the combinatorics of triangulations 390
Tetrahedrally close-packed (TCP) structures 394
Some constructions for TCP foams 397
Kelvin's problem 399

XXIV. Crystal structures as periodic foams and vice versa ................ 403
Michael 0 'Keeffe
Homogeneous sphere packings with tetrahedral holes and their duals 403
Tetrahedral packings involving two or more kinds of spheres 406
Geometric problems associated with variations on the sodalite theme 4I 6
X

XXV. Inverse micellar lyotropic cubic phases ........................ 423


John Seddon and John Robins
Lyotropic phase diagrams 424
Micellar and inverse micellar mesophases 426
Structure of the Fd3m cubic phase 429
Formation and stability of inverse micellar cubic phases 430

XXVI. Sponges ................................................... 437


S.T. Hyde
Differential geometry of elliptic, parabolic, hyperbolic surfaces 439
Differential geometry and topology 441
llomogeneoussponges 443
Embeddings of sponges in Euclidean 3-space, E3 445
Minimal surfaces in E3 446
The Gauss map 450
The Bonnet transformation 453
Faceting of sponges: infinite polyhedra 456
Triangular infmite polyhedra - close-packing on sponges 457
Generalised infmite polyhedra 462
4-connected nets as reticulations of sponges 462
3-connected nets and novel graphitic "schwarzites" 464
Cell-structures and hyperbolic surfaces 465

XXVII. Deformations of periodic minimal surfaces ................... 0 0 471


C. Oguey
Immersion in the complex space C3 473
The Bonnet transformation 475
Non isometric transformations 476

XXVIII. Aperiodic hierarchical tilings ........................... 0 •• 0 •• 481


Chaim Goodman-Strauss
Addresses 488
Keys 491
Mechanisms 492
On the utility of substitution tilings 495

XXIX. The shell map


The structure of froths through a dynamical map .....•......... 497
TomasoAste
From a cell to the whole froth, a topological map 499
Space curvature from the map 502
Construction of three-dimensional disordered structures 505
xi

XXX. Curved spaces and geometrical frustration ..................... 511


Jean-Franfois Sadoc
Simple two-dimensional examples 512
Sphere packing 513
Regular packing of tetrahedra: The polytope (3,3,5) 514
Decurving and disclinations 516
Frustration in amphiphile liquid crystals 520

XXXI. Computer simulations and tessellations of granular materials ..... 527


L. Oger, J.P. Troadec, A. Gervais and N.N. Medvedev
Computer simulations 528
Tessellation and analysis ofpackings 539

XXXII. Study of experimental and simulated evolutions of 2D foams ...... 547


V. Parfait-Pignol, R. Delannay and A. Mocellin
Experimental devices 548
Image treatment 551
Experimental results 552
Simulation models 553
Experiment versus simulation 557

XXXIII. Simulation ofthe foaming process ............................. 563


N. Pittet
Growth and tessellations 565
Sequence of contact times 567
The example of Kerroc 568

XXXIV. Voronoi tessellation in model glass systems ..................... 571


Remi Jullien, Philippe Jund, Didier Caprion and
Jean-Fran~ois Sadoc
Random packings in flat space 573
Random packings in curved space 577
Configurations generated by molecular dynamics 583

Index 589
FOREWORD

This volume is the proceedings of the School on Foams, Emulsions and


Cellular Materials, held in Cargese (Corsica), May 12 -24, 1997. The school
gathered a wide spectrum of participants and lecturers, coming from various
communities and countries, from university to industry (nuclear, petroleum,
chemical, mechanical and thermal).
The volume is intended as a general and introductory survey of the field.
The authors have tried to be clear and didactic. Because the field spans several
scientific disciplines and is relatively new, there are no textbooks with all the
basic tools necessary for research students. We trust that the present book will
serve this purpose.
The chapters are grouped in sections, but with fairly loose boundaries.
While cross-referencing has been encouraged, each chapter is intelligible on its
own, and if a few concepts are not familiar to the reader, their explanation can
be found in an earlier chapter, as referred to in the index.
The book contains all the lecture courses, and several contributions selected
because they were new and promising developments, not yet available in print
elsewhere, or because they covered aspects of the subject not discussed in the
lectures. It was not possible to include the lecture course by Yann Barrandon on
the renewal of the epidermis. It contained medical and biomedical applications,
a very wide domain, which cannot be reduced to a single chapter in a general
treatise on foams. For an introduction on the subject, see chapter 7 of the
monograph by Dover and Wright [1]. The close connection between epidermis
and foams is shown by the fact that Lewis's law was discovered on the
epidermis of the cucumber, and by this photograph of the basement membrane
of human epidermis (B. Dubertret [2]).

xiii
xiv

We are grateful to Michael Leunig for permission to reprint his cartoon, to J.


Sullivan, G. Boissonnet and B. Dubertret for photographs of computer-
generated or real foams. Other photographs are by T. Aste, B. Gardiner, D.
Weaire, K. Stebe, U. Thiele and N. Rivier. C. Oguey has helped with the edition
and the index.
We would like to thank the Centre of Cargese and its Director, Elizabeth
Dubois-Violette for their hospitality and for their help in the running of the
school. The school has been sponsored and supported by NATO, by the Institut
Fran~ais du Petrole, by Rhone-Poulenc, and the French granting agencies:
CNRS (Formation Permanente) and DGA. This volume has been published with
the help of a special grant from NATO.

Jean-Fran~ois Sadoc
Nicolas Rivier

I. Dover, R. and Wright, N.A. (1991) Physiology Biochemistry and Molecular


Biology of the Skin, 2nd. edition, Oxford University Press.
2. Dubertret, B. and Rivier, N. (1977) The renewal of the epidermis: A topological
mechanism, Biophys. J. 73, 38-44
SURFACE ENERGY AND SURFACE RHEOLOGY
RELATION TO FOAM PROPERTIES

D.LANGEVIN
Centre de Recherche Paul Pascal
Av. A.Schweitzer
33600, Pessac, France

Foams made from pure fluids are generally very unstable : bubbles
obtained by shaking pure water last only for a few seconds. When a surface
active substance is added to water, bubbles lifetime can become much
longer. The increase in surface energy due to the increase in surface area
after the creation of the bubbles is substancial : for instance if one cm3 of
solution with a surface of about one cm2 is shaken to produce a foam with
bubbles of millimetric size, the air-solution area increases by a factor of ten
and the surface energy, which is proportional to the area, increases by the
same factor. Of course, the surface tension of the solution is decreased by
the presence of the substance, but the state of equilibrium is the state of
minimal energy and is in any case the state of minimal area : the foam can
never be stable. The difference between water and surfactant solutions lies
therefore in the time scales involved in bubbles lifetime. Surface tension is
not the main surface characteristic property there, and other properties such
as surface elasticity, surface viscosity, dynamic surface tension, become
extremely important. In this chapter, we will first define all these surface
properties, indicate how they can be determined and discuss how they
influence the foam properties.

I. Definition of surface properties

1. SURFACE ENERGY

1.1 Surface excess properties


The definition of properties such as viscosity for a purely two-dimensional
system poses some difficulties. Real interfaces between two media are never
perfect mathematical surfaces, all the physical properties change from those

J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 1-20.


© 1999 Kluwer Academic Publishers.
2

of the lower medium to those of the upper medium in a thin interfacial


region, with a thickness comparable to molecular dimensions. This led Gibbs
to define .surface properties as excess properties [ 1]. For instance if z is the
vertical coordinate, A the surface area, if medium 1 of density p 1 is located
in the region z<O , medium 2 of density P2 in the region z>O, if F(Pi) is the
total energy of medium i and f(z) the energy density at heigth z, the excess
energy due to the presence of the interface in the region z - 0 is given by,
according to Gibbs :

+oo
LW =A J f dz - F(pt)- F(p2) (1)
-oo

Because the density p varies in a thin region around z=O, the integral is not
equal the sum F(PI) + F(P2) :there is an excess energy AF which is currently
referred to as the "surface" energy. The surface tension yis simply:

y= LW I A (2)

When a surface active substance is added, it spontaneously adsorbs at the


surface, and decreases the surface energy (otherwise, there would be no
spontaneous adsorption). A monolayer is formed, with the polar parts of the
surfactant molecule in contact with water, and the hydrophobic parts in
contact with air. The surface tension decrease can be identified by
dimensional arguments with a "surface pressure" rr.

Y=Yw- P (3)

where Yw is the surface tension of pure water. Extensive work done with
water-insoluble substances showed that this identification is helpful to
understand the monolayer properties and the transitions between the
different surface phases that can be found with these systems[2]. For water-
soluble substances, it becomes difficult to know the amount of surface
material, because most of the molecules are dissolved in bulk water.
Information about the surface concentration can be however obtained from
thermodynamic arguments.

1.2 Surface Thermodynamics


The free energy F of the system is the sum of the internal energy U, the
entropy term -TS and the chemical potential term ~J.li N i , where T is the
absolute temperature, S, the entropy, J.li the chemiclil potential of species i
3
and Nj the number of molecules i. The corresponding surface excess energy
is then :

~F = -T ~S + ~ J.li ~N i + 'Y A (4)


1

where L1S and L1N i are the excess entropy and number of molecules of
species i. By differentiation, one obtains :

d"( ~s
(5)
dT A

d"( ~Ni

dJ.lj A = -Gi (6)

The first equation tells us that the vanatwn of surface tension with
temperature is associated to the excess entropy. Since the surface tension
generally decreases with increasing temperature, the surface excess entropy
is positive : the surface region is less ordered than the underlying liquid. The
second equation is extremely useful, because it allows to obtain the surface
excess concentrations rj. For dilute solutions, 11 - kT In c where k is the
Boltzmann constant, and c the bulk concentration. Then :

(7)

This equation is generally called "Gibbs equation".

I. 3 Surface equations of state


The knowledge of the equation of state is obviously helpful as it is for three-
dimensional fluids. In the case of monolayers made from water-insoluble
substances, the equation of state is the relation between TI , A and T . Foams
made with water-insoluble surfactants are generally not very stable, because
the solution does not contain the amount of surfactant necessary to cover the
large new surface area created upon shaking. For water-soluble surfactants,
the amount of material at the surface is generally much smaller than the
amount dissolved in bulk water. It is therefore not easy to evaluate the
surface concentration r which is now the relevant parameter in the equation
of state (in place of A). The Gibbs equation (eq7) can be used for this
purpose, but very accurate surface tension data are needed because of the
differentiation involved in this equation. It has been recently shown that
4
surface concentrations as obtained from neutron reflectivity data are in
reasonable agreement with surface concentrations as obtained from the
Gibbs equation[3]. However, the quality of the surface tension data is
frequently not good enough to obtain accurate surface concentrations from
eq.7. For this reason, there have been many attempts to work out theoretical
equations of state for the fitting of surface tension data. The best known
equation of state is the Langmuir equation, which can be derived from the
kinetic equation :

(8)

where a and ~ are desorption and adsorption constants, respectively, and roo
the surface concentration at full coverage. At equilibrium, artat =0 and :

c
a (9)
1 - r1r 00
=

where a = a/~ and is sometimes called Szykowski concentration. When the


Gibbs equation is used afterwards, one obtains :

(10)

This equation of state only applies to monolayers which are not very dense
( r< roo), or to monolayers at oil-water interfaces where the surfactant
hydrophobic chains are not collapsed as in air and where surfactant-
surfactant interactions are less strong. In order to better account for
surfactant interactions, desorption and adsorption energies Ect and Ea can be
introduced in the kinetic equation 8:

a= ao exp (-Ect /kT) and B=Bo exp (-Ea /kT) (11)

where the desorption and adsorption energies can be assumed to be


proportional to r. This leads to the Frumkins equation of state. This
equation is frequently used, because it generally gives good fits with the
surface tension data. However, it has been shown that it leads to serious
errors when it is used to derive the surface compression elasticity [4][5]
The case of charged mono layers can be treated more rigorously. In this
case, desorption and adsorption energies depends on the surface potential 'I',
which itself depends on the surface concentration :
5

Ea =EaO- e \}' =EaO + 2 kT sinh- 1 ( 27tlB r I 2K) (12)

where e is the electron charge, lB the Bjerrum length and K-1 the Debye-
Huckellength[6]. This leads to :

(13)

n
TI =- 2 kT roo In ( 1 - rtroo) + n kT (14)

where b is a constant and n - l is the ionized fraction of the surface


molecules. These equations do not incorporate interactions between
surfactant chains and cannot be used safely at air-water surfaces. They were
however found to be useful at oil-water interfaces [7].

2. SURFACE RHEOLOGY

2.1 Parameters definition


Surface elasticity and viscosity can also be defined as excess properties[8].
In this way several couples of elastic and viscous parameters can be
introduced :
- compression parameters; by analogy with bulk systems, the elastic modulus
can be defined as :

(15)

the viscosity K can be introduced as the imaginary part of the complex


modulus describing the linear response to a sinusoidal deformation :

E' = e + iro K (16)

these parameters are also sometimes called "dilational" parameters


- shear parameters S and Tls
- transverse parameters, the elastic part being the surface tension, and the
viscous part a viscous coefficient which appears to be usally negligible,
excepted in very dense insoluble monolayers[9]
- bending parameters; the elastic part is a small correction to the surface
tension, which is only important in systems where the tension is either small,
6

either zero (microemulsions, vesicles)[lO]. In the case of foams, they do not


play any significant role.

2.2 Orders of magnitude


The orders of magnitude of these viscoelastic parameters are extremely
varied. To achieve good foam stability, a general requirement is to have a
dense surfactant layer at the foam film surfaces. In these cases the surface
pressures are typically between 20 and 50 mN/m. For sodium dodecyl
sulfate (SDS), an extensively studied model surfactant, n = 34 mN/m just
above the critical micellar concentration (erne). The surface compression
elasticities are of the same order of magnitude or larger. These surfactant
monolayers, although dense, are not in the highly condensed states of the so-
called solid monolayers : in the solid monolayers, the area per hydrocarbon
chain is close to the area of the chains in paraffinic cristals, i.e. about 20A2.
In SDS mono layers, the area per chain is typically 60 A2, and the
monolayers are in a liquid-like state. They have therefore a zero shear
modulus, and their shear viscosity is very small and difficult to measure :
about or less than w-3 g.s-1. Mixed SDS-dodecanol solutions give extremely
stable foams. In this case, the surfactant monolayer is very compact,
although not solid, with area per chains close to 30A2, and larger surface
viscosities[ll]. It must be noted that even the small viscosities quoted above
are equivalent to locally very viscous media. Indeed, the order of magnitude
of the local viscosity inside the surfactant monolayer can be obtained by
dividing rts by the thickness of the monolayer, i.e. about lnm. With Tis -Io-3
one gets a local bulk viscosity which is 1o6 times that of water. This also
means that in the case of foam films, the viscous dissipation in the
monolayers becomes comparable to that in the water layer when the
thickness of the water film becomes smaller than 1o6 times the monolayer
thickness, i.e. about lmm.

2.3 Experimental devices for surface rheology determinations


There are several types of experimental devices for the measurement of
shear and compression surface parameters. Surface shear properties have
been the most widely studied, with channel viscosimeters for insoluble
monolayers, and oscillating disks devices for soluble ones[l2]. Compression
properties have been investigated with moving barrier devices, including
excitation of surface waves, either capillary or compression
waves[l3][14][15]. Thermally excited capillary waves have also been studied
with surface light scattering techniques [9]. In these methods, the mechanical
distorsions of the monolayers are frequently sinusoidal, and large frequency
variations of the viscoelastic coefficients have been observed. These
variations are particularly large for the compression coefficients of soluble
7

monolayers, because of the important coupling with the bulk. This question
has been first addressed by Lucassen, who derived equations for the
viscoelastic parameters frequency variation when a sinusoidal compression is
applied to the monolayer[ 13]. In this model, it is assumed that upon
monolayer compression, some surfactant molecules dissolve into the
underlying water, to restore the equilibrium surface concentration. When the
monolayer is expanded again, surfactant molecules come back to the
surface, and so on. Two extreme cases are easy to understand : when the
frequency of the sinusoidal compression is low, the monolayer has always
time to reach equilibrium, and there is no resistance to the compression : E =
roK = 0. When the frequency is high, the monolayer has no time to respond,
and it behaves as if it were insoluble : K =0 and :

(17)

In the intermediate frequency range[13],

1 +
E =EO - - - - - - -
o (18)
1 + 20 + 202

ffiK =EO (19)


1 + 20 + 202

with n = --/ D/2ro dc/dr, D being the diffusion coefficient of surfactant


molecules in bulk water. A typical frequency and concentration variation is
shown in figure 1 for dodecyl trimethyl ammonium bromide (DTAB). The
characteristic diffusion time decreases when surfactant concentration
increases, so the high frequency-insoluble monolayer regime is seen at small
surfactant concentration, whereas at high concentration the monolayer
follows the compression instantaneously, and there is no more resistance to
the compression. Eqs. 18 and 19 apply in principle only to nonionic
surfactants[5]. For ionic surfactants, the model can be refined to take into
account the influence of surface potential in the diffusion process[?].
There is also certainly a frequency variation of the shear viscosity with the
frequency of the excitation, but this effect has not been much studied so far.
As we will see in the following, the compression parameters are closely
related to different dynamic phenomena occuring on foams, and it is then
useful to understand properly how do they vary with frequency and
concentration, as well as from a given system to another.
8

10 £
dyn/cm
• V: 240Hz

15

V:800Hz

10

Figure 1
Real and imaginary parts of the compression modulus versus surfactant concentration
(DTAB) for two different frequencies, 240 and 800Hz. The lines are the fits with eqs 18 & 19.
Data are from ref 52
9

II. Foaming and Foam stability

I. FOAMING AND DYNAMIC SURFACE TENSION

Foaming is in principle related to the adsorption kinetics of surfactant


molecules. Indeed, when the solution is shaken, a large surface area is
created, free of surfactant. After some time the molecules adsorb from the
bulk solution, the foam films are covered by equilibrium monolayers and
protected against film rupture. The adsorption kinetics has been extensively
studied with a variety of devices in which free surfaces are created such as
liquid jets, drops, etc., and surface tension is measured as a function of time
[16]. These devices are currently called "dynamic surface tension" devices.
When the surface is created in less than say one second, it is difficult to
control the con~ective liquid motion. In the pendent drop device, convection
can be avoided, and the surface tension variation resulting from progressive
adsorption can be conveniently modelled with diffusion controlled
processes[17]. At short times, the surface concentration can be approximated
by:

(20)

Typical equilibrium surface concentrations are of the order of one molecule


per square nm; for bulk concentrations in the millimolar range (below lQ-3
by weight), characteristic adsorption times are thus around O.ls and for bulk
concentrations of about 0.1 mM, these times increase to about lOs. For
surfactant concentrations of the order of weight %, the adsorption times are
much shorter than the characteristic times of foam formation, even assuming
that convective transport to the surface is negligible. Therefore, dynamic
surface tension is likely to be relevant only for foams made from very dilute
surfactant solutions.
In some cases, the adsorption kinetics is slower than predicted by diffusion
mechanisms. This arises when adsorption barriers are present, for instance
during the adsorption of macromolecules which currently change their
configuration at the surface. In these cases, the equilibrium tension is
reached only after long times[18]. Similar effects can be seen with smaller
molecules[19]The effect of electrostatic barriers has also be investigated
recently[20][21]. Again, the effect of convective transport on these
mechanisms has not yet been studied, and the consequence on foaming
needs to be clarified.
10

Obviously, a second requirement for good foaming if to achieve good


foam stability, since if the foam is too unstable, it can collapse during the
foaming process, and then no foam is created.

2. FOAM STABILITY

2.1 Evaporation, Ostwald ripening


Very frequently, foams are destroyed because of water evaporation. This
includes the role of temperature gradients which create surface tension
gradients and induce surface motion in the foam films. Local rapid thinning
in a foam film leads to instabilities and increase the probability of rupture of
the film at early ages[22][23]. Water evaporation across surfactant
monolayers is a well documented subject[2]. It is known that very compact
monolayers such as those made from fatty alcohols can significantly reduce
water evaporation. Ostwald ripening, which involve the transport of gas
across the foam films is affected by the presence of the surfactant
monolayers in a similar way. Again this has been studied in great detail [24].
Although the main important fact here is the compacity of the monolayer,
correlation with surface rheology can be easily made, since the denser
monolayers are also the more rigid ones.
Let us also mention the importance of the nature of the gas used for
foaming : gases soluble in water such as C02 give less stable foams than less
soluble ones such as N2, because C02 transport across water films is faster.
The stability of C02 foams can be improved by adding some nitrogen : the
gas composition in each bubble having to stay the same ( chemical potential
induced forces will prevent composition changes), the gas diffusion process
is slowed down.

2.2 Drainage with dilute surfactant solutions


When the foam is first formed, the amount of liquid in the foam films is
large, and these films are thick. They drain afterwards under the influence of
gravity and capillary pressure : the Plateau borders, connecting the foam
films are curved in such a way that the liquid pressure is smaller there than in
the flat parts of the foam films. The drainage of foam films have been
extensively studied, in particular horizontal films with devices in which the
pressure can be controlled more easily : Sleludko cells, porous plates (figure
2)[25]. Mysels distinguished several regimes in his early studies of vertical
films[22] : "mobile" films where the film thickness does not remain uniform
and sometimes surface turbulence could be observed, and "rigid" films which
drain much more regularly. Typical examples are films made from pure
SDS solutions (mobile films) and mixed SDS-dodecanol solutions (rigid
films). These features are extremely difficult to interpret. The easiest way to
II

Laser beam

Po

2R

Micrometrlc screw

Plastic box

Figure 2 a.

gas

Figure 2 b.
Experimental set-up for the study of thin liquid films ; a) Sheludko cell b) porous plate cell
(see ref 26 and 33 for details)
12
estimate the velocity of drainage is to assume that the film is flat parallel, that
its surfaces are immobile and that the fluid flows regularly from the center
towards the Plateau borders. In such a simple case, an expression derived by
Reynolds for the flow between two rigid plates brought together can be used:

v- dh - h3 L\P (21)
-- dt - 3 1l R2

where h is the film thickness, R the film radius, 11 the fluid viscosity and dP
the difference in pressure between the film center and border. This formula
shows an important fact, which is the influence of the fluid viscosity : a
simple way to slow down drainage is to increase the fluid viscosity. It is less
easy to see the influence of the surface rheology. Indeed, the film surfaces
are not solid, and surface flow can arise. Numerical calculations done in the
simple case of a flat parallel film show that the influence of surface
compression elasticity can be important for thick enough films[26]. The
surface elasticity that needs to be considered is the high frequency elasticity
EO : this is because surfactant is taken away along the surface to the Plateau
borders and there is not enough surfactant in the foam film to replenish the
surface. This result is also to be expected from emulsion films made from
the phase in which the surfactant is solubilized, for instance the water phase
for an emulsion of oil and water made with SDS. If the emulsion film were
made with the oil phase, then the two sides of the film would be in contact
with the aqueous phase containing the surfactant and the monolayer
replenishment will be much faster, because no surfactant depletion will
occur. The surface elasticity to be considered in this second case is the low
frequency elasticity, close to zero, and the corresponding velocity of
drainage is much faster. This argument has been invoked to explain the
Bancroft rule which states that the emulsion which forms is the one where the
surfactant is in the continuous phase [27] .Similar differences have also been
observed during the wetting of thin liquid films when surfactants are present
[28]. When the film thickness further decreases, the surface velocity
decreases and the role of the surface elasticity becomes negligible : for
h<O.lf.Lm and£- 10 mN/m, V can be safely approximated by eq 21.
Coming back to the case of thick films, it can be shown that a "dimpling"
instability frequently appears during the drainage of large circular
horizontal films, in which the film is pinched off around its boundary and
thicker in the center[29]. When the monolayer at the surface has a large
surface elasticity, the dimple remains centrosymmetric, and the velocity of
drainage corresponds to the velocity of flat films with thicknesses equal to
the smaller actual film thickness (eq 21). When the surface elasticity is
moderate (SDS solutions for instance), a second instability can take place in
13

which the dimple looses its circular shape and is sucked away rapidly into
the Plateau border. During this fast process, the film frequently ruptures.
The threshold for the second instability is predicted to depend upon the
shear surface viscosity[23]. Such dimpling instabilities occur more easily on
large films, and are not observed in very small films, even when these films
are made with oil-water systems of small surface tension (favouring
dimpling) [30]. This suggests that the mechanisms of drainage in emulsions
and foams are different in nature, because emulsion droplets are much
smaller than foam bubbles, and therefore much less sensitive to all the
instabilities that lead to foam film rupture in the early stages of the life of a
foam.
The foam drainage has to be distinguished from the foam film drainage.
Indeed, most of the foam liquid is located into the Plateau border, where no
surfactant depletion effects are to be expected. The foam drainage is
therefore not affected by the surfactant monolayers properties. This is
discussed in more details in the chapter by Weaire et al in this book. It must
be stressed that foam drainage can only be studied with foams that are stable
enough and where the thin film instabilities described above probably do not
occur.

2.3 Drainage of concentrated surfactant solutions


When the surfactant solutions contains aggregates, the drainage process
becomes more complex. At the beginning of the century, Johnnott and
Perrin reported that the drainage of the foam films occurs stepwise[31].
Later on, it was shown that this was due to the presence of surfactant micelles
which form layered structures below the film surfaces, and that the drainage
proceeds by expulsion of the layers, one after the other, into the Plateau
border [32][33][34]. The expulsion of a layer of micelles begins by the
nucleation of one (or more) holes, which expand in different ways(figure 3).
When the film is thick and contains several layers, the radius of the holes
increase as the square root of time. This has been modelled by Kralchevski
et al by assuming that the micelle layer flows regularly through the Plateau
border[32]. It has also been noted that the problem is analog to the
spreading of stratified drops on solids, where an equivalent diffusion
coefficient can be defined : Rbole - ..J Deff t , with :

(22)

where Ild is the "disjoining pressure", i.e. the force per unit area exerted
between the two film surfaces[35]. When this force is mainly a van der Waals
interaction, Deff - A/61t11h, where A is the Hamaker constant. This expression
gives for Deff an order of magnitude of to-8 cm2fs, close to the measured
14

numbers[34]. When the films are thinner, the opening of holes in a micelle
layer can be followed by the rupture of the liquid rim which borders the
hole into small droplets, via a mechanism similar to the Rayleigh instability
of a liquid jet.[36] The radius of the hole then increases linearly with time.
This is similar to dewetting processes[37] and although less rapid, to the
bursting of foam films[22][38].

Figure 3.
Image of a film made with an SDS solution above erne with the set up of fig. 2a. The darker
regions are thinner, and when they expand, they push a rim that can fraction into tiny little
drops visible on the picture (after ref 34)

Very similar features were observed in mixed polyelectrolytes-surfactant


solutions, at concentrations where micelles are not present[39]. The layering
in the foam films has been attributed to the strong electrostatic repulsion
between the polymer chains, which result in a "correlation hole"
phenomenon, i.e. a chain network in which the chains sits at a well defined
distance from each other. This can be seen in the small-angle scattering
spectra of the polymer solutions where a peak is clearly visible [40] . The
drainage process is more difficult to understand than in the case of micelles,
since here the layers which disappear into the Plateau borders are network
layers.
15

The drainage of other types of concentrated solutions has been less


extensively studied. The particular case of nonionic surfactant solutions
above the cloud point can be mentioned. These solutions are cloudy,
because they slowly phase separate into a micelle rich phase and a dilute
phase, with surfactant concentration close to the erne. In some cases, the
interfacial tensions are such that droplets of the concentrated phase can enter
into the air-dilute phase interface and bridge to two sides of a thin film
formed from the dilute phase[41]. This is one mechanisms of action of
antifoams. The drainage process leads then quickly to the rupture of the
foam films, once the film thickness is comparable to the size of the droplets
of the concentrated phase (microns)[42].
It has been noticed by Friberg that foam or emulsions made when lamellar
phases are present can be very stable. When the lamellar phase is viscous, one
obvious effect is the slowing down of the drainage process. A more subtle
effect could be that the interfacial tensions are such that the lamellar phase
cannot enter the surface of foam films, and is thus trapped in the Plateau
borders : this had been indeed observed [43]. Particles that cannot enter the
foam films act as foam stabilizers[42]. The drainage of foam films made
from lamellar phases has been studied in the case of a very dilute phase. The
observed behavior was very complex, and no clear understanding was
achieved[ 44].

2.4 Surface Forces


If the foam drainage has been smooth enough, so that no early rupture has
occurred, a regime where interactions between the two sides of the film
become significant can be attained : this is in a thickness region of about
50nm and less. Typical interactions are van der Waals forces (attractive) and
electrostatic forces (repulsive)[45]. When the electrostatic repulsion is strong
enough, an energy minimum can be found for thicknesses of about 1Onm.
Figure 4 shows a schematic diagram for the disjoining pressure in which
short range repulsive forces have also been taken into account

A
I1vdw---- Ileiect- B e -Kh Ilsteric, hydration- C e -h/1.. (23)
67th3

where B and C are constants, and A is the range of the short range forces,
typically a few tens of nm. If a lateral pressure L\P is applied to the film
(gravity, Laplace pressure), the thickness of the films decresases down to ht
(see fig 4), and the corresponding equilibrium film is called "common black
film". If the pressure L\P is larger than the electrostatic barrier, one reaches
after drainage a very small film thickness where the water layer thickness is
of order A : this is the so-called "Newton black film". Let us note that the
16

1td "Newton black film" ~~~


/

"Common black film"

Figure 4.
Schematic representation of the variation of the disjoining pressure fld with film thickness h.
The dootted lines correspond to different applied pressures .1P and show the final equilibrium
thicknesses of the film.

surface tension of the film is also affected by the interactions between film
surfaces[46] :

n (h)
'Yfilm =2y+ f hdTI (24)
TI(oo)

2.5 Film rupture


Different models for film rupture can be found in the literature. In the first
type of model, worked out first by Sheludko and later refined by Vrij, it is
17

proposed that thermal thickness fluctuations can be amplified in some


circumstances. If one writes (horizontal film)[47]:

h(x,y)= ho + Lq Uq exp(iqxx+iqyy) (25)

then the mean square amplitutude of the fluctuation uq is :

<u 2> = _ _kT


__ (26)
q an
2yq2- ah
In situations where I1 decreases when h decreases, one can find q values for
which the denominator of eq 26 vanishes. This happens for instance for
h<hmax (fig4); if the short range forces are neglected, this process can be
shown to lead to a film instability and rupture[47].
When the film surfaces are covered by a dense surfactant monolayer, the
short range forces cannot be neglected. It has been shown in experiments
done with a mica plate surface force machine that the fusion of bilayers
adsorbed on the mica plates can only be achieved when these bilayers are
formed from dilute surfactant solutions. Above erne, the bilayers are
compact and cannot be forced to fuse, even if pressures up to thousands of
atmospheres are exerted between the mica plates[48]. Such forces are well
above the strength of van der Waals forces. Furthermore, the Vrij's model
predicts variations of the rupture time for rupture with film size, surface
tension and other parameters, variations that do not correspond to
experimental observations. It is therefore likely that other mechanisms
operate in the process of rupture of foam films covered by dense surfactant
monolayers.
It has been proposed by other authors (Israelachvili et al for bilayer
fusion[48], Exerowa et al for foams films[24] and de Gennes for emulsions
films[ 49]) that the rupture occurs rather via thermal concentration
fluctuations in the film surfaces. This process is controlled by compression
elasticity rather than by surface tension and disjoining pressure (eq26), the
mean square amplitude of the density fluctuations in the surfactant layers
being[9] :

(27)

Once the surfactant layer is less dense at some point, the chains can tilt
towards the opposite layer, and the hydrophobic attraction between opposite
18

surfaces will be enhanced : this is the instability mechanism proposed by


Israelachvili et al [47]. Of course, the state of the models is very preliminary,
as is also the problem from the experimental side. Correlations between
pressure threshold and compression elasticity are lacking. The measurement
of pressure threshold for the rupture of foam films is not easy and the
reproducibility of the results is poor. Attempts to measure this threshold in
emulsions have been made by Bibette et al by an osmotic compression
method[ 50]. In preliminary experiments done with the same technique, a
correlation between the rupture threshold and the compression elasticity has
been found[51]. Clearly much more work is needed to clarify this complex
problem.

Conclusion

The foam behaviour is largely controlled by the properties of the surfactant


monolayers that protect the foam film surfaces. The knowledge of surface
tension alone is not sufficient to understand the foam properties. The surface
viscoelasticity, and compression elasticity in particular play an important role
in a variety of dynamic processes. Many of these processes, and especially
film rupture need further investigations both from the theoretical and
experimental point of view

REFERENCES
I. see for instance in : A.Adamson, "Physical Chemistry of surfaces" Wiley, New York,
1976
2. G.Gaines, "Insoluble Mono/ayers at Liquid-Gas Interfaces" Wiley, New York, 1966
3. J.R.Lu, E.M.Lee, R.K.Thomas, J.Penfold, S.L.Fiitsch, Langmuir 9 1352 (1993)
4. E.Lucassen-Reynders, Pro g. Surface Membrane Sci. 10 253 (1976)
5. Y .J ayalakshmi, L.Ozanne, D.Langevin, J. Colloid Interface Sci. 170 358 ( 1995)
6. J. T.Davies, E.K. Riddeal "Interfacial Phenomena" Acad.Press 1961
7. A.Bonfillon, D.Langevin, Langmuir 9, 2172 (1993)
8. F.Goodrich, Proc. Royal Soc. London A374, 341 (1981)
9. D.Langevin, editor, "Light Scattering by Liquid Surfaces" M.Dekker, 1992, chapter II
I 0. W.M.Gelbart, A.Ben-Shaul, D.Roux, editors, "Micelles, Membranes, Microemulsions
and Mono/ayers" Springer, 1994, p 62
II. D.O.Shah, N.F.Djabarrah, D.T.Wasan, Colloid Polym.Sci. 251 1002 (1978)
12. R.Miller, R.Wtistneck, J.Kragel, G.Krezschmar, Colloids Surfaces, 111, 75 (1996)
I 3. E. H. Lucassen, J.Lucassen, Adv. Colloid Interface Sci. 2 347 (1969)
14. P.Joos, M.van Uffelen, J.Colloid Interface Sci. 155 271 (1993) and references
therein
15. D.Langevin, Colloids Surfaces 43 121 (1990)
19

16. J.van Hunsel, P.Joos, Colloid Polym. Sci. 267 1026 (1989) and references therein
17. S.Y.Lin, K.McKeigue, C.Maldarelli, AICHE J. 36 1785 (1990)
18. G.Serrien, G.Geeraerts, L.Gosh, P.Joos, Colloid Surf. 62 219 (1992)
19. A.Colin, J.Giermanska-Kahn, D.Langevin, B.Desbat, Langmuir 13 2953 (1997)
20. A.Bonfillon, F.Sicoli, D.Langevin, ].Colloid Interface Sci. 168 497 (1994)
21. H.Diamant, D.Andelman, Europhys.Lett. 34, 575 (1996) and to appear
22. K.Mysels, K.Shinoda, S.Franke1, "Soap Films" Pergamon 1959
23. J.L.Joye, G.Hirasaki, C.A.Miller, ].Colloid Interface Sci., 177 542 (1996)
24. D.Exerowa, D.Kashchiev, D.P1atikanov, Adv.Colloid Interface Sci. 40 201 (1992)
25. A.She1udko, Adv. Colloid Interface Sci. 391 (1967)
26. A.Sonin, A.Bonfillon, D.Langevin, ].Colloid Interface Sci.l62 323 (1994)
27. I.lvanov, D.S.Dimitrov, "Thin Liquid Films", Surfactant Sci. Ser. 29, p379
M.Dekker, 1988
28. V.Bergeron, D.Langevin, Phys.Rev.Lett., 16 3152 (1996)
29. J.L.Joye, G.Hirasaki, C.A.Miller, Langmuir, 8 3085 (1992); 10 3174 (1994)
30. O.D.Velev, G.N.Constantinides, D.G.Avraam, A.C.Payatakes, R.P.Borwankar,
1. Colloid Interface Sci. 175 68 (1995)
31. E.S.Johnnott, Philos. Mag. 11 746 (1906); J.Perrin, Ann. Phys. 10 160 (1918)
32. A.D.Nikolov, D.T.Wasan, ].Colloid Interface Sci.l33 1 (1989)
3 3. V .Bergeron, C.Radke, Langmuir 8 3020 (1992)
34. A.Sonin, D.Langevin, Europhys.Lett. 22, 271 (1993)
3 5. P.G.de Gennes, A.M.Cazabat, C.R.Acad.Sci. Ser II 310 1601 (1990)
3 6. A.Jimenez-Laguna, C.Radke, Langmuir 8 3027 (1992)
3 7. F.Brochard-Wyart, C.Redon, Langmuir,8 2324 (1992)
38. P.G.de Gennes, in "Physics of Amphiphilic Layers" 34, p64, Springer 1987,
39. A.Asnacios, A Espert, A.Colin, D.Langevin, Phys.Rev.Lett.18, 4974 (1997)
40. M.Nierlich, C.E.Williams, F.Boue, J.P.Cotton, M.Daoud, B.Farnoux, G.Janninck,
C.Picot, M.Moan, C.Wolf, M.Rinaudo, P.G.de Gennes, ].Physique 40 701 (1979)
41. A.Bonfillon-Colin, D.Langevin, Langmuir, 13, 599 (1997)
42. P.R.Garrett, "Defoaming, Theory and Industrial Applications" Surfactant Sci.Ser. 45
p1 M.Dekker,1993.
43. S.Friberg, Adv. Liq. Cry st. 3 149 (1978)
44. A.Sonin, D.Langevin, Adv. Colloid Interface Sci. 51 1 (1994)
45. J.N.Israelachvili "Intermolecular and Surface Forces" Acad.Press, London 1985
46. A.Scheludko, B.Radoev, T.Kolarov, Trans.Faraday Soc. 64, 2213 (1968)
47. A.Vrij, Discuss. Faraday Soc. 42 23 (1966)
48. C.A.Helm, J.N.Israelachvili, P.McGuigan, Science 246, 919 (1989); Biochemistry
31, 1794 (1992)
49. P.G.de Gennes, to be published
50. J.Bibette, D.C.Morse, T.A.Witten, D.A.Weitz, Phys.Rev.Lett., 69 2439 (1992)
5 I. A.Colin, thesis, Paris 1994, unpublished
52. C.Stenvot, D.Langevin, Langmuir, 4, 1179 (1988)
20
FOAMS AND EMULSIONS: THEIR STABILITY AND BREAKDOWN
BY SOLID PARTICLES AND LIQUID DROPLETS

The Colloid Chemistry of a Dog's Breakfast

R. A VEYARD, B. P. BINKS, J. H. CLINT AND P. D. I. FLETCHER

Surfactant Science Group, School of Chemistry, University of Hull,


Hull HU6 7RX, UK

1. Introductory Remarks

From the standpoint of physical chemistry foams and emulsions are rather complex,
multicomponent colloidal systems whose behaviour is often crucially dependent on
composition, temperature and electrolyte concentration. Industrially encountered examples
usually contain commercial or natural surfactants. Commercial antifoam formulations are
themselves complex, containing as they often do a dispersed oil in which solid particles
are present. It can be appreciated therefore that a system containing both foam and
antifoam is somewhat of a "dog's breakfast", and brings to mind a description of colloidal
systems given by Hedgest : To some the word "colloidal" conjures up visions of things
indefinite in shape, indefinite in chemical composition and physical properties, fickle in
chemical deportment, things infilterable and generally unmanageable.
When faced with the problem of devising plausible mechanisms for the action of
antifoams on foams it is first necessary to identify some key processes which are thought
to occur in elements of the system e.g. a single soap film or a single oil droplet at the
surface of a surfactant solution. There will then be additional processes which can only
occur in a foaming system as a whole, such as the collection of oil drops in an assembly
of Plateau borders. It is clear that in order to test proposed mechanisms of foam breaking,
model systems must be devised which contain pure components and solid particles of
known (probably spherical) geometry, size and wettability.
The structures of foams and emulsions, and the processes which control their
stability, have much in common. The mechanisms by which particles interact with foams
and emulsions and modify (increase or decrease) the stability also have strong similarities.
Foams are dispersions of gases in liquids whereas emulsions are liquid in liquid
dispersions. We will confine our discussion of foams to aqueous foams stabilised by low
molar mass surfactant (as opposed to polymers such as proteins and polymeric

' Hedges, E. S. (1931) Colloids, Edward Arnold, London.

21
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 21-44.
© 1999 Kluwer Academic Publishers.
22

surfactants). The analogous emulsions are those in which the gas phase is replaced by say
a hydrocarbon oil as the disperse phase i.e. oil-in-water (o/w) emulsions. The inverse
emulsions (w/o) are also possible in which the aqueous phase is dispersed in the
(continuous) oil phase. Formally, one might choose to regard a liquid aerosol as an
inverted foam, although we will not be concerned with aerosols here.
In what follows we will first consider some of the important processes and
phenomena contributing to the stability of foams and emulsions. Then follows a
discussion of the behaviour of particles at fluid/fluid interfaces. A pre-requisite for solid
particles or liquid droplets to rupture thin liquid films in foams is that the particles or
droplets should first enter the surfaces of the lamellae. Solid particles can have a variety of
effects on the stability of emulsions, but again for these effects to be possible the particles
must reside in the interfaces between the droplets and the continuous phase. Finally we
give a brief overview of what is currently known (or believed) concerning the effects of
particles on foam and emulsion stability. Of course in a thermodynamic sense neither
emulsions or foams can be stable, and in the present context stability is taken to mean
"kinetic" stability or metastability unless otherwise indicated.

2. Foam and Emulsion Stability

2.1. SOME GENERAL CONSIDERATIONS

The discussion will be focused mainly on foams but as will be apparent much of what is
said will be equally applicable to emulsions. Distinction will be made where appropriate
between behaviour in the two kinds of system, which arises in the main from the different
"drop" sizes in emulsions and foams, the different interfacial tensions between dispersed
and continuous phases, and differences in density between dispersed gas in foams and
dispersed liquid in emulsions. Foams can be generated simply by shaking a surfactant
solution so that air bubbles become entrapped. The bubbles rise and liquid drains fairly
rapidly from between them. In the upper part of the foam the bubbles become distorted
giving a polyhedral structure (sometimes referred to as polyederschaum) in which the thin
aqueous films are effectively planar and joined (at 120°) at Plateau borders (Figure 1). In
the lower part the bubbles are essentially spherical and the foam here is termed
kugelschaum [ 1,2].
The bubble or cell size usually encountered in foams (say 50 Jlm to several mm
in diameter) is much greater than drop sizes in emulsions. Colloidal dimensions are of the
order of 1 om to 1 Jlm; foams are considered to be colloids because the lamella thickness
lies in the colloid range. In contrast emulsion drops are typically around 1 Jlm in
diameter. It is difficult to produce small bubbles in foams since for one thing, the surface
tension of bubbles is greater than the interfacial tension of emulsion droplets, often by a
factor of ca. 10. Further, in a polydisperse system the solubility of the dispersed phase in
the continuous medium depends on the droplet/bubble radius, r . In the case of bubbles in
foams this arises as a result of the excess pressure, Ap, within the gas bubbles, given by
23

kugelschaum

Pe

gas (b)
(a)

Figure I. (a) Formation of foam by blowing gas through a sinter into a surfactant solution. At the bottom of the
column the foam is "wet" and consists of gas bubbles dispersed in the solution (kugelschaum). As the liquid
drains from between the rising bubbles the foam becomes polyhedral, and consists of planar films joined at
Plateau borders. This kind of foam is sometimes called polyederschaum. (b) A Plateau border, at which 3
plane films meet at 120" along a line . The border has one radius of curvature r; the other radius is infinite so
r
the pressure, Ps· in the border is /r less than that, PA· in the plane films (see text). The capillary pressure
sucks liquid from the films into the Plateau border.

the Laplace equation

f1p=2ylr (1)

where y is the surface tension of the solution. The smaller r the greater f1p and hence the
greater the solubility of the gas in the continuous phase. This can lead to rapid bubble
disproportionation, in which the smaller bubbles shrink and the larger bubbles grow. A
similar effect occurs in emulsions and is termed Ostwald ripening, which is usually
discussed in terms of an analogue of the Kelvin equation

(2)

Heres and s~ are the solubilities (in the continuous phase) of the dispersed phase (molar
volume Vm), in the form of bulk liquid and of droplets of radius r respectively; y is the
interfacial tension between the two liquid phases. Since the solubilities in emulsion
systems (e.g. solubilities of hydrocarbons in water) are very much smaller than those of
dispersed gases in foams, Ostwald ripening is much slower than bubble
disproportionation.
To obtain a polyhedral structure it is necessary to distort the fluid spheres. As
noted, the pressure inside a sphere, radius r, exceeds that outside by the Laplace pressure,
f1p given in equation 1. To obtain a flat film between droplets or bubbles an "external"
stress must be applied to overcome the Laplace pressure, somewhat like pressing 2
balloons together. Although in emulsions the interfacial tension yis often about a factor
of 10 lower than that in foams, ,1p for emulsions is much larger than for foams as a result
of the much smaller droplet sizes. It follows that the stress (due e.g. to buoyancy or van
24
der Waals attractive forces across the thin films, see below) required to distort emulsion
drops is much greater than that needed to distort gas bubbles. In addition, the difference in
density between dispersed and continuous phases in foams (giving rise to the buoyancy of
the bubbles) is much greater than in emulsions. Droplets in emulsions, even after
creaming or sedimentation, commonly remain spherical.
Foams formed from surfactant solutions are much less stable than emulsions
formed with similar surfactants. When two droplets or bubbles come together a thin film
exists between them. For coalescence to occur this thin film must rupture. In a
dispersion of unflocculated spherical entities the (very small diameter) films exist only
during encounters arising from Brownian motion. In flocculated and in polyhedral
structures however, where the films are "permanent", the chances of rupture are much
increased and, importantly are greater the larger the film area, explaining in part the
fragility of foams relative to emulsions. We now consider further the properties of liquid
films which give rise to stability and to rupture.

2.2 FILM DRAINAGE [1-3]

After formation and before rupture occurs, the liquid lamellae separating foam cells or
emulsion droplets thin as a result of drainage under gravity and suction into Plateau
borders driven by the Laplace pressure. With reference to Figure 1 it is seen that the
pressure, p 8 , in the Plateau border is lower than that, p A• in the gas cells and the planar
film. In non-spherical systems, where the curvature can be described in terms of the two
principal radii r 1 and r 2, the Laplace equation has the form

(3)

For a Plateau border one of the radii of curvature is infinite so that .1p, driving capillary
suction from the films, is ylr.
For the drainage of a vertical film of pure liquid there is no velocity gradient
across (normal to) the film (Figure 2a), so that plug flow occurs [2]. This results in a
rapid thinning of the film. Further, such a film when very thin could not be stabilised by

(a)
Figure 2. Velocities of flow across liquid films of (a) a pure liquid (b) liquid bounded by monolayers of
adsorbed surfactant, which inhibit flow at the surfaces. A tension gradient in the surface is generated by the
surface concentration gradient (see text). After Garrett [2].
25
repulsive surface forces (see below). It follows that films of pure liquids are not stable; it
is necessary to have adsorbed layers of surface active material (e.g. a surfactant) at the
lamella surfaces. This can drastically reduce the drainage rate since the surfaces become
rigid, as if bounded by solid walls, and a parabolic velocity profile is produced as
illustrated (after Garrett [2]) in Figure 2b. Thus an interface with adsorbed layers can
withstand tangential stress. Liquid flow along the interface tends to increase the surface
concentration of surfactant "downstream" giving rise to an interfacial tension gradient;
movement of the interface is arrested (no-slip boundary condition) if the gradient is large
enough.
Just as liquid flow can cause a tension gradient, so a tension gradient can result in
tangential flow of underlying liquid; this phenomenon is termed the Marangoni effect. If
(part of) a relatively thick film is rapidly expanded (hence thinned), such that surfactant
adsorption from the film interior cannot keep pace with the expansion, a tension gradient
is produced which, by the Marangoni effect causes a flow of underlying liquid back into
the expanded region causing it to thicken again. In this way potential points of rupture are
healed. In very thin films, if a local expansion occurs, there may be insufficient surfactant
available in the film interior to adsorb and remove tension gradients. In the limit where
the lamella is so thin that no surfactant exists in the "bulk" interior, the lamella surfaces
are purely elastic; if some adsorption is possible on extension of a film it behaves visco-
elastically.
In summary, for stable films to exist during drainage it is necessary to have
tension gradients within the surface. The gradients, which can drastically reduce the
drainage rate, are readily provided by adsorbed layers of low molar mass surfactant.
Interfacial tension gradients also tend to heal thin spots of potential film rupture. As will
now be seen, the surfactant layers bounding thin films can also give rise to net repulsive
forces across films which have thinned sufficiently. We note however that in practice it is
common for films to rupture before they become "thin" films in which these surface forces
operate.

2.3. THIN FILM FORCES AND FILM RUPTURE [1,3-5]

The draining films so far discussed have a thickness of the order of micrometres or more.
It is not until the films have thinned to the order of a few tens of nanometres or less than
one surface begins to interact with the other through "surface" or "colloid" forces. For
these forces to aid film stability there must at some thickness be a net repulsion between
the lamella surfaces. Repulsion can arise from a variety of sources including coulombic
repulsion between the hydrophilic groups of ionic (or even nonionic) surfactants and steric
repulsion, particularly when the hydrophilic groups are chain-like entities. For
thicknesses a little greater than those where net repulsion exists however, net attraction
can occur as a result of the operation of van der Waals forces. The interactions between
surfaces result in changes in the tensions of the interfaces (expressed in terms of an excess
film tension) and in concomitant changes in pressure normal to the interfaces. This
pressure is termed the disjoining pressure, II( h), and is defined in terms of the free energy
of interaction per unit area of film, G(h), and film thickness h by
26

II (h) = - [()G(h) I ()h h (4)


It can be appreciated from equation 4 that for the film thickness corresponding to strongest
attraction (minimum in G(h) ) the disjoining pressure is zero. If capillary suction is
tending to pull liquid (laterally) out of the film then, for a plane film in equilibrium with
the adjoining Plateau border, the disjoining pressure is equal to the capillary pressure. A
schematic plot of G(h) versus h, for say a thin soap film, and the corresponding disjoining
pressure isotherm are shown in Figure 3. The capillary pressure is indicated by the
horizontal dashed line in the disjoining pressure isotherm. For a system giving rise to
such a disjoining pressure isotherm that there will exist a metastable film with thickness
h2 • If the film can surmount the energy maximum in some way then another, thinner

G(h)

capillary
II( h) pressure

Figure 3. Schematic free energy (top) and disjoining pressure (bottom) isotherms. For a planar film in
equilibrium with an adjoining meniscus the disjoining pressure in the film is equal to the capillary pressure.
Two "stable" film thicknesses are possible for the case represented, h 2 and h 1, where the disjoining pressure
equals the capillary pressure on parts of the isotherm where dll( h)ldh is negative. Since stability results from
the competition between capillary pressure tending to thin the film and disjoining pressure tending to thicken it,
it follows that equilibrium is not possible if dll( h)ldh is positive.

(meta)stable film with thickness h1 will form. Films with thickness h 2 and h 1 correspond
to respectively the common black film and the Newton black film. The black or grey
appearance arises because the thicknesses are much less than the wavelength of visible
light. A soap film with thickness h 1 is effectively a surfactant bilayer; all the unbound
water has left the film which could be described as a 2-dimensional crystal [6].
Although it has been appropriate to consider films as being plane parallel
structures up to now, capillary waves are present in the film surfaces as a result of
Brownian motion or mechanical effects. The waves are damped by the action of the
surface tension (which is the cause of the Laplace pressure). The pressure in the film
under the crest of a wave exceeds that under a trough so the Laplace pressure tends to push
liquid back into the thinner parts of the film. If however there are (only) attractive forces
between the surfaces, this will act to thin even further the thinner parts of the film,
increasing the chances of forming a hole which can lead to rupture. The capillary pressure
27
depends on the amplitude and wavelength of the waves and on the surface tension. Vrij
and Overbeek [7] have shown that there is a critical wavelength, It", given by

= ( 2n:2y )0.5 (5)


Ac dll I dh

above which disjoining forces of attraction are expected to dominate over the capillary
pressure, and the perturbation will grow. In a system in which only attractive (van err
Waals) surface forces are considered, dll(h) I dh is always positive. Above ..:tc, the greater
the wavelength the lower the damping effect of the capillary pressure, and so the rate of
growth of the perturbations increases. It is supposed that a film will rupture when the
amplitude of a perturbation is equal to the mean film thickness, that is the critical
thickness, he, (possibly around a few tens of nanometres).
In soap films there are of course repulsive surface forces in addition to van err
Waals forces and, as Garrett [2] points out, this can mean that in systems where the
surfactant concentration is high enough (to keep the monolayers close-packed) instead of
rupture occurring at the critical thickness, so-called "black spots" of thin film can form.
This phenomenon presumably results from changes occurring in the disjoining pressure
isotherms with changes in surfactant concentration. Once formed, the black spots can grow
to cover the whole film. In systems exhibiting black spots foam stability tends to be
high. Bergeron discusses thin film forces in much more detail elsewhere in this book.

2.4. PHASE INVERSION IN EMULSIONS

When say equal volumes of oil and water are homogenised in the presence of stabilising
surfactant, one emulsion type (o/w or w/o) has in a relative sense a very much greater
stability than the other, and this is the type ultimately obtained. The preferred emulsion
type is known to depend on a number of variables including temperature, electrolyte
concentration and surfactant structure. The empirical Bancroft's rule states that the
preferred emulsion type is that in which the surfactant partitions strongly into the
continuous phase. Thus, an explanation of Bancroft's rule often given runs as follows [1].
To obtain film stability during drainage, tension gradients are needed. Such gradients in
the surfaces of a thin film separating two droplets could be removed by adsorption if an
ample supply of surfactant were to be present in the droplets. On the other hand, transport
of surfactant from the continuous phase into the thin film is presumably slow and cannot
effectively remove tension gradients before the Marangoni effect acts to stabilise the film.
This explanation cannot be general however. Careful experimental examination
of emulsion behaviour shows that the preferred emulsion type is that in which aggregated
surfactant (i.e. micelles or microemulsion droplets) resides in the continuous phase.
Often, unaggregated ("monomer") surfactant distributes strongly into one (the droplet)
phase whereas the micelles form (above the critical micelle concentration, erne) in the
other (continuous) phase [8,9]. If the surfactant is present just above the erne it follows
that most of the surfactant in the preferred emulsion is actually present in the droplets.
Although it is beyond the scope of this chapter, we would remark that an alternative, and
28
attractive theoretical explanation of Bancroft's rule has recently been advanced by Kabalnov
and Wennerstrom [10] and involves the bending elasticity and preferred curvature of close-
packed surfactant monolayers at o/w interfaces.

3. Particles and droplets at fluid/fluid interfaces

The stability of thin liquid films and hence of foams and emulsions can be modified by the
presence of solid particles and, in the case of foams, liquid droplets. To be effective in
rupture of thin films however, the particles or droplets must be capable of entering the
surfaces of the films. Before looking at the effects of particles and droplets on the stability
of foams and emulsions, the factors involved in the incorporation of (spherical) solid
particles and liquid drops into liquid surfaces are described. It will be supposed that the
particles or droplets are sufficiently small for the liquid surface to remain planar up to
contact.

3.1. SPHERICAL SOLID PARTICLES AT LIQUID INTERFACES

We consider spherical particles for simplicity and because model experiments on foam and
emulsion stability, which are capable of reasonably secure interpretation, have been carried
out using spherical particles. However, in industrially important systems the particles are
often of indeterminate shape, and indeed irregularly shaped particles in for example
antifoam formulations, are often more effective in rupturing liquid films.
Imagine a spherical hydrophobic particle originally present in a bulk aqueous
phase close to the air/water surface. (Use of the term "water" here is taken to include
aqueous surfactant solutions). There are two aspects to a consideration of whether or not
the particle will enter the liquid surface. For the particle to enter and remain in the
interface there must be a stable (or a metastable) configuration in the interface. It is also
possible however that even if such a configuration does exist a metastable thin aqueous
film will form between the particle and the interface, particularly if surfactant is present in
the system, and prevent particle entry. We will not consider further the exclusion of solid
particles from the interface by such metastable films, but will return to the question in
connection with the entry of liquid droplets, where the problem has been explicitly
addressed experimentally.

When a particle leaves bulk water and enters the liquid surface (Figure 4) the
extent of the various interfacial areas changes and in addition a 3-phase contact line is
formed around the particle. The areas and the line all have energies and tensions associated
with them. The line tension, denoted by 'l", is a one-dimensional analogue of surface
tension and has units of energy per unit length or force. Whereas surface tension is always
positive (i.e. surfaces at equilibrium always tend to contract), line tensions can be positive
(contractile line) or negative (line tends to expand). In what follows we assume that line
tension is positive, in conformity with experimental results for solid/liquid/vapour line
tensions.
29
contact
line
A

rur ~
/~
water g---B water

Figure 4. Entry of a spherical particle from water into the air/water interface. An area A of air/water
interface and area B of solid/water interface are lost and area B of solid/air interface is formed . The 3-phase
contact line is also generated. The contact angle (} is measured into water.

Neglecting possible effects of line tension for the moment, 1t 1s readily shown
from a consideration of changes of interfacial areas that the free energy G of a system with
a spherical particle, radius r, resting at equilibrium in the interface, relative to that with the
particle completely immersed in the more wetting phase, is given by [11]

(6)

where Yw)s the surface tension of the aqueous solution. The positive sign in equation 6
is to be taken when ()(measured into the aqueous phase) is > 90". In Figure Sa G is
shown schematically as a function of 0, from which it is seen that the particle is most
8
0

G
~
0
c
0
~
.!;::
&H
water _ _ __ -~

90° 0 h 2r
(c)
(a) (b)

Figure 5. (a) Free energy G of particle at equilibrium in an interface, relative to the free energy in the more
wetting bulk phase, as a function of contact angle. (b) Free energy, of a system with a hydrophobic particle in
the interface relative to that for the system with particle in air, as the particle is moved through the interface.
The minimum occurs when the particle has its equilibrium contact angle and the depth of the minimum, G, is
given by equation 6. (c) Depth of immersion of a particle in an interface.

strongly attached to the interface when the contact angle is 90°. (It is interesting to note
that the reason why a particle is "adsorbed" to the interface, rather than resting entirely in
the more wetting phase, is that liquid surface is eliminated when the particle enters the
surface). One can imagine conceptually moving the particle vertically through a series of
30
non-equilibrium positions in the surface and keeping the interface planar up to the particle
i.e. letting the contact line slip around the particle and the contact angle change
continuously. The free energy associated with this, shown in Figure 5b as a function of
the depth of immersion h, is of course minimum for the equilibrium configuration. The
value of G (corresponding to the equilibrium contact angle) given by equation 6 is
indicated in the Figure.
Theoretical estimates of line tension are of the order of 10- 11 N [12], and such
values would have a negligible effect on the uptake by an interface of particles of the sizes
of interest here. Recently however carefully determined values of line tension of the order
of 10·6N and greater have been reported [13], which if correct would have a crucial effect on
particle uptake [ 14, 15]. When (positive) line tension is accounted for equation 6 becomes
[11]

- G= 11: r 2 y wv[(l ± cosB) 2 - 21".


Ywv rsmO
(I± cos e)] (7)

where as before the positive sign in the brackets is to be taken when > 90°. It is evidente
from equation 7 that a positive line tension reduces the magnitude of G. This is because
the contact line squeezes the particle further into air (for contact angles > 90") or into the
water (contact angles < 90"). In so doing it alters the contact angle, i.e. the particle
wettability. The contact angle appearing in equation 7 is the equilibrium value in the
presence of the line tension. It is interesting that the free energy calculated as a function
of the depth of immersion as previously but now with a positive line tension acting,
exhibits 2 maxima as seen in Figure 6. This means there is an activation energy for the

0 h
Figure 6. Free energy of a system with a hydrophobic particle in the interface relative to that for system with
the particle in air, as the particle is moved through the interface. Curve (a) is for zero line tension, as in Figure
5b. The action of positive line tension introduces 2 maxima into the curves so that there is an energy barrier to
entry from either bulk phase. Curve (b) is for 1: = 1:, where the minimum energy is equal to that when the
particle is completely immersed in the more "wetting" phase (air in the case of a hydrophobic particle). For
this line tension it is thermodynamically feasible for the particle to leave the interface spontaneously, although
there is an energy barrier to be surmounted. As 7: increases beyond 1:, a point is ultimately reached (at 7:m)
where the energy minimum disappears (as in curve (c)); for such a condition there is no equilibrium possible
for the particle in the interface. Between 1:, and 7:m only metastable configurations in the interface are possible.
31
process of particle entry into the interface, from either bulk phase. It turns out also that
there are 2 values of line tension of particular interest. For a value denoted 'l'c, the value of
G for the particle at equilibrium in the surface, relative to G when the particle is in the
more "wetting" bulk phase (vapour for a hydrophobic particle), is zero. For line tensions
above 'l'c, the minima in the free energy versus depth curves correspond to metastable
particle configurations. There is a maximum value of -r = 't'mo above which there is no
stable configuration possible for the particle in the interface, i.e. there is no free energy
minimum.
If positive line tensions of the magnitudes reported by Neumann and co-workers
were to operate in systems of present interest it would mean that spherical particles with
radii less than the order of 100 to 150 J..Lm would not be able to enter surfactant solution
surfaces and consequently they could not act as antifoam agents.
Finally we remark that it is possible to obtain the contact angles of small
spherical particles with the solution/vapour interface using a Langmuir trough technique
[16-18]. A monolayer of monodisperse particles spread on an aqueous subphase can be
compressed much the same as an insoluble molecular monolayer. The resulting surface
pressure-surface area isotherm exhibits a knee which is assumed to correspond to the
collapse of a hexagonally close-packed particle monolayer. The surface pressure
corresponding to collapse, nc, was equated by Clint and Taylor [17] to the free energy
required to remove particles in unit area of interface into the more "wetting" phase [16].
Noting that the free energy of particle attachment to the surface is given by equation 6, the
simple relationship is obtained between the collapse pressure and the contact angle (in the
absence of line tension)

cosO=±[ - - - - 1)
1rc2.J3 (8)
nrwv

Equation 8 has subsequently been modified to allow for inter-particle repulsion in the
monolayer [19] and for the effects ofline tension [20]. Indeed it is possible in principle to
measure line tensions using the Langmuir trough [16,20].

3.2 LIQUID LENSES AT LIQUID INTERFACES

Much the same considerations apply to the entry of an oil drop from aqueous solution into
the solution surface (Figure 7) as to the entry of a solid hydrophobic particle. The main
difference is that a liquid droplet deforms on entry to form a lens, which may subsequently
spread along the surface. We remark however that it is possible for material from a surface
active solid particle to spread as a monolayer on the surface of a solution.
Bergeron et al. [21] and Lobo and Wasan [22] have reported experimental
investigations relating to the effects which a metastable aqueous film between an oil drop
and the surface of an aqueous surfactant solution (a so-called "pseudoemulsion" film - see
Figure 7b) can have on the entry of the droplet. It was shown that, even though entry can
be thermodynamically feasible (see below), it may be that an oil drop is prevented from
entering the surface of a surfactant solution by a metastable pseudoemulsion film.
32
Bergeron et al. measured disjoining pressure isotherms for pseudoemulsion films in several
systems and showed that the films could be sufficiently robust to prevent drop entry in
practical situations. We will show below how the "classical" thermodynamic treatment of
drop entry in terms of interfacial tensions has been modified [21] by taking account of the
excess film tension of a metastable film arising from the surface forces.
metastable
pseudoemulsion film

Ot (a) (b)

Figure 7. (a) Approach of an oil drop from water to the air/water interface and (b) the formation of a
metastable pseudoemulsion oil/water/air film between drop and surface. The pseudoemulsion film tension is
approximately equal to the sum of the oiVwater and air/water tensions. The ratio of the radii of curvature of
the upper and lower caps of the lens is equal to the ratio of the metastable film tension and the oil/water
interfacial tension .

A great deal can be learned, at least in principle, concerning drop entry, lens shape
and the subsequent fate of the lens from a knowledge of the various interfacial tensions in
the system. We denote the interfacial tensions of the solution/air (or vapour) (wv),
oil/solution (ow) and oil/vapour (ov) interfaces by Ywv• Yow and Yovrespectively. It is clear
that it is not thermodynamically feasible for an oil drop to enter the wv interface if the
aqueous phase spreads spontaneously on the oil (to give a thick or "duplex" film in which
the 2 surfaces do not mutually interact). A spreading coefficient for water on oil, Sw.ov
can be defined as

(9)

Broadly, an aqueous solution will spread on oil if Sw.ov is positive or zero;


spreading will not occur if the coefficient is negative. The quantity Eo.wv is termed the
entry coefficient (for oil into the wv interface). Drop entry is feasible (aqueous solution
does not spread on the oil) if Eo.wv is positive, and not feasible if Eo.wv is negative. A
major problem in understanding the effects of oil on foam stability is that it is often
unknown if the system is at adsorption equilibrium or not. Hydrocarbon oils, often
present in antifoams are capable of mixing with (adsorbing into) surfactant monolayers
giving mixed films. Since adsorption always lowers interfacial tension, the values of the
entry and spreading coefficients will be affected by such adsorption. It turns out that for
systems at adsorption equilibrium (and assuming that Ywv is the largest of the 3 tensions)
the relationship between the various tensions is [ 12]

Ywv $ Yov + Yow (10)

so that in systems at adsorption equilibrium the maximum value of the spreading


coefficient is zero (corresponding to spreading); otherwise the spreading coefficient is
33
negative and macroscopic spreading does not occur. The entry coefficient is either zero
(non-entering drop) or positive (entering drop).
Bergeron et al. [21] have defined a "generalised" entry coefficient, Eg wv, which
takes account of thin film forces when a metastable pseudoemulsion film is formed. In
this case the lens exhibits the thin film to the vapour phase rather than an oil/vapour
interface (see Figure 7b). Correspondingly, in place of Eo,wv defined in equation 9 we have
for E3,wv

E3,wv = Ywv+Yow-Yf (11)

Here, y1 is the film tension and it replaces Yov in the expression for Eo.wv· The film
tension is the sum of the macroscopic tensions Ywv and Yow plus an excess film tension,
r; which arises from interactions across the pseudoemulsion film. It follows therefore
that the generalised entry coefficient is equal to the negative of the excess film tension. In
terms of the disjoining pressure (II) isotherm

g - e- rii(h)
Eo.wv--Yf --J 0 hdll (12)

where h is the film thickness. The generalised coefficient can be either negative or
positive, depending on the shape of the disjoining pressure isotherm. Values measured to
date [21,22] are small, ofthe order of 10·2 mN m·'. It is possible to have a large positive
classical entry coefficient (so that entry is thermodynamically feasible) and a very small
negative generalised entry coefficient such that entry is prevented.
If a drop does enter the wv interface it may or may not spread as a duplex film. A
spreading coefficient for oil on water, So.wv• can be defined in an analogous way to Sw.ov
by interchanging the ov and wv tensions. Again, in systems at adsorption equilibrium
macroscopic spreading is indicated if So,wv is zero, otherwise (So,wv is negative) the oil does
not spread as a macroscopic layer. As mentioned above however, the oil can spread
molecularly and mix with the surfactant monolayer, or it can form a very thin film (a few
nanometres thick) in equilibrium with the remaining lens [23,24]. Recently Bergeron and
Langevin [25] have treated the kinetics of radial spreading from a drop on a liquid surface.
The mixing of oil molecules with a surfactant monolayer is a more rapid process than
bulk spreading. A spreading monolayer drags along underlying liquid and when this
happens on a film surface the film is thinned, which can lead to film rupture.
The shape of a non-spreading lens resting in the surface of an aqueous solution is
determined by the relative magnitudes of the interfacial tensions. In the treatment of foam
breaking by oils the interest is in small lenses (ca. lflm 3) so that the lens consists of 2
spherical caps and the interface in which it sits is not distorted from planar. Since the
effects of gravity are negligible, the pressure is the same throughout the interior of the
lens and so (noting equation 1) the ratio of the radii of the caps is equal to the ratio of the
respective tensions. For a system in which the surface tension of an aqueous surfactant
solution close to its erne is around 30 mN m·', the surface tension of a hydrocarbon oil is
around 23 mN m·' and the oil/water interfacial tension is 5 mN m·', an oil lens would be a
34

shape similar to that illustrated in Figure 8a. The angle a (Figure 8b) is related to the
interfacial tensions by

(13)

and the depth of immersion, ha, of the lens in the subphase is given by

ha = r 1 (1-cosa) I sin a (14)

where r1 is the lens radius. The quantity ha is relevant to the bridging of thin films by oil
drops and to film rupture [26].

..
2r1
air IIIII air

water water

(a) (b)

Figure 8. (a) Typical shape of a small oil lens, radius r 1 resting on an aqueous surfactant solution at or above its
erne. (b) The angle a and the depth of immersion, hw are determined by the relative values of the three
c
interfacial tensions (see text). The angle is referred to in the discussion of the stability of liquid bridges
across liquid films.

4. Bridging of liquid films by particles and droplets

Imagine a particle or lens resting in a liquid surface in a foam or emulsion; sooner or later
another interface will approach and engage the particle or lens, which then forms a bridge
across a liquid film. Depending on particle or lens wettability by the film material, film
stability can be reduced or enhanced, and we consider several kinds of system below.

4.1. SPHERICAL SOLID PARTICLES IN FOAM AND EMULSION FILMS

In Figure 9 we illustrate the effect of both a hydrophobic and a hydrophilic particle on the
stability of an aqueous foam film. Initially the particle rests in one of the lamella
interfaces. A hydrophobic particle will be more out of the film than immersed (8
measured into the aqueous phase > 90" as shown in Figure 9a). The film will thin by
gravity and capillary suction into the adjoining Plateau borders. Ultimately the second
(lower as shown) surface will engage the particle and de-wet the surface, during which
curvature will be produced in the film surface as shown. The Laplace pressure generated
forces the liquid away from the particle (shown by arrows in the Figure), causing thinning
and rupture. It should be borne in mind that the curvature of the meniscus around the
particle in the lower surface of the film has a component of opposite sign, normal to the
plane of the paper. For a very small particle this second curvature could be greater than
35
that shown in the plane of the paper, and if so the Laplace pressure would draw film liquid
towards the particle and rupture would not occur.

film+ thins film+thins

(a) (b)

Figure 9. (a) Hydrophobic particle in aqueous film. When the bridge is formed as the film thins, the Laplace
pressure generated in the curved meniscus which forms forces liquid away from the particle and causes
rupture. (b) For a hydrophilic particle a stable bridging configuration is possible (upper diagram). As the film
drains curvature is generated which draws liquid towards the particle, which thus opposes film thinning.

For a hydrophilic particle in a film (Figure 9b) the equilibrium contact angles
(less than 90°) are achieved with one surface of the lamella on each side of the particle
equator. Subsequent film thinning will give curvature of the surfaces as shown in the
lower diagram in Figure 9b, and the Laplace pressure in this case will tend to counter film
thinning. Hence the particles do not rupture the films.
Exactly similar arguments apply to a small isolated particle in the surface of a
(larger) emulsion drop if it forms a bridge between two drops on collision (Figure 10). In
this case a potential exists for the particle to facilitate droplet coalescence when the particle
is less wetted by the continuous phase than the dispersed phase of the emulsion. Thus for
example, a hydrophobic particle should facilitate coalescence of oil drops in an o/w
emulsion and a hydrophilic particle is expected to aid coalescence of water droplets in a

?( R 9( (a) (b)

Figure I 0. (a) Solid particle in the surface of a non-wetting drop can form a stable bridge between two drops
in an emulsion. (b) The particle preferentially wets the droplet phase in the emulsion. In this case the particle
cannot stabilise an emulsion film because the drop surfaces tend to cross over, which causes film rupture and
aids drop coalescence.

w/o emulsion. Equally, reverse combinations (e.g. hydrophobic particles in a w/o


emulsion) can aid adherence of droplets whilst tending to prevent droplet coalescence.
36

4.2. LIQUID DROPLETS IN FOAM FILMS

Liquid lenses, unlike solid particles, deform upon forming a bridge over a liquid film;
otherwise they act in a similar way to solid particles. That is, a hydrophobic oil drop will
rupture an aqueous foam film. In the case of a lens the relevant contact angle is that, /5,

(a) (b)

Figure II. (a) A "hydrophobic" bridge across an aqueous film (o > 90°); the Laplace pressure forces aqueous
liquid away from the oil bridge causing thinning and rupture. (b) A "hydrophilic" oil bridge (o < 90°). The
Laplace pressure tends to draw water in the film towards the bridge, thus slowing fllm drainage. The arrows
denote the direction of flow induced by the Laplace pressure.

between the oiVwater and air/water interfaces (Figure 8b) rather than 8. The starting point
is a lens such as that illustrated in Figure 8. When, as a result of film thinning, the
second lamella surface meets the immersed part of the lens a bridge is formed which,
viewed from a static point of view (but see below) can be mechanically stable or not. The
stability is determined by the relative values of the three interfacial tensions and is
manifested in the angle /5. If 8 > 90° the bridge has the general shape given in Figure 11a;
the curvature of the oiVwater interface is convex towards the film and the Laplace pressure
forces liquid away from the droplet, causing film rupture much like the effect of a
hydrophobic solid particle on the film. For 8 < 90° the oiVwater interface in the bridge is
of opposite sign (Figure 11 b) and tends to pull liquid toward the liquid bridge. In the case
of a "stable" liquid bridge it may be that the initial readjustment of shape as the second
surface meets with the lens, will cause film rupture even though an existing bridge would
be mechanically stable. We note that the dimension which determines the film thickness
when the bridge forms is not the original oil drop radius but the depth of immersion ha
given in equation 14.

4.3. PARTICLES AND DROPS IN COMBINATION IN FOAM FILMS

In foam breaking formulations solid particles are often used in combination with mineral
or silicone oil, and synergistic effects between oil and particles are obtained. The relevant
configuration in this case is a solid particle bridging an asymmetric aqueous film between
the oil and vapour phases, as shown in Figure 12. It is seen that there are now two
relevant contact angles, that between the oil/water interface and the particle ( 80 w) and the
angle 8wvbetween the air/water interface and the particle. The stability of the asymmetrical
oil/water/air film depends on the sum of the two angles. If the sum is less than 180° the
two contact lines will not meet and the particle will tend to stabilise the aqueous film and
prevent drop entry into the interface. However, dynamic effects accompanying the
dewetting of the solid particle as it meets the air/water surface could cause film rupture
(assuming the entry coefficient for the oil is positive). Again assuming that oil entry into
37

the air/water surface is thermodynamically feasible, the aqueous film will not be stable for
a sum of contact angles exceeding 180° which means that the particle will accelerate drop
entry into the surface.

oil drop

Figure 12. A spherical solid particle bridging an aqueous film between an oil drop and the vapour phase. If the
sum of Bow and Ow, is less than 180° (as shown) the two contact lines will not meet. When the sum of the angles
exceeds 180° however the contact line would need to cross to attain the equilibrium angles, so that film rupture
would occur.

5. Effects of particles and oil drops on foam stability

There is a long history of research into foam breaking but it has to be admitted that much
of the evidence for the mechanisms proposed is somewhat circumstantial. There has been
a renaissance of interest in foam breaking recently, stimulated in part by the use of foams
as flow control agents in oil recovery. In this case the requirement is for the foams to be
stable in the presence of oil. However, unwanted foaming occurs in many industrial
processes and then the foams have to be broken. Formulations which counter foaming are
therefore of considerable practical importance. Antifoamers (which reduce the formation of
foams) and defoamers (which break down already formed foams) for use on aqueous foams
are often based on dispersions of small hydrophobic solid particles in a hydrocarbon or
silicone oil.
We have described above the basic phenomena which are believed to be involved
in film and foam breaking by oils and solid particles. It is often the case however that
experimental studies have been carried out on rather ill-defined systems. Surfactants and
oils have, understandably, usually been commercial samples of unknown purity and solids
have often been of irregular shape and unknown wettability. This means, for example,
that relevant surface and interfacial tensions and contact angles have often been unavailable
to test proposed mechanisms rigorously. In this section we review very briefly some of
the work on model systems. The interested reader will wish to consult the valuable
reviews of Garrett [2] and Pugh [27] where a much broader coverage is given.
Altho~gh there is a clear dependence of foamability (the ability of a solution to
form a foam) and foam stability (after formation) on the wettability of solid particles, the
precise nature of the dependence has not been widely tested. The use of spherical particles
of known wettability has however tended to confirm the bridging-dewetting mechanism of
foam and film breakdown described earlier. Some of the most impressive evidence is that
of Dippenaar [28] who obtained high speed cinematographic images of both spherical and
cubic (galena) particles in thin films. These results are particularly important since they
relate directly to the dynamic process of film rupture.
38
It has been observed by Tang et al. [29] and by Kumagai et al. [30] that spherical
particles can actually stabilise foams, although in their work contact angles were not
reoorted. It should be noted that although the contact angle of a hvdroohobic surface with

foams stabilised
.Q
"!!!
CD 0
""j

·1
40 50 60 70 80 90 100 110
contact angle/degrees

Figure 13. Half-life ratios of AOT (anionic surfactant) foams as a function of contact angle of AOT solution
with hydrophobised glass Ballotini beads. To obtain the range of contact angles a range of surfactant
concentrations and of solid wettabilities were used. The half-life ratio is defined as (foam half-life with
particles present less that without particles)/(half-life without particles)) where foam half-life is the time for
the foam to decay to half its initial volume. Positive values of the ratio reflect foam stabilisation by particles
and negative values destabilisation.

water exceeds 90°, with a surfactant solution it can be substantially lower than 90°.
Aveyard et al [31], using spherical particles showed that the foams used were maximally
stable for a "static advancing" contact angle of 90°. Drastic foam breakdown occurred for
contact angles in excess of about 92° (Figure 13). The stabilising effect was attributed to
the collection of particles in the Plateau borders and the concomitant reduction in film
drainage and thinning. Johansson and Pugh [32] also observed foam stabilisation by solid
particles using angular shaped quartz particles. In this case, presumably as a result of the
non-spherical particle geometry, maximum stability was observed for a contact angle
around 65°.
To explain the effect of dispersed oil droplets on the stability of aqueous foams it
is necessary to have a knowledge of the various relevant interfacial tensions. In recent
years, detailed studies have been made of the 2-dimensional solubilisation of hydrocarbons
in surfactant monolayers (see e.g. ref. 33). It is clear in a number of reports purporting to
give equilibrium values of spreading and entry coefficients, that equilibrium (with respect
to adsorption) had not been achieved. We recall that for systems exhibiting equilibrium
tensions, Eo.wv ;;::: 0 and So.wv :5: 0.
Aveyard et al. [34] and Bergeron et al. [35] have, respectively, studied the effects
of dispersed alkanes and dispersed silicone oils on aqueous foams. In the work of Aveyard
et al. the influence of a series of normal alkanes on aqueous foams formed from the pure
anionic surfactant Aerosol OT (AOT, diethylhexyl sodium sulphosuccinate) was
investigated. Entry coefficients in equilibrated systems were determined and correlated with
observations on the entry of small alkane droplets from aqueous surfactant into the
solution-(alkane-saturated) vapour interface. For those oils with positive coefficients,
entry was observed to occur whereas for systems exhibiting zero coefficients, entry was
not observed. It should be said however that the positive entry coefficients were all close
39
to zero and it would not be possible to conclude from these coefficients alone that drop
entry is thermodynamically feasible. No evidence was found in the systems studied for the
existence of robust pseudoemulsion films which could prevent entry where it is feasible.
A good correlation was found between the behaviour with respect to drop entry and the
effects of dispersed alkanes on the foam half-life, as seen in Figure 14. The lower alkanes
(hexane to undecane) are entering oils and reduce foam stability. The higher, non-entering
1500 100
(a) --·····-··-~--- (b)
all> 3600 s
80
~ ·§
5:1000
] 3!
;.= 60
..!.
! 01

.
.c 40
E
~ 500
.8
·~
20

0 0
5 10 15 20 25 30 35 5 10 15 20 25 30 35
alkane chain length alkane chain length

Figure 14. (a) Half-life of single alkane drops placed under the interface between air, saturated with alkane
vapour, and 3.8 mM AOT in 0.03 M aqueous NaCI. Drops of dodecane and higher alkanes did not enter the
surface within an hour. The alkane with "chain length" 30 is squalane. (b) Foam half-life in the same
systems. Alkanes up to decane destabilise foams whereas undecane and higher alkanes stabilise the foams.

oils were found to stabilise films and foams. Close observation revealed that oil droplets
collected in the Plateau borders, presumably reducing the rate of film drainage (thinning)
and hence rupture.
Oil drops containing dispersed hydrophobic particles are used as commercial
defoamers/antifoams since oil and particles act synergistically [2,36,37]. In a recent study,
Koczo et al. [36] proposed that dispersed oil drops (containing dispersed solid particles)
collect in Plateau borders of foams. The particles are supposed to penetrate the
pseudoemulsion films between drops and lamella surfaces, thus facilitating oil drop entry
into the surfaces. Such a mechanism is very similar to that proposed by other workers
[2,38,39] who supposed that the effects take place within the films themselves.
We have seen earlier that when oil and solid particles are present in a foam
together, there are 2 contact angles of importance, ()wv between air/water surface and
particle and ()ow between the particle and oil/water interface (see Figure 12). In a static
system the 2 fluid interfaces will be held apart if the sum of the angles is less than 180°.
Otherwise the 2 contact lines around a bridging particle are expected to meet leading to
film rupture and drop entry. Enhancement of the rate of entry presumably involves
curvature of the air/water meniscus as the solid particle enters the surface, somewhat as in
the case of entry of solid particles alone (Figure 9a), so the Laplace pressure generated
forces liquid in the pseudoemulsion film from the vicinity of the particle. We note that
this process can occur for ()wv < 90° since all that is required for film rupture is that the
40
sum of Owv and (}ow exceeds 180", and it turns out that (}ow is usually high, in the vicinity
of 140" [37].
In an attempt to explore oil/particle synergy further, a study has been made of the
effects of dispersed spherical hydrophobic particles (311m diameter) on the entry of single
oil drops from aqueous surfactant solution into the interface with air saturated with oil
vapour [16,40]. From a consideration of contact angles in static systems it might be
expected that if systems can be devised in which (}wv + (}ow < 180", particles bridging the
pseudemulsion films (illustrated in Figure 9) should stabilise the films and prevent entry
of the oil drop. In the context this would constitute oil/particle antagonism. In fact, it
was found that although drop entry can be prevented by the particles, the sum of the
contact angles required for this was as low as 50 or 60". This corresponds to
pseudemulsion film thicknesses of about 80% of the particle diameter; thinner films than
this are unstable. Such behaviour presumably reflects the importance of dynamic effects
in film rupture. It was observed in those systems where drop entry is prevented by the
particles that foams containing oil together with particles were more stable than those
containing just dispersed oil droplets i.e. oil/particle antagonism occurs in foam breaking.

6. Effects of solid particles on emulsion stability

It has been known for many years that solid colloidal particles can stabilise emulsions and
that the particles must be substantially smaller than the emulsions drops. The particles
act somewhat like surfactants in this respect. It is necessary for stabilisation that the
particles "adsorb" at the drop surfaces giving complete coverage, and the emulsions so
formed are termed Pickering emulsions. We have discussed earlier the energetics of
particle attachment to fluid interfaces (see Figure 5 and equation 6). It is found that
hydrophilic particles ((}ow< 90") stabilise oil-in-water (o/w) emulsions whereas water-in-
oil (w/o) emulsions are favoured by hydrophobic particles. Stabilisation results from the
protective film produced by the particles. Drop coalescence requires that particles be
removed from the surfaces of adjacent drops. For the favoured emulsion type (i.e. that in
which the particles are more wetted by the continuous than the dispersed phase) there is a
large energy barrier to displacement of particles into the dispersed phase (where they would
tend to go when 2 drops collide), and such large barriers (see Figure 5b) are unlikely to be
surmounted in practice [41,42]. It therefore appears likely that particle displacement prior
to coalescence must occur laterally and so the interfacial rheology of the particle-covered
drop surfaces is an important factor in emulsion stability. Tambe and Sharma [42] have
recently reviewed aspects of the behaviour of Pickering emulsions.
The emulsion analogues of the foaming systems containing particles which we
have already discussed are somewhat different to Pickering emulsions since the extent of
surface coverage by particles is much lower (say 10%). Further, the emulsions are now
stabilised by surfactant and the particles only modify the stability. There appears to be
little methodical work on these systems and we merely outline some interesting
possibilities below and refer to some preliminary experimental findings.
Reference has been made earlier to isolated particles in the surfaces of emulsion
drops and the situation is represented in Figure 10. For particles which preferentially wet
41

the dispersed phase, the bridging of 2 drops is expected to facilitate coalescence. If the
particle preferentially wets the continuous phase however, a stable bridge between droplets
can form and in principle this can lead to the formation of floes or possibly larger
networks.
After formation of an emulsion, the droplets can cream (rise) or sediment (fall)
under gravity. Creaming or sedimentation is accelerated by the formation of floes, but it
would be inhibited by the formation of a network of droplets. Drops in close proximity in
a creamed or sedimented layer are more likely to coalesce than are well-dispersed droplets.
It is clear then that the presence of a relatively low concentration of particles in an
emulsion can influence both sedimentation/creaming and coalescence.
Suppose a w/o emulsion is formed, containing surfactant and some hydrophobic
particles (sufficient say to give 10% coverage of droplets initially), in which the oil is less
(c)

~
,., (b) ..... n~
(d)
bd (e)

Figure 15. Appearance of a water-in-oil emulsion undergoing sedimentation and coalescence. Equal volumes
(say) of oil and water (a) before homogenisation to give an emulsion (b). Sedimentation of water drops results
in the formation of a clear oil layer (c), and coalescence ultimately gives a clear water layer (d).
Sedimentation and coalescence can occur simultaneously giving clear oil and water layers with an intervening
layer of unresolved emulsion (e). Shaded areas represent unresolved emulsion.

dense than the aqueous phase. Droplets will sediment, leaving a clear upper oil layer. As
coalescence proceeds, a clear water layer will begin to form at the bottom. The remaining
emulsion lies in the central portion of the container as illustrated in Figure 15. The oil to
water ratio in the emulsion layer can be calculated from the positions of the interfaces
between the emulsion and the clear oil and water layers.
We have made a preliminary study of the effects of monodisperse hydrophobic
polystyrene latex particles (diameter= 0.4821J.m) on the sedimentation and coalescence in
w/o emulsions, formed from heptane and 0.1 M aqueous NaCl and stabilised by AOT [43].
We show the positions of the upper and lower boundaries (corresponding to sedimentation
and coalescence respectively- see Figure 15) as a function of time in Figure 16. It is clear
that the particles in the early stages of resolution of the emulsion reduce the rate of droplet
sedimentation and subsequently enhance sedimentation slightly. In the time range where
sedimentation is retarded, no clear aqueous layer is observed indicating that coalescence is
much reduced. (Of course, droplet size can increase without a clear layer of droplet phase
being observed). When however sedimentation is enhanced by the presence of the
particles, the rate of formation of the clear aqueous layer becomes large.
42
Although the evidence is circumstantial, we can hypothesise as follows. When
the particles (which are hydrophobic) are mixed with the emulsion, they bridge water
droplets and initially form a weak droplet network. This inhibits both sedimentation and
coalescence. In time however the network breaks up under the influence of gravity to give
floes of droplets. Floes are expected to sediment more rapidly than isolated droplets and
this accounts for the enhanced sedimentation at longer times. By the time coalescence is
evident (ca. 200 min) the volume fraction of oil in the remaining emulsion layer is low.
It appears that the particles remain in the emulsion layer (the oil and water layers are
optically clear) and are thus concentrated as emulsion resolution proceeds. The enhanced
rate of coalescence could be due to dynamic effects as particles become attached to oil-water
interfaces in what might at this stage be a polyhedral structure (i.e. a biliquid foam). The
energy required to distort droplets into polyhedral shapes could be provided by the energy
of attachment of the particles to the oil/water interfaces.

1.0
CLEAR
OIL
0.8

~
-g 0.6
::s
0
.0
.....0
c: 0.4
0
·~
·o;;
0
0..
0.2

0.0
0 100 200 300 400 500
time/min

Figure 16. Resolution of w/o emulsions formed from 3.8 mM AOT in 0.1 M NaCl and heptane (equal
volumes). The curves represent the position in the cylindrical tube of the interfaces between unresolved
emulsion and (upper curves) clear oil layer and (lower curves) clear aqueous layer. The upper curves
represent sedimentation of water droplets and the lower curves the coalescence of water drops as explained in
the text. Open symbols are for systems without particles and filled symbols are for systems containing
monodisperse hydrophobic polystyrene latex particles (diameter= 0.482 Jlm), sufficient to give 10% coverage
of droplets in the original emulsion.

Summary

Whilst there are some important differences between the structure and behaviour of foams
and emulsions, there are also a number of common features. The kinetic stability is
closely related in both cases to the drainage of thin symmetrical liquid films between gas
cells or liquid droplets. When the films are thin (say 100 nm or less) the same kinds of
surface forces act across liquid/ liquid/ liquid films in emulsions and gas/ liquid/gas films
in foams. The stability of thin liquid films can be drastically affected by the presence of
43

solid particles or insoluble liquid droplets of dimension of the order of the lamellar
thickness. Before the particle or droplet contained within a film can cause rupture
however, it must first enter at least one of the film surfaces. Thermodynamic feasibility
does not ensure entry since the pseudoemulsion film between immersed particle or droplet
and the lamella surface may be metastable. After entry, film rupture can be effected by
either spreading (macroscopic or molecular) along the lamella surface or by bridging of the
film and subsequent dewetting of the bridging particle/droplet. From this it follows that
the effects which liquid drops and solid particles have on foam stability are intimately
associated with wettabilities of lamella surfaces, liquid droplets and solid particles.
Wetting is of course an area of great interest, to physicists, chemists and others, which
stretches far beyond the foam and emulsion systems discussed here.

References

1. Dickinson, E. (1992) An Introduction to Food Colloids, Oxford University Press,


Oxford. Chapters 4 and 5.
2. Garrett, P.R. (1993) in P.R. Garrett (ed.), Defoaming: Theory and Industrial
Applications, Marcel Dekker, New York, Chapter 1.
3. Walstra, P. (1996) in Becher, P. (ed), Encyclopedia of Emulsion Technology,
Volume 4, Marcel Dekker, New York, Chapter 1.
4. Shaw, D. J. (1991) Introduction to Colloid and Surface Chemistry, Butterworth-
Heinemann, Oxford. Fourth Edition.
5. Israelachvili, J. (1992) Intermolecular and Surface Forces, Academic Press,
London. Second Edition.
6. Ross, S. and Morrison, I. D. (1988) Colloidal Systems and Interfaces, Wiley-
Interscience, New York. p 294 et seq.
7. Vrij, A. and Overbeek, J. Th. (1968) J. Amer. Chern. Soc. 90, 3074.
8. Aveyard, R., Binks, B. P., Fletcher P. D. 1., Ye, X. and Lu, J. R. (1992) in
Sjoblom, J. (ed) Emulsions- A Fundamental and Proactical Approach, Kluwer,
The Netherlands.
9. Binks, B. P. (1996) Annual Reports of the Royal Society of Chemistry Sect.C
92, 97.
10. Kabalnov. A. and Wennerstrom, H. (1996) Langmuir 12, 276.
11. Aveyard, R. and Clint, J. H. (1996) JCS Faraday Trans. 92, 85.
12. Rowlinson, J. S. and Widom, B. (1982) Molecular Theory of Capillarity, Oxford
University Press, Oxford.
13. Duncan, D., Li, D., Gaydos, J. and Neumann, A. W. (1995) J. Colloid Interface
Sci. 169, 256.
14. Aveyard, R., Beake, B. D., and Clint, J. H. (1996) JCS Faraday Trans.
92, 4271.
15. Mingins, J. and Scheludko, A. (1979) JCS Faraday Trans. 75, 1.
16. Aveyard, R. and Clint, J. H. (1995) JCS Faraday Trans. 91, 2681.
17. Clint, J. H. and Taylor, S. E. (1992) Colloids Surf 65, 61.
44

18. Aveyard, R., Binks, B. P., Fletcher P. D. 1., and C. E. Rutherford


(1994) Colloids Surf. A 83, 89.
19. Clint, J. H. and Quirke, N. (1993) Colloids Surf A 78, 277.
20. Aveyard, R. and Clint, J. H. (1995) JCS Faraday Trans. 91, 175.
21. Bergeron, V., Fagan M. E., and Radke, C. J. (1993) Langmuir9, 1704.
22. Lobo, L. and Wasan, D. T. (1993) Langmuir 9, 1668.
23. Kellay, H., Meunier J, and Binks, B. P. (1992) Phys. Rev. Lett. 69 1220.
24. Kellay, H., Binks, B. P., Hendrikx, Y., Lee L. T., and Meunier, J. (1994)
Adv. Colloid Interface Sci 49, 85.
25. Bergeron, V. and Langevin, D. (1996) Phys. Rev. Lett. 76, 3152.
26. Aveyard, R. and Clint, J. H. (1997) JCS Faraday Trans. 93, 1397.
27. Pugh, R. J. (1996) Adv. Colloid Interface Sci. 64, 67.
28. Dippenaar, A. (1982) Int. J. Miner. Process. 9, 1.
29. Tang, F. Q., Xiao, Z., Tang J. A., and Liang, J. (1989) J. Colloid Interface Sci.
131, 498.
30. Kumagai, H., Torikata, Y., Yoshimura, H., Kato, M., and Yano, T. (1991)
Agric. Bioi. Chern. 55, 1823.
31. Aveyard, R., Binks, B. P., Fletcher, P. D. I. and Rutherford, C. E. (1994)
J. Dispersion Sci. Techno/. 15, 251.
32. Johansson, G. and Pugh, R. J. (1992) Int. J. Miner. Process. 34, 1.
33. Aveyard, R., Cooper, P., and Fletcher, P. D. I. (1990) JCS Faraday Trans.
86, 3623.
34. Aveyard, R., Binks, B. P., Fletcher, P. D. 1., Garrett P.R., and Peck, T. G.
(1993) JCS Faraday Trans. 89, 4313.
35. Bergeron, V., Cooper, P., Fischer, C., Giermanskakahn, J., Langevin, D. and
Pouchelon, A. (1997) Colloids Surf. A 122, 103.
36. Koczo, K., Koczone J. K., and Wasan, D. T. (1994) J. Colloid Interface Sci.
166, 225.
37. Aveyard, R., Cooper, P., Fletcher, P. D. 1., and Rutherford, C. E. (1993)
Langmuir 9, 604.
38. Frye, G. C. and Berg, J. C. (1989) J. Colloid Interface Sci. 127, 222;
130, 54.
39. Aronson, M.P. (1986) Langmuir 2, 653.
40. Aveyard, R., Binks, B. P., Fletcher P. D. 1., Peck, T. G. and Rutherford, C. E.
(1994) Adv. Colloid Interface Sci. 48, 93.
41. Tambe, D. E. and Sharma, M. M. (1993) J. Colloid Interface Sci. 157, 244.
42. Tambe, D. E. and Sharma, M. M. (1994) Adv. Colloid Interface Sci., 52, 1.
43. Aveyard, R., Beake, B. D., and Clint, J. H. unpublished work.
AN INTRODUCTION TO FORCES AND STRUCTURE IN INDIVIDUAL FOAM
AND EMULSION FILMS

VANCE BERGERON
Rhone Poulenc Industrialisation
85, Av. Des Freres Perret- BP 62, Saint-Fons Cedex, France

1. Introduction

The stability of a foam or an emulsion relies on the stability of the individual


films that separate the discontinuous phases, a subject central to surface and colloid
science. Moreover, many commercial processes and products rely on the fundamental
interfacial interactions that take place in the thin-film region. Thus, a great deal of effort
has been spent studying the dynamics and stability of individual thin-liquid films (1,2).
Although foams and emulsions are in an absolute sense thermodynamically unstable, it is
often found that a particular system can be categorized as a relatively short-lived
"dynamically" stabilized system (ca. minutes) or one that can remain stable for very long
periods (ca. days or years). Champagne foams are a classic example of the former, while
certain beer foams and cosmetic creams fall into the later category. This striking
difference in a dispersion' s lifetime reflects the primary mechanisms that govern the
individual film stabilities. In rapidly coalescing dispersions, the film lifetimes are
controlled by the drainage rate of the intervening continuous phase (hydrodynamics),
while the long-lived systems require additional time to overcome energy barriers that
hold the film in a metastable thermodynamic state. These barriers arise from surface-
force interactions (i.e. disjoining pressures) created by having two interfaces in close
45
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 45-72.
© 1999 Kluwer Academic Publishers.
46
proximity, see Figure 1.1. In fact, for some cases overcoming these barriers can take so
long that other factors such as Oswald ripening and gas diffusion determine the ultimate
lifetime of the dispersion. Clearly, understanding and controlling the energy barriers that
inhibit thin-film coalescence has great practical benefits for these dispersed systems. The
central theme of this work is to present an introduction to the concepts and intermolecular
forces that arise in surfactant-laden thin-liquid films and to show how they are quantified
and studied experimentally.

Dynamic Foam Static Foam


-30 sec > 30 min

• Film Thinning : Hydrodynamics • Disjoining Pressure : Thermodynamics

spinodal decompostion nucleation

Figure 1.1 Typically foams can be categorized as short-lived systems where film rupture is described as a
spinodal decomposition, or long-lived systems in which energy barriers create an activation energy and film
rupture is governed by a nucleation process.
47
2. Disjoining Pressure

2.1 Definition
Every interface has a thin interfacial region whose intensive thermodynamic
properties deviate from those of the two neighboring bulk phases. These transition
regions naturally develop from changes in the molecular interactions as we cross the
phase boundary. If two interfaces approach one another these changes manifest
themselves as macroscopic "surface forces". This situation will occur when two phases
approach each other while an intervening third phase separating them grows thinner (e.g.
foam film). When the thickness of the third phase becomes comparable to the thickness
of the interfacial regions there remains no portion of the interlayer (i.e. film) possessing
the properties of the initial third phase and further decreases in film thickness requires
work. This requirement originates from net repulsive or attractive macroscopic forces
generated by the overlapping interfacial regions. Therefore, in order to maintain a
constant film thickness after overlapping has occurred, an external force (positive or
negative) must be applied to the system. A crude but simple analogy to this process is
found by considering what happens when two magnets are brought together. As the
magnets approach and their fields start to overlap, an additional external pressure must be
applied in order to bring them closer (this can be a negative or positive pressure
depending on the interaction between the magnets). Similarly, as the intervening aqueous
solution in a thin-liquid film (e.g., foam, emulsion, etc. ) drains, the interfaces approach
one another and the phases separated by the solution interact. These interactions can be
quantified as an excess pressure versus the separation distance (i.e. the film thickness, h),
which by definition is a disjoining pressure isotherm. Note that the term disjoining is
somewhat misleading in that attractive forces produce a conjoining force. Nevertheless,
both repulsive and attractive forces are embodied in the disjoining-pressure concept.

Derjaguin originally formulated the concept of a disjoining pressure for thin-


liquid films and was the first to verify experimentally its existence (3). The more general
and strict definition of the disjoining pressure given by Derjaguin and Churaev (4) is, "In
48

mechanical equilibrium the disjoining pressure, ll(h), is equal to the difference existing
between the component Pzz of the pressure tensor in the interlayer and the pressure, PB,
set up in the bulk of the phase from which it has been formed by thinning out:

(1)

(where PN is the pressure normal to the surface of a thin plane-parallel interface and
directions are defined by the co-ordinate system pictured in Figure 2.1 ).

z z

h - -z=O--

Figure 2.1 Typical distribution of the pressure tensor components PN and PT in a thin-liquid film (after
Derjaguin and Churaev (3)).

In the simplest case of a one-component liquid phase, mechanical equilibrium under


isothermic conditions implies thermodynamic equilibrium. In that case the disjoining
pressure is a single-valued function of the interlayer thickness, h, ...". More recently,
Kralchevsky and Ivanov have extended Equation 1 and derive a general vectoral
expression for the disjoining pressure (5, 6),
49

(2)

where u 'is the three-dimensional idemfactor, e is the total pressure tensor, and n is
an outer unit normal to the reference surface. r 0 corresponds to a reference surface
dividing the film into two halves and PR is a reference pressure. The utility of Equation 2
is that it can be used for arbitrarily curved films without making model simplifications.

2.2 Thermodynamic Definition


An alternative definition for the disjoining pressure can be formulated in terms
of thermodynamic variables. In this case, the work required to change the thickness of a
film at constant temperature, T, overall pressure, P, area, A, and mole numbers, Ni, is
expressed by a change in the Gibbs free energy of the film (7,8),

ll(h) =- (aa)
ah T,P,A,N;
(3)

For symmetrical foam films Equation 3 can be used to generate the following familiar
form of the Gibbs-Duhem relation (8-11) for the film:

2da =-sfdT- lldh - 2I rf dJ.li ,


i
(4)

where 0' is the surface tension, sf is the excess entropy of the film interface, and fi and
Jli are the adsorption (i.e., surface excess concentration using the Gibbs convention) and
chemical potential of the ith component. At constant temperature and chemical potential
Equation 4 reduces to the following useful relationship,

,fda) = -n (5)
\dh T,Jli '

Integration of Equation 5 then yields an expression that relates the surface tension of the
film interfaces to the disjoining pressure isotherm,
50

2a(h) 2a(h=~) - f Ildh (6)

where O'(h=oo) is the bulk value of the surface tension. In terms of the membrane model
the equivalent expression is (10,11),

2a(h=oo) + i ii( h)

JI(h=oo)
hdJI (7)

where O'f is the overall tension of the film and is used when all of the film properties are
ascribed to a single two-dimensional plane. The later two equations can be used to
describe the effect disjoining forces have on film and three phase contact angles (10).
Thus these equations are the starting point towards developing bulk foam and emulsion
constitutive equations which incorporate thin-film forces.

The thermodynamic and mechanical approaches to the definition of the


disjoining pressure can be combined by utilizing the Bakker equation for the definition of
surface tension at each interface of a symmetrical film (12),

(8)
51

where PT = Pxx = Pyy is the tangential component of the pressure tensor. The limits of
integration are more easily understood by referring to Figure 2.1. The upper limit, o/2,
represents the normal distance from the film interface far enough away from the
interfacial zone to guarantee an isotropic bulk pressure ( i.e., PN = PT = PB). The lower
limit at z= 0, corresponds to the center of the film. Notice that as the film thickness
approaches infinity, o(h) -> o(h = oo) in Equation 8, and we recover the bulk surface
tension expression. Substitution of Equation 8 into Equation 5 relates the thermodynamic
description of the disjoining pressure to its mechanical origins (anisotropy of the pressure
tensor within the film),

II= (9)

Equation 9 can be used to relate the disjoining pressure to intermolecular potentials


through standard statistical mechanic expressions for PN and PT when care is taken to
correctly define the limits of integration (12). In a similar context Rusanov and Kuni
(13,14) have applied the statistical mechanic approach to calculating the distribution
functions and the pressure tensor for films of simple liquids.

An example of how the pressure distributions PN and PT change in the thin film
region is provided in Figure 2.1. The length of the horizontal arrows in Figure 2.1
represent the magnitude of the pressure components while the direction (right or left)
signifies the sign (positive or negative). For plane parallel films in equilibrium the normal
component of the pressure tensor can not change and remains constant through the film.
Conversely, the tangential component can change in both sign and magnitude. However,
beyond the transition zone, defined by 0 in Figure 2.1, PN = PT, and the pressure is
isotropic and equal to the bulk pressure of the contiguous phases. A more detailed picture
for a soap film stabilized by ionic surfactant is given by Eriksson and Toshev (8).
52

3. Disjoining Pressure Models

3.1 General Approach


Changes in the interfacial region that generate the disjoining pressure in a thin
liquid film originate from intermolecular forces. It is customary to separate the various
contributions of the disjoining pressure into different components, e.g.,

n (h) = lldz + llvan + llsteric + llsupra +etc ... (10)

where the subscripts in Equation 10 indicate the following contributions: dl =


electrostatic double layer forces, van =London-van der Waals dispersion forces, st =
steric and short range structural forces (e.g., entropic confinement forces), and supra=
forces arising from supramolecular structuring. Of course, models based on the
application of Equation 10 make the key assumption that the various contributions to the
disjoining pressure are additive. However, it is not always clear that this assumption is
valid and in some cases it may lead to anomalous results (15,16).

Combination of the first two components listed in Equation 10, lldl and llvan.
constitute the well-known DLVO theory. These two basic contributions are used
throughout colloid science to describe particle interactions. Typically they are treated
separately and additive as suggested by Equation 10, but , Attard et.al. (15,16) have
recently extended classical Poisson-Boltzmann theory and show that the distinction made
between van der Waals and double layer forces is somewhat of an illusion.

In what follows we briefly review each of the components in Equation 10


separately for the purpose of highlighting the molecular origins of the disjoining pressure
isotherm. For a comprehensive review several texts (17), and monographs (7,18-21) are
available.
53

3.2 Electrostatic Double-Layer Forces


One of the first and most studied contributions to the disjoining pressure arises
from "electrostatic" interactions. These interactions result from overlapping of the electric
double layers that develop at charged interfaces and a repulsive force can develop due to
entropic confinement of the counter ions. As pictured in Figure 3.1, when the separation
distance between two charged interfaces approaches twice the characteristic length for
decay of the diffuse ionic atmospheres, A., an additional external force is required to
maintain the separation distance. From classical Debye-Huckel theory the characteristic
length over which ions from a univalent electrolyte will act is given by (21),

A.= l/K"= . I ekT _ c-112 (11)


'V 8nn°e2

where A. is called the Debye length, no is the number density of ions, e is the elementary
charge, e is the dielectric constant of the medium, T is temperature, k is the Boltzmann
constant and C is the concentration of electrolyte (moles/L). Equation 11 provides the
important result that the decay length decreases as the electrolyte concentration increases
(i.e. interactions become shorter range because of ionic screening).

The electrostatic double layer forces are obtained by solving the Poisson-
Boltzmann equation under a variety of different boundary conditions. There exists an
extensive literature concerning the calculation of the electrostatic repulsion between
interfaces (17-26). Therefore, only two of the classic results will be given here as
illustrative examples.
k
54

.·.~Jl
~
A= 1
"'K

I~ h

Figure 3.1 Two charged plates with their accompanying ionic atmospheres will interact when the separation
distance, h, approaches twice the Debye length, A..

In most cases only relatively simple approximations for lldl are required. Such
approximations are typically valid for small surface charges where linearization of the
Poisson-Boltzmann equation is acceptable. Under these conditions and assuming
univalent electrolytes the constant surface potential and constant surface charge models
for lldl are given by,

Constant Surface Potential (7,22)

(12)
55

Constant Surface Charge (7,23,24)

rrft = 2na'i {1 + sech( ~<:hf2))2 (13)


e tanh ( ~<:h/2) '

where 'I'0 is the potential and cr 0 is the charge density at the interface. The constant
surface potential model can be further simplified for large separation distances and small
potentials to give the following well-known form (17),

Ildt = 64n°kTy2 exp(-K:h) (14)

r=(exp(Z/2)-1)' Z=e'lj/0 .
exp(Z/2) + 1 kT

Hunter (17) points out that under the conditions assumed, Equation 14 is also valid for
constant charge systems since little discharge occurs if the degree of double layer overlap
is small. More elaborate models for ndl include, charge regulation boundary conditions
at the surface (25,26) and effects due to ionic correlation and image forces (15,16).

3.3 Dispersion Forces


In addition to electrostatic double-layer forces London-van der Waals dispersion
forces have long been recognized as being important in thin-liquid films. The calculation
of these forces has been approached in two different ways, microscopically and
macroscopically.

The microscopic method, credited to Hamaker (27), came first and is based on
pairwise summation of the individual dispersion interaction between molecules. Casmir
and Polder (28) later supplemented this approach by including the correction for
56
electromagnetic retardation. The molecular interaction potential used is typically
represented by (29),

(15)
u (r)

where u(r) is the interaction potential between two spherically symmetric molecules, 1
and 2. his Planck's constant, Vi is a characteristic electronic frequency for each molecule
in its unexcited state, and <Xi is the polarizability of molecule i. In order to obtain the force
of interaction between two macroscopic bodies Equation 15 is integrated over the volume
of the system, which for two plane parallel surfaces separated by a vacuum gap yields
(17,18),

llvan
(16)

where A 12 is known as the Hamaker constant. At large separations retardation effects

(finite response time of the induced dipoles) can become important, and lead to a
decreased interaction that decays faster ( llvan - tfh4) (28). When the interaction
between two different bodies, 1 and 2, is mediated by a third phase, 3 (e.g., aqueous films
"3" sandwiched between phase "1" and phase "2") the potential energy of interaction,
n 132· becomes,

where (17)

and Vi corresponds to the volume of phase i and At32 is the composite Hamaker
constant. Like the electrostatic component many elaborate models have been developed
57
to handle different geometries and more complex systems (e.g. multilayered films) (17,
30-32).

The fundamental shortcoming of the microscopic approach stems from the


assumed pairwise additivity of the molecular interactions. However, this problem is
overcome if we adopt an alternative point of view and consider the interacting bodies as a
continuous media. This macroscopic approach was developed by Lifshitz (33,34) and the
theory now bears his name. The basic idea of the theory is that the interaction between
the bodies is considered to take place through a fluctuating electromagnetic field. In this
approach the fields are calculated on the basis of the exact Maxwell equations, so that the
effects of retardation, caused by finite propagation velocities of the electromagnetic
waves, are automatically taken into account. The Lifshitz result for a thin uniform film of
phase 3 between two semi-infinite phases, 1 and 2, is given by (34),

(18)

where

and E1, E2, and E3 are functions of imaginary frequency ro = il;. However, the quantity
E(i!;) is a real function that can be evaluated from,

(19)
58
E11(ro) is the imaginary part of the dielectric response function and WE (ro) measures the
11

spontaneous electric field fluctuations in a body as well as a substance's ability to


dissipate applied electrical energy (35).

Needless to say Equation 18 is a bit cumbersome and its original derivation is


rather lengthy. However, many subsequent treatments of the macroscopic theory are now
available which provide both a more readily understandable approach and many useful
approximate expressions (17,18,35). In fact, using the method of Parsegian and Ninham
(36), to determine the dielectric response function (Equation 19) from absorption data and
reflectance measurements, it is now quite easy to calculate dispersion forces from
Lifshitz's theory.

Although the macroscopic approach is a great improvement over the classical


Hamaker summations, it is not expected to hold when a film gets so thin that its dielectric
properties change with thickness or when molecular orientation is important. This may
be an important consideration for aqueous films that undergo structuring near the
interface.

3.4 Sterle (Entropic Forces)


Entropic confinement forces are a third class of forces thought to occur in
ultrathin surfactant films (< 5.0 nm). llsteric is introduced into Equation 10 to cover
contributions to the disjoining pressure that are responsible for the stability observed in
so-called Newton black soap films. In concept this component is similar to the II

adsorption component of disjoining pressure II introduced by Derjaguin (7), that arises


from the steric repulsion that occurs when adsorbed layers overlap. Recently,
Israelachvilli and Wennerstrom (37) have outlined the physical origin of these forces
more precisely and have carefully categorized the various modes by which they operate.
The general classifications given by them include:
59
Undulation - forces created by undulations of the interface (inversely proportional to the
bending modulus, Kb, llu - 1/Kb).

Peristaltic - forces generated by peristaltic fluctuations as two interfaces approach


(inversely proportional to the compressibility modulus, Ka, llp - 1/Ka).

Headgroup overlap - a steric stabilization force that becomes important in systems


containing large non-ionic headgroups. These forces can be described by theories used
for polymer "brushes".

Protrusion - molecular scale protrusions of surfactant molecules at the interface. llpro


can be approximated from (37),

n _ n/3 (hi/;) exp(-h/1;) (20)


pro- [ 1- (1 + hts)exp(-htl;)]

where ~= kT/~ is the protrusion decay length, n is the density of protrusion sites, and ~ is
an interaction parameter (J /m).

A semiquantitative treatment of these entropic confinement forces is presented by


Israelachvili and Wennerstrom in their review.

Also important in extremely thin films are solvation forces (18), or when water
is the solvent, hydration forces. These forces originate from molecular ordering at the
interface. When two interfaces approach this ordering is disturbed, resulting in forces of
attraction and repulsion. These short-range interactions can be very complex and depend
on how the molecules structure at an individual surface, then how this structure is
modified once a second surface is encountered. The simplest liquids display force curves
that oscillate with a periodicity equal to the liquids molecular diameter and can be
modeled by treating the molecules as hard-spheres between two hard walls (18). Water,
60
however, has strong dipoles that can lead to hydrogen bonding and long-range dipole
polarization. These effects generate repulsive "hydration" or attractive "hydrophobic"
interactions in addition to the short-range oscillatory interaction.

Figure 3.2 shows a schematic of a foam-film disjoining pressure isotherm which


includes the entropic force contributions, llsteric• superimposed on the classical DLVO
components, lldl and llvan· It is important to note that thermodynamically stable films
can exist only in negatively sloping regions of the isotherm. Hence, the portion of the
curve with a positive slope separates the isotherm into two stable regions, thick (- 50 nm)
common black films (CBF) and thinner (- 4 nm) Newton black films (NBF). In foam ,
CBF stability is normally due to the electric double layer forces. NBF stability is not as
well understood but can be accounted for by the short-range entropic confinement forces
outlined above.

\
\ lldl- exp(-Kh)
+ \.

''
'
-
II '
0~~~~------~~~----~~
.,..,....... h
llsteric /
/
/ llyan- 11 ~
/
I
t t
Newton black films common black films

Figure 3.2 Schematic representation of a disjoining pressure isotherm that


includes contributions from ndl, llvan and llsteric·
61

Recent x-ray reflectivity experiments (38) and molecular dynamic calculations


(39) confirm the schematic representation of the NBF pictured in Figure 3.3. In the figure
the filled circles represent water molecules in the first hydration shell that surround the
polar head groups of sodium dodecyl sulfate (SDS) surfactant molecules. The
hydrocarbon chain of the surfactant (i.e., surfactant tails) are indicated by the irregular
lines that extend out of the aqueous film. Approximate thicknesses have been indicated
and from the picture it is clear that very little space is left for unbound water and the
interior of the NBF film resembles a solid-like structure.

-13 A

A
A

.. 20A

Figure 3.3 Schematic diagram showing the profile of a Newton black film. Filled circles correspond to water in
the first hydration shell, surfactant headgroups are labeled and the hydrocarbon chain is depicted by the
irregular lines.
62

3.5 Supramolecular Forces


Relatively new types of forces, due to supramolecular structuring of
amphiphilic molecules within the film, have been recently measured in foam and
pseudoemulsion films for the first time by Bergeron et. al. (40-42). Richetti et. al. (43-45)
have also measured similar structural forces between surfactant coated crossed mica
cylinders in a surface force apparatus (SFA). The general form of the force curves
obtained for these systems is summarized in Figure 3.4.

-
"""'... ltll '
4
'
I I

~
Ill I I'
"\ ,
q I JLJ.,W
""-''' UJ I' I .UJU U I' I JJ,H£W
~

....rt""" ,, ,,,,,.,, """'..,...


... •

Structureless Film
(belowCMC)

81/ayers Micelles

h
h II
II

Structural Models
DLVOTheory

Figure 3.4 The general form of the force curves for systems with and without supramolecular structuring. Only
two types of structures are depicted, however many different types can occur depending on the system.
63
In systems studied thus far the forces can be extremely long range (> 50 nm) and
oscillatory, having a periodicity set by the effective size of the structures responsible for
the forces. One important difference found thus far between the SFA measurements and
those obtained for foam and pseudoemulsion films is the magnitude of the forces
involved. In foam and pseudoemulsion films the magnitudes are low, order 100 Pa, while
SFA measurements on similar systems exceed 104 Pa. This difference most likely comes
from physical differences between the interfaces in the two experiments. SFA
measurements confine a fluid between two solid interfaces, which support more stress
than fluid interfaces and promote a higher degree of supramolecular order. Conversely, a
fluid interface is flexible and can absorb energy through deformations (bending modes)
which will diffuse the ordering between the interfaces. Some effects of spatial and density
fluctuations on the measured forces and stability in foam and emulsions films has been
recently disscused (46).

Supramolecular forces develop when relatively high concentrations of surfactant


are present in a system. In such systems a variety of structures can form depending on the
nature of the amphiphile and composition in the system, ionic strength, and external
conditions such as temperature. Therefore modeling supramolecular forces typically
requires assumptions about the structures that are present in the film. The two most
common structures encountered in thin-liquid films are, repeating units of surfactant
bilayers (47-49) or ordered arrays of micellar domains (50,51).

Mysels was the first to suggest that micelles can contribute to the disjoining
forces in foam films (52,53). Since then attempts have been made to model this effect
(51,54). Initial models treated the problem as a series of infinite barriers (54) while the
oscillatory nature of these forces was first modeled by Pollard and Radke (51), who
utilize density functional theory (DFT) to calculate a micellar contribution to the
disjoining pressure, llmic· This method sums the force exerted on the interfaces by the
micelles in the film:
64

n.
mzc
=-1_1hp(x·h)[dUm-l(x)+
2 'dx
dUm-2<h-x)]dx -P
dx B· (21)
0

where x defines the location in the film, Um-i is the interaction potential between a
micelle and interface i, and PB is the reference bulk pressure of the fluid in equilibrium
with the .film. p( x ; h) is the number density distribution of micelles in the film which is
obtained from variational differentiation of the Helmholtz free energy, F, with respect to
p (x; h),

(22)
8F[p(x)] = _J.l + uext ,
8p(x)

where Jl is the chemical potential and uext is the external potential (i.e. a typical DLVO-
type potential). Calculations using Equation 21 and 22 show that charged micelles have
energetically preferred locations within the film which cause them to create a micellar
density profile that has an oscillatory form. Thus the density of micelles in the film forms
a series of peaks that define the most probable position of finding a micelle in the interior
of the film. As the film thickness decreases micelles are "squeezed" out of the film and
the number of density peaks decreases concomitantly. Since the micelle structuring
generates multiple values of the film thickness that are thermodynamically unstable (
oll/Oh > 0 ), the squeezing out process occurs in a discrete manner, changing from one
stable configuration to the next. That is, DFT allows us to interpret the oscillatory
branches of the force curve as arising from micellar structuring in the film.

Clearly different types of structuring (e.g., bilayer, liquid crystal) can occur and
modeling these systems has received little attention. However, Perez et. al. (55,56) have
developed a theory for the structural component of the disjoining pressure in thin films of
liquid crystals. Their thermodynamic theory is based on the concept of surface tension
anisotropy (i.e. variation of nematic liquid crystal interfacial tension with molecular
65
orientation at the interface). The primary contributions to the force results from a balance
of two torques:

'tel : elastic torque opposing disalignment of molecules


'ts: surface torque opposing an increase of surface free enthalpy or surface tension.

Although in some cases good agreement is found between theory and experiment, there
still remains many unexplored problems that should prove to be both challenging and
exciting. The most recent experimental work on bilayer containing films has also
revealed new hydrodynamic phenomena that should also stimulate novel work in this area
(42).

3.6 Hydrophobic Forces


It is now well established that a long-range(> 10 nm) attractive force operates
between hydrophobic surfaces immersed in water and aqueous solutions (57). This force
can be much stronger than those predicted on the basis of van der Waals interactions and
is termed the hydrophobic force. So far no generally accepted theory has been developed
but the hydrophobic force is thought to arise from overlapping solvation zones as two
hydrophobic species come together (18). In fact, Eriksson et. al. (58) have used a square-
gradient variational approach to show that the mean field theory of repulsive hydration
forces can be modified to account for some aspects of hydrophobic attraction.
Conversely, Rukenstein and Churaev suggest a completely different origin that attributes
the attraction to the coalescence of "vacuum gaps" at the hydrophobic surfaces (59). The
exact origins and character of the hydrophobic attraction remains an open question.

New data suggest that hydrophobic forces may also be found in so-called
pseudoemulsion films (i.e. oil/water/air). Considering the strong hydrophobic oil/water
interface in these films, this fact is actually not that surprising. This suggestion stems
from the lack of agreement found between DLVO theory and pseudoemulsion-film
disjoining pressure data (41).
66

4. Experimental Measurements

Disjoining pressure isotherms are typically measured with a Thin-Film-Balance,


TFB, based on the original design of Mysels and Jones (60). A brief description of this
device and the teGhniques to use it are provided here, for a complete review the reader is
referred to Claesson et. al. (61). A schematic of the experimental cell used is provided in
Figure 4.1. Single thin-liquid foam films are formed in the hole drilled through a fritted
glass disc, onto which a glass capillary is fused. These solution-permeable film holders,
are placed in a gas tight measuring cell with the free end of the capillary tube exposed to
a reference pressure (normally atmospheric pressure). Within the cell a precisely
controlled capillary pressure is imposed on the film, which in equilibrium is balanced by
the disjoining pressure. This is accomplished by regulating the gas pressure in the cell
with a syringe pump coupled to a pressure transducer/controller. As shown by Bergeron
and Radke (40), this capillary pressure can be related to the disjoining pressure, II, in a
plane-parallel (40,62) film by the following expression,

2y
n = PG -PR +--11pghc (23)
r

where Pg and Pr are the gas and external reference pressures respectively, y is the bulk

surface tension of the solution, r is the radius of the capillary tube, .!1p is the density
difference between the aqueous surfactant solution and the gas, he is the height of

solution in the capillary tube above the film, and g is the gravitational constant. Each term
on the right side of Equation 23 is measured independently, providing a direct
measurement of II.
67

Porous-frit __..
holder
0 0

Pressure
Transducer I controller

Syringe Pump

Figure 4.1 A schematic showing the principle elements of a typical thin-film-balance used to measure disjoining
pressure isotherms.

Film thicknesses are conveniently measured via Sheludko's microinterferometric


technique (1, 40, 61, 62) in conjunction with video microscopy. White light from a
suitable source is passed through a heat filter and focused at normal incidence onto the
individual foam film formed in the porous-plate holder. Reflected light from the film is
then split and sent to a CCD video camera and a fibre optic probe placed in the
microscope ocular. The video camera documents film drainage throughout the
experiment while light from the fibre optic is filtered (A. =546 nm) and analyzed with a
sensitive photo multiplier tube . The so-called "equivalent" film thickness is then
68
calculated from the standard Scheludko interferometric equation which assumes a
constant refractive index across the film,

heq = (~)arcsin
2nnw 1+ 4R(1-~) (24)
(1-R)2

where !:1 =(1-Imin)/(Imax-Imin). heq is the equivalent film thickness, A. is the wavelength
of light, R = (nw-1)2/(nw+1)2 and nw is the refractive index of the surfactant solution. I
is the instantaneous value of the reflected intensity while Imax and Imin correspond to the
last interference maximum and minimum values. This equivalent thickness is slightly
thicker than the true film thickness, h, because the surfactant adsorption layers at each
film interface have a higher refractive index than the aqueous core. To correct for this
difference we adopt the following multilayer correction factors, derived by Duyvis (32),

h = h - 2h - nw2) - 2h (n2 n2wJ


221
(nhcnw- pg -
(25)
eq he pg 21
nw-

where hhc is the thickness of the surfactant hydrocarbon tails at the interface and hpg is
the surfactant's polar head-group thickness. These values can be calculated from the
volume of the hydrocarbon chain and the polar head-group, together with the area per
molecule at the interface, which we evaluate from surface tension data using Gibbs'
adsorption equation. nhc and npg are the refractive indexes for the hydrocarbon tails and
polar head groups. Finally, the thickness of the film's aqueous core, haq• can be

determined by subtracting the thickness of the adsorbed layers from the total film
thickness evaluated in Equation 25,
69

(26)

Disjoining pressure isotherms are generated by measuring the equilibrium film


thickness after applying a fixed capillary pressure to the film. Equilibrium conditions for
most simple surfactant systems are reached after 10-90 minutes depending on the
magnitude and change of the imposed capillary pressure. Systematic changes to the
capillary pressure by altering the gas cell pressure, Pg• allows us to map out the entire
repulsive (positive) branch of the disjoining pressure isotherm (negative capillary
pressures cannot be imposed with the porous-plate method). The disjoining pressures are
then plotted against the film's aqueous core thickness, haq• to facilitate theoretical
comparisons. Additional experimental details can be found elsewhere (40,61,62).
70

References

1) Scheludko, A., "Thin Liquid Films," Adv. Colloid lnteiface Sci., 1 , pp. 391 (1967).

2) In Thin Liquid Films, Fundamentals and Applications; Ivanov, I.B. Ed; Marcel Dekker: New York, 1988,
Surfactant Science Series, Vol. 29.

3) Derjaguin, B. and Obuchov, E., "Anomalien dunner Fliissigkeitsschichten. III," Acta Physic. U.R.S.S., 5,
No.I, pp. 1-22 (1936).

4) Derjaguin, B. and Churaev, N.V., "On the Question of Detennining the Concept of Disjoining Pressure and
Its Role in the Equilibrium and Flow of Thin Films," J. Colloid lnteiface Sci., 66, No.3, pp. 389-398 (1978).

5) Kralchevsky, P.A., and Ivanov, I.B., "Micromechanical Description of Curved Interfaces, Thin Films, and
Membranes II. Film Surface Tensions, Disjoining Pressure and Interfacial Stress Balances," J. Colloid Inteiface
Sci., 137, No.I, pp. 234-252 (1990).

6) Kralchevsky, P.A. and Ivanov I.B., "Mechanics and Thermodynamics of Curved Thin Films," in Thin Liquid
Films. Fundamentals and Annlications, Ivanov I.B. ed., Marcel Dekker Inc., Surfactant Science Series Vol29.,
Chapter 2, pp. 49-129 (1988).

7) Derjaguin, B.V., Churaev, N.V., and Muller, V.M., Surface Forces, Kichener, J.A. ed., Consultants Bureau,
New York, (1987).

8) Eriksson, J.C., and Toshev, B.V.,"Disjoining Pressure in Soap Film Thermodynamics," Colloids and
Suifaces, 5, pp. 241-264 (1982).

9) Ash, S.G., Everett, D.H., and Radke, C.J., "Thermodynamics of the Effects of Adsorption on Interparticle
Forces," J.C.S. Faraday Trans. II., 69, pp. 1256-1277 (1973).

10) Toshev, B.V. and Ivanov, I.B., "Thermodynamics of Thin Liquid Films I. Basic Relations and Conditions of
Equilibrium," Colloid and Polymer Sci., 253, pp. 558-565 (1975).

11) De Feijter, J.A., Rijnbout, J.B., and Vrij, A., "Contact Angles in Thin Liquid Films I. Thermodynamic
Description," J. Colloid lnteiface Sci., 64, No.2, pp. 258-268 (1978).

12) Rowlinson, J.S. and Widom, B., Molecular Theocy of Capillarity, Oxford Press, New York, Chapters 1-4,
pp. 1-122 (1989).

13) Rusanov, A.l., and Kuni, F.M., "Distribution Functions and Pressure Tensor for a Film of a Simple Liquid,"
in Research in Surface Forces, Derjaguin, B.V. ed., Consultants Bureau, Vol. 3, pp. 111-121 (1971).

14) Kuni, F.M., and Rusanov, A.l., "A Microscopic Theory of Dispersion Interactions in Capillary Systems," in
The Modern Theory of Capillarity, Goodrich, F.C., and Rusanov, A.l. ed., Akadernie-Verlag, Berlin, pp 107-
140 (1981).

15) Attard, P., Mitchell, D.J., and Ninham, B.W., "Beyond Poisson-Boltzmann: Images and Correlations in the
Electric Double Layer I. Counterions Only," J. Chem. Phys., 88, No.8, pp 4987-4996 (1988).

16) Attard, P., Mitchell, D.J., and Ninham, B.W., "Beyond Poisson-Boltzmann: Images and Correlations in the
Electric Double Layer II. Symmetric Electrolyte," J. Chem. Phys., 89, No.7, pp 4358-4367 (1988).

17) Hunter, R.J., Foundations of Colloid Science, Clarendon Press, Oxford Vol.l, (1987).

18) Israelachvili, J.N., Intermolecular and Surface Forces with Annlications to Colloid and Biological Systems.
Academic Press, Orlando, Fl., (1985).

19) Derjaguin, B. V ., Theory of Stability of Colloids and Thin Films, Translated by R.K. Johnston, Consultants
Bureau, New York, (1989).

20) Verwey, E.J.W., and Overbeek, J.Th.G., Theory of the Stability of Lyophobic Colloids The Interaction of
Sol Particles Haying and Electric Double Layer, (collaboration with K. Van Nes), Elsevier Publishing Co.,
New York, (1948).
71

21) Chu, B., Molecular Forces. Based on the Baker Lectures of Peter J.W Debye, Interscience Publishers-John
Wiley and Sons, New York, (1967).

22) Usui, S., and Hachisu, S., "Interaction of Electrical Double Layers and Colloid Stability," in Electrical
Phenomena at Interfaces· Fundamentals. Measurements and Applications, Kitahara, A., and Watanabe, A. ed.,
Marcel Dekker Inc., Surfactant Science Series, Vol.l5, pp 47-98 (198 ).

23) Ohshima, H., "Diffuse Double Layer Interaction Between two Parallel Plates with Constant Surface Charge
Density in an Electrolyte Solution 1. The interaction Between Similar Plates," Colloid and Polymer Sci. 252, pp.
158-164 (1974).

24) Ohshima, H., "Diffuse Double Layer Interaction Between two Parallel Plates with Constant Surface Charge
Density in an Electrolyte Solution 1. The interaction Between Similar Plates," Colloid and Polymer Sci. 252, pp
257-267 (1974).

25) Chan, D., Perram, J.W., White, L.R., and Healy, T.W., "Regulation of Surface Potential at Amphoteric
Surfaces During Particle-Particle Interaction," J.C.S. Faraday Trans., 71, pp 1046-1057, (1975).

26) Chan, D., White, L.R., and Healy, T.W., "Electrical Double Layer Interactions Under Regulation by Surface
Ionization Equilibria- Dissimilar Amphoteric Surfaces," J.C.S. Faraday Trans., 72, pp 2844-2865, (1975).

27) Hamaker, H. C., "The London-van der Waals Attraction between Spherical Particles," Physica, 4, No.IO, pp
1058-1072 (1937).

28) Casmir, H.B., and Polder, D., Phys. Rev., 73, pp. 360-372 (1948).

29) Prausnitz, J.M., Lichtenthaler, R.N. and Azevedo de E.G., "Intermolecular Forces and the Theory of
Corresponding States," in Molecular Thermodynamics of Fluid-Phase Eqpilibria 2Ed., Prentice-Hall Inc.,
Englewood Ciffs, New Jersey, Chapter 4, pp. 48-88 (1986).

30) Vold, M.J ., "The Effect of Adsorption on the Van der Waals Interaction of Spherical Colloidal Particles," J.
Colloid Sci., 16, pp 1-12 (1961).

31) Vincent, B. "The van der Waals Attraction between Colloid Particles Having Adsorbed Layers. II.
Calculation of Interaction Curves," J. Colloid Interface Sci., 42, No.2, pp 270-285, (1973).

32) Duyvis, E. M., "The Equilibrium Thickness of Free Liquid Films," Thesis, Utrecht (1962).

33) Dzyaloshinskii, I.E., Lifshitz, E.M., and Pitaevskii, L.P., "The General Theory of van der Waals Forces,"
translated by Priestley, M.G., Advan. Phys., 10, pp 165-209 (1959).

34) Dzyaloshinskii, I.E., Lifshitz, E.M., and Pitaevskii, L.P., "van der Waals Forces in Liquid Films," Soviet
Physics JETP, 37 (10), No.I, pp 161-170 (1960).

35) Parsegian, V.A., "Long Range van der Waals Forces," in Physical Chemistry: Enriching Topics from
Colloid and Surface Science IUPAC, Olphen, H., and Mysels, K.J., ed., Theorex, La Jolla, CA., pp 27-72
(1975).

36) Parsegian, V.A., and Ninham, B.W., "Application of the Lifshitz Theory to the Calculation of van der Waals
Forces Across Thin Lipid Films," Nature, 224, pp 1197-1198 (1969).

37) Israelachvili, J.N., and Wennerstom, H., "Entropic Forces Between Amphiphilic Surfaces in Liquids," J.
Phys Chern., 96, pp 520..531 (1992).

38) Belorgey, 0., and Benattar, J.J., "Structural Properties of Soap Black Films Investigated by X-Ray
Reflectivity," Phys. Rev. Let., 66, No.3, pp. 313-316 (1991).

39) Gamba, Z., Hautman, J., Shelley, J.C., and Klein, M.L., "Molecular Dynamics Investigation of a Newton
Black Film," Langmuir, 8, pp. 3155-3160 (1992).

40) Bergeron, V ., and Radke, C.J ., "Equilibrium Measurements of Oscillatory Disjoining Pressures in Aqueous
Foam Films," Langmuir, 8, No.l2, pp. 3020-3026 (1992).

41) Bergeron, V., and Radke, C.J., "Disjoining Pressures and Stratification in Asymmetric Thin-Liquid Films,"
Colloid and Polymer Science, 273, pp. 165-174 (1995).
72
42) Bergeron, V., "Microtubes Created in Thin Liquid Films during Bilayer Adhesion and Fusion, 'Langmuir,
12, No.24, pp. 5751-5755 (1996).

43) Richetti, P., and Kekicheff, P., "Direct Measurment of Depletion and Structural Forces in a Micellar
System," Phys. Rev. Let., 68, No.l2, pp. 1951-1954 (1992).

44) Kekicheff, P., and Richetti, P., "Direct Measurements of Interactions in Supermolecular Fluids and Liquid
Crystals," Pure and Appl. Chem., 64, No.I!, pp. 1603-1609 (1992).

45) Richetti, P., Kekicheff•. P, Parker, J.L., and Ninham, B.W., "Measurement of the Interactions between
Membranes in a Stack," Nature, 346, pp. 252-254 (1990).

46) Bergeron, V., "Disjoining Pressures and Film Stability of Alkyltrimethylammonium Bromide Foam Films,"
Langmuir, 13, pp. 3474-3482 (1997).

47) Manev, E.D., Sazdanova, S.V., Rao, A.A., and Wasan, D.T., "Foam Stability- The Effect of a Liquid
Crystalline Phase on the Drainage and Transition Behavior of Foam Films," J. Dispersion Science and
Technology, 3(4}, pp. 453-463 (1982).

48) Keuskamp, J. W., and Lyldema, ].,"Stratification in Free Liquid Films," in Adsorption at Interfaces ACS
Symposium Series 8, K. L. Mittal, Eds. Washington D.C., pp. 191-198 (1975).

49) Exerowa, D. and Lalchev, Z., "Bilayer and Multilayer Foam Films: Model for Study of the Alveolar Surface
and Stability," Langmuir, 2, pp. 668-671 (1986).

50) Wasan, D.T., Nikolov, A.D., Huang, D.O., and Edwards, D.A., "Foam Stability: Effects of Oil and Film
Stratification," in Surfactant-Based Mobility Control Proeress jn Miscible-Flood Enhanced Oil Recovery,
Smith, D.H. ed., American Chemical Society, Washington, D.C., pp. 136-162 (1988).

51) Pollard, M.L., and Radke, C.J., "Density-functional Modeling of Structure and Forces in Thin Micellar
Liquid Films," J. Chem. Phys., 101, No.8, pp. 6979-6991 (1994).

52) Overbeek, J.Th.G., "Black Soap Films," J. Phys. Chem., 64, pp. 1181 (1960).

53) Lyklema J., and Mysels, K.J., "A Study of Double Layer Repulsion and van der Waals Attraction in Soap
Films," J. American Chem. Soc., 87:12, pp. 2539-2546 (1965).

54) Nikolov, A.D., Kralchevsky, P.A., Ivanov, LB., and Wasan, D.T., "Ordered Micelle Structuring in Thin
Films Formed from Anionic Surfactant Solutions II. Model Development," J. Colloid Interface Sci., 133, No. I,
pp.13-22 (1989).

55) Perez, E., and Proust, J.E., and Ter-Minassian-Saraga, L., "Structral Disjoining Pressure in Thin Film of
Liquid Crystals," Colloid and Polymer Sci., 256, pp. 784-792 (1978).

56) Perez, E., and Proust, J.E., and Ter-Minassian-Saraga, L., "Thin Films from Liquid Crystals," in Thin Liquid
Films. Fundamentals and Apolications, Ivanov LB. ed., Marcel Dekker Inc., Surfactant Science Series Vol29.,
Chapter 13, pp. 891-925 (1988).

57) Israelachvili, J.N., and Pashley, R., "The Hydrophobic Interaction is Long Range, Decaying Exponentially
with Distance," Nature, 300, pp. 341-342 (1982).

58) Eriksson, J.C., Ljunggren, S., and Claesson, P.M., "A Phenomenological Theory of Long-Range
Hydrophobic Attraction Forces Based on a Square-gradient Variational Approach," J. Chem. Soc. Faraday
Trans. 2, 85(3), pp. 163-176 (1989).

59) Ruckenstein, E., and Churaev, N., "A Possible Hydrodynamic Orgin of the Forces of Hydrophobic
Attraction," J. Colloid Interface Sci., 147, No.2, pp. 535-538 (1991).
60) Mysels, K. J., and Jones, M.N., "Direct Measurements of the Variation of Double-layer Repulsion with
Distance," Discuss. Faraday Soc., 42, pp.42-50 (1966).
61) Claesson, P., Ederth, T., Bergeron, V., and Rutland, M.W., "Techniques for Measuring Surface Forces,"
Adv. Colloid Interface Sci., 67, pp. 119-183 (1996).
62) Bergeron, V., "Forces and Structure in Surfactant-laden Thin-liquid Films," Phd. Thesis, University of
California, Berkeley, 1993.
STRUCTURE OF FOAM FILMS CONTAINING ADDITIONALLY
POL¥ELECTROLYTES

R. V. KLITZING, A. ESPERT, A. COLIN AND D. LANGEVIN


Centre de Recherche Paul Pascal - CNRS
A v. Dr. Schweitzer, F - 33600 Pessac

1. Introduction

Foams are interesting for many industrial applications like enhanced oil re-
covery, the decontamination of polluted surfaces or cosmetic industry. The
research of foam presents also an important part of the colloid and surface
science. For the comprehension, control and manipulation of the macro-
scopic properties (e.g. stability of foam) it is necessary to understand the
behaviour of the building blocks i.e. the single liquid films. The model sys-
tem of the foam is presented by gas/film/gas.
Pure foam films were the subject of many studies ([7] and references in).
Because of the molecular interactions between the two film interfaces a pres-
sure appears being either positive (disjoining) or negative (conjoining). The
two most important contributions are the electrostatic forces which stabi-
lize the film if they are repulsive and the van der Waals forces which are
always attractive and therefore destabilizing. They form the DLVO theory
[11]. Investigating films made from surfactant solutions above the critical
micelle concentration (erne) a stepwise thinning is observed which cannot
be explained by the DLVO theory. This oscillation of the disjoining pressure
is due to additional stratification forces within the film [7, 15]. This phe-
nomenon occurs also in films of liquid-crystals [13] and latex suspensions
[25].
Using polylectrolytes is one possibility to stabilize thin liquid films and
therefore the whole foam. Polyelectrolytes are well known as flocculants and
stabilizers of colloidal dispersions due to electrostatic and entropic interac-
tions [23]. In the measurements presented in this paper the film was formed
from semi-dilute anionic polyelectrolyte solutions containing a cationic sur-
factant at a concentration below the erne. The disjoining pressure isotherm
{disjoining pressure in dependence on the film thickness) shows also os-
73
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 73-82.
© 1999 Kluwer Academic Publishers.
74
cillations in spite of the low surfactant concentration. The first proof of
such a kind of stratification force in films of polyelectrolytes and surfac-
tants were found by Bergeron et al. (6]. Varying the concentration c of the
polyelectrolyte within the semidilute regime, it was found that the period
of oscillations scales as c- 1/ 2 (1] and the stratification was related to the
presence of a network of polyelectrolyte chains.
Oscillations for polymer-like structures in confined geometry are also found
with other kinds of force apparatus like SFA (14] or AFM (20].
In this paper the results for a strongly charged (Polystyrene sulfonate, PSS)
and a weakly charged polyelectrolyte (Acrylamide-Acrylamide-sulfonate,
AAS) are compared. PSS has a hydrophobic backbone while the AAS is
soluble in water even in the uncharged form. The influence of the confine-
ment on the polyelectrolyte structure in the thin film is checked. Further,
the influence of the film surface on the stratification is investigated by re-
placing the cationic surfactant with a nonionic one.
To get informations about the adsorption behaviour of the polyelectrolyte
and the surfactant at the film interface, surface tension measurements at
the liquid-air interface were also carried out.

2. Experimental Section
The strongly charged polyanion used in this study is polystyrenesulfonate
sodium salt (PSS). It has a molecular weight of "' 70000 gfmol and a
degree of charge of 100 %. The weakly charged polyelectrolyte was an an-
ionic random-block copolymer: acrylamide-acrylamidesulfonate (AAS) with
a molecular weight of"' 400000 gfmol and a degree of charge of 25 %. As
surfactants the cationic dodecyltrimethylammonium bromide (DTAB, 99
%) and the nonionic penta(ethylene glycol)mono-n-dodecyl ether (C 12 E 5 )
were used. All surfactant concentrations were below the erne. The struc-
tural formulas of PSS and AAS are shown in fig. 1.
For measuring disjoining pressure isotherms the porous-plate technique de-
scribed by Bergeron (5] in this issue was used. The films were formed from
aqueous solutions containing surfactant and polyelectrolyte. All measure-
ments were carried out under equilibrium conditions at 23°.
The equilibrium measurements of the surface tension were carried out with
an open frame version of the Wilhelmy plate technique.

3. Results
Fig. 2 shows the disjoining pressure isotherms for the mixture of DTAB
and PSS. All data points are measured in equilibrium. The concentration of
PSS was varied up to 8000 ppm. The concentration unit ppm refers to the
polyelectrolyte weight with respect to the weight of the solvent, i.e. water.
75

PSS AAS

j- 3000
m- 340 i- 1000

Figure 1. Polystyrenesulfonate sodium salt (PSS, m.....,340) and Acrylamide-acrylamide-


sulfonate (AAS, i.....,lOOO, j.....,3000).

1600 1200
(?
a.
a) 1000 • b) -
; 1200 •
800 •
:::J
(/)

(/)
~ 800 600 •
a.
•• •
OJ
c:
·c: • 400 •••
"6
"00'
400
• 200 •
'6 ·~·
•• •••
0 0
20 40 60 80 100 120 20 40 60 80 100 120

BOO 800

•·~·•
c) d)
(?
e:. •
• •• ••
600 600
~
• • ••
:::J
(/)

•• ••
(/)

-
c. 400
• 400
Q)

OJ
c:
c: ·+-•
• • 200 •• •
• \+-.
"6 200
"00'
'6 • \~·~·
0 0
20 40 60 80 100 120 20 40 60 80 100 120
thickness (nm) thickness (nm)

Figure 2. Disjoining pressure isotherms for the system DTAB/PSS at different PSS
concentrations: a) 200 ppm, b) 2000 ppm, c) 6000 ppm, d) 8000 ppm. The DTAB
concentration was fixed at 10- 4 mol/!.

Obviously, the isotherm profile becomes steeper with increasing polyelec-


trolyte concentration. From concentrations about 1000 ppm up the thinning
is not any longer continuous. The film thins stepwisely by increasing the
disjoining pressure. The jumps are not reversible. The number of jumps
76
75

-zE
....... 70

-E
c
0
65

-
"iii
c
Q)
60
Q)
0
as
't: 55
:::J
C/)

50
101 102 103 104
c (ppm)
Figure 9. Surface tension in dependence on the PSS concentration: pure PSS solution
{empty circles) and the mixture between DTAB {fixed concentration, 10- 4 mol) and PSS
{filled circles). Dashed line: surface tension of the pure DTAB solution at a concentration
of w- 4 mol.

increases with the PSS concentration and their size decreases from 43 nm
for 1000 ppm to 15 nm for 8000 ppm. For a fixed PSS concentration the size
of the different jumps is almost the same. For the branch with the smallest
thickness the film thickness becomes thinner with increasing polyelectrolyte
concentration.
Qualitatively, the results for the mixture of DTAB and AAS are similar.
The isotherms for the system DTAB/ AAS show also a decrease in jump
size with increasing polyelectrolyte concentration but the jumps already
occur from a concentration as low as 300 ppm. The jumps for DTAB/ AAS
are smaller than for DTAB/PSS at a fixed polyelectrolyte concentration.
The transition is faster and the equilibrium is reached in a shorter time for
PSS than for AAS.
To get information about the interactions between the polyelectrolyte and
the surfactant at the film surface, the surface tension of DTAB and PSS
respectively and for a mixture of both components were measured. The
results presented in fig. 3 show that the surface tension for the pure PSS
solution is almost the same as for water up to a polyelectrolyte concentra-
tion of 1000 ppm. Above this concentration, the surface tension decreases.
Qualitatively, PSS behaves as a surfactant [9]. Using the mixture between
DTAB (fixed concentration at 10- 4 mol) and PSS even for polyelectrolyte
concentrations well below 1000 ppm the surface tension is reduced. The
difference between the curve for the pure PSS solution and that for the
mixture corresponds to the formation of complexes between the cationic
surfactant and the anionic polyelectrolyte at the surface. Films of pure
77

-"' 800

-a.....
Q) 600
0

0 •

::l
CfJ
CfJ 0
....

Q)
c.. 400 0


0>
c
·c: 0
·a 200
0
0 •
"(j)
-<--- -- -.------- 0
:0 0

0
20 40 60 80 100
film thickness (nm)
Figure 4. Comparison between the isotherms of DTAB/PSS (filled circles) and
C12Es/PSS (empty circles). The concentration of PSS was 1000 ppm.

PSS solutions are not stable, although the surface tension can be consider-
ably lowered compared to water.
For the system DTAB / AAS the results of the surface measurements indi-
cate also a formation of complexes at the surface.
To get informations about the influence of these surface complexes on the
properties of the film the cationic DTAB was replaced by the nonionic
C12E5 which forms no complexes with the PSS. No changes in the sur-
face tension are observable by adding PSS to the C12E5 solution which
confirms results of Radlinska et al. [26]. PSS was used at a concentration
of 1000 ppm. In fig. 4 the isotherms of the films made from solutions of
C12E5/PSS and DTAB/PSS respectively are shown. The film containing
C12E5 is thinner.

4. Discussion and Conclusions


4.1. EFFECT OF THE SURFACE

With increasing polyelectrolyte concentration complexes between the polyan-


ion and the cationic surfactant are formed at the interface. The counteri-
ons of both polyelectrolytes and surfactants are exchanged by the opposite
charges of surfactant and polyelectrolytes respectively [2]. An exchange of
counter ions explains for example also the interdigitation between neigh-
bouring oppositely charged polyelectrolyte layers in a multilayer system
[18]. In both cases it is assumed that the concentration of counterions in
the complex layer is very small. The formation of complexes is also well
studied for mixtures of insoluble amphiphiles and oppositly charged poly-
78
electrolytes [21, 27].
Because of the complexes between polyelectrolyte and surfactant stable
films of DTAB/PSS can be formed while the films of the pure DTAB and
PSS respectively are unstable. In addition to the reduction of surface ten-
sion, the adsorption kinetics seem to play an important role in the formation
of stable films. The stability of films could also be influenced by fluctuation
at the interface [4].
FUrthermore, the repulsion between surface layers with increasing polyelec-
trolyte concentration is reduced which explains the lowered film thickness
for higher polyelectrolyte concentration. It is assumed that the surface po-
tential is lowered in comparison with that of the pure surfactant at the
interface.
Preliminary results of ellipsometry measurements at the air-liquid interface
lead to a thickness for one complex layer of about 3 nm [2].
The concentration of counterions increases with the polyelectrolyte concen-
tration. The charges of the polyelectrolyte chain are more screened which
leads to a steeper profile of the disjoining pressure isotherm. The data points
are in good agreement with exponential fits of the form exp( -Kh) (K: in-
verse Debye length) as expected for screened electrostatic repulsion. The
fits result into Debye lengths between 22 nm for 200 ppm and 1.5 nm for
8000 ppm using PSS. For the Debye length the concentration of counterions
is important which is determined by the charged monomer units. In view
of the different molecular weights and the counterion condensation at the
PSS chain the concentration of free counterions at a fixed polyelectrolyte
concentration (in [ppm]) is similar for AAS and PSS which leads to similar
Debye lengths.
The surface tensions measured at the air - liquid interface are not directly
transferable to the film surfaces. The electrostatic interaction between the
two interfaces could lead to a change of surface excess and dynamics. But
the results give a qualitative idea of the influence of polyelectrolyte concen-
tration at the film surface.

4.2. EFFECT OF THE BULK

For low polyelectrolyte concentrations the isotherm is continuous as ex-


pected. It is assumed that there are no micelles inside the film because the
used concentration of DTAB is 150 times lower than the erne. Even if there
is a higher concentration in the film than in the solution bulk caused by
the confined geometry in the film, it is not probable that the erne will be
reached. The surfactant concentration was always below the critical aggre-
gation concentration (cac) which is about 7*10- 3 M for AAS [2]. In addition
a surfactant concentration above the cac leads to dense polyelectrolyte -
79

surfactant complexes and therefore to a heterogeneous film thickness with


almost immovable domains [6]. Though the film thicknesses in the mea-
surements presented in this paper were always homogeneous. So the jumps
of the isotherm are not caused by molecule aggregates inside the film.
By replacing the cationic DTAB with the nonionic- C 12E5 the surface is
changed from a mixed surface to a pure surfactant surface. The whole
amount of polyelectrolyte is assumed to be in the film bulk now. The re-
sults show that the jump size is independent of the surfactant used. There-
fore, we conclude that the jumps are caused by the structure formed by the
polyelectrolyte chains in the film bulk.
Moreover, there is no observable interaction between the polyelectrolyte
chains and the nonionic surfactant molecules in the bulk. The indepen-
dence of the macromolecular structuring on the surface properties is also
well known from measurements of pure surfactant films. The jump size and
therefore the size of the micelles in foam films and in pseudoemulsion films
(oil/film/gas) are almost the same [8].
The image which is geometrically most similar to that of micelle layers in
the film is a layering structure of polyelectrolyte coils. In this case the os-
cillation period of the isotherms corresponds to the size of the coils and the
jump size would increase with the molecular weight because of R 9 ex N 3 / 5
(R9 : gyration radius, N: number of monomers per chain). Though AAS
measurements with different molecular weights show that the jump size is
independent of the chain length [1).
A polyelectrolyte bridging [6) could only explain one jump during film thin-
ning but in the presented measurements in some cases more than one jump
can be observed.
A network of polyelectrolytes inside the film is proposed where the mea-
sured oscillation period of the disjoining pressure is related to the network
mesh size [1] which will be considered more detailed in the following. The
structure of polyelectrolytes in solution is the subject of many studies ([3]
and references in). One of the topics is the transition between the dilute
and the semi-dilute regime indicated by the concentration c*. From this
critical polyelectrolyte concentration on, the chains begin to interdigitate
e.
and form a network with the mesh size Along the strongly charged chain,
the distance between two charges is smaller than the Bjerrum length (7 A
in water at 25°C) which is the natural distance in a solvent determined
by the compensation of electrostatic and thermal energy. In this case, the
Manning condensation of counter ions [19) plays an important role. For
the used PSS, the distance between two charges is about 2.5 A. So, it can
be described by the theory of Odijk [22), Skolnick and Fixman [28) for a
network of strongly charged polyelectrolytes. The behaviour of a weakly
charged polyelectrolyte like the used AAS (distance between two charges is
80
35 ~~~~~--~~~~~""

30
50 • b)

~25 40
E
.s 20
Q)
30
N
"iii 15
Q. 20
E 10
.2.
10
5
0 ~~~~~~~~~~-L~ 0
0 2000 4000 6000 8000 0 1000 2000 3000 4000
c (ppm) c (ppm)

Figure 5. Measured jump size (divided by factor .J2) of isotherms in dependence on


the polyelectrolyte concentration for a) DTAB/PSS and b) DTAB/AAS. The solid lines
correspond to fits with a power law resulting in an exponent of -1/2.

24 A) is described by the theory of Khoklov and Khatchaturian [16]. Both


theories show that the mesh size scales as c- 1/ 2 .
In the following the application of the bulk theory to the behaviour of the
polyelectrolytes in the confined geometry of a film is checked. Fig. 5 shows
the measured jump sizes of the isotherm in dependence on the polyelec-
trolyte concentration. To take into account that the oscillation period of
the force measured between to interfaces mediated by the polyelectrolyte
solution corresponds to V2e [10] the values of the jumps are divided by
factor V2. For both DTAB/PPS (fig. 5a) and DTAB/AAS (fig. 5b) the fit
results in an exponent of almost -1/2 which is in good agreement with the
theoretical relation~ <X c- 1/ 2 .
Because of the lower charge density along the AAS chain, the electrostatic
repulsion between different chains is lower and the mesh size of the network
smaller. This is in good agreement with a lower jump size for the system
DTAB/ AAS than for DTAB/PSS at the same polyelectrolyte concentra-
tion. The fact that the chains of the used AAS are weakly charged, and
longer leads to the conclusion that the interdigitations are stronger which
is related to a higher viscosity than in the case of PSS [17]. This could
explain the slower transitions and longer times needed to equilibrate the
system with AAS.

From the facts presented in this paper it seems straightforward to derive


the film structure shown in fig. 6. At the interface there are complexes be-
tween surfactant and polyelectrolytes. The polyelectrolyte network within
the film has a correlation length which is related to the measured jump size
between the branches of the isotherms. As for films of pure surfactant solu-
81

I··-~

Figure 6. Model of the film structure with surfactant/polyelectrolyte complexes at


the interface and a polyelectrolyte network within the film. The correlation length e
corresponds to the measured jump size t:..h of the disjoining pressure isotherms

tions the oscillations of the disjoining pressure can be related to a structure


inside the film.
In contrast to spherical particles (e.g. micelles, latex particles or polymer
coils) where the stepwise thinning is due to the expulsion of layer by layer of
the particles the discontinuous thinning of foam films containing polyelec-
trolytes is not easy to understand. If the stepwise thinning is understood as
the expulsion of a part of the network with the dimension of the correlation
length the whereabouts of the rest of the chain involved in other meshes
will not be clear. There would have to be a complete rearrangement of
the network. Another possibility is that the layers of the network collapse
one after an other. Such an image is proposed for numerical simulations of
foams under stress [24]. It has to be checked carefully if these results can
be applied to networks of polyelectrolytes.
We are currently investigating the behaviour of rigid polyelectrolytes and
other polymer-like molecular formations [12].

References
1. Asna.cios, A., Espert, A., Colin, A., & Langevin, D. {1997). Structural forces in thin
films made from polyelectrolyte solutions. Phys. Rev. Lett., 78, 4974-4977.
2. Asnacios, A., Langevin, D., & Argiller, J.-F. (1996). Complexation of cationic surfac-
tant and anionic polymer at the air-water interface. Macromolecules, 29, 7412-7417.
3. Barrat, J. L. & Joanny, J. F. {1996). Theory of polyelectrolyte solutions. Adv. Chern.
Phys., 94, 1-66.
4. Bergeron, V. {1997a). Disjoinig pressures and film stability of alkyltrimethylammo-
nium bromide foam films. Langmuir, 13, 3474-3482.
5. Bergeron, V. {1997b). An introduction to forces and structure in individual foam
82

and emulsion films. In J.-F. Sadoc & N. Rivier (Eds.), Foams, Emulsions and Cellular
Materials. Kluwer Acad. Publ.
6. Bergeron, V., Langevin, D., & Asnacios, A. (1996). Thin-film forces in foam films
containing anionic polyelectrolyte and charged surfactants. Langmuir, 12, 1550-1556.
7. Bergeron, V. & Radke, C. J. (1992). Equilibrium measurements of oscillatory dis-
joining pressures in aqueous foam films. Langmuir, 8, 3020-3026.
8. Bergeron, V. & Radke, C. J. (1995). Disjoining pressure and stratification in assym-
metric thin-liquid films. Colloid Polym. Sci., 273, 165-174.
9. Caminati, G. & Gabrielli, G. (1993). Polystyrene sulfonate adsorption at water-
graphon and water-air interfaces. Colloids and Surfaces A, 70, 1-14.
10. Chatellier, X. & Joanny, J.-F. (1996). Adsorption of polyelectrolyte solutions on
surfaces: A Debye Hueckel-theory. J. Phys. II France, 6, 1669-1686.
11. Derjaguin, B., Churaev, N. V., & Muller, V. M. (1987). In J. A. Kitchener (Ed.),
Surface Forces (pp. 293). New York: Consultants Bureau.
12. Espert, A., v. Klitzing, R., Colin, A., & Langevin, D. (1997). unpublished.
13. Ivanov, I. B. & Dimitrov, D. S. (1988). Thin film drainage. In I. B. Ivanov (Ed.),
Thin Liquid Films chapter 7, (pp. 379-496). New York: Marcel Dekker, Inc. Surfactant
Science Series Volume 29.
14. Kekicheff, P., Nallet, F., & Richetti, P. (1994). Measurement of depletion interaction
in semi-dilute solutions of worm-like surfactant aggregates. J. Phys. II France, 4, 735-
741.
15. Kekicheff, P. & Richetti, P. (1992). Direct measurement of depletion and structural
forces in a micellar system. Phys. Rev. Lett., 68, 1951-1954.
16. Khokolov, A. R. & Khatchaturian, K. A. (1982). On the theory of weakly charged
polyelectrolytes. Polymer, 23, 1742-1750.
17. Klitzing, R., Espert, A., Asnacios, A., Hellweg, T., Colin, A., & Langevin, D. (1997).
Forces in foam films containing polyelectrolyte and surfactant. submitted to Colloids
and Surfaces A: Physicochemical and Engineering Aspects.
18. Klitzing, R. & Mohwald, H. (1995). Proton concentration profile in ultrathin poly-
electrolyte films. Langmuir, 11, 3554-3559.
19. Manning, G. S. (1969). Limiting laws and counterion condensation in polyelectrolyte
solutions i. colligative properties. J. Chern. Phys., 51, 924-933.
20. Milling, A. J. (1996). Depletion and structuring of sodium poly(styrenesulfonate)
at the silica-water interface. J. Phys. Chern., 100, 8986-8993.
21. Miyano, K., Asano, K., & Shimomura, M. (1991). Adsorption kinetics of water-
soluble polymers onto a spread monolayer. Langmuir, 7, 444-445.
22. Odijk, T. (1977). Possible scaling relations of semidilute polyelectrolyte solutions.
J. Polym. Sci., Polym Phys. Ed., 15, 688-693.
23. Ortega-Vinuesa, J. L., Martin-Rodriguez, A., & Hidalgo-Alvarez, R. {1996). Col-
loidal stability of polymer colloids with different interfacial properties: Mechanisms.
J. Coll. Inter. Sci., 184, 259-267.
24. Papka, S. D. & Kyriakides, S. (1994). In-plane compressive response and crushing
honey comb. J. Mech. Phys. Solids, 42, 1499-1532.
25. Perez, E., Proust, J. E., & Ter-Minassian-Sagara, L. {1988). Thin films from liquid
crystals. In I. B. Ivanov (Ed.), Thin Liquid Films chapter 13, (pp. 891-925). New
York: Marcel Dekker, Inc. Surfactant Science Series Volume 29.
26. Radlinska, E. Z., Gulik-Krzywicki, T., Lafuma, F., Langevin, D., Urbach, W.,
Williams, C. E., & Ober, R. (1995). Polymer confinenment in surfactant bilayers
of a lyotropic lamellar phase. Phys. Rev. Lett., 74, 4237-4240.
27. Royappa, A. T. & Rubner, M. F. (1992). Novel langmuir-blodgett films of conduct-
ing polymers. i. polyion complexes and their multilayer heterostructures. Langmuir,
8, 3168-3177.
28. Skolnick, J. & Fixman, M. {1977). Electrostatic persistence length of a wormlike
polyelectrolyte. Macromolecules, 10, 944-948.
DRAINAGE OF FOAM FILMS

ROUMEN TSEKOV
Max Planck Institute for Colloids and Interfaces
University of Mining and Technology, 09599 Freiberg, Germany

1. Introduction

The science of foams is as old as classical mechanics since the first systematic study in
this area has been carried out by Newton [1]. He has noticed that soap films, losing
liquid in time, pass through various thicknesses with characteristic colours. When the
thinning process advances black spots appear which are sometimes unstable and the
film ruptures at their place. These structures are nowadays known as Newton black
films. During the next centuries the interest to thin liquid films (TLFs) has increased
[2-4] due to their importance for disperse systems such as foams, emulsions and sus-
pensions widely employed in technology. The stability of dispersions is directly related
to the process of thinning and rupture of TLFs dividing bubbles, droplets and particles
(5). At present, intensive investigations are performed on foam films formed from
aqueous solutions of surfactants with biological origin. Special attention [6] is paid to
TLFs stabilised by phospholipides because of the important role of these substances in
the structure of biological membranes and metabolism of the living cells.
The simplest TLF system, a single microscopic foam film, has proven to be a useful
tool for model investigations. In many aspects the results of such a study are applicable
to other types of TLFs as well, e.g. emulsion films, films on solid and liquid substrata.
The contemporary versions of the interferomertic method proposed by Scheludko and
Exerowa [7) enable measurement of the local kinematics in time of the film thickness.
It is a challenge presently for theoreticians to supply a detailed theory of the TLF
83
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 83-90.
© 1999 Kluwer Academic Publishers.
84

drainage. Scheludko [8] has also proposed a formula for the thinning rate of horizontal
microscopic films fom1ed in the centre of a double concave drop
2h3/lp
V• JrtR 2 • VRc (1)

where V is the drainage velocity, h and R are the thickness and radius of the film, re-
spectively, 11 is the dynamic viscosity and llp is the difference between the capillary
and disjoining pressures. Equation (1) is analogous to an expression known as the Rey-
nolds law (9) for the velocity of approach of two parallel rigid discs separated by a liq-
uid slit. The necessary conditions for applicability of the Reynolds law to foam films
are tangentially immobile plane-parallel film surfaces and equilibrated Plateau border.
In practice, however, many deviations from Eq. (1) have been observed and the reason
should be found in violations of these compulsory conditions.
Deviations from the Reynolds law due to tangential surface mobility have been ex-
perimentally established [10-12] and theoretically described [13-15]. The usual way to
produce relatively stable axisymmetrical films is to stabilise them with surfactants.
Since the film drainage causes an interfacial gradient of the surfactant adsorption, a
gradient of the surface tension appears and the corresponding Marangoni force slows
down the interfacial flux (14-16]. The theory predicts a thinning rate satisfying Eq. (1)
again but with a reduced viscosity Ti=ll/(l+Jr)D/aE) where a, E and D are the
surfactant adsorption length, Gibbs elasticity and diffusion coefficient, respectively.
This means that V is again inversely proportional to the square of the film radius R.
Many experimentally determined rates of thinning of foam and emulsion films, how-
ever, exhibit much weaker dependence on the film size: V is inversely proportional to
the film radius R to the power of about 0.8. Such a behaviour cannot be explained by
surface mobility, the effect of which is not expected to alter the functional dependence
of the thinning rate on the film radius. Moreover, tangential immobility on the film
surfaces can be achieved at higher surfactant concentrations in the film: for aE >> 3rJD
the film surfaces become immobile and 'ij = 11·
Malhotra and Wasan [17] have proposed a model explaining the deviations of the
R -dependence of V from the Reynolds law by disturbances of the equilibrium state in
the meniscus due to liquid outflow from the film. The latter are confirmed by exact
numerical calculations [18), too. This effect diminishes, however, when the volume of
85
the meniscus is much larger than the volume of the film and the film drainage is rela-
tively slow. In addition, the relaxation in the Plateau border is as quicker as deeper it is.
Therefore, for a sufficiently thin films with sharp Plateau borders one can expect that
the meniscus remains quasi-equilibrated during the film drainage.
Obviously, the thickness non-uniformity of the films is the basic reason for devia-
tions from Eq. (1) in the case of TLFs stabilised by surfactants. The influence of the
surface deformability on film thinning is the most complicated effect. Equation (1) re-
quires drainage of the liquid to be strictly symmetrical with respect to the centre of the
circular film confined between two parallel flat surfaces. With solid surfaces this con-
dition is provided by their rigidity while the profile of a film with fluid surfaces is de-
termined by the unifonn external pressure upon the film area and capillary and surface
forces. Many experimental investigations [19-22] have shown that for large radii the
film surfaces are not strictly planar and parallel, i.e. the films are non-homogeneous in
thickness. Due to the film geometry, the liquid in the film centre flows initially slower
as compared to that in the film periphery. In this way a characteristic shape is formed
which is known under the name of "dimple" (see Figure 1a). Ruckenstein and Sharma
[23) have put forward an explanation of the deviations from Eq. (1) through the peri-
staltic action of surface hydrodynamic propagating waves formed in the process of the
film thinning. Observations show, however, quasi-static surface corrugations rather
than running surface waves.

Figure 1. Film shape during tbe film drainage: (a) single and (b) multiple axisymmetrical dimples.

2. The Classical Theory

The theoretical problem of dimple formation and its evolution has been widely studied
[24-29). The outflow from TLFs is described by the Navier-Stokes equations in the
frames of the Reynolds lubrication approximation valid if h << R [3]. For an axisym-
mctrical dimple, the corresponding continuity and dynamic equations in cylindrical co-
86
ordinates are aP(pu) + pa,w = 0 and aPP =lJa ••u where u and ware the radial and

nom1al components of the hydrodynamic velocity, P ( a,P = 0) is the pressure in the


film, and a indicates partial derivatives. Integrating the dynamic equation twice on z
with the boundary condition of tangentially immobile film surfaces u(z = .z.H I 2) .. 0,

one has Stlll • ( 4z 2 - H 2 )o Pp where H is the local film thickness. Substituting now u

in the continuity equation and integrating once more under the symmetry condition
w(z .. 0) • 0, the normal component of the hydrodynamic velocity can be expressed by
the relation 24l)pw= op[pz(3H 2 -4z 2 )opP]. Finally, employing this equation and the

kinematics relation o,H • 2w(z • H /2) yields a differential equation governing the

evolution of the local film thickness in time


1211 pa,n = ap(pH 3opP) (2)

The further analysis requires a relationship between the film pressure and the local
film thickness. It is supplied by the normal force balance on the film surfaces
P=G-oap(PapH)/2p-ll(H) (3)

where G is the constant pressure in the gas phase, o is the film surface tension and n
is the disjoining pressure. The capillary term in Eq. (3) is one for relatively flat film
surfaces.
In general, the evolution of the film thickness is governed by Eqs. (2) and (3) and a
set of relevant boundary conditions. There are three main problems, however, to calcu-
late rigorously the film profile evolution. First, Eq. (3) requires an expression for the
dependence of the disjoining pressure from the film thickness. Apart from the classical
van der Waals and electrostatic components of n [3] many other interesting effects of
macroscopic interaction have been discovered. Here one can mention hydrophobic,
structural, protrusion, undulation, etc. forces. In the case of micellar and vesicular sus-
pensions the disjoining pressure exhibits a periodic dependence on H which is respon-
sible for the film stratification. Second, Eqs. (2) and (3) are non-linear and one may
expect a chaotic behaviour of their numerical solutions. Indeed, it is the conclusion of
many experiments that the film evolution is very sensitive to the manner of film prepa-
ration, i.e. the initial conditions are important. Unfortunately, the latter are not strictly
specified from an experimental point of view. Third, the rigorous boundary conditions
could be defined only far away from the film in the meniscus where Eqs. (2) and (3) are
87

no longer valid owing to violation of the lubrication approximation. For this reason
some approximate conditions suggested by the experiments are employed, e.g. mini-
mum of the film shape at the border and constant thinning rate, which are questionable
and less general.
Let us consider now the derivation of Eq. (1) and, hence, the range of validity of the
Reynolds law to foam filll1S. Taking the film surfaces to be flat the local thickness H in
the film is everywhere equal to the average thickness h. In this case Eq. (2) reduces to
121wV = -il 3uP(puPP) where V = -a,h. Integrating twice this equation under the con-

strain that the pressure at the firm rim is equal to the pressure M of the liquid in the
meniscus, one obtains P = M + (R 2 - p2 )3t] VI h 3 • Now, combining this result with Eq.
(3) and integrating once again, one gets

apH= pllp/a- p(2R 2 - 2


p )3rtV !2ah
3
(4)

where !J.p • G- M- n(h). As seen in Figure 1a, the film shape exhibits minimum at
the film border. Hence, (uPH)R = 0 and it follows immediately from Eq. (4) that the

rate of drainage of the film obeys the Reynolds law (1). Integrating Eq. (4) once again
and remembering that h is the average film thickness, one gets the film profile
H = h + (2R 4 - 6R 2 p2 + 3p4 )llp I 12aR 2 (5)
As is seen, Eq. (5) describes a dimple with a thicker part in the film centre and a
thinner ring at the film border as is presented in Figure 1a. Of course, the film is not
more planar and to obtain the range of validity of the Reynolds law one should require
at least a positive thickness in the film thinnest part, i.e. the barrier rim. Using Eq. (5)
one can transfonu the corresponding inequality H(R) ~ 0 to

R s ~12ah I llp (6)


Therefore, the Reynolds law is valid only for small filll1S with radii restricted by ine-
quality (6). This is exactly the conclusion of many experiments which clearly show that
the Reynolds law is only applicable to sufficiently small filll1S which are relatively ho-
mogeneous in thickness. For instance, Radoev eta/. [20] have reported that filll1S with
h =250 nm, !J.p =35 N/m2, a= 35 mN/m and ll = 1 mNs/m2 obey Eq. (1) if they are
smaller than 0.05 nun which coincides with the value of the transition radius predicted
by Eq. (6). Note that all the films approach their equilibrium state with the Reynolds
88

rate of thinning since in the limit Ap- 0 inequality (6) is fulfilled for arbitrary film
radii. This is not surprising because the equilibrium foam films are flat.

3. A New Theory

The question is what happens with TLFs with radii out of the range of inequality (6).
The experimental results (20] indicate an increase of the thickness non-homogeneity in
amplitude and number with increasing film size. This corresponds to increasing diver-
gence from the idealised models of planar film profile and single axisymmetric dimple
to fonna tion of several centres of concurrent thinning in the film. The single dimples in
such films arc not stable and break spontaneously down to more complex structures, an
ensemble of several smaller films (sec Figure 1b). The minimum entropy production
theorem says that this should result in a more favourable hydrodynamic regime and,
therefore, in an increased rate of drainage.
The existence of multiple dimples schematically presented in Figure 1b was recently
demonstrated experimentally (30] in large TLFs. This new phenomenon was theoreti-
cally explained as a result of lost of dynamic correlation in the behaviour of the local
parts in the film [31]. The effect of thickness non-homogeneity in the multiple dimple
on the TLF drainage was also studied both experimentally [30] and theoretically [31).
A new fonuula describing the thinning rate of dimpled thin liquid films was derived

1 vh12
I!J.p8
V=-S - - - (7)
6rJ 4dR4
In contrast to the Reynolds law, the predicted rate of drainage in Eq. (7) is inversely
proportional to the 4/5 power of the film radius which is in excellent agreement with
the experimentally determined functional dependence of V vs. R (sec Figure 2). The

theory predicts the number n = (tJ.pR 2 /16ah) 215 of the film thickness non-homogeneity,

too. As is seen, the number of corrugations increases simultaneously with the film ra-
dius and driving pressure as required by the experiments. It is easy to show now that

the thinning rate of large TLFs V = n.J; VRe is always larger than VRe and reduces to
the Reynolds one only for films with radii given by Eq. (6) since n = 1.
89

Q1

Film radius R[mm)

Figure 2. Dependence of film thinning rate on tbe film radius. The squares [20) and circles [32) represent
=
experimental resul5 and tbe line is tbe prediction ofEq. (1) for films with h 250 nm, Ap =35 N/m2,
a =35 mN/m and 11 =1 mNs/m2• The slope of tbe line is 6/5.

4. Concluding Remarks

The theoretical models describing the film profile as axisymmetrical dimple take into
account the defomtation of the film surfaces during the thinning process. Sometimes,
however, these models do not correspond well to the real film shapes, especially in
large TLFs formed by quick initial withdrawal of liquid. More complex phenomena
take place in such films (33] and the cylindrical symmetry of the thinning process as-
sumed in the present consideration, is violated (34]. As a rule, the asymmetric film
drainage is much quicker than the axisymmetric one due to the peristaltic pumping ef-
fect (23,33). Owing to the stochastic nature of the film shape in this case, the descrip-
tion of the film drainage is more complicated and may require a statistical approach.

S. Acknowledgements

The author is thankful to the Alexander von Humboldt Foundation for the granted fel-
lowship and to the organisers of the NATO Advanced Study Institute "Foams, Emul-
sions and Cellular Materials" for a financial support to attend in the school in Cargese.
90

6. References

1. Newton, I. (1704) Optics, Smith & Walford, London.


2. Mysels, K., Shinoda, K. and Frenkel, S. (1959) Soap Films, Pergamon, New York.
3. Ivanov, I. B. (ed.) (1988) Thin Liquid Films, Dekker, New York.
4. Exerowa, D.R. and Kruglyakov, P.M. (1997) Foam and Foam Films, Elsevier, Amsterdam.
S. Dubin, S.S., Rulev, N.N. and Dimitrov, D.S. (1986) Coagulation and Dynamics of Thin Liquid Films,
Naukova Dumka, Kiev.
6. Langevin, D. and Sonin, A.A. (1994)Adv.Colloid Interface Sci. 51, 1.
7. Scheludko, A. and Exerowa, D. (1959) Commun.ChemJnst., Bulg.Acad.Sci. 7, 123.
8. Scheludko, A (1967)Adv.ColloidlnterfaceSci.l, 391.
9. Reynolds, 0. (1886) Phii.Trans.Roy.Soc. A (London) 177, 57.
10. Manev, E., Scheludko, A and Exerowa, D. (1974) Colloid Polym.Sci. 252,586.
11. Manev, E., Vassilieff, C.S. and Ivanov, LB. (1976) Colloid Polym.Sci. 254, 99.
12. Sonin, A., Bonfillon, A. and Langevin, D. (1994)J.ColloidlnterfaceSci.162, 323.
13. Cooper, D. F., Davey, B.H. and Smith, J.W. (1976) CanJ.Chem.Eng. 54, 631.
14. Davis, R.H., Schonberg, J.A. and Rallison, J.M. (1989) Phys.Fiuids A 1, 77.
15. Chesters, A.K. (1991) TransJnst.Chem.Eng.A 69,259.
16. Barber, A.D. and Hartland, S. (1976) CanJ.Chem.Eng. 54,279.
17. Malhotra, A.H. and Wasan, D.T. (1987)AIChEJ. 33,1533.
18. Joye, J.-L, Miller, C.A. and Hirasaki, GJ. (1992)Langmuir8, 3083.
19. Scheludko, A (1962) Proc.Kon.Ned.Akad. Wet. B 65, 86.
20. Radoev, B., Scheludko, A and Maoev, E. (1983)J.Colloid Interface Sci. 95, 254.
21. Manev, E., Sazdanova, S. and Wasao, D.T. (1984)J.Colloid Interface Sci. 97,591.
22. Li, D. and Slattery, J.C. (1988)A/Ch£ J. 34,862.
23. Ruckenstein, E. and Sharma, A. (1987)J.ColloidlnterfaceSci.ll9, 1.
24. Frankel, S. and Mysels, K. (1962)J.Phys.Chem. 66, 190.
25. Radoev, B. and Ivanov, I. B. (1971n2)Ann.Univ.Sofia, Fac.Chim. 66, 631.
26. Jain, R.K. and Ivanov, LB. (1980)J.Chem.Soc., Faraday Trans. 2 76, 250.
27. Lin, C.-Y. and Slattery, J.C. (1982)AIChEJ. 28, 786.
28. Chen, J.D. (1984) }.Colloid Interface Sci. 98, 329.
29. Tsekov, R. and Ruckenstein, E. (1994) Colloids Surf. A 82, 255.
30. Manev, E., Tsekov, R. and Radoev, B. (1997) Effect of Thickness Non-Homogeneity on the Kinetic
Behaviour of Microsoopic Foam Films, J.Disp.Sci.Technol. in press.
31. Tsekov, R. (1997) The R 4" -Problem in the Drainage of Dimpled Thin Liquid Films, Colloids Surf. A
in press.
32. Exerowa, D. and Kolarov, T. (1964!6S)Ann.Univ.Sofia, Fac.Chim. 59, 207.
33. Joye, J.-L, Hirasaki, GJ. and Miller, C.A. (1996)J.ColloidlnterfaceSci.l71, 542.
34. Joye, J.-L (1994) Mechanisms of Symmetric and Asymmetric Drainage of Foam Films, Ph.D. Thesis,
Rice University.
FOAM EVOLUTION IN TWO DIMENSIONS
A Particular Limit of Domain Growth

JOEL STAVANS
Department of Physics of Complex Systems,
Weizmann Institute of Science
Rehovot 76100, Israel

1. Introduction
Foams are examples of a large class of systems endowed with a cellular
structure pattern: they consist of a tiling of space with compact polygonal
(two dimensions) or polyhedral (three dimensions) domains, separated by
what we will consider structureless boundaries [1, 2, 3, 4]. Besides the in-
trinsic interest in the properties of foams (e.g. their viscoelasticity), which
make them so useful in a large variety of applications [5], foams have been
the focus of intensive research within the last two decades, as a paradigm
of cellular structure evolution [6, 7, 8, 9]. This evolution is driven by the
fact that the boundaries between bubbles, simple soap films, are endowed
with surface tension and are typically curved. This curvature, which be-
trays pressure differences between neighbouring bubbles, induces a transfer
of gas between them. As a result, some bubbles shrink while others grow,
so that the average bubble size in the foam grows with time, and the total
film area decreases (see Fig. 1). Thus a foam should be viewed as a system
which is out of equilibrium, and evolving towards it as it lowers more and
more the energy of its boundaries. A remarkable fact about this evolution
is that it is characterized by statistical properties which are independent
of scale: distributions such as the number of sides of polygonal bubbles
do not change as a result of the evolution. Furthermore, this evolution is
universal: systems disparately different in their microscopic structure such
as polycrystalline materials (10], monolayers of lipids on water [11, 12],
and magnetic domains [13] which also posess a cellular structure, display
essentially the same behavior. In fact, the interest in foams is due in no
small part to C. S. Smith [14], a metallurgist at the University of Chicago,
who had the original insight of using foams as caricatures of polycrystalline
metals.
91
J. F. Sadoc andN. Rivier (eds.), Foams and Emulsions, 91-104.
© 1999 Kluwer Academic Publishers.
92

Figure 1. Snapshots of the evolution of a two-dimensional foam in the asymptotic scaling


state.

It turns out that one can view the evolution of cellular structures from a
broader point of view, and encompass their evolution within a wider frame-
work. To understand this let us digress and think how a polycrystalline
structure in a material such as a metal is produced. One takes such a ma-
terial in its high temperature liquid phase, and quenches it to some final
lower temperature within the solid phase. Crystallites of independent crys-
tallographic orientations nucleate within the molten metal, and grow until
their interfaces meet, so that the solid phase fills completely all space, and
a structure which looks just like a foam is formed . If the final temperature,
while still within the solid phase, is still high enough so that atoms in the
93

a b

Figure 2. Snapshots of a coarsening thin film of succinonitrile (a) in the cellular regime
( 4> = 1), and (b) in the Ostwald ripening regime with 4> = 0.3. For details see [34].

material have appreciable mobility to migrate along grain boundaries, the


cellular structure coarsens, a process called recrystallization. This process is
illustrated in Fig. 2a, in the case of an organic material called succinonitrile.
Alternatively, one may choose the final temperature after quenching to
be within the liquid-solid coexistence region of the material. In that case,
the crystallites that nucleate grow, until they attain a final volume or area
fraction </>different from one. However, this is not the end of the process: the
system, which consists of many small crystallites within the liquid matrix
may still lower its energy by reducing the amount of interfacial area between
the liquid and the crystallites. Just as in the cellular structure case, small
solid domains shrink and melt completely while others grow, in such a
way that the total solid volume fraction is nearly conserved. This process,
called Ostwald ripening occurs in general after a first order phase transition
[15], and one can study it not only in metals but also in binary mixtures
of either two polymers or liquids, organic materials etc. An example of
Ostwald ripening of solid domains of succinonitrile, a plastic crystal, is
shown in Fig. 2b.
Ostwald ripening and cellular structure evolution share many features in
common and can be viewed as different limiting cases of the same problem:
94
in the former case, the domains occupy a small volume fraction while in
the latter, the domains comprise essentially the whole system. It is within
this framwork, which encompasses both problems, that I wish to discuss
the evolution of foams. In these lectures I will concentrate on the case of
two dimensional systems. These latter are much easier to study than their
three dimensional counterparts. From the experimental point of view, an
observation of a two-dimensional system gives essentially all the informa-
tion there is to it, without recourse to complicated procedures to deduce
structure from cuts of a three-dimensional system or from scattering. There-
fore, one can follow the dynamics of all the system in real time, and study
the elementary processes of evolution in detail. From the theoretical point
of view, much more is known about the geometrical properties of plane
tilings than from partitions of three-dimensional space. Moreover, one can
perform computer simulations of large two dimensional systems whereas
simulations of three-dimensional structures can be rather costly. As a final
word, these notes are not intended to be a review, and therefore will not
present neither Ostwald ripening nor the evolution of cellular structures in
full detail. I will just skim over salient features. The persevering reader is
referred to several recent reviews and papers on these subjects for details,
and a more comprehensive list of references.

2. Cellular Structures: the Foam as a Paradigm

A two-dimensional foam can be produced quite easily: take two nearly-


spaced glass plates (a few millimeters will suffice), seal the space in between,
and blow bubbles inside. Assuming that the bubble size is larger than the
plate spacing, bubbles will form beautiful patterns of polygons tiling the
whole window such as the ones shown in Fig. 1. Close scrutiny of these
patterns reveals the following features:

1. the films separating bubbles are not straight but are actually arcs of
circle.
2. at each vertex in the network, only three films meet.
3. the angle between films meeting at a vertex is 120 degrees.

As mentioned above, the curvature of films is due to a pressure difference


between adjacent bubbles. The latter is given by the Laplace relation: l:!..p =
2u / r where r is the radius of curvature of the film and u the surface
tension. The fact that only three films meet at each vertex is a consequence
of local energy minimization. A generic configuration of four points linked
by lines in a four- vertex is characterized by a larger film area (line length)
and consequently energy, than a configuration in which a new film forms,
linking two vertices with three films each, as shown in Fig. 3.
95

Figure 9. Generic configuration of four points on the plane. The configuration with one
vertex in which four films meet has a larger overall length than the one with two vertices
with three films each.

Under these conditions, a theorem by Euler applies: the average num-


ber of sides per domain in a polygonal network in which each vertex has
coordination number three is six: < n >= 6. Finally, since all films are char-
acterized by the same value of the surface tension, mechanical equilibrium
requires the angle between films meeting at a vertex to be 120 degrees.
In spite of the close similarity between the appearance of the foam
and the patterns of polycrystals shown in Fig. 2a, there are some subtle
differences: grain boundaries do not have the shape of circle arcs. The reason
is that the energy associated with a grain boundary is a highly local quantity
which varies along its contour. It is a complicated function determined
both by the difference in crystallographic orientation between the solid
grains, and by the angle the boundary makes with these. As a consequence,
each grain boundary pulls with a different force at each vertex, and the
mechanical balance of forces does not yield an angle of 120 degrees between
boundaries. In addition to these, there are also differences in the microscopic
processes taking place in these two systems: while in a foam the timescale
governing diffusion of gas across films is orders of magnitude slower than
the mechanical equilibration of a film, the kinetics of atom migration along
96
a grain boundary is of the same order of magnitude as changes in the shape
of the boundary. Notwithstanding these differences, the evolution of both
systems is remarkably similar, and the possible differences in statistical
characterizations fall within experimental error.
Following closely the evolution of a two-dimensional foam reveals a re-
markable fact: the bubbles that shrink and disappear have all less than six
sides. This is the experimental manifestation of a theorem by John von
Neumann who proved the following mathematical statement:
dan (
--=Kn-6 ) (1)
dt
where an is the area of ann-sided bubble and K a constant which depends on
the permeability ofthe films to the gas filling the bubbles [16]. He based the
proof of this statement on the basic geometrical features of two-dimensional
froths we have discussed above. The von Neumann theorem expresses the
physical idea that on average, bubbles with less than six sides look convex,
while those with more than six sides have concave boundaries. A bubble
with convex boundaries has a larger inside pressure than its neighbours, and
therefore the gas inside will tend to flow out and the bubble will shrink.
The opposite argument applies to bubbles with concave sides. A similar
argument exists for polycrystalline materials [17].
We have seemingly arrived at a contradiction: on one hand, Euler's
theorem imposes the existence of bubbles with less than six sides (unless we
have a uniform hexagonal tiling), while on the other, von Neumann theorem
tells us that if we wait long enough, these bubbles will disappear. There is in
fact no contradiction, as Fig. 4 depicting bubble disappearance illustrates.
If a triangular bubble disappears, its three neighbours loose a side each, and
a normal vertex is left in its place. If a bubble with four sides disappears,
two adjacent bubbles loose a side while the number of sides of the two others
remains constant. A new film is created in this process, linking two normal
vertices. Finally, when a bubble with five sides disappears, two neighbours
loose a side, two remain unchanged, and the fifth neighbour gains a side.
As a result of all these processes, new bubbles with less than six sides are
formed, and the contradiction is removed dynamically.
The combination of bubble disappearance, with the concomitant changes
in the number of sides of adjacent bubbles contrive, after a sufficiently long
time has elapsed, to produce a dynamical steady state in which the rela-
tive proportions of bubbles with given number of sides are conserved. In
other words, the distribution of sides calculated at different times becomes
time-independent, even though the unique length scale in the system, the
average bubble size, keeps growing (as t 112 ) [8]. Using more sophisticated
words, patterns become statistically scale invariant. The distribution of the
number of sides is shown in Fig. 5.
97

->-<

Figure 4- Creation of new vertices in which only three films meet, as a result of bubble
disappearance.

There have been many theoretical approaches aiming at reproducing


the experimentally observed distribution [18, 19, 20, 21, 22, 23, 24]. In
particular one can try to model the evolution by a mean-field theory in
which coupled evolution equations for the concentration of bubbles with
n sides are constructed. Since this has been done in detail elsewhere [18,
19, 3], I will limit myself to point out a surprising finding emerging from
these calculations: one does not get a unique distribution as a fixed-point
solution, but rather a one-parameter family. The question that arises then
is, which member of the family is the one consistently observed in Nature.
A dynamical selection principle has been proposed according to which the
98

2 4 6 8 10 12

Figure 5. Distribution of sides of a two- dimensional soap froth in the asymptotic scaling
state. The solid line is the solution of the mean-field equations in [18] with the fastest
decaying tail.

distribution whose tail decays the fastest with increasing n, is the observed
one among all the members ofthe family. The reason is that any distribution
f( n) obtained from experiment must have the following property: there is
a value n = n* so that for all n > n*, f( n) = 0. Any such distribution used
as initial condition in the evolution equations will flow to the "closest"
solution, namely, to the one with the shortest tail.
Up to now, we have regarded the evolution of individual bubbles as
independent, except for the transfer of gas between them. It turns out that
there are actual correlations between neighbouring bubbles: on average,
bubbles with less than six sides (shrinking ones) are surrounded by bubbles
with more than six sides (growing ones). This is embodied in the Weaire-
Aboav law [25, 26] which gives the average number of sides m( n) of the
bubbles adjacent to an n-sided bubble:

b
m(n) =a+- (2)
n
Experimental data are very well fitted with by relation when a 5 and
b = 6 + }1 2 , where }1 2 is the second moment of the distribution of the
99
number of sides. This relation, which arises from an exact sum rule, has
been discussed by Nick Rivier and others in their lectures in more detail.
Not only is there a correlation between the rates of growth of neighbour-
ing bubbles but also between their size. If one averages the area of n-sided
bubbles, one readily notices that this quantity grows monotonically with
n, and that < an > is proportional ton for large values of n [27, 28, 19].
This fact was first observed by Lewis, a botanist who studied the tissues
of cucumbers among others. Together with the Weaire-Aboav law, Lewis's
law implies that small bubbles are surrounded on average by large ones and
viceversa.

3. Ostwald Ripening

We now go back to Fig. 2b, where we show the evolution of well-separated


solid domains in a liquid matrix of the same material. In fact the material
is not pure, it contains a small amount of impurities which open further
the liquid solid coexistence region. These impurities are to a large extent
expelled from the material when it solidifies, forming a diffusion field around
each solid domain. Lifshitz and Slyozov [29], and Wagner [30] proposed
decades ago a model appropriate for small volume fractions (in this case
area fractions) of solid, assuming diffusion and curvature as the driving
forces in the system. In this situation, the domains are supposed to evolve
independently of one another. The model can be solved exactly, and predicts
that: ( i) the average size of the domains grows as the third power of time
< r >= Kt 113 with K a universal constant, and ( ii) that the system reaches
a scaling regime where the distribution of sizes becomes scale invariant,
which they calculated in closed form.
While a large number of experiments observed the 1/3 exponent in the
growth law, they also revealed that ( i) K is volume-fraction dependent,
and ( ii) the size distributions are consistently broader than the LSW pre-
diction. It was later realized that domains do not evolve independently of
one another, but rather interact via their diffusion fields. Correlations then
develop, which perturbative calculations in V'¢ [31, 32, 33] showed to be
of two types: ( i) direct correlations between domain sizes: large domains
are on average surrounded by small ones, and ( ii) correlations between
rates of growth: shrinking domains are on average surrounded by growing
ones and viceversa. These correlations extend within a finite neighbourhood
whose scale is determined by the volume fraction. The existence of these
correlations in the case of the two-dimensional system in Fig.2b has been
explicitely demonstrated recently [34]. As an illustration of these correla-
tions, one can divide crystals into two classes, according to whether they
are smaller or larger than the average size. One can then calculate the pair
100

$=0.40
0.12 •• •


~
.... 0.08
•o
~

·~ • 0 0 0
0 0 0 0
a. 0 0
• 0 0 0 ••
•o " •·~~ooo
• • • •• •
0
0
0.04 oo 0


0

0 •
2 3 4 5 6 7
r
0.08

$=0.40
•••
0.06 • ••
• •
....
••

~
e 0 o 0 oo
.If 0.04 ~ oooo ollo•••l'l!'l~l'loo!'l~o~ll~~oo
Cl • 0
0
•••••.•••• u0 o
• ooooo
0

0.02
• 00


0
0
()

0
• 0
0

0 2 4 6 8 10

Figure 6. Top: size-size pair correlation function as function of normalized distance


for crystals being both smaller or both larger than the average size (empty circles), one
being smaller and the other larger (full circles). Bottom: rate pair correlation function
between pairs of crystals which both shrink or grow (empty circles), or one shrinks while
the other one grows (full circles). From [34]

correlation functions shown in Fig. 6 (top): these measure the likelihood of


finding two crystals within the same class or in different classes, as func-
tion of distance (normalized by the average size). Clearly, small crystals are
preferably surrounded by large ones. One can also calculate analogous pair
correlation functions by dividing crystals into classes according to whether
they grow or shrink. As Fig. 6 (bottom) shows, shrinking crystals are on
average surrounded by growing ones.
Notice the close similarity between the correlation of growth rates and
101

the Weaire-Aboav law. In both cases shrinking domains are "screened" by


growing ones. In addition, direct size correlations also have their counter-
part in cellular structure evolution if the monotonic increase of < an >
with n is taken into account.
Correlation effects were originally invoked to explain the difference be-
tween the distributions measured in experiments and the one calculated
by Lifshitz and Slyozov. This difference also gave rise to another interest-
ing contention: that the solution of the equations in the Lifshitz-Slyozov
model may not be unique [35]. Further theoretical support for this con-
tention has been provided recently [36]. However, no selection mechanism
has been proposed, and the connection between physical distributions and
those belonging to the family of solutions has yet to be established.

4. Conclusions
The main point I have tried to get across is that foam evolution can be en-
compassed within a larger framework, and viewed as a particular limiting
case of domain growth. Both two-dimensional foams as paradigms of two-
dimensional cellular structures (high area fraction coverage), and Ostwald
ripening in the small area fraction regime exhibit scaling regimes of evo-
lution characterized each by a unique lengthscale, which grows with time
as a power law. In both cases, there are finite-range correlations between
the rate of change of domains and their size. Mean-field models of cellu-
lar structure evolution lead to a one-parameter family of distributions. A
dynamical selection principle which singles out a member of this family,
corresponding to experimentally-observed distributions has been proposed.
While there is also strong evidence for the existence of a family of solutions
to Lifshitz-Slyozov model, the classical mean-field description of Ostwald
ripening, no selection principle has been proposed yet. It is not even clear
whether such a principle should exist. An issue remaining for future studies
is then the connection between the family of solutions and those obtained
in experiments in finite systems.

References
1. Stavans, J. (1993) Evolution of Cellular Structures, Rep. Prog. Phys. , Vol. no. 56,
pp. 733-789.
2. Weaire, D. and Rivier, N. (1984) Soap Cells and Statistics-Random Patterns in Two-
Dimensions Contemp. Physics, Vol. no. 25 , pp. 59-99.
3. Stavans, J. (1993) Evolution of Two-Dimensional Cellular Structures: the Soap Froth
Physica A, Vol. no. 194, pp. 307-314.
4. Glazier, J. A. and Weaire, D. (1992) The Kinetics of Cellular Patterns J. Phys.
Condens. Matter, Vol. no. 4, pp. 1867-1894.
5. Aubert, J. H., Kraynik, A. M. and Rand, P. B. (1986) Aqueous Foams Sci. Am. ,
Vol. no. 254 , pp. 74-82.
102

6. Glazier, J. A., Gross, S. P. and Stavans, J. (1987) Dynamics of Two-Dimensional


Soap Froths Phys. Rev. A, Vol. no. 36 , pp. 306-312.
7. Stavans, J. and Glazier, J. A. (1989) Soap Froth Revisited: Dynamic Scaling in the
Two-Dimensional Froth Phys. Rev. Lett. , Vol. no. 62 , pp. 1318-1321.
8. Stavans, J. (1990) Temporal Evolution of Two-Dimensional Drained Soap Froths
Phys. Rev. A, Vol. no. 42, pp. 5049-5051.
9. Aste, T., Szeto, K. Y. and Tam, W. Y. (1996) Statistical Properties and Shell Analysis
in Random Cellular Structures Phys. Rev. E, Vol. no. 54, pp. 5482-5492.
10. Fradkov, V. E., Kravchenko, A. S. and Shvindlerman, L. S. (1985) Experimental
Investigation of Normal Grain Growth in Terms of Area and Topological ClassScripta
Metall., Vol. no. 19, pp. 1291-1296.
11. Stine, K. J., Rauseo, S. A., Moore, B. G., Wise, J. A. and Knobler, C. M. (1990)
Evolution of Foam Structures in Langmuir Monolayers of Pentadecanoic Acid Phys.
Rev. A, Vol. no. 41 , pp. 6884-6892.
12. Berge, B. Simon, A. J. and Libchaber A. (1990) Dynamics of Gas Bubbles in Mono-
layers Phys. Rev. A, Vol. no. 41 , pp. 6893-6900.
13. Weaire, D., Bolton. F., Molho, P. and Glazier, J. A. (1991) Investigation of an
Elementary Model for Magnetic Froth J Phys.: Condens. Mat., Vol. no. 3, pp. 2101-
2114.
14. Smith, C. S. (1952) Grain Shapes and other Metallurgical Applications of Topology
Metal Interfaces, American Society for Metals, Cleveland, OH
15. Langer, J .S. (1992) An Introduction to the Kinetics of First Order Phase Transi-
tions, in Solids Far From Equilibrium, Cambridge University Press, Cambridge.
16. von Neumann, J. (1952) Discussion: Shape of Metal Grains Metal Interfaces, Amer-
ican Society for Metals, Cleveland, OH.
17. Mullins W. W. (1956) Two-Dimensional Motion of Idealized Grain Boundaries J.
Appl. Phys, Vol. no. 27, pp. 900-904.
18. Stavans, J., Domany, E. and Mukamel, D. (1991) Universality and Pattern Selection
in Two-Dimensional Cellular Structures Europhysics Lett, Vol. no. 15, pp. 479-484.
19. Segel, D., Mukamel, D., Krichevsky, 0. and Stavans, J. (1993) Selection Mecha-
nism and Area Distribution in Two-Dimensional Cellular Structures Phys. Rev. E,
Vol. no. 47, pp. 812-819.
20. Flyvbjerg, H. (1993) Model for Coarsening Froths and Foams Phys. Rev. E ,
Vol. no. 47, pp. 4037-4054.
21. Iglesias, J. R. and de Almeida, R. M. C. (1991) Statistical Thermodynamics of a
Two-Dimensional Cellular System Phys. Rev. A, Vol. no. 43, pp. 2763-2770.
22. Holm, E., Glazier, J. A., Srolovitz, D J. and Grest, G. S. (1991) Effects of Lattice
Anisotropy and Temperature on Domain Growth in the Two-Dimensional Potts Model
Phys. Rev. A, Vol. no. 43, pp. 2662-2668.
23. Nagai, T., Ohta, S., Kawasaki, K., and Okuzono, T. (1990) Computer Simula-
tionof Cellular Pattern Growth in Two and Three Dimensions, Phase Transitions
Vol. no. 28, pp. 177-211.
24. Levitan, B. and Domany, E. (1996) Dynamical Features in Coarsening Soap Froth:
Topological Approach Inti. J. Mod. Phys. B, Vol. no. 10, pp. 3765-3805.
25. Aboav, D. A. (1970) The Arrangement of Grains in a Polycrystal, Metallography,
Vol. no. 3 , pp. 383-390.
26. Lambert, C. J. and Weaire, D. L. (1981) The Arrangement of Cells in a Network
Metallography, Vol. no. 14 , pp. 307-318.
27. Lewis, F. T (1928) The Correlation Between Cell Division and the Shapes and
Sizes of Prismatic Cells in the Epidermis of Cucumis Anat. Rec. , Vol. no. 38 , pp.
341-362.
28. Rivier, N. (1982) On the Correlation Between Sizes and Shapes of Cells in Epithelial
Mosaics J. Phys. A, Vol. no. 15 , pp. L143-L148.
29. Lifshitz, I.M. and Slyozov, V.V. (1961) The Kinetics of Precipitation from Super-
saturated Solid Solutions,]. Phys. Chern. Solids, Vol. no. 19, pp. 35-50.
103

30. Wagner, C. (1961) Z. Elektrochem. , Vol. no. 65, pp. 581-587.


31. Tokuyama, M. and Kawasaki, K. (1984) Statistical Mechanical Theory of Coarsen-
ing of Spherical DropletsPhysica A, Vol. no. 123, pp. 386-411.
32. Marder, M. (1985) Correlations and Droplet Growth Phys. Rev. Lett. , Vol. no. 55,
pp. 2953-2956.
33. Zheng, Q .and Gunton J.D. (1989} Theory of Ostwald Ripening for Two-Dimensional
Systems Phys. Rev. A, Vol. no. 39, pp. 4848-4853.
34. Krichevsky, 0. and Stavans, J. (1993) Correlated Ostwald Ripening in Two Dimen-
sions Phys. Rev. Lett. , Vol. no. 70 , pp. 1473-1476; (1995) Ostwald Ripening in
a Two-Dimensional System: Correlation Effects Phys. Rev. E , Vol. no. 52 , pp.
1818-1827.
35. Brown, L. C. (1989) A New Examination of Classical Coarsening Theory Acta met-
all. , Vol. no. 37, pp. 71-77.
36. Meerson, B. and Sasorov, P. V. (1996) Domain Stability, Competition, Growth, and
Selection in Globally Constrained Bistable Systems Phys. Rev. E , Vol. no. 53 , pp.
3491-3494.
104
STATISTICAL THERMODYNAMICS OF FOAM

NICOLAS RIVIER
Laboratoire de Dynamique des Fluides Complexes
Universite Louis Pasteur
3, Rue de l'Universite, 67084 STRASBOURG, France

1. Introduction
Instead of this absurd division into sexes
they ought to class people as static and dynamic
(E. Waugh)

These lectures are divided into two parts.


The first deals with foam in statistical equilibrium. It discusses the structural
equations of state (laws of Aboav-Weaire, Lewis, Peshkin and Lemaitre) which
are the signatures of statistical equilibrium, as PV = NkT is the thermodynamic
signature of an ideal gas. Here, the foam is (quasi-) static, and its statistical
equilibrium is maintained by local, elementary topological transformations
(ETT). These transformations (Tl: neighbour exchange, T2: disappearance of a
3-sided cell. Fig.l) shuffle the shape of a cell (the number n of its sides in two
dimensions) and define the thermodynamic characteristics of the foam (energy,
average cell size, etc.), exactly as collisions shuffle the mechanical state (x,v) of
an atom in the gas and define its pressure and temperature as moments of the
velocity distribution (Bernoulli and Maxwell).
The thermodynamics of foaming is discussed in the second part (§7-8), with
a real, three-dimensional material, the expanded glass Kerroc as an example. Its
structure and material properties will depend essentially on two thermodynamic
parameters, the temperature and duration of the expansion process.
Like a gas or a liquid, a foam is able to explore all local, microscopic
configurations, compatible with macroscopic constraints (1). But, because its
structure is disordered, a foam also contains within itself all accessible local
configurations. Structurally, a solid foam is identical to a liquid foam like soap
froth. The only difference is that the latter coarsens extremely slowly.
105
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 105-125.
© 1999 Kluwer Academic Publishers.
106

I +I

Ti ~I
_,
Figure 1. Elementary topological transformations (E'IT). Tl: neighbour exchange.
1'2: disappearance of a 3-sided cell. Cell division (right) is a combination of the
elementary transformations Tl and 1'2.

Dominique Langevin has warned us in her lectures that 'thermodynamics is


somewhat boring ... and equations of state are dangerous' (or at least, that was
the gist of what she said). This may be sufficient (if hardly necessary) to put the
reader off this chapter. Nevertheless, if equations of state are dangerous, they
are certainly important, as diagnostic of pathologies in the renewal of the
epidermis (psoriasis, melanoma, ... ).As for boredom, I can only try to emulate
John Cleese on 'why accountancy is not boring'. See [1] for an elementary
introduction, and [2] for some generalisation to 3D.

2. Two-dimensional foams

A foam is a random, space-filling cellular network, a planar, regular graph


consisting of C cells (a large number), E edges and V vertices. Edges are
incident on vertices and bounded by vertices. Cells are incident on edges and
bounded by (a cycle ot) edges. Their numbers are related by incidence relations,

3V =2E =<n>C ,

and by the Euler relation,

C-E+V=l
107

(The cell at infinity is not included in C). The number n of edges bounding a
cell is the only local random variable in the foam (all other incidence and boun-
ding numbers are fixed; in particular, randomness imposes 3 edges and cells
incident on a vertex). It is shuffled by the local, elementary topological transfor-
mations (Fig.l). Thus, a foam is a topological pattern in statistical equilibrium,
the local, microscopic variable n attached to an individual cell being shuffled by
topological collisions, like the velocity of an atom in a gas.
The Euler and incidence relations combine to yield <n> = 6. (Corrections
for a finite foam are of order 1/C, and thus negligible in the thermodynamic
limit). A non-hexagonal cell can be regarded as a source of curvature, carrying
a topological charge 6-n, for which there is a topological equivalent of Gauss
theorem [3] (§3.4). The elastic energy of the foam is proportional to the fluctu-
ations of the local random variable J.l2 = L Pn (n-6)2 (Eq.(lO)).
The evolution of the foam is curvature-driven: An n-sided cell shrinks or
expands at a rate proportional to 6-n (von Neumann's law). This is simply the
generalisation to a a bubble in the foam, with n neighbours, of Laplace's law for
an isolated bubble L\p = 2y/R (y is the tension of the soap film interface, L\p is
pressure excess of the gas inside the bubble of radius R). See Ch.VI and XXIT.
Ferrofluid foams (which are in a stable state of equilibriun and do not coarsen),
discussed in Ch.IX, are an interesting exception to this general rule.

3. Structural equations of state (Lewis, Aboav-Weaire and Peshkin)

3.1. CONSTRAINTS

Each cell in the foam is characterized by the number n of its neighbours. The
random variable n has a distribution Pn over the C cells in the tissue. This
distribution is constrained by the following inevitable relations

LkPk =1 normalisation
LkPkk =<k>=6 topology
Lk Pk A 1(k,n) =n topological correlations.
LkPkAk = AtotiC space-filling (1)

where Ak ;::: 0 is the average area of k-sided cells, Atot is the area covered by the
foam, and Pk At(k,n);::: 0 is the number ofk-sided cells neighbouring any given
n-sided cell. The second constraint is a consequence of Euler's relation. The
third states that an n-sided cell has n-neighbours.
For the foam to remain in a steady state, its structure must be a random
distribution of cells (any long-range order would be visibly offset by a local
transformation) maximizing the entropy which is unchanged by further trans-
formations. The most probable distribution (maximizing the entropy) is not only
108

overwhelmingly so (it can be achieved by the largest number of microscopic


configurations), but it is also robust under the local elementary topological
transformations. ETI are also responsible for the plasticity and the coarsening
of the foam,.

3.2. MAXIMUM ENTROPY INFERENCE


Compressing the largest amount of words
into the smallest amount of thought
(Churchill)

Statistical equilibrium is the thermodynamic state of maximum entropy. Entro-


py is decreased by the imposition of a physical constraint. Conversely, removal
of a constraint increases the entropy: the over-all structure is more likely,
because it can be realized by many more local microscopic configurations. This
result is achieved by forcing one constraint ( 1) to duplicate another, thereby
making it redundant [4,5]. The duplication condition is a structural relation for
foams in statistical equilibrium (equation of state in thermodynamics). The main
structural equations of state, Lewis's (2) and Aboav-Weaire's (4), have been
discovered empirically [6,7]. They are diagnostic: A foam which does not obey
them is restricted by other forces. Furthermore, how they are violated may
reveal the nature of these forces. Lewis's law indicates how topological space is
most likely to be filled by cells. Aboav's gives the most probable correlations
between neighbouring cells.
If the average area of k-sided cells, Ak is linear in k,

Ak =(AtotiC) A. [k-(6-1/A.)] , Lewis's law (2)

then the fourth constraint ( 1) is a linear combination of the first two and dupli-
cates them [4] (with some characteristic parameter A.). The foam has used the
arbitrariness in the functional form of the relation Ak =A(k) to increase further
its entropy. Lewis's law, a relation between size and shape of cells, has been
observed throughout the biological world [8,9].
The same redundancy argument gives the functional dependence for topolo-
gical correlations (third constraints (1)). At(k,n) is linear ink, and thus inn, by
symmetry, and nm(n), the total, average number of sides of all neighbours to a
n-sided cell, is linear inn [5],

At(k,n) = (k-6)crt(n-6) + (n+k-6) Peshkin's law (3)


nm(n) =Lk Pk kAt(k,n) =(6+0"tJ.12)n + 6J.12(116-o"I). Aboav's law (4)
109
Smaller cells surrounding larger cells, and vice-versa, impose crt<l/6. In nature,
<1]<0; Slope 5 = (6+<11Jl2) for Aboav's law follows from local balance under T2
or mitosis [2,10]. nm(n) is linear, experimentally [9] and in simulations [5,11].
It has been established over the last fifteen years that foams are in statistical
equilibrium, described by a state of maximum entropy restricted by constraints,
linear relations between the p 0 • There is abundant indirect evidence for statis-
tical equilibrium and the fact that all local configurations are actually present
and accessible in foams. The best direct evidence is given by:
(i) The state of human epidermis, which is structurally and physically
invariant for eighty years, whereas every individual cell undergoes a
topological transformation (it divides or dies) every week [12].
(ii) The Voronoi foam produced in Rennes by disks on an air table [13].
(iii) Simulations [11,14,15].

3.3. CORRELATIONS BEYOND NEAREST NEIGHBOURS


The English never draw a line without blurring it
(Churchill)

Figure 2. Example of the shell structure around a germ cell (in white).
Layers 1-5 are shaded. Defect inclusions are outlined.

The foam can be structured into concentric shells around one arbitrary central
cell [16]. (Fig.2). The central cell constitutes layer 0. Each cell in layer j has
neighbours in layers j-1, j and j+ 1, except for defect inclusions (in bold in
Fig.2), which only have neighbours in layers j and j-1. 3-sided cells are always
defect inclusions. Notice that the outer boundary of layer j is meandering, as if
110

disorder induced a negative curvature in the systeml. Defect inclusions fill in


the meanders and are inevitable in a uniform foam on a flat substrate.
Consider a shell-structured foam with N cells, out of which N(n) have n sides.
There are Cj(n,k) couples of n-cells and k-cells separated by j-1 layers. Define
the correlator Aj(k,n) = Aj(n,k) = Cj(n,k)N/[N(n)N(k)] ~ as the average number
of of pairs of k- and n-cells at separated by j-1 layers, given that the two cells
exist. PkAj(k,n) = Cj(n,k)/N(n). Let Kj(n) = Lk Pk Aj(k, n) be the number of
cells in the jth layer of neighbours to a central, n-sided cell. K 1(n) = n.
For first neighbours, maximum entropy (redundancy of the constraints (1))
infers (2) that A 1(k,n) is linear in k and in n, by symmetry. Thus, K 1(n) is,
indeed, linear in n. This is completely general.
Suppose now that the system does not contain any defect inclusions. The
number K2(n) of cells in the second layer can be expressed, through A 1(k,n), as
a map from layer 1 to layer 2 [11],

Lk~4 Pk (k-4) A1(k,n) = K2(n). (5)

Linearity of A 1(k,n) in n implies that K2(n) = a2n + ~2 is linear in n. But,

(6)

is also a constraint on Pk• additional to the set (1), and compatible with it (the
right-hand sides are all linear in n). Maximum entropy (redundancy of the
constraints) infers that A2(k,n) is linear ink, and thus inn, by symmetry,

A2(k,n) = (k-6) 02 (n-6) +(J.12<11 + 2)(k+n-6) + J.12(1-6o1). (7)

The map can be iterated:

L~4 Pk (k-4) Aj(k,n)- Kj-1(n) = Kj+1(n) =<Xj+1(n-6) +~j+1• (8)

is also linear inn, etc. [11]. It is the shell map, described inCh. XXIX, building
the froth layer by layer, from the central n-sided cell outwards. Thus, the Aj(k,n)
are all linear ink and n. Sum rules on the Oj, involving J.12 yield <Xj independent
of j and ~j = 6j+O(J.12.i2). The term O(J.12.i2) is the negative curvature imparted by
disorder; it is cancelled out by the contribution of defect inclusions. Both contri-
butions are independent of the central cell. (The first term, <Xj+ 1(n-6), is the

1 The ratio of the perimeter to the radius of the boundary is a measure of the curvature
of the substrate (theorem of Gauss-Bonnet) [17]. The perimeter P and the radius R of a
circle drawn on a surface of revolution of gaussian curvature K, are related by the
celebrated formula (Gauss), P =[1t/.V(-K)]{ exp[R.V(-K)] - exp[-R.V(-K)]}.
111
cuvature imparted to the whole froth by the central topological charge. It is
screened by other cells, but independently of n).

3.4. TOPOLOGICAL GAUSS THEOREM

The map (8) for the number of cells in successive layers is simply the differential
form of a topological analogue to Gauss's theorem in electrostatics [3,11].
Since <n> =6, each cell of a flat (neutral) foam carries a topological charge
Q = 6-n. Q is the local curvature that the cell imparts to the froth. Consider a
cluster of cells G, bounded by a contour oG with v+ vertices with an edge
outside G, and v- vertices with an edge within G. The total charge of the cluster
is LG Q. The "flux" of edges across the contour aG is v+-v-. A topological
Gauss theorem is obtained, almost by inspection.

(9)

This expression can be "diffferentiated", by comparing two clusters differing by


one layer of cells. The cluster contains cells at distance ~j from an arbitrary, n-
sided central cell. One obtains a relation between the v± of two successive
contours, one layer apart, and V/ is the number of cells Kj+ 1(n) in layer j+ 1,
including defects. Obviously, the number of defect inclusions is independent of
n, at least beyond a few layers; defects do not affect the linearity of Kj(n).

4. Disorder control. Lemaitre equation of state


We demand rigidly defined areas of doubt and uncertainty
(D. Adams, The Hitchhiker's Guide to the Galaxy)

Foams may be disordered, but they are certainly not disorganised. Disorder is
measured by the thermodynamic variable Jl2 = Ln Pn (n-6)2. It manifests itself
locally by Gaussian curvature fluctuations, i.e. by the presence of non-
hexagonal cells or by their generation. The average curvature is zero for a foam
on a plane, because <n> =6 (a statement of the theorem of Gauss-Bonnet).
If left uncontrolled, these fluctuations destroy statistical equilibrium in the
foam. For examples:
(i) Systematic decoration of vertices (inverse T2) leads to a fractal Sierpinski
foam [18] with Jl2 = oo.
ii) Non-hexagonal cells are produced by cellular division. If our cells were to
divide one after the other in a row or on a circle, our skin would develop long
scars of softer tissue or pimples [19], and disorder would be concentrated on
isolated defects, visible and felt far away, instead of being spread out
throughout the tissue.
112

iii) A foam constructed layer-wise from a central cell [16] has, as we have
seen, an effective negative curvature due to disorder [3]. To keep the foam
uniform on a plane, defects (outlined in bold line in Fig.2) are intercalated.
Disorder is controlled by the topological energy of the foam.l. Cells have
correlation and self energies [2, 19].

U(Jl2) = E Ln,k Pn Pk (n-6) A1(n,k) (k-6) + 11 Ln Pn (n-6) 2 (£,11>0)

= (EGtJl2 + 11) Jl2 (10)

(The first term is the average correlation energy of two neighbouring cells with
topological charges (6-n) and (6-k). The second is the average self-energy of the
cell). The correlation energy is highest for uncorrelated foams (Gt=1/6) and
vanishes for hexagonal foams (Jl2=0). The self-energy, proportional to the
fluctuations of the Gaussian curvature, makes U positive, thus Tt>aE.
We will maximize the entropy, keeping the energy U, i.e. Jl2 constant. This
is an additional constraint besides system (1). The new constraint cannot be
made redundant since it is quadratic in n. It is therefore an additional restriction,
but to the dynamics rather than to the geometry. The result is a relation
(equation of state) between the thermodynamic variables Jl2 and P6. found by
Lemaitre et al. [20]. (Fig.3).
Lemaitre et al. noticed that this relation between Jl2 and P6 = 1-Ln~ Pn was
universal in two-dimensional foams. Now, Jl2 and 1-p6, even moments in (n-6)
of the distribution Pn. are both measures ("order parameters") of disorder. Uni-
versality of the relation Jl2(P6) may be thought, at first sight, to express their
equivalence. This is not so: Lemaitre's law is the virial equation of state for
foams. It is non-linear (non-trivial virial expansion). In a gas, non-triviality re-
flects specific interactions between the atoms or molecules. In foams, the full
equation of state, is universal for 0.34<p~1 (Fig.3).
For P6>0.67, the foam has only 5-,6- and 7-sided cells, with (1-p6)/2 = P5 =
p7, thus Jl2 = 1-p6; it has trivial virial expansion and corresponds to an ideal
gas.

2physically, the energy in a foam is carried by the interfaces E (surface tension). There
are three (topological) contributions:
1) Total interfacial length.
2) Topological curvature: An n-sided cell is a source of curvature, of charge (6-n) in an
elastic medium.
3) Topological correlations: A pair of neighbouring, opposite charges,is a dislocation, a
defect costing less energy.
We also require that the topological energy is always positive and only vanishes for a
hexagonal lattice (no topological defects).
113

2,0

1,5

J.12
1,0

0,5

0' 0 +-"""T"--,-..--""T"".......-.--"T""""""T'"__,._.,.
0 ,0 0,2 0,4 0,6 0,8 1 ,0
p6

Figure 3. Lemaitre law: J.1.2 as a function of P6·


Full line: Maxent theory (Mayer coefficients (14)) [21].
Squares: Experimental and numerical data [20].

5. Lemaitre's law as the virial equation of state for foams


Ha, thou mountain foreigner!- Sir Jacques ...
(Shakespeare, The Merry Wives of Windsor)

The following procedure is standard maximum entropy inference [21]. One ma-
ximises the entropy

(where qn is the a priori probability of having an n-sided cell, unknown so far),


restricted by the independent constraints. The latter fall into two classes and are
either even or odd in (n-6).
(i) Inevitable, topological
114

LnPn = 1 (even)
I.n Pn (n-6) = 0 (odd) (1)

(ii) Metric. The foam is at constant energy if the variance J.l2 of the
distribution is constrained.

Ln Pn (n-6) 2 = J.l2 (even) (10)

The entropy is maximised by the Boltzmann distribution

(11)

where ~and 'Yare Lagrange multipliers enforcing the constraints. (<In includes
the normalization factor.)
Since we are interested in J.l2 and P6· i.e. in the even moments, let us
introduce a fugacity z = exp(-'Y) and define an even moment ~enerating function

00 2
Z= ~a zs =(1-p6)/ (12)
£..J s /P6'
s=1

with s = ln-61 and coefficients as =[q6+sexp(-~s) + q6-sexp(~s)]/q6 > 0,


dependent on the prior probabilities qn. This is known in statistical physics as a
Mayer series (for p = NN in a gas) in the fugacity z. ll211P6 is given by a second
Mayer series, derived from the first (PikT for a gas)

(13)

Reversing the first Mayer series3 and putting it into the second yields the virial
expansion J.L2IP6 as a series in (1-p6)1p6 (PikT as a series in p for a gas).
For small z, ( 1-p6)1P6 = a 1z = J.L 2/p6. the virial expansion is limited to the
first term (ideal gas law PlkT = p) regardless of the Mayer coefficients as or the
priors qn. At higher z, the Mayer series involves several as and the virial expan-
sion should accordingly also involve the priors. It does not. This is the universa-
lity found by Lemat"tre et al ..

3A series y(x) (y as an expansion in powers of x) is reversed, x(y), when it is expressed


as x in powers of y [22].
115
6. Universality of Lemaitre's law

How can universality result from series whose coefficients a5 are, at first sight,
specific functions of the prior probabilities qn? The dependence on n of the
priors qn is unknown and arbitrary (your guess is as good as mine), except for
two, essential items of information:
(i) The as deal with even moments in (n-6). Thus, a1 bins together priors q5
and q7 , etc., and the odd moment constraint <n>=6 is considered as prior
information in this context.
(ii) n ~ 3.
We can now, for the third time in this chapter, use maxent inference. (We
used it to make constraints redundant and in the derivation of the Boltzmann
distribution). Ignorance is a symmetry (whether one takes your guess or mine
doesn't affect the physical relation between 112 and P6· or any other macroscopic
variable), so that, in the absence of any further information, every coefficient as
should have equal weight a priori, except for its multiplicity (degeneracy):

a1 = [q7exp(-p) + qsexp(p)]/q6 =2
a2 = [qgexp(-2P) + q4exp(2p)]/q6 =2
a3 = [qgexp(-3p) + qsexp(3p)]/q6 =2
~ = qwexp(-4P) lq6 =as= ... =1 (14)

(In the statistical mechanics of gases, the phase space (of the mechanical state
(x,p) of an atom) is partitioned into equal volumes (dx.dp)3 = h3, all equally
probable a priori, according to Gibbs's postulate).
The generating function Z would be a Gaussian sum (Jacobi's theta function)
2
L,zs '
except that as~4 = 1 instead of 2 (because cells have at least 3 sides, the prior
information (ii)). Contributions of s ~ 4 to the sum are utterly negligible for z
small, y large Thus,

for 0 < z < 0.7 (to O(z16)).


The Poisson summation formula (or Jacobi's imaginary transformation
(Abramowitz and Stegun, 16.38.5-6))
116

(15)

gives an analytic expression for the Gaussian sums (the second is the derivative
by (1t't) of the first), now valid for z large, y = 1tt small,

(16)

to 0(exp(-n2fy) ), i.e. for 0.25 < z < 1.


In the common range 0.25 < z < 0.7, we obtain a simple, analytic
expression,

(17)

which had already been guessed (as f.l2P62 = 0.15) by Le Caer and Delannay
[23]. For z < 0.25, the Gaussian sum reduces to the first term (to 0(z4)) and one
has the trivial virial equation (P/k:T = p).
In summary, here is the universal Lemat"tre's law

f.l2 == 1-p6 0.66<p6< 1


f.l2 P62 = 1/(21t) 0.34 < P6 < 0.66 (18)

Note the universality (1121t), the range and the non-triviality. Known foams
departing from the universal curve (Feynman diagrams, random fragmentation,
Poisson random Tl, Sierpinski foams [24-26,5,1], all artificially generated) have
P6 < 0.22, large f.l2 and dominant contribution of cells with very large n (~ 10)
which are unlikely to be equiprobable a priori. Lemaitre's law does indeed
control disorder, in a universal fashion.
A final remark. The distribution Pn is not yet fully determined, and not a
Gaussian in general. By concentrating on the even moments, we have obtained
ps+P7, P4+P8 and P3+P9 binned together. Differences between P5 and P7, etc.,
are free, restricted only by the odd moment constraint <n> = 6. The distribution
has two adjustable parameters, one of which can be its third moment [15]. These
two adjustable parameters are not universal. They vary from foam to foam.
We have used here maximum ignorance, not as an excuse, but to obtain a
scientific result. Maximum ignorance is a state of equilibrium, a symmetry,
independant of local reshuffling,unknown facts, unstated assumptions or out-
side interference. This equilibrium is not frustrated: unlike knowledge ("To be
117
or not to be ... "). A system in statistical equilibrium (the state of maximum
entropy) is able to respond locally and efficiently to any outside disturbance
offsetting this equilibrium, like the epidermis to a wound.

Jl2*p6A2


0,15

+
+
+
0,10 +
+
+
+
+
+
0,05 +
+
+
+
+
+
+
+
0,00
0,0 0,2 0,4 0,6 0,8 1,0
p6
Figure 4. Lemrutre law: J..l2 P62 as a function of P6· •: Maxent theory (as:53=2, as2!4=1).
-- :Jacobi theta function (as= 2). +:Trivial virial expansion. Lemrutre law (virial
equation of state) is given accurately by the Jacobi theta function for P6>0.34. It is
trivial (ideal gas) for P6>0.75. The Mayer series have been computed through s=IO.

7. Foaming
He is what you might call an avant-gardener. He believes in randomness.
His yard is a wilderness of weeds and [Kerroc] ... and cabbages...
(D. Lodge, Changing Places)

7.1. THE FOAMING TRANSffiON

In a foam, two phases, gas (bubbles) and (interfacial) liquid coexist as distinct
entities. Foaming is the transition whereby an homogeneous mixture of gas and
118
liquid segregate. This is a first-order phase transition (except at the critical
point) from the one-phase region to the two segregated phases (Fig.5). For
example, blood is an homogeneous mixture of a complex fluid and air, in
thermodynamic equilibrium at a given pressure. If the pressure is suddenly
lowered, blood enters the two-phases region of the phase diagram and boils
(bubbles are formed). To prevent it, deep-sea divers must go through
successive depressurisation stages.
The free energy F has, as a function of the density cl>, one minimum (at the
average density <cl>> of the homogeneous mixture) in the single-phase region,
and two minima separated by a hump in the two-phase region of the phase
diagram (Fig. 5, inset).

-1
f
I
\ I
I
\ I
I
\ I
\ I

F/k.T

V~
Figure 5. Schematic pahase diagram of a liquid foam.
Inset: Free energy F(<l>)/k:T. Dashed line: spinodal line.

Consider a mixture of a concentration c 1 of gas and c2 of liquid. The gas by


itself has density ci>1 and free energy F 1; the liquid, on its own, ci>2 and F2. The
free energy of the mixture is F(<cl>>) if it is homogeneous, Fsegr = ctFt + c2F2
if it is segregated. Its density is <cl>> = CJ ci>1 + c2 ci>2 (whether homogeneous
or segregated). Thus, the homogeneous mixture is stable (F(<cl>>)<Fsegr)
wherever the function F(cl>) is convex; it is unstable (Fsegr<F(<cl>>) in the region
of the hump ofF( ci>).
119

The destabilisation of the homogeneous mixture can be through segregation


(as in foaming or bubbling, where ci> = cl>1 (bubble) or ci>2 (interface) in different
regions of space) or through spinodal decomposition (whereby ci> oscillates in
space about <cl>>). Segregation occurs in the region of the phase diagram where
the curvature (()2Ft()cf>2) of F(cl>) is positive, i.e. if dPidcl> > 0. Where the
curvature of F(cl>) is negative (dP/dcl> < 0, the mixture is mechanically unstable
and cannot exist even as supersaturated), destabilisation takes place as density
waves (spinodal decomposition).4 The spinodal line dPidcl> = 0 (dashed line in
Fig.5, ill-identified experimentally or in simulations) separates the two regions.
Foaming is a segregation mode. The energy of a bubble of radius R in the
supersaturated fluid has two contributions.
i) Ev =-a R3, negative, proportional to the volume of the bubble, with a a
measure of the supersaturation (the distance into the two-phase region of the
phase diagram).
ii) Es = y R2, positive, proportional to the surface of the bubble, is the energy
necessary to create an interface between liquid and gas regions, with y, the
tension of the interface.
The bubble has maximum energy for a critical radius Rc = 2y/a. If the
nucleated bubble has a radius R<Rc, energy is reduced by its shrinking. If its
radius is supercritical (R>Rc), the bubble expands until it touches other
expanding bubbles. This argument has first been proposed bv Griffifth for the
propagation of cracks in materials under stress. The tougher the material,
the longer the critical crack length~-
Supercritical bubbles are nucleated at random positions and time. They grow
uninhibited until they touch. Then, if growth stops upon contact, the dry foam
has fixed interfaces which are arcs of hyperbola (in 20) or hyperboloids (in 3D)
(Johnson-Mehl [27]). On the other hand, if space is partitioned between two
bubbles by their radical axis, the interfaces are straight and the dry foam
continues to coarsen after contact, with the larger bubble growing at the expense
of the smaller one (Laguerre-Telley [28]). Thus, foaming can be viewed as the
establishement of contacts between bubbles, or of edges of a network dual to the
foam, through the dynamics of its vertices [29]. Foaming is the simplest model
for the dynamics of a first-order phase transition, of the establishment in real
time of a network of contacts.
Foaming requires supercritical bubbles. This is done by reducing the critical
radius, i.e. by decreasing interface tension or increasing the level of supersatu-
ration. Examples:

4This is easily seen in 10: The Euler-Lagrange equation minimizing the free energy is
(d2/dx2)«1> = oF/o«<>, i.e. Newton's equation for the "motion" (in x) of a "particle" (at«<>)
in a "potential" -F (inverted). One distinguishes clearly the profile «<>(x) of spinodal
oscillations from the soliton-like bubbles, depending on the value of the density «1>(0) at
the point x=O ("initial position of the particle").
120

1) Add surfactant (this reduces the interfacial tension) and shake or expand.
2) Internal source of gas (AlN in Kerroc).
3) Cell division (bubble proliferation) as in the epidermis [12].
4) Charged polymers (cheese), in a solvent with free ions (an acid, e.g. wine)
to screen the charges on the polymer chain and make it flexible. This forms (at
normal pressure and moderate temperatures) a stable, homogeneous mixture
(fondue). Just before serving, add a weak alkali (sodium bicarbonate) to coun-
ter the free ions of the acid; the polymer chains, now unscreened, stretch, the
fondue expands and foams (see also [30]).
The fmal morphology of the foam depends only on a few thermodynamic
parameters (temperature, time of expansion, pH). Foaming is a standard phase
transition, occurring in real time.

8. Thermodynamics of solid foams

Closed

Open

I
I

Time
Figure 6 . Phase diagram ofthe expanded foam Kerroc (N. Pittet [29])

A range of solid foams Kerroc® is produced by the firm Cemix in Thorigny-


Fouillard, near Rennes. They are expanded glass foams, made of crushed
(bottle) glass, mixed with AlN powder (the foaming agent). The mixture is then
121

taken into an oven for an isothermal expansion stage, at a temperature T, for a


duration t (sometimes preceded with a pre-expansion stage). The fmal structure
(a light, solid foam) and its porosity depend essentially on the two parameters T
and t, so that it is a genuine thermodynamic state [31,32]. A schematic phase
diagram of Kerroc is shown in Fig.6 reproduced from ChJCXXIII. The electron
micrographs of Kerroc have been taken by S. Graf, from samples supplied by Y.
Laurent (Cemix).
In practice, the expansion stage lasts for 100 min < t < 1000 min, at a tempe-
rature 700°C < T < 11 00°C. As the duration of the expansion stage increases,
the structure changes from a wet foam with spherical bubbles, (Kugelschaum),
almost monodispoerse at first, then with double porosity (small spherical
bubbles inside the inerfaces between large bubbles), to apolyhedralfoam. The
porosity of the material is closed at the higher temperatures, and open at the
lower temperatures of expansion (and long duration), because the glass is much
more viscous at lower temperatures, and the stretched interface between two
rapidly expanding bubbles may tear.
The foaming reaction is a reduction of Fe3+ in the iron oxyde present in the
glass, by nitrogen N3- from AlN, liberating molecular nitrogen:

for Fe203; the solid foam has a grey colour. If the glass contains Fe2+ instead of
Fe3+, less nitrogen is liberated per AlN molecule, the expansion is slower, and
the solid foam is green.
Mechanically, the solid foam Kerroc is very tough on any scale larger than a
bubble size. Although its constituting material (glass) is fragile, it can hit or fall
on a hard surface without shattering. This is because there is always a bubble to
stop a crack, relax the elastic stress at its tip, and prevent it propagating further.
Solid foam owes its physical toughness not to its constituting material (glass),
but to the empty space inbetween (the bubbles), Moreover, the voids must be
located at random for the material to be strengthened by their presence (compare
solid foam with a perforated sheet of stamps). Kerroc owes its strength to
disorder and voids (essence through non-existence) [29,31].
Solid foams constitute therefore a class of thermodynamic systems, resulting
from a foaming phase transition. Their structural and physical properties depend
on two thermodynamic parameters, the temperature and duration of the expan-
sion process, and only weakly on the chemical nature of the glass and of the
foaming agent.
122

8.1. THE LIQUID-PASTE TRANSffiON IN AN EXPANDING FOAM

The viscosity of glass decreases very rapidly with increasing temperature, above
the glass transition temperature Tg. which conventionally corresponds to a vis-
cosity of 1013 poise (1012 Pa.s, i.e a relaxation time of several weeks). In fact,
there is a transition at Ttp. below which the molten glass behaves as a paste
(viscoelastic material), and above, as a viscous fluid. If a paste is extended
rapidly, it breaks, like chewing gum. Thus, within the time-scale set by the
expansion process, the film separating two bubbles in the expanding foam can
tear at lower temperatures and remain intact above T 1P.
Kerroc has open porosity if expansion takes place below Ttp• closed porosity
if it occurs above Ttp· Should expansion stop before physical contacts between
bubbles are established, the foam is spherical with closed porosity, regardless of
the temperature. (Fig.6). For Fe3+ Kerroc (grey), the liquid-paste transition
takes place at about Ttp"" 980°C [32,29].
There is more direct thermodynamic evidence for a liquid-paste transition in
molten glass, than open or closed porosity of expanded foam. It comes from two
measurements.
(i) M. Tasserie ([32], fig.3.6) has identified a weak, endothermal peak in
differential scanning calorimetry of Kerroc samples expanded below Ttp·
(ii) The ratio of the specific heats Cp/Cy (and thus of the adiabatic to
isotherrmal bulk elastic moduli EsiET, or s~ds of sound vs/vT), has a
maximum as a function of the temperature [33]. The peak occurs at the
temperature where fast (adiabatic) and slow (isothermal) responses of the
materials differ most. It may be identified with Ttp .
The liquid-paste transition is seen as a knee in the Angell fragility plot [34]
of the (logarithm of the) viscosity as a function ofTg!T (Fig.7). The viscosity
has, qualitatively, the standard Vogel-Fulcher (-WLF) behaviour,

Tl(T) =Tloo exp [-B/(T-To)], (19)

where To is the temperature at which the entropy frozen in a glass-forming


liquid cooled infinitly slowly, would vanish ("Kauzmann's paradox" [35]). The
fragility is the parameter 0 < 9o =TotTg < 1, characteristic of the material. Then,
T 1p = 2Tg- T0 .

Slndirectly (in ZnCl2 and glycerol), since Cy cannot be measured experimentally. What
is measured is the dynamical structure factor S(q,m), and the thermodynamic parameters
are extracted from a fit of the whole spectrum to the hydrodynamic expression resulting
from the inversion of the Navier-Stokes equations.
123

0
~/T 1
Figure 7. Angell plot: Viscosity (log 1111loo) as a function ofTg!T (schematic, with
lloo===l0-4 Pa.s, say). The fragility 9o =TofTg measures the deviation from Arrhenius
viscosity (dashed line). The liquid-paste transition is seen as a knee(*= T giTip).

Thus, solid foam is a fossil record of the foaming transition, with its
structure and porosity determined by the duration and the temperature of the
foaming process.

9. References

1. Weaire, D. and Rivier, N. (1984) Soap, cells and statistics- Random


patterns in two dimensions, Contemp. Phys. 25, 59-99.
2. Rivier, N. (1993) Order and disorder in packings and froths,
in D. Bideau and A. Hansen (eds.), Disorder and Granular Media,
Elsevier, Amsterdam, pp. 55-102.
3. Rivier, N. and Aste, T. (1996) Curvature and frustration in cellular
systems, Phil. Trans. A 354, 2055-2069.
4. Rivier, N. and Lissowski, A. (1982) On the correlation between sizes
and shapes of cells in epithelial mosaics, J. Phys. A 15, L143-148.
5. Peshkin, M.A., Strandburg K.J. and Rivier, N. (1991) Entropic
predictions for cellular networks, Phys. Rev. Letters, 67, 1803-1806.
6. Lewis, F.T. (1928) The correlation between cell division and the shapes
and sizes of prismatic cells in the epidermis of Cucumis, Anat. Rec. 38,
341-376.
124

Lewis, F.T. (1931) A comparison between tne mosaic ot polygons in a


film of artificial emulsion and the pattern of simple epithelium in
surface view (cucumber epidermis and human amnion), Anat. Rec. 50,
235-265.
7. Aboav, D.A. ( 1970) The arrangement of grains in a polycrystal,
Metallogr. 3, 383-390.
8. Smoljaninov,V.V. (1980) Mathematical Models ofTissues, Nauka.
9. Mombach, J.C.M., Vasconcellos, M.A.Z. and de Almeida, R.M.C.
(1990) Arrangement of cells in vegetable tissues, J. Phys. D, 23 600.
10. Rivier, N. (1993) Geometry and evolution of biological tissues, Mater.
Sci. Forum 123-125, 363-392. [In K.-H. Anthony and H.-J. Wagner
(eds.), Continuum Models of Discrete Systems, Trans. Tech. Publ.]
11. Dubertret, B., Rivier, N. and Peshkin, M.A. (1998) Long-range correla-
tions in two-dimensional foams, J. Phys. A: Math. Gen. 31, 879-900.
12. Dubertret, B. and Rivier, N. (1997) The renewal of the epidermis: A
topological mechanism, Biophys. J. 73, 38-44.
13. Lemaitre, J., Troadec, J.-P., Gervois, A. and Bideau, D. (1991)
Experimental study of densification of disc assemblies, Europhys. Lett.
14, 77-83.
14. Ohlenbusch, H.M., Aste, T., Dubertret, B. and Rivier, N. (1998)
The topological structure of 2D disordered cellular systems,
Eur. Phys. J. B2, 211-220.
15. Dubertret, B., Aste, T., Ohlenbusch, H. and Rivier, N. (1998) Two-
dimensional froths and the dynamics of biological tissues,
Phys. Rev. E, Nov. 1998.
16. Aste, T., Boose, D. and Rivier, N. (1966) From one bubble to the whole
froth: A dynamical map, Phys. Rev. E 53 6181-6191.
17. Petit, J.-P. (1980) Le Geometricon, Belin.
18. Le Caer, G. and Delannay, R. (1995) Topological models of2D fractal
cellular structures, J. Phys.l France 5, 1417-1429.
19. Rivier, N., Dubertret, B. and Schliecker, G. (1997) The stationary state
of epithelial tissues, in W. Alt, G. Dunn, A. Deutsch (eds.), Dynamics
of Cell and Tissue Motion, Birkhaeuser, Basel, pp. 275-282.
20. Lemaitre, J., Gervois, A., Troadec, J.-P., Rivier, N., Ammi, M.,
Oger, L. and Bideau, D. (1993) Arrangements of cells in Voronoi
tessellations of monosize packings of discs, Phil. Mag. B67, 347-363.
21. Rivier N. (1994) Maximum entropy for random cellular structures, in
P. Grassberger and J.-P. Nadal (eds.), From Statistical Mechanics to
Statistical Inference and Back, Kluwer Academic Publ., pp. 77-93.
22. Whittaker, E.T., and Watson, G.N. (1962) Modern Analysis,
Cambridge University Press, p. 133 and footnote.
125
23. Le Caer, G. and Delannay, R. (1993) Correlations in topological models
of 2D random cellular structures, J. Phys. A: Math. Gen. 26, 3931-3954.
24. Godreche, C., Kostov, I. and Yekutieli, I. (1992) Topological
correlations in cellular structures and planar graph theory,
Phys. Rev. Letters 69, 2674-2677.
25. Delannay, R. and Le Caer, G. (1994) Topological characteristics of 2D
cellular structures generated by fragmentation,
Phys. Rev. Letters 73, 1553-1556.
26. Fortes, M.A. ( 1995) Applicability of the Lewis and Aboav-Weaire law
to 2D and 3D cellular structures based on Poisson partitions,
J. Phys. A: Math. Gen. 28, 1055-1068.
27. Meijering, J.L. (1953) Interface area, edge length and number of
vertices in crystal aggregates with random nucleation,
Philips Res. Reports, 8, 270-290.
28. Telley, H., Liebling, Th.M. and Mocellin, A. (1996) The Laguerre
model of grain growth in two dimensions. I: Cellular structures viewed
as dynamical Laguerre tessellations; II: Examples of coarsening
simulations, Phil. Mag.B 73,395-408; 409-427.
29. Pittet, N. (1998) Simulation of the foaming process, Foams and
Emulsions, Ch.XXXIII.
30. Lugeon, M. (1950) La recette de La fondue vaudoise, Lausanne, Ed. du
Capricome.
31. Pittet, N., Rivier, N., Laurent, Y. and Troadec, J.P. (1999)
Thermodynamics of solid foams, in preparation.
32. Tasserie, M. (1991) Optimisation physicochimique d'un materiau
expanse, PhD Thesis, Univ. de Rennes I.
33. Grimsditch, M. and Rivier, N. (1991) Anomaly in CpiCv: A possible
signature of the liquid-glass transition, Appl. Phys. Lett. 58, 2345-2347.
34. Angell, C.A. (1995), Formation of glasses from liquids and
biopolymers, Science 267, 1924-1935.
35. Rivier, N. (1987) Continuous random networks. From graphs to glasses,
Adv. Phys. 36, 95-134.
126
POLYGONAL NETWORKS RESULTING FROM DEWETTING

U. THIELE, M. MERTIG AND W. POMPE


Institut fiir Werkstoffwissenschaft, TU Dresden,
Hallwachsstr. 3, D-01069 Dresden, Germany
AND
H. WENDROCK
Institut fiir Festkorper- und Werkstofforschung Dresden,
P.O.Box 16, D-01171 Dresden, Germany

The occurence of polygonal structures is widespread in nature [1]. Ex-


tensive investigations on the statistics of two-dimensional networks have
been performed for biological tissues [2, 3], clusters of metal grains [4, 5],
systems of soap bubbles [6, 7, 8], emulsion lattices [9], gas bubbles in Lang-
muir monolayers [10], magnetic froth [11] or convective patterns in hydrody-
namics [12, 13, 14]. The strong similarity between structure and evolution
of two-dimensional soap froth and grain boundary networks has become a
subject of growing interest [15, 16, 17, 6]. These similarities make it dif-
ficult to differentiate the networks occuring in different experimental sys-
tems. Additionally, one faces a problem if the system is two-dimensional
because it is a planar cut of a three-dimensional systems (grain bound-
ary network) or through putting the three-dimensional structure between
two narrowly spaced glass plates (soap froth, emulsion lattice, magnetic
froth). Here, we will introduce two new experimental systems representing
dewetting processes of a thin liquid films on a solid substrates. The oc-
curing polygonal networks are intrinsically two-dimensional. After a short
introduction of the concepts of wetting and dewetting, the dewetting exper-
iments of polystyrene on silicon and of collagen solution on highly oriented
polygraphite are explained. The different stages of the dewetting process
will be discussed at these examples. Main features of the resulting struc-
tures are analysed by means of stochastic geometry of polygonal networks.
The resulting distributions are compared with distributions obtained for
two-dimensional soap froth. Typical differences between dewetting patterns
and soap froth and between the two dewetting patterns are explained by
distinct driving forces behind structure formation.
127
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 127-136.
© 1999 Kluwer Academic Publishers.
128

Putting a macroscopic drop of liquid on a surface one can observe two


different scenarios. Macroscopic means here bigger than the range of long
range molecular interactions like Van-der-Waals forces but small enough
that effects of gravity need not be taken into account 1 . The drop of liquid
can spread until a thin film covers the whole surface. The liquid wets the
surface. The case is called complete wetting. On the other hand the drop
can remain as a small spherical cap. The liquid does not wet the surface.
This case is called partial wetting.
In both cases the driving force is the minimization of surface energy.
The final state is characterized by the equilibrium contact angle (}E between
the liquid-gas and the liquid-substrate interface at the three-phase contact
line. It can be calculated from the surface tensions with the help of the
Youngs-relation:
/SG = /SL + /LG cos (}E (1)
where /SG, 'YSL and 'YLG denote the surface tensions of the solid-gas, solid-
liquid and liquid-gas interface respectively. For /SG- 'YSL > /LG the liquid
wets the substrat.
Now, consider the situation, where a thin film of fluid is placed on the
surface. This may be realized by spin-casting, floating or painting. What
will happen? If the liquid wets the substrate the film remains. But taking a
liquid that does not wet the substrate the film tries to reach its equilibrium
configuration, i.e. it tries to form a single drop. A thicker film will retract
at the borders in order to reach this state. If, however, the film is very thin
(when it is brought onto the substrate by spin casting at high frequencies),
it does not only draw back at the borders but also breaks up at many
locations 'within' the film. In the course of this process, inner edges appear
that also draw back. Or in other words, holes appear that grow with time.
This process is called dewetting. We will discuss here patterns resulting
from this process. (For an introduction into the subject see [18, 19, 20].)
Before the stages of the process are explained in more detail, the exper-
imental systems are shortly introduced. Experiments on polystyrene (PS)
films on silicon are described in [21, 22]. Thin films (thickness 20-300nm)
are produced at room temperature by spin casting a solution of PS in toluen
onto the silicon. Toluen evaporates and a smooth thin film of glassy PS is
formed. When one brings the sample above the glass transition tempera-
ture, PS becomes a liquid at once and dewets from silicon. The process
is observed time resolved with an optical microscope. All holes arise at
nearly the same time at random distributed spots on the substrate[22].
The mechanism of formation of initial holes is still controversial. It may
be a spontaneous instability of the thin film under the influence of long

1 The latter is the case for drops smaller than the capillary length.
129
range molecular forces [21, 23], or heterogeneous nucleation caused by de-
fects [22, 24]. In the next stage of the process the holes grow till they meet
each other. At first, neighbouring holes touch leaving a thin liquid bridge
inbetween. The liquid bridge between the holes can either rupture, leading
to hole coalescence, or remain stable. If the bridge remains stable, a thin
rim of liquid is formed between the holes. The rims form the edges of a
two-dimensional polygonal network resembling at the first sight to a two-
dimensional aged soap froth. The diameters of the polygonal cells are in
the range of 10-100J.Lm (see Fig.1d).
But this is only a transient state. On a longer timescale the liquid rims
are not stable. They may decay into rows of drops via a Rayleigh instability.
We have investigated acedic collagen solution (CS) that is spin-casted
on highly oriented polygraphite (HOPG) [25]. After the deposition of the
film (thickness 10-15J.Lm) the solvent begins to evaporate. Evaporation con-
tinues during all the process. In competition with evaporation, hole nucle-
ation sets in. The holes grow, meet and form a polygonal network as in
the PS experiment. But in contrast with PS films, the nucleation of holes
continues during all stages of the process. The evolution of the structure
only stops when all the solvent is evaporated. The pattern resulting from
dewetting is fixed in the dried collagen and can be imaged by scanning
force microscopy (cell diameter below 1J.Lm ). The rate of evaporation and
therefore the observed stage of the dewetting process can be controlled by
humidity. In Fig.1a-c three stages of this process are shown. In order to get
this series of images, different collagen concentrations and different humidi-
ties are used. The edges of the polygonal network are stable, because they
have approximatively the same size as the collagen molecules (relatively
rigid rods of 300nm length). Thus coalescence of holes and the evolution
of the rims into drops are supressed. At this point it should be mentioned
that both systems can show a dynamical instability of the moving liquid
rim during hole growth: Liquid rims may lag behind the moving circular
rim (so called back-fingering). In the case of the CS, this gives a transition
between network-like pattern and tree-like patterns. Here, we restrict our
attention to networks.
In order to characterize the network structures that form an intermedi-
ate state of the PS film and the final state for the collagen film evolution, we
use methods of stochastic geometry of polygonal networks (SGPN) which
are part of stochastic geometry. These methods were used extensively to
analyse the evolution of 2d soap froths. They allow comparison between
soap froth and dewetting structures. We are investigating the statistical
distributions of network variables, such as number of edges, edge length,
cell area, cell perimeter or angles between the edges. The mean values and
second moments of these distributions give a first characterisation of the
130

Figure 1. Structures obtained by dewetting for two different experimental systems:


collagen solution on highly oriented polygraphite (a-c), and polystyrene on silicon [26]
(d). For the collagen solution three stages of the evolution of holes are illustrated by
final images taken at different experimental conditions. The images a-c show an area of
5 micron x 5 micron. (a) The formation of holes just started. (b) Intermeadiate state
during hole growth. Some holes have touched. (c) Developed polygonal network, final
state of the dewetting process of collagen films. (d) Developed polygonal network for PS
on silicon. Some rims are ruptured, leading to coalescence of pores.

structures. One can further take into account correlations between neigh-
bouring cells (for edge numbers: Aboav- Weaire law) or between different
variables like for example cell area and edge number (Lewis law). Here we
are interested in the distribution functions of the single variables only.
Fig.2a-d show the distributions of edge number, edge angle, cell perime-
ter and cell area. We show data from two samples of aged soap froth, dewet-
ting network of PS films and of collagen films [27].
With respect to the edge angles we have to remark that the data rep-
131

0.50 r - - - - . - - - - - - . - -- - - - . - --;
0.4 - PolySiyn:n - PolyJtyn:n
---Collagen ---Collagen
- ·-· Soap frolh 0.40 -·- · Soap fro<h
OJ
~ & 0.30
:.0
_g 0.2 ~
£ £ 0.20

0. 1
0.10

0.0 )k::~ :_~_._~-~~~~=---~


2 3 4 5 6 7 8 9 10 II 12 60 120 180
Edge number Edge angle

0.40 r-----~-----~--, 0.3 .----~---~---~--,

- PolySiyren - PolySiyren
- - - Collagen --- eou.1en
-·-· Soap fro<h -·-· Soap fro<h
0.30
0.2
.~ ~
:E :.0

i
0.20
"
.D

£ 0. 1
0.10

0.00
0.0 1.0 2.0 1.0 2.0 3.0
Perimeter Area

Figure 2. Comparison of distributions characterising the polygonal networks occuring in


the dewetting process (polystyrene, collagen) and aged soap froth [27]. (a) Edge number
distribution. (b) Distribution of angles between edges, where the straight lines between
vertices are taken as edges. (c) Distribution of normalised cell perimeter. (d) Distribution
of normalised cell area.

resent not the actual angles under which the edges meet locally at the
vertices, but the angles that are given if one connects neighbouring vertices
by straight lines (resulting from the image analysis procedure). This might
be misleading if one wants to analyse the physics of a soap froth near the
vertices however it gives nevertheless a good measure for comparison of
different experimental systems as it is done here.
The distributions are in general quite similar for the same experimen-
tal system (meaning that they are reproducible), but show characteristic
differences with respect to each other. Differences are more pronounced for
132

edge angle and edge number than for cell perimeter and area. To quantify
these differences one has to study the characteristic variables of the distri-
bution functions. Note first, that the mean values of the distributions are
not relevant for a comparison. Networks with threefold vertices only have
generally a mean value of the edge number (n) = 6 and a mean value of the
edge angle that is 120°. The mean value of the area gives only the length
scale for the experimental system.
However as demonstrated in the following the normalised second mo-
ments are a good measure to compare networks resulting from different
experimental systems. They are defined as follows:

(2)

where f denotes the relevant variable of the distribution ( n edge number, p


cell perimeter, a cell area and wangle between edges), (!),its mean value,
fi, its value for cell i and N the number of cells. Normalisation of metric
variables like perimeter and area is necessary in order to compare different
networks. The normalisation is also used for the edge angles. The second
moment of the edge number distribution f-L2 is given without normalisa-
tion as in the literature. Higher moments are not discussed because of the
restricted number of cells available for the statistical analysis. The second
moments are listed in Tab.l. The second moment of the edge number dis-

TABLE 1. comparison of the second moments of colla-


gen and polystyrene dewetting networks and soap froth

Property Collagen Polystyrene Soap (aged)

1-''2 2-5 ::::::2 :::::: 1.4


I'~ 0.4- 1.5 ::::::0.3 0.4- 1.2
I'~ 0.10- 0.35 0.05-0.10 0.01-0.03
I''{ 0.05-0.07 0.03- 0.05 ::::::0.015

tribution f-L2 that is usually taken as a measure of disorder indicates that


dewetting structures are more disordered than soap froth. Among dewetting
structures, PS patterns are more ordered than the collagen networks. This
can be explained by coalescence of holes occuring during the evolution of
the PS system. Coalescence happens through rupture of the thinnest rims,
found mainly in between holes of very different sizes. Therefore, coalescence
equilibrates the cell sizes. Indeed, f-L2 is considerably smaller in PS than in
cs.
133
By contrast, the evolution process of a soap froth is driven by pressure
differences between neighbouring cells that are equilibrated by diffusion
of gas through the cell walls. Cells with more than six sides grow at the
expense of the cells with less than six sides that shrink (von Neumann's law)
[15, 28]. Therefore, one finds allways small and large cells, i.e. P2 does not
decrease in time in the steady state of an aged froth. Locally, the vertices
stay in equilibrium; three edges meet at angles of 120°.
The evolution of soap froth and dewetting networks are therefore very
different, as reflected by the second moments of the area distributions P2
and the edge angle distribution p,?,.

A Collagen
1.5 • Polystyren
c: +Soap froth
s
0

••
.0
·;::
u;


'5
!tl
~
1.0
!tl
"E


Q)
E
0
E


-c
c: 0.5

0

• ••
(.)
Q)
(/)
A

0.0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Second moment edge angle distribution

Figure 3. Shown is the second moment of the area distribution vs. the second moment of
the distribution of edge angles for networks resulting from dewetting and two-dimensional
soap froth respectively [29].

This is visualised in Fig.3 where P2 is plotted over p,?,. The dewet-


ting pattern and the soap froth occupy different regions of the p,'?, - P2
plane. Moreover dewetting patterns are clearly split into separate PS and
CS regions. The soap froth has a very small p,?,. We will call this vertex
equilibrium. By contrast the PS pattern has a large p,?, but a smallp,2. We
call this edge equilibrium because the system tends to form edges that are
stable on the time scale of the growth of the holes.
134
Because the collagen molecules suppress the coalescence of holes almost
completely, the observed structures result from hole formation only. Neither
edges nor vertices are in local equilibrium. This is reflected by the fact that
both Jl2 and J.L2 are large.
All the statistical distributions are a measure of the initial distribution
of holes. Therefore they should resemble the distributions found for spatial
tesselations generated with the Johnson-Mehl model [30, 31]. In this model
one generates a random distribution of points (Poisson distribution). These
points are activated at timet with a probability p(t). Activated points grow
into circular holes at constant velocity until they touch, ultimately filling
the whole plane by a polygonal network. Let p(t) rv tUJ-l) be the probability
of activating a point at timet. The exponent f3 can be negativ or positive.
For f3 < 1 the activation probability decreases with time, for f3 > 1 it
increases whereas for f3 = 1 one gets a activation homogeneous in time.
A delta function for p( t) activates all points at ones, and one obtains the
Voronoi tesselation. This simultaneous activation can be ruled out, because
the values of the second mo~ents of a Voronoi tesselation are smaller than
the collagen values (Voronoi: J.L2 ~ 1.8, J.L2 ~ 0.3, J.L~ ~ 0.06).
The evolution of the dewetting pattern itself suggests new extensions of
the Johnson-Mehl model to interpret the experimental distributions. The
standart Johnson-Mehl model can only be used for a qualitative compari-
son. For collagen films, it is not possible to extract a value for f3 by compar-
ing model and experiment. The problem is that, experimentally, the rate of
growth of the holes is not the same at all times. It decreases with increasing
viscosity, as the solvent evaporates.
In conclusion dewetting of a thin liquid film is an interesting phenomena
showing two-dimensional polygonal network formation. The direct com-
parison of networks of different physical origin with methods of stochastic
geometry gives means to quantify the differences. It is possible to distin-
guish two local equilibriums: the vertex equilibrium of soap froth and the
edge equilibrium of the coalescing dewetting network.

References
1. D. Weaire and N. Rivier. Soap, cells and statistics - Random patterns in two
dimensions. Contemp. Phys., 25(1):59-99, 1984.
2. F. T. Lewis. The geometry of growth and cell division in columnar parenchyma.
Am. J. Bot., 31:619-29, 1944.
3. J.C.M. Mombach, M.A.Z. Vasconcellos, and R.M.C. de Almeida. Arrangement of
cells in vegetable tissues. J. Phys. D, 23(5):600-6, 1990.
4. D.A. Aboav. The arrangement of cells in a net. iv. Metallography, 18(2):129-47,
1985.
5. D.J. Srolovitz, M.P. Anderson, P.S. Sahni, and G.S. Grest. Computer simulation of
grain growth. II. Grain size distribution, topology, and local dynamics. Acta Met.,
32(5):793-802, 1984.
135

6. J.A. Glazier, M.P. Anderson, and G.S. Grest. Coarsening in the two-dimensional
soap froth and the large-q potts model: a detailed comparison. Phil. Mag. B,
62(6):615-45, 1990.
7. J. Stavans. The evolution of cellular structures. Rep. Prog. Phys., 56(6):733-89,
June 1993.
8. T. Aste, K. Y. Szeto, and W. Y. Tam. Statistical properties and shell analysis in
random cellular structures. Phys. Rev. E, 54(4):5482-92, 1996.
9. D.A. Noever. Statistics of emulsion lattices. Coli. Surf., 62(2):243, 1992.
10. B. Berge, A.J. Simon, and A. Libchaber. Dynamics of gas bubbles in monolayers.
Phys. Rev. A, 41(12):6893-900, 1990.
11. F. Elias, C. Flament, J.-C. Bacri, and S. Neveu. Macro-organized patterns in fer-
rofluid layer: Experimental studies. J. Phys. I France, 7:711-28, 1997.
12. B. Simon and M. Belmedani. Cellular convection in shallow layers of aqueous solu-
tions of sucrose: Lewis law. C. R. Acad. Sci., 319(8):865-71, Oct. 1994.
13. P. Cerisier, S. Rahal, and N. Rivier. Topological correlations in benard-marangoni
convective structures. Phys. Rev. E, 54(3):5086, 1996.
14. U. Thiele and K. Eckert. Stochastic geometry of polygonal networks- an alternative
approach to the hexagon-square-transition in Benard convection. Preprint, 1997.
15. C. S. Smith. Metal Interfaces, p. 65. American Society for Metals, Cleveland, Ohio,
1952.
16. D.A. Aboav. Foam and polycrystal. Metallography, 5:251-63, 1972.
17. M.A. Fortes and A.C. Ferro. Topology and transformations in cellular structures.
Acta Met., 33(9):1697-708, 1985.
18. P.G. de Gennes. Wetting: statistics and dynamics. Rev. Mod. Phys., 57(3):827-63,
1985.
19. S. Dietrich. Wetting phenomena. Phase Transitions and Critical Phenomena,
Vol. 12, p. 1. Academic Press, London, 1988.
20. F. Brochard-Wyart and J. Daillant. Drying of solids wetted by thin liquid films.
Can. J. Phys., 68(9):1084-8, 1989.
21. G. Reiter. Dewetting of thin polymer films. Phys. Rev. Lett., 68(1):75-8, 1992.
22. K. Jacobs. Stabilitat und Dynamik flussiger Polymerfilme. Konstanz, 1997. Phd-
thesis, ISBN 3-930803-10-0.
23. A. Sharma and G. Reiter. Instability of thin polymer films on coated substrates:
Rupture, dewetting and drop formation. J. Coli. Inter/. Sci., 178:383, 1996.
24. K. Jacobs, S. Herminghaus, and G. Schatz. Dominance of defects in thin liquid
polymer film rupture. Preprint, 1997.
25. M. Mertig, U. Thiele, J. Bradt, G. Leibiger, W. Pompe, and H. Wendrock. Scan-
ning force microscopy and geometrical analysis of two-dimensional collagen network
formation. Surf. Int. Anal., 25:514-521, 1997.
26. PS: unpublished image by G. Reiter.
27. Aged soap froth (reanalysed images taken from [1] and [6]; one of them did not
yet reach the scaling state), dewetting network of PS films (analysed images taken
from [32] and Reiter, G., unpublished) and of collagen films (two different samples,
Fig.1c shows part of one of them).
28. James A. Glazier, S. P. Gross, and J. Stavans. Dynamics of two-dimensional soap
froths. Phys. Rev. A, 36(1):306-12, 1987.
29. Same samples as in Fig.2. Additionally analysed are unpublished soap froth images
by J. Glazier, PS images from [23] and images of other collagen samples.
30. J. M!llller. Random Johnson-Mehl tessellations. Adv. Appl. Prob., 24:814, 1992.
31. J. M!llller. Topics in Voronoi and Johnson-Mehl tesselations. In Proc. Seminaire
Europeen de Statstique, Toulouse, 1996.
32. G. Reiter. Mobility of polymers in films thinner than their unperturbed size. Eu-
rophys. Lett., 23(8):579-84, 1993.
136
TWO-DIMENSIONAL MAGNETIC LIQUID FROTH

F. ELIAS (1), C. fLAMENT Ol, J.-C. BACRI Ol, F. GRANER <2l


rn Laboratoire des Milieux Desordonnes et Heterogenes (case
78), 4 place Jussieu, 75252 Paris cedex 05, France
and Groupe ferrofluide, Universite Paris 7 Denis Diderot
(2) Laboratoire de Spectrometrie Physique, Universite Joseph
Fourier, BP 87, 38042 Saint-Martin d'Heres, France

We present a new cellular structure, obtained in a pseudo two-dimensional


(2D) layer of magnetic fluid (MF) submitted to a perpendicular magnetic
field. This froth is analogous to 2D soap froths, because its energy contains
an interfacial contribution. Nevertheless, whereas a soap froth coarsens in
time to minimize the surface of its interface, the 2D MF froth can be stable in
time. Indeed it also has a magnetic energy, which allows the cellular pattern
to reach an equilibrium state. The properties of such a froth in its
equilibrium state are thus driven by competition between the surface energy
and the magnetic energy. The latter depends on the amplitude of the applied
magnetic field, which is a control parameter of this system. An evolution of
the structure is obtained on decreasing the amplitude of the field: the
number of cells decreases so that the pattern turns into a single drop of MF
in zero field. We present here a study of this coarsening with decreasing of
the field. For high amplitudes of the applied field, we have observed out of
equilibrium states of the 2D MF froth: the area of the cells evolves in time,
depending on their number of sides. This behaviour looks like a Von
Neumann one (which drives the time-evolution of 2D soap froths), but with
the opposite sign: 5-sided cells grow in time, 7 -sided cells shrink, whereas 6-
sided cells do not evolve. In section 6, the 2D MF froths are shown to obey
the Aboav and Weaire law which describes the topological interactions
between cells, as all the 2D cellular structures. Finally, we present a potential
application of the 2D MF froth, for the study of topology in 2D cellular
patterns.

1. Two - dimensional instabilities in magnetic fluids

A magnetic fluid (MF), or ferrofluid, is a colloidal suspension of


ferromagnetic grains. The particles are monodomains (diameter roughly 10
nm, inferior to the width of a Bloch wall), and act as permanent magnets of
magnetic moment 1711"" 10 4 JlB, where JlB is the Bohr magneton. They are
small enough to remain in suspension in the carrier liquid by Brownian
motion. To ensure the colloidal stability of the suspension, the attractive
137
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 137-150.
© 1999 Kluwer Academic Publishers.
138
interactions between the particles (van der Waals attractions or magnetic
dipolar interactions) must be balanced. The MF we use are water-based, i.e.
are in a polar solvent. An additional repulsion between particles is
superimposed on the mean of an electrostatic charge at the surface of each
particle. These so-called ionic ferrofluids [1] are synthetized by S. Neveu in
the Laboratoire de Physico-Chimie Inorganique (University Paris 6),
following Pr. R. Massart's method.
If no external magnetic field is applied, the magnetic dipoles are randomly
orientated because of the Brownian motion, thus the MF magnetization
equals zero. When a magnetic field is applied, the dipoles progressively align
along the field direction and the MF magnetization increases. Finally it
saturates above a magnetic field of the order of 1 Tesla, because all the
dipoles are parallel to the field. Therefore the magnetic behaviour of a
ferrofluid is (in spite of its name) well described by the Langevin
paramagnetism.
To observe 2D instabilities, we use a binary mixture made of a ferrofluid
and an immiscible oil of lower density. The mixture is confined in a cell
made of Altuglass©, the thickness of which is small (h = 1 mm) in
comparison with the other dimensions (10 em). The presence of the oil
avoids wetting phenomena of the MF, because a thin film of oil always lies
between the MF and the cell walls. The cell is placed in the horizontal
position and is located between two coils in the Helmholtz configuration, in
order to get a vertical spatially homogeneous magnetic field. The images are
recorded by a CCD camera and digitized in a computer.
The macroscopic instabilities observed in such a geometry are shown in
Fig. 1, where the MF appears in black. The control parameters are Ho, the
amplitude of the applied magnetic field, 4>, the MF volume fraction and h,
the cell thickness. If the external magnetic field is static, one can have a
roughly hexagonal pattern of monodisperse MF bubbles (Fig. 1a), or a stripe
structure (Fig. lb). An analogy between the MF stripe pattern and a 2D
smectic system has been theoretically established and the elastic constants
measured [2]. A cellular pattern, with cell boundaries consisting of MF, is
obtained if the applied magnetic field is alternating, as shown in Fig. 1c. A
review of these experimental patterns is given in ref. [3]. Some properties of
the 2D MF cellular structure are studied in ref. [4], such as the evolution with
the magnetic field and the topological correlations between cells. Such
patterns are also observed in a lot of different physical systems. Let us
mention the thin films of magnetic garnets submitted to a perpendicular
magnetic field. The patterns consist in arrangements of domains of
magnetization parallel to the applied field, and domains whose magnetization
is antiparallel to the field. Bloch walls separate the domains; the short range
attraction between magnetic moments in the Bloch wall acts as an effective
surface tension.These systems are analogous to ferrofluids, because they are
driven by the competition between short range attraction and long range
magnetic dipolar repulsion. In the magnetic garnet cellular structure, the
magnetization of the bubbles is parallel to the field. Such 2D magnetic froths
coarsen on increasing the strength of the field, because all the dipoles tend to
139
align along the field direction[5, 6]. The surface fraction of bubble domains
is not conserved in the coarsening, contrary to 2D MF froths.

~··~ (b)
~ r""'
~·.
em
....
r"

I~ -:7 · ·~I

I ~ ~ '~

@H
~

sta tic @ H alternating


Figure 1. Macro-organized patterns in a ferrofluid layer, under a spatially homogeneous external
magnetic field. The MF appears in black. If the external magnetic field is static, one observes either a
hexagonal array of monodisperse bubbles (a), or parallel stripes of MF and oil (b). A cellular structure is
formed under an alternating external magnetic field (c).

A wide variety of MF froths is observed. Figure 2a shows a so-called wet


froth: the oil cells are circular and there is a large quantity of surrounding
liquid at the vertices. It is also possible to build a dry froth (Figure 2b ), in
which the cells are facetted. The Figure 2c shows an anisotropic froth built
using an additional magnetic field in a direction parallel to the cell: the oil
bubbles become elongated in this additional field direction. Finally, in high
magnetic field, a deformed magnetic froth is obtained as shown in Figure 2d.
Starting from a dry froth (figure 2b) and rapidly increasing the amplitude of
the magnetic field above a threshold value, the MF boundaries undulate and
the bubbles become strongly deformed. A theoretical model developed by
Andrejs Cebers to account for this latter instability is presented in ref. [7],
which also includes numerical simulations performed by Ivars Drikis, and
comparison with experiments.

1 em
alternating
Figure 2. a) Wet liquid magnetic froth consisting in circular oil bubbles surrounded by a large quantity
of MF. b) Dry froth with faceted oil bubbles. The transition between (a) and (b) is obtained by draining the
MF using a little magnet introduced at the edge of the cell, which attracts the MF. c) Anisotropic liquid
magnetic froth. The alternating magnetic field is perpendicular to the cell ; the oil bubbles are then
elongated by adding a static homogeneous magnetic field parallel to the cell . d) Deformed liquid magnetic
froth obtained for high values of the magnetic field.
140

2. Formation of a magnetic fluid froth


Figure 3 shows the formation of a MF froth. Starting from a big drop of MF
surrounded by the oil (Fig. 3a), the perpendicular alternating magnetic field
is applied at t = 0. After several minutes some bubbles of oil are nucleated
within the MF (Fig. 3b ). The number of oil bubbles increases in time (Fig.
3c). After nuclueation, the bubbles grow until the froth occupies all of the
available space, and after several hours the equilibrium pattern is established
(Fig. 3d).
(a)

Figure 3. Formation of a 20 magnetic liquid froth. A spatially homogeneous alternating magnetic field
is applied perpendicular to the plane of the cell. Its amplitude is Ho = 18.4 kAm· 1 and its frequency v = 50
Hz. a) t =0: in the initial situation a MF drop is surrounded by the oil. The white bar represents I em.
b) t =7 min: several minutes after applying the magnetic field , some oil bubbles appear within and near the
edges of the MF drop. c) t = 15 min: the number of bubbles increases in time and the holes already formed
at the edges grow. d) t = 97 min: the equilibrium froth is formed.

The formation of the MF froth is due to a surface instability. The MF is


separated from the Altuglass© plates by a thin film of oil; therefore two
interfaces between the MF and the oil exist at the top and the bottom of the
cell as illustrated in Fig. 4. When an alternating magnetic field is applied,
they undulate. This dynamical instability is analogous to the Faraday
instability and has already been studied in [8]. When both interfaces
undulate 180° out of phase (i. e. the MF layer is in the peristaltic mode), the
MF layer narrows locally; a very thin film of MF is formed, which collapses
and gives way to a hole of oil.
Let us note that the structure of these MF froths is different from that of
20 soap froths confined between two transparent plates. Indeed, in the latter
the liquid wets the plates, so the bubbles are disconnected one from each
141

other; they can grow or shrink in time by gas diffusion through the liquid
boundaries. In 20 MF froths, the bubbles are connected to one another by
the oil film which separates the MF and the transparent plates, making the
structure rather similar to a lace structure; the flow of the oil within this film
can change the bubble size.

Figure 4. Nucleation of an oil bubble within the MF. The applied alternating magnetic field generates
an undulation of both interfaces between the oil and the MF at the top and the bottom of the cell. The top
sketch represents a transverse section of the cell in the initial state, corresponding to Figure 3a: the MF
shape is circular when observed in the direction perpendicular to the cell, and the MF is surrounded by the
oil. Below are sketched the stages of the oil hole formation. When the amplitude of the undulation is about
half the cell thickness , h, the MF layer narrows locally. The dipole-dipole repulsions between the magnetic
particles tend to increase this narrowing, and the oil flows over the MF into the region where the MF layer
is a thin film. Finally, the film breaks and gives way to an oil hole.

3. Topological characterization

The MF froths respect the basic topological rules which characterize all
cellular structures that are governed by surface tension [9-11]. The number
of MF lines impinging on a vertex is always equal to 3. As a consequence,
the cells have on average six sides (in fact the mean number of cell sides
varries between 5.6 and 6; this small discrepancy is due to the finite size of
the system and is observed in all experimental cellular structure). The MF
boundaries meeting at a vertex form an angle of 120° in a dry froth, thus the
edges of cells which are not 6-sided are curved.

4. Artificial coarsening: statistical study

Once it is formed, a MF froth can be in an equilibrium state. This means that


contrary to 20 soap froths, it can be stable in time. Indeed the energy of a
142
MF froth contains two contributions: the interfacial MF/oil energy, and the
magnetic energy. When a magnetic field is applied, the dipoles align on
average along the field direction, i. e. perpendicular to the MF layer. In such
a geometry, the magnetic dipoles repel each other. Therefore this tends to
stretch the MF/oil interface, whereas the interfacial energy tends to minimize
it. The competition between these two antagonistic energies allows an
equilibrium state to be reached, which stabilizes the pattern in time. Let us
note that the period of the alternating magnetic field is greater than the time
during which the magnetic particles rotate within the solvent, so the particles,
and consequently the MF magnetization, follow the motion of the magnetic
field.
Whereas a 2D soap froth coarsens in time to minimize its interfacial
energy, a comparable evolution is obtained in a 2D MF froth on decreasing
Ho, the amplitude of the magnetic field. Indeed the equilibrium width of the
MF walls is a function of H o. This point is the crucial feature for
understanding the 2D MF froths behaviour. As in the stripe pattern [2], the
equilibrium width of the stripes increases on decreasing H o. because the
repulsion between magnetic dipoles in the MF decreases. Since the total
amount of MF is fixed in the froth, the total length of the MF boundaries
must then be decreased. Hence some bubbles have to disappear on
decreasing H o; in the limit of a zero magnetic field, the magnetic energy
vanishes and the pattern turns into one single bubble. This evolution is called
artificial coarsening, in analogy with the time-evolution of 2D soap froths.
Two elementary process occur during this evolution. They are shown in
Fig. 5. A Tl rearrangement (Fig. 5a) is a process by which an edge between
two cells shrinks until its length becomes zero. Since a fourfold vertex is
unstable, a new edge is created perpendicular to the first one. This process
changes the local connectivity of the cells and their number of sides. Let us
note that such a Tl process is not reversible if the field is reversed to its
initial value (before the Tl). The process by which cells disappear in the
froth on decreasing Ho is a coalescence (Fig. 5b).
We have performed a statistical study of the topological and the
geometrical properties of the froth in the artificial coarsening. These are
extensively described in ref. [4]. We give here a brief summary of the main
results. Fig. 6a represents the variation of N, the total number of cells, during
the coarsening, for different values of the MF volume fraction C/J. This graph
must be read from right to left because the froth is created with a high
magnetic field, and then Ho is decreased. Two regimes can be distinguished:
for Ho > 9 kA/m (Regime 1), N remains constant: no coalescence is observed,
but T1 processes occur. We will see in the following section that this regime
corresponds to out of equilibrium states of the pattern. For H o < 9 kA/m
(Regime 2), coarsening occurs: cells coalesce and Tl rearrangements are
observed, and the number of bubbles decreases. Consequently, the average
cell area, <A>, increases on decreasing Ho. It is shown in ref. [4] that <A> is
proportionnal to H o -2 in this regime. Fig. 6b shows the evolution of the
distribution function of the cells with n sides, P(n), during the coarsening
process. P(n) is defined as the fraction of n-sided cells in the froth, for a
given value of Ho.
143

a)

b)

coalescence
)
Figure 5. a) A T1 process in a MF froth. On decreasing the amplitude of the magnetic field, the edge
between bubbles 2 and 4 disappears and a new perpendicular edge is created between cells I and 3.
b) Breaking of an edge between two cells. This is the mechanism by which cells disappear in a MF froth
when the magnetic field is decreased. Starting from an initial froth, the amplitude of the field is decreased.
The photographs represent respectively the situation prior to, during and immediately after the wall
destruction. The characteristic time of each of these two processes is of the order of one second. The white
bars represent I em.

Regime 2 Regime 1

.. .. ..
250
Tl + coa lescence OnlyTI - !l.19kA/m
z 200
p (n) - -~.77 k;\/m
-.- · 7.33kA/m
.!l
A A 6 • 6
<I> = 0.21
0 .8
··.a· ·5.98 kA/m
;:;

.•
v 15 0
<1> = 0.35
0
...
..0 100 ++
++
++ + +
<I> = 0.16
0 .6

e
:I
• + ++ 0 .4

z 50
+
+ C!>=0.2 l
0 .2

6 8 9 10 11 12 0
H (kA/m) 3

< 0
Coarsening number of sides: n

(a) (b)

Figure 6. a) Evolution of the number of cells N during the coarsening process for three different values
of the MF volume fraction C/J. This graph must be read from right to left, because the coarsening process
occurs as the ma~netic field is decreased. Two stages are clearly apparent, regardless of the value of C/J.
For Ho > 9 kAm- (Regime l ), N remains constant: no cells disappear. but T1 processes still occur. This
regime corresponds to out of equilibrium states of the froth (see section 5). For H 0 < 9 kAm- 1 (Regime 2),
oil bubbles coalesce and so N decreases; T1 rearrangements are still observed. In this regime, the froth is in
its equilibrium state.
b) Distribution function of cells with n sides, P(n), and evolution with the amplitude of the external field, for
lP =0.21. This evolution accounts for the two regimes quoted in Fig. 6a): for Ho above 9 kNm (Regime I),
P(n) does not evolve, whereas it broadens and becomes asymmetric below 9 kA/m (Regime 2).
144

In regime 1, P(n) remains constant, implying that it is not affected by T1


processes; it is concentrated around its mean value of 6. The froth is roughly
monodisperse: the second moment of P(n) is 112 "" 0.2, which is very small;
consequently the froth is well-ordered. In regime 2, P(n) broadens on
decreasing H o, becomes asymmetric and its maximal value is shifted from
n = 6 to n = 5. The topological polydispersity of the froth is nevertheless
fairly small. At the end of the coarsening 112 :::: 0.8, which is quite small (for
2D soap froths, 112 "" 1.5 in the long time regime). Typically, in a 2D MF
froth most of the cells have 5, 6 or 7 sides, and there are sometimes rare
events with 4, 8 or 9-sided cells.

5. The role of time: quasi Von Neumann behaviour

Although the dipolar magnetic repulsive energy can stabilize the MF froth in
time, we have observed out of equilibrium states in the high field regime.
Indeed in the regime 1, a time-evolution of the area of cells is observed. This
evolution depends, for each cell, on its number of sides: the cells with less
than six sides grow in time, whereas those which have more than six sides
shrink, and the area of 6-sided cells does not evolve in time. We present an
experiment performed for a dry froth and a high value of H o (i. e. in the
regime 1). We used a stabilized power supply for the coils, to reduce the
fluctuations of Ho to 0.1 %. The evolution of the froth was followed during
15 hours. The growth rate of the area of each individual cell was measured
every 15 minutes. We report below the averages of the growth rates, which
have been performed over all the cells in the froth which have the same
number of sides, and over the duration of the experiment:
( oA I 8t )n = 5 = 0. 07 mm 2 . min -1 for the 5-sided cells,
( oA I 8t )n = 6 = 0. 00 mm 2 . min -1 for the 6-sided cells, and
( oAI 8t )n = 7 = - 0.16 mm 2 . min -1 for the 7-sided cells.
Let us note that this time-evolution of the 2D MF froth in high field
regime does not induce a coarsening, as in the case of soap foams. Indeed
the 7 -sided cells shrink in time, but do not disappear: the magnetic dipolar
repulsions between the MF edges prevent the area of the cells from
decreasing to zero. A topological T1 process occurs when the cell is too
small, making it lose one side and stopping its time-evolution. The 5-sided
cells are also involved in T1 processes: the vertices are displaced during their
growth, consequently the neighbouring edges shrink. When the length of
one of these neighbouring edges tends to zero, a T1 process occurs to avoid
an unstable fourfold vertex. The cell acquires an additional side, i. e.
becomes 6-sided, hence its time-evolution is stopped (until it is involved in
another T1 process). The number of cells is thus conserved during the time-
evolution.
145

This evolution must be compared to the time-evolution of 2D soap froths.


Indeed, the latter coarsen following the Von Neumann law [9-11]:
dA n I dt = K. (n- 6},
where An is the area of an individual n-sided cell in the froth, and K is a
positive constant which depends on the gas diffusion process across the cell
edges. In a 2D soap froth, the 6-sided cells do not evolve in time, the cells
with more than six sides grow, and the cells with less than six sides shrink
until they disappear in a T2 process.
The 2D MF froths in high field regime seem to follow a Von Neumann-
like law as the growth rate of the area of a n-sided cell depends on n, but
their surprising peculiarity is that the sign of this growth rate is given by the
sign of (6 - n ), instead of (n - 6) for 2D soap froths. We try to give here a very
qualitative explanation for this behaviour. The main idea is that an
equilibrium state can be defined for a 2D MF froth, due to the competition
between the interfacial MF/oil energy and the magnetic dipolar interaction
energy. These two terms depend on L, the total length of the MF/oil
interface. The minimization of the froth energy with respect to L
gives Leq(Ho, tP), the equilibrium value of L as a function of the control
parameters. For a fixed value of tP,
aLeq !aH0 >0 (1)
because the magnetic dipolar repulsions, which tend to increase L, become
stronger as H o is increased.
How can the froth reach its new equilibrium state when Ho is changed? Let
us calculate L. For a dry froth, neglecting the Plateau borders, L is roughly
equal to the sum over the whole froth of the length of the cell boundaries:
L:N(n)(l),
where (n )::::: 6 is fixed by the topology and (l) is the mean length of a MF
wall: (l)::::: 'V'\AI = vSoi/ N, where <A> is the mean cell area and Soil is the
total area occupied by the oil. For a fixed value of tP, S 0 u is given. We
obtain:

L = C.N 112 (2),


where Cis a constant (which depends on tP). Therefore, the only way for L to
reach its equilibrium value when H o is changed is by a change in the total
number of cells. Two cases must be distinguished.
If H o is increased, Le is increased according to (1), and then N must
increase according to (2).13ut in a dry froth, it is not possible to create a new
oil bubble within the MF. Thus L remains below to Leq• its equilibrium value,
and the froth is in an out of equilibrium state: that is why a time-evolution is
observed in the high field regime. Let us now compare the time-evolution in
2D MF froths and in 2D soap froths. The energy of the latter is purely
interfacial: they are out of equilibrium systems because L is always superior
to Leq• which is zero (neglecting the interactions between the interfaces). The
2D soap froths coarsen in time in order to reach their equilibrium state, and
their time-evolution is driven by the Von Neumann law: 5-sided cells shrink
146
and 7-sided cells grow to decrease L. Contrary to the 2D soap froths, the 2D
MF froths in the high field regime must increase the total length of the
interface to reach their equilibrium state. That could explain that 5-sided
cells grow and 7-sided cells shrink in this regime.
Suppose now that H o is decreased, starting from an equilibrium state of the
MF froth. According to Eq. (1), Leq is then decreased, and some cells have to
disappear (see Eq. (2)) for the froth to reach its new equilibrium state. This is
possible because cells can coalesce. That corresponds to regime 2 in Fig. 6a.
Indeed in this regime, no time-evolution of the cells is observed, implying
that the structure is equilibrated.

6. Topological correlations
2D MF froths respect the Aboav and Weaire law, which describes the
topological correlations between neighboring cells [9, 10]:
<n> a+ J.L
m(n)=<n>-a+ 2,
n
where m(n) is the mean number of sides of the cells surrounding an n-sided
cell. J.l2 is the second moment of P(n), the distribution function of the cells
with n, and a is an empirical coefficient. This law implies that cells with a
small number of sides (less than <n>) will be on average surrounded by cells
with a large number of sides (more than <n> ), and vice versa. We obtained
values of a which do not seem to depend on the magnetic field strength, but
which depend on the MF volume fraction: a = 1 to 1.2 for tP = 0.13 to 0.35.
Fig. 7 shows the experimental data for tP = 0.16. They are obtained for
different values of Ho (from 6 up to 10 kA. m-1), and fitted according to the
Aboav-Weaire law.
• Hext = 6.3 kAm- 1 ( <n> = 5.80 and
<Jl2> = 0.42),

X Hexr = 8.6 kAm- 1 ( <n> = 5.91 and


<Jl2> = 0.22),
n.m(n)
C Hext = 10.3 kAm- 1 ( <n> =5.95 and
<Jl2> = 0.14).

30
/ <{) = 0.16
25+--+--~4--4--+--+--~~
3,5 4 4,5 5 5,5 6 6,5 7 7,5

number of sides: n

Figure 7. Curve of the product nm(n) versus n for tP = 0.16 and for different values of the magnetic
field. We also indicate in brackets the value of the average number of sides and the second order moment
of the side distribution for a given field. We obtain a"' 1.0 for tP = 0.16 and for all values of the applied field.
147

Let us note that the obtained value of a is very close to that obtained in other
20 cellular structures: a = 1 in 20 soap froths, and a = 1.2 to 1.5 in
magnetic garnet films [ 10]. The physical signifiance of this empirical
coefficient is not yet known, but we can remark that its value is roughly the
same for the evolving 20 cellular structures.

7. Two-dimensional liquid magnetic froth: a model for studying the


topology of two-dimensional cellular structures.
Here we present a potential application of 20 MF froths. This work is
performed in collaboration with Fran~ois Graner and James Glazier.
The 2D MF froth presents an advantage in that the vertices may be pinned
by a local magnetic force. This force is created by a small piece of magnetic
material, which concentrates the magnetic field lines, creating a local
magnetic field gradient, which attracts the MF. Moving this piece above the
experimental cell, it is possible to displace the vertices and to force a fourfold
vertex. When the piece is removed, the fourfold vertex spontaneously splits
into two threefold vertices, linked to one an-other by a new edge. There are
two potential configurations for this new edge: it can be either in one
direction, or in the perpendicular one; one can switch from one of these
configurations to the other one with a T1 process. Spontaneously, the system
chooses the stablest one. One sees then that if the initial configuration is not
stable, the perturbation then allows the system to cross the potential barrier
between the two states, and an artificial T1 process is induced. Furthermore it
is possible to go back to the initial situation (before the Tl ): the piece of
magnetic material is put above the experimental cell, again forcing a
fourfold vertex, then it is moved rapidly back and forth in the direction of
the edge that the experimenter wants to create, and afterwards it is quickly
removed. Figure 8 shows images of such an experiment. Four cells within the
froth have been represented. Fig. 8a corresponds to the initial situation. Fig.
8b represents the configuration spontaneously adopted after the fourfold
vertex has been forced and the piece of magnetic material removed. A Tl
has been induced between the two figures, implying that Fig. 8b corresponds
to the stablest configuration. Fig. 8c represents the configuration obtained
from Fig. 8b after another artificial Tl by the method described above.

Figure 8. Configurations of a cluster of four cells isolated in the froth. a) Initial configuration. b)
Configuration spontaneously adopted after a fourfold vertex has been forced using an additional local
magnetic force. c) Configuration adopted after a second artificial Tl process has been induced, using the
method described in the text.
148

Such an experiment gives informations about the range of a Tl


rearrangement. Figure 9 is a superposition of the images of the whole froth
before the Tl(thick line, corresponding to Fig. 8a), and after the Tl (thin
line, corresponding to Fig. 8b ). Motion of vertices is observed up to the third
neighbours of the vertices involved in the Tl. We can be sure that these
motions are due to the Tl process, and not to a possible evolution of the
froth. Indeed the whole froth is exactly the same after the two consecutive
Tls, implying that the displacement of the vertices observed in Fig. 8b is
reversible. Therefore, we can conclude that a Tl is not a local process. That
complicates somewhat the problem of predicting a T1 process in a froth. The
analysis of the induced displacements of vertices near an artificially
displaced one should allow one to quantify the probability for a Tl to occur
in the froth, during its evolution. In future works, we plan to use 20 MF
froths as an experimental model system to study such topological
interactions between cells in a 20 froth.

Figure 9. Comparison of the froth before and after a Tl process. This figure is the superposition of two
images corresponding to the situations represented in Fig. Sa (thick line) and Fig. 8b (thin line). One sees
the non-locality of a Tl process, because motion of the froth is observed far from the edge involved in the
Tl.

8. Conclusions
The 20 MF froths are new experimental cellular structures. They are close to
soap froths because they have surface tension; due to the magnetic dipolar
repulsions between the cell walls. they are also close to cellular structures in
magnetic garnet films.
The characteristics of these systems are determined by the value of a
control parameter: the strength of the applied magnetic field, which fixes the
MF wall thickness. These froths can be in equilibrium (i. e. stable in time).
149
An evolution is obtained on decreasing the amplitude of the external
magnetic field. Two regimes then clearly appear then. For high values of the
applied field, the system is out of equilibrium: the magnetic dipolar
repulsions are strong, and the system does not have enough interface to
equilibrate these interactions. Thus a time-evolution of the pattern occurs,
following a Von Neumann-like law, but with an opposite sign: five-sided
cells grow in time, and seven-sided cells shrink, for the system to reach its
equilibrium. For low values of the applied field amplitude, on the other hand,
the froth does reach its equilibrium. On decreasing the field, cell coalescence
is observed, and the distribution function of cells with n sides, which is very
narrow and symmetric in high field, becomes broad and asymmetric.
This macroscopic cellular pattern can be employed for topological
studies because it can be in an equilibrium state for a given value of the
external field. In a soap froth, gaseous diffusion is superimposed on the
topological processes, implying that the froth is only in a quasi-equilibrium
state. In garnet systems it is not possible to observe directly a topological
rearrangement because it occurs very quickly. The 2D MF froth, however,
can be effectively used as a model system to differentiate the diffusion
effect and the topological changes. Furthermore the topology can be forced
by inducing artificial Tl processes, as described in section 7. For instance,
it is possible to create a defect in an equilibrated froth and study the
relaxation towards equilibrium. Such experiments should allow us to
determine the role played by the topological T 1 processes.

9. References

[1] Neveu-Prin S., Tourinho F. A., Bacri J.-C. and Perzynski R. (1993)
Magnetic birefringence of cobalt ferrite ferrofluids, Colloid Surf A 80 1.
[2] Flament C., Bacri J.-C., Cebers A., Elias F. and Perzynski R. (1996)
Parallel stripes of ferrofluid as a macroscopic bidimensional smectic,
Europhys. Lett. 34 225.
[3] Elias F. , Flament C., Bacri J.-C. and Neveu S. (1997) Macro-organized
patterns in ferrofluid layer: experimental studies, J. Phys. I 7 711.
[4] Elias F., Flament C., Bacri J.-C., Graner F. and Cardoso 0. (1997) Two-
dimensional magnetic liquid froth: coarsening and topological correlations,
Phys. Rev. E 56 3310.
[5] Babcock K. L. and Westervelt R. M. (1989) Elements of cellular domain
patterns in magnetic garnet films, Phys. Rev. A 40 2022.
[6] Weaire D., Bolton F., Molho P. and Glazier J. A. (1991) Investigation of
an elementary model for magnetic froth, J. Phys. Condens. Matter 3 2101.
[7] Elias F., Drikis F., Cebers A., Flament C. and Bacri J.-C., Undulation
instability in two-dimensional foams of magnetic fluid, in preparation.
[8] Bacri J.-C., Cebers A., Dabadie J.-C., Neveu S., and Perzynski R. (1994)
Threshold and marginal curve of magnetic Faraday instability, Europhys.
Lett. 27 437.
150
[9] Weaire D. and Rivier N. (1984) Soap, cells and statistics-random patterns
in two dimensions, Contemp. Phys., 25, 1 59.
[10] Stavans J. (1993) The evolution of cellular structures, Rep. Prog. Phys.
56 733.
[11] Glazier J. A. and Weaire D. (1992) The kinetics of cellular patterns, J.
Phys.: Condens. Matter 4 1867.
CELLULAR STRUCTURES IN METALLURGY

Y.J.M.BRECHET, D.WEYGAND

L.T.P.C.M. Groupe "Physique du Metal",


Domaine Universitaire de Grenoble,
ENSEEG, BP75, 38402 Saint Martin d'Heres, France

Abstract; A review of cellular structures in metallurgy is given, insisting both on the


physical origin of these structures, on their influence on mechanical properties, and on
the current methods used to model them. Successively will be presented grain structures
in a polycrystal, cellular structures issued from deformation and fracture, and metallic
foams, both regular and irregular.

!.Introduction

This lecture is a layman viewpoint on cellular structures, from a metallurgist. It is


obvious, with a somewhat vague notion of cellular structures, that they are often present
in metallurgy. The most well known example is the structure formed by grain
boundaries in polycrystals. But cellular structures appear also as dislocation patterns
emerging from plastic deformation, as dimples patterns observed on ductile fracture
surfaces, and even more directly in this new class of materials: the metallic foams. The
question is therefore not so much "are there cellular structures in metallurgy?" than
"what can the metallurgist learn from the results obtained in your community?" and
"what examples of cellular patterns observed in metallurgy can be of some interest to
you?". In order for these questions to be addressed in a meaningful! manner, i.e. beyond
a simple visual analogy which can be misleading, an introduction to the field of
metallurgy is necessary. Though it will be at a very elementary level, it will be
sufficient, I hope, to provide a common vocabulary. Beyond the necessary "zoology" of
cellular structures in metallurgy, we'll try to present the physical origin of these
structures, the mesurable quantities which can describe them, the properties related to
these cellular structures, and the current understanding of the relations between the
process, the structures and the properties. The references given in this course will be
divided into "general references" providing a more detailed introduction to the field of
physical metallurgy, and "specialised references" aiming at a more focuss insight into
some questions which can be of interest for collaborations between different fields.
I will first focuss my attention on grain structures in polycrystals, and their thermal
stability (§2). Then I will present the most usual mechanisms of plastic deformation
and ductile fracture and the cellular structures emerging from these processes (§3). The
last exemple will be devoted to relatively new materials with interesting properties,
metallic foams, for which ideas develloped for other foamed materials can help the
undestanding (§4).
151
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 151-174.
© 1999 Kluwer Academic Publishers.
152

2. Grain structure in metals

2.1 THE ORIGIN AND STABILITY OF GRANULAR STRUCTURES

The most natural cellular structure in a metal comes from its crystalline structure:
different regions Qf space with different crystalline orientations will meet (to fill the
space) along structural defects, the grain boundaries, which will delimitate a cellular
partition of the space [1] . Each grain boundary will be characterized by the
misorientation between the two crystals separated by this grain boundary, and by the
inclination of the boundary with respect to one of the lattice frame. These grain
boundaries, as structural defects, have an energetical cost, and therefore if enough
mobility is given to the atoms (which means in metallurgy at high enough temperature
to allow atomic diffusion) the system will reorganize progressively in order to diminish
the total area of grain boundaries: this is the origin of grain growth. This granular
structure can be obtained either from the liquid state, or from the solid state when a solid
has been heavily deformed and then annealed at higher temperature [2].
From the liquid state [3], germs of crystal will appear in the liquid (often on
heterogeneous nucleation sites) and will grow till impeagement occurs. The grain size
resulting from this process will depend on the nucleation and growth rate of the germs
from the liquid. These quantities depend both on the driving force for nucleation, and the
mobility to ensure growth. As a result, the grain size after a solidification process can be
controlled by the number of nucleation sites and the rate of cooling the liquid metal. For
instance a higher cooling rate will lead to a smaller grain size.
Another way of controlling the grain size is to deform heavily the material (for instance
rolling it) and then to heat it. The result of rolling is to introduce in the metal energetic
defects (dislocations, see §3) which increase subsancially the free energy of the solid.
While annealing, one may under certain conditions have new grains, free from defects,
growing at the expense of the defective regions. This mechanism known as
recrystallisation is a method to control grain size either changing the rolling reduction,
the rolling temperature or the annealing temperature. For instance a higher rolling
reduction will decrease the grain size [2].

2001Jffi
Granular structure in pure Aluminium
153
These granular structures in pure metals have been extensively studied, and they are
familiar to the people interested in cellular structures. Less familiar are some structures
issued from solidification of alloys rather than pure metals [4]. In an alloy, with a
typical phase diagram shown in figure 2, solidification cannot occcur at constant
composition except at the so called "eutectic composition". For instance for a
hypoeutectic alloy, the frrst solid will be solidified at the composition indicated by the
phase diagram which is lower than the nominal composition of the liquid, therefore
while solidification proceeds, the liquid will become enriched progressively up to the
eutectic composition at which solidification will terminate. The result of this process
will be to have grains of metal, progressively richer in solute while they grow, arxl
finally surrounded by a "eutectic phase" which is a two phase region. If the initial
composition is at the eutectic composition, the solidification will occur at constant
composition, starting from nuclei, and the final structure will be composed of "eutectic
grains" wich are the equivalent of the "dendritic grains" previously described, but with a
two phase situation inside each grain. Such a structure, when annealed, certainly will
undergo coarsening, just as the dendritic grain did, but with an extra degree of
complexity associated with the coarsening of the two phases which may occur inside the
eutecti grains themselves. Surprisingly, the thermal stability of these eutectic grains has
recieved very little attention compared to the thermal stability of the dendritic grains.

Figure 2

. ~

~ -
0
0 0 ()
"
.
0
no 0 ~
0
0

..
0

0
0 \;,\
o'tL () 0

'~ ~ '11-~1.7
'\'
~~~

.. I ·J. ~ ~.~
•• ~ f9!
l L II ' I

AI IO Sl,. . . u~fj
Schematics for hypoeutectic (a) and eutectic (b) solidification

2.2 Tiffi IMPORTANCE OF GRAIN BOUNDARIES

Most of metallic alloys are used as polycrystals. It is therefore of utmost importance to


understand the role of grain boundaries and of the scale of the grain structure on their
properties. Grain boundaries separate crystals of different orientation. Grain boundaries
are regions of defective metal, often associated with chemical gradients. They will
influence both the mechanical properties and the chemical properties. They will
influence the mechanical properties because plasticity of metal depends on the
orientation of the lattice with respect to the loading directions. They will influence
corrosion because of the chemical variations across grain boundaries.
154
During high temperature defonnation, the larger the grains, the larger the resitance to
plasticity. On the opposite at low temperature, the smaller the grains the higher the
resulting resistance to plasticity. This different behaviour results from a different
defonnation mechanism. But in any case, grain boundaries are at best inocuous as far as
fracture behaviour is concerned, and at worst dramatic. Chemical segregation at grain
boundaries, can lead to dramatic embrittlement of the material. As a result of these
properties, the grain boundaries are a major concern for the optimisation of materials,
and both their individual properties and their collective behaviour is of practical interest.
That could be a wonderful pretext to study soap froth since they have been often
presented as analogous to metallic grain structures. However, before going further in the
analogy, we need to precise the difference between a soap froth and a polycrystal, these
difference being related to the fact that a crystalline structure underlies the grain
boundaries [5].

2.3 GRAIN BOUNDARY PROPERTIES: ENERGY AND TRANSPORT.

Grain boundaries are surfaces separating two crystals. As such their structure and their
properties depends on the misorientation, and in a lesser effect, on the inclination, of
the grain boundary. In contrast with the soap froth, the grain boundary energy and the
grain boundary diffusivity can be very different (by a factor 10 for the energy) depending
on the nature of the grain boundary. It is usual to distinguish the low angle grain
boundaries (corresponding to misorientations less than 10°, and whose energy is roughly
proportional to the misorientattion) and the high angle grain boundaries, most of which
have the same properties, except from some "special grain boundaries" corresponding to
particular grain boundary structures. Figure 3 shows an illustration the dependance of
grain boundary energy with respect to misorientation for a family of grain boundaries.

Energy vs misorientation for symmetrical <100:> and <110> tilt grain boundaries in aluminium [9]

Similar result exist in the litterature for the mobility or the diffusivity of grain
boundaries [10]. These data are difficult to obtain experimentally, that's why they are so
valuable. In fonner times, bicrystal experiments had to be perfonned. Nowadays, thanks
to the development of experimental techniques allowing a local measurement of the
misorientation, it seems possible to get some of these infonnations much more rapidly
using polycrystal experiments.
155

It may seem to the physicist a bit "zoological" but these differences in energy aro
mobility for different misorientations play a crucial role in texture development (i.e. in
selecting the crystallographic orientations dominant after say a recrystallisation
treatment). It is unlikely that a realistic description of grain growth can be obtained
without taking into account these specificities, and that is certainly a major difference
with soap froth.

2.4. GRAIN BOUNDARY DYNAMICS: SOLUTE DRAG AND PINNING

Another difference with soap froth is that grain boundaries are moving inside a crystal
and as such they may be submitted to retarding forces. The most frequent forces come
either from the elements present in solid solution, or from the precipitates. The physical
reason for the flrst retarding force stems from the interaction between solute and the
grain boundary which make as a result the grain boundary equilibrium concentration
different from the nominal concentration. There will exist a concentration profile around
the grain boundary, and when it is migrating, this profile becomes asymmetrical. From
this asymmetry results a drag force which is a non linear function of the velocity. This
drag force has been computed by several authors, and the net result is a non linear
relation between the driving force which drives the migration of the grain boundary aro
the velocity [11][12]. This drag force is zero when the velocity is zero and therefore
does't result in a threshold for migration.
By contrast, the interaction between precipitates and grain boundary gives a threshold
force below which no migration is possible. The physical reason for this interaction, put
forward by C. Zener, is the reduction in grain boundary surface if a precipitates sits on
the grain boundary [2].

Figure 4

Experimental evidence of grain boundary pinning in aluminium


(Transmision Electron Microscopy observation).

If y is the grain boundary energy, r the size of the precipitates and f their volume
fraction, the resulting pinning force is 3yf/2r. The driving force to drive grain growth is
the reduction of surface and therefore is equal to 2y/R. The result is than t.'lere exist a
limiting grain size at which the growth is stopped by Zener pinning: it is given by
4r/3f.
156
This very simple estimation has been refined many times by many authors: the
corrections given by more realistic models are always minute, and we can take that as a
reasonable estimation of the threshold force due to pinning. More complex situations
when the precipitates can be dragged by the grain boundary are also possible, but beyond
the scope of this introduction to physical metallurgy.

The two examples provided here show that, though at first sight soap froth and grain
structures in metals seem similar, the very physical nature of grain boundaries is
responsible for important differences in the behaviour of the elementary walls. A natural
question to be asked to the physicist is then "what are the properties of froth wich are
insensitive to these differences? What modifications do these specificities of the grain
boundary introduce into the collective behaviour of a cellular structure?".
As we have seen, the presence of a grain boundary increases the free energy of the
system. Therefore, grain growth appears as the natural mechanism leading to the
reduction of the total grain boundary area. This phenomenon will be reviewed in the next
8.

2.5. COLLECTIVE GRAIN BOUNDARY DYNAMICS: NORMAL GRAIN


GROWTH.

The easiest way to observe grain boundaries in metals is to take advantage of their
chemical sensitivity. A mild corrosion called etching reveals the grain boundary at the
surface of an polished sample. Then the observation of grain growth is possible by
etching samples after different durations of the coarsening heat treatment. It is quite easy
then to measure the average grain size as a function of time, and the grain size
distribution. Unfortunately, what we get by this procedure is a two dimensionnal section
of a phenomenon which occured in three dimensions. Stereology allows, with some
hypothesis, to get from these measurements some informations concerning the true 3D
behaviour. Alternatively, experiments can be done on thin films for which the growth is
two dimensionnal. But really 3D measurements are quite difficult. For instance one
would like to investigate experimentally the equivalent of Von Neuman law (6 as the
magic number of sides) for three dimension. In principle the experiment is easy to
perform: start with an aluminium polycrystal and embrittle the grain boundary with
gallium, and then count the faces for a given grain size ... easy to say, tedious to do ...
Most of the available informations concerning grain growth in metallurgy are in fact 2D
informations on a 3D phenomenon, most of the available computer simulations for
grain growth are 2D simulations [13], some of them are 3D [14].

The very first question to be asked concerning grain growth is the overall average
kinetics. Many years ago, Burke and Turnbull provided a simple phenomenological
model. The density of surface energy is the driving force for grain growth. It scales as
1/R where R is the grain size. The rate of evolution for the dimension of the structure is
assume to be proportional to the driving force, and therefore R 2 should be proportional
to time. Difficult to find a simpler derivaton for a kinetic law ... Experimentally, this
natural exponent is seldom observed. Numerically, for large enough systems, it gives
accurately the asymptotic behaviour, as any domain growth in a conserved order
parameter system.
157

Since this very simple derivation, the classical work by Hillert [15] has provided a better
foundation for this parabolic law, assuming that the grains smaller than average would
shrink at the expense of the grains larger than average: the shift of the average grain size
is parabolic in time, and Hillert was able to derive a grain size distribution ... which is
never observed experimentally. Experimentally, the distribution seems rather log
normal. More recently, a totally different approach, closer to the usual topological
considerations issued from Von Neuman's law was proposed and applied to 20 grain
growth, allowing also to take into account the specificities of grain boundaries ( energy
and mobility depending on the misorientation). But clearly some important issues
concerning grain growth in metals, such as the effect of solute drag, of precipitate
pinning, of different stored energies in the bulk of different grains, were out of the scope
of these classical derivations. That fact motivated the development of computer
simulations, first to describe grain growth in the most simple situations, then to apply
these methods to the more complex problems which occur in realistic situations. A
short overview of these computer simulation methods will be presented now.

2.6. COMPUTER SIMULATIONS OF GRAIN GROWTH: EXAMPLES OF


APPLICATIONS

Only recently intensive computer simulations have been applied to materials science,
but the variety of problems involving collective behaviour encountered in this field has
triggered a very productive activity (for a review, see [6]). The main problem in
materials sciences is the variety of scales involved. For grain growth for instance, the
grains may be 100 microns, whereas the distance between the atoms crossing the
interface is of the order of 2 angstrom. In order to have a reasonable statistic on the
grains, at least a sample of lmm is needed. The result is that the scales involved in the
phenomenon span 6 orders of magnitude!!!

The first models [14] proposed to simulate grain growth were adapted from the Pott's
model. Each orientation for a grain corresponds to a "level", and the hamiltonian
included higher energies for bonds between dissimilar sites. For numerical reasons, each
site was in fact composed of thousand of atoms, which as result led to a "frequency of
jump" which had little physical meaning. Monte Carlo time in this simulation was only
a "fake time". Still, many problems were adressed with this simulation which had the
non negligible advantage of simplicity. The parabolic growth kinetic was recovered, log
normal distribution were obtained. But it seems that the local fluctuations inherent to
the method makes its application dubious as far as problems such as pinning, or
recrystallization are concerned. As a general rule, it seems that this method is well
adapted to problems which are not too dependant on local configurations.

The other class of models proposed were more "continuous" in spirit. The grain
boudaries are simulated as "elastic membranes", the triple junctions between grains
being driven by the capillary forces. Several models of this type have been proposed,
first with triple junctions as the only vertices, then allowing virtual vertices along the
grain boundary in order to allow non circular shapes. The most general formulation for
this class of models is given by Kawasaki [16]. Rather than entering into the technical
details for these models, for which the reader is referred to references [13], [6] we will
158
then illustate their applications to some "real problems" corresponding to experimental
observations in metallurgy.

2.6.1. Grain growth in systems with wetted grain boundaries (Monte Carlo model)
A puzzling observation [19] is that adding SOppm of Gallium to pure aluminium
accelerates grain growth. More specifically, in situ experiments with bicrystals, using a
special X-Ray device to follow the grain motions shows that it goes faster. The
proposed explanation is the occurrence of a "liquid like layer" of Gallium along some
grain boundaries, this liquid like layer increasing enough the mobility to overcome the
effect of solute drag. As segregation is expected to depend on misorientation, only some
grain boundaries of high enough misorientation will exhibit a liquid layer. This
motivates the following simulation: in the Monte Carlo Model, the grain boundaries of
large misorientation are affected with a 10 times larger mobility. This effect is
characterised by a critical misorientation .1qc above which the mobility is larger. When
all grain boundaries are "slow", or when all are "rapid", the parabolic kinetic is observed.
and a scale invariant log normal distribution is measured (Figure 5) . When only some
of the grain boundaries are rapid, the kinetic is no longer parabolic and the distribution is
no longer scale invariant (Figure 6), the proportion of fast and slow grain is evolving
with time.

Figure 5

a) b)

-
3000 o.oe
MCSo2COO
0.10
2000 0.04
T
1000
!""o.02 ~
-o.os

0.00
-3 0.00-3
-2
~ -1 0
~
log(-)

MCS=lOOO MCS=2000 MCS=SOOO

Grain growth with a uniform mobility: a) kinetic of the average grain size <a> in function of MCS(Monte
Carlo Step); b) distribution of grain size for two different stages of growth; c) grain structure at three
different stages of growth
159

Figure 6

a) b)
2000 r-- -- --------, 0.10 r----------,

!
0 .08

a
0.00

0.00
1000 ~D.O.
0.04

0.02
0 .02

10000 MCS 15000

MCS=lOOO MCS=5000 MCS=l5000

Grain growth with a variable mobility: the grains with a misorientation lower than &CJ.: are slow, the others
are rapid.(&qc::0.77):a) kinetic of the average grain size <a> in function of MCS;b) distribution of grain
sizes for two different stages of growth;c) grain structure at three different stages of growth

2.6.2. Abnormal grain growth in a grain size gradient (Monte Carlo model)
A classical problem [18] in grain growth is when a large grain is "eating all the others".
This is called "abnormal grain growth". It is often observed in alloys with dissolving
precipitates. A simple way to understand this fact is to look at the driving force driving a
flat interface in a region with grain size R. The force will be 3y/R instead of 3y/2R for
normal grain growth. If there exist a "large grain", it may grow whereas the generalised
grain growth is still impossible due to Zener pinning. Therefore when precipitates are
dissolving, i.e. when Zener pinning decreases, abnormal grain growth can occur. 1bere
are other situations associated with large differences in mobilities of grain boundaries in
which abnormal grain growth, as a transient regime, can occur. A related, much less
studied, question is the grain growth phenomena in an initial grain size gradient. This
for instance the situation in granular structures directly issued from solidification where
the zone close to the cooler ("chilly zone") has very small grains.
This can define a "macroscopic interface" [17] between two regions of different stored
energies. The result is the drift of the interface, with a mobility related to the individual
grain boundary mobility, while the stored energy (and thus the driving force) is evolving
with time.
160

Figure 7

a) b) so
40

e30

20

1000 2000 3000 4000


MCS

MCS=250 MCS=500 MCS=750

a) Definiton of the interface coordinate system; b) Migration kinetics of a macroscopic interface;


Displacement <X> of the interface in function of MCS; c) Evolution of an interface between regions of
different grain sizes

2.6.3. Triple junction mobility (Vertex model)


A recent experiment [20] has shown that triple junctions may have mobilities different
from the grain boundaries themselves. This is a simple situation to apply the vertex
model, on a geometry which is exactly the experimental one. Figure 8 shows the
morphological evolution for different triple junction mobilities

Figure 8
a)
r-----------------·

~>
Morphological evolution of a tricrystal with different mobilities for the triple junction, as simulated by the
vertex model. The influence can be seen on the angle at the triple point: a) equal mobility; b) triple junction
less mobile
161

2.7. SOME OPEN QUESTIONS

From the metallurgist view point there are still many open questions concerning the
stability and evolutions of granular structures. For all these questions, some
experimental investigations can be performed, and they are therefore good ways of
testing models. The most obvious questions relate to the influence of drag forces and
pinning forces: what is their influence on the average kinetics and on the grain size
distributions? But a question of more general interest for the physicist is: knowing the
differences at the level of the boundary between walls in a soap froth and grain
boundaries in metals, what are the expected differences at the level of the collective
behaviour of a soap froth and a polycrystal.

3.Plasticity and ductile fracture

Whereas grain structures are quite familiar to the community of cellular structures, if not
in details, at least in the siming with soap froth, there are in physical metallurgy
occurrence of cellular structures during plastic deformation and fractures of metals which
have never been investigated with the view point of "cellular structures". Before giving
two examples of these structures, lets briefly review the ideas necessary to get un
understanding of the mechanical properties of metal, beyond elasticity.

Figure 9

A tensile stress strain curve for a typical pol ycrystalline ductile metal [ 1]

A typical stress strain curve for a ductile metal is shown on figure 9. At low applied
stresses, the strain is linear with stress and reversible: it is the elastic regime which is
the macroscopic response of the harmonic crystal. Beyond a threshold stress called "yield
stress", deformation ceases to be reversible and the relation between stress and strain
becomes non linear. The stress removed, the material is left with a non recoverable
deformation, the plastic deformation. If the stress is removed, the unloading curve is
parallel to the elastic part of the stress strain curve, indicating so that the harmonic
crystal has remained essentially unchanged by the plastic flow. These facts are the basis
of the microstructural interpretation of plasticity.
While deformation proceeds in the plastic domain, the increase of stress needed to
increase the strain of a given amount decreases : it is said that the "work hardening" is
162
diminishing. If locally on a sample a geometrical defect is present, locally the stress is
increased and the strain increases accordingly, thus reducing further the effective section
of the sample: this mechanism would immediately lead to fracture if work hardening (the
fact that a deformed material is harder than a non deformed one) were not to compensate
it. When deformation proceeds, stress increases while the workhardening rate decreases.
When they become equal (Considere criterion) plastic flow localizes into what is called a
"neck" and the sample breaks. The microscopic mechanisms for this type of fracture,
known as ductile fracture, is the growth and coalescence of voids developped mainly due
to the plastic flow. In the next §, a short overview of the microscopic interpretation of
mechanical properties will be given.

3.l.MECANISMS FOR PLASTIC DEFORMATION IN METALS

At temperatures below half of the melting point, plasticity of metals is mainly


controlled by the motion of linear structural defects called "dislocations" [1][21][22]. The
defects are associated with a stress field, and therefore couple to the applied elastic
stresses. As a result, when the stress is sufficient (i.e. above the yield stress),
dislocations are moving and are responsible for a non reversible deformation called
plasticity (Figure 10).

Figure 10
!I II I I I
I I I I I I I I I I
I I I I ' I i I ;
I '' ! I I I 0 I

''
I i ' I I
:'
I

I I I I I I I
! i

--
I I I I I \ I I

- :
I
:
i
'
:
I
I
,
:
~
:
:
i
1

:
I

'
I

!
'
I i

'

.-· -

- -- --
a) Schematic of a dislocation; b) Its motion by glide due to a shear stress

Dislocations are associated with a stress field, which causes them to have a line energy.
Were there is no pinning force on dislocations, they should not be present in the crystal
since they are far from equilibrium and the configurational entropy which is limited for
topological reasons (a dislocation cannot end in the crystal) cannot counterbalance the
important elastic energy of these defects.
These defects having a stress field associated with them, they interact, and their
interaction is long ranged (force decreasing as 1/r). Similar dislocations repel each other,
163
opposite dislocations attract each others. Beside this long range interaction, dislocations
have a short range interaction due to the fact that they are structural defects. They can
make junctions, nodes... An extra complexity is added to the problem by the crystalline
nature of metals. Dislocation have to glide along prescribed dense planes, they can
"climb" out of these planes only with the help of point defects. As a result of these
constraints on their motions, dislocations can never really screen each other their stress
fields.
Beside not having an equilibrium density, dislocations are also non conservative species.
Figure 11 shows schematically the most frequent mechanisms for annihilation mxl
multiplication of dislocations. These mechanisms are active all along plastic
deformation, and their dynamic equilibrium is responsible for the existance of a "steady
state" of constant stress for a given strain rate and a given temperature. It is observed
that while deformation proceeds, the dislocation density p and the stress cr are related by
the equation cr=0.5Gbp 112, b being the atomic spacing and G the shear modulus [23].

Figure 11

6
0 Ill l'!ll
Q
'

HI hm
G
' I '
'

a) Multiplication of dislocations by the Frank Read source mechanism


b) Annihilation of dislocations of opposite sign

This law can be derived very simply from the condition under which the Frank Read
mechanism can operate: the stress necessary to reach the unstable half circular position
scales as 1/d where d is the distance between anchoring points. If the pinning points are
provided by other dislocations, one must have p=lld2 , and that explains the experimental
164
result. It also explains why a materials "work hardens": while deformation proceeds,
dislocations accumulate and the distance between them diminishes, making the operating
of Frank Read source more difficult. To sustain the same strain rate, the stress has to
increase.
As will be shown later, these linear defects responsible for plasticity do not remain
positioned at random, but under certain conditions, they organize themselves into
cellular structures [24].

3.2.MECHANISMS FOR DUCTILE FRACTURE

In ductile metals, the microscopic mechanism for fracture [7][24] is the nucleation,
growth and coalescence of cavities. At high temperatures, under a constant applied stress
("creep conditions") fracture occurs in general along the grain boundaries which are
weakened by cavities growing on the grain boundaries or a the edges. The cavity growth
is controlled either by diffusion, or by plastic flow. The fractographic patterns are in
general difficult to observe because of oxydation at high temperature.
Figure 12

Mechanisms of creep fracture by cavity growth along grain boundaries

At lower temperature and in a constant strain rate testing, ductile fracture is most often
transgranular, which means that the micromechanisms involved operate inside the
grains. In most of alloys, inclusions act as nucleation sites for the cavities, which then
grow by plastic flow up to a critical size for which they coalesce leading to fimal
fracture [25].

Figure 13

Fracture mechanism by cavity growth at inclusions


165

The growth rate of cavities depends exponentially on the hydrostatic stress, and therefore
the ductility of the material, and even the failure modes can be considerably modified by
applying an external pressure [26].
This mechanism (growth of cavities from inclusions) leads to characteristics fracture
surfaces exhibiting dimples at the center of which inclusions can be observed.

3.3. CELLULAR DISLOCATION STRUCTURES

Dislocations produced during plastic deformation are not randomly distributed. They
organize into cell structures as shown on figure 14.

Figure 14

a) Dislocations cellular structures after a creep test


b) Dislocations cellular stuctures after cold rolling
c) Scaling law for dislocation cell size as function of the applied stress
166

It seems that the yield stress is controlled by the dislocation density much more than by
the details of the organisation. The work hardening behaviour though can be very
different depending on the microstructure. A very simple scaling law of inverse
proportionality relates the dislocation cell size to the applied stress [23]. This scaling
law can be readily obtained from dimensionnal analysis and is once again related to the
stress necessary to operate a Frank Read source. The origin of these cellular structures is
still not understood, in spite of numerous attempted models. Is it a "memory" of the
first "hard points" formed between interacting dislocations where dislocation storage
would have been more efficient? Is it an over all instability? Is it a self screening
structure for dislocation fields (a sort of coulomb gaz screening with limited degree of
freedoms)? The current approach for this problem is intensive 3D computer simulations
of dislocations dynamics allowing both creation and annihilation of dislocations.
However the strains attainable in these simulations are still too low to show a
convincing cellular structure. The formation of dislocation cellular structures remains to
now a mystery, and one may wonder is the analysis of the statistical distributions of cell
sizes could bring some informations on their generation.

3.4. CELLULAR STRUCTURES IN FRACTOGRAPHY

The dimple structure observed on fracture surfaces in ductile metals clearly provide a
space filling pattern in two dimensions (figure 15).

Figure 15

Fracture surface of Aluminium observed in scanning electron microscope: the dimple pattern defines a
cellular structure
167

The mechanism responsible for this patterns is the nucleation, growth and final
instability of voids initiated at inclusions. The number of nucleation sites can be
measured experimentally. However numerous questions are still open: is the growth of
one cavity coupled to the development to cavities in its surrounding? What is the final
cavity density leading to fracture? Is the nucleation rate constant or not during plastic
flow? May be the analysis of statistical features of the dimple cellular structure would
provide more informations on these questions. Experimentally a number of experiments
are possible such as varying the number of nucleation sites, or initiating the cavities,
and then developping them at different rates by applying an hydrostatic pressure. The
metallography and image analysis of such fracture patterns is relativly easy, an so this
may be an interesting field to apply ideas corning from the physics of cellular structures.

3.5. SOME OPEN QUESTIONS

The statistical analysis of dislocation cellular structures and of dimple patterns has never
been performed having in mind to get from these informations some hints about the
conditions for their formation. In this respect, all is still to be done, both experimentally
and theoretically.

4. Mechanical behaviour of metallic foams

Even closer to the classical foams (such as the polymeric foams) metallurgists have
develloped metallic foams which present interesting properties as thermal insulators
resitant to fire, good specific properties (properties divided by the density), good energy
absorbers in a mechanical crash. Possible applications of these materials are expected in
building as well as transport industries. Seen as a composite metal-air, the simplest
structure is the so called honeycomb structure. In order to be efficient mechanically it
has to be loaded perpendicular to its plane. Metallic foams such as the Al-SiC foam (the
matrix is a metal matrix composite) have the advantage of having isotropic mechanical
properties. Their obvious drawback, since they are not regular, their properties are much
less accurately controlled.

Figure 16

~
2cm
....
a) Honeycomb structure [29); b) Al-SiC foam [30)
168

There are different methods to obtain metallic foams [8]. A clever but expensive one is
to start from a polymeric foam and deposit the metal from the vapour phase and then
burn the foam. The foams we are discussing now are "cheap" ones: the matrix is a metal
matrix composite, th~ metal is Aluminium alloys, the reinforcement is either Al20 3 or
SiC particles. The foam is obtained by bubbling a gaz while the metal is semi-solid (i.e.
between the solidus and the liquidus line). The role of particles in stabilizing the foam
is not yet completely established, whether they increase the effective viscosity of the
liquid, or they prevent holes in the wall from growing before the wall is fully solidified.
The whole procedure to obtain the cellular solid is still. very empirical, the size am
volume fraction of particles need to be in a definite range to form a stable foam, as
shown in Figure 17.
This process allows to make foams with variable densities, and for a given density, the
size of the bubbles may be varied. Though the foam is in principle formed of closed
cells, most of the metal is in the "Plateau Border" so that the mechanical properties are
expected to be closer to those of an open cell structure.

Figure 17


~

-~

~..... 20
PARTICLE
w
:::E s~
3 SEVERE
~
~ 10
~ DIFFICULT
TO MIX
~
0 "10
0.1 LO 100
PAfmCLE SIZE, J.l m

Processing diagram for MMC foams [27]

4.1. REGULAR AND IRREGULAR METALLIC FOAMS

The standard textbook on cellular materials derives the properties of foams from the
properties of the materials assuming that the foam under consideration is regular. As a
result of this assumption, the compression curve is expected to look like Figure 18. Fist
an elastic response corresponding to the reversible bending of the cells, then a critical
169
stress at which all cells should collapse simultaneously since they are supposed to be all
identicals, then when the collapse is complete, a densification sequence during which the
solid's properties come closer to the one of the bulk material.

Figure 18

\~
I.

Theoretical curve describing the behaviour in compression of a regular foam (such as a honeycomb).

4.2. MECHANISMS FOR ELASTICITY AND PLASTICITY IN IDEAL REGULAR


FOAMS: MATERIALS VS STRUCTURE

Ashby and Gibson [8] have proposed models of a remarquable elegance and simplicity
to predict the properties of foams from the properties of the bulk material and the
characteristics of the assumed regular cellular structure. The elastic deformation of the
foam is governed by the elastic bending of the beams (or the walls), the plastic threshold
by the collapse stress of the cells, which may occur by elastic buckling of the bealms,
plastic yielding of the metal or fracture of the metal.
As an example, here is the derivation of the elastic properties for an open cell foam. The
notations are the ones of figure 19. The applied stress CJ gives on each bar a fon:e
o
F=CJIL2• The bending of the "horizontal" bar gives a deflection proportional to FL31El
where E, is the modulus of the bulk material. The resulting macroscopic strain is oiL.
Therefore the elastic modulus E of a foam of relative density pip, is proportional to
E.(p/p.)2 • This scaling law is remarquably well satisfied for a wide range of foamed
materials.

Figure 19

Schematics of an open cell foam


170

Similar scaling laws using the same method were derived for the stress of the plateau.
For a plateau corresponding to cell buckling, the plateau level is also proportional to
E.(plp.)2• For a plateau corresponding to plastic yielding, it is proportional to crsy(p/pJ312
(cr.y is the yield stress of the bulk solid). For a plateau corresponding to fracture of the
walls, it is proportional to cr.~plpJ 312 ( crst is the yield stress of the bulk solid).

4.3. SOME REAL FOAMS: POROUS Sll...ICON AND ALUMINIUM FOAMS

Porous Silicon was certainly not developped to please specialists of the mechanical
properties of cellular solids: it has the advantage (still mysterious) to be a semiconductor
with a direct gap (by contrast bulk silicon has an indirect gap). It is obtained by
electrochemical etching of bulk silicon in a liquid electrolyte. The size on the pores is
nanometric and therefore the drying of porous silicon induces considerable capillary
stresses leading to beautiful fracture patterns. (By the way they are cellular patterns just
like for the drying of clay, but in spite of their nice aspect, people from microelectronic
do not seem to appreciate this phenomenon ... ).
In order to understand the conditions in which this cracking can be avoided, the
knowledge of the mechanical properties of porous silicon is needed. P doped porous
Silicon appears to be a remarquable model system for cellular solids. Figure 20 shows
the Young modulus (measured by instrumented nanoindentation) as a function of
porosity, and the excellent agreement with Ashby-Gibson formula.

--=..
II
60
Figure 20

""'=
._,
50

=-,::
0
40

30
e
:zo
"'
-co
=
=
10
0
> 0
0 0.1 O.l 0.3 0.4 0.5 0.6 0.7
Relative density
Elastic properties of p-doped Porous Silicon [28]

An interesting feature of the simple Ashby-Gibson formulae is that, among the


structural parameters of a cellular solid, only the relative density of the foam matters.
Therefore, testing two foams with the same relative density and different average cell
sizes should'nt make any difference. Figure 21 shows that this is not the case. The
elastic properties are roughly identical, and approximately given by Ashby-Gibson
formulae. But the plastic properties are quite dissimilar. There is no horizontal plateau,
and the beginning of the collapse is about 10 times lower than predicted by the models.
The qualitative reason for this discrepancy is that the collapse doesn't start
homogeneously, but at the weakest cell (the largest one for buckling of plastic yield).
For larger average cell size, collapse starts at a lower stress. The slope of the plateau is
related to the dispersion in cell size: a larger stress is necessary to provoque further
171

collapse of smaller cells. Assuming independant cells, a collapse by elastic buckling,


and with a cell size distribution of width L1 around an average cell size D, the plateau
slope should scale as !l/D3•

Figure 21
rr6
a."
~ 5
...
"'
~ 4

"' 3

o_s ,f.
Compression curves for two aluminium foams having the same density and different cell size [30).

4.4. SIMPLE IDEAS ON AN IRREGULAR CELLULAR SOLID: THE MODIFIED


HONEYCOMB

The previous § has clearly shown that irregularities in cellular structures introduce new
features in their mechanical properties. A simple way to control disorder is to start from
a perfectly ordered structure, say a honeycomb, and introduce progressively the disorder.

Figure 22

a) honeycomb partially filled with wax during deformation


b) honeycomb with a weak cell during deformation
172

Figure 23

100

80

..
~ 60

!
40

zo
Compression Direction L

00 20 40 60 80 100
Deformation (%)

Compression curve for a honeycomb partially filled with wax

Two experiments were performed in this direction, in order to introduce either stronger
cells or weaker cells in the regular structure: either filling some cells of the honeycomb
with wax, or cutting some bonds in some cells. The results are shown in Figure 22.
Figure 23 shows the influence of disorder on the slope of the plateau, as expected. But
another feature is that the collapse of the weak cell (figure 22b) induces a localisation of
plastic flow. This indicates that coupling between cells has to be taken into account in
order to understand plasticity of cellular solids. A simple model to include this aspect is
to describe the foam by a Voronoi tesselation, to attribute to each cell the behaviour
given by Ashby-Gibson formula for its own characteristics, and to couple the cell with
its neighbours in order to include the fact that a collapsed cell will trigger the collapse of
its neighbours. The ratio of the dispersion in strength to the coupling stress controls
both the slope of the plateau and the localised nature of plastic flow [30].

4.5 OPEN QUESTIONS

Metallic foams are definitely fascinating materials. The conditions for their stability
(which has to last longer than the time to solidify) are still poorly understood. Their
elastic behaviour seems appropriately described by the classical formula, but the disorder
plays a crucial role in their plastic behaviour. Contrarely to soap froth they cannot create
topological defects (each cell keeps its neighbourgs) but they have structural defects
(cells weaker or stronger than average) wich control the onset of plasticity and its future
spatial structure. The instabilities observed may serve as simpler analogs to plastic
instabilities observed in bulk metals.
173

5. CONCLUSIONS

In this short introduction to physical metallurgy, we have frequently come across


cellular structures: the well known granular structures, dislocation and dimple patterns in
ductile metals, and finally metallic foams. The questions addressed by physical
metallurgists for these different situations are mainly, for the grain structure its thermal
stability, for the dislocation structure , the reason of its apparition, for the metallic foam
the influence of the structure on the mechanical properties. The tools used in physical
metallurgy to model these phenomena range from the "back of the enveloppe"
calculations to the "intensive computer simulation".

In all these questions, open problems remain to be solved. What physical metallurgist
are interested in is not so mutch "what is general in cellular structures?" but "what can
the structure teach us about the underlying mechanism?". What thay are interested in is
not so much "how does behave a regular cellular structure?" but "what are the roles of
defects and irregularities on the macroscopic behaviour?". The various examples
proposed in this lecture would certainly benefit greatly from the joint approach of
physical metallurgy and physics of cellular structures.

ACKNOWLEDGEMENTS

The original work presented in this paper has benefited from collaborations with
Pr.D.Bellet, Pr. D.Embury, Dr.P.Lamagniere, Dr.J.Lepinoux, Pr. F.Louchet,
Pr.H.J.Roven, Dr.O.Prakash, and graduate students P.Bichebois, G.Heiberg and
O.Belmont.

REFERENCES

General references
[1) M.Ashby, D.Jones, Engineering Materials volt and 2, Pergamon Press (1986)
[2) P.Haasen, Physical Metallurgy, Cambridge University Press (1992)
[3] "Materials science and technology", R.Cahn, P.Haasen, E.Kramers ed., Vol 5, "Phase transformation",
P.Haasen ed. (1993)
[4] W.Kurz, D.Fisher "Fundamentals of solidification" Transtech.Pub, (19 .. )
[5] R.Baluffi, A.Sutton, "Interfaces in crystalline materials", Oxford University Press (1995)
(6] "Computer Similations in Materials science" , H.Kirchner, L.Kubin, V.Pontikis ed. , Kluwer Academic
Publisher, (1995)
[7] "Materials science and Technology", R.Cahn, P.Haasen, E.Krarners ed., Vol4., "Plasticity and fracture",
H.Mughrabi ed., VCH Publ. (1993)
[8] L.Gibson, M.Ashby "Cellular solids, Structure and properties", Pergamon (1988)

Specialised references
[9] Hasson G.C. and Goux C. (1971), Scripta Metall., 5, 889.
[10] Kaur 1., Gust W. and Kozma L. (1989), Handbook of Grain and Interphase Boundary Diffusion,
Ziegler Press, Stuttgart.
[11] Cahn J.W., (1962), Acta Metall., 10, 789.
[12] Anderson M.P., Srolovitz D.J. , Grest G.S. and Sahni P.S. (1984), Acta Metall., 32, 783.
174

[13] Atkinson H.V. (1988), Acta Metall., 36,469.


[14] Anderson M.P., Grest G.S. and. Srolovitz D.J (1989), Phil. Mag. 8 ,59, 293.
[15] Hillert M (1965), Acta Me tall., 13, 227.
[16] Kawasaki K., Nagai T. and Nakashima K. (1989), Phil. Mag. 8, 60, 399.
[17] Weygand D., Brechet Y. and Neda Z. (1997), Phil. Mag. 8, 75,937.
[18] Rollett A.D., Srolovitz D.J. and Anderson M.P. (1989), Acta. Metall., 37, 1227.
[19] Molodov D.A., Czubayko U. Gottstein G., Shvindlerman L.S., Staumal B. and Gust W. (1995), Phil.
Mag. Lett., 72, 361.
[20] Schwindlerman L.S., Molodov D.A., Gottstein G., To be published (1997)
[21] Friedel J. "Dislocations", Pergamon (1966)
[22] Nabarro F. ed. "Dislocations", Vol I to 9, North Holland, (1992)
[23] L.Kubin in "Materials science and Technology", R.Cahn, P.Haasen, E.Kramers ed., Vol 4., "Plasticity
and fracture", H.Mughrabi ed., VCH Pub!. (1993),
[24] Ashby M., Ghandi C., Taplin D. (1979), Acta.Met. 27, 699
[25] Rice J., Tracey D., (1969) J.Mech.Phys.Sol. 17, 201
[26] Brown L., Embury J.D. ( 1973) Proc.ICSMA 3. , Institute of Metals
[27] Prakash 0., Embury J.D., (1995), Mat.Sc.Eng., A199, 195
[28] Bellet D, Lamagniere P., Brechet Y., Vincent A., (1996) J. Appi.Phys, 80, 3772
[29] Prakash 0., Bichebois P., Brechet Y., Louchet F., Embury J.D., (1996) Phil.Mag. 73, 739
[30] Heiberg G., Flaujac.L, Brechet Y., Roven H.J. To be published
THE COMPRESSION OF CLOSED-CELL POLYMER FOAMS

N. J MILLS & H. X. ZHU


School of Metallurgy and Materials,
The University of Birmingham, Bl5 2TT, UK.

Abstract The tensile response of oriented polyethylene and polystyrene films provides
material data for the modelling of closed-cell polymer foams. We analyse a Kelvin
foam, with 60% of the polymer in the cell faces, compressed in the [001) direction. Cell
edges bend and compress axially, while cell faces act as membranes. The tensile strains
across some face diagonals are 40% of the foam compressive strain, so tensile yielding
can occur. The predicted Young's moduli are slightly low, because compressive face
stresses are ignored, but the Poisson's ratio is correctly predicted. The predicted yield
stresses for polyethylene foams are close to experiment. Polystyrene extruded foams
may collapse in compression when face yielding commences.

1 Introduction
Three aspects of the mechanics of closed cell foams, made from polystyrene (PS), low
density polyethylene (LDPE) and polypropylene (PP), will be considered: -
a) The material responses: the microstructural state of the thermoplastic polymer must
be considered, as must heat transfer from the gas enclosed in the cells.
b) The foam elements: edges, if long and slender can be considered as beams, and thin
faces, that they buckle easily, can be considered as membranes. The high strain
deformation of these elements must be related to the applied forces.
c) A cell model: the proportions of edges and faces should be the same as in the real
foam. A regular lattice of cells or an irregular microstructure can be analysed.
Prior research includes a Finite Element Analysis (FEA) of the Young's modulus in the
[001] direction of the Body Centred Cubic (BCC) lattice oftetrakaidecahedral cells (the
Kelvin foam) shown in figure 1 (Renz and Ehrenstein, 1982). They assumed that all the
polymer was in the cell faces and that it was linearly elastic. Their graphical results for
the variation of the Young's modulus E1 of polyvinylchloride (PVC) foam with its
relative density R follows a linear relationship,
E 1 = 0.33£ R (1)
where E is the Young's modulus of the polymer. This was confirmed by Kraynik
(1997), who computed the constant in equation (1) to be 0.299 for compression in the
[001] direction, and 0.322 for compression in the [Ill] lattice direction. The upper
bound (uniform strain) modulus of a composite of polymer and air is ER. Equation (1)
predicts a modulus that is a third of this upper bound, as a result of the variety of cell
face directions.
175
J. F. Sadoc and N. Rivier (ed.<.). Foams and Emulsions. 175-192.
© 1999 Kluwer Academic Publishers.
176

z [00 1) O"z

-------'if
I
b

edge secllon

Fig. I A tetrakaidecahedral closed cell, plus a vertical face BCRS. The section SS' contains the side face
of the structural cell. The insert shows a Plateau border edge section, with the neighbouring parts of the
faces, which act as reinforcing ribs.

No quantitative analysis of high strain compression exists. Gibson and Ashby (1988)
asserted that elastic collapse occurs in some closed cell foams at a constant stress a:,
when axially compressed cell edges buckle. They ignored the contribution of the cell
faces, which were treated as thin membranes. Their effectively two-dimensional model
lead to the conclusion that a:,
scales with the square of Re , the volume fraction of cell
edges, but they could not make quantitative predictions of the yield stress. Their Euler
buckling analysis was for the onset of buckling, so did not prove that the stress remains
constant at higher strains. The Elastica analysis for high deformations of a strut that is
initially axially loaded (Timoshenko and Gere, 1961) predicts that the load rises
significantly after initial buckling.

2 Material responses
The response of closed-cell foams varies. After compression to 80% strain, LDPE foam
recovers its shape completely, whereas high density polyethylene (HDPE) foams of
density 50 kg m· 3 have permanent strains of less than 10%, (Loveridge and Mills, 1991 ).
There are wrinkled faces in the foam after recovery, due either to permanent stretching
of the faces, or permanent bending of some of the thicker edges. Plastic hinges occur in
PS foam cell faces if the compressive strain exceeds 10% (Mills, 1994). The permanent
strain is approximately half of the maximum foam compressive strain, for PS bead foam
of density 50 kg m·3 .
Modelling predicts that high tensile strains occur in the cell faces, but it is not currently
possible to obtain tensile stress-strain data for cell face material. Therefore polymer
films, in a similar microstructural state to the foam faces, were examined. The foaming
177

process stretches the molten polymer biaxially, then the faces solidify rapidly. If the
polymer melt deforms entirely elastically, rather than flowing viscously, the biaxial
extension ratio (assuming uniformly spaced nuclei on a BCC lattice) is a maximum of
4.6 for a relative density of 0.05, and 6.3 for a relative density of 0.02. The LDPE foams
tested (from Zotefoams pic, Mitcham, UK) are crosslinked, so molecular orientation
remains as the polymer crystallises.
LDPE film, biaxially stretched in the blown-film process, can be highly oriented. Figure
2a shows the tensile stress-strain curve of a LDPE packaging film, of density 910 kg m'3
and thickness 45 !liD, having a draw ratio of 6.5 (measured by reversion in hot silicone
oil). The film yields at a stress of 10 MPa at a strain of 8%, and the stress rises linearly
to 20 MPa at a strain of 90%. Its Young's modulus up to 1.5% strain, measured on a
200 mm long specimen at a strain rate of 7 x 104 s· 1, was 202 MPa. The sample extends
uniformly, unlike LDPE film of lower orientation, which necks and cold draws. No
necks were observed in the faces of deformed LDPE foams, so their yielding response is
similar. The Young's modulus increases with strain rate. Creep compliance data for the
solid LDPE in the linear viscoelastic region (Mills and Gilchrist, 1997b) was
extrapolated to an impact time scale of 0.01 s to give Young's modulus of 300 MPa for
impact modelling. Since the modulus of PE depends on the crystallinity and on
crystalline orientation, the film Young's modulus is approximation to the foam face
modulus. Figure 2b shows the tensile stress-strain curve of Dow Window Film 6003£
of thickness 32 !liD, measured in the machine direction. This biaxially oriented PS film,
of draw ratio 5.5, yields at a stress of 70 MPa and strain of 3.5%, then extends nearly
uniformly at a constant yield stress. This contrasts with bulk unoriented PS, which fails
by crazing at a stress of about 40 MPa. The Youngs modulus, for 0 to 1% strain, is 3.08
0 5 10 strain% 15 20
80

20
60
CIS
a. 15
:!
t/) 40

-...
t/)
Cl)

t /)
10

20
5

0
0 20 40 strain% 60 80 100 °
Fig 2. Tensile stress strain curves for oriented polymer films
a) 45 J.liD thick LDPE with an extension ratio of 6.5, b) 32 J.liD thick PS with an extension ratio of 5.5.
178

GPa at a strain rate of 7 x I o-4 s- 1, the same as for unoriented bulk PS_
The polymer is assumed either to be a linearly-elastic isotropic material, with Young's
modulus E and Poisson's ratio v, or to be an elastic-plastic material, with a yield stress
that increases linearly with strain_ Although polymers are viscoelastic materials,
viscoelasticity will not be considered_ The predictions are strictly of isochronous stress-
strain curves, and should use polymer moduli for the appropriate time scale_
Viscoelasticity is important in open cell foams, where the creep buckling of edges leads
to compressive collapse (Zhu and Mills, 1997, Zhu, 1997).
The gas in the cells is air, since gases can diffuse in and out of LDPE foams on a time
scale of weeks. However there is negligible gas escape by diffusion through the cell
faces for load application times less than thirty minutes. The surface area of the cell
faces is so high that there is effective heat transfer, by conduction and forced
convection, from the compressed air to the polymer (Mills and Gilchrist, 1997), even
when the foam is impacted at 5 m s- 1• As the thermal mass of the polymer greatly
exceeds that of the air, the absolute gas pressure p in the compressed foam can be
calculated assuming that it remains at ambient temperature, using
pVK = PoVKo (2)
where p 0 is the gas pressure in the uncompressed foam, and Vg is the total volume of gas
in the foam (initially Vag). The cell gas occupies a fraction 1-R of the volume of the
undeformed foam. If a foam of constant Poisson's ratio Vf has a uniaxial compressive
strain liz applied, its volume v is Vo (I- liz XI + v fli z r' where Vo is the initial foam
volume. Therefore the gas pressure is
p 1-R
Po = (1 - &J1 + v j - R
1&
(3)

3 Element mechanics
The cell edges deform by bending and by axial compression. When 90% of the polymer
is in the faces, the edges have a length to diameter ratio >5 when the foam relative
density < 0.2. Therefore the edge deflection due to shear will be neglected. The analysis
is not valid when the foam relative density exceeds 0.2, but this is above the commercial
limit of PS and LDPE foams. The edges are assumed to have Plateau border sections,
consisting of three touching circular arcs (figure 1), because of the surface tension of the
melt. The radius of curvature is equal to the edge breadth b. To calculate the elastic
curvature of an edge from the applied bending moment M, the second moment 1 of the
edge cross-sectional area A is needed. This is given (Kraynik & Warren, 1994) by
1=0.1388A 2 (4)
For high deflections, a curvilinear co-ordinates is used for the position along the edge,
e
and the edge curvature is dB Ids, where is the local orientation of the edge relative to
the z axis. The shape of edge BC (Figure 3) is found by integrating its curvature and
using the end condition that (} = n/4 at B, assuming that it remains elastic, using
179

I
I
I
I
z
·1
I
I
I
I
I
I
I
I

t.-----------y--------------
1

Fig 3 The forces and moments acting on the deformed edge.

Eld(} =M (5)
ds
At vertices, reinforcing webs between the edges prevent vertex deformation. An
unknown moment M"' acting at B on the edge BC, maintains the end condition that
0= rc/4. Mx does not affect the shear force on the edge, which can be related to the
differential of equation (5)

Ed2()
I- !"'
2 = -FsinO+ NcosO+cos() [/1(s)- / 2 (s)]ds (6)
ds o
The terms on the right hand side of equation (6) are from the vertical and horizontal
forces F and N at the vertex B, and from the integral of the face tensions j/ and h (see
later). The axial compressive strain in the edge is calculated by resolving the vertical
and horizontal forces along the edge axis
EA& 8 c=Fcos0+Nsin0+ fC.t;(s)-/2 (s))sinOds (7)
The full analysis (Mills and Zhu, 1997) considers the yielding of an edge in bending.
The parts of the edge furthest from the neutral surface begin to yield at a foam strain of
about 3%, but a plastic hinge (yielding across the section) does not form until the
applied moment is three times as great. The edge, supported by three elastic faces, still
cannot collapse. Hence edge yielding alone has little effect on the foam response.
The faces are considered as membranes. They cannot transmit compressive stresses to
the edges, because they can buckle at a very low strain. Consider a face of uniform
width wand thickness o, with built-in ends a distance L apart. The compressive force Fe

e:r
to initiate elastic buckling is given by

~= EI where I= ~~3 (8)


The critical strain to cause buckling is therefore

c, =E~o =~(:or (9)


180

For a relative density of 0.025, and face fraction= 0.9, &L = 0.0189 and the strain to
cause buckling is 0.11 %. The buckling strain will be higher in narrower parts of a non-
rectangular face. The vertical square faces are linked comer to comer in the direction of
the applied stress, so, prior to buckling, the compressive strain across the face diagonal
is the same as the applied strain. The foam Young's modulus is measured at a strain of
about 1%, at which the faces are buckled. The buckling assumption is unrealistic for
higher foam densities, so the predicted Young's modulus will be a lower bound value.

4 Foam microstructures
Thermoplastic foams have variable cell sizes and shapes. Williams (1968) showed that
the average number of faces per cell in a variety of cellular solids is close to 14, and the
average number of edges per face is 5.1. The modelling uses a tetrakaidecahedral cell,
having six square and eight hexagonal faces, shown in figure 1. Each edge, of length l ,
links two hexagonal faces to a square face. The cell diameter D = 2J2 L . All the flat
faces have a uniform thickness t5, related to the face relative density Rt by
li= 16/Rr
(10)
3.fi +6..f6
The Plateau border breadth b is related to the edge cross-sectional area A and edge
relative density R. by

b 2 (J3-1t/2)=A= 2~t 2 R. (11)


Scanning Electron Microscopy (SEM) of sectioned foams shows that the face
thicknesses t5 are small and slightly non-uniform, so difficult to measure accurately.
The faces are flat (figure 2), because of the high biaxial tensile stresses in the faces as
they solidify. The mean edge breadths and lengths were measured (Table 1), then the
edge relative density calculated using equations (10) and (11), and compared with the
total relative. The fraction of polymer in the faces¢= R1 / R varied from 0.97 to 0.99
for PS foams and was slightly lower for LDPE' foam.

Table I Microstructure of the foams examined


Polymer density mean edge length mean edge breadth face fraction
kgm"3 [ f.lm b f.lm ¢
Polystyrene 20 124 4.7 0.98
LDPE 24 620 62 0.93
Polypropylene 30 81 5.4 0.98

Although the edge fraction I - ¢ is very low, parts of the faces close to an edge act as
reinforcing ribs for that edge. When this ribbed edge beam is bent, its ribs resist
buckling more than does the central region of a face. Therefore it is appropriate to treat
the foam as having a lower effective ¢value than those in Table 1.
181

Fig. 4 SEM micrographs ofLDPE foam of density 24 kg m·3 a) complete cells showing the flat faces, b)
detail of the section through an edge and the 3 adjoining faces.

5 Foam structural model

5.1 GEOMETRY AND CELL FACE DEFORMATION


A regular lattice model was chosen because it is easier to analyse than a random
collection of cells. A BCC lattice of tetrakaidecahedral cells had been used successfully
to analyse high strain open cell foam deformation. When the lattice is loaded along a
symmetry axis, the number of edges and faces to be analysed is minimised.
Compression in the [00 I] lattice direction is easier to analyse, since the edges only bend
and compress axially, than compression in the [III] direction, where the edges
compress, bend and twist. The cell size, assumed to be much smaller than the foam
block dimensions, does not affect the compressive response. Figure 5 shows the
structural cell, bounded by mirror planes acting for both the initial and deformed
structures. For a uniaxial compressive stress in the z direction, the structural cell shape
does not shear. The co-ordinate axes are chosen so that the edge BC lies in the yz plane.
Inside the structural cell there are three two-fold rotational symmetry axes through C,
parallel to the structural cell edges, so only a quarter of the structural cell needs to be
analysed. This contains a quarter of a vertical square face, a quarter of a horizontal
square face, and two quarters of hexagonal faces. The term horizontal is used as an
abbreviation for ' lying in the xy plane' and vertical for ' containing the z axis'. Since the
hexagonal faces are equivalent (figure 4b) only three types of face are analysed, together
with the slanting edge BC and half the horizontal edge AB .
The edge BC, bending while remaining in the yz plane, is considered to consist of I 00
segments, initially of uniform length. Its shape determines the strains in the hexagonal
and vertical square faces . Figure 5 shows a quarter of the vertical square face BCRS.
182

I
I
I
'!I :·nL
I
I I
I I

- -- - - - - - --
I I
I I
-~-
I
~-
I
' I I
', I I
" I I
' I
--- -- ----- --- ------ -~
~----------2L -----------4~
Fig 5 Structure cell with mirror plane boundaries that act for the deformed structure. The three types of
faces are shaded differently, and the horizontal tensile forcesjj(s) andfH(s) are shown

The face can be considered to consist of horizontal parallel ribbons, of initial length
(!- s) sin 45°. The horizontal tensile strain evs in the ribbon is given by
& 1x ==--
l- s '
f' .
.fi smBds-1 (12)

If the polymer is elastic, the tensile stress a- vs is this strain multiplied by the Young's
modulus E. However the tensile stress cannot become negative, since the face buckles at
a negligible stress. In the elastic-plastic version of the model the tensile stress is related
to the strain by
If &< 0 a- == 0
If t: < t: r a- == Et:
(13)
If t: > t: r a- == 1'; + t:J";
where Y1 and Y2 are yield stresses. The ribbon cross-sectional area (in the yz plane) per
o
edge segment is ds0 / .fi . This area does not change as the edge deforms, but the face
may form horizontal wrinkles. The horizontal tensile force /J(s) per segment of BC is
therefore
0 (14)
J,(s)ds == ~O"v~ds0
-v 2 ·
It is assumed that horizontal square faces such as ABQP remain square, but that their
edges contract in length by an amount 2L1 . Each edge of the square is connected to two
183

inclined hexagonal faces. The latter are assumed to contain only horizontal tensile
stresses, so exert no forces on the edges of the square face. Horizontal forces N act at B,
along the positive y-axis, on edges BC and BS. These are in equilibrium with the axial
compressive forces in edges AB and AQ which are N .fi . !1 is small compared with the
other deformations; it is large enough to cause the face ABQP to buckle, but the
force N .fi is not large enough to make the edges AB buckle.
The hexagonal face ABCDEF stretches horizontally, but face buckling relieves any
compressive stresses in the direction from A to E. It acts as a set of horizontal ribbons
in which the tensile strain is
eHu = ( .fi J:sinBds-s-2!1) j{z+s) (15)
In the elastic model, the stress a Hlix is Young's modulus times eH,,x, with the condition
that the stress cannot be negative, and in the elastic-plastic model, it is given by
equation ( 13 ). The ribbon cross-section per segment of BC is ods0 .J3/2 , so the
horizontal force fH(s) per edge segment is
o.J3
fH(s)ds=-aHEXdsO (16)
2
Two hexagonal faces exert tensile forces of the same magnitude on BC, and their vector
sum.fi(s) acts along the negative y-axis, having magnitude

/ 2
r;;
(s)ds=v.t.fH(s)ds=
o.J3
J2 am;xds0 (17)

5.2 THE FORCES ON THE STRUCTURAL CELL


The vertical forces F, acting on the top face of the structural cell (figure 5), are in
equilibrium with the external stress and the gas pressure. The absolute gas pressure p in

l
the foam is initially atmospheric Pa· Using equation (3), the relative pressure p, is

p,=p-pa=Pa ( 1-R 2 -1 (18)


(1- sz}(l + v 1 &,) - R
The gas contribution to the true stress is p, so the engineering stress on the foam is
a,= : 2 + p,(l+v1s} (19)

The left side face of the structural cell cuts: -


1) the hexagonal face ABCDEF, with a tensile stressfH(s) per unit length ofBC
2) the buckled square face ABPQ, and the edge AB with a compressive force N .fi
3) the cell gas under relative pressure p,.
Edges AB and DE are shared by two structural cells, so on average there is one such
edge per structural cell. The force equilibrium perpendicular to the side face requires
-N.fi + 2 JjH(s)ds = [Ac(1- sJ(1 + Vf&z)- Ap]p, (20)
184

where the undeformed cross-sectional areas of the structural cell side face and the
polymer are given respectively by
A,. = 2-Ji L2 A
p
= .J3
4
b 2 + lt5 + .f3tt5
2 (21)
Substituting equation (18) in equation (20) gives
N= Jj (s)ds-G I .J2
2 (22)
where G is the horizontal force exerted by the compressed gas on the side face, given by
G= p,[
A, (I- &;}(1 + v1 sJ- AP] (23)
The compressive force .JiN causes an elastic contraction Ll in half of edge AB of
.JiN(l 12)
~= 9~
EA

5.3 ITERATIVE SOLUTION


Equation (6) is solved by iteration, using equations (7) and (12-24) in the algorithm. The
main contributions to the shear force on the edge are from the vertical force F and the
tensile stresses in the vertical square and hexagonal faces. Starting with the undeformed
shape Bo1d(s) = ff/4, the engineering stress was incremented by an amount chosen to
give a foam strain increment of 1% (or 0.1% when the Young's modulus was being
determined). A new value ofF was obtained from equation (19), then the rate of change
of curvature d 2 (} I ds 2 obtained from equation (6). This was integrated twice
numerically. The constants of integration are determined by the end conditions
B(l) = B(O) = n/4, which means that the average edge curvature is zero. The edge
orientation is therefore
(} . (s) = f' ( f' e" (s)ds)ds- !_ r' ( f' e" (s)ds)ds + !!__ (25)
""w Jo Jo f Jo Jo 4
where (}"represents the second derivative d 2 (} I ds 2 • The iteration was repeated, using a
revised value of (} that is a weighted average of (}new and Bold, until the maximum
change in any value of 8( S) was less than 5 X10-J radians.
The projected lengths YI and z 1 of the edge BC along they and z axis (figure 3) were
used to calculate the macroscopic compressive strain ofthe foam&z
& = t/.Ji -zi = I-.J22 (26)
z t/.Ji l
and the foam lateral strain, using the elastic contraction 2 ~ of the edge AB from
equation (24)
& . = YI +.Ji~-t/.Ji _l:J___~-- (27)
) .Jiz .Jiz z 2
185

The Poisson's ratio of the foam is the ratio of the lateral tensile strain to the compressive
strain along the z axis. The horizontal tensile strains in the vertical square and
hexagonal faces are given respectively by equations (12) and (15). The maximum
bending strain in edge BC occurs at the point furthest from the neutral surface, and is
b dB
E: =-- (28)
e J3 ds

6 Young's modulus and Poisson's ratio


Calculations were made for a foam relative density of 0.025, with a range of relative
face densities ¢(Table 2). Some calculations fail to converge for ¢ > 0.65, since the
edge bending stiffness is low compared with the face tensile stiffnesses. Much of the
predicted behaviour is insensitive to ¢values> 0.6, so extrapolation to ¢ = 0.98 for PS
foam is possible. Face regions close to an edge may act as extensions of the edge if they
do not buckle, so the insensitivity to ¢is understandable.

Table 2. Predictions for LDPE foam of relative density 0.025, edges of Plateau section
and Elastic material withE = 300 MPa.
relative face edge face Foam F G N Poisson's
density width thickness modulus ratio
<I> b 8 Er at 3.0% strain
~m ~m kPa ~N ~N ~N
0.01 38.0 0 203 1210 1 11 0.44
0.2 34.1 0.42 396 197 68 30 0.32
0.4 29.6 0.84 455 224 96 47 0.22
0.6 24.1 1.26 423 193 61 87 0.14
0.65 22.6 1.37 396 185 76 72

The edge widths and face thicknesses in Table 2 relate to an edge length of 100 jlm; if
the edge widths and lengths and the face thickness are increased in proportion, this has
no effect on the predicted modulus and strain. If L changes the forces F, G and H change
in proportion with L 2 . An increase in relative face density causes a significant increase
in the foam Young's modulus when ¢ < 0.2. A check was made that the Young's
modulus at zero face fraction is equal to the gas contributionpa(l-20 plus the open cell
foam contribution (Zhu, Mills and Knott, 1997).
Young's modulus, predicted for PS foam using¢= 0.6, is compared with the authors'
unpublished experimental data for the secant Young's moduli at 1% strain in figure 6.
Experimental data, for PS bead foams with densities from 21 to 80 kg m-3 , is fitted by a
line with the equation
E I = 0.977 E R 1627 (29)
taking the Young's modulus ofPS as 3.0 GPa. There is data for PS extruded foams of
densities 30 and 44 kg m-3 . The latter point does not fit on the bead-foam trend line,
186

100

-- - - -
I'll
a.

- ---- - •
--- •• •
::!iii

-----
1/)
:::J FEA
:::J
'1::1
0
E 10
1/)
"tn
c

:::J
0
>-

10.01 0.1
relative density
Fig 6 Predicted variation of Young's modulus of PS foam with density compared with the authors'
e
experimental data for PS bead and extruded • foam, and FEA predictions.

probably because there is a significant density variation across the sample. The
predicted foam modulus (for ¢ = 0.6) increases nearly linearly with the foam relative
density, following
El = 0.0598E R 1066 (30)
As the experimental Young's modulus increases with the 1.63th power of density, the
deviation between the prediction and experiment increases with density. The model
predictions are good at low relative densities, where the face buckling strain is low.
Clutton and Rice's (1990) experimental data, for the I% strain tensile Young's modulus
of LDPE foams, is compared with the predicted variation of relative foam modulus with
relative density in figure 7. The prediction line, which has the equation
E 1 =0.0807E R 1155 (31)
is smaller than the experimental tensile Young's moduli by a factor of 3.5. However the
measured compressive Young's modulus of PVC foams was 52% of the tensile Young's
modulus (Menges and Knipshild, 1982), presumably because cell faces buckle only in
compressive tests. This, and the uncertainty about the face Young's modulus for LDPE,
means that the predictions in figure 7 are reasonable.
The Young's moduli of closed cell foams cannot be considered as the sum of separate
functions of the face, edge and gas parameters. The gas pressure interacts with the
polymer stresses, and the edges and faces reinforce each other in a complex way.
Nevertheless the face contribution dominates the Young's modulus for foamed
thermoplastics with relative densities lower than 0.1 0, and the gas contribution is
insignificant for higher density foams made from high modulus polymers.
For an LDPE foam compressive strain of 10% (using the simulation in the fourth row of
table 2), the central horizontal band of the vertical square, where s < 0.2, had a tensile
187

w
;;::: 0.01
w
t/)
:;,
-- --
_,
_..,. -- -- ---
:;,
"'0 data

.
0
E
Cl)
>
.... -... -- ........... model

co 0.001
~
•-- --
0.01 relative density 0.1
Fig ?Predicted variation of relative Young's modulus ofLDPE foam with relative density, compared with
Clutton and Rice's tensile data (1990).

strain of about 4%. The central band of the hexagonal faces, where s > 0.6, had a tensile
strain of 1.5%. Thus the vertical square faces are the most highly strained. The
maximum edge bending strain was small in the central region but reached 15% and -6%
respectively at B and C, at the Plateau border cusp farthest from the neutral surface. The
axial compressive strain in the edge ranged from 2 to 3%.
The predicted Poisson's ratio for ¢; > 0.6 (table 2) is close to the experimental value.
For PS bead foam of density 50 kg m-3 , the compressive Poisson's ratio is initially about
0.08. When yielding occurs at a strain of 5% the lateral strain becomes constant ± 0.2%,
indicating a change in deformation mechanism. The low strain Poisson's ratios for
extruded PS foams of densities 30 and 44 kg m-3 were 0.10 and 0.13 respectively,
showing that the lack of voids at bead boundaries only slightly increased Poisson's ratio.
In contrast the Poisson's ratio of a 24 kg m- 3 density LOPE foam, compressed on a 1
minute timescale, was 0.17 (Mills and Gilchrist, 1997a).

7 Yielding
LOPE foam of relative density 0.025 was modelled using tensile yield stresses Y 1 = 12
MPa and Y2 = 10 MPa, E = 300 MPa, ¢;=0.6. The change in shape of the edge BC is
shown in figure 8; the origin of the y-axis is at the comer of the structure cell. The size
of the horizontal square face hardly changes, whereas the edge BC buckles slowly as the
foam strain increases. This causes the point at the centre of the structure cell to move in
the y direction. Poisson's ratio at high foam strains is predicted to be 0.2. Figure 9
compares the predicted stress strain curve with experimental data for a foam of density
24 kg m- 3• The agreement of the loading curve with the theory is excellent.
188

0.6

-"'
C1l

c
...0
"0
0.4
I
0
C)

0.2

0.8 1
y co-ordinate
Fig 8. The predicted shape of the edge DC at 0, 10, ... 50% strain, for LOPE foam of relative density
0.025 , with elastic edges ofE = 300 MPa, face fraction 0.6, face yield stresses Y,= 12 and Y 2=10 MPa.

250

200

Ill
0..
.::.: 150
lh
lh
...
Q)

Cii
100

50
-- -- -- ----
0
0 10 20 30 40 50 60 70 80
strain%

Fig. 9 Predicted compressive stress-strain curve - - and polymer contribution - - - for LDPE
foam of relative density 0.025, modelling parameters as in fig. 8, compared with impact loading data
for an LOPE foam of the same relative density. .. .............. ..
189

The stress at E = 30% is 60.7 kPa, of which 32.6 kPa is from the polymer, if both the
edges and faces can yield (this contrasts with a predicted total stress of 90.3 kPa if the
polymer remains elastic). The edges begin to yield at 2.5% foam strain, and the faces
begin to yield at 9.3% strain. The gas pressure provides the main part of the response at
high strains, and the stress borne by the solid polymer falls at strains > 25%. The
predicted Poisson's ratio increases, from 0.13 prior to the faces yielding, to 0.2 at high
strains, so only the pre-yield predictions agree with experiment.
Figure 10 shows the predicted compressive stress strain curve for a PS foam of relative
density 0.03, assuming that the polymer yields at 2.33% strain with a constant yield
stress of70 MPa. As the Young's modulus ofPS is 10 times that of LOPE, the effect of
the gas pressure is small. A stress plateau is predicted to occur at a strain of 15%, but
the stress is double that observed for an extruded foam of the same density, probably
because the predicted yield strain is too high.
4UU

300
--- --
Ill
c..
-"

./
VI /'• ., •• , • • ,, . . . ... - • • u • • •• <;
~ 200
0
/
100
,.

..· ..··
0
0 10 20 30 40 50
strain%

Fig I 0 Predicted stress strain curve curve-- and polymer contributioR- - - for a PS foam of
relative density 0.03, E = 3.0 GPa, face fraction of 0.6, and constant face yield stress of 70 MPa,
compared with loading and unloading data for an extruded PS foam of density 35 kg m·3 . .. . ..... .. .. .. .. ..

Table 3 compares the predicted yield stresses with experimental data for bead foams of
a range of densities. The compressive yield stresses of the bead foams appears to
exceed those for extruded foams of the same density, possibly because the beads have
denser cells near their (moulded) surfaces, strengthening the structure. The predicted
stresses, at which the vertical square faces begin to yield, are close to but slightly
smaller than the experimental yield stresses. Face yielding, starting at a foam strain of
5% to 6%, could trigger a non-symmetrical form of collapse. The predicted yield
stresses at 15% strain are significantly larger than the experimental yield stresses. When
the predicted stresses are plotted against the relative density, they are found to vary with
190

1.11 th power of R. This contrasts with the experimental 1.5 th power variation (Mills,
1994).

Table 3. PS foam yield stresses, forE= 3.0 GPa, ¢1= 0.6, Y1 =70 and Y2 = 0 MPa, with
and WI"thout ed1ge y1e
. ld"mg, compared WI"th expenmen
. tal v alues.
relative stress for elastic edges edges yielding experimental
density face yield stress at E = 0.15 stress atE= 0.15 yield stress
R kPa kPa kPa kPa
O.Ql 65 128 unstable 63
0.02 140 250 238 170
0.03 210 384 371 (220 extruded)
0.04 300 527 494 440

8 Discussion
The model provides quantitative predictions of foam high strain compression, relevant
to cushioning and shock absorption applications, in terms of the polymer modulus and
its tensile yield stress. It emphasises the high face content, molecular and crystalline
orientation in the faces, and the tensile yielding of faces which act as membranes. In
spite of a symmetrical lattice is used, the predictions are good first approximations to
the observed behaviour.
The analysis does not consider edge torsion, which is less likely than in open-cell foams,
since connecting faces stabilise the edges. FEA analysis predicts Young's moduli that
are much higher than experimental values for low density PS foams (figure 6). The
modelling predictions and the FEA analysis are lower and upper bounds on the
experimental data. Equations (5) and (10) show that the compressive strain, at which
face buckling starts, increases with the square of the foam relative density, reaching
1.0% for a PS foam of relative density 0.065 and ¢1 = 1. A better model should consider
the face buckling resistance. The model successfully predicts the low Poisson's ratios of
PS and PE foams prior to yield, but the post-yield Poisson's ratios are too high,
suggesting that the collapse mechanism differs from that shown in figure 8.
An elastic-material model cannot explain the compressive collapse stresses of PE and
PS foams. Firstly, there are non-linear or yielding effects in experimental loading -
unloading data. Secondly the predicted collapse stress of PS foam is too high. Thirdly
the effect of face buckling, assumed in the analysis here but not in the FEA modelling,
is to reduce the Young's modulus but not to cause collapse. Hence Gibson and Ashby's
generalisations that LDPE foams are elastomeric, collapsing elastically at a stress =
0.05 ER2 , are not justified. The tetrakaidecahedral cell shape is a good approximation to
real cell geometry, and for PS and PE foams, gas compression and the tensile yielding of
cell faces are the most important deformation mechanisms. The foam collapses when
yielding spreads to a significant fraction of cell faces, and the slanting edges buckle.
Although the high strain compression of LDPE foams is predicted well, the yield
stresses and strains for PS foams are too high. The stored elastic energy in PS prior to
191

yield is much higher than that in LDPE, which may allow another collapse mechanism
to operate. It is possible that the onset of face yielding, at a 5% strain, triggers a non-
symmetrical collapse mechanism. This would lead to more realistic predictions of yield
stresses and strains. Abramowicz (1994) considered collapse modes of T and Y
junctions between steel plates, and modelled these using folded paper. Local tensile
yielding of the steel 'faces' provides the required misfit displacements where the folded
faces meet. A similar deformation mechanism may occur in PS foam.
The non-linear face analysis is performed for a regular lattice, in which all cells have the
same collapse stress. It would be possible, with greater computational effort, to consider
irregular collections of cells, or other loading directions. The elastic-plastic model
predicts that the polymer contribution to the foam stress reaches a maximum, then
declines significantly when the edge buckle. In a real foam, with a variation of cell
shapes and orientations, the weaker cells will collapse before stronger cells, possibly
smoothing out the peak in the compressive stress.
More details of the analysis are available in a PhD thesis (Zhu, 1997) and in a
forthcoming book on the Mechanics of Polymer Foams (Mills, 1998).

Acknowledgement Mr Zhu is grateful for the support an ORS scholarship. Zotefoams


pic and BASF kindly supplied foams, and Dow Germany supplied polystyrene film.

References
Abramovicz, W.(1994) Crush resistance ofT, Y and X sections, M.I.T. Department of
Ocean Engineering, Report no 24, Cambridge, Maryland.
Clutton E. Q. & Rice G.N. (1991), Structure property relationships in thermoplastic
foams, Cellular Polymers conference, 99-107, RAPRA Technology, Shawbury.
Gibson, L. J. & Ashby, M. F., (1988), Cellular Solids, Pergamon Press, Oxford.
Kraynik, A.M. & Warren, W. E.(1994) in Low Density Cellular Plastics, Ed. N.C.
Hilyard & A. Cunningham, Chapman and Hall, London.
Kraynik, A.M. (1997), NATO/ASI school Foams, Emulsions and Cellular Solids,
Cargese, Corsica.
Loveridge, P.L. & Mills, N.J.(1991), The mechanism of recovery of impacted high-
density polyethylene foam, Cellular Polymers, 10, 393-405.
Menges G. & Knipschild, F.,(1982) in Mechanics of Cellular Plastics, Ed. Hilyard, NC,
Applied Science.
Mills, N.J.(1994) in Low Density Cellular Plastics, Ed. Hilyard N.C. & Cunningham A,
Chapman and Hall, London.
Mills, N.J.(1998) Mechanical Properties ofFoams, Chapman and Hall, London.
Mills, N.J. & Gilchrist A. (1997a), The effects ofheat transfer and Poisson's ratio on the
compressive response of closed cell polymer foams, Cellular Polymers, 16, 87-109.
Mills, N.J. & Gilchrist, A. (1997b), Creep and recovery of polyolefin foams -
deformation mechanisms, J. Cellular Plastics, 33, 264-292.
Mills, N.J. & Zhu, H. X., (1997), The high strain compression of closed-cell polymer
foams, J. Mech. Phys. Solids, submitted.
192

Timoshenko, S. P., & Gere, J. M., (1961), Theory of Elastic Stability, McGraw-Hill,
New York.
Warren, W. E., Neilsen, M.K. & K.raynik, A. M., (1996), Torsional rigidity of a Plateau
Border, to be published.
Williams, R. E., (1968), Space-filling polyhedron: its relation to aggregates of soap
bubbles, plant cells, and metal crystallites, Science, 161, 276-277.
Zhu, H. X.,(1997) PhD thesis, University of Birmingham.
Zhu, H. X., Knott, J. F. & Mills, N. J., (1997), Analysis of the elastic properties of open-
cell foams with tetrakaidecahedral cells, J. Mech. Phys. Solids., 45,319-343.
Zhu, H. X., Mills, N.J. & Knott, J. F., (1997), Analysis for high strain compression of
open-cell foams, J. Mech. Phys. Solids, to appear.
Zhu, H. X., & Mills, N. J. (1997), Analysis of high strain creep in open cell foams,
Cellular Polymers IV conference, RAPRA, Shawbury.

Table of symbols
A,Ac cross-sectional area of a cell edge, the structural cell.
b breadth of a Plateau border edge cross section
E, E1 Young's modulus of polymer, foam
Ji(s), h(s) tensile forces per unit edge length from square and hexagonal faces
G force exerted by compressed gas on a section of the structure cell
F vertical force acting on edge BC in the z axis direction
I cell edge length
I second moment of area of the cell edge cross section
M bending moment acting on edge BC at B
N horizontal force acting on edge BC at B
p,po,Pa pressure in the deformed foam, undeformed foam, atmosphere
R, R., R1 relative density offoam(foam density I polymer density), edges, faces
s length co-ordinate for (bent) edge BC
x,y, z Cartesian co-ordinates
YI ZJ projected lengths of the edge BC along they and z axes
y tensile yield stress of polymer
(} angle of inclination of cell edge BC to the z axis
t5 cell face thickness
Ll extension of a half horizontal edge in the foam
&y Oz lateral and compressive strains in the foam
OHfiX, 6vs, 6e strains in the horizontal and vertical square faces, and edges
¢ fraction of polymer in the faces
Poisson's ratio of the polymer, foam
stress causing elastic collapse in compressed foam
compressive stress applied to the foam, initial yield stress of foam
HARD CELLULAR MATERIALS IN THE HUMAN BODY: PROPERTIES
AND PRODUCTION OF FOAMED POLYMERS FOR BONE REPLACEMENT

C.S. PEREIRA 1•3 , M.E. GOMES 1•3, RL. REIS 1•2•3 , A.M. CUNHA3
1- !NEB, Institute for Biomedical Engineering, Rua do Campo Alegre
823, 4150 Porto, Portugal
2 - Department of Metallurgical Engineering, Faculty Engineering,
University of Porto, Rua dos Bragas, 4099 Porto Codex, Portugal
3 - Department of Polymer Engineering, University of Minho, Campus
de Azurem, 4800 Guimariies, Portugal

ABSTRACT

Hard tissues such as bone are composed structurally by a compact surface and a
cellular core, which is very similar to a synthetic structural foam material. In this paper
the structure of human bone is discussed and related to that of synthetic foamed
materials that are aimed at copying bone structure.
This chapter also reports on the development of processing routes, based on
injection moulding for producing cellular materials that are intended to reproduce the
structure of bone. Two different polymeric materials are used: novel starch based
polymers (biodegradable) and polyethylene (bioinert), in order to develop materials
both for temporary and permanent applications.

1 - INTRODUCTION

Bone is a living organism composed of mineral substances and tissues. The latter
consists of cells, fat substances, natural polymers (like polysaccharides, collagen
polyphosphates) and other substances. Although the properties of bone vary from point
to point and the proportion of the diverse substances changes according to the different
parts of the skeleton; bone tissues contains about two-thirds of inorganic and one third
of organic substances, on average [1]. Bone may be, consequently, seen as a traditional
engineering composite material, that is composed of a polymeric phase, and a
inorganic ceramic reinforcement. Furthermore, it presents rather anisotropic properties
and a non-linear viscoelastic behaviour.
Bone is a unique structural material, which is projected to support and protect
the surrounding tissues. In order to meet these requirements, and to be able to adjust
itself to its functional duties, bone presents a high mechanical strength, and stiffness
193
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 193-206.
© 1999 Kluwer Academic Publishers.
194

and most important of all the capability to remodel and to adapt, not only to variable
conditions of mechanical loading, but also to lesions [2].
Bone has three types of cells: (i) osteoblasts, by which the matrix organic
component is produced and which controls the mineralisation mechanisms, (ii)
osteocytes, which favour the ionic exchanges between the bone and the blood and (iii)
osteoclasts which are responsible by the osseous resorption [3]. For a materials
scientist bone looks just like a structural foam, i.e., a material composed by a compact
skin and a cellular core. In fact, two different types of tissues may be found on any
bone structure [2]. The outside layer is dense in texture and is termed compact bone;
and the interior core consists of a meshwork of trabeculae containing easily visible
intercommunicating spaces; this is called spongy or cancellous bone. Fig.1 presents a
schematic representation of the complex structure of a human long bone. Close
examination of the compact bone shows it to be extremely porous so that the difference
between it and the spongy bone depends merely upon the relative amount and size of
the porosity [2]. In fact, in the compact bone the pores are small (10-20 Jlm) [1,3] while
in the spongy bone the pores are large (they can be over 1 mm) [1,3]. An interesting
work [41 points out that porosity can be basically regarded as being on the borderline
of an additional second phase so that bone, in this perspective, should be considered as
a composite between a mineral phase, an organic phase and porosity [1].
The trabeculae of the spongy bone are organised and oriented in such a way,
as to enable them to withstand the stresses to which they are normally subjected [3].
The architecture of bones is designed to provide the necessary functional strength and
to eliminate unnecessary weight. For instance, when the alignment of a bone is altered
as a result of a badly performed union following fracture, the trabeculae undergo a re-
alignment adapting themselves to the new conditions [3].

Figure I. Drawing illustrating the constitution of the diaphysis of a long bone.


195

The process of bone mineralisation is very complex and has been object of
extensive investigations in the last few years. Several advances have been made, but
many mechanisms are still not fully understood. The main constituents of inorganic
salts are already present in bone in the form of an aqueous solution in the physiological
liquid (of lymphatic nature) which derives from the numerous blood vessels that
vascularize bony tissue [1]. Bone exhibits the presence of large quantities of cations
and complex anionic groups, mainly: Ca2+, Pol·, col·. Other ions are present in
smaller quantities, such as: Mg2+, Fe2+ (in the blood), F, and cr. Due to cellular
turnover, also very small amounts of Na+ and K+, ascorbic acid, citric acid, and
different polysaccharides are present. Some specific heavy atoms may occasionally be
found - such as Ba2+, Sr+, Pb2+ - which show a preference for fixation precisely in bone
tissues. Calcium and phosphate ions lead to formation of salts, primarily calcium
phosphates - present in the amorphous mineral fraction dispersed within the organic
fraction - while the crystallised mineral fraction is practically constituted by
hydroxylapatite [I].
Anisotropy in bone has frequently been observed and reported [5-7], being
manifested as preferred orientation at two levels of organisation within its structure.
These are the microstructural and the ultrastructural levels. At the microstructural
level, it is well established that osteons or Haversian systems of mammalian long bones
are arranged largely parallel to the long axis of the bone, and that the trabeculae of
cancellous bone are aligned along the principle lines of stress in the volume of bone.
Osteons and trabeculae are composed of layers of a composite material, of which the
major constituents are the fibrous protein - collagen, and a calcium-phosphate ceramic
- hydroxylapatite. The properties of bone, particularly the mechanical behaviour, are
dependent on the structure of the mineral, the three-dimensional arrangement of the
collagen fibres within the individual layers and the interfacial adjustment of the two
components [5].
The architecture of bone seems to have been designed, as only living tissues
can be, for its function as a component of a structure. Its most remarkable properties
are the ability to repair and readjust itself by forming new bone along the mechanical
solicitation stress lines [1]. This makes living bone an unique composite, not
comparable to any other traditional engineering structural material. The bone dynamic
response to a particular load allows it to remain highly fatigue resistant. Furthermore,
fractures can only occur under extreme conditions or in unhealthy bones affected by
some malfunction of the body metabolism.
The study of the mechanical properties of natural bone is a complex subject.
Since the osteonic structure in general, and also the trabeculae organisation in long
bones, makes bone a very anisotropic material with respect to its behaviour towards
stress, the mechanical resistance values reported by different authors differ
substantially, depending deeply on individual loading methods and testing protocols.
The evaluation of the mechanical performance of bone samples is strongly dependent
not only on the conditions under which the tests are carried out, but also on other facts
such as the freshness of the samples tested, and the age of the individual from whom
the samples are obtained [1]. The importance of taking into account these observations
196
when performing mechanical tests on materials aimed at being used in bone
replacement is discussed in a review by Reis et al. [8].
Nevertheless, values for the mechanical properties of cortical and cancellous
bone may be found in the literature. Cortical bone presents a tensile strength in the 50 -
150 MPa range [9], a tensile modulus of 7-21 GPa (longitudinal direction) [10,11] and
6-12 GPa (transverse direction) [1 0,11], a critical stress intensity factor (KIC) of 2-12
MNm"312 [10,11] and a critical strain energyrelease rate (GIC) of0,5-6,0 KJm-2 [10,111
The compressive strength and modulus of cancellous bone are typically in the range of
5 to 20 MPa and 0.02 to 1.7 GPa, respectively (9]. Trabecular bone may be described as
a natural porous solid, which tends to fail in compression by crushing. Many kinds of
trabecular bone appear to behave mechanically as a standard open cell foam [12].
Our average life span has increased from 55 to 75 years over the past century.
However, this increased longevity is often at the expense of a decreased quality of life.
Bones do break, joints become painful, vertebrae collapse and teeth fall out; all because
of a progressive deterioration of our skeletal tissues. By retirement age human bones
have lost as much as 20-50% of their density and strength, depending upon the
individual sex [131 Biomaterials, namely prostheses or implants, are now routinely
used to replace body parts, including diseased or damage bones, joints and teeth (13].
At this stage it is important to define biomaterial, as being any substance (other than
drugs) or combination of substances synthetic or natural in origin, which can be used
for any period of time, as a whole or as a part of a system which treats, augments, or
replaces any tissue, organ, or function of the body [14].
There are examples of virtually all classes of biomaterials used in orthopaedic
applications, including metals, alloys, ceramics and glasses, modified natural
biomolecules, synthetic polymers and composites consisting of various combinations of
these material types [14-16]. A pre-requisite for any synthetic material implanted in
the body is that it should be biocompatible [17], in the sense of not producing any
significant inflammatory reaction. The majority of implants and medical devices used
in medicine today are, in the present days, made from polymers [16]. There are
thousands of possible applications for synthetic polymers in clinical medicine. The
primary application areas are cardiovascular, ophthalmic, general soft tissue
replacement, dental, orthopaedic and biotechnology [16].
Based on their behaviour when implanted in living tissue, polymeric
biomaterials can be divided into biostable, biodegradable and partially biodegradable
[ 18, 19]. Biostable polymers are inert, cause minimal response in the surrounding tissue
and retain their bulk properties for years. In many cases a biomaterial needs to be in
place only temporarily, and in such cases biodegradable or partially biodegradable
polymeric materials are more appropriate than biostable ones. Biodegradable surgical
implants are best used for temporary internal fixation when tissue has been
traumatised. This is owing to the fact that at an early stage of healing of, for example,
bone, tendon or skin, the biodegradable implant preserves the structure of the tissue.
As time goes by, the implant gradually decomposes and the stresses are progressively
transferred to the healing tissue. The time necessary for this process depends on the
material, its molecular weight, and its structural and surface properties.
197

Biodegradation of a material means degradation induced by the metabolism and


agressive fluids of the organism. This factgives a further advantage to biodegradable
surgical devices - they do not need a second surgery for its removal and this
substantially benefits the individual as well as saves money to the healthcare system
[15,18,19].
There is a wide variety of orthopaedic devices that are currently in clinical
application. These devices may present very distinct properties, depending deeply on
the prostheses main function. However, the most important constrains are always those
related to the biocompatibility and mechanical behaviour of the devices [8). Some of
the most well known usages of polymeric systems in orthopaedics are [8]: (i)
biodegradable polymers, such as poly(lactic/glycolic acid) (PLAIPGA) or
polyhydroxybutyrate (PHB) in bone plates and screws, (ii) ultra high molecular weight
polyethylene (UHMWPE) in acetabular cups and several joints, (iii)
polymethylmetacrylate (PMMA) as bone cement and (iv) UHMWPE reinforced with
hydroxylapatite (HA) in several small permanent implants. Recently also starch based
polymers and composites (reinforced with hydroxylapatite) have been proposed for
being used in temporary orthopaedic load-bearing applications [20-22].
The use of porous materials has proved to be an affective way of inducing a
good osteointegration in-vivo [2, 12]. The bony tissue capability to ingrow in contact
with systems having an adequate porosity is well established. For the achievement of
bone ingrowth inside a porous structure, the pores must be large enough (150- 400Jlm)
[1, 12] to be able to host the development of the bone constituents (both organic and
inorganic), as well as to host the cells typical of bone. The minimum pore size
(150Jlm) is on the order of the diameter of osteons in normal haversian bone [12]. The
major problem presented by these systems is related to its mechanical resistance since
it is necessary to develop a porous structure with an appropriate mechanical resistance,
not only to support the initial load but also the additional loads generated by the
ingrow of tissue into the porous structure [2].
The main aim of the ongoing research on our group is to develop materials
with an adequate morphology and mechanical behaviour that allow for their use in
bone replacement applications. This includes not only to replace load-bearing bone, but
also the development of 3-D scaffolds and architectures for tissue engineering
applications. In the present work results on the development of foamed materials with
a structure similar to that of human bone (compact surface and cellular core) based on
two different polymers (biodegradable and bioinert) for both temporary and permanent
applications are reported. These materials are produced by a conventional injection
moulding process, by means of using masterbatchs containing blowing agents. This
allows for the production of3-D parts with complex shapes.
198
2 -MATERIALS AND METHODS

The polymers used in this study were: - (i)a high molecular weight polyethylene
(HMWPE) grade (trade name Hostalen GM 9255F, Hoechst, Germany), with a melt
flow index of 2.66g/600s (98N, 240° C) and (ii)a com starch/poly(ethylene-vinyl
alcohol) blend (SEVA-C) (trade name Mater-Bi 1128RR, Novamont, Italy), with a
melt flow index of8.22g/600s (49N, 180°C).
The blowing agents selected were masterbatchs (with different concentrations)
based on carboxylic acids and their salts, with trade names Hostalen P 9937 (BAt) and
Hostalen P 9947 (BA2) system, Hoechst, Germany.
The experimental plan included the preparation of compounds with 0, 5, 10
and 20% (wt.) of blowing agents. The polymers and the blowing agents were pre-
mixed together in a rotating drum prior to injection moulding. Materials were then
moulded in two different machines: a Klockner Ferromatic FM-20 and a Krauss Maffei
KM60-120A in order to produce tensile and impact samples, in the first case of two
different geometries.
The foamed materials were characterised in terms on their morphology and
physical properties. An optical polarised light transmission microscopy (PLM) (in an
Olympus BH-A microscope) and scanning electron microscopy (SEM) (in a JEOL JSM
35C and JEOL JSM 6301F) were used for the morphological characterisation.
The foamed materials were tensile tested in order to determine the secant
modulus at 1% strain (E 1%) of different rectangular cross-section samples (in order to
study the effect of using different moulding geometries in the samples's
structure/mechanical behaviour). These tests were performed on a Instron 4505
universal mechanical testing machine fitted with a resistive extensometer (gauge
length, 1Omm) in a controlled environment (23°C and 55% RH). The cross head speed
was 5mm/min (8,3x10"5m/s) until 1% strain, in order to determine with higher
precision the E 1% value, and then increased to 50 mm/min (8,3x104 m/s) until fracture.
The produced mouldings were also impact tested in order to study the
materials behaviour under a more severe solicitation, to observe the resulting fracture
morphology and to determine the respective properties, namely, peak force (Fp) and
peak energy (Up). Instrumented impact tests were performed using a mass of 25kg and
an testing velocity of 3m/s, in a Rosand IFW SA instrumented falling-weight
equipment.
In order to verify the possible release of toxic elements by the blowing agents
used, samples of the foamed materials were calcinated in a oven at 625 ± 25°C during
30 min and then the solid residue obtained was analysed using energy dispersive
spectroscopy (EDS) in a NO RAN instruments equipment, in a 3-point bending loading
scheme.
For the biodegradable foamed materials based on SEVA-C, the degradation
behaviour was assessed after several pre-fixed ageing periods (0,3,7,14,30 and 60 days)
in a simulated isotonic physiological solution (NaCl 0,154 M). After being removed
from the solution, the samples were dried up to exhaustion in order to determine the
respective weight losses.
199

3- RESULTS AND DISCUSSION

Figure 2. PLM photographs showing a cross-section ofHMWPE injection moulded with 10% blowing agent.

Figure 3. PLM photographs showing a cross-section of SEV A-C injection moulded with I 0% blowing agent

Figures 2 and 3 show PLM photographs of cross-sections of respectively a


foamed HMWPE (with 10% blowing agent) and a foamed SEVA-C with the same
percentage of blowing agent respectively. These foams, also referred as integral skin
foams comprise a low density core surrounded by a high density skin of the same
material [23,24]. This structure mimics the bone architecture which has a cellular core
(spongy bone) and a compact surface (compact bone).
200
However, the pore size, the number of pores, their distribution and the
distance between the pores was highly dependent on the polymer utilised to fabricate
the foamed materials and on the relative amount of blowing agent used. The increase
in the relative amount of blowing agent results in more pores within the final
moulding. The average pore sizes of foamed HMWPE with different amounts of
blowing agent are summarised in table I.

TABLE I. Pore size of different structures obtained by injection moulding ofHMWPE using different amounts of
blowing agent (BA2 ).

SAMPLE AVERAGE PORE SIZE


m

HMWPE + 5% BA2 20-200

HMWPE + I 0% BA2 250-300

HMWPE + 20% BA2 500- 1000

Figure 4. SEM micrograph showing a pore within the core part of a the cross-section of HMWPE injection
moulded with I 0% blowing agent.

Figure 4 shows a SEM micrograph of a pore within the core part of a cross-
section of foamed HMWPE with I 0% blowing agent. The pores formed appeared to be
closed pores and to be at least partially interconnected.
The average pore size of foamed SEVA-C with 10% blowing agent may be
divided in two different classes, one is between 10 to 25J.1m and the other is between 50
to 100J.1m. No pores with intermediate size could be found. Matrixes with porous
surfaces and interconnected pore structures are often desirable in tissue engineering
applications [251 as cell and growth factor delivery devices, and in orthopaedic
prosthesis (26,27] (as cancellous bone replacement) .
201

The nonporous surface and the closed pores may provide barriers to diffusion
in tissue engineering applications [25).This will prevent cells from migrating between
the inferior and exterior of the matrix until it has begun to erode. This may be avoided,
if needed, by the development of ways of transplanting cells also to the materials
surface. However, these features may also be advantageous as they would limit the
interaction of host inflammatory cells with transplanted cells during the acute phase of
the almost inevitable inflammatory response [25]. Additionally, these features may
allow one to control more precisely the release ofbioactive molecules from the matrix.

: 2600 ll veli. . : 90
Li vel ia e : 90

a b

FigureS. EDS spectra of residue solid obtained for calcined samples of: a) HMWPE + 20% blowing agent I and
b) SEVA-C + 20% blowing agent I.

Figure 5 shows the EDS spectra of the solid residue obtained by calcinating
both foamed polymers (in this case for 20% BAI). Only Ca and Na were detected on
calcined samples which means that the blowing agent used is releasing elements that
should not be toxic for the human body. The same type of results was obtained also for
BA2.
The results of the tensile tests are represented in Table II. These results show
that the mechanical properties (namely the modulus) of the cellular materials are not
significantly different from those obtained with the compact polymer (unfoamed). In
fact, one would expect the modulus of the foamed materials to be disproportionately
low because the low-density core is also a load-bearing area. Instead, the modulus is
only slightly lower than the values for the solid material because this property is very
surface dependent [23,24] and the surface of a structural foam test specimen is nearly
solid. This is also related to the additional holding pressure effect generated by the
blowing reaction occurring in the mould, which contributes to obtain more compact
surface for the foamed materials.
Similar results were also obtained in instrumented impact tests. However, due
to the higher solicitation speed in these tests, the measured properties of the foamed
materials show a more significant decrease with respect to those measured for the
compact materials (Table III).
202
The mechanical properties of these materials can be further improved by
several methods which are currently used to attain better properties for several
polymeric materials to be used in orthopaedic implants. For instance, SEVA-C
properties can be significantly improved by means of using non-conventional injection
moulding processes [20-22,28,29] or adding hydroxylapatite (as a reinforcement) to the
polymeric matrix [20,21 ,28,29]. These methods can also be applied to materials based
on HMWPE [30].

TABLE II- Results of tensile tests for two difterent sample cross-sections.

MATERIAL MODULUS (GPa)


samples 2x4mm sampes4xl0mm
HMWPE 1.56±0.59 1.55 ± 0.22
HMWPE + 5% BAI 1.49 ± 0.15 1.86 ± 0.59
HMWPE+ IO%BAI 1.60 ± 0.73
HMWPE + 20% BA I 1.45 ± 0.09 1.87 ±0.54
HMWPE + 5% BA2 1.19±0.17 1.28 ± 0.16
HMWPE + I 0% BA2 1.24 ±0.20 1.38 ± 0.25
HMWPE +20% BA2 1.44 ± 0.19 1.31 ±0.16

SEVA-C 2.13 ± 0.17 2.34±0.27


SEVA-C + 5%BAI 2.28±0.49 2.28 ±0.59
SEVA-C + 5% BA2 1.99± 0.24 1.76 ± 0.09
SEVA-C + I 0% BA2 2.06 ± 0.19 1.74 ± 0.12
SEVA-C + 20% BA2 2.05 ± 0.34 1.81±0.12

BAI-blowing agent trade name Hostalen P 9937 BA2 - blowing agent trade name Hostalen P 994 7

TABLE Ill- Results of instrumented impact tests (sample cross-section= 6xl2mm).

MATERIAL Fp(N) Up(J)


HMWPE 749.31 ± 33.54 1.17 ± 0.25
HMWPE + 5% BAI 687.89 ± 60.27 0.80 ± .0.17
HMWPE+ l0%BA1 714.35 ± 36.78 0.98 ± .0.29
HMWPE + 20% BAI 526.11 ± 149.66 0.69 ± .0.27
HMWPE + 5% BA2 680.43 ± 61.20 0.82 ± 0.13
HMWPE + 10% BA2 744.63 ± 27.65 0.98 ±0.16
HMWPE + 20% BA2 657.36 ± 37.60 0.71 ± 0.09

SEVA-C 688.52 ± 89.57 0.58 ± 0.14


SEVA-C + 5% BAI 429.44 ± 103.10 0.33 ± 0.05
SEVA-C + IO%BAI 170.93 ± 53.51 0.05 ± 0.01
SEVA-C + 5% BA2 493.36 ± 76.50 0.28 ± 0,05
SEVA-C + 10% BA2 394.29 ± 44.88 0.23 ± 0.04

BAI-blowing agent trade name Hostalen P 9937 BA2 - blowing agent trade name Hostalen P 994 7
203

Figure 6 . SEM micrographs of an impact fracture surface for a foamed HMWPE with 5% blowing agent showing
a) a general aspect and b) a detail of the porous structure obtained after the instrumented impact test.

a b

Figure 7. SEM micrographs of an impact fracture surface for a foamed SEVA-C with 5% blowing agent showing
a)a general aspect and b)a detail of the porous structure obtained after the instrumented impact test

Figure 6 and 7 present typical SEM micrographs of impact fracture surfaces


respectively for foamed HMWPE and foamed SEVA-C, both with 5% of blowing
agents. The high speed (3m/s) impact solicitation leads to the cavitation of the internal
pores, producing at the same time, a more uniform distribution of the pores throughout
the polymer matrixes and reducing the final «skin layer» of the solid polymer.
204
The degradation behaviour of the biodegradable materials (SEVA-C with
different amounts of blowing agent) is plotted in figure 8, in terms of the dry weight
loss versus degradation time in NaCl (0.154M) at 37 ± l°C. The weight loss depends
strongly on the blowing agent additions, and higher amounts of blowing agent tend to
increase the degradation rate. This indicates that foamed materials are degraded faster
due to its higher water uptake capability, and to the easier access of the degradation
fluids to the bulk of the polymer. The degradation behaviour of the materials obtained
in this work (based on SEVA-C) is similar to that of the solid polymer (studied in
previous works by Reis et al [29] ), but the degradation rate is faster for foamed
materials. This fact assures that the compact surface of the material is rapidly degraded
in order to allow the contact between the bony tissue and the interior porous structure
of the implant inducing the bone ingrowth.

......-SEVAC
........ SEVA-{. ' 5""' AA2
.......-SEVA.C • 10'\1-SA.;
........_SEVAC+20%8A2

12

Time(days)

Figure 8. Dry weight loss vs. degradation time fur SEVA-C and fuamed SEVA-C materials.

4- CONCLUSIONS

It was possible to obtain materials which copy the structure of human bone, that is,
materials with a cellular core and a compact surface, by the injection moulding
process, using blowing agents. This was possible both for bioinert and biodegradable
polymers.
The process used to obtain the foamed materials proved to be less effective
when using the SEVA-C because the blowing agent is more effective at temperatures
between 200 - 250°C and this polymer can only be processed at temperatures below
180°C while the HMWPE can be processed at 2400C.
205

For the foamed SEVA-C {biodegradable material), this compact surface is


especially important because allows it, to maintain the necessary mechanical strength
in the initial period of implantation, while after some degradation the pores will
promote osteointegration.
The mechanical properties of the foamed materials were not significantly
different from that of the original compact (unfoamed) materials.

REFERENCES

I. Ravaglioli, A. and Krajewski, A. ( 1992) Bioceramics, Chapman & Hall, London.


2. Katz, J.L. (1995) Mechanics of hard tissue in J.D. Bronzino (ed), The Biomedical Enginering Handbook,
CRC Handbook Publishers, USA, pp. 273-290.
3. Davies, D.V. (1972) Gray'S Anatomy, Longmans, London.
4. Ondracek, G. (1986) Microstructure-thermomechanical property correlation of two-phase and porous
materials, Materials Chemistry and Physics 15, 281-313
5. Isaac, D.H. and Green, M. (1994) The origins of preferred orientation in bone, Clinical Materials 15,79-87.
6. Dabestani, M. and Bonfield, W. (1990) Fracture behaviour of compact bone, Advances in Biomaterials 9,
7-12.
7. Isaac, D.H. and Green, M. (1991) Preferred orientation in bone, Interfaces in Medicine and Mechanics 2,
76-85.
8. Reis, R.L. and Cunha, A.M. (1995) Mechanical characterization of polymers and polymer matrix composites
aimed at orthopaedic applications, in H. Skov (ed), Medical Plastics 9, Soc. Plast. Eng. , pp 7 .I- 7 .12.
9. Matthews, F.L. and Rawlings, R.D. (1996) Composite Materials: Engineering And Science, Chapman &
Hall, London.
10. Bonfield, W. (1993) Properties ofbone as a material in F. Bumy, R. Puers (eds), Monitoring of Orthopaedic
Implants 4, Elsevier Science Publishers, Amsterdam.
II. Bonfield, W. (1987) Advances in the fracture mechanics of cortical bone, J. Biomechanics 20, 1071.
12. lakes, R. (1995) Composite biomaterials in J.D. Bronzino (ed), The Biomedical Enginering Handbook,
CRC Handbook Publishers, USA, pp.598-610.
13. Hench, L. (1995) Bioactive implants, Chemistry & Industry vo13, 541-550.
14. Williams, D.F. (1987) Definitions in biomaterials, Progress in Biomedical Engineering 4, Elsevier.
15. Williams, D.F. (1989) Polymer degradation in biological environments, in G. Allen, J.C. Bevington, G.C.
Eastmond, A. Ledwith, S. Russo, P. Sigwolt (eds), Comprehensive Polymer Science 6, Pergamon Press,
Oxfurd, pp 607-619.
16. Ratner, B.D. (1989) Biomedical applications of synthetic polymers in G. Allen, J.C. Bevington, S.L.
Aggarwal, (eds), Comprehensive Polymer Science, Pergamon Press, Oxfurd, pp 201-249.
17. Bonfield, W., Hydroxyapatite-reinfurced polyethylene as an analogous material fur bone replacement, in P.
Duchyne, J .E. Lemons ( eds), Annals ofthe New York Academy ofScience 253, New York, pp 173-177.
18. Rokkanen, P.U. (1991) Absorbable materials in orthopaedic surgery, Annals a/Medicine 23, I 09-115.
19. Anderson, J.M. and Zhao, Q.H. (1991) Biostability ofbiomedical polymers, MRS Bulletin, 15-11.
20. Reis, R. L., Cunha, A.M., Allan, P. S. and Bevis, M. J. (1996) Improvement of the mechanical properties of
hidroxylapatite reinfurced starch based polymers through processing, Polymers in Medicine & Surgery,
Institute of materials, London, ppl995-2002.
21. Reis, R. L., Cunha, A. M., Allan, P. S. and Bevis, M.J. (1997), Structure development and control of
injection-molded hydroxylapatite-reinfurced starch/EVOH composites, Advances in Polymer Technology
16, in press
22. Reis, R. L., Cunha, A. M. and Bevis, M.J. (1997) Non-convencional processing routes on the development of
anisotropic and biodegradable composites of starch based thermoplastics reinfurced with bone-like ceramics,
Plastics-Saving Planet Earth, Procedings of the SPE 551hAnnual technical Conference, ANTEC'97,
Toronto, 2849-2853
23. Brewer, G.W. (1989) Properties of thermoplastics structural fuams in Engineered Materials Handbook 2,
ASM International, New York, pp 508-513.
206

24. Khakhar, D.V. and Joseph, K.V. (1994) Optimization ofthe structure of integral skin foams for maximal
flexural properties, Polymer Engineering and Science 34, 726-733.
25. Mooney, D.J., Baldwin, D., Suh,N., Vacanti, J. and Langer, R. (1996) Novel approach to filbricate porous
sponges of poly(d,l-lactic-co-glycolic acid) without the use of organic solvents, Biomaterials 17, 1417-1422.
26. Devin, J.E., Attawia,M.A. and Laurencin, C.T. (1996) Three-dimensional degradable porous polymer-
ceramic matrices for use in bone repair, Journal ofBiomaterials Science - Polymer Edition 7, 661-669.
2 7. Szivek, J.A., Thompson, J.D. and Benjamin, J .B. ( 1995) Characterization of three formulations of a synthetic
foam as models for a range of human cancellous bone types, Journal ofApplied Biomaterials 6, 125-128.
28. Reis,R. L., Cunha, A. M., Allan, P. S. and Bevis, M.J. (1996) Mechanical behavior of injection-molded
starch-based polymers, Polymers for Advanced Technologies 7, 784-790.
29. Reis, R. L., Mendes, S.C., Cunha, A.M. and Bevis, M. J. (1997) Processing and in vitro degradation of
starch/EVOH thermoplastic blends, Polymer lnternational43, 347-352
30. Reis, R. L., Cunha, A. M. , Oliveira, M.J., Campos, A.R. , Bevis, M.J., (1997), Relationship between
processing and mechanical properties of injection molded high molecular weight polyethylenelhidroxylapatite
composites, Polymer Engineering & Science, in press.
RHEOLOGY AND GLASSY DYNAMICS OF FOAMS

M. E. CATES AND P. SOLLICH


Department of Physics and Astronomy
University of Edinburgh
JCMB, Kings Buildings
Mayfield Road, Edinburgh EH9 3JZ, UK

Abstract. After a brief 'warm-up' discussion of osmotic pressure of foams,


the basic phenomena of foam rheology are reviewed, focusing on linear vis-
coelastic spectra (elastic and loss moduli) with brief mention of nonlinear
effects. Theoretical models for some of these properties are then described,
starting with Princen's model for the elastic modulus Go of an ordered foam
in two dimensions. There is a basic conflict between this model, which pre-
dicts a step-function onset of the modulus when droplets first contact at vol-
ume fraction</>= </>o, and the experimental data (which show Go rv </>-</>0 ).
The three dimensional ordered case is reviewed next, focusing on anhar-
monic deformation theory which predicts a logarithmic softening of the
modulus near </>0 ; this is still not soft enough to explain the observations.
The 3D disordered case is then addressed; a combination of disorder and
the anharmonic effect finally seems able to explain the data. We then con-
sider the problem of the frequency-dependent loss modulus G"(w) which
describes dissipation in a foam. Somewhat alarmingly, the data suggest
behaviour incompatible with linear response theory; reconciliation is possi-
ble if one invokes some very slow relaxation processes at timescales longer
than experiment. We briefly describe the search for foam-specific slow re-
laxation mechanisms of surfactant and water transport, which so far has
yielded no viable candidates. Since similar anomalies in G"(w) are observed
in several other systems, they are instead tentatively ascribed to a generic
phenomenon: glassy dynamics. A recent model for the rheology of "soft
glassy matter" is then reviewed; though phenomenological, this suggests
that glassy dynamics may be a useful concept in foam rheology.
207
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 207-236.
© 1999 Kluwer Academic Publishers.
208
1. INTRODUCTION
In these lectures we examine the rheological properties of liquid/gas foams
(froths) and liquid/liquid foams (dense emulsions). We treat the foam as an
array of droplets in a continuous matrix, where the volume of each droplet
is constant in time. This neglects the compressibility of the fluid in the
droplets; this is a good approximation whenever the characteristic scale of
Laplace pressures u / R (with u surface tension and R the mean droplet
radius) is small compared to the bulk modulus of the fluid, which is usually
the case even for a froth. It also neglects coarsening effects which lead the
droplet volumes to change slowly over time.

1.1. OSMOTIC PRESSURE

As a warmup exercise, we recap ideas about the osmotic pressure in a


foam. To measure such a pressure, a foam is compressed by a semipermeable
membrane through which the continuous phase (which we take to be water)
can pass, but not the droplet phase. The force per unit area which must be
exerted on the membrane is the osmotic pressure llo, which depends on ¢,
the volume fraction of the droplet phase. (We stick to this notation in these
notes although others are used elsewhere in this Volume.) Alternatively TI0
can be viewed as a function of V, the volume of the compressed foam.
The osmotic pressure obeys TI0 (¢) = -oFfoV with the derivative taken
at fixed number of droplets. To a very good approximation (which neglects
the tiny translational entropy of the droplets), F = uA where A(V) is the
total surface area of the droplets. Hence TI0 (¢) is zero for¢ < ¢ 0 , where
¢ 0 is the threshold at which droplets first come into contact and start to
deform. For higher volume fractions, the droplets develop flattened facets;
these are small just above the threshold but soon become fully developed
bilayer films as ¢ is further increased. The remaining water then resides
in Plateau borders along the edges of the polyhedral droplets with further
water pockets at the vertices.

2. BASIC RHEOLOGY
2.1. LINEAR VISCOELASTICITY

Linear viscoelastic theory describes the response of a material to small


deformations. Consider, for example, an imposed simple shear through a
strain angle 1 (assumed small) suddenly imposed at time zero and then
held constant. The resulting shear stress is

s(t) = 1G(t) (1)


209
For a Hookean solid, s = Go/ and there is no time dependence; Go is the
shear modulus of the material:

Go= (fPF)
812 v
(2)

For a Newtonian fluid, instead s = 7J'Y where 7J is the viscosity. (Formally


this corresponds to a delta-function transient for G(t).) Viscoelastic solids
show behaviour intermediate between these extremes, with a nontrivial
transient in G(t) followed by a plateau, G(oo) -+ Go at long times. Vis-
coelastic fluids are similar but there is either no plateau or else one which
lasts a finite time before G(t) finally decays to zero. Foams do show a well-
developed plateau, of amplitude Go which we call the plateau modulus; they
are usually viewed as viscoelastic solids.
A related experiment is oscillatory shear, 1 = ~['Yoeiwt], for which the
stress is
(3)
with G*(w) = G'(w) + iG"(w) where G', G" are called the storage modulus
and the loss modulus. These control respectively the in-phase (elastic) re-
sponse and the out-of-phase (viscous) response, of the medium. Their ratio
obeys G"/G' = tan8 where 8(w) is the phase shift between the measured
stress and the applied strain.
Linear response theory (see e.g. [1]) shows that

G*(w) = iw fooo G(t)e-iwt dt (4)

Since G(t) is real, this requires that G'(w) = G'( -w) whereas G"(w) =
-G"(w). Thus the loss modulus is an odd function of frequency which
means that it must vanish as w -+ 0.

2.2. FOAM PHENOMENOLOGY

Foam are found to have a well defined elastic modulus G(t-+ oo) = Go(<P)
which depends interestingly on the volume fraction ¢, as well as on surface
tension a and (mean) droplet size R. Specifically, G 0 rises smoothly from
zero as <P is raised above ¢ 0 • Surprisingly perhaps, most theories cannot
explain this, and below we discuss this problem in some detail.
According to linear response theory, the plateau modulus Go can also
be measured as G'(w-+ 0); and in most foams a limiting value can readily
be extracted from the frequency dependent spectra. However, there is a
strongly anomalous behaviour in G"(w), which, in apparent contradiction
to linear response theory, does not vanish at low frequencies but instead
210

seems to approach a constant value or (in some cases) even to be rising as


the frequency is lowered. (Typical G" values at low frequency are between
1/5 and 1/50 of Go.) This is a cause for concern, to which we return in
Section 4.

2.3. NONLINEAR EFFECTS

These can arise when either the stress 8 or the strain 1 is not small. In most
foams one observes the phenomenon of yielding: if a stress 8 is maintained
in steady state, then for 8 < 8y the material attains a finite deformation
and then stops moving (finite 'Y) whereas for 8 > 8y the steady state
motion involves finite i' and the sample deforms indefinitely (creep). This
behaviour defines a steady-state flow curve which is a plot of stress against
strain rate i'· If the "yield stress" 8y is nonzero (solid-like behaviour) the
curve comprises a vertical line segment at i' = 0 followed by a smooth
continuation. Although many other nonlinear measurements are possible,
in these lectures we avoid discussion of these. Moreover, up until Section 5,
we will focus exclusively on linear viscoelastic phenomena.

3. THE ELASTIC MODULUS


3.1. PRINCEN'S MODEL

The simplest model [2, 3] for foam elasticity considers an array of cylin-
drical "droplets" in a two-dimensional ordered hexagonal packing. The ge-
ometry is shown in Fig. 1. We consider a sample of unit length into the
page, in which case the free energy per cylinder is

Fcyl = u(6L + 21rr) (5)

where (6L + 21rr) is the perimeter of the cylinder, L = a- 2r/v'3 the


facet length, a the side of a hexagonal cell and r the radius of the Plateau
borders.
The volume fraction 4> is the ratio of the cross-sectional area of a droplet
to that of the hexagonal cell

(6)

where R is the radius of a free (circular) cylindrical drop. The contact point
4>0 is fixed by requiring a= 2R/v'3 and hence (after some algebra)

(7)
211

Figure 1. Geometry of a "droplet" in a two-dimensional ordered hexagonal foam.

and ¢ 0 = 0.91. Likewise one finds

1 (1-
Fey!= 21ruR [ <Po¢- ~
¢0)1/2 (1-
-¢-
¢)1/2] (8)

which is completely fixed by geometry.


We first use this result to find the osmotic pressure TI 0 = -(8F(V)/8V)R
where F(V) = Fey!/¢:

(9)

(This is shown in Fig. 2.) Note that TI 0 rv (u/R)f(¢) where f is a dimen-


sionless function. According to the model, this function vanishes smoothly
as ¢-+ <Pt and diverges as ¢-+ 1, when the Laplace pressure 0' /r formally
becomes infinite (since r -+ 0). Both features are broadly confirmed by
experiment [4, 5].
Turning now to the elastic modulus Go, we must find the change in
surface area of the droplets when the lattice is deformed. We choose for
simplicity a shear direction parallel to the direction joining droplet centres,
and note that by the Young-Laplace law, the 120° angles between films
(extrapolated through the border regions) are preserved on deformation;
212

Til( /R)
2

1.5

0.5

Figure 2. Osmotic pressure of a two-dimensional ordered hexagonal foam

otherwise there would be unbalanced forces on the borders. Each deformed


droplet therefore occupies an irregular hexagon whose corners each contain
one third of a plateau border (whose radius is still r). The resulting free
energy per cylinder is

(10)

where L 1 ,2 ,3 are the film lengths, or equivalently

(11)

where a 1 ,2 ,3 are the sides of the deformed hexagonal cell. Notice that for
given a, this expression is independent of volume fraction. In other words,
the area change on deformation, which is entirely attributable to stretching
of the film regions, does not depend on how large these films are originally
- except that of course the calculation breaks down for </> < </>o when the
films simply do not exist.
From this calculation one can find Go from the stored free energy using

F(V, -y)- F(V, 0) = VG 0 -y 2 /2 (12)

which holds in the limit of small strain ')'. This gives the result

G = _!!__ = _!!_
0 V'3 2R
(.!t)
</>o
1/2 = 0 525 u 4>1/2
. R
(13)

which has a finite value of 0.525u / R as </> ~ <f>ci. Accordingly the Princen
model predicts a step discontinuity in the elastic modulus Go at </> = </>o.
213

G/(cr/R)

Figure 9. Shear modulus of a biliquid foam, from [4]. The solid line is a (semi-)empirical
fit.

This contrasts with the prediction for osmotic pressure II0 (which rises lin-
early from zero) but is similar to that for the bulk modulus B0 = -8II/8V.
The step discontinuity is not consistent with experiments on either poly-
disperse [4] or near-monodisperse [5] foams: see Fig. 3. Candidates for ex-
plaining the discrepancy include: (a) the fact that real foams are three di-
mensional; (b) the fact that they are disordered; or, (c) some combination
of these.

3.2. 3D ORDERED FOAMS

To find G0 (<P) for a three-dimensional foam is a hard task, because the


Plateau borders have a more complex geometry (see other lectures in this
volume!). There are three main approaches. The first is to guess the shape
of the borders; so long as volume constraints are respected, the resulting
surface will always have a larger area than the true one (which is the state of
least free energy and hence minimal area). So the "guessing" method gives
an upper bound on F(V) and can in principle be refined systematically,
for example by including variational parameters in the guessed shape, and
minimizing the computed area over these parameters. However, in what
follows, we shall follow Refs. [6, 7] and simply guess. The second line of
attack involves an exact analysis for the limit <P -+ <Po in which droplets
are weakly compressed. This is informative since it addresses precisely the
region where the Princen model fails to explain the data. Finally, numerical
214

approaches based on the surface evolver programme [8) can be used. We


describe these aspects in turn.
Before doing this, however, we make a technical amendment to the cal-
culation: it turns out to be much easier to calculate the uniaxial elongation
modulus, P,o(<P), than the shear modulus Go(<fo). The former is defined by
stretching a unit cube of material so that the lengths of its sides become
1+f along the stretch direction and 1-f/2+3f2 /8 along the remaining two.
(This deformation preserves volume.) In terms of the elongational strain f,
we have
(14)

which serves as the definition of /-tO· For an isotropic material such as a real
foam, one can prove Go = p,0 /3, whereas for an ordered array of droplets,
both Go and p,0 in principle are functions of orientation. So, given that
results from an oriented model have to be compared with experimental
data on an isotropic system, it is no less valid to compute p,0 (<P) than
G0 (<fo). Moreover, at least for the purpose of qualitative comparison, one
can choose the direction for our elongational distortion so as to minimise
the algebra involved.
Buzza and Cates [6) performed such a calculation with a simple cu-
bic array of droplets. Though in obvious violation of the Young Laplace
condition, this allows even more algebraic simplification and should not
qualitatively affect the nature of the singularity at <P -t <Po (i.e., smooth
onset versus jump discontinuity). Their guess for the droplet shape was
a truncated sphere: that is, the droplet surface is the inner envelope of
the intersection between a sphere and the deformed (cubic) unit cell. To
maintain constant droplet volume, the radius of the sphere must change as
the unit cell is deformed (Fig. 4). Results for this model (Fig. 5) show a
jump discontinuity at <P = <Po = 0.52, closely resembling that of the Princen
model.
In the second approach to this problem, the limit of weak compression,
<P -t <Pci is addressed using the theory of Morse and Witten [9). This is a
perturbative treatment which exploits the fact that at weak compression,
the droplet shape is almost spherical, with small faceted regions. Accord-
ingly it can be studied by analysing the response of a spherical drop to a
weak external force field. The latter induces pressure shifts on either side
of the droplet surface, Pout(Q) (where Q is an angle) and Pin· The latter is
an angle-independent quantity whose value is chosen to maintain constant
volume for the droplet overall. In units where the free droplet radius R and
surface tension u are both unity, the Laplace equation reads

-CV 2 + 2)p(S1) =Pin- Pout(Q) (15)


215

disc

unit
cell

> 2L'

2L
2L"

Figure 4. Trnncated sphere model. Each droplet is modelled as a trnncated sphere


within a cubic unit cell. On uniaxial deformation the radius of the drop, as well as the
geometry of the unit cell, are changed.

where the left hand side is the local curvature (the 2 comes from the refer-
ence shape of a sphere) and the right hand side is the local Laplace pressure
across the interface. In this equation p(O) is the local (normal) displacement
of the droplet surface.
Morse and Witten calculated the response function G determining the
displacement field, p(O) = G(Q, Q')J, in response to a point force Pout(O) =
fo(Q- Q'). I~ terms of the angle 8 between Q and 0', this reads:

G(O,O')= ~: [~+~cos8+cos8ln(sin 2 (8/2))] (16)

This allowed them to calculate the free energy of the weakly compressed
droplet. For a small facet subject to a localised force J, they found a free
energy contribution

oF= _.!._ j2 [-ln(J j81r)- 4/3) (17)


87r
This can be compared with Hooke's law for a spring of stiffness k, oF=
P /2k, revealing that small facets act as logarithmically soft springs, having
an effective spring constant keff"' -1/ ln /, which tends to zero as f -t 0.
The Morse-Witten effect can be incorporated into the simple cubic
model of an emulsion to find a more accurate expression for the elastic
modulus G0 near onset (also shown in Fig. 5). Although this offers a for-
mal improvement, in that the step discontinuity is replaced by a smooth
curve, the slope at <Po remains infinite, and the prediction still has scant
resemblance to the experimental data [4, 5).
216

J.l)( cr/R)
4

1
~ <I>
0
0.6 0.7 0.8 0.9
Figure 5. Results for the truncated sphere model (upper curve) and those incorporating
the Morse-Witten correction for small compressions.

We now discuss the third approach to the problem of 3-D ordered foams:
the numerical one. Lacasse, Grest and Levine [7] used the surface evolver [8]
to study the free energy of spherical droplets squeezed into a series of n-
faced polyhedra. The case n = 2 (a droplet between parallel plates) is
an unrealistic but interesting limit, because the surface shape can then be
found exactly using the calculus of variations. As well as benchmarking the
surface-evolver calculations, the exact solution can be compared with the
Morse-Witten theory (at small compressions) and also with the truncated
sphere model. In fact, the truncated sphere model is never very good: at
small compressions the Morse-Witten result is approached whereas at larger
ones the data approach a model in which the droplet shape is a disc sur-
rounded by a toroidal "bulge" of semicircular cross-section. (Such a bulge
joins onto the films tangentially, unlike the truncated sphere, so it is not
surprising that this gives a better result; however, the extension to n > 2
is not obvious.)
For larger n, the free energy data was found to be well fit by the em-
pirical form

(18)

where a(n) varies between 2.1 (n = 2) and 2.6 (n = 20), and C(n) be-
tween about 7 and about 70. Here R is the free droplet radius and h the
perpendicular distance from the centroid of the polyhedron to the centre
217

of a face. Note that most of the n values correspond to shapes that do


not tesselate in three space: this is probably not crucial since in reality n
is not the same for all droplets in a foam. Note that because a > 2, the
model corresponds to a soft (power-law) spring interaction at contacting
facets. Using the results for simple cubic and face-centered cubic unit cells,
Lacasse et al. predicted the shear modulus and found curves rather simi-
lar to those found by Buzz a and Cates using the Morse-Witten potential.
In other words, they again found a smooth curve for Go(</>), but with an
infinite slope at </> = </>o rather than the finite slope seen experimentally.

3.3. ROLE OF DISORDER

This was studied in two dimensions by Hutzler and Weaire [10, 11], and
Durian [12]. As emphasised long ago by Weaire and others, one would
expect the disorder to wash out the step discontinuity, replacing it with a
much smoother curve, and this indeed was seen computationally. However,
it is difficult to perform accurate simulations very close to </>o, and these
authors did not get extremely close.
The three dimensional version was studied by Lacasse et al [13], with-
out, however, including explicitly the details of the foam structure (thereby
limiting the computational load). Instead an anharmonic pair potential was
constructed by appeal to the findings quoted above for 8F in polyhedral
boxes:

{19)

where averaged parameters a = a(n) ~ 2.3 and C = C(n) were chosen.


Here dis the distance between droplet midpoints; the U's are summed over
droplets within ranged< 2R (that is, summed over facets). This potential
was fed into a very large MD simulation and the response to compression
and shear distortions measured. Data for the shear modulus is compared
with experiment in Fig. 6; that for the osmotic pressure is also in reasonable
agreement. It appears that by combining disorder and anharmonicity, one
finally gets close to the experimental behaviour; this certainly cannot be
achieved using anharmonicity alone and arguably also cannot be achieved
solely with disorder [14]. The significance of disorder was highlighted by a
study of the individual droplet displacements under shear: these were found
to be highly non-affine (in other words, different droplets moved in different
directions).
218

0.45 ,-----...,------.-----..- ----,-----..-----,

0.40

0.35 0
0
0
~0 *
0 p.Cf0
.-... t +.
~ 0.25 0 ; d
~ 020
,_ * 0
t:§ . 0 ·' +G
IQ+

0.15 ~ .d!\+
0 c§> .'t)T•••

0.10 0° *:~:ll*·*·
&> 4l
0.05 ,:Q, .~.tit
.'!'. 0 *~·· ·
'"' a s""_.,.
0.00 ~~~~--~---~ --~~--~-~
0.65 0.70 0.75 0.80 0.85
<p

Figure 6. Shear modulus of a 3d foam found by simulation ( +) [13] and experiment [5].

4. DISSIPATION IN FOAMS
Above we have described calculations of the linear elastic modulus G0 =
G'(w-+ 0). W now return to the problem of understanding the linear loss
modulus, G"(w), whose low frequency behaviour remains very puzzling.
Specifically, it shows no sign of vanishing at low frequencies, instead re-
maining constant or even increasing as frequency is lowered.
The lowest frequencies easily accessed by experiment are of order 0.1
Hz. Hence the observed anomaly might be resolved if one could find re-
laxational modes of the foam having characteristic frequencies well below
this. Candidates for such modes fall into two classes: specific relaxation
mechanisms arising from the physics of films, Plateau borders, etc., which
are restricted directly to foams; and generic mechanisms associated with
more universal aspects of disorder. In this Section we restrict attention to
the first type, and show that there are no really obvious candidates. In
Section 5 we therefore review a recent model for generic dynamics in "soft
glassy materials", which we believe may include foams.

4.1. QUALITATIVE ANALYSIS OF SLOW MODES

If there are slow relaxation modes in foams, the observed "low fre-
quency" viscoelastic spectrum can be interpreted as shown schematically
in Fig. 7. In other words, at still lower frequencies, relaxation modes would
be seen, each causing a drop !:l.G in G'(w) and a bump in G"(w) as the fre-
quency is t racked downward. For a simple (Maxwell) process, the maximum
219

G"

r-----::>~ experimental window?

ro
Figure 7. Sketch of viscoelastic spectrum, showing possible relaxations at frequencies
below the measurement window.

height of the bump in G" is also D..G; for a. distribution of relaxation times
both the drop and the bump are smeared out. Given that G" is an apprecia-
ble fraction (say 1/10) of G' at the measured frequencies, this explanation
requires not only that the characteristic frequencies of the mode(s) be low
(say < 0.1 Hz), but also that the corresponding amplitudes D..G be a. rea-
sonable fraction of Go.

The exact calculation of the relaxation modes is far beyond our present
capabilities. However, qualitative estimates can be made by the following
strategy. (i) Identify sources of dissipation and express these as steady
state viscosities drJii (ii) Identify the corresponding driving forces and
express these as amplitudes D..Gi (found from the excess stored free en-
ergy in the strained state); (iii) Construct the characteristic frequency as
Wi = D..Gi/ d'TJi· The following discussion summarises that of Buzza., Lu and
Cates [15]. Numerical estimates are generally those appropriate to small-cell
biliquid foams as studied by Mason, Weitz and Bibette [5], and dissipation
estimates are quoted per unit volume.
220
4.2. SOURCES OF DISSIPATION

Consider first a wet foam. This has films of thickness d (so far neglected,
but finite in practice), in equilibrium with borders of mean radius r. The
Laplace pressure in the borders is balanced, within the water films, by the
disjoining pressure IId. This arises from the direct interaction between the
surfactant monolayers across the thin water film; without it, foams would
not be stable. We assume that the wet foam has R ~ r ~ d and that the
borders provide a good reservoir of surfactant. That is, we assume that static
changes in film surface area negligibly perturb the chemical potential of the
surfactant in the foam. (Transient changes may however set up temporary
gradients in the chemical potential.) Possible dissipation sources include
the following.
Fluid viscosity: In most cases, this contribution is dominated by the
shearing of the water films (as opposed to the fluid within droplets).
Diffusion resistance: If a diffusive current j of surfactant is driven rela-
tive to stationary water by a chemical potential gradient, there is an entropy
production. This is precisely analogous to the Joule heating by electrons
flowing down a wire from high to low chemical potential (voltage). An ex-
pression for the heat production is given later.
Intrinsic dissipation in monolayers: The surfactant monolayers are 2-D
fluid films characterised by a shear viscosity p, and dilational viscosity K,.
(In fact, several different K, 7S can be defined for films in different states of
equilibrium with a reservoir; this is discussed in Ref. [15) but ignored here.)
Rough estimates for p, are in the range 10- 7 - 10- 6 kg s- 1 and for K,, about
10-6 -10- 4 kg s -l, although the latter is hard to measure (and/or define).

4.3. DRIVING FORCES

Each driving force can be associated with a free energy contribution 6.F =
V 6.Gr 2 /2 which contributes to the elastic stress until such time as the
stored free energy relaxes.
Gibbs elasticity: For example, the free energy increment in a stretched
film is
8F = u8A + E(8A) 2 /2A (20)
where E is the Gibbs elastic constant. As with K,, different E's can be defined
depending on the state of equilibrium of the film; if it is instantaneous
exchange contact with a surfactant reservoir, E is zero. But if the density
of surfactant per unit area is reduced by a sudden expansion of the film,
then a finite Gibbs elastic storage will occur until this transient density
deficit is rectified (by surfactant diffusion or some other mechanism). The
value of E used below refers to this situation.
221

Surface tension: We distinguish this from the surface tension gradi-


ents arising from Gibbs elasticity. The latter can become large, and may
represent a stronger driving force than surface tension itself. Pure surface
tension remains relevant, however, since it is what causes the relaxation of
a nonminimal surface toward a minimal one, under conditions where the
surfactant has reached equilibrium.
Disjoining pressure: This provides a third driving force, which drives
the equilibration of the film thickness d to the proper value (at which it
becomes balanced by Laplace pressure).

4.4. A THEOREM

To summarise the above, shear deformation requires currents of both water


and surfactants to be set up in the foam, to transport these materials from
their old to their new positions. The driving forces responsible are Gibbs
elasticity (strong), surface tension (weaker) and disjoining pressure (weak,
except in dry or nearly dry foams [15]) and the dissipations involve water
viscosity, film viscosities, and diffusion resistance.
A theorem of nonequilibrium thermodynamics [16] states in essence that
"If alternative pathways exist for relaxing the same driving force, the least
dissipative is chosen". Note that this is only true strictly within the lin-
ear response regime - but that is enough for us. Versions of the theorem
involve the well-known variational principles of least dissipation in Stokes
equation of fluid motion and Kirchoff's laws for resistor networks. Accord-
ingly, for qualitative purposes we should think of possible patterns of water
and surfactant fluxes and choose the least dissipative. (This is a poor mans
substitute for the full solution of the problem, which would involve finding
the full equations of motion from a variational principle, and solving these
explicitly.)

4.5. DRY LIMIT

It is easier to think of candidate flux patterns in the dry limit where the
borders are not present; the film thickness dis, however, finite. Consider a
hexagonal array under slow shearing: there are regions where films contract
and others where they expand. In the contracting (expanding) regions there
is too much (too little) of both water and surfactant. Considering the water
current first, the required transport can arise via Poiseuille type flow of
water with a film (Fig. 8). The associated dissipation rate (per unit volume)
is of order T S = 17W R-1 2 / d as may be confirmed by checking the typical
shear rate within a film (~ .YR/d) and taking account of the film volume
fraction (~ d/ R). The driving force for this motion is essentially lld.
222

Figure 8. Poiseuille flow of water with a film of thickness h; the typical fluid velocity
near the film centre is of order R.Y.

Because there is a nontrivial stress boundary condition at the surfactant


monolayers, the mean velocity in the centre of a moving film is, according
to this mechanism, larger than at the edges. Thus, although part of the sur-
factant is swept along with the water, this does not give the right surfactant
flux overall. An obvious way to balance the flux is by diffusion of surfactant
within the watery part of the films. For typical biliquid foams (where the
surfactant concentration in this region is very small) this is extremely dis-
sipative [15]: T S ~ RT.Y 2 I (dDcE 2 ) with D the surfactant diffusivity and E
the area per molecule in a monolayer. (Note that this scales inversely with
surfactant concentration.) So if a less dissipative route can be found, the
system will choose it.
Such a route involves "Marangoni flow", that is, collective motion of
the surfactant layers along the film surface (dragging some fluid along for
the ride). Clearly, by combining this with the Poiseuille flow in some lin-
ear combination, the required ratio of surfactant to water fluxes can be
achieved. Such a resultant flow is shown in Fig. 9: note, however, that
to get a suitable overall flow pattern requires shearing within the mono-
layers as well as within the water; the dissipation is thus found to be
T S ~ 17W R.Y 2 Id + k.Y 2 IR. The driving force for this motion is essentially
Gibbs elasticity.

4.6. WET FOAM

For a wet foam, the preceding flow patterns are still possible, but an ex-
tra possibility arises. This is called Marginal Regeneration and involves the
push/pulling of fully-formed bilayer films into and out of the Plateau bor-
der regions. There is then negligible dissipation within the body of the film,
223

>

<
Figure 9. A superposition of Marangoni and Poiseuille flows leads to strong shear in
some films (bold arrows) and compression and dilation in others.

but dissipation arising from and water flow, diffusion resistance and surface
dilational viscosity. All three processes take place in the "neck" regions con-
necting the films to the borders. (It should be noted that in this mechanism
there is a singular limit for d --+ 0, in which the slow drainage of water out
of the pulled film gives a dominant dissipation [17). So long as d remains
finite, this becomes a nonlinear effect, which is, at least formally, irrelevant
to the limit under discussion.)
The dissipation in the Marginal Regeneration mechanism is estimated
as [15] TS = (1Jw + T /(Dc'L}) + K,/ p)'Y 2 where p is the size of a neck
(somewhere between r and d). Comparison with the Marangoni process
described above suggests the latter to be less dissipative for typical foam
parameters. An exception is for ultra-dry foams (d--+ 0) where the Marginal
Regeneration mechanism becomes obligatory.

4.7. MODE STRUCTURE

With the above ideas about water and surfactant flux patterns, we can
identify the following candidate modes:
224

Poiseuille Mode: Driven mainly by disjoining pressure; elastic storage


modulus increment !:l.Gp ~ ITdd/ R ~ crdj(Rr); viscosity increment !:l.TJp ~
'f/W R/ d, hence relaxation frequency

!:l.Gp crd 2
Wp = - - ~ ----::--:--- (21)
/::l.'f/p R 2 dTJw
For wet foams (and parameters relating to the small cell biliquid case)
!:l.G p ~ Go, and wp ~ 1-100Hz. For dry foams, !:l.G p ~Go but wp ~ 100
Hz.
Marangoni Mode: Driven mainly by Gibbs elasticity; elastic storage
modulus increment !:l.G M ~ E / R; viscosity increment !:l.TJM ~ 'f/W R/ d +
"'/ R; relaxation frequency
!:l.GM E
(22)
'f/W R 2 d + K,
WM=--~-----:::,.....,--
/::l.'f/M
For wet foams, !:l.G M ~ Go, but WM ~ 100 - 1000 Hz. For dry foams,
!:l.GM ~Go but WM ~100Hz. Note that in circumstances where Marginal
Regeneration prevails (instead of the Marangoni mode), this again gives
!:l.G ~ Go but w ~ 1000 Hz.

4.8. BOTTOM LINE

As summarised above, the authors of Ref. [15] did not find a candidate
mode which combines both a reasonable amplitude (!:l.G of order 0.1 G 0 ,
say) with a low characteristic frequency (w < 1Hz). Obviously the analysis
is not complete, nor quantitative. Therefore, we do not categorically rule
out slow relaxation modes associated with foam-specific dynamics; but this
approach does not look promising. It at least seems worthwhile to seek
other, more general, mechanisms for slow relaxations. This idea is also
motivated by similarities in the rheology of many soft materials.

5. SOFT GLASSY RHEOLOGY


5.1. SOFT GLASSY MATERIALS
The class of materials whose (low frequency) rheology is very similar to that
of foams and emulsions is actually quite large; clay slurries, pastes, dense
multilamellar vesicles and colloidal glasses are just a few examples [5, 18,
19, 20, 21, 22, 23]. Their elastic (G') and loss (G") moduli often depend only
weakly on frequency w, with G" being typically between one and two orders
of magnitude smaller than G'. In fact G" often shows no sign of decreasing
to zero for w -+ 0 (as it should if the response is truly linear) even at the
lowest frequencies that are accessible experimentally; sometimes it even
225

seems to rise as w decreases. These similarities also extend to nonlinear


rheology. For example, 'flow curves' of shear stress s versus shear (strain)
rate i' in steady shear flow are generally well described by a relation of the
form
8 =By+ c.YP (23)

with an exponent p between 0.1 and 1. If there is no yield stress (sy = 0),
this is called a 'power law fluid', otherwise a 'Herschel-Bulkley model' [24,
25, 26]. Either way, the materials are 'shear-thinning' in that the apparent
viscosity s / i' decreases as the shear rate i' increases.
Such qualitative similarities in the rheology of many soft materials sug-
gest an underlying common cause. An obvious candidate, common to all the
materials listed above, is (mesoscopic) structural disorder. (The importance
of this feature has been noted before for specific systems [5, 12, 13, 15, 27,
28], but we feel that its unifying role in rheological modelling has not been
properly appreciated.) In a foam, for example, the droplets are normally
arranged in a disordered fashion rather than as a regular, crystalline array.
The latter would give a lower free energy, and the disordered state is there-
fore only metastable. The dynamics of transitions between such metastable
states will be slow, because typical energy barriers for rearrangements of
droplets are much greater than kaT. Qualitatively, the same features are
found in all the other materials that we have mentioned. More importantly,
they are very close to what we normally refer to as a glass (except that
the disorder there is on a molecular scale). We express this similarity by
referring to the class of materials, that we now consider, as "soft glassy
materials" (SGM) [29]; the "soft" is added to emphasise that they deform
and flow easily, in contrast to many ordinary glasses.

5.2. BOUCHAUD'S GLASS MODEL

We are aiming for a phenomenological model that can explain the main
features of SGM rheology (both linear and nonlinear) as described above.
To apply to a broad range of SGMs, such a model needs to be reasonably
generic. It should therefore incorporate only a minimal number of features
common to all SGMs, leaving aside as much system specific detail as possi-
ble. We start with the 'glassiness', i.e., the effects of structural disorder and
metastability. An intuitive picture of a glass is that it consists of local 'ele-
ments' (we will be more specific later) which are trapped in 'cages' formed
by their neighbours so that they cannot move. Occasionally, however, a
rearrangement of the elements may be possible, due to thermal activation,
for example. This idea was formalised into an effective one-element model
by Bouchaud (see Refs. (30, 31], where references to earlier work on similar
226
models can also be found): an individual element 'sees' an energy landscape
of traps of various depths E; when activated, it can 'hop' to another trap.
Bouchaud assumed that such hopping processes are due to thermal fluc-
tuations. In SGMs, however, this is unlikely as kaT is very small compared
to typical trap depths E (see later). We assume instead that the 'activation'
in SGMs is due to interactions: a rearrangement somewhere in the material
can propagate and cause rearrangements elsewhere. In a mean-field spirit
we represent this coupling between elements by an effective temperature (or
noise level) x. This idea is fundamental to our model.
We can now write an equation of motion for the probability of finding
an element in a trap of depth E at time t:

:t P(E, t) = -foe-Ef:c P(E, t) + f(t) p(E) (24)

In the first term on the r.h.s., which describes elements hopping out of their
current traps, r 0 is an attempt frequency for hops, and exp(-Efx) is the
corresponding activation factor. The second term represents the state of
these elements directly after a hop. Bouchaud made the simplest possible
assumption that the depth of a new trap is completely independent of that
of the old one; it is simply randomly chosen from some distribution of trap
depths p(E). The rate of hopping into traps of depth E is then p(E) times
the overall hopping rate, given by

f(t) = fo ( e-E!:r:) P = fo j e-E/:c P(E, t) dE (25)

Bouchaud 's main insight was that the model (24) can describe a glass transi-
tion if the density of deep traps has an exponential tail, p (E) "' exp (- E / x g),
say. Why is this? The steady state of eq. (24), if one exists, is given by
Peq(E) "'exp(E/x)p(E); the Boltzmann factor exp(E/x) (no minus here
because trap depths are measured from zero downwards) is proportional to
the average time spent in a trap of depth E. At x = xg, it just cancels the
exponential decay of p(E), and so the supposed equilibrium distribution
Peq(E) tends to a constant for large E; it is not normalisable. This means
that, for x ~ Xg, the system does not have a steady state; instead, it 'ages'
by evolving into deeper and deeper traps [30, 31]. We therefore identify
x = Xg as the glass transition of the model (24). In the following, we choose
energy units such that this transition occurs at x = Xg = 1.
We now have a candidate model for describing the glassy features of
SGM. Its main advantage is that it is simple. Its disadvantages are: (i)
It has no spatial degrees of freedom, hence cannot describe flow-this we
shall fix in a moment. (ii) The assumption of an exponentially decaying
p(E) is rather arbitrary in our context. It can be justified in systems with
227
'quenched' (i.e., fixed) disorder, such as spin glasses, using so-called 'ex-
treme value statistics' [32], but it is not obvious how to extend this argu-
ment to SGM. (iii) The exponential form of the activation factor in (24) was
chosen by analogy with thermal activation. But x describes noise arising
from interactions, so this analogy is by no means automatic, and functional
forms other than exponential could also be plausible. In essence, we view
(ii) together with (iii) as a phenomenological way of describing a system
with a glass transition. We now ask how such a system will flow.

5.3. MODEL FOR SGM RHEOLOGY

To describe deformation and flow, we now incorporate strain degrees of


freedom into the model [29]. As our 'elements' we take mesoscopic regions
of our SGM. By mesoscopic we mean that these regions must be (i) small
enough for a macroscopic piece of material to contain a large number of
them, allowing us to describe its behaviour as an average over elements;
and (ii) large enough so that deformations on the scale of an element can
be described by an elastic strain variable. For a single droplet in a foam,
for example, this would not be possible because of its highly non-affine
deformation; in this case, an element should therefore be at least a few
droplet diameters across. We choose the size of the elements as our unit
length (to avoid cumbersome factors of element volume in the expressions
below).
We denote by l the local shear strain of an element (more generally,
the deformation would have to be described by a tensor, but we choose a
simple scalar description). This is measured from the nearest equilibrium
configuration of the element, i.e., the one it would relax to if in complete
isolation and without external stresses. When an element is deformed, l
will first increase from zero. Assuming the deformation in this regime to
be elastic, there will be a local shear stress s = kl; k is an appropriate
elastic constant, which we take to be the same for all elements. On further
deformation, however, the element must eventually yield and rearrange into
a new equilibrium configuration; the local strain l is then again zero. This
happens when the elastic strain energy ~kl 2 approaches a typical yield
energy E; due to the disordered structure of the material, the value of this
yield energy will in general be different for each element. We can view
such yielding events as 'hops' out of a trap (or potential well), and identify
the yield energy E with the trap depth. As before, we assume that yields
(hops) are activated by interactions between different elements, resulting
in an effective temperature x. The activation barrier is now E- ~kl 2 , the
difference between the typical yield energy and the elastic energy already
stored in the element.
228
We haven't as yet specified how elements behave between rearrange-
ments. The simplest assumption is that their strain changes along with the
macroscopically imposed strain 1. This means that, yielding events apart,
the shear rate is homogeneous throughout the material; on the other hand,
the local strain l and stress s are inhomogeneous because different elements
yield at different times. We therefore now need to know the joint proba-
bility of finding an element with a yield energy E and a local strain l to
describe the state of the system. An appropriate model for this was set up
in Ref. [29]; within the model, the probability evolves in time according to

The first term on the r.h.s. describes the motion of the elements between
rearrangements, with a local strain rate equal to the macroscopic one, i = -y.
The interaction-activated yielding of elements is reflected in the second
term. The last term incorporates two assumptions about the properties of
an element just after yielding: It is unstrained (l = 0; this assumption can
be relaxed without qualitative changes to our results) and has a new yield
energy E randomly chosen from p(E), i.e., uncorrelated with its previous
one. Finally, the total yielding rate is given by the analogue of (25),

Eq. (26) tells us how the state of the system, described by P(E, l, t), evolves
for a given imposed macroscopic strain 1(t). What we mainly care about is
of course the rheological response, i.e., the macroscopic stress. This is given
by the average of the local stresses

s(t) = k (l)p =: k j l P(E, l, t) dEdl (28)

In the absence of yielding events, the equation of motion (26) then predicts
a purely elastic response: D..s = s(t) - s(t') = kt1{. This is a consequence
of our assumption that in between rearrangements, the response of each
individual element is purely elastic. In reality, there are also viscous contri-
butions; in foams, these are due to the flow of water and surfactant caused
by the deformation of the elements. In the low frequency regime of interest
to us, such viscous effects are insignificant (see Section 4) and can be ne-
glected. At high frequencies, this is no longer true and the model (26,28)
would have to be modified appropriately to yield sensible predictions.
With (26,28), we now have a minimal model for the (low frequency)
rheology of SGM: It incorporates both the 'glassy' features arising from
229
structural disorder (captured in the distribution of yield energies E and lo-
cal strains l) and the 'softness': for large macroscopic strains, the material
flows because eventually all elements yield. One interesting consequence of
this is that 'flow interrupts aging': Above, we saw that below the glass tran-
sition (x < 1), the system evolves into deeper and deeper traps; it ages. In
the presence of steady shear flow (i' =canst), however, this doesn't happen:
As the local strain l increases with the macroscopic one, the activation bar-
rier E- ~kl 2 of any element decreases to zero in a finite time, for any trap
depth E. The system can therefore not get stuck in progressively deeper
traps; aging is 'interrupted' [33] and ergodicity is restored.

5.4. PREDICTIONS FOR LINEAR RHEOLOGY

We now summarise the predictions [29] of the model defined by (26,28)


for the linear (shear) rheology of SGM. We choose units for energy and
time such that Xg = f 0 = 1. We also set k = 1, which can always be
achieved by a rescaling of the strain variables 1 and l. In these units, typ-
ical yield strains ..}2E/ k are of order one. Finally, for the density of yield
energies ('trap depths') we assume the simplest form with an exponential
tail, p(E) = exp( -E). The only parameter that distinguishes between dif-
ferent systems is then the effective noise temperature x. Note that this
is not a parameter that we can easily tune from the outside; rather, we
expect it to be determined self-consistently by the interactions in the sys-
tem. This should be borne in mind when we use expressions like "as x
increases/decreases" below.
The complex linear modulus predicted by the model [29] turns out to
be rather simple:
G*(w) = (. iwr ) (29)
ZWT +1 eq
This an average over Maxwell modes with relaxation times r. The distri-
bution of r follows from the equilibrium distribution of energies, Peq(E),....,
exp(E/x)p(E). Here Tis given by r = exp(E/x), the 'lifetime' (time be-
tween rearrangements) of an element with yield energy E, and this leads
to a power-law relaxation time distribution Peq(r) ,...., T-x (forT ~ 1). As
x decreases towards the glass transition, the long-time tail of the spec-
trum becomes increasingly dominant and causes anomalous low frequency
behaviour of the moduli:
G" ,...., w for 2 < x, for 1 < x < 2
G' ,...., w2 for 3 < x, (30)
for 1 < x < 3

These are illustrated in Fig. 10. The main point to note is that for 1 < x < 2,
i.e., not too far from the glass transition, G' and G" vary as the same power
230

/
/
/
/
/
/
/
/
/ x=2.5 x= 1.5
/
/

------- .......... , f---------- . . . ,


"'
' "'
x= 1.1 x= 1.05 \
10-3
10-4 10-3 10-2 10-1 10° 10-4 10-3 10-2 10-1 10°
(1) (1)

Figure 10. Linear moduli G' (solid line) and G" (dashed) vs frequency w at various
noise temperatures x above the glass transition.

of frequency (wx- 1 ); their ratio is therefore constant. Furthermore, as we


approach the glass transition (x---+ 1), this power law becomes increasingly
'flat'. These predictions of the model are compatible with many experimen-
tal results [5, 18, 19, 20, 23).
The above linear results only apply for x > 1, where there is a well
defined equilibrium state around which small perturbations can be made.
However, if a cutoff Emax on the yield energies is introduced (which is
physically reasonable because yield strains cannot be arbitrarily large), an
equilibrium state also exists for x < 1, i.e., below the glass transition.
(Strictly speaking, with the cutoff imposed there is no longer a true glass
phase; but if the energy cutoff is large enough, its qualitative features are
expected to be still present.) One then finds for the low frequency behaviour
of the linear moduli:

G'""' const. (31)

This applies as long as w is still large compared to the cutoff frequency,


Wmin = exp( -Emax/x). In this frequency regime, G" therefore increases
231

=
Figure 11. Flow curves: shear stress s vs shear rate .Y, for x 0.25, 0.5, ... , 2.5 (top to
bottom on left); x = 1, 2 are shown in bold. Inset: small .Y behavior, with yield stresses
for x < 1 shown by arrows.

as w decreases, again in qualitative agreement with recent experimental


observations [5, 21, 22, 23].

5.5. PREDICTIONS FOR NONLINEAR RHEOLOGY

The model of Ref. [29] can also be used to predict nonlinear rheological fea-
tures. This is especially important, because arguably the linear behaviour
described above follows inevitably from the existence of a power law distri-
bution of relaxation times: if we were only interested in the linear regime,
it would be simpler just to postulate such a power law! But in fact, an
exact (scalar) 'constitutive equation' relating the stress at a given time to
the strain history up to that point can be derived [34]. Therefore the model
allows one to probe the nonlinear regime in detail.
Here, we only discuss results for the flow curves, i.e., shear stress s vs
shear rate i' in steady flow (Fig. 11). For high shear rates, strong shear
thinning is observed for all x; the stress increases only very slowly with
i' as s "' (x In )') 112 . More interesting is the low shear rate (i' ~ fo = 1)
232
behaviour, where three different regimes can be distinguished. (i) For x > 2,
i.e., far above the glass transition, the behaviour is Newtonian, s = rry. The
viscosity, which is simply the average relaxation time, diverges as x -t 2
(i.e., at twice the glass transition 'temperature'). This signals the onset of
a new regime: (ii) For 1 < x < 2, one has a power law fluid, s -yx-l.
f"V

The exponent decreases smoothly from 1 to 0 as the glass transition is


approached. (iii) In the glass phase (x < 1), finally, there is a nonzero yield
stress (as one would intuitively expect for a glass). This shows a linear
onset, sy f"V 1 - x, as x decreases below the glass transition temperature.
Beyond yield, the stress again increases as a power law, s- sy -y 1-x.
f"V

The behaviour of our model in regimes (ii) and (iii) therefore matches
respectively the power-law fluid [24, 25, 26] and Herschel-Bulkeley [24, 25]
scenarios as used to fit the experimental nonlinear rheology of many SGMs.

5.6. INTERPRETATION OF MODEL PARAMETERS

Our model for SGM captures important rheological features that have been
observed in a large number of experiments, at least in the region around
a 'glass transition'. Using a mean-field (one element) picture, it is also
simple enough to be generic. The main challenge now is the interpretation
of the model parameters, namely, the 'effective noise temperature' x and the
'attempt frequency' r 0 • To answer these questions, we should really start
from a proper model for the coupled nonlinear dynamics of the 'elements'
of a SGM and then derive our present model within some approximation
scheme. At present, we do not know how to do this.
We can nevertheless interpret the activation factor exp[-(E- !kl 2 )/x]
in (26) as the probability that (within a given time interval of order 1/ro)
an element yields due to a 'kick' from a rearrangement elsewhere in the ma-
terial. Therefore x is the typical activation energy available from such kicks.
But while kicks can cause rearrangements, they also arise from rearrange-
ments (whose effects, due to interactions, propagate through the material).
So there is no separate energy scale for kicks: Their energy must of the
order of the energies released in rearrangements, i.e., of the order of typical
yield energies E. In our units, this means that x should be of order unity.
Note that this is far bigger than what we would estimate if x represented
true thermal activation. For example, the activation barrier for the simplest
local rearrangement in a foam (a T1 or neighbour-switching process) is of
the order of the surface energy of a single droplet; this sets our basic scale
for the yield energies E. Using typical values for the surface tension and a
droplet radius of the order of one p,m or greater, we find E ;::: 10 4 kBT· In
our units E = 0(1), so thermal activation would correspond to extremely
small values of x = kBT :::; 10-4 •
233
We now argue that x may not only be of order one, but in fact close
to one generically. Consider first a steady shear experiment. The rheologi-
cal properties of a sample freshly loaded into a rheometer are usually not
reproducible; they become so only after a period of shearing to eliminate
memory of the loading procedure. In the process of loading one expects a
large degree of disorder to be introduced, corresponding to a high noise tem-
perature x ~ 1. As the sample approaches the steady state, the flow will
(in many cases) tend to eliminate much of this disorder [35] so that x will
decrease. But, as this occurs, the noise-activated processes will slow down;
as x --+ 1, they may become negligible. Assuming that, in their absence,
the disorder cannot be reduced further, xis then 'pinned' at a steady-state
value at or close to the glass transition. This scenario, although extremely
speculative, is strongly reminiscent of the 'marginal dynamics' seen in some
mean-field spin glass models [36].
Consider now the attempt frequency f 0 • It is the only source of a char-
acteristic timescale in our model (chosen as the time unit above). We have
approximated it by a constant value, independently of the shear rate i'; this
implies that f 0 is not caused by the flow directly. One possibility, then, is
that ro arises in fact from true thermal processes, i.e., rearrangements of
very 'fragile' elements with yield energies of order kaT. This mechanism can
give a plausible rheological time scale if one local rearrangement can trigger
a long sequence of others [29], as may be the case in foams [12, 28]. Other
possible explanations for the origin of r 0 include internal noise sources, such
as coarsening in a foam, and uncontrolled external noise sources (traffic go-
ing past the laboratory where the rheological measurements are performed,
for example). The rheometer itself could also be a potential source of noise;
this would however suggest at least a weak dependence of r 0 on the shear
rate i'· We cannot at present say which of these possibilities is most likely,
1or rule out other candidates. The origin of f 0 may not even be universal,
mt could in fact be system specific.

:. CONCLUSION

'hese lectures were intended to summarise our current understanding of


near and nonlinear viscoelasticity in foams. The study of the low frequency
1ear elastic modulus G'(w --+ 0) is well-established, but, as described in
)Ction 3, there is a clear discrepancy between the predictions of simplis-
: ordered models (in both two and three dimensions) and the observed
1lume fraction dependence of this quantity. This is partially explained
· the anomalous spring constant between droplets at weak contact (the
orse-Witten effect) but a full explanation also requires disorder. The lat-
r idea was proposed several years ago by Dennis Weaire and others, but
234

has only recently been implemented in a three dimensional model [13]. The
low frequency loss modulus, G"(w) is much harder to understand; the data
shows a clear anomaly in that, at the lowest attainable frequencies, this
quantity appears to be constant or even rising as the frequency is low-
ered. Attempts to explain this in terms of foam-specific mechanisms were
summarised in Section 4. Although the qualitative analysis of surfactant
transport that this entails is certainly of some value, the basic conclusion is
that there is no obvious candidate among such mechanisms for explaining
the anomalous dissipation in foams. Again, one is drawn to disorder as a
general explanation.
Accordingly in Section 5, we have described a recent phenomenological
model for foam rheology. It captures in a simple yet generic way the effect of
mesoscopic structural disorder and metastability; these features are shared
by many other 'soft glassy materials'. Thus the model can account for the
main qualitative features of the rheology, not only of foams, but of other
systems such as slurries and pastes which are commonly observed to show
weak power law behaviour and/or near constant loss modulus. The model
offers an intriguing link between the linear viscoelastic spectrum and the
nonlinear flow curves. However, the interpretation of its parameters, no-
tably the 'effective noise temperature' x, remains to be clarified. To do this
may require study of a more fundamental model involving strongly coupled
degrees of freedom (as undoubtedly are present in soft glassy materials),
rather than the mean-field description used so far.

6.1. ACKNOWLEDGEMENTS

We are indebted to our colleagues Martin Buzza, Pascal Hebraud, Francois


Lequeux, and David Lu, much ofwhosejoint work is described in Sections 4
and 5. We are especially grateful to Martin Buzza for permission to use
Figs. 1-5, 8 and 9 (37], and to Dov Levine for providing Fig. 6.

References
1. L D Landau and E M Lifshitz. Statistical Physics, Part 1. Pergamon, Oxford, 1980.
2. H M Princen. J. Coli. Inter/. Sci., 71:55, 1979.
3. H M Princen. Rheology of foams and highly concentrated emulsions. 1: Elastic
properties and yield stress of a cylindrical model system. J. Coli. Inter/. Sci.,
91:160-175, 1983.
4. H M Princen and A D Kiss. Rheology of foams and highly concentrated emulsions.
3: Static shear modulus. J. Coli. Inter/. Sci., 112:427-437, 1986.
5. T G Mason, J Bibette, and D A Weitz. Elasticity of compressed emulsions. Phys.
Rev. Lett., 75:2051-2054, 1995.
6. D M A Buzza and M E Cates. Uniaxial elastic-modulus of concentrated emulsions.
Langmuir, 10:4503-4508, 1994.
7. M D Lacasse, G S Grest, and D Levine. Deformation of small compressed droplets.
Phys. Rev. E, 54:5436-5446, 1996.
235

8. K Brakke. Exp. Math., 1:141, 1992.


9. D C Morse and T A Witten. Droplet elasticity in weakly compressed emulsions.
Europhys. Lett., 22:549-555, 1993.
10. S Hutzler and D Weaire. The osmotic-pressure of a 2-dimensional disordered foam.
J. Phys. Cond. Matt., 7:L657-L662, 1995.
11. S Hutzler, D Weaire, and F Bolton. The effects of plateau borders in the 2-
dimensional soap froth. 3: Further results. Phil. Mag. B, 71:277-289, 1995.
12. D J Durian. Foam mechanics at the bubble scale. Phys. Rev. Lett., 75:4780-4783,
1995.
13. M D Lacasse, G S Grest, D Levine, T G Mason, and D A Weitz. Model for the
elasticity of compressed emulsions. Phys. Rev. Lett., 76:3448-3451, 1996.
14. D Levine. Private communication.
15. D M A Buzza, C Y D Lu, and M E Cates. Linear shear rheology of incompressible
foams. J. Phys. (France) II, 5:37-52, 1995.
16. S R de Groot and P Mazur. Non-equilibrium thermodynamics. Dover Publications,
New York, 1984.
17. L W Schwartz and H M Princen. A theory of extensional viscosity for flowing foams
and concentrated emulsions. J. Coli. Inter/. Sci., 118:201-211, 1987.
18. M R Mackley, R T J Marshall, J B A F Smeulders, and F D Zhao. The rheological
characterization of polymeric and colloidal fluids. Chern. Engin. Sci., 49:2551-2565,
1994.
19. R J Ketz, R K Prudhomme, and W W Graessley. Rheology of concentrated microgel
solutions. Rheol. Acta, 27:531-539, 1988.
20. SA Khan, C A Schnepper, and R C Armstrong. Foam rheology. 3: Measurement
of shear-flow properties. J. Rheol., 32:69-92, 1988.
21. P Panizza, D Roux, V Vuillaume, C Y D Lu, and M E Cates. Viscoelasticity of the
onion phase. Langmuir, 12:248-252, 1996.
22. H. Hoffmann and A. Rauscher. Aggregating systems with a yield stress value. Coli.
Polymer Sci., 271:390-395, 1993.
23. T G Mason and D A Weitz. Linear viscoelasticity of colloidal hard-sphere suspen-
sions near the glass-transition. Phys. Rev. Lett., 75:2770-2773, 1995.
24. S D Holdsworth. Rheological models used for the prediction of the flow properties
of food products. Trans. Inst. Chern. Eng., 71:139-179, 1993.
25. E Dickinson. An introduction to food colloids. Oxford University Press, Oxford,
1992.
26. H A Barnes, J F Hutton, and K Walters. An introduction to rheology. Elsevier,
Amsterdam, 1989.
27. D Weaire and M A Fortes. Stress and strain in liquid and solid foams. Adv. Phys.,
43:685-738, 1994.
28. T Okuzono and K Kawasaki. Intermittent flow behavior of random foams - a
computer experiment on foam rheology. Phys. Rev. E, 51:1246--1253, 1995.
29. P Sollich, F Lequeux, P Hebraud, and M E Cates. Rheology of soft glassy materials.
Phys. Rev. Lett., 78:2020-2023, 1997.
30. J P Bouchaud. Weak ergodicity breaking and aging in disordered-systems. J. Phys.
(France) I, 2:1705-1713, 1992.
31. C Monthus and J P Bouchaud. Models of traps and glass phenomenology. J. Phys.
A, 29:3847-3869, 1996.
32. J P Bouchaud and M Mezard. Universality classes for extreme value statistics.
LPTENS preprint 97 /XX. To be published.
33. J P Bouchaud and D S Dean. Aging on Parisis tree. J. Phys. (France} I, 5:265-286,
1995.
34. P Sollich. Exact constitutive equation for soft glassy rheology. In preparation.
35. D Weaire, F Bolton, T Herdtle, and H Aref. The effect of strain upon the topology
of a soap froth. Phil. Mag. Lett., 66:293-299, 1992.
36. L F Cugliandolo and J Kurchan. Analytical solution of the off-equilibrium dynamics
of a long-range spin-glass model. Phys. Rev. Lett., 71:173-176, 1993.
37. D M A Buzza. Theory of Emulsions. PhD thesis, Cambridge University, 1994.
236
SURFACTANTS AND STRESS CONDITIONS AT FLUID INTERFACES

Kathleen J. Stebe & Charles D. Eggleton, The Johns Hopkins University, Baltimore,

Maryland 21218

1. Introduction

Surfactants are amphiphilic molecules that adsorb on fluid interfaces, where they
reduce the surface tension. If the interface is moving, surface convection can
distribute surfactant non-uniformly, creating surface tension gradients or Marangoni
stresses. These two phenomena are the keys to understanding the stress response of
surfactant-laden interfaces in multiphase flows.
Consider the simple example of a droplet suspended in a surfactant solution at
concentration C..,. If the droplet is not moving, surfactant adsorbs onto the interface,
establishing a surface concentration r.q in equilibrium with the bulk, and the
corresponding surface tension Yeq· If the droplet begins to translate, (as shown in
Figure 1, in the drop-fixed reference frame), the interface becomes a streamline in the
flow, moving from the leading to the trailing pole. Surfactant adsorbed on the
interface is swept by

dlr~-fton of
? ' · '·
\ -X·f
' ?
dlurtlon of
surface
•· ··'J-,_ 'l
~
Maran~onl concnetration

"'
stress ~~ r'\.'\. er&dlent
... ,
+- i
' \:"'- /~
.7'
·· - -x --~

u • + •
Figure I . Non-defonning drop translating in a surfactant solution.

237
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 237-258.
© 1999 Kluwer Academic Publishers.
238

surface convection from the leading pole toward the trailing pole. Non-uniform
surface concentrations develop if the rates of mass transfer that restore equilibrium
(adsorption-desorption and diffusion) are slow compared to the surface convection.
If the surfactant does collect at the trailing pole, the surface tension y there is
less than at the leading pole, creating a Marangoni stress. The interface pulls from
regions of low tension toward those with high tension, opposing the surface velocity
and increasing the drag resisting the flow.

The interfacial deformation is also altered by y gradients. The ability of the


interface to resist normal stress jumps is giver, by the Laplace pressure, the product of
y and the mean curvature of the interface 2H. TherefQre, regions of lower y need
higher curvatures to support a stress jump, and tend to become pointed.
The stress balance at an interface is expressed: [1]

(1)
where Pi denotes the pressure evaluated at the interface in each phase and 'tj denotes
the viscous stress tensor. The first term on the right hand side of (1) is the Laplace
pressure; the second is the Marangoni stress, where the symbol v. denotes the surface
gradient operator, and n denotes the surface normal. Through y, these stresses are
highly coupled with the surfactant mass transfer kinetics.
The manner in which surfactants can effect fluid flows can be understood by
considering simple flow geometries such as the settling drop depicted in Figure 1.
Interfacial regions from which the surface flow diverges, such as the leading pole of
the droplet, can be depleted of surfactant. Regions of converging surface flow, like
the trailing pole, can accumulate surfactant. Stretching dilutes the interface. The
discussions given in this chapter can be generalized to more complex flows if such
regions are considered for the flow field.
2. Basic Equations

When an interface is formed in a quiescent surfactant solution of concentration Coo,


surfactant in the fluid sublayer immediately adjacent to the interface adsorbs. This
depletes the sublayer concentration C., causing surfactant to diffuse from the bulk
toward the interface. This continues until equilibrium is established. For the
purposes of this chapter, the Langmuir framework is adopted to describe the
adsorption process. The adsorption flux is assumed to be first order in sublayer
concentration, C., and first order in space available on the interface. The desorption
239
flux is assumed first order in surface concentration r. So, the ad/desorption flux to the
interface is :

(2)
where ~ is the kinetic constant for adsorption, a is the kinetic constant for desorption
and r"' is the maximum packing of surfactant.

The diffusion flux from the bulk is equal to the adsorption flux. The diffusion
flux is given by:

(3)
At equilibrium, these fluxes are equal to zero. The bulk concentration becomes
uniform at C"' and r obeys the equilibrium adsorption isotherm:

r.q - _k_. k - f3Coo


roo - 1 + k' - a (4)

where k is the adsorption number, the ratio of characteristic desorptive timescales to


characteristic adsorptive timescales. Equation (4) accounts for monolayer saturation;
as k becomes large, the surface concentration r approaches its upper bound r ro· The
isotherm is related to the surface equation of state by the Gibbs adsorption equation.
For the Langmuir isotherm, the surface equation of state is:

y =Yo+ RTf'"' ln(l- r) ['

<Xl (5)

where Yo is the surface tension of the surfactant-free interface, and RT is the product of
the ideal gas constant and the temperature.
This equation is used to understand dynamics at interfaces as well, i.e. it is
assumed that the surface tension instantaneously reflects the surface concentration.
Given the surfactant distribution, the state of stress of the interface is known. In the
limit of small r If'"'' (4) and (5) can be approximated by linear functions of the surface
and bulk concentrations. This linear dependence of y(r) is often assumed in studies
of surfactant effects. However, important insights are missed if this assumption is
adopted; the simplest framework that allows reasonable comparison to experiment is
the Langmuir model used here.
For interfaces in motion, the surface mass balance is: [2,3]

(6)
where the flux jn is given in general by the equality between ja1d and jn, as defined in
(2) and (3), above. This balance states that the rate of change of surfactant
240

concentration along the interface is determined by the surface convection tangential to


the interface, the dilution of the interface by stretching (the second and third terms on
the left hand side of (6}), and the flux of surfactant from the bulk jn. Equation (6)
determines the instantaneous surfactant distribution, r, and, through (5), the
instantaneous surface tension profile. (Surface diffusion is usually negligible compared
to surface convection, and is neglected here.) The Marangoni stress can be expressed,
using a chain rule, as

(7)
Consider again the example of a droplet of radius a which translates in a
surfactant solution with a velocity U. If the Laplace pressure and pressure Pi are scaled
with the quantity y.Ja, the surface concentration scaled with r eq• and lengths scaled
with a, and viscous stresses scaled by JJ.U/a, the stress balance can be written in
dimensionless form as:

(8)
where, the dimensionless surface tension is:

y = _ro + E ln( l - xD
Ye~~ (9)
which is unity at equilibrium. In (8}, a number of dimensionless groups appear.
• Ca is the capillary number, a ratio of viscous stresses which act to deform the
droplet to surface tension which resists the deformation.
Ca= JJ.U/yeq (10)
For Ca <<1, interfacial deformation can be neglected; the drop in Figure 1 will remain
spherical. The normal stress balance is then satisfied in an integral sense, and only the
tangential stress balance is retained at the interface.
• E is the elasticity number, a measure of the sensitivity of the surface tension to the
surface concentration.
For typical surfactants at room temperature, this number ranges from roughly 0.1 to
0.2. (see Eggleton et al. )
(11)
• The quantity x is the fraction of the interface that is covered by surfactant at
equilibrium
241

x= r.q rr 00 (12)
For x<<l, (9) becomes:
y = y,Jy.q -Exr (13)

ForE of 0.2 and x<<l, the slope Ex is extremely small. The non-linear form of the
surface tension and Marangoni stress in (8) and (9) show that the coupling between
y(r) can be very strong for x approaching unity without adopting unrealistic parameter
values. Furthermore, the singular form of the Marangoni stress as x approaches unity,
and the non-linear form of the adsorption isotherm, dictate the system hydrodynamics,
as will be shown below.
Consider (2)-(7) in the context of a drop of radius a translating through a
surfactant solution. Several characteristic timescales can be defined in this system.
• The timescale for desorption is a·1; for adsorption is (~Coor 1 . Their ratio is given
byk.
• The characteristic timescale for diffusion from the bulk to the interface is h 2/D. In
this expression, the adsorption depth h = r.qiCoo is a characteristic depth that is
depleted by surfactant adsorption.
• The characteristic timescale for surfactant to be convected from one drop pole to
the other is a/U.
The ratio of convective to desorptive timescales is the Biot number:
Bi = aa/U (14)
Defining the Peclet number:
Pe= aUlD (15)
the ratio of timescales for diffusion from the bulk to convection along the drop surface
is:
(16)
(In fact, the ratio A must be modified for cases of large Pe to account for diffusion
boundary layers that develop near the droplet interface. This is discussed in detail in
Stebe eta/. [4]). For fixed bulk concentration (fixed k) the occurrence of Marangoni
stresses depends on the magnitude of A and Bi. If both A<<I and Bi >> 1, mass
transfer is rapid compared to the surface convection, and no Marangoni stresses
develop. If A~l, or Bi ::;I, mass transfer is too slow to prevent non-uniform surface
concentrations from developing. Marangoni stresses develop which alter the flow
field.
The material parameters that determine the state of stress of the interface are
those that combine to form the dimensionless groups E, A and Bi. The parameters roo ,
Yo and ~Ia are found from equilibrium studies of the surface tension as a function of
242

bulk concentration. The mass transfer coefficients D and a. are found from studies of
the dynamic surface tension by such methods as the oscillating bubble, pendant bubble,
growing drop, capillary wave or other techniques [5-10]. Typically, the kinetic
constants for adsorption-desorption ~ and a. are a function of the chemical structure of
the surfactant and the fluids of which the interface is composed.
Note that while Bi does not depend on the bulk concentration of surfactant, A
does through the adsorption depth h. This allows the control of stresses exerted at
interfaces by the manipulation of the bulk concentration. The ramifications of this are
significant, and are discussed in section 5, below.

3. Stagnant Interfaces

Under certain conditions, Marangoni stresses oppose the surface velocity so strongly
that the surface velocity is reduced nearly to zero, and the drop interface acts like a
solid, no-slip surface. For example, small settling drops ordinarily move with Stokes'
velocity, (the velocity one would find for a rigid sphere), which is slower than the
theoretical prediction for a freely convecting interface. It has long been understood [1]
that this difference is attributable to surfactant adsorption, and to Marangoni stresses.
Consider again the problem of a non-deforming sphere of radius a translating
through a fluid. If the timescales for adsorption-desorption or diffusion are sufficiently
slow (i.e. Bi<<1, A>>1) the surfactant can be treated as insoluble. The mass balance
of surfactants along the interface is given by:

v. • (rv,) = o (17)
On an axisymmetric interface, this can be integrated to show:
rv, = 0 (18)
10

0.8

06
D.<

02

.0.2 '·
·IU

.06
.......
.0.8 ~ ····-··········
0 aJ ~ 60 Ill 100 1al 1<0 160 100

Figure 2. A qualitative sketch of stagnant-cap behavior. The cap angle, divides stress e,,
free, surfactant-free mobile regions from highly stressed, immobile regions.
243

Thus, the interface can be divided into surfactant covered regions near the trailing end
of the drop which are immobile and surfactant-free regions which are mobile. A cap
angle, Sc, at which the interface changes behavior can be defined. For 8 measured
from the leading toward the trailing pole, a qualitative sketch Of the profiles for r, Vt
and the Marangoni stress are shown in Figure 2. The distribution of surfactant and the
details of the flow field are determined by the solution of the hydrodynamic equations
subject to the tangential stress condition, given in dimensionless form as:
E 1 or
Tro = Ca {1- xr) 88
(19)

Consider the development of the steady distribution from a system at rest and at
equilibrium. Suppose that the drop is initially standing still, with an equilibrium
distribution of surfactant (i.e. r =1). lfthe drop then begins translating, the tangential
velocity at the surface sweeps adsorbed surfactant toward the trailing pole. lf x is
sufficiently small, the Marangoni stresses resisting the surface convection are weak.
Surfactant is swept away from the leading end of the drop, and toward the trailing end.
The result is a spherical droplet with a stagnant cap at the trailing end. When the
gradients in rare significant, Marangoni stresses develop, locking in the r, v. profiles.
lf the same experiment were performed for x near one, the stresses resisting even
small surface concentration gradients are large. The surface velocity is driven to zero
everywhere. Detailed solutions to this problem for the case of linear y(r) [11,12] and
Langmuirian formulations have been performed [13]. The Marangoni stresses
increase the drag resisting the flow, decreasing the drop translation velocity U.

The existence of the stagnant cap was determined from the mass balance for
an insoluble surfactant. lf a surfactant is soluble, the entire range of surface behaviors
can be realized. In the limit of Bi<< 1 or A>> 1, the surface approaches the insoluble
behavior discussed above. As mass transfer rates increase, the surface concentration
approaches its equilibrium distribution. For Bi>>l and A<<l, only small
perturbations from equilibrium distributions are realized and the velocity is only
slightly retarded everywhere along the interface. Small increases in drag are realized.
This is the uniform retardation regime discussed by Levich [1] and Holbrook and
Levan [14, 15]. Finally, in the limit of instantaneous mass transfer rates, the interface
is always in equilibrium with the bulk solution. No Marangoni stresses are realized,
and the droplet moves as if no surfactant were present. The full range of stress
behaviors for a settling droplet for an adsorption-desorption controlled surfactant have
been demonstrated for a settling droplet [16].

Interfacial deformation was neglected in this discussion, but is considered


below.
244
4. Surfactant Effects on Strongly Deforming Interfaces

Recently, the problem of an initially spherical drop of radius a deforming in a Taylor


extensional flow of strain rate G has been used as a means to understand the effects of
surfactants on strongly deforming interfaces [17-21]. The flow is depicted in Figure 3.
Since the characteristic velocity scale for this flow is Ga the capillary number is
defined:
Ca= j.WG/ Yeq (20)

v.....G

s ,

Tip

Equotor

Figure 3. A drop centered in a pure axisymmetric extensional flow of strain rate, G, is shown.

In the absence of surfactants, the droplet deformation has been studied extensively (22,
23]. For a given drop radius and fluid pair, increasing Ca corresponds to increasing
the applied strain rate. As the straining flow becomes stronger, the drop deformation
from its spherical shape increases, until surface tension cannot balance the applied
viscous stress and no steady shape can be realized. For the case where the inner and
outer fluids have the same viscosity, most of the steady shapes are ellipsoidal. The
deformations are quantified as the difference between the drop breadth b and length a:
Df= (a-b)/(a+b) (21)

From the numerical solution of Stokes equations for a drop of the same viscosity as the
surrounding fluid, Df can be found as a function of Ca. The results for a surfactant-
free drop are graphed in Figure 4 as the curve labeled C. (Attention is limited here to
drops with the same viscosity as the external fluid).

Clearly, surfactant adsorption will reduce the surface tension. This reduction
is accounted for in the definition of Ca in (20). However, the interplay of the
245

surfactant mass transfer and the stress conditions alters the interfacial deformation in a
complex manner that is not captured by re-scaling.

0 .4 c
x=0 .996
0 .3

DF o.2

0.1

0 .01 0.02 0 .03 0 .04 0 .05 o.oe 0 .07 o.oa


Ca ..

Figure 4. Drop defonnation as a function ofCa is shown for a drop with a clean interface C, with
dilute surface concentration (x=O.l) and with a high surface concentration (x=0.996).

The tangential velocities along the interface are indicated by the arrows in
Figure 3. Consider first an insoluble surfactant. Surface convection tends to sweep
surfactant from the drop equator towards the tips (or poles). If strong surface
concentration gradients are realized, then the surface tension at the poles is reduced
relative to the equatorial region. The normal stress balance in (1) demands that the
local curvature increase; the droplet becomes more pointed, deforming outward into
the highly straining regions of the flow. This is termed tip-stretching; through this
mechanism, surfactants can lead to deformations far larger than would be realized by a
clean droplet at a given Ca. If however, weak gradients in surface concentration are
realized, the surface tension remains nearly constant along the interface. As the
droplet deforms, the surfactant concentration becomes increasingly dilute, and the
mean surface tension increases above Yeq· The surface resists stretching more strongly
than is indicated by the scale for Ca, and lower deformations are realized. Thus, the
details of the surfactant distribution are key to understanding the manner in which
surfactants alter the behavior of these interfaces; this distribution is determined by the
Marangoni stress. ( 17]

Much of the previous work in this area has been limited to a linear y(I)
relation. Recently, we have considered this problem for an insoluble surfactant using
(5) to relate y(I) [21]. This allowed us to investigate the role of surface saturation in
determining the stress behavior. Consider (5), recast in terms of the surface pressure
246

n= and the normalized area/molecule of adsorbed surfactant, 1/x. The graph of


1to-1teq
1t vs. 1/x is shown in Figure 5. The surface pressure 1t is the 2-D stress that must be
applied to compress the interface. As the surfactant molecular area approaches its
minimum, the surface pressure diverges. This is the 2-D analogue of a pressure-
volume isotherm that accounts for the excluded volume posed by spherical particles in
three dimensions.

\
rr i
R'T['.
\
\
-" .......... ,
---------·-------- --
1.0 I." L2 l.J I_L 1.~ 1.& U 1.9 1.9 2.(

-
1
X

Figure 5. The surface pressure n for a monolayer forming surfactant


as a function of normalized area/molecule.

For dilute interfaces, the area/molecule can change without provoking a large
change in 1t. However, as x approaches 1, the surface pressure is extremely sensitive to
small changes in r , enforcing the excluded area limit. The divergence of 1t as x
approaches unity gives rise to the pole in the Marangoni stress in (7). Marangoni
stresses are weak for small x, large for x approaching unity.

This problem is studied using a quasi-static approach. Initially, the droplet


has an equilibrium distribution on the interface. The extensional flow is applied at a
given Ca. The flow tends to sweep surfactant toward the drop tips. For small x,
Marangoni stresses are weak. Large gradients in r can be realized. The local surface
tension near the drop poles is strongly reduced. Tip stretching dominates, and
interfacial stretching is significantly higher than in the absence of surfactants. These
droplets break at smaller Ca. The profiles presented here include the effect of weak
surface diffusion. In the absence of surface diffusion, for Ca at which stable shapes are
attained, the interface can be divided into stagnant regions at the drop tips, and mobile
regions near the drop equator. The deformation as a function of Ca for x=O.l is
graphed in Figure 4. The profiles for r, v1 and the Marangoni stress are given in
247

Figure 6. Note that r is normalized by its equilibrium value, which is small for x=O.l,
large for x=0.996.

Recall that r=l before the flow is initiated. For x near 1, the Marangoni
stress in (7) has a first order pole. The surface velocity would sweep surfactant
toward the drop tips, but this would allow x to approach unity, and let the Marangoni
stresses diverge. In order that the Marangoni stresses remain bounded in this limit, the
flow adjusts so as to force v.r to zero. Thus, r is nearly uniform. Recall that the
surfactant is insoluble, so the interface dilutes as the deformation proceeds. The
surface tension becomes higher than its equilibrium value. The droplet deforms
significantly less than in the absence of surfactant, and is far more stable against break-
up. The mass balance reduces to regulating tangential fluxes of surfactant along the
interface so as to balance any local stretching, keeping r uniform. For stable drop
shapes (for which Vn is zero), this requires that vt=O, and the interface is stagnant.
Interestingly, the same physics that dictate that the interface remain stagnant if the
droplet has a steady shape require that the surface have tangential flow if the droplet is
deforming. The deformations for a droplet with x=0.996 initially are presented in
Figure 4; r, Vt and the Marangoni stress profile are shown in Figure 6.

I== r- 0.1 I
x-0.996.
6
~ \
5
\
4
\\
\
r
3

0.0 0.1 0.2 0.3 0.4 0.5


s
Figure 6a. Profiles for rat dilute (x=O.l) and high (x=0.996) surface concentration.
248

I== :~:~g 6 1
0 .6
.r-·-,

I ~
0 .5

I \
0.4

I \\
v,
0 .3

I '

\
0 .2
I
0 .1

0.0
~~·:::::
_//
_____ __ _,_ .. _. _____ __....._,. ............... .-......... .....__ .... _., ...... __ , ____ ~ ... . . _..
, ,.~

0.0 0.1 0.2 0.3 0 .4 0.5


s

Figure 6b. Profiles for v, for dilute (x=O.l) and high (x=0.996) surface concentration. Note the nearly
stagnant tips for x=O.l, and the near-stagnant interface for all s at x=0.996.

0.5

;;
0. 4
I: :::.·::. =~~I
-·-·
0.3
.
,./ /
dy
ds

!'
0.2 j
//
I

0.1

0.0

0.0 0.1 0. 2 0.3 0.4 0.5


s

Figure 6c. Marangoni stress profiles for dilute (x=O.l) and high (x=0.996) surface concentration.
249

Throughout this discussion. the surfactant has been assumed to be insoluble.


The Marangoni stress was the sole mechanism that the system had to prevent
surfactant from approaching its limiting area/molecule at the drop tips. Soluble
surfactants have two mechanisms with which they can enforce this limit. When swept
from the drop equator toward the drop tips, the surfactant desorbs and diffuses away
rather than accumulate and provoke a high stress. In the elevated surface coverage
limit, the surface concentration remains nearly uniform, but approaches its
equilibrium value as the rates of mass transfer increase. For slow mass transfer, the
insoluble results are recovered. In the rapid mass transfer limit, the deformation and
velocity profiles of a clean interface droplet are recovered. However, the deformations
for intermidiate mass transfer rates are not bounded by the insoluble and clean results
[19,24].

5. Remobilizing Surfactants: Controlling Stresses

It is often desirable to control the stresses exerted along fluid interfaces. Surfactants
are usually present as impurities or components of fluids in multiphase flows. The
removal of all surface active molecules is not a practical option. It is often thought
that Marangoni stresses become more pronounced as the concentration of surfactant
increases. Below, we will sketch out arguments and present experimental data that
show that this is not the case for a diffusion-controlled surfactant [4,25].

Suppose that a droplet is translating through a surfactant solution in which


the droplet is immiscible. Once the fluids and surfactant are chosen. the Biot number
Bi= a.a!U is fixed. If Bi>> 1, the surfactant distribution is entirely controlled by the
bulk diffusion rate, and hence by the dimensionless group A=Peh2/a2 • If A>> 1, the
surfactant will cause Marangoni stresses to develop. Consider this group in detail.
The Peclet number Pe is fixed for a given droplet and surfactant system, and is
typically quite large; swfactants typically have a diffusivity of 10-6 cm2/s; for a drop of
radius 1 em translating at 1 cm/s, Pe ~ 106 • Thus, for hla of roughly unity, A is large,
and significant Marangoni stresses develop. However, the adsorption depth h is a
strong function of concentration, and can be used to regulate the size of the ratio A.
Recall that h is defined:

h=reql C""= ~/c:x.rool(l+k); k= ~Coolc:x. (22)

Thus, if the concentration can be brought to sufficiently elevated values, h , and


therefore A can approach zero. Marangoni stresses should decrease, until they are
completely eliminated at sufficiently high concentrations. This reflects the fact that
once r approaches it saturation value, the interface becomes ineffective at depleting the
bulk. and diffusion timescales become rapid. This simple scenario has not been tested
experimentally, although we have identified surfactants whose adsorption depths can
be forced to extremely small values [26].
250
The complication in applying this reasoning to most surfactants is that they
cannot be brought to high bulk concentrations without forming micelles. Micelles are
aggregates in which surfactant monomers self-assemble to remove their hydrophobic
regions from the water in which they are dissolved. These aggregates form when the
bulk concentration of surfactant in monomeric form reaches the critical micelle
concentration (CMC). Concentrations of surfactant in excess of the CMC are
incorporated into micelles, and the monomer concentration remains constant at the
CMC.

While the monomeric adsorption depth h cannot be reduced to zero, the


micelles themselves allow for the control of the bulk concentration. They act as
sources or sinks of monomer, and, can, when present in sufficient quantity, maintain
the concentration of monomeric surfactant uniform and equal to the CMC. The sole
additional requirement is that the rate of micellar dissociation to form monomer must
be rapid compared to the surface convection rates.

Suppose that a droplet is settling in a surfactant solution. At dilute


concentrations, as discussed above, the diffusion timescales are sufficiently slow that
surfactant swept to the trailing pole collects there. The local sublayer concentration is
high at the trailing pole, low at the leading pole. Diffusion is too slow to remove the
surfactant accumulated at the back, or to supply the front. Marangoni stresses
develop. In this regime, adding more surfactant causes greater Marangoni stresses.
This is the situation depicted in Figure 1.

Now consider the same scenario just below the CMC. Surfactant will be
swept to the trailing pole, where, because of slow bulk diffusion, it will accumulate.
This will cause the local surfactant concentration to exceed the CMC. Micelles will
form in the region of the trailing pole, maintaining the concentration of monomer
constant at the CMC. It is monomeric surfactant that adsorbs on interfaces. The
interface in this micellar cap region will maintain local equilibrium with the monomer
in solution there. The surface tension will be uniform in this region, and therefore free
of Marangoni stresses. Outside of this micelle-containing region; however, there are
gradients in the surface concentration. These regions still have Marangoni stresses.
This is sketched in Figure 7.
251

'
.? "-

'
.?
.? ~
.,.,-
' .? / ~

' .?

'
.?

' '.,.,-
'
' '
'
Marangoni
stress
direction
of surface
concentration
gradient
'
' .? '
' '
'
.?

' u + + +
Figure 7. A drop translating in a surfactant solution just below the CMC. The appearnace of a stress-free cap in
the micellar region caused by rapid monomer mass transfer is depicted.

If the bulk concentration is elevated above the CMC, enough micelles can
develop to completely eliminate monomer concentration gradients. The surface
concentration is in equilibrium with a uniform sublayer concentration. Therefore, the
surface tension is uniform. No Marangoni stresses are exerted. Surfactant adsorbs and
desorbs freely all along the interface. In the sublayer, c. remains uniform, with excess
surfactant being incorporated into micelles, or depleted regions being rapidly supplied
from the micelles. This situation is shown schematically in Figure 8.
*
252

• ' ' ..... .


.r'
~ .~
'

*
.~~

·t-'
_.J ~ ' • .r' • "
~ .~
'
?

•L
·*
' *
.--·t-
~\,

''·
·*
~
*'
u +
Figure 8. Depiction of a drop with no Marangoni stresses because of fast mass transfer to and from the interface.

This is termed remobilization, in that an interface can be restored to free


motion or remobilized by the addition of a diffusion-controlled, micelle-forming
surfactant. Remobilization has been verified experimentally using a three phase flow
that is extremely sensitive to Marangoni stresses. In the experiments, a train of
alternating air and aqueous fluid segments (or slugs) were drawn into a teflon
capillary. The inner surface of the capillary tube was coated with a fluorocarbon oil
film. The aqueous segments were made of surfactant solution. The flow geometry is
depicted in Figure 9; the train was moved at constant velocity, and the pressure drop
required to drive the flow was measured as a function of surfactant concentration.
253
Marangoni stresses exerted in this flow increased the pressure drop required
to drive it at constant U. The flow field in Figure 9 is drawn in the train-fixed
reference frame. Stagnation points at the poles, and rings along the interface are
indicated by bold dots. The tube wall moves to the left with velocity U. The
streamlines along the aqueous slug diverge from the leading ring, and converge at the
trailing ring. Surfactant adsorbed along the aqueous-oil interface could then
accumulate at the stagnation points, creating a Marangoni stress that resists the flow.
The stresses so generated are very pronounced. The oil film is thin (~ 10 microns
thick); any shearing in this region results in large stresses.

fluorocarbon oil fil~


.....___ u

air slug
..
aqueous slug

Figure 9. The three phase periodic flow used to demonstrate remobilization is shown.

The nonionic surfactant Triton-X 100 was used in the experiments. The
results are graphed in Figure 10. The pressure drop per unit cell (made dimensionless
with surface tension scalings) is graphed as a function of the bulk concentration,
expressed as percentage of the CMC. The pressure drop initially increased with added
surfactant, peaking roughly an order of magnitude below CMC. As the bulk
concentration of surfactant is increased further, the pressure drop declines. An order
of magnitude above the CMC, the pressure drop has reduced to be equal to that
required to drive the clean flow. This demonstrates that Marangoni stresses have been
eliminated by the addition of a diffusion controlled, micelle-forming surfactant in
sufficient concentration. These results are discussed in detail in [4] and [25].
254

TRITON-X 100 SERIES

• ID•.097cm - • lD•.UScm -----

3.00
0u
...u 2.40
c.
c.
...0
'0
...
u 1.80
=
U1
en
...c.
0

1.20
-
U1
en
0
;::
0
·;;
c0 0.60
a
~

0.00
to·, 10'' 10° 10'

percent CMC TRITON-X 100

Figure 10. The dimensionless pressure drop per unit cell of the three phase flow is shown as a function of
Triton 100 concentration in the aqueous phase. (Reprinted with permission from the American Physical Society,
K.J. Stebe, S.-Y. Lin & C.Maldarelli, "Remobilizing sutfactant retarded fluid particle intetfaces,. I. Stress-free
conditions at the intetfaces of micellar solutions of sutfactants with fast sorption kinetics", Physics of Fluids A 3
(1) January 1991, p3-20)

6. Conclusions

The material parameters that allow us to predict or control the Marangoni effects
caused by a given surfactant are those that determine the surface equation of state and
the mass transfer dynamics of a surfactant. The equation of state relates the surface
tension to its local surface concentration, and the mass transfer dynamics determine the
surfactant distribution.

Surfactants are often assumed not only to alter the surface tension of an
interface, but also to alter the rheological response in the interfacial region. Surface
viscosities intrinsic to the interface are defined [27-29]. The increased resistance to
interfacial flow caused by surfactants is often attributed to this mechanism. This
255
cannot be the mechanism for the experiments presented above, however. Recall that
r eq increases monotonically with c_ until the CMC is reached, thereafter staying
constant. While there is little data available for the surface viscosity as a function of
r, one would expect the viscous dissipation in the interface to increase with r. It is
difficult to imagine why these viscosities would drop to zero above the CMC.
However, the elimination of monomeric diffusion as a controlling timescale can be
straightforwardly explained. Therefore, we favor the mass transfer and surface
tension based formalism used here and have neglected intrinsic surface rheology.
Surface rheological effects may well be important for polymeric surfactants or proteins
at interfaces, which are not addressed here.

The term surface viscosity also appears in the literature for describing the
manner in which surfactants alter the damping of capillary and longitudinal waves.
However, in the literature for these experiments, it is understood that surface tension
variations (Marangoni stresses) are generated which increase the damping rate of the
surfactants. The in-phase part is termed a surface elasticity, the out-of-phase part a
surface viscosity [8,9].

Marangoni stresses are highly coupled with the mass transfer rates of a
surfactant. For a diffusion controlled surfactant, Marangoni stresses do not increase
monotonically with the amount of surfactant added, but can in fact decrease to zero at
elevated concentrations. The tangential surface velocity can range from complete
stagnation to free convection.

Insoluble surfactants on non-deforming axisymmetric interfaces can cause the


interface to become stagnant. This is not the effect of a strong Marangoni stress on a
deforming interface. For a deforming interface, strong Marangoni stresses force the
surface concentration to remain uniform. This is discussed above for a drop in a
Taylor extensional flow. If the interface stretches locally, it tends to become diluted
locally. The Marangoni stress regulates the tangential flow which sweeps surfactant to
the diluted area, keeping r uniform. Therefore, for a deforming interface, the high
Marangoni stress limit must have a tangential flow.

Finally, a uniform distribution of adsorbed surfactant does not imply that the
interface is stress-free. For the insoluble limit discussed above, the Marangoni stresses
keep r uniform in order to stop the surface convective flux from allowing r to reach
r aa· In this case, r is uniform, but the interface is highly stressed. Compare this with
the case of remobilization, in which r is uniform and in equilibrium with the
surrounding fluid. Rather than ever being packed above its equilibrium concentration,
it desorbs. Depleted regions are rapidly supplied by adsorption. Surfactants then
convect freely along the interface, and no retarding Marangoni stress develops.
256

References

[1] Levich, V.G., Physicochemical Hydrodynamics, Prentice Hall, Englewood Cliffs, N.J. , 1962
[2] Stone, H.A., "A Simple Derivation of the time-dependent convective diffusion equation for
transport along a deforming interface", Phys. Fluids A 2, 111-112, (1990)
[3] Wong, H. , Rwnshitzski, D. and Malderelli, C. ,"On the surfactant mass balance at a
deforming fluid interface", Phys. Fluids 8, 3203, (1996)
[4] Stebe, K.J. , Lin, S.Y. and Maldarelli, C. , ''Remobilizing surfactant retarded fluid particle
interfaces. I. Stress-free conditions at interfaces of micellar solutions of surfactants with fast
sorption kinetics", Phys Fluids A, 3 (1), 3-20, (1991)
[5] Chang, Chien-Hsiang and Franses, Elias, "Adsorption dynamics of surfactants at the
air/water interface: a critical review of mathematical models, data and mechanisms", Colloids
and Surfaces A, 100 1-45 (1990)
[6] Lin, S.Y., McKeigue, K., and Maldarelli, C., AIChE J., "Diffusion-Controlled Surfactant
Adsorption Studied by Pendant Drop Digitization", 36, 1785-1795, (1990)
[7] Johnson., D. and Stebe, K. "Experimental Confumation of the Oscillating Bubble Technique
with Comparison to the Pendant Bubble Method: The Adsorption Dynamics of 1-decanol", J.
Colloid Int. Sci., 182, 526-538, (1996)
[8] reviewed in van den Temple and Lucassen-Reynders, "Relaxation Processes at Fluid
Interfaces", Advances in Colloid Interface Sci. 18, 281-30 I, ( 1983)
[9] reviewed by Langevin, D. see chapter in this text
[10] Macleod, C.A. and Radke, C.J., "Surfactant Exchange Kinetics at the Air/Water Interface
from the Dynamic Tension of Growing Liquid Drops", J. Colloid Int. Sci. 166, 73-88, (1994)
[11] Davis, R. and Acrivos, A. , "The influence of surfactants on the creeping motion of
bubbles", Chern Eng. Sci. 21, 681-685, (1966)
[12] Sadhal, S.S. and Johnson, R.E., "Stokes Flow Past Bubbles and Drops Partially Coated
With Thin Films: Exact Solution", J. Fluid Mech., 126, 237- 250, (1983)
[13] He, Z. Dagan., Z. and Maldarelli, C. "The Size of Stagnant Caps of Bulk Soluble
Surfactant on the Interfaces of Translating Fluid Droplets", J. Colloid Int. Sci. 146, 442-451,
(1991)
[14] Holbrook, J.A and Levan, M.D., "Retardation of Droplet Motion by Surfactant. Part I.
Theoretical Development and Asymptotic Solutions", Chern Eng. Cornrn. 20, 191-207, ( 1983)
[15] Holbrook, J.A and Levan, M.D., "Retardation of Droplet Motion by Surfactant. Part II.
Numerical Solutions for Exterior Diffusion., Surface Diffusion and Adsorption Kinetics", Chern
Eng. Cornrn. 20,273-290, (1983)
[16] Chen, J. and Stebe, K.J. "Marangoni Retardation of the Terminal Velocity of a Settling
Droplet: The Role of Surfactant Physico-chemistry", J. Colloid Int. Sci. 178, 144-155, (1996)
[17] Stone, H.A. and Leal, L.G, " The Effects of Surfactants on Drop Deformation and
Breakup", J. Fluid Mech. 220, 161-186, (1990)
[18] Milliken, W.J., Stone, H.A., and Leal, L.G.,"The Effects of Surfactant on the Transient
Motion of Newtonian Drops",Phys. Fluids A 5, 69-79, ( 1993)
[19] Milliken W.J. and Leal, L.G., "The Influence of surfactant on the deformation and breakup
of a viscous drop: the effect of surfactant solubility", J. Colloid Int. Sci, 166, 275-285, (1994)
[20] Pawar, Y.P. and Stebe, K.J. "Marangoni effects on drop deformation in an extensional flow:
the role of surfactant physico-chemistry. I. Insoluble Surfactants", Phys. Fluids 8 (7) ,1738-1751,
(1996)
257

[21] Eggleton, C.D., Pawar, Y.P. and Stebe, K.J., "Insoluble Surfactants on a Drop in an
Extensional Flow: A Generalization of the Stagnated Interface Surface Limit to Deforming
Interfaces", in review, J. Fluid Mech.
[22] Barthes-Biesel , D. and Acrivos, A.," Deformation and Burst of a Liquid Droplet Freely
Suspended in a Linear Shear field", J. Fluid Mech. 61, 1-21 (1973)
[23] Rallison, J.M. and Acrivos, A., "A Numerical Study of the Deformation and Burst of a
Viscous Drop in an Extensional Flow'', J.Fluid Mech. 89, 191-200, (1978)
[24] Eggleton, C. and Stebe, K.J. " Surfactants and stresses on deforming interfaces: A soluble
surfactant on a drop in an extensional flow'' in preparation for J. Colloid Int. Sci.
[25] Stebe, K.J. and Maldarelli, C. " Remobilizing Surfactant Retarded Fluid Particle Interfaces
II. Controlling the Surface Mobility at Interfaces of Solutions Containing Surface Active
Components" J. Colloid Int. Sci. 163, 177-189 (1994)
[26] Ferri, J.K. and Stebe, K.J. "The Rapid Reduction of Surface Tension: a Structure-Property
Study of Acetylenic Diol Surfactants", in preparation for J. Colloid Int. Sci.
[27] Scriven, L.E., "Dynamics of a fluid interface: Equation of motion for Newtonian surface
fluids", Chern. Eng. Sci. 12, 98-108, (1960)
[28] Slattery, J.C. Interfacial Transport Phenomena Springer-Verlag, New York, 1990
[29] Edwards, D., Wasan, D. and Brenner, H. Interfacial Transport Processes and Rheology ,
Butterworth Hienemann, Boston, MA, 1991
258
FOAM MICROMECHANICS
Structure and Rheology of Foams, Emulsions, and Cellular Solids

ANDREW M. KRAYNIK
Sandia National Laboratories
Department 9112 MS0834
Albuquerque, New Mexico 87185-0834 USA
MICHAEL K. NEILSEN
Sandia National Laboratories
Department 9117 MS0443
Albuquerque, New Mexico 87185-0443 USA
DOUGLAS A. REINELT
Southern Methodist University
Department of Mathematics
Dallas, Texas 75275-0156 USA
AND
WILLIAM E. WARREN
Texas A&M University
Department of Civil Engineering
College Station, Texas 77843-3136 USA

1. Introduction

Foam evokes many different images: waves breaking at the seashore, the
head on a pint of Guinness, an elegant dessert, shaving, the comfortable
cushion on which you may be seated ... From the mundane to the high tech,
foams, emulsions, and cellular solids encompass a broad range of materials
and applications. Soap suds, mayonnaise, and foamed polymers provide
practical motivation and only hint at the variety of materials at issue.
Typical of multi phase materials, the rheology or mechanical behavior of
foams is more complicated than that of the constituent phases alone, which
may be gas, liquid, or solid. For example, a soap froth exhibits a static
shear modulus-a hallmark of an elastic solid-even though it is composed
primarily of two Newtonian fluids (water and air), which have no shear
259
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 259-286.
© 1999 Kluwer Academic Publishers.
260

modulus. This apparent paradox is easily resolved. Soap froth contains a


small amount of surfactant that stabilizes the delicate network of thin liq-
uid films against rupture. The soap-film network deforms in response to a
macroscopic strain; this increases interfacial area and the corresponding sur-
face energy, and provides the strain energy of classical elasticity theory [1].
This physical mechanism is easily imagined but very challenging to quantify
for a realistic three-dimensional soap froth in view of its complex geome-
try. Foam micromechanics addresses the connection between constituent
properties, cell-level structure, and macroscopic mechanical behavior.
This article is a survey of micromechanics applied to gas-liquid foams,
liquid-liquid emulsions, and cellular solids. We will focus on static response
where the foam deformation is very slow and rate-dependent phenomena
such as viscous flow can be neglected. This includes nonlinear elasticity
when deformations are large but reversible. We will also discuss elastic-
plastic behavior, which involves yield phenomena.
Foam structures based on polyhedra packed to fill space provide a unify-
ing geometrical theme. Because a two-dimensional situation is always easier
to visualize and usually easier to analyze, the roots of foam micro mechanics
lie in the plane packed with polygons. There are striking similarities as well
as obvious differences between 2D and 3D.
Many of the issues investigated here have been reviewed by Weaire &
Fortes [2], who also discussed stress and strain in liquid and solid foams.
Kraynik [3] has surveyed fluid-fluid systems. Useful reviews on the mechan-
ics of cellular solids include Gibson & Ashby [4] and Kraynik & Warren [5].
The rigorous theoretical foundations and computational details associated
with foam micromechanics are beyond the scope of this article. The inter-
ested reader is refered to the literature to pursue more technical matters.

2. Theoretical Approach

We only consider spatially periodic models, which are based on a repre-


sentative volume (unit cell) of foam, countless identical copies of which fit
together to fill space. The obvious advantages of spatial periodicity include
rigorous mathematical formulation and tractability. The unit cell in the
simplest models only contains one foam cell. These perfectly ordered mod-
els composed of identical cells are most likely to admit analytic solutions,
which reveal important physics and can be used to validate computer codes.
The unit cell in more complete models will contain many foam cells of dif-
ferent size and shape; a large number of cells N is required to represent the
disorder and polydispersity found in real foams.
261
Spatial periodicity refers to situations where any position-dependent
property 'P(x) = 'P(x + Lk ikLk), where x is position, the ik are arbitrary
integers, and the Lk are lattice vectors. A homogeneous deformation is
imposed on the entire foam by applying displacements through the lattice
vectors. The local forces or stress u'(x) within the foam are determined by
solving balance equations that reflect the appropriate physics. The effective
macroscopic stress tensor u is given by the volume average of local stress
u' as
u = ~ Ju'dv (1)

where V is the volume of the unit cell. Components of u can also be ob-
tained by using energy methods, which involve derivatives of energy with
respect to strain; however, calculations based on (1) frequently offer compu-
tational advantages in addition to more information. Force methods, which
are based on (1), are especially convenient for evaluating effective stress
when the foam is not undergoing macroscopic deformation.
The specific problem to be solved can cross the boundaries of tradi-
tional fluid mechanics and solid mechanics, reminding us that foam science
and technology, in general, are highly interdisciplinary pursuits involving
scientists and engineers from many fields.

3. Two Dimensions
3.1. FOAM STRUCTURE

The structure of gas-liquid foams and liquid-liquid emulsions under static


conditions is governed by energy minimization. We neglect gravity so liquid
drainage and the resulting spatial inhomogeniety of the foam structure
will not be considered. Plateau's laws [6] determine film-network geometry
in the limiting case of a dry soap froth, which contains negligible liquid.
Polyhedral cells in 3D are separated by surfaces (faces) with constant mean
curvature (CMC); three curved faces meet at equal dihedral angles of 120°
along each edge; and four edges meet at each vertex at equal tetrahedral
angles of cos- 1 ( -1/3)=109.47°.
In 2D, Plateau's laws require polygonal cells separated by circular arcs
that meet at equal angles of 120°. The 2D structures, shown in figure 1, can
be viewed as cross sections of 3D honeycombs. All lines, which represent
liquid films with zero thickness, have equal length in a perfectly ordered
foam that is undeformed. Cell 'volumes' and edge lengths vary in the poly-
disperse hexagonal structure, but all edges are straight indicating that each
cell has the same internal pressure. All cells do not have six sides in the
disordered structure, which has curved edges and different pressures.
262

Figure 1. Perfectly ordered honeycomb, polydisperse hexagons, Voronoi polygons, which


do not satisfy Plateau's laws, and disordered foam produced by relaxing Voronoi polygons.

3.2. LIQUID FOAMS


Princen [7] pioneered the microrheology of 2D foams. Khan & Armstrong [8]
derived the shear stress u for a perfectly ordered foam in simple shear

G'Y 31/4 T T
= --~r.:=-1=+='Y=;;=2/:;:::;:4 ' G = lil - 1 = 0.931 - 1 {2)
(J'
v2 V2 V2
where G is shear modulus, 'Y is shear strain, T is interfacial tension, and V
is the cell 'volume.' This result deserves several comments. The shear stress
does not depend on the foam orientation so the nonlinear elastic response is
isotropic; hexagonal symmetry alone only guarantees isotropic linear elastic
response. The shear modulus, as well as the stress and energy density, scale
1 1
as T fV75 where D is the dimension and V75 is a characteristic cell size. A
soap froth with smaller cells is stiffer.
Equation (2) applies up to strains where some film length goes to zero,
which creates an unstable situation with film connectivities of four. This
triggers topological transitions called T1s that cause discontinuities in the
energy and stress-strain curve. The first T1 determines the elastic limit;
the corresponding stress and strain depend on orientation of the foam. For
certain orientations, the foam structure and stress are periodic functions
of strain for simple shear and planar extensional deformations [9, 10]. The
average stress (u) for each strain-periodic condition corresponds to the
dynamic yield stress. The stress-strain curves for the perfectly ordered foam
exhibit large fluctuations and (u) is very sensitive to orientation.
The energy {edge length) density of a polydisperse hexagonal foam with
average cell volume V is identical to that of a perfectly ordered foam; con-
sequently, the elastic behavior is the same [11]. The strain energy function
W that describes this nonlinear elastic response is given by
(3)

where the Ai are principal stretch ratios, Pb is the bubble pressure in the
undeformed foam, and ¢ is the volume fraction of liquid. Equation {3) is
263

valid as long as the foam structure remains polydisperse hexagonal. For a


dry foam (¢=0) this means that all film lengths remain finite so there are
no topological changes that are precipitated by unstable edge connectivities
of four. A wet foam (¢>0) remains polydisperse hexagonal as long as no
film length in between Plateau borders goes to zero.
For isochoric deformations such as simple shear, where the foam vol-
ume does not change, the product >.1>.2 is unity; consequently, the shear
modulus G does not depend on liquid content ¢. This lack of dependence
on ¢ can be explained by considering the threefold Plateau borders where
liquid collects at the cell corners. These Plateau borders translate and ro-
tate as the foam deforms but they do not change shape, therefore, they do
not contribute to the strain energy since their interfacial 'area' does not
change with deformation. It is ironic that nonlinear response of the poly-
disperse hexagonal foam is isotropic even though polydispersity breaks the
hexagonal symmetry that guarantees isotropic linear response.
The complete lack of dependence between G and ¢ also holds for dis-
ordered 2D foams as long as all Plateau borders form threefold junctions
between films, which only occurs near the dry limit. The decoration theo-
rem of Bolton & Weaire [12] sheds additional light on this situation. The
equilibrium structure of a wet foam can be regarded as a dry foam whose
vertices have been decorated with threefold Plateau borders. The dry foam
can be constructed by continuing the circular arcs, which represent films,
into the Plateau borders where they meet at 120°. Plateau borders increase
in size as ¢ increases and eventually merge with neighbors to form higher
connectivity junctions. The continued arcs in such structures do not meet
at 120°; the connectivity of film junctions and the stress both depend on ¢.
Simulations indicate that G decreases with increasing liquid content when
the decoration theorem does not apply [13, 14]. Eventually, bubbles lose
contact with neighbors and become circular beyond the percolation limit
where G is zero.

3.3. CELLULAR SOLIDS

We now consider solid foams where the cell structure is patterned after
liquid foams. Just imagine that liquid films are transformed into solid struts
by some chemical or thermal process. The in-plane elastic constants of a
regular honeycomb are given by

K - 1 G - 1 (4)
h- 4J3 M' h- 2J3 (M +N)
where Kh is the bulk modulus and Gh is the shear modulus [15]. The
parameters M and N are compliances that relate strut-level forces and
264

displacements due to stretching and bending. The undeformed struts have


mirror symmetry about their axis and midpoint. The compliances can be
evaluated for various strut shapes but we only consider struts with uniform
thickness. The volume fraction of solid ¢ is often referred to as the relative
foam density. For low-density foams where ¢«1, M- 1 = v'3 E¢ ~ N- 1 =
3v'3/ 4 E¢3, where E is Young's modulus of the solid material in the struts.
The inequality emphasizes that long slender struts are much easier to bend
than to stretch. The elastic constants evaluate to

Gh = ~E ¢3 . (5)
Both moduli scale with E. In sharp contrast with liquid foams, the stiffness
of cellular solids does not depend on cell size; the moduli are identical for
self-similar structures. The bulk modulus is order ¢ because hydrostatic
loading only induces axial strut displacements. The shear modulus is order
¢ 3 because strut bending is the dominant mechanism when ¢«1.
The strut lengths and the corresponding strut-level compliances vary in
a polydisperse hexagonal cellular solid, so analytic solutions are not feasible.
Force and moment balances and compatibility conditions provide a system
of llN algebraic equations that are solved numerically to evaluate effec-
tive elastic response. We only consider struts with uniform thickness but
distinguish two cases: constant strut thickness (CT) throughout the foam,
and constant strut mass (CM), which means that strut lengthxthickness
does not vary. In general, the elastic behavior is not isotropic, however ori-
entation dependence is rather mild and decreases with N, as expected. All
results are averaged over foam orientation.
Polydispersity does not change the bulk modulus Kcr when all of the
struts have the same thickness because joint displacements are affine, and
each strut carries the same axial load as the perfectly ordered foam. The
results for the constant-strut-mass case are represented very accurately
by KcM = Kh/(1 + 11-2), where 11-2 = N- 1 l::[(L/L) - 1]2 is the second
moment of the strut-length distribution. The joint displacements are not
affine however bending is negligible. The axial load in each strut is reduced
by a factor of (1+~t2); this can be calculated by assuming that all of the
struts, which have different length but the same mass, are connected in
series.
The shear modulus for constant-strut-thickness is well represented by
Gcr = Gh (1 + 11-2), which indicates that polydispersity increases stiffness.
In constrast, GeM decreases very slightly with /L2· We can not offer a simple
explanation for the effect of polydispersity on the shear modulus.
The moduli of solid foams do not scale with cell size; however, they
do depend on the distribution of cell size as well as on the distribution
of material within the network of struts. With this in mind, we note that
265

Kelvin Weaire-Phelan (A 15) Friauf-Laves (C15)

Bergman (T) Monodisperse Polydisperse

Figure 2. Kelvin cell, three TCP structures, and two random foams.

experimental data, which appear to support a correlation between foam


modulus and cell size, could actually be explained by systematic variations
in cell-size distribution or cell-level material distribution. Viscous and sur-
face tension effects interact when bubbles grow and deform during foam
formation. These effects exhibit different scaling with bubble size that can
cause systematic variations in foam structure with cell size.

4. Three Dimensions
4.1. LIQUID FOAMS IN THE DRY LIMIT

4.1.1. Cell Structure


A dry soap froth based on the Kelvin cell [16] is the only perfectly ordered
structure known that is consistent with Plateau's laws. This establishes
the Kelvin foam as a natural starting point for developing micromechanics
models in 3D, just like the honeycomb in 2D. The Kelvin cell and other
foam geometries are shown in figure 2.
The discovery of the elegant Weaire-Phelan (WP) structure [17] was
a watershed event in foam research. Its impact goes beyond providing a
counterexample to Kelvin's candidate for the most efficient division of space
266
into equal-volume cells. Weaire and Phelan brought Ken Brakke's Surface
Evolver computer program [18] to the attention of the foam community.
The Surface Evolver is publically available from the Geometry Center at
the University of Minnesota (by anonymous ftp from geom.umn.edu); it
has been used to study a wide range of problems involving surfaces shaped
by surface tension and other energies. The Surface Evolver was used to
compute all of the liquid foam geometries discussed in this section.
The Weaire-Phelan phenomenon has also stimulated a search for mono-
disperse foams with even lower energy. Rivier [19] has pointed out that
WP is one of two dozen or so structures known as tetrahedrally close-
packed (TCP) to crystallographers and Frank-Kasper to metallurgists and
materials scientists. TCP foams are restricted to four polyhedra having
12, 14, 15, or 16 faces. Each n-hedron has 12 pentagonal faces and n-12
hexagonal faces. The TCP structures shown in figure 2 include Weaire-
Phelan (A15), Friauf-Laves (C15), and Bergman (T), all of which have
cubic symmetry. This high symmetry is convenient because one only has
to probe in two principal directions to determine orientation dependence of
elastic properties [20].
Voronoi polyhedra produced from random points in a unit cell have the
proper connectivity of a soap froth but they do not have minimal surface
area. We have used the Surface Evolver to relax Voronoi partitions of space.
Consider the analogous 2D problem (figure 1). Each Voronoi cell associated
with a random seed consists of all points closer to that seed than to any
other. The edges of a Voronoi polygon lie on perpendicular bisectors be-
tween neighboring seeds; but the straight edges do not meet at 120°. During
the relaxation process, vertex positions and edge curvatures are adjusted
until pieces of circles meet at equal angles and satisfy constraints on cell
volumes. Relaxation also involves topological changes (T1s) that are trig-
gered by edges going to zero length. Many cell neighbors and edges switch
in going from Voronoi cells to minimal structures.
In 3D, Voronoi polyhedra have flat faces and straight edges. No Voronoi
partition can satisfy Plateau's law requiring each edge to meet at equal
angles because no polygon with straight edges has all vertex angles equal
to the tetrahedral angle. As the surface evolves during relaxation, cell edges
shrink and trigger topological transitions.
Voronoi polyhedra were computed with a program written by John Sul-
livan. The initial Voronoi seeds were generated by two methods: random
sequential adsorption (RSA) and random close packing (RCP) of hard
spheres. In RSA, a randomly generated seed is accepted in the unit cell
only if the distance to existing seeds is greater than some diameter, which
is chosen to be as large as possible to pack N spheres in a cube with
spatially periodic constraints. Relatively loose packings of monodisperse
267

0.5
--Matzke
N • 500 cella
••• ••• Monod lapera e RSA
- - - Monodlaperae RCP
0.4 · ·· ·· ··Lognormal cr = 0 .• 74

0.3

u..
0.2

0.1

10 12 14 16 18 20
Faces/Cell

Figure 3. Distribution of polyhedra with n faces in random foams.

o .sr----r--.....---r---;:::::::::=:c:=:::::~
--Matzke
N • 500 cella
•• •••· Monodlaperae RSA
- - - Monodlaperse RCP
• ·• ·• ·• Lognormal cr ~ 0.• 74
0.6

u.. 0.4

0.2

0.0 -+--........~-+-----+---M--+-------1
2 4 6 8 10
Edges/Face

Figure 4. Distribution of faces with p sides in random foams.

spheres with volume fractions up to ¢>8 =0.35 were achieved with N =512.
Molecular dynamics simulations, using software provided by Frank van
Swol, produced much denser configurations with ¢ 8 =0.64. The standard
deviations on Voronoi polyhedra volumes were about O"v=0.13 for RSA and
O"v=0.04 for RCP.
Typical distributions of polyhedra with n faces and polygons with p
sides, are shown in figures 3 and 4. These results are compared with ex-
perimental observations by Matzke [21] on six hundred bubbles in foams
268

that were believed to be monodisperse. The computed statistics for relaxed


monodisperse foams are in substantial agreement with the experimental
data. Matzke found averages of 13.70 faces/cell and 5.124 edges/face. We
performed eight relaxation simulations on Voronoi partitions based on RCP
with N=500, 512 and found 13.85 faces/cell and 5.133 edges/face.
Triangular faces are common among Voronoi cells but very rare in
monodisperse soap froths; Matzke did not find any three-sided faces. Nor
did he find a single Kelvin cell in his experiments, although others have.
We found a few Kelvin cells among our relaxed structures but they were
very rare, which is not surprising since pentagonal faces are so common.
The Voronoi foams based on RSA have broader distributions of polyhedra
and polygons than those based on RCP, which are much closer to the final
relaxed systems.
Relaxed foams with lognormal distribution of cell volumes were simu-
lated to show the effects of polydispersity; in the most polydisperse, Uv =
0.474, individual cell volumes ranged from 0.224 to 3.65 with an average of
one. The distribution of n-faced polyhedra, shown in figure 3, is obviously
broader for the polydisperse system, which even contained cells with 20
to 23 faces that were absent in the monodisperse foams. Larger polyhedra
tend to have more faces and smaller polyhedra tend to have fewer faces.
By contrast, the distribution of p-sided faces, shown in figure 4, is surpris-
ingly similar for the polydisperse and monodisperse structures. Clearly, this
would not be expected in a polydisperse foam that contains a large number
of extremely small cells; tetrahedra would be common.
Two geometric properties, surface area and edge length, are graphed
in figure 5 for several types of relaxed foams. Both quantities fall in a
relatively narrow range for monodisperse systems. The polydisperse foams
have significantly lower area and edge length than the others. Different size
bubbles pack more efficiently, just like different size spheres.

4.1.2. Shear Modulus


The shear moduli of various monodisperse foams are given in Table 1.
All of the structures except the random foams have cubic symmetry, so
orientation dependence of the shear modulus is evaluated by applying two
specific deformations [22], which give two distinct shear moduli

(6)

where the Cij are elastic constants used by Love [1] and Nye [20]. An effective
isotropic shear modulus G is calculated by averaging G over foam orien-
tation. We use the Voigt average, which is performed at constant strain.
The Kelvin foam is significantly more anisotropic than the TOP foams.
269

5.45 -r---T--.......T-"--r.....--,r-...--r-.....,.......-T""""....,

Williams 0

~ 5.41
"''> • 0

-
.o
..r:::
Z Frlaui-Laves (C15)


Cl 5.37
c::
Q)
...J Kelvin
Q)
.~
Cl Wealre-Phelan
~ 5.33 P64


5.29 +-.....--1--+.....--1--+ ..........,--+..........,---t
5.23 5.25 5.275.29 5.31 5.33 5.35 5.37 5.39
Energy T/V 113

Figure 5. Cell edge length graphed against interfacial energy for dry foams with minimal
surface area. The label 'R' refers to random, monodisperse structures and 'P' refers to
random, polydisperse structures.

The T foam, which has many faces of different orientation, is essentially


isotropic. As measured by G, our results for monodisperse structures fall in
a relatively narrow range, but the TCP foams are perceptibly stiffer than
the Kelvin and random foams. The influence of polydispersity on the shear
modulus of disordered foams is currently under study.

TABLE 1. Shear moduli of dry foams

Foam structure G1 G2 G
Kelvin 0.5706 0.9646 0.8070
Weaire-Phelan (A15) 0.8902 0.8538 0.8684
Friauf-Laves (C15) 0.8448 0.8860 0.8695
Bergman (T) 0.858 0.856 0.857
Random (monodisperse) 0.78 ± 0.08

The bidisperse Wearie-Phelan foams, shown in figure 6, represent a sim-


ple model system that is polydisperse [23]. The unit cell contains six 14-
hedra with volume V14 and two pentagonal dodecahedra with volume ,BV14 ;
these particular volume constraints preserve cubic symmetry. This system
exhibits counter-intuitive results for bubble pressure: the pressure inside
the dodecahedra is always greater than the pressure inside the 14-hedra,
270

Figure 6. Bidisperse Wearie-Phelan foams with f3 = 0.04, 1.0, 2.4.

even when the dodecahedra are larger. The pressure difference P12-P14 de-
creases but remains positive as {3 increases. In contrast with isolated soap
bubbles, the pressure inside foam cells depends on their size and topol-
ogy. The interfacial energy density increases with {3; this is consistent with
diffusion-driven coarsening because the total energy of the foam decreases
as gas diffuses from the higher pressure dodecahedra into the tetrakaidec-
ahedra. The total cell edge length per unit volume also increases with {3.
The shear moduli of bidisperse WP foams were evaluated over the range
0.04<{3<2.4. The effective isotropic shear modulus G varies less than 0.4%
from the monodisperse case over the entire range of {3, which indicates that
the shear modulus is relatively insensitive to bidispersity.

4.1.3. Large Extensional Deformation


We now focus on large deformations of a Kelvin foam. Consider uniaxial
extension in a (100) direction with no volume change [24, 25]. This is the
most symmetric distortion of a perfectly ordered foam. The evolution of
cell geometry and tensile stress with Hencky strain E are shown in figure 7.
The area of the 'square' face that is being pulled decreases with increasing
strain but remains finite in the limit t:-tfTI =0.254. Beyond ETI, no stable
solution satisfies Plateau's laws and maintains contact between the original
cell neighbors. This situation triggers a topological transition (T1), which
marks the limit of reversible elastic behavior and the onset of plasticity.
The stress-strain curve exhibits a maximum in the tensile stress and
a turning point. The unstable solutions on the curve below the turning
point have the same topology but higher surface area than their stable
counterparts. The unstable solution at t:=!log 2=0.231, where the tensile
stress is zero, corresponds to a rhombic dodecahedron on a face-centered-
cubic (fcc) lattice, which will be discussed later.
In view of the cost of computing minimal surfaces, it is useful to consider
solutions based on oversimplified foam geometry. In all three approxima-
271

1 . 5o-y-;:::::::::~'::!:~~~:!::::'::::::::~~...........,.........,....... ..,....,l
••·••··••·· Affine Deformat ion
---Constant Angles • Force Method
)( 1.25 ··· · · · Constant Angles • Energy Met hod
)(

b - - Minimal Surfacea
I
N
- - Minimal Flat Surfaces
N
1.00
b

-..
C/1
~ 0.75
( /)

~ ........... ......................
...................... ........
0.50
en
c:
{:!. ..................
0.25

0.00 ~....._.___._+ .......--'-+.-..11......_._+.._.............+ ..................-f


0.0 0.1 0 .2 0.3 0 .4
Hencky Strain e

Figure 7. Evolution of cell geometry and tensile stress for uniaxial extension in a
(100) direction . The horizontal sequence of minimal structures can be compared with
approximate geometries with fiat faces. The highly stretched cell corresponds to affine
deformation .

tions to be considered, the undeformed cell is a regular tetrakaidecahedron;


the faces remain flat as the foam deforms but the dihedral angles are dif-
ferent in each case. Under affine deformation, all points in the foam move
exactly with the macroscopic strain. The pulled square continues to shrink
for all values of strain (figure 7) but there is no criterion for topological tran-
sition so there is no elastic limit. The affine assumption gives the largest
tensile stress. Next, we assume constant dihedral angles that do not vary
with strain. The pulled square shrinks to zero area when E= ~log 2=0.462
causing edge connectivity of eight, which violates Plateau's laws and trig-
gers a Tl. There are two different stress-strain curves in figure 7, which
correspond to different methods for calculating the stress. The force and
energy methods give different results in this situation where film-level forces
are not balanced and energy is not minimized. In the third approximation,
dihedral angles are determined by minimizing the total surface area of all
the flat faces . The evolution of foam structure and stress are qualitatively
272

0.45

0.30

~>-······ ~
(/)
0.15

-...
(/)
CD
... /'·····-····
..:.:;.:·········· ..
--- ........,, , 7····:.::.:-
( /}
0.00 ,,
,'
~ ~~

..........
', .....
', '
N, ' '
-0.15
.....
',
-0.30 -+-'................+-o-............o...+................-+...........+ ................+-'-............o...+....................
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Figure 8. Evolution of cell geometry, shear stress, and normal stress differences for
simple shearing flow. This cell orientation has the smallest strain period "fp=(3/2) 1 12 •
The dotted curve refers to the minimal planar approximation.

similar to the complete solution; the stress exhibits a local maximum and a
turning point but is lower. Results for this case indicate that approximate
solutions do not necessarily provide bounds for exact solutions.

4.1.4. Simple Shearing Flow


Under simple shearing flow, the structure and stress of a Kelvin foam are
piecewise continuous functions of shear strain 1 [26]. Each discontinuity
corresponds to a topological change that is preceded by shrinking faces.
Each Tl reduces surface energy, results in cell-neighbor switching, and pro-
vides a cell-level mechanism for plastic yield behavior during foam flow.
In figure 8, the foam is oriented so that structure and stress are periodic
functions of strain.
There are two Tls per cycle; each is triggered by opposite edges of
shrinking quadrilateral faces going to zero length, which leads to unsta-
ble connectivity. This standard transition is very different from the point
transition in the extensional deformation, where all edges on a face shrink
together. Symmetry imposes strong restrictions on the type and outcome
273
0 .7

0.6

~
-> 0.5
i===
<I) 0.<4
<I)

~
U5 0.3
:;;
Q)
.t::. 0 .2
en
0 .1

0 .25 0.50 0 .75 1.00 1.26 1.50 1.75 2 .00 2.25


7

Figure 9. Shear stress versus shear strain for simple shearing flow of a Weaire-Phelan
foam . This cell orientation has the smallest strain period ')'p=l.

of topological transitions when the foam is perfectly ordered. Kelvin cells


beget Kelvin cells.

TABLE 2. Topology changes for a Weare-Phelan foam in simple shearing flow

Faces Polyhedra
")' 4 5 6 7 8 d D X X y y z z
0-0.7 0 48 6 0 0 12 12 14 14 14 14 14 14
0.7-0.98 14 22 18 2 0 18 16 12 12 13 13 14 14
0.98-1.1 17 20 16 4 1 16 16 14 14 14 14 14 14
1.1-1.7 0 48 6 0 0 12 14 14 14 14 14 12 14

A Weaire-Phelan foam has less symmetry and exhibits more diverse


topological transitions than a Kelvin foam in simple shearing flow. Table 2
contains statistics on the outcome of topological transitions that corre-
spond to the stress-strain curve in figure 9. Up to !=0.70 where the first
Tl occurs, there are only pentagonal and hexagonal faces, which is charac-
teristic of TCP foams . The first Tl eventually produces 4-sided and 7-sided
faces, the total number of faces increases from 54 to 56, the original dodec-
ahedron 'd' gains faces to become an 18-hedron, etc. We say 'eventually'
because this structure is not achieved in a single step. A local topology
change that corrects an unstable condition leads to other edges vanish-
ing and induces another topology change, etc. The cascade of Tls that
occurs at !=1.10 involves at least nine steps before the structure regains
274

Figure 10. Wet Kelvin cells, 4>=0.01, 0.06; a wet rhombic dodecahedron, 4>=0.06; Plateau
border segments from a Kelvin cell, 4>=0.001, 0.01; and a rhombic dodecahedron.

Weaire-Phelan topology and begins a new stress-strain cycle; but notice


that bubbles have changed identity: an original dodecahedron has become
a 14-hedron and vice versa. Bubbles have shuffled within the lattice.

4.2. WET FOAMS

The Surface Evolver can also be used to study the structure and microrheol-
ogy of wet foams (and liquid-liquid emulsions) for which the volume fraction
of continuous phase ¢ is finite [27]. We consider the case where all of the liq-
uid (continuous phase) collects in Plateau borders that form interconnected
channels along bubble 'edges' where films join. Assuming dry films corre-
sponds to physical situations where film thickness is negligible compared
to the cross section of Plateau borders. The distribution of liquid between
the films and borders is set by a balance between disjoining pressure and
Laplace pressure.
A wet Kelvin cell with ¢=0.01 is shown in figure 10. The Plateau border
interfaces have tension T and the films have tension 2T from two interfaces.
The bubble and border have volume (1-¢)V and ¢V, respectively. Figure 10
illustrates an important fact about the geometry of Plateau borders and
the analogous struts in a solid foam with open cells, which will be discussed
in Section 4.3. When ¢«1, the borders can be considered long and slender
and the size of the junction can be neglected; however, ¢=0.01 might not
always be small enough for these asymptotic conditions to give accurate
results.
Unlike the dry limit, a perfectly ordered wet foam can have more than
one stable structure, e.g., bubbles compressed on a face-centered-cubic (fcc)
lattice, which relate to closest packed spheres (figure 10). The polyhedron
associated with fcc packing, the rhombic dodecahedron (RD), has twelve
275

1.25
•• • • • Kelvin a2 - c..
....... Kelvin a, - (c 11 ·c12 )/2
-•-Kelvin a
"'...~> 1.00
···•···•
...........
··•·· RD
••O•• RD
aa 2 -•
1
c44
(c 11 ·c12 )/2
;::: •···•···--··•···:. - • - RD a

.
.._0:-• ·········•·· --- Prlncen & Klu
Ill
;::,
"3
"tJ
0.75 ',.~."

...····o. ......,,
·--·~·-·"·--
,,.....,
~~-.~~~~--~'1
~................ ..... . ...
............ ..
0 .........
···o... ....., '•
~
...Ill 0.50
_, . .c···
.o•loDo•••D•••··~•-D••
..... ' . . . , ········o. ... o. •
CD o •.
.·. ,, ··o

. , .., ,,
~ '-. ....,
en 0.25 II '-.

~
o.oo+-----+---+-----r----+----t---~"'""1


0.00 0.05 0.10 0.15 0.20 0.25 0.30

Figure 11. Shear moduli for wet Kelvin and wet rhombic dodecahedron foams compared
with the empirical relation of Princen & Kiss.

identical rhombic faces. The obtuse angle of these faces equals the tetrahe-
dral angle, and the vertex shares four edges, so Plateau's laws are satisfied
locally. However, a dry foam with rhombic dodecahedral cells is unstable
because eight edges meet at the corners with acute angles. Plateau's laws
do not pertain to wet foams. The connectivity of Plateau borders can ex-
ceed four in wet foams, which gives a richer variety of topologies than in
dry foams.
The shear moduli for wet Kelvin and wet RD foams are graphed in fig-
ure 11 and compared with the empirical correlation of Princen & Kiss [28],
obtained from data for concentrated oil-in-water emulsions with polydis-
perse drop-size distributions. Note that in the dry limit this correlation
extrapolates to a value that is consistent with the results in Table 1. Both
wet structures have positive Gi and are stable over some overlapping range
of </J. The smaller shear modulus for the Kelvin foam decreases rapidly as
</J--t</J*:=::::O.ll; at </J* the smaller (original 4-sided) films shrink to zero area
prior to being consumed by surrounding borders that grow with </J. When
these borders converge to form eight-way junctions, the bubbles lose con-
tact with all next nearest neighbors on the bee lattice, which creates an
unstable situation.
Consider the uniaxial extension described in Section 4.1.3 but applied
to a wet Kelvin foam. A bee lattice becomes fcc when f.Jcc = log2=0.231. i
Figure 12 shows the evolution of bubble shape with strain. At some f.K,RD <
f. fcc, the structure undergoes a transition from wet Kelvin topology to wet
RD topology, which we label TK,RD· Films that are perpendicular to the
276

o.12r......................,.----r-.,........,.-~~=~:::!:~~::::1
- • - Kelvin (bee) energy
- • - Kelvin (bee) streao
~ RD (fcc) energy
o.oa - - - RD (fcc) etreu

Ill

-...
Ill
...
G) 0.04

en

--
0

wI
u
~
0.00

w -0 .04
0
C\1

._.._._+---+-.......-+.............-+.............-""'
-0.08 ...............
0.00 0.05 0.10 0.15 0.20 0.25
Hencky Strain e

Figure 12. Evolution of bubble shape as a wet Kelvin cell on a bee lattice stretches to
a wet rhombic dodecahedron on an fcc lattice, </>=0.04. The cell orientation is the same
as in figure 7. Graph of tensile stress and energy density, </>=0.06.

stretching axis shrink and eventually vanish at EK,RD, resulting in bubbles


with twelve films and some eight-way Plateau border junctions. Further
stretching takes these bubbles to wet RD with isotropic stress at f. fcc· Fig-
ure 12 shows the tensile stress u and interfacial energy density for this pro-
cess. The slope of the stress-strain curve evaluated at E=O and at f.= f. fcc re-
lates to the smaller shear modulus of each structure. The energy maximum
Emax that occurs when the stress changes sign at Emax, determines the en-
ergy barriers between the undeformed structures, e.g., tl.Ebcc=Emax-Ebcc,
which also represents the area under the stress-strain curve. Ebcc and Efcc
both decrease with¢> and cross when ¢~0.064; Ebcc is lower when¢> is small.
The process just described appears to be completely reversible-the wet
RD foam can be compressed to a Kelvin foam. Reversibility implies that the
energy and its first derivitive are continuous at EK,RD=ERD,K · Reversibility
might be expected because of the symmetry involved. The reverse topologi-
cal transition TRD,K involves bubbles that are aligned with the compression
axis and separated by an eight-way junction, becoming neighbors. The new
neighbors are separated by a film that forms where the bubbles first contact
277

0 .45

Dry Kelvin

0 .30

VI

...
VI

-- --~--,,._.............................-
Q) 0.15
U5
...
al
Q) 0.00
~
en / ',, RD /'•lvln
Kelvin ... ______: .......····

-0.15

-0.30 ...,........._.-+_......~1--'--........-+-.......--+-......-1--'--.......-1
0.00 0.25 0 .50 0.75 1.00 1.25 1.50
-y

Figure 13. Evolution of bubble shape and shear stress for simple shearing flow of a wet
Kelvin foam with r/>=0 .06. Cell orientation is the same as in figure 8. Topology changes
from wet Kelvin to wet RD to wet Kelvin. The lower bubbles show a different view of the
second transition, which involves the formation of a new film when opposite interfaces of
an eight-way junction come together.

and grows in area as E decreases below ERD,K. A transition of type TRv,K


is initiated when interfaces on opposite sides of an eight-way junction come
into contact.
Both topological transitions occur in simple shearing flow but they are
not reversible. Consider the flow described in Section 4.1.4 but applied
to the wet Kelvin foam with ¢=0.06. Figure 13 contains representative
structures and the stress-strain curve. Instead of two Tis per cycle like the
dry case, there are four distinct topology changes in the wet case: a TK,RD,
then a TRv,K, then a TK,RD, and then a TRD,K to complete the cycle.
Both TK,RD transitions involve the 'same' shrinking film as the dry foam,
but in each case, a stable wet RD results. Both TRD,K transitions involve
contact between opposite interfaces of an eight-way junction. Different from
the reversible uniaxial extension, the stress and energy both decrease with
each transition in simple shear. The magnitude of these jumps is much
smaller for the wet foam than the dry foam (figure 13). The same is true
278

Figure 14. Beam element models for open-cell foams including: Kelvin, Weaire-Phelan,
and random with N=64. Continuum element mesh for a Kelvin foam with ¢=0.0267.

for the average shear stress (a), which corresponds to the (viscometric)
yield stress of the foam. Consistent with measurements of yield stress by
Princen [29], (a) decreases very rapidly with ¢.
When ¢ is very small, we anticipate that wet RD structures will be
unstable intermediates that lead to wet Kelvin structures. There will be
two Kelvin branches in the stress-strain cycle, just like the dry limit.

4.3. SOLID FOAMS WITH OPEN CELLS

In nature and industry, liquid foams undergo phase changes to produce


solid foams so the cell-level structure of the former heavily influences the
latter. This justifies and motivates using the geometry of soap froth and
related materials as templates for developing micromechanics models for
cellular solids.

4.3.1. Linear Elastic Behavior of a Kelvin Foam


A low-density Kelvin foam with open cells is composed of identical, straight
struts of length 2£ that meet at identical, rigid joints. Force-displacement
relations at the strut level are expressed through compliances for stretching
M, bending N, and twisting .:1. Analytic solutions for arbitrary homoge-
neous deformations of the foam provide all forces, moments, and displace-
279

ments at the strut level, and the rotation at each joint [30]. The effective
elastic constants of the foam include the bulk modulus K and two shear
moduli, which take the form
1
K =
24¥2 .C M
1
4¥2 .C (M + 3N)
1
(7)
8¥2 .C (M + N + .C2.J)
The bulk modulus only depends on M; struts neither bend nor twist when
this highly symmetric structure experiences hydrostatic compression. Bend-
ing and twisting are the dominant strut-level deformation mechanisms for
foams under shear because M «N when the foam density <P is small. The
modulus G 1 corresponds to compression along a (100} axis (pushing on
squares); the absence of .J indicates that struts do not twist (figure 14).
The modulus G2 corresponds to compression along a (111} axis (pushing
on hexagons); struts bend and twist. Beam theory can be used to evaluate
the compliances

.C2 .J = ~N(1 + 9) (8)

where E is Young's modulus of the material in long slender struts with


uniform cross section of area A, radius of gyration q, and moment of inertia
I =q2 A. The elastic constants for the foam become

G = 16¥'2 E (q2) ,~..2 (9)


1 9 A'~''

when ¢«1. The parameters q2 j A and 9 depend on the strut cross sec-
tion. Shapes of interest include circle, equilateral triangle, and Plateau
border, which corresponds to the space between three identical, mutually
tangent circles. The elastic response of the foam is isotropic (G1=G2) when
9=0, which occurs for circular struts with a Poisson's ratio v of zero. In-
compressible struts with Plateau border shape cause the most anisotropy:
9=2/13, G1 =0.336E¢2, G2=0.320E¢2. An open-cell Kelvin foam exhibits
weak anisotropy for strut shapes and material properties of interest; this
insensitivity to orientation is surprising since G2 involves twisting but G 1
does not.
The effective isotropic Young's modulus E for incompressible struts with
Plateau border shape is
- 2
E = 0.979 E ¢ . (10)
280

Gibson & Ashby [4] chose a coefficient of one to correlate experimental


data. The coefficient in (10) is 40% smaller for circular struts.
Zhu et al. [31, 32] have performed an independent analysis of the linear
problem under consideration and used the Elastica approach to analyze
uniaxial compression at finite strains.

4.3.2. Large Deformation Behavior: Finite Element Analysis


Foams composed of nonlinear material, which is arranged to form a com-
plex cell structure, are subjected to large deformations in a wide range of
applications. A standard industrial test involves foam compression between
two flate plates. A variety of finite element techniques are well suited for
solving micromechanics problems that are relevant in these situations.

Beam Elements The primary microstructural feature found in low-density


open-cell foams is the network of slender struts that meet-most often with
a connectivity of four-at joints. Beam elements can be used to discretize the
struts along their axis. We use the B31 beam elements that are contained in
ABAQUS, a general purpose finite element program [33]. The struts in an
undeformed foam are assumed to have uniform cross section and a straight
axis, but these approximations can be relaxed. The location of joints and
the connectivity of struts are based on the foam geometries described in
Section 4.1.1; these include perfectly ordered, TCP, and random structures.
Beam section properties are chosen according to the strut shape and the
relative foam density, which is calculated from ¢ = v- 1 :Ei=l Ai.Ci, where
Ai and .Ci are the section area and length of strut i. The geometry and
deformation of the joint region are neglected in the simplest beam models,
which focus entirely on the mechanics of connecting struts.
Consider situations where a foam is subjected to uniaxial compression
without confining the sides. The shear stresses and lateral normal stresses
are zero. Stress-strain curves for three representative structures are con-
tained in figure 15. Calculations for ¢>=0.001, 0.01 scale with E¢> 2 • Two
curves indicate the range of orientation dependence when a foam has cubic
symmetry. The results for a Kelvin foam agree with the Elastica analysis of
Zhu et al. (private communication, N.J. Mills). The large-strain behavior
of the Kelvin foam is more sensitive to orientation than the shear modulus.
The Weaire-Phelan foam is stiffer than the Kelvin foam and exhibits as
much anisotropy, even though it contains eight different cells. The Weaire-
Phelan response is characteristic of TCP foams, except the T structure,
which has 81 different cells, is significantly less anisotropic. A single curve
represents the essentially isotropic behavior of two different random foams
with 64 cells. The Young's modulus of the random foam is about 10% larger
than the Kelvin foam; this relates to the leading coefficient in (10).
281

0.5
••••• • WP <100>
........... WP <111>
- ·-Kelvin <100>
0.4 ---Kelvin <111>
..·················
..
- Random N• 84
"'o&
w 0.3
.... .--····
U) .....··
.. -
-
. . ········::··~:~~::::======~::::~
...
U)
Q)
0.2
en
0 .1
.....· ,..,.~·:,;;,.· ._­

0 .0
.· ... [,joO"--

0 .0 0 .1 0.2 0.3 0 .4 0 .5
Strain

Figure 15. Stress-strain curves for Kelvin, Weaire-Phelan, and random foams under
uniaxial compression.

Since the strut material is linear elastic, the significant nonlinearity


of the curves in figure 15 stems entirely from large deformation of the
microstructure.
In view of the simplicity of Kelvin's cell and the complexity of random
foams, their macroscopic response to uniaxial compression is very similar.
It is unnecessary-in some cases-to model large disordered structures, espe-
cially when perfect order will do.
The foam response is quite different for hydrostatic loading, which pro-
duces pure volume change. The pressure for the Kelvin foam is well repre-
sented by linear theory (9) up to the critical buckling load. The pressure
scales as E¢ because struts only carry axial loads in a perfect Kelvin foam.
Structures that are not perfectly ordered contain struts of different length
meeting at joints at different angles. Under hydrostatic loading, the struts
do not experience pure axial loading and joint displacements are not affine.
Off-axis loading causes strut bending and twisting at very small volume
strains, which leads to the stress plateaus in figure 16. The level of these
plateaus does not scale with E¢ since bending is involved. The insert of
a deformed Weaire-Phelan foam in figure 16 illustrates the strut bending
generated by hydrostatic compression. The onset to the plateau is very
gradual for the random foam. The plateau level in figure 16 is the same for
the random and the T structure.
Recent studies of buckling instabilities, which will not be discussed here,
indicate that small imperfections in the structure of Kelvin and Weaire-
Phelan foams will lead to strut buckling and also produce a plateau level
282

0 .010
---Kelvin
........... Wealre - Phelan
• • •• •. T
0. 008
- - Random N=84
0)
.......
.eo
UJ 0 .006

....
Q)

::l
C/)
C/)
0 .004
....
Q)
[l.

0 .002

0. 000~:;.,.,---+-----+-----+----+----!
0 .000 0.002 0.004 0 .006 0.008 0 .0 10
Volume Strain

Figure 16. Stress-strain curves for Kelvin, Weaire-Phelan, T, and random foams under
hydrostatic compression.

similar to the random foam.


Future analysis of hydrostatic compression and other complex load
paths promises to shed more light on cell-level deformation mechanisms
in foam mechanics. Large-deformation response of foams may be easier to
study on a computer than in the laboratory.

Continuum Elements Struts become thicker and joint regions become


larger as the density of an open-cell foam increases. Eventually, the ap-
proximations used to formulate beam elements are violated. Deformation
within the joint regions also becomes more important as the distinction
between joint and strut becomes more arbitrary (figure 10). Under these
circumstances, continuum elements can be used to discretize the entire
solid phase within the foam. This approach permits more accurate model-
ing of the cell-level geometry and the use of more sophisticated constitutive
equations to describe nonlinear behavior of the foamed material such as vis-
coelasticity, plasticity, etc.
Figure 14 contains a continuum element mesh of an open-cell Kelvin
foam with ¢=0.0267. This corresponds to a flexible polyurethane foam with
density: two pounds per cubic foot , which is often referred to as low-density.
It is obvious that the struts are thick and the joints have significant volume.
Computations with continuum elements are b eing used to study the ¢-
dependence of properties in dense foams where details of cell-level geometry
are more subtle.
283

Figure 17. Shell element meshes for closed-cell Kelvin and random foam with N=lO.

4.4. SOLID FOAMS WITH CLOSED CELLS

The cells in a closed-cell foam are separated from neighbors by cell walls,
which are absent in an open-cell foam. We consider an extreme case where
all cells are closed and all of the solid material is located in the cell walls,
which is similar to neglecting the joints in a beam analysis of open-cell
foams. We also assume that the cell walls are thin and flat with uniform
thickness. Even with these simplifications, analytic solutions for linear elas-
tic behavior of a Kelvin foam are not feasible. We use finite element analysis
and discretize the faces with the quadrilateral shell element S4R contained
in ABAQUS . With the exception of the Kelvin cell, the vertices of each face
in a dry foam do not lie in a plane; therefore, we use Voronoi partitions
to model the geometry of closed cell foams. Typical shell-element meshes
are shown in figure 17. The Weaire-Phelan model is based on the weighted
Voronoi construction of Kusner & Sullivan [34], which is monodisperse.
284

The elastic constants that were computed for closed-cell Kelvin and
Weaire-Phelan foams with v=0.49 for the wall material and ¢«:..1 are

K = 0.435 [0.434] E<P


G1 - 0.108 [0.113] E<P
G2 - 0.116 [0.111] E<P (11)

where brackets refer to Weaire-Phelan. Note that all elastic constants, even
the shear moduli, scale withE¢, which indicates that in-plane deformation
of cell walls dominates the cell-level mechanics. This is very different from
2D where the shear modulus scales with E¢ 3 because cell walls bend. In
3D, each face is connected to two other faces around its entire perimeter;
the resulting mechanical constraints suppress bending in the linear elastic
regime.
The elastic constants for the Weaire-Phelan foam are very close to the
Kelvin foam. The same is true for other structures (Friauf-Laves, a random
Voronoi partition with ten cells), whose constants are not shown. All of the
closed-cell foams exhibit very similar linear elastic response. It will be very
interesting to see how widely this observation applies. Also different from
our results for low-density open-cell foams, all of the closed-cell foams are
nearly isotropic.
Hashin & Shtrikman [35] calculated upper bounds on elastic constants
for two-phase solids. For low-density porous materials with v=0.49 for the
solid phase, KHs = 0.4357 E<P and GHs = 0.1996 E</J. The bulk modulus of
the closed-cell foams considered here is very close to the Hashin-Shtrikman
bound. The shear moduli are about 50% smaller than the upper bound.
Compare the elastic constants of open-cell foams and closed-cell foams
given in (9) and (11). Closed-cell foams are stiffer. The bulk modulus has
the same scaling in both cases but different numerical coefficients. Open-
cell foams are much softer than closed-cell foams in shear since the modulus
scales with E¢ 2 •

Acknowledgements We thank Jean-Francais Sadoc and Nicolas Rivier


for creating a magical experience in Cargese. Sandia is a multiprogram
laboratory operated by Sandia Corporation, a Lockheed Martin Company,
for the United States Department of Energy under contract #DE-AC04-
94AL85000. This work was also supported by the Dow Chemical Company
under a Cooperative Research and Development Agreement (CRADA).
285
References
1. Love, A.E.H. (1994) A Treatise on the Mathematical Theory of Elasticity, Dover,
New York.
2. Weaire, D. and Fortes, M.A. (1994) Stress and strain in liquid and solid foams,
Advances in Physics, 43, 685-738.
3. Kraynik, A.M. (1988) Foam flows, Ann. Rev. Fluid Mech., 20, 325-357.
4. Gibson, L.J. and Ashby, M.F. (1997) Cellular Solids: Structure and Properties, 2nd
Ed, Cambridge University Press, Cambridge.
5. Kraynik, A.M. and Warren, W.E. (1994) The elastic behavior of low-density cellular
plastics, in N.C. Hilyard and A.C. Cunningham (eds.) Low Density Cellular Plastics,
Chapman & Hall, London, 187-225.
6. Plateau, J.A.F. (1873) Statique Experimentale et Theorique des Liquides Soumis aux
Seules Forces M oleculaires. Gauthier-Villiard.
7. Princen, H.M. (1983) Rheology of foams and highly concentrated emulsions. I. Elastic
properties and yield stress of a cylindrical model system, J. Coll. Int. Sci., 91, 160-175.
8. Khan, S.A. and Armstrong, R.C. (1986) Rheology of foams. I. Theory for dry foams,
J. Non-Newtonian Fluid Mech., 22, 1-22.
9. Kraynik, A.M. and Hansen, M.G. (1986) Foam and emulsion rheology: A quasistatic
model for large deformations of spatially periodic cells, J. Rheology, 30, 409-439.
10. Reinelt, D.A. and Kraynik, A.M. (1990) On the shearing flow of foams and concen-
trated emulsions, J. Fluid Mech., 215, 431-455.
11. Kraynik, A.M., Reinelt, D.A. and Princen, H.M. (1991) The nonlinear elastic be-
havior of polydisperse hexagonal foams and concentrated emulsions, J. Rheology, 35,
1235-1253.
12. Bolton, F. and Weaire, D. (1991) The effects of Plateau borders in the two-
dimensional soap froth. I. Decoration lemma and diffusion theorem, Philos. Mag. B.,
63, 795-809.
13. Bolton, F. and Weaire, D. (1992) The effects of Plateau borders in the two-
dimensional soap froth. II. General simulation and analysis of rigidity loss transition,
Philos. Mag. B., 65, 473-487.
14. Hutzler, S., Weaire, D. and Bolton, F. (1995) The effects of Plateau borders in the
two-dimensional soap froth. III. Further results, Philos. Mag. B., 11, 277-289.
15. Warren, W.E. and Kraynik, A.M. (1987) Foam mechanics: The linear elastic re-
sponse of two-dimensional spatially periodic cellular materials, Mech. Materials, 6,
27-37.
16. Kelvin, Lord (Thompson, W.) (1887) On the division of space with minimum par-
titional area, Philos. Mag., 24, 503-514.
17. Weaire, D. and Phelan, R. (1994) A counter-example to Kelvin's conjecture on
minimal surfaces, Phil. Mag. Lett., 69, 107-110.
18. Brakke, K.A. (1992) The surface evolver, Experimental Mathematics, 1, 141-165.
19. Rivier, N. (1994) Kelvin's conjecture on minimal froths and the counter-example of
Weaire and Phelan, Phil. Mag. Lett., 69, 297-303.
20. Nye, J.F. (1985) Physical Properties of Crystals, Clarendon Press, Oxford.
21. Matzke, E.B. (1946) The three-dimensional shape of bubbles in foam-an analysis
of the role of surface forces in three-dimensional cell shape determination, Am. J.
Botany, 33, 58-80.
22. Kraynik, A.M. and Reinelt, D.A. (1996) The linear elastic behavior of dry soap
foams, J. Coll. Int. Sci., 181, 511-520.
23. Kraynik, A.M. and Reinelt, D.A. (1996) The linear elastic behavior of a bidisperse
Weaire-Phelan foam, Chern. Eng. Comm., 148-150, 409-420.
24. Reinelt, D.A. and Kraynik, A.M. (1996) Large elastic deformations of three-
dimensional foams and highly concentrated emulsions, J. Coll. Int. Sci., 159, 460-470.
25. Kraynik, A.M. and Reinelt, D.A. (1996) Elastic-plastic behavior of a Kelvin foam,
Forma, 11, 255-270.
286
26. Reinelt, D.A. and Kraynik, A.M. (1996) Simple shearing flow of a dry Kelvin soap
foam , J. Fluid Mech., 311, 327-343.
27. Kraynik, A.M., and Reinelt, D.A. (1996) The microrheology of wet foams, in A.
Ait-Kadi, J .M. Dealy, D.F . James and M.C. Williams (eds .) Proceedings of X!Ith
International Congress on Rheology, Quebec City, Canada, August 18-23, 625-626.
28. Princen, H.M. and Kiss, A.D. (1986) Rheology of foams and highly concentrated
emulsions. III. Static shear modulus, J. Call. Int. Sci., 112, 427-437.
29. Princen, H.M. (1985) Rheology of foams and highly concentrated emulsions. II.
Experimental study of the yield stress and wall effects for concentrated oil-in-water
emulsions, J. Call. Int. Sci., 105, 150-171.
30. Warren, W .E. and Kraynik, A.M. (1997) Linear elastic behavior of a low-density
Kelvin foam with open cells, ASME J. Appl. Mech ., to appear.
31. Zhu, H.X., Knott, J .F . and Mills, N.J. (1997) Analysis of the elastic properties of
open-cell foams with tetrakaidecahedral cells, J. Mech. Physics Solids, 45, 319-343.
32. Zhu, H.X., Mills, N.J. and Knott , J .F . (1997) Analysis of high strain compression
of open-cell foams with tetrakaidecahedral cells, J. Mech . Physics Solids, to appear.
33. ABAQUS User's Manual, Version 5.6 (1996) Hibbitt, Karlsson and Sorensen Inc.,
Providence, Rhode Island.
34. Kusner, R. and Sullivan, J .M. (1996) Comparing the Weaire-Phelan equal-volume
foam to Kelvin's foam, Forma, 11, 233-242.
35. Hashin, Z. and Shtrikman, S. (1963) A variational approach to the theory of the
elastic behavior of multiphase materials, J. Mech. Physics Solids, 11, 127-140.
THE STRUCTURE AND GEOMETRY OF FOAMS

D. WEAIRE** , R. PHELAN* AND G. VERBIST*


*Shell Research and Technology Centre, P. 0. Box 38000,
1030 BN Amsterdam, The Netherlands.
**Department of Physics, Trinity College, Dublin, Ireland.

Abstract. The basic equilibrium rules which define the ideal foam model
are reviewed. This model may be accurately represented by simulations in
two or three dimensions. The techniques used are sketched and numerous
examples given, with some advice on the evaluation of results.

1. Introduction: The Ideal Foam Model

Our subject is the basic geometry of foams, much of which was elucidated in
the classic monograph of Plateau (1873), fig. 1, together with the structures
which arise from this, both ordered and disordered. We shall however not
stray far into the various considerations of statistics of disordered foam
structures: these will be covered elsewhere in the present volume.

Figure 1. The geometry of soap films spanning wire frames as modeled using the Evolver.
Plateau observed such arrangements experimentally and from them deduced the rules of
equilibrium for dry foam structure.
287
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 287-302.
© 1999 Kluwer Academic Publishers.
288

Some of the rules which we shall state are quite general, but in most
cases a model is implied, in that the films which constitute the faces of
the foam cells have infinitesimal thickness, and a surface tension 20" which
is twice that of a single liquid/ gas interface. Let us call this the the ideal
foam model. It is an extremely successful idealisation, which lends itself
to accurate simulations in both two and three dimensions, by techniques
which will be described. Thus we can determine many of its properties with
arbitrary precision.
In the absence of any further elaboration the model excludes many
dynamical effects: only those capable of being described quasi-statically
are included. This means that the processes in question are described in
terms of equilibrium states, punctuated by topological changes which are
regarded as being instantaneous. Examples are stress-strain relations for
slowly varying strain, which are described in a subsequent article {Kraynik,
1997) and coarsening due to the slow diffusion of gas between cells.
To further simplify the model without losing much of its accuracy, both
liquid and gas components are usually regarded as incompressible.

2. Laplace Law and Curvature


First-year physics courses usually include the formula for the excess pressure
inside a bubble within a liquid:

A 20"
up=- {1)
r
This is a particular case of the Laplace Law, which relates the difference
in pressure across a surface ~p, associated with surface tension O", to its
curvature, here expressed in terms of the radius of curvature, r. This is
simply the radius of the bubble. The law represents the equilibrium of
surface tension and pressure forces on an element of surface. More generally,
such a surface is not spherical (see fig. 2). Two local curvatures are defined,
corresponding to radii r1 and r2, but the same formula is valid, taking the
average:
2 1 1
- = - + - = 2H {2)
r r1 r2
where we have defined H, the mean curvature.
In the ideal foam model, a further factor of two enters equation {1), in
describing a soap film with two surfaces, as when a bubble is blown in air.
If gravity is neglected, the pressure on each side of the surface does not
vary with position, so the above mean curvature is fixed. Nevertheless this
allows the individual components of curvature to vary and the surface to
take subtle shapes, difficult to capture in mathematical form. In particular
289

Figure 2. A single bubble from a disordered dry foam structure. The faces are not simple
spherical surfaces and have no elementary mathematical description

the surfaces to be found in foams and bubble clusters have this double
curvature and are not spherical, except a few special cases (Sullivan, 1997).
The case of a 2D foam, such as is made by squeezing a sample of liquid
foam between two glass plates, is much simpler. One component of curva-
ture is taken to be zero (but beware the imperfect experimental realisation
of this!) and the surfaces are represented as lines, which are arcs of circles.

3. Thin Films

Let us take a closer look at the basis for the ideal foam model, which re-
duces the modeling of foam structure to simple equilibrium rules. A typical
foam is composed of bubbles of an average diameter which is of the or-
der of millimeters, and where the faces (the bubble-bubble contacts) have
thicknesses on the scale of microns. The films are held at this thickness
by repulsive forces between their two surfaces, which are covered with sur-
factant molecules. If these forces are of sufficiently short range, we can
think of them as hard wall interactions, dictating a minimum separation
but contributing zero to the energy of the system.
Precise measurement and interpretation of the disjoining pressure, which
is the force between the two faces of the foam expressed per unit area, is a
large subject in itself (Langevin, 1997).
The idealisation suggested here, and the difference in scale of bubble
size and film thickness, give us the required rationalisation of the ideal
foam model.
290

4. Vertices and Plateau Borders in the Ideal Foam Model


In the ideal model of a dry foam, the cells {bubbles) are surrounded entirely
by the thin films. Accordingly they are polyhedra, whose curved faces meet
on curved lines, and the lines meet at vertices. The liquid fraction is negligi-
ble, essentially zero. An ordinary soap froth, left to stand for a few minutes
reaches an equilibrium state, close to the dry foam limit, as gravity causes
most of the liquid to drain away. In the 2D case, we have polygonal cells
consisting of circular arcs, in the same limit.
The cell faces in 3D, or cell sides in 2D, cannot meet in arbitrary ar-
rangements. Only three can meet, and the angles which they make with
one another are 120° . Furthermore, the lines so defined in the 3D case {the
edges of the cells) can only meet four at a time, in symmetric tetrahedral
vertices (corners of the cells). These are the celebrated rules of Plateau,
which state that all other configurations are unstable, fig. 3. This is easily
justified, except for the fourfold restriction of 3D vertices, which requires
an elaborate proof, only recently completed in all technical details {Taylor,
1976 and Almgren and Taylor, 1976).

Figure 3. An illustration of Plateau's equilibrium rules. Cell films meet at 120°and lines
meet at arccos(- t) ~ 109.5°.

Where conditions are such that a significant amount of liquid is con-


tained in the foam, we speak of a wet foam. In the 3D ideal foam model,
291

the liquid is incorporated in the cell edges, which become a network of


Plateau borders, fig. 11. As the liquid fraction increases, these borders swell
and the cell faces shrink and eventually disappear. Ultimately the bubbles
must come apart, in what has been termed the rigidity loss transition, be-
cause mechanical stability is lost at this point. This may also be called the
wet limit: the corresponding critical value of liquid fraction depends on the
structure, but it typically about 0.35 for a disordered 3D foam. In 20, the
liquid is to be found at the vertices, as cross sections of transverse Plateau
borders. {In some experimental realisations there will be horizontal Plateau
borders at the confining plates)
Recall that the Plateau rules just stated are strictly true only for dry
foam. We may expect, and find, that foams of a low finite liquid fraction
conform well to the rules, but these break down progressively as the liquid
fraction increases, so that multiple junctions are found which are excluded
at the dry limit, as in fig. 4.

Figure 4. A wet eightfold vertex. Such multiple vertices are common in foam of finite
liquid fraction but are unstable in the extreme limit </>1 -+ 0.

5. Gas and Liquid Pressures


An important experimental variable in some types of experiment on foams
and emulsions is the osmotic pressure. We will define it here in terminology
appropriate to the case of a foam.
292

From an experimental point of view it may be defined in terms of the


derivative of the energy (or free energy) with respect to volume, keeping the
volume of the (incompressible) gas fixed. It therefore refers, in particular,
to a situation in which liquid can leave or enter the system through an
appropriate filter and the entire volume change is due to this. The definition
is

II=_ (8F)
av v 9
(3)

Alternatively, in a calculation it may be more convenient to use

II =p- Pl (4)
where p is the total pressure and pz is the pressure in the liquid.
When gravity is neglected, pz must be the same throughout the system.
In general the gas pressure p 9 is not, but rather varies from bubble to
bubble.
On the other hand, when gravity is considered, pz must follow the usual
hydrostatic equilibrium rule:

Pl =Po- pgh (5)

where p 0 is atmospheric pressure, and it is sometimes judicious to set the


gas pressure constant, because its variation is comparatively insignificant
(Weaire et al., 1997).
Just as pressure is paired with volume as its complementary variable,
so the osmotic pressure may be associated with gas fraction, and we may
ask what the function II(¢9 ) is, which is equivalent to asking how the total
energy varies with ¢ 9 . We return to this question later (Verbist et al., 1997).

6. Inversion and the Decoration Theorem


In 2D it is possible to transform any equilibrium foam structure into an-
other one by the geometrical operation of inversion, fig. 5. This is true in
great generality, applying to arbitrary liquid fraction and even allowing for
different values of surface tension on different lines. One nice application
is the proof of the Decoration Theorem, which states that any 2D foam
with only three-sided Plateau borders can be regarded as an equilibrium
dry foam, "decorated" with these borders. That is, each border can be re-
moved and replaced by a vertex, at which the corresponding sides meet,
when extrapolated with constant curvature.
The Decoration Theorem has been used to neatly estimate, inter alia,
the effects of the Plateau borders on gas diffusion between cells (since they
293

(0,1)

Figure 5. Inversion through a point. Under this transformation an equilibrium 2D


foam (in this case just a single wet vertex) maps into another equilibrium structure. The
inversion centre is at (6, 1) relative to the co-ordinates shown in the left-hand figure.

simply block diffusion across those parts of the cell sides which they ob-
scure). It is also helpful in thinking about mechanical properties, since the
Plateau borders can have no effect on relative energy at different strain
(i.e. the stress) until the point at which they meet and undergo topological
transformations.
Note the limitations of the Decoration Theorem: it does not apply at
high liquid fractions, when multiple vertices are formed. Nor can it be
proven in 3D, where the inversion theorem stated above does not hold
generally.

7. The Hexagonal Honeycomb in 2D

In 2D a very special role is reserved for the hexagonal honeycomb structure,


which satisfies the equilibrium condition with straight sides and hence equal
pressure in all cells. It is fairly obvious that, for equal-sized cells, this must
have the global minimum of line length and hence energy. Nevertheless this
has not so far been fully proved (e.g. see Sullivan, 1997).

8. Simulation

The representation of foam structure can be pursued in a variety of ways.


Here we refer only to those which are essentially exact, i.e. numerical sim-
ulations which may be made as accurate as we wish, given the necessary
computing resources. They therefore add little or nothing to the approxima-
tions and limitations inherent in the ideal foam model itself. Among other
294
things, such accurate simulations may serve to validate more approximate
theories, by comparison.

9. 2D Simulation
In 2D a straightforward simulation may be pursued as follows
- Define some provisional structure with periodic boundary conditions
and stipulate the required area of each cell.
- Iteratively adjust the following variables to adjust the structure so
that the equilibrium rules are satisfied everywhere and the areas are
as specified. These variables are the vertex positions where the circular
arcs which represent the cell sides are joined and the pressures in the
cells and Plateau borders. The cell pressures are different while those
in the Plateau borders are usually the same everywhere because these
are connected at the plates. A simulation based on these principles
was originally done for dry foam by Weaire and Kermode (1983), but
eventually a similar approach to the wet 2D foam was also successful,
(Bolton and Weaire, 1991). Alternatively a hydrostatic variation of
Plateau border pressure may be incorporated, for a vertically disposed
2D foam.
The program which accomplishes this must be able to accommodate
topological changes. Figure 6 shows examples of such simulations.

(a) (b)

Figure 6. Illustration of 2D foam simulations showing foam samples at both low (a)
and high (b) liquid content (Hutzler and Weaire, 1995)
295

10. 3D Simulation
Three-dimensional simulations present some challenging computational dif-
ficulties, the main complication being that the individual bubble-bubble
interfaces have no simple geometric description and can only be modeled
in some approximate form. There is a general numerical tool called the
'Surface Evolver 1 ' specifically designed to represent and analyse complex
surfaces and is well suited to the particular case of foam structure.

11. The Surface Evolver


The software is capable of modeling surfaces driven by surface tension and
other forces and subject to a variety of constraints. A continuous surface
is approximated as an elementary mesh of flat triangles each defined by an
ordered loop of straight edges2 , the edges themselves being ordered pairs
of vertices.
A body is declared simply as those facets which define its boundary
and can be constrained to have a fixed volume or pressure. It is possible to
define an arbitrary periodic unit cell so that bulk foam structures may be
simulated. Boundary constraints can also be defined, forcing parts of the
surface to lie in a particular plane, for example. This may represent the
walls of a container.
Once the mesh has been defined, bodies declared and any necessary
constraints specified the software evolves the surface towards a minimum
using one of a number of gradient descent methods (Brakke, 1977), the
conjugate gradient method being the one most commonly used.

12. Example Evolver Calculation


Clearly there is a trade-off between the computational resources required
and the level of tessellation used. Such a model is only useful if the cells
and surfaces as represented at the available levels of refinement are good
approximations to the limiting infinitely refined structure. To give some
feel for the issues of accuracy and convergence consider the following simple
example calculation: the evolution of a unit volume Kelvin cell.
Starting from the input configuration shown in fig. 7(a), the surface is
successively minimised (using the conjugate gradient method) and refined.

1 The Evolver was written by the U.S. mathematician Kenneth Brakke. It is avail-
able via anonymous ftp from the Geometry center at geom.umn.edu in the directory
pub/software/evolver.
2 1t is also possible to represent each edge as a 'quadratic curve' and the triangles by
quadratic patches. This method is more accurate but somewhat slower to implement.
Extra work is also needed to use this method for periodic surfaces.
296

(a) RO (b) Rl (c)R2

Figure 7. Stages in the minimisation of a unit volume Kelvin cell; (a) shows the initial
plane sided input structure while (b) and (c) show the first two refined structures with
surface area minimised.

5.315
RO
5.314

5.313
5.312
5.311
5.310

5.309 Rl

5.308
R2
5.307 R3 R4
5. 306 l:--::,:j····i:.::···=····::.::···=····:::···=····:::··-~----=---=----= ======d
---=----=---=----s:---:::--
0 200 400 600
Iteration number

Figure 8. Minimisation of a unit volume Kelvin cell. The steps indicate increasing
refinement level (see fig. 7) . The x-axis shows the number of conjugate gradient minimi-
sation steps used. The dashed horizontal line is the energy value obtained after further
calculation and is accurate to 5 decimal places.

Figure 8 shows a plot of the energy as a function of the total iteration


number covering the first five refinements. We have allowed a sufficient
number of iterations at each refinement (R-level) to reach the best mesh
configuration for that level as shown by the extended plateaus in fig. 8.
297

5.30702
- 0
(.)
5.30700
=
-
......
~
>
5.30698
~
->=
0
.....
......
5.30696

=
= 5.30694
~
.....0
5.30692
~ R2
=
0
~
5.30690

0 20 40 60 80 100
Iteration number

Figure 9. A blow up of the iteration of the second refinement level (R2) shown in fig. 8.

I Refinement Level I Final Energy I Extrapolated Energy I


RO 5.31474

Rl 5.30874
R2 5.30690 5.30608

R3 5.30643 5.30627
R4 5.30632 5.30627

TABLE 1. The final energies at each of the refinement levels


plotted in fig. 8. Based on the current and previous levels
the Evolver also estimates the final minimised energy us-
ing a power law extrapolation. Further refinement, obtaining
the energy accurately to five decimal places gives a value of
5.30627(5).

There is some structure within this overall picture however; fig. 9 shows a
blow up of the 100 iterations used at the second refinement level (R2}. Care
must be taken to avoid confusing the small flat regions with the minimum
298
for that R-level. Table 1 lists the final energies for each level as well as a
power law extrapolation based on the current and previous final energies.
Combining the gaps between these final energies at subsequent levels
and the extrapolated value we get a good feel for the probable error. This
knowledge is particularly important for more complex structures where it
is not possible to refine to the same degree used here.

13. Application to Wet Foams


The example presented above was an application of the Evolver to the case
of a perfectly dry foam. What of the simulation of wet foam structures? In
an idealised dry foam three faces meet at an edge and four edges at each
vertex. A wet foam has a network of liquid Plateau borders in place of these
triple lines and tetrahedral vertices. It may be represented in the Surface
Evolver by having one Evolver 'body' for each cell and one body for the
entire Plateau border, each body of fixed volume. The Evolver cannot de-
tect surfaces intersecting each other so double-sided films between adjacent
cells must be explicitly included as single surfaces, along with the surfaces
between cells and the Plateau border. The surfaces between cells are given
twice the surface tension of the surfaces shared by cells and the Plateau
border.
The initial Evolver datafile for a wet foam is considerably more compli-
cated than that for a dry foam. To ease the work, a program was written
in the Evolver command language to construct a wet foam datafile from an
existing dry foam 3 . The program inserts a triangular tube in place of each
triple edge, and inserts an octahedron in place of each tetrahedral point, see
fig. 10(a). Half of the octahedron faces are omitted, since they are attach-
ment faces for the triangular tubes. The tensions of the new surfaces are set
as indicated above, and the body volumes recomputed to account for the
volume in the Plateau border. The resulting datafile can be run without
manual editing, but it is necessary to adjust the volumes to get the desired
liquid fraction. Figure 10 shows a single vertex of the type found in wet
foams and the stages in its minimisation. Figure 11 illustrates an extended
network of Plateau borders from a wet Kelvin bulk structure.

14. Examples of Simulated Structures


We end this survey with some further examples of foam structures simulated
by the Evolver. For details of the motivation and interpretation of these
calculations, the relevant publications should be consulted (Weaire and

3 This may be obtained from the same site as the Evolver itself. See for example
(Phelan et al., 1995).
299

(a) (b)

(c) (d)

Figure 10. Example of a single wet foam vertex and its four attached Plateau border
arms. The initial input of uniform triangular pipes meeting at a partial octahedron is
shown in (a). As the vertex is refined and minimised we obtain the structures shown in
(b) , (c) and (d) . The inter-bubble films have been removed but their positions can easily
be inferred.

Figure 11. Part of a wet bulk Kelvin foam showing the network of Plateau borders. The
inter-bubble films have been removed for clarity.
300

Phelan, 1994(a)), (Weaire and Phelan, 1996), (Phelan et al., 1995), (Phelan,
1997), (The Kelvin Problem, ed. D. Weaire, 1997).
The Kelvin foam shown in fig. 7 is based in a very natural way on the
b.c.c. structure of solid state physics. In the same way we can construct
foams with more complicated topologies based on the crystalline phases
of complex alloys. See for example Rivier (1995) for a discussion of an
important class of such topologies, the Frank-Kasper phases. The structure
found by Weaire and Phelan (1994) is drawn from this class. Figure 12
shows two further examples of such foams.
We can model a wet foam based on an existing dry foam in the manner
described previously. Figure 13 shows both wet f.c.c. and Weaire-Phelan
structures. It is interesting to consider which bulk foams are stable (and/or
which are optimal) as the liquid fraction is varied. As </>l increases from 0
we may expect that mechanical instabilities and topological changes will
be triggered as cell faces shrink and disappear. For example in the case of
the Kelvin foam, the four fold faces are seen to pinch apart at </>l ~ 11%,
as shown in the sequence of images in fig. 14.
There are considerable experimental obstacles to recording and analysing
bulk foam structure. They scatter light strongly, particularly when the bub-
bles are small or the liquid fraction high, making ordinary microscopy diffi-
cult. The exact method used to construct the foam and the timescale over
which the observations are made also appear to have a large impact on
what is seen, (Matzke, 1946), (Weaire and Phelan, 1994(b)).
In contrast if the bubbles are large compared to the sample container
then it is remarkably easy to observe a whole range of beautiful and complex
surface induced ordering. Pittet et al. (1995) have observed and catalogued
cylindrical foam structures formed when relatively large monodisperse bub-
bles are collected in an open-ended glass tube. An Evolver model of one of
the simplest such arrangements is shown in fig. 15.

15. Acknowledgments
We are grateful for the opportunity to present this survey at the Cargese
Summer School. The research upon which it is based is supported by For-
bairt (Irish Science and Technology Agency) and by a TMR Fellowship
contract number ERBFMBICT961741 (RP).

References
Almgren, F., and Taylor, J.E., (July 1976), Scientific American, 82-83.
Bolton, F., and Weaire, D., (1991), Phil. Mag. B, 63, 795.
Brakke, K., (1977), The Motion of a Surface by its Mean Curoature, Princeton University
Press, Princeton, NJ.
Brakke, K., (1992), E:r;p. Math., 1, 141.
301

Figure 12. Two examples of more complex monodisperse foams modeled using the
Evolver. Shown on the left is a foam based on the topology of ,13-Uranium and on the
right, one based on the 'C15' structure.

Figure 13. Wet foams. A foam with the f.c.c . topology (left) and a liquid fraction of
a few percent. Note the multiple vertices. Also shown is the WP structure for a liquid
fraction of 1%

Hutzler, S., and Weaire, D., (1995), J. Phys.: Condens. Matter, 7, L657.
Kraynik, A., (1997), this volume.
Langevin, D., (1997), this volume.
Matzke, E.B., (1946), Am. J. Botany, 32, 58.
Phelan, R., (1997), "Generalisations of the Kelvin Problem and Other Minimal Prob-
lems", in The Kelvin Problem, Taylor and Francis Ltd., London.
Phelan, R., Weaire, D., and Brakke, K. , (1995), Exp. Math ., 4, 181.
Pittet, N., Rivier, N., and Weaire, D., (1995), Forma, 10, 65.
Rivier, N., (1994), Phil. Mag. Lett., 69, 297.
Sullivan, J., (1997), this volume.
Taylor, J .E ., (1976), Ann. Math., 103, 489.
Verbist, G., Weaire, D., and Phelan, R., (1997), this volume.
Weaire, D. (ed.), (1997), The Kelvin Problem, Taylor and Francis Ltd., London.
302

Figure 14. The Kelvin structure for a liquid fraction of (left) 1%, (centre) 10%, and
(right) just over 11%. Note the disappearance of the initially four fold faces

Figure 15. The Evolver can also be used to model the geometry of foam in an arbitrary
container. A whole range of ordered structures have been observed experimentally (Pittet
et al. (1995)) in the case of a monodisperse foam collected in a cylinder. One of these is
the so called 422 structure shown above.

Weaire, D ., Hutzler, S., Verbist, G., and Peters, E.A.F .G., (1997), Adv. Chern. Phys.,
accepted for publication.
Weaire, D ., and Kermode, J . P., (1983), Phil. Mag. B, 48, 245.
Weaire, D., and Phelan, R., (1994), Phil. Mag. Lett., 69, 107.
Weaire, D., and Phelan, R., (1994) , Phil. Mag. Lett., TO, 345.
Weaire, D., and Phelan, R., (1996), Phil. 11-ans. R. Soc. Lond. A, 354, 1989.
RHEOLOGY AND DRAINAGE OF LIQUID FOAMS

G. VERBIST*, D. WEAIRE** AND R. PHELAN*


*Shell Research and Technology Centre, P. 0. Box 38000,
1030 BN Amsterdam, The Netherlands.
**Department of Physics, Trinity College, Dublin, Ireland.

Abstract. The rheology of liquid foams is reviewed, concentrating on the


quasi-static regime, in which calculations can be based on accurate simula-
tions. Equilibrium and drainage under gravity are analysed, leading to the
Foam Drainage Equation. This has many interesting solutions, including
that which describes a solitary wave.

1. Introduction

The practical physics of foams includes many problems which involve dy-
namics. As the structure is modified by stress or otherwise, liquid flows in
the films and Plateau borders in ways that we cannot yet describe with full
confidence. These notes are confined to circumstances in which the system
remains very close to equilibrium, apart from sudden topological changes
(bubble rearrangements). The theory can then be based on the description
of equilibrium states. In the case of rheology this type of approximation is
called quasistatic.
The present frontier of research lies in the more subtle dynamic effects
which occur when this approximation is a poor one. It also calls for a
better understanding of the wet limit, since present theory relies heavily on
formulae which are strictly appropriate to the dry limit, just as in the case
of transport (see the following article).

2. Foam: Solid or Liquid?


A foam remains a solid when under low stress, but cannot sustain a stress
greater that some critical yield stress. This is true of all solids but the case
of a foam (or emulsion) is such that it generally undergoes plastic (i.e.
303
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 303-314.
© 1999 Kluwer Academic Publishers.
304

LIQUID

------
~cmodulus

I -- - -,-;--
yield stress
--
bubbles
rearrange,' PLASTIC " bubbles
\ separate

'\I
SOLID
___ ! _______________ _
I
stress

ELASTIC
SOLID

-strain-

Figure 1. Schematic illustration of the stress-strain relations in a liquid foam for in-
creasing increments of strain. Beyond the initial elastic region there are hysteretic effects
which are not indicated here. The dependencies of elastic modulus and yield stress on
liquid fraction are also shown.

irreversible) deformation over an essentially infinite range, while remaining


in much the same state. The bubbles rearrange, with little other effect.
This behavior is like a liquid, so foam presents both solid and liquid-like
responses, depending on the stress, see Figs. 1 and 2.

3. Elastic Moduli
For very low stress/strain, foam is a linear elastic solid, and the conventional
definitions of its elastic properties may be adopted. Usually it is isotropic,
because of disorder, in which case only two elastic moduli are required, the
bulk modulus K and the shear modulus G. In general K » G, and both
fluid components can be regarded as incompressible.

4. Bulk Modulus of a Foam


If we allow a finite compressibility for the gas component, the bulk modulus
may be written

K = Kgas + Ksurface (1)

The contribution of the gas is simply the usual bulk modulus, equal to
the pressure p in the ideal gas model, for isothermal processes, and much
greater than the other term.
305

Only in very extreme circumstances (small bubble size, low pressure)


can the surface term outweigh the other. If ever achieved, this regime would
introduce an interesting instability into the subject, but so far it has only
been observed in computation (Aref and Herdtle, 1990).

5. Shearing a Disordered Foam


In describing this we concentrate on a typical disordered foam. For examples
of the response of ordered foams in simulations see the chapter by Kraynik.
Firstly, the internal displacements reduce the shear modulus G of a dry
foam greatly, with respect to an estimate based on affine deformation. This
may be inferred from:
- Exact calculation for the honeycomb in 2D
- The estimate of Stamenovic, (1991), in 3D (not exact, but derivable in
various ways and successful in practice)
- Various simulations
As stress in increased, the internal displacements take the system to the
point at which significant numbers of topological changes begin to occur so
that there is a transition from an elastic response to a plastic one. Figures
1 and 2 illustrate these different regimes, see Weaire and Fortes, (1994), for
a complete discussion.

6. Foam Drainage
Turning to the problem of drainage, we may ask: how does the local flow of
liquid in the Plateau border network relate to the pressure difference (and
gravitational force) which drives it?
The simplest reasonable assumption is that of fixed boundary conditions
on the walls of the Plateau borders, and Poiseuille flow within them. If
we also neglect any change of structure due to liquid flow, the resulting
theory addresses the flow through a network of pipes. This has much in
common with porous medium theory, with one important difference: the
pipes expand or contract in order to remain in equilibrium under the lateral
pressure forces. The system "breathes" in this way and the result is a
nonlinear partial differential equation, with subtle and interesting solutions.

7. Foam Drainage Equation


In developing an equation that describes the flow of liquid through foam,
we assume that the liquid flows only through the network of Plateau bor-
ders (the cells edges) while the underlying structure of the foam remains
unchanged. That is to say: we consider an ideal foam as defined in the
306

Figure 2. Elongational shear of a two-dimensional dry foam. Increasing shear results in


topological changes and a plastic response.

preceding chapter. The local wetness of a foam (or its liquid fraction ci>t)
is then essentially given by the cross section of the Plateau borders A, to
which it is proportional to lowest order.
If the gas pressure in the different bubbles is treated as constant, the
307

R= ~ .....

\/
i

bubble
pg

Figure 3. Relation between pressure and wall curvature for a Plateau border.

shape of the Plateau border can be described as triangular with circular


sides whose radius of curvature R is given by Laplace's law due to the
pressure difference between liquid (border) and gas (bubbles), see Fig. 3.
J
Note that A = C 2 R 2 where the constant C = J3- 1f /2 accounts for the
Plateau border shape. As a result the liquid pressure PP. can be expressed
as
'Y c, (2)
PP. = Pg - R = Pg - VA '
where 1 denotes the surface tension and p 9 is the gas pressure.
For simplicity we focus now on a single vertical Plateau border. The
forces acting on the liquid are gravity and the Laplace-press ure gradient.
Neglecting inertial effects, we can consider the liquid flow through the bor-
der as Poiseuille flow. The average liquid velocity will therefore be pro-
portional to the applied force (gravity + pressure gradient) and the pipe
cross section A. The constant of proportionalit y will depend inversely on
the liquid viscosity ru :

u = ~ (pg - ope)
ox
' (3)
!'TJP.

where f ~ 50 is a factor which depends on the shape of the border (Leonard


and Lemlich, 1965) (Peters, 1995). In a circular pipe this factor would be
81r ~ 25 much lower than for the concave border shown above in which
dissipation is indeed more important.
The foam drainage equation is finally obtained by inserting the above
expressions for u and PP. in the equation of continuity

0 0
ot A(x, t) + OX (A(x, t)u(x, t)) = 0 (4)
308

and transforming into convenient units (x = exo, t = Tto and A = ax5)


given by a capillary-type length scale xo = J G-y / pg and a "viscous" time
scale 1 to = 3fru/ JG-ypg,

(5)

The physical origin of the different terms is still apparent in the form of
the scaled equation: the time (T) derivative denotes the variation of the
local liquid fraction in time which is balanced by the variation of the flow
(term in brackets), one part of which is due to gravity (a2 ) and the other
due to the pressure gradient as it includes aafae. It is worthwhile to note
that all dimensions could be scaled out: no dependence remains on dimen-
sionless groups such as a Reynolds number for the Navier-Stokes equation.
Mathematically the equation belongs to the class of non-linear partial dif-
ferential equations which generalise the (non-linear) diffusion equations.
Surprisingly enough analytical solutions have been obtained and will be
discussed in the following section. For a survey of the solutions to the foam
drainage equation we refer to a recent review, (Verbist et al., 1996).

8. Some Solutions and Experimental Verification

The starting point for our investigation of foam drainage has been the
steady-state drainage experiment (Weaire et al., 1993) in which a stable
(dry) foam was wetted at its top by a steady addition of liquid. This type
of experiment is quite different from the classical experiment in which foam
is created by sparging gas bubbles through a liquid solution and recording
the foam height (at constant gas flow) and its decay in time (after switch-
ing off the gas flow). The merit of the steady-drainage experiment is that
it singles out the drainage process whereas in the classical experiment it
is combined with film breakage, bubble rise and coalescence and possibly
coarsening. Experimentally the local liquid fraction can be monitored with
a variety of techniques including tomography, MRI (or NMR) imaging and
capacitive or conductive probing. We will show results obtained with the
latter technique. For a more elaborate review we refer to Weaire et al.,
(1997). There are many previous contributions to the subject, both exper-
imental and theoretical, and the cited review attempts to put them all in
perspective and give due credit.

1 In a real foam not all Plateau borders are vertical but to a good approximation we

may consider their orientation to be isotropic. The factor 3 in to appears because of an


orientational averaging for an isotropic border distribution, see Verbist et al., (1996).
309

.. .
40
0. 1
E'
~ 35

§ .
0.08
~!>£ 30
0.06 't 25
>
0.1)4 "';.!; 20
c
0
0.02 ·;::
·;;;
0
()0 0. 5
10 20 10 40 10 20 30 40

segment# [~em] time [seconds]

Figure 4. In the left panel, snapshots of the drainage profiles of solitary-wave type are
plotted. After fitting these profiles with Eq. (7), the position of the front is plotted in
the right panel as a function of time demonstrating the constant velocity of the wave.

8.1. STATIC EQUILIBRIUM

The equilibrium solution aeq can be obtained by imposing a no-flow con-


dition and it leads to
-1/2 )-2
aeq (~ ) = (al +6 - ~ {6)
which is the familiar capillary-rise curve as discussed by Princen and Kiss,
{1987) . The integration constant a1 is the value at~= 6 .

8.2. STEADY DRAINAGE

The foam drainage equation has the simple solution a = constant which
leads to steady-state flow . The flow rate (a2 ) is then proportional to the
square of the liquid fraction {since ll>e ex a), exactly the dependence origi-
nally found experimentally by Weaire et al., (1993) .

8.3. SOLITARY WAVE

By imposing a solution of the form a= a(~ -VT), i.e., one that depends only
on the "running" coordinate ~ - VT , the form of the solution is preserved
in time; it is only displaced along the ~-axis by an amount v6.T during a
time 6.T . Solutions of this type are called solitary waves with a constant
velocity v. For boundary conditions corresponding to a dry foam that is
wetted from above with a constant flow rate, a solution can be constructed
of the form
(7)
310

2000

:>-.
--+-)
·~ 1500
rn
~
(l)
--+-)

~ 1000
........
......-l
cO
~
•rl
Ql) 500
rn

0
30 40 50

segment number

Figure 5. Snapshots are plotted for a fast (high flow) wave overtaking a slow (low flow)
wave. The dotted curves represent the slow wave, dash-dotted curves are the final fast
wave and full curves are drawn at intermediate times during which the coalescence takes
place. The catch-up velocity v is seen to be the sum of the original solitary wave velocities
Va and Vb ·

Fig. 4 shows snapshots of the liquid fraction as a function of vertical


position at different times. The liquid fraction was obtained by conductance
measurement in this experiment. The curves through the data points are
taken from Eq. (7) with equal velocity v for each snapshot as is apparent
in the left panel.
It is possible to find other solitary-wave solutions to the foam drainage
equation, e.g., by considering the constant wetting of an already wet foam
(rather than the dry one considered above). In that case the solution can
only be obtained in implicit form, (Verbist et al., 1996). The experimen-
tal profiles which are snapshots of such a double wave experiment, i.e.,
a solitary wave corresponding to a high flow rate catching up with and
overtaking a low flow rate (and hence slower) wave, are shown in Fig. 5.
They were obtained by capacitance probing. From the implicit solution it
is possible to predict that the catch-up velocity should be the sum of the
fast and slow solitary wave velocities. Although the plotted data agree well
311

with this prediction the situation may be more complicated for higher flow
rates as suggested by recent conductivity data reported by Peters, (1995).

0 .2

0 .15 ( 11 profiles with fl"T=lOO)


c:!
ro
Q)
~ 0 .1
ro
a:!
p.,.
0 .05

0
0 20 40 60 80 100

(downward) vertical distance ~

Figure 6. Numerical simulation of free drainage: an initially homogeneously wet foam


(profile at r = 0) is allowed to drain with a top boundary condition corresponding to
zero flow, while the lower boundary is kept at constant liquid fraction .

8.4. FREE DRAINAGE


Suppose a foam sample is prepared while a constant flow of liquid is added
from above. The foam will then be homogeneously wetted such that its
liquid fraction is constant along the vertical direction. If we now suddenly
switch off the the liquid flow and let the foam drain freely, the observed
process is called free drainage. Fig. 6 shows a numerical solution to the foam
drainage equation under these conditions. We find that for a large portion
of intermediate times the curves are linear. An approximate analysis, orig-
inally due to Kraynik, (1983), neglects the pressure-gradient contribution
to the flow and predicts linear profiles of the form

a(~,T) = ~-~-, (8)


2T +Ta
where Ta is an integration constant. This approximation is appropriate for
the linear parts in Fig. 6. It is also an excellent approximation for the tail of
a drainage pulse, the subject of the following section. Experimental results
obtained using a conductive segmented probe (Fig. 6) reveal that a major
312

E
,.....,.....,1
- 0 0.00.1

g.
C1)
Q./)()2

-;;;
0.001

O L--4~
0 --~
W~~80~~~~QQ---1~
20

segment# [::::::em] time [seconds J

Figure 7. Free drainage profiles are shown on the left panel. The region between the
vertical lines was chosen for linear fitting. In the right panel the fitted slopes are plotted
versus time. The solid curve is a hyperbolic fit based on Eq. (8) .

part of the profile can be fitted well with the linear form but deviations
clearly show at the bottom where the foam column rests in the liquid. Note
that in this region the foam drainage equation itself becomes questionable
because of the underlying assumption of a dry foam structure. Numerical
fitting of the profiles in the linear regime allow us to determine how the
slope varies with time. Such an analysis has been reported by Peters, (1995) ,
and Weaire et al., (1997), showing excellent agreement with the expected
hyperbolic form suggested by Eq. (8), see Fig. 7.

8.5. PULSED DRAINAGE

It is possible to combine the conditions of a solitary-wave front with a free-


drainage tail in a pulsed drainage experiment (or simulation) as follows .
Start off with a dry foam and wet it from above with a constant flow
of liquid as in a solitary-wave experiment. Then, before the front reaches
the bottom of the foam column, switch off the flow as in a free-drainage
experiment. Based on the analysis presented above one would expect a
sharp "solitary-wave" front and a linear "free-drainage" tail leading to a
triangular-shaped pulse. The linear tail will extend to the top of the column,
but since liquid is no longer being added, volume conservation dictates that
the height of the triangle will be reduced in time. Eventually the triangular
form will be rounded off since a decreasing height corresponds to a lower
"solitary-wave" flow rate which has a larger front width. Adopting volume
conservation and a Kraynik-type description, Eq. (8), for the tail, suggests a
temporal variation of the front height <P~top) ex (T+Ta)- 112 and of the front
313

0.08

0.06

0.04

0.02

0 0 10 20 30 40
segment# [~em]

Figure 8. Snapshots of drainage profiles as a pulse propagates through a foam.

position ~(top) ex (T + Ta) 112 . Fig. 8 shows the profiles of a pulsed-drainage


experiment obtained by conductive probing. See Weaire et al., (1997) for
an extensive discussion of these experiments and their analysis which show
good agreement with the predicted time dependence.

8.6. QUANTITATIVE ANALYSIS

In the discussion above we have merely indicated the form of different


solutions of the foam drainage equation and have shown experimental data
which confirm these solutions. It also possible to make a more quantitative
statement by comparing the values of the fitting constants obtained in
matching various types of drainage experiments. It is found that there is a
close relation between the different constants in the solitary-wave, free and
pulsed drainage experiments giving us more confidence in both the foam
drainage equation itself and the simple assumptions on which it is based.
For a detailed discussion of these matters and a more extensive treatment
of the subject we refer to a review, (Weaire et al., 1997).
Often we have indicated that a major assumption lies in the fact that
the foam is rather dry resulting in Plateau borders of constant cross section.
A first correction might take into account that the Plateau borders swell
at the tetrahedral junctions where they meet. A similar correction for the
volume and the conductance has been obtained recently, (Phelan et al.,
314

1996), and is quite successful in describing the conductivity versus liquid-


fraction curve up to 14% while the original Lemlich model (constant cross
section) already becomes questionable above 3%, see also the following
chapter.
Another approximation made is that the flow through a Plateau border
is of Poiseuille type with a no-slip condition at the gas/liquid boundary.
It would be worthwhile to introduce the effect of surface viscosity, which
corresponds to a finite slip at the boundary. A partial indication that the
latter correction is small might already be inferred from drainage experi-
ments using liquids of different (bulk) viscosities and which compare well
with the simple theory, see the recent work in Stoyanov et al., 1998.

9. Acknowledgments
We are grateful for the opportunity to present this survey at the Cargese
Summer School. The research upon which it is based is supported by For-
bairt (Irish Science and Technology Agency) and by a TMR Fellowship
contract number ERBFMBICT961741 (RP).

References
Aref, H., and Herdtle, T., (1990), Topological Fluid Mechanics, (ed. H. K. Moffatt and
A. Tsinober), Cambridge University Press, p. 745.
Kraynik, A.M., (1983), 'Foam Drainage', Sandia Report SAND 83-0844.
Leonard, R.A., and Lemlich, R., (1965), Amer. Inst. Chem. Eng. J., 11, 18.
Peters, E.A.F.J., (1995), 'Theoretical and Experimental Contributions to the Under-
standing of Foam Drainage', MSc Thesis, University of Technology Eindhoven (the
Netherlands).
Phelan, R., Weaire, D., Peters, E.A.J.F., and Verbist, G., (1996), J. Phys. Condens.
Matter, 34, 1475.
Princen, H.M., and Kiss, A.D., (1987), Langmuir, 3, 36.
Stamenovic, D., (1991), J. Colloid Interface Sci., 145, 255.
Stoyanov, 8., Dushkin, C., Langevin, D., Weaire, D., and Verbist, G., (1998), 'The Effect
of Surface Rheology on Foam Drainage', to be published.
Verbist, G., Weaire, D., and Kraynik, A.M., (1996), 'The Foam Drainage Equation', J.
Phys. Condens. Matter, 8, 3715.
Weaire, D., Hutzler, 8., Verbist, G., and Peters, E.A.J.F., (1997), 'A Review of Foam
Drainage', Adv. Chem. Phys., 102, 315.
Weaire, D., Pittet, N., Hutzler, 8., and Pardal, D., (1993), Phys. Rev. Lett., 11, 2670.
Weaire, D., and Fortes, M.A., (1994), Advances in Physics, 43, 685.
ELECTRICAL AND THERMAL TRANSPORT IN FOAMS

R. PHELAN*, G. VERBIST* AND D. WEAIRE**


*Shell Research and Technology Centre, P. 0. Box 38000,
1030 BN Amsterdam, The Netherlands.
**Department of Physics, Trinity College, Dublin, Ireland.

Abstract. The Lemlich theory of electrical conductivity of a foam is re-


viewed, together with its nonlinear extension, which is very successful. This
confirms that Plateau border conduction dominates in a typical liquid foam.
Conductivity is therefore a good measure of liquid fraction. The same the-
ory can be applied to solid foams to estimate the contribution of solid
conductance to thermal conductivity, an important practical problem.

1. Introduction
These notes address the transport of electrical and thermal current through
a foam. In the former case we usually have in mind a liquid foam and in
the latter case a solid foam, such as polystyrene or polyurethane, because
of the practical motivation of such research. In particular, the thermal con-
ductance of plastic foams is a critical parameter in setting standards for a
major industry. The challenge being the optimisation of insulation panels.

2. Electrical Conductance
Consider a liquid foam, with the structure described in the previous paper,
that is, a network of Plateau borders, spanned by thin films. What is its
electrical conductance u, if that of its liquid component is u 1?
The geometry of the foam structure is such that scaling arguments sug-
gest a very simple formula for the dimensionless quantity u / u 1 for a rel-
atively dry foam. It should simply be proportional to the liquid fraction.
More detailed geometrical arguments lead to:
u 1
- = -¢l Lemlich Law (1)
0'! 3
315
J. F. Sadoc and N. Rivier (eels.), Foams and Emulsions, 315-322.
© 1999 Kluwer Academic Publishers.
316

where ¢1 is the liquid fraction. Its derivation depends on the following


assumptions and approximations.
The main assumption is that conduction takes place through the net-
work of Plateau borders only. In this case we may consider the contribution
to the current from all Plateau borders crossing a given plane, taking the
variation of potential in each to be simply (V cos 0) where 0 is the angle
between the direction of the Plateau border and the normal to the plane.
The correct average (which is not entirely obvious!) brings in a factor of
i.
< cos2 0 >, the spherical average of cos2 0 and this is This assumes an
isotropic conductivity.
In practice this estimate, which looks crude, is a very good one. This
has to do with the fact that Plateau borders are almost straight (fig. 1),
and meet in symmetric tetrahedral junctions. In fact the estimate is close to
being exact for a wide range of reasonable foam structures. For example, the
Kelvin structure has a/ a1 = 0.3313¢1, as an elementary exact calculation
will show. (It requires the precise length of each Plateau border, which is
given by the Evolver) .
This linear relation is based on the formula for the liquid fraction in the
dry limit
(2)

where Lv is the line length per unit volume, i.e. the sum of all the border
lengths per unit volume, and A is the cross-sectional area of a uniform,
straight Plateau border. This has the geometry shown in fig. 1 where we
have also defined 6, the radius of curvature of the cross-section or equiv-
alently the 'width' of the border. It is easy to show that A = c9 62 where
c9 = V3 - ~ is a geometrical constant. These definitions will be used in
what follows.

''

Figure 1. A uniform, straight Plateau border. The cross-section consists of circular arcs
of radius a as shown.
317

The treatment outlined above neglects extra liquid stored at the junc-
tions where the Plateau borders meet, fig. 2. These junctions effectively
shorten the Plateau borders as well, in the sense that a "short cut" is avail-
able for the flow of current between the borders. These two effects combine
to make the conductivity of the network nonlinear in </Jz. One might well
expect that the linear formula would have a very narrow regime of useful-
ness but the nonlinearities largely cancel, so that the Lemlich Law holds
well up to liquid fractions of a few percent.

Figure 2. A single Plateau border junction from a wet foam structure. Note the curvature
of the edges and the thickening of the junction itself. Calculating the conductance of such
a vertex allows us to extend Lemlich's analysis and gives a nonlinear relationship between
u / u1 and 1>1-

For higher liquid fractions, the experimental results (some of them now
quite old) show a large departure from the law. But is this the effect of the
nonlinear corrections to the theory based on Plateau borders or is it the
contribution of the films, or indeed some other effect?
Happily calculations have now shown that the Plateau border theory is
sufficient to describe the observed nonlinear variations very well, leading us
to dismiss the film contribution.

3. Calculation of Nonlinearities
The calculation which demonstrates this is performed on a single junction,
simulated by the Evolver, fig. 2. Knowing the exact geometry of the junction
we can calculate its conductance. With some reasonable approximations,
this can be used to extend the Lemlich theory. The best representation of
the results is the following parametric dependence of a/ az on <Pt :
318

(3)

{4)
The quantities Lv and c9 were defined previously and L is the average bor-
der length. The parameter 8 has been generalised as follows. As mentioned
in the previous article {Weaire et al., 1997), Laplace's law relates the mean
curvature, H, of an interface to the pressure drop across it. For the surface
shown in fig. 2 we define 8 to be 1/{2H) which has a direct relation to
the liquid pressure in the border. In the dry foam limit where the borders
have a uniform cross section this definition reduces to one used before. It
is worth pointing out the these relations are completely determined by the
geometrical constants Lv, L, and c9 , and by the two calculated corrections
due to the presence of thickened vertices, 1.27 in the case of conductance
and 1.50 in the case of liquid fraction.
Figure 3 shows some typical experimental data {Peters, 1995) for com-
parison with the nonlinear theory. Similar agreement is found with other
measurements {Clarke, 1946) (Curtayne, 1996).

4. Measuring cr (and hence ¢L)


The consistency of various experiments with each other and the extended
Lemlich theory implies that relative conductance is a good measure of liquid
fraction.
This has motivated recent work devoted to conductance measurements,
particularly with segmented electrodes which detect the local conductance
in a vertical column. The apparatus used is shown in Fig. 4.

5. Thermal Conductivity
The conduction of heat rather than electricity through a foam presents a
more complex challenge to theory. Correspondingly its minimisation (to
create good thermal insulators) can be a subtle problem.
It is conventional to consider thermal conductivity A. to be the sum of
various independent contributions.

A.* =A.*+
s A.*+
g A.*+
c A.*r {5)
This follows the notation of Gibson and Ashby's book {1988), where
representative data may be found. The contributions are: solid conduction
{A.;), gas conduction {A.;), gas convection {A.~) and radiation {A.;).
319

0 .10

.c
...... 0.08
......
;>
.....
u
.g
c 0.06
0
u
<!)
;>
·g 0.04 ....
--<
<!)
~ Lemlich equation
0.02

0.00 Lo::::::......:..-..&...-- -'------1.------'"-- - " " ' - - - - - ' - - - 1


0.00 0.02 0.04 0.06 0.08 0.10 0.12

Liquid Fraction, <1>1

Figure 3. Compaxison of experimental conductivity measurements and the nonlineax


theory described here. The experimental measurements ( +) axe from foam samples con-
sisting of 3mm bubbles in a perspex cylinder of diameter 2cm and height 70cm. The

dashed line is the Lemlich limit u =


solution used was ordinaxy tap water and a non-ionic surfactant (dobanol). The lower
uJiuz = ¢z/3. The two upper curves show the
relationship based on equations (3) and (4) which include the vertex corrections. For
comparison we have plotted the relationship based on L., and L corresponding to both
the Kelvin and WP structures

At first sight this is puzzling: why should these effects be additive in the
conductivity? Indeed they are not, strictly speaking, but closer analysis will
show that there are many circumstances in which this is not unreasonable.
To some extent, this will emerge in what follows.
In many practical foams, >.~ is negligible, the cell size being small enough
to suppress convection, so that only three terms are necessary. At steady
state we may consider the heat flux due to radiation, qradiation and that
due to conduction, qconduction, to be independent so that that the total flux
q is given by.
q = qradiation + qconduction (6)
In such foams we are justified in expressing the total thermal conductivity
320

Conductance
Profile
Multiplexer

Figure 4- The segmented conductivity apparatus developed at Shell for use in the
monitoring of foaming processes. In the example shown here the cylinder contains a
sample of foam. A liquid pulse has been added and is traveling vertically downward.
The dark region indicates the more dense, high liquid fraction foam . A schematic of the
corresponding conductance profile is also shown.

as a sum of a radiative and a conductive term:

). * = A~onduction + >.;adiation {7)

If in addition, the volume fraction of the solid material, is low (as it


</J8 ,
is for a foam) we may expand in that small parameter and hence separate
gas and solid conductivity to arrive at equation {5) (without the convective
term).
321

The largest contribution is usually A;,


the conduction though the gas.
If the cell faces of the foam are thin (with negligible thermal resistance),
then it is clear that A;
is simply the thermal conductance of the gas present
in the foam, in the absence of the solid component and is given by:

{8)

where Ag is the gas conductivity and ¢9 is the volume fraction of the mate-
rial occupied by the gas. It is not surprising that in the past most effort has
gone into developing manufacturing processes based on low conductivity
gases, CFC's for example. Changes in environmental legislation have re-
sulted in the use of less damaging blowing agents but at the cost of higher
conductivities. Hence the need for a good understanding of all the contri-
butions.
The radiative term is quite difficult to measure and depends strongly
on temperature. The solid contribution however should have the same form
as the electrical conductivity described earlier.
The case of a rigid polyurethane foam satisfies all of the conditions
mentioned above. The cell size is small ( < lmm) so convection can bene-
glected. The solid volume fraction is low (a few percent typically) and the
structure of the solid veins is essentially that of a liquid foam Plateau bor-
der network. Figure 5 shows a confocal laser scanning microscopy (CLSM)
image of an open-cell PU foam. The corresponding closed-cell foam, of the
type we are considering here, shows the same vein structure. Initial esti-
mates of the solid component of conductivity for such a material agree well
with the theory outlined above {Fleurent et al., 1997).

6. Acknowledgments
We are grateful for the opportunity to present this survey at the Cargese
Summer School. The research upon which it is based is supported by For-
bairt {Irish Science and Technology Agency) and by a TMR Fellowship
contract number ERBFMBICT961741 {RP).

References
Clarke, N.O., (1946), TI-ans. Faraday Soc., 44, 13.
Curtayne, P., (1996), Stagiaire Report, Shell Research and Technology Centre Amster-
dam.
Fleurent, H., Verbist, G., Phelan, R., and Weaire, D., (1997), unpublished.
Gibson, L.J., and Ashby, M.F., (1988), Cellular Solids, Pergamon Press, Oxford.
Lemlich, R., (1977), J. Colloid Interface Sci., 64, 107.
Peters, E.A.J.F., (1995), MSc Thesis, Eindhoven University of Technology, Eindhoven.
Phelan, R., Curtayne, P., Weaire, D., Hutzler, S., Verbist, G., and Peters, E.A.J.F., (1998)
to be published.
322

Figure 5. A CLSM image of a low density open-cell polyurethane foam.

Phelan, R., Weaire, D., Peters, E.A.J.F., and Verbist, G., (1996), J. Phys. Condens.
Matter, 34, L475.
Weaire, D., Phelan, R., and Verbist, G., (1997), this volume.
DECONTAMINATION OF NUCLEAR COMPONENTS THROUGH THE USE
OF FOAMS

G. BOISSONNET, M. FAURY, B. FOURNEL


Commissariat al'Energie Atomique, DCCIDESD/SEPILETD
CEA Cadarache 13108 St Paul Lez Durance, FRANCE

KEYWORDS/ABSTRACT: decontamination I nuclear waste I foam I chemical reagents


I gas-cooler I image analysis I bubble size distribution I rheology I Ostwald de Waele fluid

During the decommissioning of a nuclear facility, most of the material requires disposal and
may therefore warrant decontamination. Decontamination is an important procedure
because it not only reduces radiation during decommissioning operations, but also has the
potential for the decategorization of waste to achieve lower disposal costs.
A few techniques are available to decontaminate large internally contaminated components
with complex shapes. The main advantage of the foam application process is that it only
requires small amounts of liquid. Thus, the volume of secondary waste produced can be
significantly reduced. This process was successfully validated during the decontamination of
a graphite-gas cooler representing a developed area of 1000m2. Since then, there has been
an increasing number of studies concerning new foam formulation adapted to maintenance
or dismantling purposes. For a better mastery of this process, comprehensive studies have
begun. They consist of the characterization of decontamination foams, using image analysis
and of an approach of the rheological behaviour of such a fluid

1. Introduction

1.1. NUCLEAR DECONTAMINATION

Decontamination of solid waste consists in transferring the radioactivity of a material


to another medium which is easier to manage. It is based on the retrieval of a surfacic
layer containing most of the activity (grease, paint or oxide layer). Decontamination is
performed for two main reasons. First, to reduce the dose rate to the workers during
dismantling and maintenance operations. Second, to enable a change in the
classification of nuclear waste. Thus, waste initially destined for a geological disposal
can be transferred to a surface storage roughly 20 times less expensive.
Decontamination can be performed using various means: physical, electrochemical and
chemical. Chemical decontamination consists of conducting a degreasing or an erosive
323
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 323-334.
© 1999 Kluwer Academic Publishers.
324
treatment on the material to be cleaned. Chemical reagents can be supplied through
various means depending on the geometry and the size of the nuclear component. For
example:
- liquid solutions for the decontamination of tubes,
- gel spraying for the decontamination of walls,
- foam filling for the internal decontamination of large facilities.

1.2. FOAMS FOR DECONTAMINATION PURPOSES

A few techniques are available for decontaminating large internally contaminated


facility with complex shapes such as valves, reservoirs, heat exchangers, turbines,
vessels, boilers. Spraying does not reach all internal surfaces evenly, due to preferential
flow patterns. Immersion systems are not feasible. First, because the volume of
reactants would be too high, second because reactants are ineffective on large volumes
due to boundary layer phenomena. Mechanical methods are inoperative for it is
impossible to reach all the internal surfaces. Moreover, the retrieval of the abrasion
residue would be difficult. However, the foam application process provides good
solutions to all these problems.
The foam process has the major advantage of using only small amounts of liquid.
Thus, the volume of secondary waste can be significantly reduced. Moreover, foams are
able to forcefully penetrate everywhere, reaching all the interstitial gaps, acting on all
contaminated surfaces. Furthermore, the effectiveness of the process is enhanced due to
the dynamic motion of the foam and its permanent regeneration on the component
walls. Common problems resulting from boundary layers should thus be minimized.
And finally, the recirculation of the foam and the continuous filtration of the stock
solution allow the treatment to be pursued until decontamination is achieved.

2. Industrial application of the foam process.

The foam process was successfully validated on an industrial scale during the
decontamination of a graphite gas-cooler in 1993. This work was conducted within the
context of a CEE program [1].
The characteristics of the gas-cooler were the following :
-total volume : 18m3 ; contaminated volume: 13m3
-contaminated surface : 1000m2
- main contaminated materials : carbon steel cooler body (90m\ 1950 admiralty brass
tubes (873m2 )
- main contamination : 6°Co, 137 Cs.
The decontamination device presented in Figure I consisted of two tanks (COl and
C02) and three units : one foam generation unit and two filtering units. The process
was bas~d on the filling of the equipment with liquid foams containing suitable
reagents.
Clp
.,.
I <
' ~ ~/ Cap
SECTION OF
THE GAS COOLER

AIR
-
MIXer
c 02

\-- · - · - · - · - · - · - · - · - · - · -
I Dripping pan I I Dripping pan II Dripping pan I ' Dripping pan I
w
N
Figure I. Foam decontamination device- Industrial demonstration VI
326

The foam exiting from the top of the equipment as well as the liquid collected from
drainage were filtered and recycled though a recycling loop.
The foam process was applied in the three steps described in Table I : degreasing,
descaling and rinsing steps. The foam formulation was taken from a CEA patent [2].
The advantage of this formulation is that it can be used either in an acid or alkaline
environment (up to 5 mol/1). During this industrial demonstration, proof was made of
the feasibility and the effectiveness of the foam process.
The circulation of the foam was mastered, the quality of the foam was excellent and
grease particles and large copper chips (about 10 cm2) were thus removed. Moreover,
the decontamination operation only produced 6m3 of eftluents i.e. 6.2 Vm2 of treated
component. Decontaminating an identical cooler using a chemical solution would have
produced approximately l 0 times more eftluents for a similar result. A
decontamination factor (ratio between the initial and final activity) up to 190 was
obtained leading to a residual activity below l Bq/cm2 • This decontamination therefore
established that, for geometrically complex equipments, it was possible to limit secondaiy
waste while maintaining (and even improving) adequate efficiency standards.
Since this achievement, numerous studies have been undertaken to develop new foam
formulations adapted to maintenance or dismantling purposes. There are obviously
many future possible applications for the foam process.
For a better mastery of this process, comprehensive studies are necessary. These studies
rely on the characterization of decontamination foams through the use of image
analysis and on a better comprehension of the rheological behaviour of such a fluid.
The final aim of this work is to predict the flow properties of foams in facilities of
various geometries.

Table 1. Foam process & operating conditions

Formulation Treatment Remarks Operating conditions


I

Degreasing [NaOH]: 2M lh graphite particles Foam rate: 50 to


removed by foam 70 m 3/h
Descaling [H2S04) : 1.5M 5h copper chips Expansion factor: 14
JHN03) : 0.09M removed by foam
Rinsing buffer solution 2h

3. Structure and Stability of Decontamination Foam

3.1. AIM OF THE STUDY

The aim of this study is to characterize the liquid film in contact with the wall since
this is where decontamination occurs. Bubble size distribution and interfacial areas are
of great interest considering that film characteristics are linked to foam structure.
Drainage is also studied because it directly influences film thickness.
327
3.2. EXPERIMENTAL DEVICE

The experimental device (Figure2) consists of a CCD camera, linked to a numerizing


card. Foam flows upstream in a I m high, 30 liter, octogonal column.

FOAM OUTLET

Reagents Alcohol
Inlet Aspersion

Air Inlet

Foam

CFOAMINLET
Solution Inlet
Figure 2. Experimental device

-5 -5 3 -1 .
Flowrates are low (3.9xl0 to 9xl0 m s at atmosphenc pressure) and residence
times are high ( 10 to 20 min.) in order to favour drainage. 7 photographs per second
can be taken. The main parameters are termed as follows :
- '-<': 1qm"d fl owrate (m3s-1 ), Q80 :gas flowrate measured at atmosphenc. pressure (m3s-1 )
r\. • .

°
- Qm0 = Q8 + Q1: foam flowrate measured at atmospheric pressure (m3s"\
0
0 Ql +Qg
- F = : expansion ratio at atmospheric pressure
QJ
- The parameters of the study are : foam generator geometry, Qm0, F0, observation
height in column.
- The 2D measurements made by image analysis are : bubble size distribution, mean
equivalent diameter, specific interfacial area and 2D expansion ratio corresponding to
the ratio betwen total area and liquid film area.
- The 3D measurements are : 3D expansion ratio (sampling) and global drainage
flowrate.
328

3.3. FOAM STRUCTURE STUDY USING IMAGE ANALYSIS

Several methods are used to study the foam structure. Biswal et al. [3] measured the size of
foam bubbles flowing separately through a capillary tube. Bisperink et al. [4] used an optical
fiber moving through a beer foam blanket. The interaction between the medium and the
measuring device is a limit of these invasive methods. Durian et al. [5) used a non-invasive
method based on light transmission in foam. This type of method was also used by Lachaise
et al. [6) on foam samples.
Photographic devices have been used since the middle of the 80's. Kroezen et al. [7]
used a photographic device in order to measure bubble size distribution in a rotational
rheometer. Calvert et al. [8] and Gido et al. [9] used it to study the structure of a
flowing foam. This device has the major advantage of preventing interaction with foam
and does not influence foam structure. Recent progress in electronics and especially in
CCD sensors and numerization cards allows a CCD camera coupled with an
appropriate computer to deliver 25 numerized images per second with a high
resolution.
Some limiting aspects of this measurement method must be mentionned.
Measurements are always made on the wall which induces obvious drifts. For internal
pressure reasons, little bubbles wedge the bigger ones in the bulk, which causes a
diameter discrimination (Cheng et al. [10]). Furthermore, wettability makes the
interbubble liquid film larger at the wall than in the bulk foam, as illustrated in Figure
3. Then, the measured diameter is reduced compared to the actual diameter and the 2D
expansion ratio is underestimated.

Figure 3. Liquid film at the wall

Nevertheless, the drifts only have a minor impact on surface decontamination and the
use of the CCD camara remains relevant.

3.3.1. Image processing


The CCD camera gives coulored images (presented in 256 grey levels, in Figure 4a).
The role of the image treatment is to isolate the significant information contained in an
image, such as bubble outline, by increasing image contrast thanks to linear and non-
linear filters. After thresholding, a binary image which contains too much noise to be
analyzed (Figure 4b) is obtained. Image cleaning and bubble filling are necessary, after
which (Figure 4c) measurements are possible.
329

100 18
90 16
ro 14
~ 70

j
I 12 ~
~
·i
10 ~

~ 40 8

]~
6 t.t.

4
2
0 0
IQ-5 !()-' to->
Equivalert diamer (m)
(d)
Figure 4. (a) Original image (b) Processed and thresholded image (c) Labelled image ready for measurements
(d) Bubble size distribution

3.3.2. Experimental results


Bubble Size Distribution. The bubble size distribution represented in Figure 4d is
based on I 0 images corresponding to almost 700 bubbles which do not touch image
edges. Several distributions have been built for different operating conditions.
Distribution shapes are identical whatever the operating conditions.
In a first approximation, bubble size distribution seems to be log-normal, which is in
agreement with other studies using other measurement methods or under other
operating conditions [8, 6]. This curve could also be fitted by an exponential
distribution such as :
R = A r exp(-sr 2 ) (1)

where R is the cumulative fraction of bubbles with a radius smaller than r. Some
bibliographical research is in progress to answer the question of bubble size
distribution shape.
Mean diameter. Studying the bubble mean diameter, we observed that it does not
vary much with operating conditions. Then we made the following assumption : wall
effects could be more powerful on the external structure of the foam than bulk effects,
such as foam generation or flowrates.
330
Expanstion Ratio - First Relationship with Drainage and Stability. In Figure 5, 3D
expansion ratios measured by sampling and 20 expansion ratios measured at the wall
by image analysis are presented.

80~---~
n----~----.-----rr---------~
'"' [\_ bottom --o- F30,
70 +---~'-<:-"'---+----+-----H-- F30, middle
""'- --<>-- F30 , top
60 +----1---~'\.. ------+------H .. a .. Fw, h=O.l m
Ll... 50 ""., ·· •·· Fw, h=0.4 m
-~ • ..._ .._______ ·· <> ·· Fw. h=0.7 m
~4ot---~~~--~--~--~~~~==~
-~§.. 30 +----+~
------'"rl--+----+---------il---or--------"!
",
~ 20 r-----~
~ =====+~":=~::~==~====t=~~~
e= :===~~:; ;; ; ;;:::a:::;;; ;;;;;:;::;;; ;;;;;;;; ;;;; """""ill
10+---~~--~-----+-----r----~--~

0 +-----~--~-----+-----r----~-~

3 4 5 6 7 8 9
Foam flowrate at atmospheric pressure, Om0 (x 10"4 m3s· 1)

Figure 5. 3D expansion ratio and 2D expansion ratio vs height and flowrate

One can note that : - The expansion ratios increase with the observation height,
- For a set height, the expansion ratios decrease when the foam
flowrate increases.
These observations can be linked to the fact that the longer the residence time, the
greater the drainage effect. A general study is in progress to obtain information on the
influence of foam flowrate on stability in order to find an optimal residence time for
foam drainage.
Comparing F2o and F30, we observe that curve shapes are the same for varying height
and tlO\\Tate. F30 is greater than F2o and F30 increases of a maximum factor of 5
instead of a maximum factor of 2 for F2o. As for foam structure, the resulting
hypothesis is that the wall phenomena seem to be more important than the bulk ones
on F 20 . In particular, wettability and liquid film influence the structure at the wall.

3.3.3. Conclusion
Wall effects seem to influence local expansion ratios and bubble size distributions,
more than bulk effects. Considering that F2o variations with height are lower than F3o
ones, the liquid film could be relatively uniform along the column. This is a positive
factor in the context of a decontamination operation.
331
4. Foam rheology

4.1. EXPERIMENTAL DEVICE

The experimental device is a 32 m long pipe \\ith an inner diameter of 30 mm. Foam
velocity ranges from 0.1 to 0.8 m.s- 1.and residence time is low (30 sec. to 5 min.). Thus,
°,
drainage is limited, whereas it was favoured in the previous study. Q, Q8 Qm0, and F0 are
defined as in the structure and stability study. F is the expansion ratio using Q8, the gas
flowrate, which is corrected with the pressure value, assuming that the gas is ideal.

4.2. EXPERIMENTAL RESULTS

Figure 6 represents L\P/L, the linear pressure drop as a function of foam flowrate, for
three different sections of the pipe. It increases along the pipe because foam expansion
due to pressure drops induces a bigger volume flowrate.

4.5
-s
"=
p.. 4

-
....
0
X
........ 3.5
e::!
~ 3
c:
"0
e 2.5

r
~
:;:1
Vl
Vl
2
~
c.
- - 1st section

/I --
~ 1.5 +-----.tF+----t---.;.----;------'J - • - 2nd section
=
<I)

3rd section
:.:3
0 2 3 4 5 6 7
Foam flowrate at atmospheric pressure, Qm0 (xi0- 4 m3s- 1)
Figure 6. Linear pressure drop vs foam flowrate

The values of total pressure drop (Figure 7) over the pipe confirm the influence of the
volurnic gas fraction . Indeed, pressure drops increase with the volumic gas fraction.
These two results show that volumic gas fraction plays a role in rheological behaviour
and that compressibility cannot be neglected. Considering that foam is a dispersed
medium, a specific flow model must be defined. A model defined by Valko et al. (11,
12] seems to be well adapted to foam rheological study because it combines a
conventional model with the foam compressible property.
332

1.4
--fO=lO
··•··f0=14
1.2
--f0=18
;f
-
--f0=25
0 1.0+------T.-----:-'----::I~-t-----!"---+--~---:-----+--i

. .. ~ . . ....···r··
X
: .. . ... .. F0:30
'-'
--f0=35
~ 0.8 -t--:t-:"-r~--:;.;;-=------;---...;;--~~----i----L--,-----!
c:
e
~ 0.6 -t----r---:-:-:---+----+-=-~-=t---~---------t-----;
::::3
"'
"'
J:0.4+----':....,.:...-;._,.:::::.::....--+----;----+------;-----'--- f----...;

0.2 +---+----+-----+-----+-----+-----+-----+---~

0 2 3 4 5 6 7 8
Foam flowrate at atmospheric pressure, Qm0 (xi0·4 m3s-1)
Figure 7. Total pressure drop vs foam flowrate for different expansion ratio values

4.2.1. Model Construction


The aim of this model is to work with a reference state (marked*) in which the friction
factor .f" and the wall shear stress 1:*P are constants along the pipe.
6=p*/p is defined with p* as a constant. Then u*=u/6 is a constant because of mass
conservation in steady flow. The local friction factor is defined by:
(2)
f!ocal = p* 2
!-.(u*.s)
s
If 1:p/6 is a constant (noted 1:* p) then .fiocal = .f" is constant
We assume that foam matches the following condition : "tp* is a constant. This
hypothesis is verified if a bijective relationship between "tp * and y* can be fitted on
the plot "tp * vs y * . The arbitrary reference state is defined as incompressible liquid
with an unknown rheological behaviour. Then the specific expansion ratio is defined
by 6 = PIIP and "tp* = "tp/6 and y* = y I 6 are the variables of the study.

4.2.2. Hypothese Verification and Curve Fitting


Figure 8 shows a flow diagram using the equalized shear rate y *and the equalized
shear stress "tp *. It has been built with experiments made on foams with expansion
ratio F0 values of 10, 14 and 18. Repeatability is better than what Calvert et al.[l3]
333
announced. Indeed, scattering is less than 5% instead of 20%. A bijective relationship
can be established between the two variables and the diagram shows that an Ostwald
de Waele model can be fitted. Only a drift appears for high flowrates and experimental
points corresponding to the pipe inlet. Under these conditions, the actual expansion
ratio F, is of about 5 or 6. Near this limit, Calvert et al. [8] noted that foam strucure
changes, which could induce a modified rheological behaviour.

;;.1.4


'a' f O:: lQ
!fl
- . 1.2 D

'@'
• f 0 = l4 0
~ 1.0 tt,,,.-H5
o f O= 18
"'
"'
,s. t~ ulf!8w•
I S~.Ii!il.fl.iH!I"
~ 0.8
a
l! 0.6
"'
'"0 ~ ~
l!:! 0.4
;.::::: j f~
:
~
§.0.2 0
<U

~0 ~
~
'0-0.2
0 I
i ~2 3 4

~-~-0.4 b
0.0
I
_s-0 .6
Logarithm of wall equalized shear rate (s· 1) , ln(y *)
Figure 8. Flow diagram

The values of the Ostwald de Waele model parameters do not depend on the expansion
ratio F0 . They have been calculated by linear regression of curves plotted in Figure 8.
Thus, the rheological equation is :
-r *P =o.26 (r •) 0·6 (3)
Knowing the rheological characteristics, a numerical method is used to calculate the
pressure drop in a capillary pipe.

4.3 . CONCLUSION

A rheological model of a flowing foam taking compressibilty into account has been
found in the F 0 and Qm0 domain we worked on. In the future studies, the effect of a slip
layer and the influence of pipe material will be thoroughly investigated.
334

5. General Conclusion and prospects


The aim of this paper was to show how foam can be used in nuclear decontamination.
The foam decontamination process has been industrially validated and two types of
comprehensive studies are in progress in order to master it. The first one is the study of
structure and stability using image analysis under operating conditions which favour
drainage. The second one is the study of foam rheology under operating conditions
which limit drainage. Gathering the results of both studies, we intend to be able to
predict the hydrodynamic behaviour in equipments of different geometry.
Understanding the hydrodynamics will also allow us to improve our knolwledge on
mass exchanges during a decontamination.

Bibliograpic references

[1] Faury M.L, Gauchon J.P., "Decontamination of large volume nuclear


components using foams", Contrat CCE n° FI2D-0035, (1994)
[2] Gauchon J.P., Brunei G., Costes J.R, Faucompre B., "Mousses de
decontamination a duree de vie controlee et installation de decontamination
d'objets utilisant une telle mousse", Patent n° 9109286, (1991)
(3] Biswal S.K., P.S.R. Reddy, S.K. Bhaumik, "Bubble size distribution in flotation
column", The Canadian Journal of Chemical Engineering, 12, 148-151 (1994)
[4] Bisperink C.G.J., A.D. Ronteltap, A. Prins, "Bubble size distibution in foams",
Advances in Colloid and Interface Science, 38, 13-32 (1992)
[5] Durian D.J., D.A. Weitz, D.J. Pine, "Multiple light-scattering probes of foam
structure and dynamics", Science, 252, 686-688 (1991)
[6] Lachaise J., S. Sahnoun, C. Dicharry, B. Mendiboure, J.L. Salager, "Improved
determination of the initial structure of liquid foams", Progr. Colloid. Polym.
Sci., 84, 253-256, (1991)
[7] Kroezen A.B.J., J. Groot Wassink, "Bubble size distribution and energy
dissipation in foam mixers", JSDC, 103, 386-394 (1987)
(8] Calvert J.R., K. Nezhati, "Bubble size effect in foams", Int. J. Heat & Fluid
Flow, 8, 102-106 (1987)
(9] Gido S.P., D.E. Dirt, S.M. Montgomery, R.K. Prud'home, L. Renbenfeld, "
Foam bubble size measured using image analysis before and after passage
through a porous medium", J. Dispersion Science and Technology, 10, 785-793
(1989)
[10] Cheng B.C., R. Lemlich, "Errors in the measurement of bubble size distribution
in foam", Ind. Eng. Chern. Fundam., 22, 105-109 (1983)
[11] Valko P., M.J. Economides, "Volume equalized constitutive equations for
foamed polymer solution", J. Rheol., 36, 1033-1055 (1992)
[12] Winkler W., P. Valko, M.J. Economides, "Laminar and drag-reduced
polymeric foam flow", J. Rheol., 38, 111-127 (1994)
[13] Calvert J.R., "Pressure drop for foam flow through pipes", Int. J. Heat & Fluid
Flow, 11, 236-241 (1990)
FOAM IN POROUS MEDIA

WILLIAM R. ROSSEN
Department of Petroleum and Geosystems Engineering
The University of Texas at Austin
Austin, TX 78712-1061, U.S. A.

1. Introduction

Foams are used in the production of petroleum and natural gas for several
purposes: in drilling wells, in cementing piping into a well, in hydraulically
fracturing the surrounding rock to give better communication with a well, and in
lifting out liquid accumulated in gas wells. In all these applications, however,
foam acts primarily in a flow channel (wellbore, hydraulic fracture) relatively
large compared to the bubble size. In such cases foam can be treated as a
homogeneous, albeit complex, fluid. Foam is also used for two other purposes
in oil and gas production: In processes where gas (usually steam, carbon
dioxide or hydrocarbon gas) is injected into an oil field to improve oil recovery,
foam can help divert gas into untouched portions of the reservoir. In acid well
treatments, foam can help divert acid into those layers most in need of
treatment. In these applications, where foam enters the rock pores of the
geological formation, foam bubbles are as large as, or larger than, the flow
channel (rock pores). As a result, foam bubbles interact with the solid matrix as
much as with each other. These applications are the focus of this article.
There are several reviews of foam applications in porous media available [ 1-
5], and I attempt here only a brief account of this field. In particular, I will
focus on the distinctions between the properties of bulk foams, the subject of
most of this volume, and foams in porous media, and on some topics likely to be
of interest to an audience largely comprising physicists and mathematicians.

2. The nature of foam in porous media

2.1. THE POROUS MEDIUM

The porous medium in which oil and natural gas is found comprises the
microscopic spaces between particles of sediment that form rock, modified by
335
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 335-348.
© 1999 Kluwer Academic Publishers.
336

subsequent processes of cementation, dissolution and compaction [6,7]. In its


simplest form, it comprises a multiply interconnected network of wider spaces,
called pore bodies, connected to each other by narrower spaces called pore
throats. The pore bodies are of order 1O's or 100 Jlm wide, and the throats are of
order l's or tO's of Jlm wide. Figure 2 below is a schematic representation of
this network in two dimensions (2D) . The pore walls are usually strongly
wetted by the liquid phase (water) over gas, so that pore walls are covered by a
thin film of liquid, and any crevices or comers in the pores are filled with liquid
as well. (In this application, liquid films between bubbles are called lamellae to
distinguish them from the liquid films that coat the rock surface.) In a
heterogeneous pore network, liquid, as the wetting phase, tends to occupy the
narrower pores and gas the wider ones.
In the conventional flow of two or more Newtonian phases such as water and
gas, Darcy's law governs the flow of each phase [6,7] :

(1)

where Uj is the volumetric flux (volume/area/time) of phase j, Aj is the phase


mobility, Y'pj is the pressure gradient in phase j, k is permeability (a property of
the porous medium), Jlj is the viscosity of phase j, and krj is the relative
permeability to phase j. For a given rock and pair of fluids, krj is a property of
the fluid phase saturations Sj, i.e. volume fractions of each phase in the
porespace. Figure la illustrates how relative permeabilities to gas and water
depend on water saturation Sw (gas saturation Sg = 1 - Sw) . Because water
occupies the narrower pores in the pore network, its flow is relatively

a) 1 b)

> .... ...


..... "' w
~

:;j '/ a:
::;)
iil
<t .1
'\ / CJ)
CJ)
w w
a:
~
a: a..
w ¥1 Fr >
a:
a.. '"'"
w
:::!.01
1/ :3...J
~ 0::
...J <t
(.)
w I
a:
.001
0 20 40 60 80 1 00 0 100
WATER SATURATION,% WATER SATURATION,%

Figure 1. Fluid properties in porous media (schematic).


a) Relative permeabilities (eq.1 ). b) Capillary pressure (eq. 2).
337

inefficient, and water relative permeability stays low up to large values of water
saturation. Note that the two relative permeabilities sum to less than one at
intermediate saturations. This may remind readers of network conductivity in
percolation theory [8], which is always less than the fraction of conducting
elements on the network. Indeed, percolation theory can explain many of the
aspects of flow of fluids in porous media [9-11]. The division ofmobility Aj in
eq. 1 into a ratio of separate properties of the rock (k) , fluids (Jlj). and fluid
saturations (krj) is valid for Newtonian fluids, but not for foams.
Even in the absence of foam, there is a positive capillary (i.e., osmotic)
pressure Pc between gas and liquid in a porous medium. For nonwetting phase
(gas) to enter the porous medium, its pressure Pg must exceed the wetting phase
(liquid) pressure Pl by an amount given by

(2)

where y is gas-liquid surface tension, 8 is the contact angle of the gas-liquid


surface on the pore wall, and Rt is the radius of the narrowest pore throat in the
porous medium entered by gas As liquid saturation decreases, gas enters
narrower and narrower pore throats, and capillary pressure rises, as shown
schematically in Figure 1b.

2.2. FOAM IN THE POROUS MEDIUM

No one can see foam in porous media of direct interest in oil recovery (i.e.,
rock). One can observe the effects of foam in rock. One can directly observe
foam in artificial porous media - sand- or beadpacks, or 20 pore networks
etched into glass plates - but various artifacts, including small sample size, 20
network geometry and relatively low capillary pressure make the relevance of
such observations moot [4]. Therefore, the basic structure of foam in porous
media is still subject to debate.
The two proposed foam structures are illustrated in Figure 2 [12]. In a
continuous-gas foam, lamellae block a fraction of the gas flow paths, reducing
gas relative permeability. Much of the gas may be completely surrounded by
lamellae and thereby trapped, but some gas still flows freely. In a
discontinuous-gas foam, all gas flow paths are blocked by lamellae. Gas cannot
flow (except perhaps by diffusion through lamellae) unless lamellae are
displaced from pore throats and bubbles are set moving. The continuous-gas
model for foam has never been directly disproven, but the remainder of this
article, like most of the current literature on foam mechanisms, will adopt the
discontinuous-gas foam model.
In porous media relevant to oil and gas production, foam bubbles are as large
or larger than pores. Thus the movement of foam in pores is less like the flow
of shaving cream through a pipe than like the movement of sausage-shaped
338

LAMELLA BUBBLE

\ ...
------·
---'
.....
....
' ,...

.........
I
~
~~

,...
A
~~
CONTINUOUS-GAS DISCONTINUOUS-GAS
FOAM FOAM
Figure 2. Schematic of continuous- and discontinuous-gas foams in rock.

bubbles through a periodically constricted tube. This finding is based on limited


measurements of bubble size in foams flowing out of rock core samples and
theoretical estimates based on foam mobility [13] . It may be that when two or
more bubbles occupy the same pore, diffusion rapidly eliminates the smaller
bubbles, as discussed below.
Like bulk foams, foam in porous media has a non-Newtonian rheology that
depends on bubble size [13,14]. This rheology is complex, however, because
bubble size can change drastically upon a small change in flow conditions (or,
sometimes, with no deliberate change at all). To make this complexity
manageable, here we apply the approach of the "population balance" [12,15,16]
and consider, first, the rheology of a foam of fixed bubble size, and then the
factors that alter bubble size.

3. Foam mobility at fixed bubble size

"Foam" in rock is not a separate phase. As bulk foam enters rock, most of the
water in the foam segregates into the same tiny rock pores it would occupy, as
the wetting phase, in the absence of foam. A small amount of liquid stays with
the gas as lamellae and Plateau borders between foam bubbles. (One can show
from geometrical arguments that if bubbles are several pores long this fraction
of liquid is small). As a result, the flow of liquid is unaffected by the formation
of foam - i.e. its flow is governed by eq. 1 with no change in the relative-
permeability function krw(Sw) (Figure 1a) that applies in the absence of foam
[ 17]. The presence of lamellae between gas bubbles enormously reduces gas
mobility, however, by factors of up to order 1Q6.
339

Like bulk foam, foam in porous media has a yield stress that is related to
bubble size [14]. For bulk foam, as discussed elsewhere in this volume, this
yield stress arises from the stretching of liquid films as sheets of bubbles slide
past each other. In porous media, however, the lamellae must stretch much
more greatly to permit flow. For instance, a lamella in a pore throat of diameter
10 J.Lm must increase in area by a factor of 25 if it is to advance into a pore body
of diameter 50 J.Lm. As a results, yield stress plays an important role in gas
mobility in foam. Remarkably, from 75 to 99% of the gas in foam remains
trapped even at huge pressure gradients (of order lO's to lOO's of atrnlm)
[15,18]. Models for this effect must account for both the converging-diverging
nature of individual pore pathways and the interconnections in the pore network;
a brief account is given in the next section. The existence of a substantial
effective yield stress for gas flow in foam has led some to conclude that foam
cannot flow over significant distances underground without prohibitively large
pressure gradients [19]; but others dispute this conclusion [5].
The bubbles that do flow are opposed by both capillary forces resisting
stretching of lamellae and the drag of Plateau borders along the pore wall. The
result is a strongly shear-thinning mobility for gas [13,14].
Clearly the trapping of much of the gas introduces an effect like a decrease in
gas relatively permeability in eq. 1, and the resistance to flow of bubbles that
move is like an increase in viscosity of gas that does flow. Models that break
gas mobility into these two parts (and focus primarily on the latter) have been
successful in fitting laboratory data [16]. Fundamentally, however, the two
effects are inseparable, since the same capillary resistance that traps gas in place
resists the movement of bubbles that do move.
The following section briefly describes a theoretical approach to determining
foam yield stress in porous media.

3.1. A MODEL FOR FOAM YIELD STRESS

At the onset of gas flow, where the pressure gradient is just sufficient to initiate
flow, bubbles move as a "bubble train" along a single sinuous path in the pore
network [14]. Figure 3 illustrates a model for estimating capillary resistance to
movement of bubbles along this path; real bubbles, each thought to extend for
several pores, are shrunken for illustration here. At the onset of flow, we
assume that flow rates are low, and therefore that viscous drag is negligible. At
any given moment, as illustrated, some lamellae bulge forward, resisting
forward movement by virtue of their curvature and surface tension, and some
bulge backwards, pulling the train forward. We assume for simplicity that pores
are biconical and symmetrical as illustrated in Figure 4; other simple pore
shapes can easily be accommodated. Significantly, even if individual pores are
symmetrical from front to back, the population of lamellae is skewed toward
340

lamellae bubbles

Figure 3. Model for estimating capillary resistance to gas flow


(effective gas yield stress) in foam.

Figure 4. Schematic of passage of a single lamella through a single pore.


From (a) to (c), lamella bulges forward; surface tension resists forward
movement during this period. At (c), volume of rearward bubble is greater
than (1/2) volume of pore; i.e., lamella has already spent more than half its
transit time resisting forward movement. From (c) lamella jumps to (d) to
conserve bubble volume. Trip from (d) to (f) mirrors first portion of trip, but
with surface tension pulling bubble forward.

those that resist flow [20]. The reason is illustrated in Figure 4 - when a lamella
reaches the midpoint of the pore according to its point of attachment to the pore
wall, its volume already accounts for over half the volume of the pore. In other
words, if volumetric flow rate is constant (Figure 3) the lamella spends more
than half its time of travel in shapes that resist forward movement. For given
pore shapes, one can compute the average capillary resistance to forward
movement and the effective yield stress of a population of bubbles in a bubble
train. The computation can account for contact-angle hysteresis and finite width
of Plateau borders, shown as points in Figure 4 [20] .
Gas compressibility causes lamellae to stay longer in pore throats, where
capillary resistance to forward movement is greatest [21]. In effect, each
lamella pauses in the pore throat, waiting for the rearward bubble to compress
enough to build up enough pressure to force it to move on. Populations of
compressible bubbles tend to get stuck in pore throats simultaneously and all
jump at once to the next throat, further increasing resistance to foam flow.
341

Liquid films, Plateau borders and trapped gas bubbles surrounding the bubble
train also expand and contract in response to changes in pressure in the bubbles
in the train, further increasing effective compressibility of the moving bubbles.
These effects can be significant. For instance, in one study of foam at
atmospheric pressure in a beadpack [14], the heaving and sighing of
surrounding bubbles gave an otherwise effectively incompressible foam a large
effective compressibility, greatly increasing the effective yield stress of the
foam [21].
The above analysis assumes that lamellae in radially symmetric pores, like
biconical pores, are radially symmetric themselves. This is not the case, as one
can show any evening washing the dinner dishes. Take a wine glass and slosh
the dishwater around to make a bubble partly filling the bottom of the glass as
illustrated in Figure 5. Though the wine glass is radially symmetric, the bubble
need not be. Lamella shapes in 3D pores are complex in such cases. Attempts
to modify the theory using a 2D analog [22] fit the basic trends in lamella
shapes, including asymmetric shapes, observed as a lamella traveled across a 3D
glass pore, but failed to reproduce the capillary resistance quantitatively [23].
This issue is ripe for resolution using the Surface Evolver program described
elsewhere in this volume.
wine glass
soap bubble

Figure 5. Simple demonstration of asymmetric lamella shape


inside a radially symmetric solid.

This approach presumes a given pathway of pores through which the bubble
train passes. The foam selects this path from all possible paths through the pore
network by minimizing capillary resistance along all the paths (in effect, by
looking for a relatively short path with wide pore throats). This problem is
similar to that of "minimal paths" in percolation theory. The foam selects a path
primarily of large-diameter pore throats, but occasionally chooses to pass
through a narrow throat if it thereby can access a cluster of wide throats. This
problem can be addressed either numerically [24] or using simplified network
models like the Bethe (or Cayley) tree [25].

4. Processes that change bubble size.

4.1. PROCESSES THAT CREATE LAMELLAE

Gas mobility depends on bubble size, or, more strictly, on the number of
342

lamellae in the pore network trapping some bubbles and resisting the movement
of others. Gas mobility therefore depends on the dynamic processes that create
and destroy lamellae. Four processes create lamellae [4,26]:
1) When gas invades the porous medium for the first time, lamellae are
created by "leave-behind." "Leave-behind, illustrated in Figure 6, occurs when
gas approaches a pore throat from both opposing pore bodies. In principle, an
enormous number of lamellae can be created in this way.

INITIAL STATE:
SATURATED LEAVE-BEHIND
WITH WATER LAMELLAE

•••
•••
•••
Figure 6. Schematic of "leave-behind."

2) When a lamella traverses a pore network, each time it passes through a


pore junction (pore body) as illustrated in Figure 7, it must either divide to
occupy all unoccupied throats, or else break. A huge number of lamellae can be
created quickly, if lamellae are moving (i.e., if the pressure gradient is sufficient
to overcome capillary resistance to flow).

time - - - - - - - - - - - - - - - - - - - - -
Figure 7. Schematic of lamella division as
a lamella passes through a pore body.

3.) As illustrated in Figure 1b, capillary pressure must rise to allow


increasing gas entry into the porous medium. If capillary pressure then falls,
"snap-off' can create lamellae in pore throats, as illustrated in Figure 8. Snap-
off can occur under at least six different scenarios [4], all of them involving
fluctuating capillary pressure.
4) If gas evolves in the midst of liquid, due to either chemical reaction or
changing temperature or pressure, new bubbles are created.
Whether snap-off or lamella division is the primary mechanism of foam
generation in porous media is the subject of some controversy. One can develop
343

liquid film

DECREASING CAPILLARY PRESSURE


Figure 8. Snap-off in a pore throat.(schematic).

a model for foam generation, triggered by mobilization of lamellae and lamella


division, based on predicting the pressure drop across individual pore throats
using percolation theory [27]. Others see this process as triggered by snap-off,
however [26].

4.2. PROCESSES THAT DESTROY LAMELLAE

In contrast to bulk foams, diffusion plays a minor part in determining bubble


size foam in porous media, beyond eliminating bubbles smaller than a single
pore. For instance, if the foam illustrated in Figure 3 were at rest, diffusion
would cause the first lamella (counting from the left) to retreat to the pore throat
behind it; the second and third lamellae would meet at the throat between them,
eliminating one bubble; the third lamella would retreat slightly; and the fourth
and fifth would meet, eliminating another bubble; and then diffusion would
cease. Diffusion would play no further part even if (as not shown in Figure 3
for lack of room) one bubble were tens of pores long and its neighbor only one
pore long. In porous media, diffusion tends to bring all lamellae to pore throats
and leave them there, with no further effect on bubble sizes.
Lamellae disappear then primarily because they dry out and break. Others in
this volume describe how lamellae break at high applied capillary (i.e., osmotic)
pressures. Moving lamellae break at somewhat lower capillary pressure
because they are under constant stress, stretching and contracting, dragged along
pore walls, sometimes encountering new pore walls as in Figure 7, and having
to form new Plateau borders along these walls. In bulk foams, as described
elsewhere in this volume, high capillary pressure evolves over time as wet foam
dries out under gravity. In porous media relevant to petroleum production,
capillary pressure is nearly always relatively high as a reflection of the presence
of gas in the medium at all (Figure lb). Moreover, there is no delay in reaching
this capillary pressure; the medium at high capillary pressure is all around the
foam, in intimate contact with it, and the foam reacts to it instantaneously. The
relation between Pc and water saturation (Figure 1b) means that lamellae break
344

at low water saturation in the rock. This idea is exploited further in the next
section.
The foam formed in different rock layers is not identical; in general, stronger
foam forms in layers with higher permeability, partially blocking those layers.
This helps to equalize flow among layers, and is one of the chief benefits of
foam. The reason, it appears, is related to bubble stability and capillary pressure
[28]. Higher-permeability layers have wider pores and, therefore, lower
capillary pressure (eq. 2). Lower capillary pressure means that foam bubbles do
not break as easily in these layers; smaller bubbles in higher-permeability layers
reduce gas mobility more and divert flow away from these layers.

5. A simplifying paradigm for foam in porous media

Accounting for both the non-Newtonian mobility of gas at fixed bubble size
and the many factors that affect bubble size is a daunting task, though progress
is being made toward this goal [ 12-16]. Some studies indicate, however, that
foam collapses, not progressively as capillary pressure Pc rises, as discussed
above, but catastrophically at a single value of Pc, the "limiting capillary
pressure" Pc * [28]. Enormous simplifications are possible when this
assumption is justified. The assumption appears to be most valid for strong
foams flowing under relatively dry conditions [13]. The collapse of foam at Pc*
acts in effect as a strong feedback controller, keeping foam in the vicinity of Pc *
over wide ranges of conditions. Since capillary pressure is related to water
saturation (Figure 1b), and the relation between water relative permeability and
water saturation (Figure 1a) is unaltered by foam, then water transport with
foam is controlled by eq. 1 with a single value of krw· Analytical models for
foam flow based on these concepts have thus far performed nearly as well as
population-balance simulators in matching experimental results, with much
greater simplicity [29]. There clearly are conditions under which a more-
complete model, like the population balance, will be needed, however.
Recent work suggests that, in contrast to the limiting-pc regime, which
applies under relatively dry conditions, a different simplification may apply
under very wet conditions [30]. In this case it appears that gas mobility may be
dominated by trapping and mobilization of gas bubbles, with relatively little
change in bubble size.

6. Other considerations in application of foam

There are many other considerations in application of foam to petroleum and


gas production that I have not attempted to address here. These are discussed in
the reviews cited above. [1-5]
Most crude oils destabilize foam, just as some oils are used to destabilize bulk
345

foams. This is especially true in oil reservoirs where a film of crude oil has
reduced the wettability of the rock surface toward water.
Just as gravity progressively dries out and destabilizes columns of bulk foam,
gravity tends to dry out and break foam at the top of oil reservoirs. This makes
it difficult to block gas channeling along the top of oil reservoirs, one of the
principle goals of foam application in improved oil recovery [31].
An important issue for field application of foam is the level of surfactant
adsorption on the rock surface. Surfactant adsorbed on mineral and clay
surfaces is unavailable to stabilize lamellae, and unfortunately one loses
surfactant to coating the rock surfaces over the entire pore volume in which one
hopes to place foam. This can be prohibitively expensive if foam is expected to
fill a significant portion of the oil reservoir in order to maximize gas diversion
in an improved-oil-recovery project. Therefore, most foam research today
focuses on less-ambitious near-well treatments, while some research looks for
less-expensive surfactants or surfactants with reduced levels of adsorption.

7. References

1. Smith, D. H., ed. (1988) Surfactant-Based Mobility Control: Progress in Miscible-


Flood Enhanced Oil Recovery, ACS Symp. Ser. No. 373, Am. Chern. Soc.,
Washington, DC.
2. Hirasaki, G. J. (1989) The Steam-Foam Process, J. Petr. Technol. 41, 449-456;
also (1989) A Review of the Steam Foam Process Mechanisms, paper SPE 19518,
Society of Petroleum Engineers (SPE), Richardson, TX.
3. Schramm, L. L. (ed.) (1994) Foams: Fundamentals and Applications in the
Petroleum Industry, ACS Advances in Chemistry Ser. No. 242, Am. Chern. Soc.,
Washington, DC.
4. Rossen, W. R. (1995) Foams in Enhanced Oil Recovery, in R. K. Prud'homme and
S. Khan (eds.) Foams: Theory, Measurements and Applications, Marcel Dekker,
New York.
5. Patzek, T. W. (1996) Field Applications of Foam for Mobility Improvement and
Profile Control, SPE Reser. Eng. 11, 79-85.
6. Collins, R. E. (1961) Flow of Fluids Through Porous Materials, Reinhold Publ.
Co., New York, NY.
7. Lake, L. ( 1989) Enhanced Oil Recovery, Prentice Hall, Englewood Cliffs, NJ.
8. Stauffer, D., and Aharony, A. (1992) Introduction to Percolation Theory, 2nd ed.,
Taylor and Francis, London.
9. Larson, R. G., and Morrow, N. R. (1981) Effects of Sample Size on Capillary
Pressures in Porous Media, Powder Technol. 30, 123-138.
10. Heiba, A. A., Sahimi, M., Scriven, L. E., and Davis, H. T. (1992) Percolation
Theory of Two-Phase Relative Permeability, SPE Reser. Eng. 7,123-132.
346

11. Sahimi, M. (1995) Flow and Transport in Porous Media and Fractured Rock:
From Classical Methods to Modem Approaches, Weinheim, New York, NY.
12. Falls, A. H., Hirasaki, G. J., Patzek, T. W., Gauglitz, D. A., Miller, D. D., and
Ratulowski, J. (1988) Development of a Mechanistic Foam Simulator: The
Population Balance and Generation by Snap-Off, SPE Reser. Eng. 3, 884-892.
13. Ettinger, R. A., and Radke, C. J. (1992) The Influence of Texture on Steady Foam
Aow in Berea Sandstone, SPE Reser. Eng. 1, 83-90.
14. Falls, A. H., Musters, J. J., and Ratulowski, J. (1989) The Apparent Viscosity of
Foams in Homogeneous Bead Packs, SPE Reser. Eng. 4, 155-164.
15. Friedmann, F., Chen, W. H., and Gauglitz, P. A. (1991) Experimental and
Simulation Study of High-Temperature Foam Displacement in Porous Media, SPE
Reser. Eng. 6, 37-45.
16. Kovscek, A. R., Patzek, T. W., and Radke, C. J. (1994) Mechanistic Prediction of
Foam Displacement in Multidimensions: A Population Balance Approach, paper
SPEJDOE 27789, Proc. SPEIDOE Symp. on Improved Oil Recovery, Tulsa, OK,
Apr. 17-20.
17. Bernard, G. G., Holm, L. W., and Jacobs, W. L. (1965) Effect of Foam on
Trapped Gas Saturation and on Permeability of Porous Media to Water, SPEJ 5,
195-300.
18. Gillis, J. V., and Radke, C. J. (1990) A Dual Gas Tracer Technique for
Determining Trapped Gas Saturation During Steady Foam Aow in Porous Media,
paper SPE 20519, Proc. SPE Annual Tech. Conf. and Exhibition, New Orleans,
Sept. 23-26.
19. Friedmann, F., Guice, W. R., Gump, J. M., and Nelson, D. G. (1994) Steam-Foam
Mechanistic Field Trial in the Midway-Sunset Field, SPE Reser. Eng. 9, 297-304.
20. Rossen, W. R. (1990) Theory of Minimum Pressure Gradient of Flowing Foams in
Porous Media. I. Incompressible Foam, J. Colloid Interface Sci. 136, 1-17.
21. Rossen, W. R. (1990) Theory of Minimum Pressure Gradient ofAowing Foams in
Porous Media. II. Effect of Compressibility, J. Colloid Interface Sci. 136, 18-37.
22. Rossen, W. R. (1990) Theory of Minimum Pressure Gradient of Flowing Foams in
Porous Media. III. Asymmetric Lamella Shapes, J. Colloid Interface Sci. 136, 38-
53.
23. Rossen, W. R. (1990) Minimum Pressure Gradient for Foam Aow in Porous
Media: Effect of Interactions with Stationary Lamellae, J. Colloid Interface Sci.
139, 457-468.
24. Kharabaf, H., and Yortsos, Y. C. (1996) A Pore-Network Model for Foam
Formation and Propagation in Porous Media, paper SPE 36663, Proc. SPE Annual
Tech. Conf. and Exhibition, Denver, CO, Oct. 7-9.
25. Rossen, W. R., and Maroun, C. K. (1993) Minimal Path for Transport in Networks,
Phys. Rev. B 47, 11815.
26. Ransohoff, T. C., and Radke, C. J. (1988) Mechanisms of Foam Generation in
Glass-Bead Packs," SPE Reser. Eng. 3, 573-585.
347

27. Rossen, W. R., and Gauglitz, P. A. (1990) Percolation Theory of Creation and
Mobilization of Foams in Porous Media, A/ChE J. 36, 1176-1188.
28. Khatib, Z. 1., Hirasaki, G. J. , and Falls, A. H. (1988) Effects of Capillary Pressure
on Coalescence and Phase Mobilities in Foam Flowing Through Porous Media,"
SPE Reser. Eng. 3, 919-926.
29. Rossen, W. R., Zeilinger, S. C., Shi, J.-X., and Lim, M. T. (1994) Mechanistic
Simulation of Foam Processes in Porous Media, paper SPE 28940, Proc. SPE
Annual Tech. Conf. and Exhibition, New Orleans, LA, Sept. 26-28.
30. Rossen, W. R., and Wang, M.-W. (1997) Modeling Foams for Matrix Acid
Diversion, paper SPE 38200, Proc. SPE European Formation Damage Symp., The
Hague, The Netherlands, June 1-2.
31. Shi, J.-X., and Rossen, W. R. (1996) Simulation and Dimensional Analysis of
Foam Processes in Porous Media, paper SPE 35166, Proc. SPE/DOE Improved Oil
Recovery Symp., Tulsa, April 21-24.
348
APPLICATION OF THE VORONOI TESSELLATIONS TO THE
STUDY OF FLOW OF GRANULAR MATERIALS

J. LEMAITRE, L. SAMSON, P. RICHARD AND L. OGER


Groupe Matiere Condensee et Materiaux,
UMR CNRS 6626, Universite de Rennes 1,
F35042 Rennes Cedex
N.N. MEDVEDEV
Institute of Chemical Kinetics and combustion
3.lnstitutskja, 630090 Novosibirsk 90, Russia

Abstract. This chapter provides a very simple method to simulate the displace-
ment of a sphere in interaction with a bi-dimensional or tri-dimensional structure.
These structures are modeled as assemblies of spheres glued on a plane (2D) or
packed inside a tank (3D). The displacement mainly occurs in the space in between
the spheres, so a subnetwork made by the Voronoi tessellation of the previous struc-
tures can describe nicely this open space. Each vertex of this tessellation is linked
to its neighbors (3 at 2D and 4 at 3D) and creates a continuous network through
which the moving sphere can travel. By using a series of assumptions for the cal-
culation of the probabilities to use each link and a Monte Carlo method for the
choice of them after, we have simulated the diffusion process of the moving sphere.
These simulations correspond to two experimental setups used in our laboratory
and agree very well with the experimental results.

1. Introduction
Granular materials have recently been the subject of increasing interest,
motivated by their crucial role in geophysics and industry and by the con-
siderable scope for investigation offered to physicists. However, in spite of
all this attention, our understanding of granular flows or movements is far
from being complete. For instance, a striking property of granular media is
segregation when a homogeneous mixture of granular materials of different
sizes or masses is shaken[l], rotated[2] or poured[3]. Two limit cases of these
segregation processes have been intensively studied in our laboratory. The
349
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 349-358.
© 1999 Kluwer Academic Publishers.
350

first one consists of a surface segregation in which a big sphere rolls on the
rough surface created by an assembly of small spheres (2D problem)[4]. The
second one is quite the opposite: a small sphere passes through an assembly
of large spheres (3D problem)[5].

2. Presentation of the bidimensional results


2.1. EXPERIMENTAL SETUP

In order to study the surface segregation, we have built an experimental


device where a moving sphere rolls down on a fixed rough inclined plane.
The plane is two meter long by one meter wide which define respectively
the x- and y-direction. The rough surface is made of a monolayer of glass
beads of radius r = 0.5 mm stuck on self-adhesive sheet (Figure 1). This
layer is disordered and homogeneous with a packing fraction around 0.7.
The moving sphere is a steel sphere, whose radius R ranges from 1.5 to
3.5 mm.

Steel baU,R

Figure 1. Schematic representation of the experimental setup. The launcher provides a


constant initial velocity of the steel ball.

Riguidel et al.[6] showed the existence of three different dynamic regimes


depending on the plane inclination angle (} and the smoothness </> = R/r
For small inclination angle (} or small ¢>, the ball decelerates and suddenly
gets trapped regardless of its initial velocity. For larger angle or </>, the
ball quickly reaches a steady state with a mean constant velocity regime,
which is independent of the initial velocity. For even larger values of(} and
¢>, the ball moves down the plane with big bounces and does not reach a
steady state on the two meter long of the plane. All the experimental and
351

numerical studies described in this paper deal only with the second regime
where the ball attains a steady state.

2.2. EXPERIMENTAL RESULTS

In all these experiments, the radius R of the moving sphere is at least twice
larger than thoses of the spheres constituting the rough plane. The moving
ball is released with an initial velocity close to the mean velocity it attains.
In order to analyze the transverse diffusion of the sphere along the falling
direction ( x), a collector made of a series of metal sheet separated by a
distance little larger then 2R is placed at a distance L from the launcher.
This distance L is chosen large enough to assume that the falling sphere has
already reached the mean constant velocity regime[4] . Typically L varies
from 80 em up to 170 em. Several sets of200 spheres are launched for a given
value of 0, a distance L and a radius R of the steel ball. The distribution
of the final positions (Yi) of the spheres inside the collector is obtained.
This distribution can be well fitted by a gaussian curve. The width of the
gaussian distribution is analyzed versus the distance L. From the figure
2, it can be clearly seen that < !:::.y 2 > increases linearly with L. As the
ball attains a mean constant velocity, it means that the ball has a diffusive
behavior.

60
50
2 40
<(t~ y)>
30
(em 2)
20
10
0
0 50 100 150 200
L(cm)

Figure 2. Va1iation of< Cl.y 2 > for a given value of 0 at different distances L from the
launcher.

Let a(R, 0) be the slope of the corresponding straight line, which can
be analyzed in terms of the radius of the falling sphere and the angle of the
plane ( N.B. the size of the glued glass beads is kept constant in our studies,
r = 0 ..5 mm). From figure 3, the parameter a(R, 8) decreases with increasing
352

values of sin( 8) for five different values of R and the range of inclination
angles 8 for which the mean falling sphere velocity remains const ant. The
characteristic length of diffusion a(R, 8) does not depend on the size of the
moving ball for a given inclination angle.

0.6 , - - - -- -- - -- - - - - - - - ,
0
0 R= J.S mm
0.5
= R=2mm
0.4 "' R=2.5mm
X
0
9x 6 x R=3mm
a (R,e) (em) 0.3 X
og %'< 0 R=3.5mm
0.2 = 6_
::: "'
- 0 () 0
0.1

0 ~-------------~
0 0.05 0.1 0.15
sin (e)

Figure 3. Variation of the slope of a(R,B) for values of R versus sin (8).

2.3. NUMERICAL SIMULATIONS

The displacement of the falling sphere on the inclined rough surface is


controlled by two main groups of properties. The first one concerns the
mechanical parameters of the problem such as the friction coefficient, the
restitution coefficient and the density effect. The second one is more closely
related to our approach of the geometry of the problem: the local positions
and arrangements of the glued spheres on the plane. Indeed, the positions
of the fixed spheres control partially the displacement of the moving sphere.
The number of possible impacts per unit length is directly related to the
mean value of the packing fraction i.e. to the number of possible changes
of direction of the trajectory during the falling. We are using this analogy
to simulate the displacement of the sphere along a well defined network.
Inst ead of choosing the impact position of the collision as the node
of the possible network of the available path , we are assuming that the
sphere follows the edges of the Voronoi tessellation made on a 2D assembly
of disks[7, 8]. Indeed, it is well known that a 2D stereological cut of an
assembly made of monosize spheres lying on a plane gives a set of disks
having all the same radii. So using directly a 2D packing of disks we can
model our glued ball layer. In order to generate a dense packing with a large
disorder in the ball positions, we use a packing program due to Powell[9].
353
Each disk, having radius chosen with a wide size fluctuation(!'= 1.2±0.2),
is placed in contact with two disks already placed. Then all the radii are
shrunk to a same minimal value which gives the required packing fraction (C
= 0.7) of the experimental surface. We build a plane with 160 disks wide and
200 disks long. As all the disks have equal size, a Voronoi tessellation is built
on this surface[7] (Figure 4). Each node of this tessellation is connected to its
neighbors by three links which follow the valley between two "sphere". Now,
the problem is to decide, starting at one given node, what path the sphere
will follow to go down. This problem is solved by a Monte-Carlo simulation
procedure after having defined the probabilities to use one peculiar node.

! lopm

Figure 4. Part of the numerical plane with a packing fraction equal to 0.7 and schematic
representation of the displacement of a falling sphere along the network of possible path
made with the Voronoi tessellation.

In order to model as much as possible the behavior of the ball along


the rough plane observed in the experiments, we are using two combined
probabilities, one related to the ratio of the radius of the falling sphere to
those of the glass beads constituting the roughness of the plane and one
related to the inclination of the plane. The first probability deals with the
height (hi) of the falling sphere (R) when it reaches the middle of the valley
between two adjacent spheres (r) of the plane, hi= r+J(R + r)2- (di/2)2
where di is the distance between the centers of two adjacent spheres. The
second probability corresponds to the biased effect of the gravity (i.e. the
inclination angle of the plane). It is obvious that the larger the angle () ,
354

the more straight is the direction of the falling sphere. We are using the
angle ai formed by the direction of the next link with the projection of
the gravity direction on the plane to define the second probability. The
higher ai, the smaller is the probability to use this link. The application
of the "gravity field" permits only the choice between the maximum of
two lower possible links for a given node. So the probability is defined as
P(ai) = 0.5+adx where xis related to the inverse of the inclination angle
0. The resultant probability is the normalized product of the two previous
P(h )*P(a )
terms: Pi_ = 3 ' ' •
2:;=! P(hi)*P(a;)
The Monte Carlo simulation is performed with 50 000 successive tra-
jectories with a random choice of the initial input of the falling sphere.
The relative lateral displacement is recorded at 20 different decreasing "al-
titude" along the plane and a best gaussian fit is calculated after that in
order to determine the width of the distribution. The figure 5 represents
the results of this diffusion along the falling axis for a value of R/r=2 and
x=10. We can see the linearity of the width of the gaussian versus the
distance which is in a very good agreement with the experimental results
shown in the figure 2.

2~r-----------------------~
_/!
/~w-

1500 ---
J -

2
<~y) >I~ - -,_
(a.u.) ------
500 -,...-,_

ov
0 so 100 150 200 250 300 350
L(a.u.)

Figure 5. Variation of < t:.y 2 > for a set of input parameters at different distances
from the initial layer. The two axes are defined in unity of the radius of the constitutive
spheres of the plane ( r).

We have performed this simulation for a large range of variation of the


ratio Rjr from 0.1 up to 15 and a series of values for x (Figure 6).
We can see that the slope of the lateral diffusion is independent of the
ratio Rjr for the large values for a given inclination value (x). This can
be well compared to the experiments (Figure 3) and easily understood: the
355

6.5 . - - - - - - - - - - - - - - - - - - - - - - - - .

6 <>o
o<P'<;Pooooooo o o = x:=sl

c. x=2
0 x=10
5.5

a(R.x) - x-.100
<> infinity
s
=- -
4 .5

4 L-----------------------~
0 5 10 .IS
R/r

Figure 6. Variation a(R,x) of the slope of< 6.y 2 >versus L for all possible values of
the input parameters versus the ratio R/r.

effect of the roughness diminishes with the increase of the ratio, hi becomes
quite independent of the distance di so the two probabilities P(hi) are very
close to one another. The interesting part is the possibility to analyze ratio
smaller than 1 where the sphere "sees" a large roughness of the structure.
This part cannot be reached experimentally due to the difficult conditions
to get the mean constant velocity regime. The slope increases greatly with
the decrease of this ratio which means that the displacement process deals
mostly with a percolation problem instead of a diffusion one. The choice
is made according to the most accessible path and less at random than
previously.
When we increase the influence of the gravity (i.e. decrease of the pa-
rameter x) we observe a diminution of the slope a. This can be easily
understood because we favor the smallest vertical deviation of the particle
so the lateral displacement becomes smaller. This effect is also shown in
the experiments (Figure 3) .

3. Presentation of the tridimensional results


3.1. 3D SETUP
The good agreement found in the comparison between 2D simulations and
experiments allows us to assume that this basic approach will be also valid
to simulate the displacement of a small sphere through an assembly of
larger spheres packed inside a cylindrical tank (Figure 7). In the ex peri-
356
mental setup, we can performed two kinds of measurements, the first one
is temporal and cannot be accessed by our simulations, the second one is
spatial and can be performed by using the same approach as in the previous
paragraph.

COMPUTER

X-YTABLE
FLOWING BEAD (r) H

MICROPHONE

ADHESIVE SHEET

Figure 7. Schematic representation of the tridimensional setup used to follow the dis-
placement of a small sphere through a packing of large ones.

3.2. 3D SIMULATIONS AND RESULTS

The experiments are performed with different sizes and matters for the
fixed beads (radius R = 4, 5, 8 mm, glass, styrene) and different sizes for
the falling steel spheres (radius r = 0.35, 0.5, 1 mm). These parameters
give different results for some mechanical parameters like the density and
the coefficient of restitution. These two parameters can be indirectly intro-
duced in the 3D version of the previous Voronoi-Monte Carlo procedure[8].
The tridimensional assemblies of fixed spheres are built following the same
packing algorithm [9] with a small size distribution. A new sphere is placed
at random on three already placed spheres. The packing is made with 16000
spheres placed inside a parallelepiped box with periodic boundaries on the
horizontal direction. The Voronoi tessellation is then performed without
degeneration for the vertices (i.e. each vertex is connected to 4 vertices
through 4 edges).
Following the same approach as the 2D one described previously, we
have "simulate" two properties of this process : the restitution coefficient
and the density effects. The first parameter is modelled as the possible
357

height of the bounce (z) of the falling sphere according to the material of
the two kinds of spheres. The second parameter is closely related to the
approach of the parameter O:i defined in the 2D version. In this case the
fall of the particle is vertical but the change of density of this particle
can be also related to the probability to accept a non-vertical direction
for the fall. This measure is controlled by an azimuthal variable A. So the
probability for one sphere to find the possible output path i is defined as
Pi = L:t(Ad*Q(z;) for the four connected links at a given vertex.
J=l P(A 1 )*Q(zj)
Then the classical Monte-Carlo algorithm can be performed for this
structure with different values of z and A. The initial input position of the
sphere through the packing is chosen at random between all the available
vertices on the top of the packing. The relative transverse displacement of
the falling sphere is recorded at different heights inside the packing. We
found that this displacement follows a classical diffusion process modelled
by a gaussian dispersion versus the traveled distance. The figure 8 shows
the evolution <R2 > / H and the number of changes of the direction with
the radius ratio between the falling sphere and the fixed ones. These results
are obtained for parameters which are not too far from the experimental
ones (5].

14

12

10

8
NIH
6

0
16 26 0.03 0.08 0.13
Rlr r/R
a)
b)

Figure 8. Evolution of < R 2 > / H and the number of changes of the direction with the
ratio between the falling sphere and the fixed beads.

These results permit to relate linearly the number of collisions to the


inverse of the packing sphere radius R. Indeed, the number of collisions is
intuitively related to the ratio of the height of the packing to the radius
of the spheres R inside the packing. This is proven by the figure Sb. The
358

figure 8a shows also the linearity of the radial diffusion with the radius R.
Finally, we have shown that only the radius of the spheres constitutive of
the packing are involved in the transverse diffusion process and that this
diffusion varies linearly with this radius.

4. Conclusions
We have developed a very simple algorithm using the structure of a Voronoi
tessellation of a hi and tri dimensional packing of spheres in order to simu-
late the transport of an unique sphere along or through these packings. We
have obtained a good agreement between our numerical simulations and
the experimental results performed in our laboratory. The fact that the
Voronoi tessellation can match quite well the geometrical arrangement of
the porous structure is very important for the future use of this technique.
Indeed, this approach can be extrapolated to a lot of transport problems
in packings of grains, such as percolation problem, deep bed filtration, seg-
regation, mercury porosimetry measurements, etc ... where the importance
of the paths is well known.

References
1. J.B. Knight H.M. Jaeger and S.R. Nagel Phys. Rev. Lett. 70, 3728 (1993).
2. S.B. Savage in Disorder and Granular Media , D. Bideau and A. Hansen eds., North-
Holland,Amsterdam, 255, ( 1993).
3. H.A. Makse, S. Havlin, P.R. King and H.E. Stanley, Bull A.P.S., 41, 384 (1996).
4. L. Samson, I. Ippolito, G.G. Batrouni and J. Lemaitre, J. Physique I (submitted).
5. L. Samson, Ph. D thesis, University of Rennes 1, (1997).
6. F.X. Riguidel, A. Hansen and D. Bideau Europhys. Lett. 28, 13 (1994).
7. J. Lemaitre, A. Gervais, J.P. Troadec, N. Rivier, M. Ammi, L. Oger and D. Bideau,
Phil. Mag. B, 67 347 (1993).
8. N.N Medvedev, V.P. Voloshin and Y.l. Naberukin, Phys A:Math. Gen. 21, L247
(1988).
9. M.J. Powell, Phys Rev B 20 4194 (1979).
DETERMINATION OF REAL THREE DIMENSIONAL FOAM
STRUCTURE USING OPTICAL TOMOGRAPHY

C. MONNEREAU and M. VIGNES-ADLER.


Laboratoire des Phenomenes de Transport dans les Melanges du
CNRS
4ter, Route des Gardes F-92190 Meudon (France)

1. Introduction

Foams are presently a very active field of research as it is commonly encountered in the
daily life. Until now, many studies have been devoted to the morphology of two
dimensional foams, which are much more simple than the three dimensional ones. A 20
foam consists of a single layer of bubbles between two flat and parallel glass plates. The
liquid ftlms are perpendicular to the glass plates, and the 20 foam appears as a netwott:
of polygons. Its structure is very easy to observe and the way the 2D foam structure
evolves is now well-known. The film network reorganises during time because of ftlm
ruptures (coalescence) or because of gas diffusion from a small bubble towards a bigger
neighbour due to the difference of Laplace pressure (coarsening or disproportionation).
Von Neumann has demonstrated that the law of coarsening in 20 foams for a bubble
depends only on its topology (its number of edges, n) and not on its area :

-dAn = K (n - 6), (1)


dt
with An the area of the n-edges bubble and K the diffusion constant Thus, a 6-edges
bubble has a constant area with time, a bubble with less than 6 edges shrinks and one
with more than 6 edges grows. 6 is the mean value of edges per bubble in a 20 foam
with an infmite number of bubbles. It also evolves because of morphological
rearrangements such as a topological change (Tl) where two neighbouring bubbles
generate a new face and the transformation (T2) where a bubble disappears

Let us consider now three dimensional foams. When the foam is dry, the bubbles
are polyhedral and the structure is therefore much more complex and difficult to observe.

Glazier et al. [1] have used magnetic resonance imaging to probe the interior of a
foam at different time in order to study coarsening and foam drainage; the size
359
J. F. Sadoc and N. Rivier (eels.), Foams and Emulsions, 359-378.
© 1999 Kluwer Academic Publishers.
360

distribution of the bubbles versus time was obtained. Dorian et al. have measured the
average bubble size by the multiple light scattering method [2]. A thin tip of an optical
fibber has been used to scan vertically a foam; the signal received is different when the
tip is in the gas or in the liquid. and this provides the avemge distribution size of the
bubbles [3]. An original experiment has been done by Muller and Di Meglio who
studied cascades of films ruptures in a foam by means of a microphone [4] .

All these methods only give global information on the foam structure. However,
the topological elements of the foam, such as the avemge number of faces per bubble
<f>, is of large interest for theoretical and numerical investigations. The foam structure
is such that the liquid ftlms of the foam organise themselves to minimise their surface
at a given gas volume. From the area-minimising principle, Plateau derived the first
laws of foam geometry : (i) the edges (Plateau borders) are formed by three liquid films
equally inclined toward one another, with mutual angles equal to 120°; (ii) the vertices
are formed by four edges equally inclined toward one another in space at an angle of
109.5° (the tetrahedral angle, cos·1 (-1/3)). Moreover, each liquid film has a constant
curvature, because the pressure difference between the two sides of a film is inversely
proportional to its mean curvature (Laplace's law).

In 1911, Kelvin stated that foams solve the mathematical problem of "the division
of space with the minimum partional area" [5]. Applying Laplace's law and Plateau's
laws for a monodisperse (bubbles of equal volume) dry foam, he proposed the
tetrakaidecahedron as a unit cell; it is a 14-faced polyhedral bubble with six planar
quadrilateral faces and eight hexagonal non-planar faces with a zero mean curvature. In
1987, Princen and Levinson calculated the surface area and the volume of the Kelvin
cell. They proved that it was the minimum area cell in a monodisperse foam proposed
until that time, but smaller ones could exist [6]. In 1946, Matzke detemlined the
structure of 1000 bubbles of a soap foam by looking at photographs [7]; he found that
<f> = 11.00 for 400 bubbles belonging to the frrst three layers and <f> = 13.7 for the
600 internal bubbles. In 1995, Weaire and Phelan generated numerically a bubble
structure in a monodisperse foam whose area is smaller than the Kelvin cell; this
structure has six 14-sided polyhedra and two irregular pentagonal dodecahedra with <f> =
13.5 [8].

As well-known, a liquid foam is never in stable thennodynamic equilibrium. It


evolves because its energy is very high due to its large surface toward either a metastable
state or its destruction. Even a very stable foam suffers some rearrangements as time
elapses. Glazier and Weaire have proposed two different types of rearrangements that
could occur in a 3D foam by analogy with the 2D foams : a topological change (f1)
where two neighbouring bubbles generate a new face, and the transformation (f2) where
the tetrahedral bubbles disappear [9]. A series of film rearrangements was studied by
Schwarz [10].
361

There is an obvious lack of data about the local structure of 3D foams and the
purpose of this paper is to supply some. Hence, optical tomography was developed as a
non-destructive method to investigate the 3D geometry of transparent foams. The
principle is to get images of a series of foam cross sections, to determine the locations
of all the foam vertices and to numerically reconstruct the equivalent real foam. This
paper describes our optical tomography method and the foam reconstruction procedure. It
was applied to two physical situations. In a first one, a soapy foam was dried and the
subsequent morphology changes were followed. In another experiment, the dynamics of
a coarsening foam was investigated. As far as known, this work provides the first
experimental evidence of the existence of a 3D Von Neumann's law.

2. Experimental

2.1 CELL

The foam cell consists of a cylindrical container (5 em diameter and 10 em height)


covered by a glass lid (Fig. 1), whose lower part is a sintered glass plate. Its bottom is
connected to the liquid reservoir by a flexible Teflon tube in order to control the
capillary pressure in the foam by changing the height level (ilh) between the container
and the reservoir. .ih is measured by a cathetometer. Both container and reservoir are
fixed to the same motorised table (z) to keep Llh constant during scanning. The origin of
the vertical axis is located at the bottom of the foam; z =H corresponds to the top of the
foam column.
z

porous plate
reservoir

Figure 1. Foam cell at controlled capillary pressure.

If mechanical equilibrium is assumed, the capillary pressure P1 (or Plateau border


suction) inside a foam column is determined by the difference between the gas pressure
Pb in a bubble and the liquid pressure PPbin a Plateau border [11] :
362

p'Y = pb - PPb = -
r (2)
rPb
where rPb is the radius of the Plateau border located at the height z in the foam. Since the
gas phase can be assumed to be in hydrostatic equilibrium. we also have
dl\
~=~~. m
where pb is the gas density, and g is the gravity. Combining (2) and (3), it comes

~
~
=-pbg-r~(..!...).
~ rpb
(4)

The net force F : :::~v:::i~p~:::7; ~f(oo; i)s : : : :i~e: b(y..!...), (5)


~ ~ rPb ~ rPb
where p is the liquid density. At mechanical equilibrium F = 0, and

pg = -r~(-1
~rPb
J· (6)

Interpretation of (6) shows trivially that the top of a foam column should be dry with
polyhedral bubbles since rPb is small; the bottom of a foam column should be wet with
more spherical bubbles since rPb is large.

When a foam column is located above a porous plate cell [11], as the one described
in Fig. 1, the pressure Pb in the bubble is equal to the atmospheric pressure and the
capillary pressure is directly related to &I by the following equation
Py(z) = Pb- PPb(z) = pg (&I+ z) . (7)
If pgAh >> pgz, it becomes:
Py(z) = pg &I. (8)
If relation (8) is fulfilled, the capillary pressure is constant in the foam column.

In the wet/dry experiment, the same foam column of 4 bubbles thick was samned
for &I= 0 mm (wet foam) and immediately after for &I= 74.8 mm (dry foam), with an
identical topology of the foam since no film ruptures was observed. For the wet foam,
gravity must be considered. In the dry foam, the capillary pressure is constant in the
foam column and it equals 752 Pa. In the coarsening experiment, &I is fixed to 60 mm,
i.e. the capillary pressure is constant and fixed to 595 Pa. Thus, the film thickness was
maintained nearly constant in the coarsening foam column and during the experiment

2.2 OPTICAL SET-UP

Visual observations of the foam are made with a Lhesa CCD camera equipped with a
Micro-Nikkor 105 mm f/2.8 objective and connected to a graphic digitisation card (Fig.
2). The maximum aperture of the objective is used to get a very thin depth-of-fleld
(lmm thick). The camera is focused on the centre of the upper surface of a foam sample
to only visualise the bubbles that are not in contact with the cell walls. The cell is lit
363

with a cold illumination plate, the light of which is polarised in order to minimise
reflection in the foam.
z

visualisation 2D slices
CCD Lhesa camera r--- image acquisition
D 3D reconstruction
image plane
Micro Nikkor v L.oo"
objective lOS mm f/2.8
bJ
3D
TRANSPARENT
FOAM
motorised stage z glass lid

c::::7 ,::: ~
..IIIII
~ 20mm
... lll ,..__ depth-of-field =I mm
focus plane _11~ ~

.__v
I h
'?
analyser
polaryser

(' ?J71
cold illumination table
X

Figure 2. Experimental set-up of the optical tomography.

2.3 PROCEDURE

In order to prevent contamination by antifoaming materials, great care is taken in


cleaning the cell. After degreasing with acetone (NP) and alcohol, the glassware is
washed with fresh sulfocbromic acid, the Teflon tube with boiling aquaregalia, and they
are rinsed profusely with pure water.
An aqueous soapy solution is poured into the cylindrical container. A polydisperse foam
is then created by blowing filtered U-nitrogen through a pro-Pasteur pipette directly into
the soapy solution at a flow rate equal to 55 mm 3 .s- 1• This process produced bubbles of
5 mm mean diameter size. In order to reduce the hydrostatic pressure gradient in the
foam column, a single layer of foam of 4 or 5 bubbles-thick corresponding to a
thickness of 20 mm, respectively 25 mm, is created in the sintered glass plate cell.
364

During scanning, the optical system is maintained fixed while the foam cell, that is
fixed on a motorised stage is displaced in the focus plane, by steps equal to ~z = 1 mm
(Fig. 2). Images of the slices are a:quired and stored in a microcomputer. A whole
scanning takes 45 seconds for 28 slices. Since nitrogen has a low solubility in water,
the foam coarsening is slow and it should be negligible during the time of a scanning.

3. Foam reconstruction

Images of foam slices are fJCSt processed with Visilog software from N~sis on a Silicon
Graphics RSOOO workstation. Then, the whole reconstruction procedure involves three
main steps
(i) determination of the equivalent polyhedral structure having the same number alll
arrangement of bubbles, faces, edges and vertices.
(ii) calculation of the total surface energy E of the equivalent polyhedral structure.
(iii) minimisation of E with Surface Evolver assuming that the vertices are fixed in
space; at the contrary, the edges that do not belong to the foam boundary and the faces
are free. Both gas and liquid are treated as incompressible. During and after energy
minimisation, the 3D foam structure is displayed by Geomview software.

The physical assumption underlying the reconstruction procedure is that only


capillarity is acting on the foam structure. This can only be strictly verified for dry
foams with negligible liquid fraction. The «equilibrium» structure is then obtained when
the total surface of the foam is minimum.

3.1 IMAGE PROCESSING

The liquid films are transparent, only the Plateau borders' network (edges) appears in the
images (Fig. 3). In the individual images, there are sbarp parts which belong to the
actual focus plane z = constant, and fuzzy zones which do not belong to it. The sharp
points correspond to the bubbles vertices, and their co-ordinates (x,y,z) are determined by
means of an appropriate software Mousse that was developed with N~is Inc. The series
of images is visually inspected and the vertices are approximately located by the operator
just by clicking on the vertices that look sbarp. Then, for each clicked vertex, Mousse
software analyses the whole series of images to find in which one the vertex has the best
sharpness. This determinates the x, y, z vertex co-ordinates. The spatial resolution is 30
fJID in the x and y directions, and the accuracy of the measurement is 0.5 mm in the z
direction, that corresponds to half the distance between two successive images. Each
vertex is numbered; vertices numbers and co-ordinates are stored in a datafile readable by
Surface Evolver.
365

3.2 RECONSTRUCTION

To reconstruct the foam, the equivalent polyhedral structure is then determined (Fig. 4).
It consists in defining, orientating and naming each edge, each face and each bubble of
the foam, knowing that the faces are oriented with their normal going out. This stage
has been programmed in Mousse software and the operator has just to click on the
vertices in the order, i.e. such that the face normals are pointing out . The data are also
stored in the above datafile.

Figure 4. Reconstructed 12-faced bubble with nonnals out, 30 edges and 20 vertices.

At this stage, the structure of the 3D foam is a set of polyhedra with straight edges.
Surface Evolver minimises the surface area of the reconstructed foam. The area-
minimisation makes the faces, the edges more curved and the polyhedral angles closer to
the tetrahedral angle (Fig. 5). The reconstructed foam is the one whose total surface area
is minimised.

Measurements can then be made on the reconstructed foams and a C++ program
was developed to determine the bubbles area and volumes, the film area and the edge
366

lengths. The numbers and arrangements of the topological elements of the foam, i.e. the
bubbles, the faces, the edges, and the vertices can therefore be analysed.

Figure 5.
w 00
Example of a reconstructed foam (a) before and after (b) minimisation.

4. A WeUDry foam

A polydisperse foam was generated from an aqueous solution of saponin (0.3 % wt) am
glycerol (3.5 % wt); its density p, surface tension y, and dynamic shear viscosity Jl are
1.025 g.cm·3, 38.0 erg.cm· 2 and 1.11 cP, respectively.

Examples of a slice of the wet foam (a) and of the dry foam (b) are shown in Fig.
6. In the dry foam, the edges are seen as segments, the vertices as points with negligible
volume, and the films are transparent. In the wet foam, some liquid excess can be seen
in the edges and in the vertices, while the ftlms still appear transparent.

In Fig. 6, a scheme explains the geometrical changes that occur on the Plateau
borders between the dry and the wet foam; an edge becomes a prism and a vertex an
octahedron with triangular faces. To some extent, the dry foam is the skeleton of the wet
one. This method is known as the decoration theorem.
367

.~

...' ...
I

!
I
~
'
'
I
(a)
\
' - - - - - -"'"""'"oo;;{b") '
; - (-
Figure6. Same foam slice (a) in the wet foam, Ml = 0 mm and (b) in the dry foam, Ml = 76 mm.

The dry foam is first processed since it is much more simple. In that case, the total
foam surface energy Edry is calculated assuming that only the films contribute; the film
energy per unit area is taken as 2y (the film interaction energy is neglected).

The reconstruction of the wet foam is much more complicated, because the Plateau
borders have no longer negligible volume and labelling of the foam elements become
very cumbersome. Since the previous polyhedral structure equivalent to the dry foam is
the skeleton of the wet foam, its datafile is used to generate the names of the various
elements of the wet foam.

Plateau border
Plateau border

octahedral vertex vertex

drainage

4 wet Plateau borders 4 dry Plateau borders

reconstruction

Figure 7. Geometrical changes between (a) wet and (b) dry foam.
368

One dry vertex generates six vertices and eight triangular faces (Fig. 7). At a vertex,
three fore-vertices are usually visible on the series of images and their co-ordinates
(x,y,z) are detennined by zooming on the images. Then, the co-ordinates of the three aft-
vertices are calculated by a C-language program, knowing that they form an octahedron.
Afterwards, the program creates the three edges of each wet foam Plateau border that link
the previous octahedra two by two, referring to the dry foam ftle.

Calculation of the total surface energy Ewet is also more complex for wet foams. In
a first approximation, it is assumed that gravity effect can be neglected in the foam
energy minimisation. The contribution of the surfaces of the vertices is first evaluated.
The four faces of the octahedral vertex which are shared by the four bubbles meeting at
this vertex contribute to Ewet with the surface energy per unit area r. The other four
faces are common to the four prisms (Plateau borders); actually, they are artefactual
since they do not exist as a liquid-gas interface, and no surface energy term has to be
associated with them. Then, the film contributions are calculated assuming again that
the surface energy per unit area of the films is 2y. Ewet is minimised with the vertices of
the bubbles fixed whereas the edges and the faces are free.

The reconstructed wet and dry foams are displayed in Fig. 8. Only the faces am
Plateau's borders network of its entire bubbles, i.e. not cut by the visualisation field are
given in Fig. 8. As expected, the Plateau borders of the wet foam are thicker at the
bottom of the foam column than at its top.

Figure B. Wet (a) and dry (b) reconstructed

Although thirty bubbles have been reconstructed, the numerical results are only
given for the 9 complete bubbles. The other bubbles are needed to calculate the liquid
fraction and to minimise the area of the system.
369
4.1 NUMBER OF FACES PER BUBBLE

For convenience, the bubbles are labelled in order of increasing number of faces. The
average number of faces <f> of the nine reconstructed bubbles is 11.1, which seems
very low since <f> should range between 13 and 14, depending on the model; of course,
statistics are poor. The mean value of faces per bubble found in the present experiment,
<f> = 11.1, is however in good agreement with Matzke value for external bubbles <f>
=11.00 [7].
4.2 EVOLUTION OF TilE BUBBLES FROM TilE WET TO TilE DRY FOAM
STA1E

The volume, area, and Isoperimetric Quotient IQ of the entire bubbles were calculated in
the wet and in the dry reconstructed foams. Results are reported in Fig. 9.

IQ compares the area S of a polyhedron to the area S0 of the sphere of same volume

e: r.
[6]:

IQ= 3~;2 = (9)


For a sphere, the area is minimum and IQ = 1. For a polyhedron, IQ is always smaller
than 1, and the smaller, the more its shape departs from a sphere; moreover, the larger
IQ, the more stable bubble.

With the present soapy solution, there is no obvious general law of evolution upon
drainage of the geometrical quantities.

The volumes of all the bubbles decrease of about 10% except for the two bigger
ones (8 and 9) which inflate, and for bubble 2 which remains constant (Fig. 9 (b)). The
area changes follow the volume ones (Fig. 9 (a)). The bubbles that belong to the top
surface (bubble numbers <6) have a small number of faces and their Isoperimetric
Quotient is low (Fig. 9 (c)): none have an IQ as high as Kelvin tetrakaidecahedron (IQ =
0.757) or Weaire-Phelan structure (IQ =0.764); the largest value is IQ = 0.727 for wet
bubble 7. Upon drainage, except for bubble 8, IQ decreases since the bubbles are then
less spherical and they interact more.
370
160 - . - - - - - - - - - - - - , 160 - r - - - - - - - - - - - ,
t
al••
140 •• & • 140
120 • :;" 120
1i 100 I ~ 100 I
g 80
g 60 ll
40
20 • •
0~~+-+-+-+-+-+-~
I 2 3 4 5 6 7 8 9 3456789
Bubble Bubble

(a) (b)

• ••
0,8
0,7
0,6
I
% !
I
0,5
IQ 0,4 I %
0,3
0,2
0,1
0
@] @] ~~ [!!][Q] ~[ll:l.
15* ~

2 3 4 5 6 7 8 9
Bubble

(c)

Figure 9. Evolution of the (a) area, (b) volut ,,,, and (c) Isoperimetric Quotient of the 9 bubbles from
wet state c!' )to dry state ~ ). The framed numbers are the face numbers. (*) is the number of faces in
dry state.

The volume changes can be assigned to the disproportionation phenomenon. For


the bubbles belonging to the top surface, the atmosphere constitutes a huge bubble ani
it seems that bubbles 8 and 9 have increased at the expense of their smaller neighbours.
Upon drainage, the ftlms thin drastically and this decreases the time for the gas transfer
across the film. However, the volume decrease of bubble 7, which is the biggest bubble,
contradicts this explanation. Actually, things are more intricate.

Although, no films broke during both scannings, the structure of bubbles 7 and 8
have changed upon drainage (Fig. 10). The bubbles 7 and 8 share face A (dotted); they
also share the edge FG with a third bubble during the whole experiment which is not
drawn for sake of clarity. Now after drainage, face B (hatched) has slid from bubble 7 to
bubble 8. Upon drainage, the 4-edged face B that was the smallest face of the bubble 7
in the wet state (2 mm2) totally shrinks in that bubble by a T2 topological
371
transformation (as in 2D foam). As a result, this gives 5 edges meeting along the same
vertex which is a physically unstable configuration; it evolves with the creation of a
new 4-edged face which now belongs to bubble 8. In the swapping, bubble 7 has gained
one 4-edged face and lost two 5-edged faces; the creation of the face B in the bubble 8
has generated one more 6-edged face. Hence, the T1 face swapping occurred at constant
edges and faces numbers.

drainage
WET DRY

face A

faceB

Figure 10. Tl face swapping in three dimensions.

The observed volume changes of bubbles 7 and 8 are likely a consequeoce of local
convections accompagnying the face swapping; this usually accelerates gas transfer
across the film they have in common and explains their fast gas volume
increasing/decreasing.

4.3 LIQUID FRACTION <i>I

The liquid fraction of the wet foam was calculated assuming that the thickness of the
films is negligible and that all the excess liquid is in the Plateau borders. The liquid
fraction of the foam can be estimated as
372

(10)

with VPb the total volume of the Plateau borders and Vb the gas total volume of the 24
reconstructed bubbles of the wet foam.

The measured value of liquid frnction of the wet foam is 1.10 %, which is in good
agreement with the values used by Phelan et al. for their numerical foams [8].

5. A Coarsening Foam

A polydisperse foam was generated from an aqueous solution of SDS (0.1 % wt) am
dodecanol (0.003 % wt); its density p, surface tension y, and dynamic shear viscosity Jl
are 1.01 g.cm-3, 40.0 erg.cm-2 and 0.99 cP, respectively. The films of SDS am
dodecanol in these proportions are so rigid that they evolve immediately into black
films, extremely stable in time. The foam is then easy to scan.

Foam coarsening could be followed during 30 hours. The foam topology did not
change at least for 9 hours (the scanning was not done ovemight).This is consistent
with the global results of Prud'homme and Khan who monitored the decay of the same
foam [12].

Figure 11 displays a same foam slice of the 9 and 24 hours aged foam. At 24
hours, some small bubbles have appeared and gas has diffused; no sharp vertex can be
seen on the right part of the image b. Actually, there is no more bubble at this height
because of film ruptures.

\.- '

I /
/

'·'

a
Figure 11. A foam slice at 9 hours (a) and at 24 hours (b).
373
The foam has been reconstructed at 8 different times 0, 1, 3, 6, 9, 24, 27 and 30
hours after its creation. Some constraints on foam boundaries have been applied for the
minimisation
(i) vertices with 4 intersecting edges, i.e. that do not belong to the boundary of the
reconstructed foam were allowed to move slightly.
(ii) vertices, edges and faces in contact with the porous plate are only allowed to move
in the plate plane.

In Fig. 12, the reconstructed 3D foam at 9 hours and 24 hours is presented. Until 9
hours, no change of foam structure is observed; it consists of 5 layers of bubbles among
which 48 bubbles are reconstructed, which correspond to 412 faces, 800 edges and 437
vertices. At 24 hours, small bubbles have appeared and that some films have broken. h1
fact, less than 5 bubbles layers still compose the foam; there are now 44 bubbles, 341
faces, 651 edges and 355 vertices.

Figure 12. The foam reconstructed at 9 hours (a) and at 24 hours after its creation (b).

5.1 RESULTS ON RECONS1RUCTED FOAMS

Two types of bubbles should be distinguished :


- the external bubbles that belong to the top layer or to the bottom layer in contact with
the porous plate.
-the internal bubbles that belong to the others layers (layers 2, 3 and 4 before 9 hours).
When foam ages the number of internal bubbles much decreases, and at t=30h, all
belong to a layer separated from the porous plate by only one bubble thick layer.

5.1.1 Evolution of foam topology

The discussion is focused on the statistics of the topological elements with time.
The distributions are given for internal bubbles in Figs. 13 a, b where n is the number
374

of edges per face, and f the number of faces per bubble. The foam disorder can be
appreciated by the second moment of the faces distribution
J.l2,f = (f2) - (f)2. (11)
V aloes are reported in Table 1.

DistributiOD of faces per bubble (t) Distribuliao of edges per face (t)

0~,------------------. 0,9

0,8

0,7

0,6

~ O,S
c. 0,4

0,3

0,2
0,1

0
3 4 6 7

(a) (b)
Figure 13. Topological distributions of internal bubbles: (a) number of faces per bubble; (b) number
of edges per face: ( <> ) :s; 9h; r? )
24h; (' ) 27h; ( 0 ) 30h. ~ ) edges distribution of the Kelvin cell ; (' )
edges distribution of the Weaire-Phelan cell.

Most of the faces are pentagonal and <n> = 5.11 ± 0.01 remains almost constant
with time : <n>, even in the destabilised foam, at 30h. Neither Kelvin nor Weaire-
Phelan structures were found in this foam although they are the more stable structures in
monodisperse foam (largest Isoperimetric Quotients).

TABLE 1. Distribution of the bubble topological elements.


time 01)
t:S96 t-246 t-276 t-306
# ail tlie bubbles 48 44 40 27
<f> 12.6 11.14 11.30 11.37
J.12,f 3.15 8.25 7.96 2.68
Hmternai &lbbles 28 12 12 6
<f>lnt 13.4 13.7 13.7 13.5
ll2,rno 1.81 0.89 0.89 0.92
# external bubbles 20 32 28 21
11.55 10.19 10.29 10.76
<f>ext
0.50 0.64 0.68 0.53
J.12,fext

<f> is more significant to appreciate the topological changes in 3D. There are two
bubbles populations, the internal and the external bubbles (Table 1). <f> solely
calculated on internal bubbles remains remarkably constant (13.5 ± 0.2), and it is in
375

=
good agreement with the literature [5, 7, 8, 14], whereas <f>ext 10.7 ± 0.8. For both
populations, the disorder is very low. When <f> and ~.r are calculated for all the
bubbles, their variations are much larger, and the disorder is higher at a given time. The
topological changes are likely due to the top surface and the porous plate influence.
Their influence does not extend far since <f>int =13.5 ± 0.2 even at 30h when it
remains only one bubble layer.

5.1.2 Evolution of bubble volume


The evolution of the volume distribution with time was calculated for all the bubbles,
internal and external (fable 2). The polydispersity is low at the beginning (small
standard deviation), and increasing with time because of coarsening.

TABLE2. Distribution of the bubble volumes.


tnne (h)
0 3 6 9 24 i7 3o
#bUbbles 48 48 48 48 44 44 40 27
<V(t)> (mm3) 59 59 59 58 57 59 49 49
standard deviation 14 15 15 17 17 30 23 19
(mm3)

5.1.3 Foamdynamics
The foam dynamics could be quantified by a 3D analogue of Von Neumann's law (Eq.
1), where the volume Vr of a f-faced bubble would be the equivalent in 3D of the area A,
of an-sided bubble in 2D. In the average

d(Vr} = k' (f- <f>) (12)


dt
ink' constant. Since <f> = 13.4, a bubble with less than 13.4 faces should shrink ani
at the opposite, a bubble with more than 13.4 faces shonld expand.

In Fig. 14, <Vr (t)> is plotted for the internal bubbles. The number of internal
bubbles fort~ 9h, and t =24, 27 and 30h, and the mean volume <V(t)> are reported in
Fig. 14. Before 9h, no ftlm rupture or face swapping occurred, and the bubble number
was constant. <V(t)> was nearly constant, the diffusion occurring between the bubbles
themselves since they are internal. When t > 9h, many film ruptures and face swappings
have been seen; as 2D Von Neumann's law is only valid at constant topology, relation
(12) is meaningless after 9h.
376

<Vf(t)>

100
I<V(t)> I
90

80

70

'iE 60
~

h 50
i>
!i
y 40

30
number of
20 bubbles

10

0
0 3 6 9 12 15 18 21 24 27 30

t (h)

Figure 14. Evolution of mean volumes of the f-faced internal bubbles : ~ ) f = 9; r) f = 12; {J )f
= B;f
)f= 14;P )f= 15;(-)f= 16.

Figure 14 shows that some tendency to <V ft)> equalisation occurs at the
beginning as in 2D foams [9]. Then coarsening becomes much more regular. Only
points at 1 s t s 9h are considered and the time origin is shifted by 1h. < Vt<t)> was

fitted by linear regression analysis and~dtvf) and k' were calculated for each f, assuming
that Eq. 12 is followed The slope sign changes between 13 and 14, i.e. near <f> and k'
is changing by a factor 2. Results are given in Table 3.

TABLE3. Distribution of the f-faced bubbles volumes.

#internal f- f d<Vc>idt k' k"


faced bubbles (104 mnr/s) (104 mnr/s) (10 5 mm2 /s)
9 -1.69 1.14 3.3
4 12 -0.48 1.33 2.8
11 13 -0.11 0.76 2.8
6 14 0.37 1.20 2.8
5 15 0.22 0.69 2.8
16 0.94 1.16 3.3
total= 28 <f> = 13.4
377

Since in 20 Von Neumann's law, k is a diffusion coefficient with the dimension


L 2 T 1 (Eq. 1), Glazier et al. proposed the following normalisation
1 dV:f
V f 3 - = k" (f- fo) (13)
dt
where k'' is now a diffusion coefficient. f 0 was determined numerically using a Potts
model and it is not equal to <f> but to 15.8 in [15].
Now, integmting (13), it comes in the avemge
2 2 2
(Vf(t))3 =3k'' (f - fo) t + (Vf(0))3. (14)
k" was calculated for each fby a mean square method. The best fit of Eq. 14 was found
for f0 = 13.4 = <f>. Results are reported in Table 3. k" is constant= 2.8 10"5 mm 2/s
=
except for f 9 and f =
16 for which k" =
3.3 10·5 mm 2/s. Based on the present
experimental results, the dynamics of a coarsening foam with rigid films is likely given
by the following law

(Vf r~
d(Vf} = k" (f - {f)). (15)
dt
As far as we know, this is the first experimental evidence of a 3D analogue of Von
Neumann's law.

6. Conclusion

Optical tomogmphy has proved to be a powerful non-destructive method to make


topological measurements on polyhedral and transparent real foams. The accuracy is
good enough to reconstruct the Plateau borders network of a wet foam; that yields its
liquid fraction, assuming that the liquid volume in the films is negligible. A
tridimensional topological transformation T1 was observed and quantified in terms of
area and volume. Two major results were obtained. The extend of the influence of the
wall on the foam is only one bubble thick and it decreases roughly the number of faces
where the mean face number of the internal bubbles is 13.4. The foam dynamics is
described by a analogue of the Von Neumann's law (Eq. 15).

Acknowledgements
We thank Dr B. Prunet-Foch for fruitful discussions. The partial financial suppor1 from the Centre
National des Etudes Spatiales is also acknowledged.

References
1. Glazier, J.A and Prause, B., Gonatas, C.P., Leigh, J.S. and Yodh, A.M. (1995) Magnetic resonance
images of coarsening inside a foam, Phys. Rev. Lett. 75, 573-578.
2. Durian, D.J., Weitz, D.A. and Pine, D.J. (1991) Dynamics and coarsening in three-dimensional foam
Science 252, 686.
378

3. Bisperink, C.G.J., Ronteltap, A.D. and Prins, A. (1992) Bubble-size distributions in foams Adv. Colloid
Interface Sci. 38, 13.
4. Muller, W., DiMeglio, J.-M. (1997) Avalanches in draining foams, Europhys. Lett., submitted.
5. Thomson, W. (1911) Mathematical and Physical Papers, Vol. V, p.297 Cambridge Univ. Press,
London/New York.
6. Princen, H.M. and Levinson, P. (1987) The surface area of Kelvin's minimal tetrakaidecabedron: the
ideal foam cell(?), J. Colloid Interface Sci. 120, 172-175.
7. Matzke, E. (1946) The three dimensional shape of bubbles in foam - an analysis of the role of surface
forces in three dimensional cell shape determination, Am. J. of Botany 33, 58.
8. Phelan, R., Weaire, D. and Brakke, K. (1995) Computation of equilibrium foam structures using the
Surface Evolver, Experimental Mathematics 4, 181-192.
9. Glazier, J.A. and Weaire, D. (1992) The kinetics of cellular patterns, J. Phys.: Condens. Matter 4, 1867-
1894.
10. Schwarz, H.W. (1965) Rearrangements in polyhedric foam, Recueil, 771-781.
11. Krugliakov, P.M., Exerowa, D.R. and Khristov, K. (1991), I.angmuir7, 1846.
12. G. Narsimhan and E. Ruckenstein, Foams : Theory, Measurements, and Applications, Eds.: R.K.
Prud'homme, S.A. Khan, Surfactant Sci. Series, vol57, Marcel Dekker, Inc. , New York (1996).
13. Weaire, D., Rivier, N. (1984) Soap, cells and statistics - random patterns in two dimensions, Contemp.
Phys. 25, 59-99.
14. Aste, T., Boose, D. and Rivier, N. (1996) From one cell to the whole froth: a dynamical map, , Phys.
Rev. E 53, 6181-6191.
15. Glazier, J.A. (1992) Grain growth in three dimensions depends on grain topology, Phys. Rev. Lett. 70,
2170-2173.

Surface Evolver is a surface minimisation software developed by K. Brakke to find an equilibrium


structure of an element under given conditions. It studies the surfaces shaped by surface tension and other
energies . Geomview is a graphic display software. Both are available via anonymous ftp from the server
geom.umn.edu, in the directory publsoftware.
THE GEOMETRY OF BUBBLES AND FOAMS

JOHN M. SULLIVAN
University of Illinois
Department of Mathematics
Urbana, IL, USA 61801-2975

Abstract. We consider mathematical models of bubbles, foams and froths,


as collections of surfaces which minimize area under volume constraints.
The resulting surfaces have constant mean curvature and an invariant
notion of equilibrium forces. The possible singularities are described by
Plateau's rules; this means that combinatorially a foam is dual to some
triangulation of space. We examine certain restrictions on the combina-
torics of triangulations and some useful ways to construct triangulations.
Finally, we examine particular structures, like the family of tetrahedrally
close-packed structures. These include the one used by Weaire and Phelan
in their counterexample to the Kelvin conjecture, and they all seem useful
for generating good equal-volume foams.

1. Introduction

This survey records, and expands on, the material presented in the author's
series of two lectures on "The Mathematics of Soap Films" at the NATO
School on Foams (Cargese, May 1997). The first lecture discussed the dif-
ferential geometry of constant mean curvature surfaces, while the second
covered the combinatorics of foams.
Soap films, bubble clusters, and foams and froths can be modeled math-
ematically as collections of surfaces which minimize their surface area sub-
ject to volume constraints. Each surface in such a solution has constant
mean curvature, and is thus called a CMC surface. In Section 2 we examine
a more general class of variational problems for surfaces, and then concen-
trate on CMC surfaces. Section 3 describes the balancing of forces within
a CMC surface. Then in Section 4 we look at existence results for bubble
379
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 379-402.
© 1999 Kluwer Academic Publishers.
380

clusters, and at the singularities in such clusters and in foams; these are
described by Plateau's rules.
These rules for singularities imply that a foam is combinatorially dual
to some triangulation of space, as explained in Section 5. Triangulations
of a surface are related through their Euler number to the topology of the
surface. In three dimensions, not as much can be said along these lines,
but we describe certain results in Section 6. Among the most interesting
triangulations, from the point of view of equal-volume foams, are those of
the tetrahedrally close-packed (TCP) structures from chemistry, described
in Section 7. The TCP structures observed in nature fit into an infinite
mathematical class; constructions for some infinite families are described
in Section 8. Finally, in Section 9, we discuss the application of these ideas
to Kelvin's problem of partitioning space.
The author is partially supported by NSF grant DMS-9727859, and
would like to thank Karsten GroBe-Brauckmann, Andy Kraynik, Rob Kus-
ner, Frank Morgan, Mike O'Keeffe and Nick Rivier for helpful conversa-
tions on foams and related structures, and J.-F. Sadoc for the invitation
to speak at the NATO school in Cargese. Some of the figures here were
originally created by the author to illustrate [26]. Fig. 2 was created by
GroBe-Brauckmann with software by Bernd Oberknapp.

2. Variational problems for surfaces

We want to examine the geometry of a surface which (like those in foams)


minimizes its surface area, given constraints on the enclosed volume. Re-
member that a surface in space has (at each point) two principal curvatures
k1 and k2; these are the minimum and maximum (over different tangent
directions) of the normal curvature. Because there is no way to globally
distinguish the two, only symmetric functions of the k1 and k2 are phys-
ically meaningful. The mean curvature is their sum H = k1 + k2 (twice
the average normal curvature), while the Gauft curvature is their product,
K = k1k2. The importance of K is that it is an intrinsic notion, meaning
that it doesn't change if we bend a surface without stretching it (like rolling
up a flat piece of paper). But our problems depend explicitly on the way
the surface sits in space-its extrinsic geometry-and thus we will find the
mean curvature H more useful.
Instead of minimizing the area of a surface, we might consider mini-
mizing a more general energy for surfaces which could include curvature
effects. The most natural candidate would be an integral (over the surface)
of symmetric quadratic terms in the principal curvatures k1 and k2 . Such
381
an energy E for a surface M can be written as

E(M) = JM (a+ b (H -H 0)2 + cK) dA,

where a, b, c and Ho are constants and dA is the area form on the surface.
By the Gaufi-Bonnet theorem, however, the integral of Kover M is just
a topological constant: it is 27rX on a closed surface of Euler characteristic x,
and, for a surface with boundary, depends additionally just on the boundary
data. Thus the final term in E stays constant as we vary M continuously, so
it is uninteresting for our variational problem. Furthermore, if the two sides
of the surface M are equivalent, then Ho must be zero, and we are left with
just two terms in E: a constant times the area of M, plus another times the
so-called Willmore energy W(M) := J H 2 dA. This scale-invariant elastic
bending energy W for surfaces [55, 19] has been proposed as a good model
for the geometry of certain biological membranes like lipid vesicles [30, 56,
20]. A surface which minimizes W subject to constraints fixing both area
and volume, for instance, can look quite like the usual shape of a red blood
cell.
In foams, however, the interfaces behave like soap films. Here, the cur-
vatures don't seem to affect the energy, so the only term we will consider
is surface tension (trying to minimize the area), perhaps balanced by a
pressure differential (fixing the volume to one side of the surface). A per-
turbation of a surface Misgiven by a vector-valued function iJ on M, telling
how each point is to move. Let us consider the first variations of area A
n
and volume V under this perturbation. If is the unit normal to M and
fi = Hn is the mean curvature vector, then

8V = JM iJ · ii dA, 8A = JM v· fi dA,
so we say that ii is the gradient of volume and fi is the gradient of area. If
we write H = 8A/ 8V, we obtain a second, variational, definition of mean
curvature.
These Euler-Lagrange equations show that a surface is critical for the
area functional exactly if H = 0. Such a surface is called a minimal surface,
and indeed any small patch of M is then area-minimizing for its boundary.
On a minimal surface, k1 = -k2, so necessarily K :S 0. Thus we see that
a minimal surface is saddle shaped, as in Fig. 1. Thus it is never convex,
meaning that a plane brought towards a patch of a minimal surface must
touch first at the boundary of the patch, not in the interior. More generally,
the (very useful) maximum principle says that the same happens whenever
any two minimal surfaces are brought together (so that one is never even
"relatively convex" compared to the other).
382

Figure 1. On the left, the least-area surface spanning the fixed edges of a cube. We see
that, as in any minimal surface, each point is saddle shaped, with equal and opposite
principal curvatures, and thus has negative Gaussian curvature. At the right, three min-
imal surfaces meet along a triple line at 120° . This surface would be achieved by soap
film spanning three wire Vs, and when extended by two-fold rotation about these lines
it forms the Kelvin foam.

If we realize that H = !:::,.Mii, then we can think of minimal surfaces as


a nonlinear version of harmonic functions. There is an extensive classical
theory of minimal surfaces; good introductions can be found in [38, 10, 17]
for instance. Much of the theory involves complete embedded minimal sur-
faces [18], and is based on the fact that the Gau:B map ii: M - 8 2 is
conformal; this leads to the Weierstra:B representation, which parameter-
izes M in terms of holomorphic data on a Riemann surface (and can be
reinterpreted in terms of spinors [25]).
A soap film spanning a wire boundary, with no pressure difference, will
be a minimal surface, but in general we need to consider the effect of a
volume constraint. In this case, we find that the surface has constant mean
curvature H = c, so it is called a CMC surface. Here the constant c is the
Lagrange multiplier for the volume constraint, which is exactly the pressure
difference across the surface; this relation H = t::.p is known as Laplace's
law.
We have seen that a CMC surface is a critical point for area A, given
the constraint of fixed volume V, with H being the Lagrange multiplier.
Equivalently, if we know the pressure difference H, the surface is a critical
point for A- HV. In physical bubble clusters or foams, each cell contains a
certain fixed amount of air or other gas (neglecting slow diffusion effects).
Of course, air is not an incompressible fluid. But for our mathematical
theory, we will usually pretend it is, fixing a volume (rather than a mass
of gas) in each cell. The point is that no matter what the relation between
pressure and volume is for the real gas, the equilibrium structures will be
the same for either problem.
383

For instance, a single bubble will adopt the form of a round sphere.
This isoperimetric theorem, that a sphere minimizes surface area for its en-
closed volume, was known to the ancient Greeks. (Indeed, a famous story
credits Queen Dido with making use of the two-dimensional version when
founding the city of Carthage.) But a proper mathematical proof (which
must demonstrate the existence of a minimizer rather than merely describ-
ing properties of one if it does exist) was only given in the last century by
Schwarz [44], using a symmetrization argument.
Note that the unit sphere (with H = 2) is an unstable critical point for
the functional A- 2V, even though it is the unique minimizer for area A,
given volume V = 47r /3. Although these problems have the same critical
points, as discussed above, the stability of a critical point may differ.

3. Forces in CMC Surfaces


One of the most important ideas in the mathematical study of CMC sur-
faces is that of force balancing, as described in the work ofKorevaar, Kusner
and Solomon [21]. In general, Noether's theorem says that symmetries of
an equation lead to conserved quantities. Here, the equation H = c for a
CMC surface is invariant under Euclidean motions (translations and rota-
tions) and the corresponding conserved quantities on a CMC surface can be
interpreted as forces and torques.
The physical interpretation is as follows. Start with a complete embed-
ded CMC surface M, and cut it along a closed curve 'Y· The two halves of
M pull on each other with a force due to surface tension and to pressure
(exerted across an arbitrary disk D spanning 'Y)· If ii is the normal to D,
and ij is the co normal to 1 in M (that is, the unit vector tangent to M but
perpendicular to 1), then the force can be written as

F(T) = i ij - H fv ii.
Here we think of pressure p = 0 outside M, and p = H inside M. Torques
around any origin can be defined in a similar fashion.
By Stokes' theorem, ff is independent of the choice of D. It is also
somewhat independent of 1, in the following sense. The fact that M is in
equilibrium, with surface tension balancing pressure differences, means that
no compact piece feels any net force or torque. If 1 and 1' are homologous
curves on M, their difference bounds such a compact piece, so their forces
must be equal. That is, F('Y) actually depends only on the homology class
of 'Y·
The CMC surfaces of revolution, called unduloids, were first studied by
Delaunay [9] in the last century. An unduloid is a wavy cylinder as in
384

Figure 2. Left, a typical unduloid (a CMC surface of revolution), whose force vector
points along the axis. llight, a typical triunduloid (embedded CMC surface with three
ends). Each end is asymptotic to an unduloid; the three force vectors sum to zero.

Fig. 2; the force vector, for a (homologically nontrivial) curve r around


the cylinder, necessarily points along the axis of revolution. The family of
unduloids (for given H) can be parameterized by the strength of this force,
which is maximum for a straight cylinder. In fact, Delaunay described the
generating curves for unduloids as solutions of a second-order ODE, and the
force is a first integral for this equation. As the force decreases from that of
a cylinder, we see alternating necks and bulges. The necks get smaller and
smaller, like shrinking catenoids, and in the limit of zero force, the unduloid
degenerates into a chain of spheres, as in a beaded necklace.
The idea of force balancing has been very useful in the classification
of complete embedded CMC surfaces. The only compact example is the
round sphere [1], with all forces equal to zero. In general, each end of
such a surface is asymptotic to an unduloid [21]. Thus is has a force vector
associated to it, directed along the axis of the unduloid, pulling the "center"
of the surface out towards infinity. The conservation of forces (the homology
invariance of F) means that these forces on the ends must sum to zero.
This balancing condition is the essential ingredient in classifying such CMC
surfaces, and combined with some spherical trigonometry can lead to a
complete classification of the three-ended surfaces [13, 14].
Of course, such infinite surfaces do not exist in nature; bubble clusters
and foams are collections of CMC surfaces meeting along singular lines as
described below in Section 4. We note that, if these singular lines were
known in advance, each piece of surface spanning them could be viewed as
a solution of the CMC Plateau problem [16] on its boundary. The ideas of
forces and torques carry over to these bubble clusters and foams. We might
most naturally cut a cluster M with a plane P (of normal ii); the force
exerted across this cut is

knM fi - k\M pn,


where p is the pressure in each region of the complement of the surface M
(again normalized to have p = 0 outside the cluster). For a finite cluster of
385

bubbles, these forces are always zero, since they are again constant as we
move the plane P, and are evidently zero once P fails to hit M.
In an infinite periodic foam, however, there can be non-zero forces, and
we can use them to calculate the stress tensor for the foam. This stress
tensor 0' shows how the lattice of periodicity for a periodic foam would like
to change (by stretching or shearing), and is usually computed (as in [40]
or [23]) as
1
0' =
Mo
I - llll
-+-+T
'

plus a multiple of the identity matrix which depends on the choice of ab-
solute zero pressure. Here Mo is one fundamental domain for the periodic
foam, and v at any point is the normal vector to Mo. We claim that this
tensor can also be computed in terms of the forces. Essentially, given any
n,
vector the force when M 0 is cut by a plane with normal n gives u(fi). We
will use these force or stress calculations later to help minimize the total
area of an equal-volume periodic foam of given combinatorics, by adjusting
its lattice parameters.

4. Bubble Clusters and Singularities in Soap Films


So far, we have been rather cavalier about assuming that solutions to our
problem (of minimizing surface area with constraints) exist and are smooth
surfaces (hence of constant mean curvature). Of course, the surfaces in a
foam or bubble cluster are only piecewise smooth, and the pieces meet in
the singularities observed by Plateau [39].
Almgren proposed the following bubble cluster problem: given volumes
Vi, ... , Vk, enclose and separate regions in space with these volumes, using
the least total area. He then proved that a minimizer exists and is a smooth
surface almost everywhere [2] (see also [31]).
The existence proof uses a direct approach which takes the limit of a
sequence of surfaces whose area approaches the infimal value. The difficulty
here (and one reason the theorem is hard even for the case of a single region)
is that surfaces of decreasing area can have growing "filigree". Perhaps the
nth surface Mn looks like the minimizer, but with a (bent or even branched)
tube of length n and radius 1/n attached. (There is a nice figure in [32,
p. 5].) The areas of these surfaces decrease to the proper minimum value,
but every point in space might be a limit of the filigree. To prove the
theorem, Almgren used a more sophisticated notion of limit surface from
geometric measure theory, which automatically discards the filigree.
This suggests that the proof does not give any control on the topology
of the regions; in fact we are not allowed even to assume that each region
is connected. For all we know, it may be that the best way to make a
386

>

Figure 3. The X junction at the left is not an (M, e:, a)-minimizing set in the plane
because it can deform (at an arbitrarily small scale) to the double Y at the right. In the
plane, the Y is the only singularity allowed.

cluster with some four fixed volumes is to split one of them into two pieces
(with the correct total volume), and make what looks like a cluster of five
bubbles. A reasonable conjecture, of course, is that in any minimizer, each
region is connected.
Almgren's result does, however, say something about the shape of the
minimizers: they are (M, c, &)-minimizing sets. This means, by definition,
that a Lipschitz deformation of space in a small ball (of radius r < 6)
can reduce the area of the set at most by a factor of 1 + c(r) (where
c --t 0 as r --t 0). If there are no volume constraints, a soap film, even with
singularities, is (M, 0, &)-minimizing. The function c(r) is a clever way to
allow for non-zero mean curvature while not expanding the allowed set of
singularities for soap films.
In the plane, an (M, c, d)-minimizing set is a finite network of smooth
curves meeting in threes at equal 120° angles. Note that the deformation
(shown in Fig. 3) of an X into a double Y (which is the Steiner tree on the
four vertices of a square) shows that the X is not minimizing-it can be
improved at any small scale. (Of course, the models we use here are meant
to describe dry foams; in a wet foam with positive liquid fraction <p > 0,
there are Plateau borders near each singularity. In this case one might
wonder if additional singularities might be possible. Although a wet X is
stable for large <p, Brakke and Morgan [4] have shown that it is unstable for
small positive <p. The analogous question is three dimensions is still open.)
In space, Taylor was able to show [49] that the singularities observed em-
pirically in soap films by Plateau are the only ones possible in an (M, c, 8)-
minimizing set. That is, such a minimizer consists of a finite union of smooth
surfaces, meeting in threes at 120° dihedral angles along a finite number
of smooth curves; these curves in turn meet at finitely many tetrahedral
corners (which look like the soap film obtained when dipping a tetrahe-
dral frame into soapy water-six sheets come together along four triple
curves into the central singularity). The proof relies on ruling out seven
other candidate cones, like that on the twelve edges of a cube; if we dip a
387

Figure 4. At the left we see the standard cluster of three soap bubbles, with all interfaces
spherical, presumably minimizing total area for its enclosed volumes. At the right we see
a possible competitor for the least-area double bubble, where one region is a toroidal belt
around the other.

wireframe cube into soap, the film formed breaks the symmetry, inserting
a small square at the center whose four vertices are all tetrahedral corners.
Similarly, the other candidate cones all break apart into configurations of
less area with several tetrahedral corners.
We thus know that the minimizing bubble clusters guaranteed by Alm-
gren's theorem follow Plateau 's rules, having just these possible singular-
ities, as seen in the standard triple bubble in Fig. 4, left . Actually, we
know more: here each smooth piece has constant mean curvature, and the
mean curvatures (being pressure differences) satisfy a cocycle condition.
This means that if we add up the mean curvatures of each surface we cross,
as we follow a closed loop in space, the total is zero. Equivalently, numbers
(call them pressures) can be assigned to each component of the comple-
ment such that the mean curvature of each interface is the difference of
these pressures. Although we cannot guarantee that each region (of fixed
volume) is a single component, the different components making up a single
region will necessarily have the same pressure.
Of course a minimizing cluster of just one region will be a round sphere.
It is surprising, though, how little is known about other small clusters. The
case of two regions, a double bubble, is one of the geometrical problems where
"many mathematicians believe, and all physicists know," that the obvious
solution is correct. 1 The best double bubble should be the one seen when
playing with soap bubbles, with three spherical caps meeting along a single
circular triple junction. Surprisingly, this is not yet a theorem. However,
1 The quote is from Rogers (42] , describing Kepler's sphere packing problem.
388
Hutchings showed (following a suggestion of White) that the mm1m1zer
must have rotational symmetry, and gave further conditions, which, in the
case of equal volumes left only one family of competitors, as in Fig. 4 (right),
allowing Hass and Schlafl.y to rule them out with a computer search [15]
(see also [33]).
For other small clusters, the author has suggested [48] that the min-
imizer for up to four regions in any volume ratios should be a standard
cluster, obtained by stereographic projection of a simplex in S 3 ; here the
interfaces are all pieces of spheres and the triple junctions are all circular
arcs. Perhaps even all stable clusters of up to five bubbles are spherical
in this sense, but there is evidently a non-spherical cluster of six bubbles
(shown in the author's illustration on the cover of [32]). Many more con-
jectures along these lines can be found in [48].
Almgren's theorem holds for finite clusters of soap bubbles in Euclidean
space or certain other symmetric manifolds. Mathematically we would like
to consider a foam as an infinite cluster. If we ask that the foam be peri-
odic, then it is equivalently a finite cluster of bubbles in a torus (IR3 mod the
lattice of periodicity), so Almgren's theory does apply, and we get a min-
imizer. But for the general case of infinite foams (even with equal-volume
cells) very little is known, though Morgan has made some progress on the
two-dimensional case [34].

5. Combinatorics of Foams
We will define a foam mathematically as a (locally finite) collection of CMC
surfaces, meeting according to Plateau's rules, and satisfying the cocycle
condition: pressures can be assigned to each component of the complement
so that the mean curvature of each interface is the pressure difference. We
will call the components of the complement the cells of the foam, call the
interfaces between them simply the faces, call the triple junction lines where
they meet the (Plateau) borders, and call the tetrahedral singularities the
corners.
As we have mentioned, finite clusters minimizing area for fixed volumes
are known to have this structure. Although there are no analogous results
about infinite clusters (except when they are periodic) this is a reasonable
definition for foams, and we expect it to describe all physical foams whose
shape is determined by surface tension.
For many purposes, we can ignore the geometric parts of Plateau's rules,
and only pay attention to the combinatorial aspects of the foam's cell com-
plex. The combinatorial rules mean that the dual cell complex to a foam is
in fact a simplicial complex, that is, a triangulation of space. To construct
this dual, we put a vertex in each cell of the foam. Vertices in a pair of
389

Figure 5. The heavy lines show a sample Voronoi partition in the plane, for the sites
marked with dots. The thin lines are the edges of the dual Delone triangulation, which
connect sites with adjacent Voronoi cells.

adjacent cells are connected by an edge, which is dual to a face of the foam.
Where three faces come together along a border, we span the three corre-
sponding edges with a triangle; and where four borders come together at a
corner, we fill in a tetrahedron.
It is possible that in a foam, a cell or face might fail to be simply con-
nected; or perhaps some face might have the same region on both sides or
might meet itself along a triple junction. To avoid such technical compli-
cations, we will henceforth deal only with simple foams, where each cell
(whether open, including only the interior, or closed, including the bound-
ary) is a ball, and each (open or closed) face is a disk, and each border is
an interval. Dually, these conditions mean, for instance, that no two tetra-
hedra intersect in more than one triangle; these conditions are often part
of the definition of a triangulation. In any case, the foams we care about
are all simple, having cells which approximate convex polyhedra.
Combinatorially, the Plateau rules mean that a (simple) foam and its
dual triangulation are like a Voronoi decomposition of space and the dual
Delane triangulation. Given a set of sites in space, the Voronoi cell [45, 35]
for each site is the convex polyhedron consisting of points in space closer
to that site than to any other. If the sites are in generic position (with no
five on a common sphere) then the dual Delane complex (whose vertices
are the original sites) is completely triangulated; each Delone tetrahedron
has the property that no other sites are inside its circumscribed sphere. An
example in the plane is shown in Fig. 5.
This similarity suggests that that we might look for foams as relaxations
of Voronoi decompositions. For instance, we can give a modern interpreta-
tion to Kelvin's construction [50] of his candidate for a least-area partition
of space into equal volume cells as follows. Start with sites in the body-
390

centered cubic (Bee) lattice. Their Voronoi cells are truncated octahedra,
packed to fill space. If we let the films in this packing relax (until the
geometric parts as well as the combinatorial parts of Plateau's rules are
satisfied) we should get a periodic foam, the one proposed by Kelvin.
Starting from other periodic arrays of sites, we can carry out the same
procedure to generate a foam. Mathematically, there is no theory for the
relaxation step, which would follow a mean-curvature flow for surfaces with
triple junctions and volume constraints. But in practice this procedure is
a useful way to compute foam geometries. The author's vcs software for
computing Voronoi diagrams [47] in three dimensions can send its out-
put to Brakke's Surface Evolver [5]. This latter program can interactively
manipulate, refine and evolve triangulated surfaces into energy-minimizing
configurations, modeling the physical effects of surface tension and pressure.
New symmetry features in the Evolver [6] allow computation of just one
fundamental domain in a foam of arbitrary symmetry, thus giving greater
accuracy for given computational effort.

6. Bounds on the combinatorics of triangulations


Consider a triangulation of any compact three-dimensional manifold, and
let V, E, F and T be the numbers of vertices, edges, triangles and tetrahe-
dra, respectively. Let z be the average number of edges at a vertex (which
is the average number of faces on a cell of the dual foam), and n be the
average valence of an edge (the average degree-or number of borders-of
a face in the foam). Then a multiple-counting argument shows that
4T = 2F, 3F =nE, 2E = zV.
The Euler characteristic x of a triangulation is the alternating sum of
the number of simplices in different dimensions. For a triangulation of a
surface, x = V- E + F, while multiple counting shows 3F = 2E = zV.
It follows that 6/ z = 1 + 2x/ F, meaning that the sign of x determines the
sign of 6 - z: triangulations of a sphere have z < 6, those of a torus have
z = 6, and those of a surface of higher genus have z > 6.
Returning, however, to our three-dimensional triangulation, we know
that the Euler characteristic of any compact 3-manifold is zero: 0 = x =
V - E + F - T. (In general, the Euler characteristic distinguishes differ-
ent manifolds only in even dimensions.) Combining this with the multiple-
counting results gives
6- n = 12/z.
Unlike the case of surfaces, where z was related via x to the topology of the
manifold, equation (*) relates z to n independent of the ambient manifold.
We have been talking about triangulations of compact manifolds, but a
391

periodic triangulation of an infinite manifold (like Euclidean or hyperbolic


space) covers a triangulation of the compact quotient manifold, so it will
satisfy the same relation (*).
For the regular triangulations of spherical space 8 3 , n and z will be
integers: (n,z) = (3,4) or (4,6) or (5,12). We can consider the regular
ideal triangulation of hyperbolic space as having (n, z) = (6, oo). Of course,
a regular Euclidean tetrahedron cannot tile fiat space; its dihedral angle is
i-
arccos ~ 70.53°, so we would need

211'
n = no := 1 ~ 5.1043
arccos 3

in such a tiling. This would correspond by (*) to

z = zo := 12/(6- no) ~ 13.397

which is 411' divided by the measure of the solid angle at the vertex of a
regular tetrahedron.
i-,
A regular tetrahedron has dihedral angle arccos and if we view average
dihedral angle as a function on the space of tetrahedra, by symmetry it has
a critical point at this regular tetrahedron (as the Shoemakers [46] have
pointed out-see Section 7). Thus, if we tile Euclidean space by nearly
regular tetrahedra, we expect the average dihedral angle to be very close
i-·
to arccos But the geometric part of Plateau's rules says that each corner
in a foam has exactly tetrahedral angles. So, if the borders in the foam
are not too curved, the dual triangulation should consist of nearly regular
tetrahedra. Thus we expect n ~ no, so also z ~ z0 by (*). 2
Another way to think about this is to consider a face in a foam. If the
face is nearly fiat, and its borders are nearly straight, then its equal angles
of arccos( -1/3) ~ 109.5° suggest that it should be an n 0 -gon. Or, rather,
that perhaps pentagons (and hexagons) might be favored in a foam. This
has led many people to conjecture [41, 54, 8] that efficient foams should
have many pentagonal faces, unlike the ordered Kelvin foam, but more like
the disordered equal-volume foams observed in nature [28].
In the case of a foam with equal-pressure cells, all the interfaces are
minimal surfaces. Here, Kusner [24] has given a clever argument to show
that we must have z ~ zo and n ~ no, by applying the Gauf3-Bonnet
theorem to the boundary of each cell in the foam. The corners of the cell
each contribute a known quantum of Gauf3 curvature (equal to 411' /(2zo -4))
because of their tetrahedral angles. The borders contribute an unknown

2 Note that Coxeter [8] outlines a different argument (involving Petrie polygons) which
suggests that n ~ (13 + V313)/6 ~ 5.115 and hence z ~ 13.56.
392

amount of curvature to each of the three cells they bound, but the equal
120° dihedral angles along each border imply that these amounts sum to
zero, and can be neglected in the average. Finally, the faces, being minimal
surfaces, contribute a negative amount of curvature. We see that the average
cell (having total curvature 471") must thus have at least 2zo - 4 corners, or
equivalently at least zo faces.
Choe showed that, given any compact manifold, in its universal cover
it has a fundamental domain of least boundary area [7]. For us, this means
that if we look at any of the ten crystallographic groups that act freely on
JR3 (see for instance [51, §4.3]), there is a least-area foam with a single type
of cell, tiling space by that group action. For instance, if the group con-
tains merely translations, the quotient manifold is a torus, and, no matter
which lattice parameters we choose, this least-area Choe foam presumably
is combinatorially equivalent to Kelvin's foam. In particular, if we choose
the BCC lattice, the conjecture [48] is that this foam is exactly Kelvin's
foam. It seems likely also that the Williams foam [54] is the Choe foam for
its symmetry group (which involves one screw motion with a 90° twist).
For any of the ten groups (including four with nonorientable quotients)
Choe's construction gives a foam with congruent equal-pressure cells, so by
Kusner's result, each cell must have at least 14 faces. We know of no other
examples of equal-pressure foams.
From the examples of regular triangulations, it is tempting to think
that somehow, triangulations with higher values of z (or, equivalently, n)
should tile hyperbolic manifolds, while those with lower values should tile
spherical manifolds. Of course, we know that the Euler number can give no
such information in three dimensions, but Luo and Stong [27] have given
one result in this direction. They show that any triangulation with z < 8
(n < 9/2) must be a triangulation of (some quotient of) 8 3 • And if z = 8,
the only other possibility is (some quotient of) 8 2 x lll
However, this seems to be the only topological information that can be
extracted from z, because simple refinement constructions change the value
of z for a triangulation. We can apply these iteratively (as shown in [27])
to get a sequence of triangulations with arbitrary z > 8 in the limit, on any
three-manifold.
One construction refines a tetrahedron by adding a vertex in its interior,
replacing it by four tetrahedra. (Dually, we blow a small tetrahedral bubble
at one corner in the foam.) This replaces z by a value closer to 8. For
instance, O'Keeffe [36] has noted that if we start with the Kelvin foam
(or sodalite structure) and decorate each corner with a new tetrahedral
bubble, 3 we change z from 14 to 62/7, already close to 8.
3 0'Keeffe also proposes packing spheres at the corners of this "decorated sodalite" to
give a sphere packing of least density.
393

That first construction suffices to reduce z. A second construction refines


an edge of the triangulation, by adding a vertex at its midpoint. (Dually
we inflate one fact of the foam to form a new prismatic bubble.) This
construction applied repeatedly to a k-fold edge brings z close to 2 + 2k.
Since an edge of high degree can be created in any triangulation by the first
construction, we can then achieve arbitrarily high z values this way.
As we have seen, with no restrictions on the triangulation, a value of
z > 8 tells us nothing about the topology of the ambient manifold. Perhaps,
however, if we first repeatedly reversed our two constructions, popping all
tetrahedral and prismatic bubbles in a foam, the value of z for the resulting
simpler triangulation might give some topological information. Presumably
also, knowledge that every edge has valence no more than 5 (stronger than
the assumption n ~ 5) is enough to conclude that the manifold is spherical.
Aste, Boose and Rivier [3] define a quantity n*, similar to n, and relate
this to z with an intriguing argument which looks at the growth of the
number of bubbles in layers outward from a central bubble in the foam.
In terms of the triangulation, we can outline their idea as follows. Number
the vertices such that all neighbors of a vertex labeled t - 1 are labeled at
most t. For instance, we might start with one vertex labeled 0, then label
its neighbors 1, and so on. Any edge, triangle or tetrahedron now has its
vertices all labeled t or t + 1 for some t. We will say it has type tk if exactly
k of the vertices are labeled t.
The main assumption of [3] now is that there are no tetrahedra of type
t 4 . This should mean that the t 3 triangles fit together into a surface Lt,
called a layer of the triangulation. Edges of type t 1 go from this layer to
the next, and we let n* be the average valence of such edges. Another
assumption is that this average is independent of t, or at least quickly
approaches a limit value n* as t increases. Finally, we assume also that the
topology of each layer is bounded (so that we don't see increasing numbers
of handles as t -+ oo). Then the argument of [3] sets up a recursion for
the number of edges and vertices within Lt. Assuming that these numbers
grow quadratically with t (rather than growing exponentially or staying
bounded) in a triangulation of Euclidean space, the result is that
8
z=6+--4 .
n*-
As we increase z, this equation shows that n* decreases, while (*) shows
that n increases. Naively, we might expect n = n* in most foams, since it
is hard to see why edges between layers should be different from those
within layers. By the monotonicity of n and n* means that there is only
one solution ton= n*, namely
n = (10 + 2VJ)/3 ~ 5.097, z = 8 + 2VJ ~ 13.29;
394

note that these values are less than no and zo. Actually, most known foams
(including all the TCP foams of Section 7) have values greater than these,
suggesting that they have n* < n.
For the BCC triangulation dual to Kelvin's foam (which has z = 14 and
n = 5t ), if we start by giving label 0 to all vertices in the z = 0 plane,
then we find label 1 is assigned both to the shifted vertices at z = 1 and
to the vertices directly above at z = 2. Moving upwards, each label t > 0
is assigned to such a "double layer". This means that there are many t 4
tetrahedra, so [3] does not apply, and indeed n* = 5i does not satisfy their
equation.
But if we instead start with label 0 on all vertices in the diagonal plane
x + y + z = 0, then parallel planes have increasing t, and everything works
out. For this layering, n* = 5, which agrees with z = 14. Note also that
if we start with a single cell numbered 0, then as we move outward, the
layers look like octahedra. That is, they grow to essentially behave like the
diagonal layers with n* = 5. We might hope that other periodic foams will
have the same property. That is, if we start with a single cell, and build
successive layers around it, perhaps they will naturally find the appropriate
directions, to give a value of n* fitting with the prediction of [3].
In this context, we would like to suggest a reason why it might be natural
to expect n* < n. In a slightly distorted regular tetrahedron, we expect that
longer edges will tend to have greater dihedral angles, and hence a smaller
valences. However, longer edges are also more likely to reach into a new
layer, that is, to enter in the calculation of n*. If they have smaller than
average valence, then indeed n* < n, as observed for the Kelvin foam and
predicted for any foam with z > 13.3.

7. TCP structures

As we have observed, the tetrahedral angle arccos( -1/3) ~ 109.47°, found


at foam corners, is between the average angle of a pentagon (108°) and that
of a hexagon (120°). We are led to consider foams with only pentagonal or
hexagonal faces.
Chemists have studied transition metal alloys in which the atoms pack
in nearly regular tetrahedra. These tetrahedrally close-packed (TCP) struc-
tures were first described by Frank and Kasper [11, 12], and have been
studied extensively by the Shoemakers [46] among others. In all of these
structures, the Voronoi cell of each atom has one of four combinatorial
types; these are exactly the four polyhedra which have only pentagonal
and hexagonal faces, with no adjacent hexagons. Mathematically, we can
define of TCP structures in this way [26]: triangulations whose combinato-
rial duals (called TCP foams) have only these four types of cells. Since this
395

Figure 6. There are four types of cells found in TCP foams, each with 12 pentagonal
faces like the dodecahedron at the left. The remaining three have in addition two, three
or four hexagonal faces, arranged antipodally, equatorially, and tetrahedrally.

definition deals only with the combinatorics of the triangulation, presum-


ably it allows some examples with too much geometric distortion to work
chemically as TCP structures.
Each of the four types of cells found in TCP foams has 12 pentagonal
faces, as seen in Fig. 6. One type is the (pentagonal) dodecahedron. Cells of
the other three types have additionally two, three, or four hexagonal faces,
which are arranged antipodally, equatorially, or tetrahedrally (respectively).
The 14-hedron can be viewed as a "~-fold unwrapping" of a dodecahedron,
since it has six-fold symmetry through the centers of the opposite hexagons.
Perhaps some geometric condition on the tetrahedra (saying that their
edge lengths or dihedral angles are close to equal) would guarantee that
the dual cells could only be one of the four TCP types. Note however, that
edge lengths differ less in BCC tetrahedra (where they are 2 and v'3) than
in the TCP structure A15 (where they are 2, y'5 and v'6).
It is an interesting open question just what foams are possible with
these four types of cells. Dually, we are asking for triangulations of space
in which each edge has valence five or six, and no triangle has two edges of
valence six. We know we cannot use only dodecahedra, since a foam made
of these alone (meeting tetrahedrally) will fill a spherical space, not JR3 .
This seems to be the only mathematical theorem restricting how these
cells can fit together, but the chemists have made interesting additional ob-
servations. We can define three basic periodic TCP structures. A15, observed
for instance in Cr3Si, has one 12-hedron and three 14-hedra in a fundamen-
tal domain. Z, observed in Zr4Al3, has three 12-hedra, two 14-hedra, and
two 15-hedra. And C15, observed by Friauf and Laves in MgCu 2 , has two
12-hedra and one 16-hedron. (These are shown in Fig. 7 and described in
more detail in Section 8 below.)
If we describe potential TCP structures by the ratio of cells they have of
the four types, we could plot them all within an abstract tetrahedron, as
convex combinations of the four vertices. But the observation ofYarmolyuk
and Kripyakevich [57] is that all known TCP structures are, in fact, convex
combinations of the three basic ones just mentioned, so they all get plotted
396

Figure 7. The three basic TCP structures. A15, left, has two 12-hedra (one upper-left)
and six 14-hedra (in colums like lower right) in a cubic cell. Z, center, has three 12s
(top), two 14s (in vertical columns but separated in the figure, with unequal horizontal
hexagons), and three 15s (with vertical hexagons, one lower front). C15, right, has four
12s and four 16s (with hexagons darkened) in a cubic cell.
z

J1.
M
.P
I 0
XT R /1

C15
C14 SM A15
Figure 8. The basic TCP structures A15, Z, and C15 lie at the corners of this triangle,
where the horizontal axis plots the average number of faces per cell (ranging from 13~
to 13t) and the vertical axis plots the fraction of cells which are 15-hedra (ranging from
0 to ¥>·Every known TCP structure, when described by its numbers of 12-, 14-, 15-, and
16-sided cells, is a convex combination of these three. Distinct structures may appear at
the same point, but need not then have the same properties.

within the triangle of Fig. 8. As we mentioned before, the Shoemakers [46]


attempt to explain this observation by noting that even for a tiling by
tetrahedra somewhat distorted from regular, we expect to have z ~ z0 .
Certainly the three basic TCP structures, with z = 13~ for A15, z = 13¥
for Z, and z = 13~ for C15, are close to this value. We will describe in
the next section how to construct some convex combinations of these basic
structures; the resulting foams will also have z ~ z0 .
It is interesting to note that there are other chemical structures, which
exhibit the structure of the TCP Voronoi cells (that is, the dual TCP foams)
more explicitly. These are described in detail by O'Keeffe [36], but we out-
line the ideas here. In clathrates, large gas molecules are trapped (at the lo-
cation of the TCP sites) inside water cages: oxygen atoms sit at the Voronoi
corners, bonded by hydrogen along the borders (Voronoi edges). For in-
stance, chlorine hydrate (Ch)6-46H20, has the structure ofVoronoi cells for
A15, called a Type I clathrate. Many salt hydrates have instead the struc-
397
ture of the Z foam (called Type III), while others use non-TOP structures
like the BOO cells of the Kelvin foam. Some zeolites have similar structures,
with Si (and possibly some Al) atoms at the corners, bonded tetrahedrally
to oxygen atoms along the borders. For instance sodalite Na4AhSh012Cl
has the BOO foam structure, with Cl atoms at the Voronoi sites, and Na in
the centers of the hexagonal faces. Sodium silicide, when heated properly,
can generate silicon cages in the A15 or C15 foam pattern (Type I or Type II
clathrates) with sodium trapped inside. It should be noted, however, that
in other zeolites the 4-connected silicate nets are less dense and do not close
up into cages. (See [37] for more information on many interesting crystal
structures, and [29] for instance for the structures of zeolites.)

8. Some constructions for TCP foams

Perhaps the most common partition of space into equal-volume cells is by


cubes, which form the Voronoi cells for the simple cubic lattice. Of course
this is a degenerate case of the Voronoi construction, and the cells fail to
meet tetrahedrally, so this does not make a good foam. These degenerate
cubic corners (at the "holes" in the lattice) themselves form another cubic
lattice. The two taken together form the BOO lattice. This structure, of
course is the one whose Voronoi cells give rise to the Kelvin foam.
We might consider now repeating the process, adding some holes as new
sites. The BOO lattice has only one kind of Voronoi corner (or hole). If the
lattice is scaled so neighboring sites are at distance 4 and 2v'3 then the hole
is at a distance of v's from the nearest sites. Adjacent corners are much
closer to each other, at distance J2, so if we added them all as sites, we
would not be close to having regular Delane tetrahedra. But because all the
faces in the foam have even numbers of sides (4 or 6), these corners can be
colored black and white so that adjacent corners are different colors. If we
look now just at black corners, adjacent ones are at distance 2 or v'6. The
structure with Si atoms at the BOO lattice points, and Cr atoms at these
black corners, is the A15 structure of Cr3Si. Its Voronoi decomposition has
cells of two types (12- and 14-hedra) and forms the TOP foam of Weaire
and Phelan [53], corresponding to a Type I clathrate.
If we start with the face-centered cubic lattice, it has rhombic dodeca-
hedra for its Voronoi cells. The corners include the deep holes and (up to
lattice translation) two kinds of shallow holes. The original lattice together
with one kind of shallow hole forms the diamond network. If we take all
its points as sites and repeat the Voronoi construction, the new corners in-
clude all the other original holes, plus certain new holes. The C15 structure
for MgCu2 has Mg atoms in the diamond network and Cu atoms at these
new holes. Its Voronoi diagram in turn is the C15 foam, corresponding to
398


Figure 9. Here are two examples (u, left, and H, right) of the construction for TCP
structures from tilings by squares and triangles. In each case we see only a fundamental
domain. The thick and thin lines correspond to the red and blue edges in the text. Small
dots are sites at height 4k + 1, large dots are sites at height 4k - 1, and the square and
triangle vertices have sites at height 2k.

a Type II clathrate. We could construct Z in a similar fashion, filling in


certain holes in a lattice of hexagonal prisms.
The work of the Frank and Kasper suggests the following construction
for infinitely many new TCP structures, as convex combinations of A15
and Z. Consider an arbitrary tiling of the plane by squares and equilateral
triangles. (The regular square 42 and triangular 36 tilings will give A15
and Z, respectively.) We will describe the location of sites in space for a
corresponding TCP structure. Over each vertex we will find a vertical stack
of 14-hedral cells, sharing hexagons, and centered at even integer heights.
Each edge of the tiling makes an angle some multiple of 30° with the
horizontal. We mark it red or blue depending on whether this is an odd or
even multiple, as in Fig. 9. The edge colors are the same around any triangle,
but alternate around a square. At odd heights we find two different kinds
of layers, with centers of 12-, 14-, and 15-hedral cells. The layers at height
4k + 1 have sites at the midpoints of the blue edges, at the centers of the
red triangles, and also halfway between the center of any square and each
of its red edges. The layers at height 4k - 1 are constructed in the same
fashion, after interchanging red and blue.
There are infinitely many possibilities here; if the tiling we start with
has triangles and squares in the ratio 2a : b, then the foam obtained has
12-, 14- and 15-hedral cells in the ratio 3a + 2b : 2a + 6b : 2a, which
is just the appropriate convex combination of the ratios for Z and A15.
Periodic tilings lead to periodic structures; the simplest ones have been
observed in nature as TCP structures. For instance, the semi-regular tiling
with vertices 33 42 (which alternates layers of squares and triangles) gives
the H structure, while the one with vertices 32 434 (the snub square tiling)
gives the u structure.
Sadoc and Mosseri [43] have suggested a refinement procedure for a
three-dimensional triangulation which does not distort the shapes of the
399

tetrahedra very much. Applying this procedure to a TCP structure leads to


another one. For instance, starting from the A15 structure, they derived
the TCP structure called SM, which at the time was unknown in nature,
although it is quite similar to the T structure previously observed.
Their construction adds two new vertices along each edge of the trian-
gulation (trisecting the edges), and one new vertex at the center of each
tetrahedron. New edges are drawn from this center vertex to the trisecting
vertices and also to the centers of adjacent tetrahedra.
Each tetrahedron is divided into four small tetrahedra (at its corners)
and a truncated tetrahedron, which is coned from its center to form four
more tetrahedra and four hexagonal pyramids. When the hexagonal pyra-
mids from two adjacent tetrahedra are combined into a bipyramid, the edge
connecting the two center vertices divides it into six tetrahedra.
Compared to other ways to refine three-dimensional triangulations, this
construction preserves the shape of tetrahedra quite well. If we start with
a triangulation having two adjacent regular tetrahedra (with side 3), then
some of the tetrahedra created in this construction are regular (with side
1), but others are somewhat irregular, have some edges of length JIT78
or fi/2. Of course, adjusting the positions of the new vertices might lead
to slightly less variation in new edge lengths. But no matter how nice a
triangulation we start with, if we apply this construction repeatedly, it
seems we may be forced to get tetrahedra further and further from regular.
Thus the suggestion of [43], that such repeated refinement applied to a
regular triangulation of 8 3 might give a Euclidean TCP structure, seems
questionable.

9. Kelvin's problem

Over one hundred years ago, Kelvin proposed the problem of partitioning
space into equal-volume cells using the least interface area per cell [50]. He
suggested that the solution might be what we have been calling the Kelvin
foam, a relaxation of the Voronoi diagram for the BCC lattice. Although
Weaire and Phelan now have a better partition, Kelvin's is still conjec-
tured [48] to be the best if the cells are required to be congruent or to have
equal pressure. Mathematically, there is no theory to suggest that such a
best infinite cluster should exist, but we expect that it will, and will have
the structure of a foam.
Weaire and Phelan [53] were the first to consider using TCP structures
as foams. Although they initially thought of these as models for wet foams,
they quickly discovered that the (dry) Al5 foam was a more efficient parti-
tion of space than Kelvin's candidate. To go beyond their good numerical
evidence (from Brakke's Evolver) and give a rigorous proof that their foam
400
is better than Kelvin's, we need to give a bound on how much the Kelvin
foam can relax.
Although, as we have mentioned, there is no general mathematical the-
ory for the relaxation step in constructing foams from Voronoi cells, for
Kelvin's foam there is enough symmetry that we need only consider mean-
curvature flow on a single surface. In Fig. 1, right, we saw a symmetric unit
of Kelvin's foam, bounded by lines of rotational symmetry. In that picture,
the vertical sheet is in a mirror plane, so we need only solve for one of the
other sheets, a minimal surface with two fixed boundary lines and one free
boundary with 120° contact angle. We can use this analysis to show that a
unique foam exists in Kelvin's pattern, and a slicing argument then gives
a lower bound on its area [26]. This bound, it turns out, suffices to prove
that even the unrelaxed Weaire-Phelan A15 foam beats Kelvin.
Given this example, it is natural to look for good equal-volume par-
titions among the other TCP foams. Rivier [41] proposed that since C15
has (among the known TCP structures) the lowest z and thus the highest
proportion of pentagons, it might give an even better partition. Instead,
computer experiments [22] suggest that among all TCP foams, A15 is the
most efficient, and C15 among the least. (Here, we are looking at equal-
volume foams. For each pattern (like Z) without cubic symmetry, we adjust
the lattice parameters to get rid of any stress tensor; this ensures that we
have the most efficient partition in that pattern.) Probably the 12-hedral
and 16-hedral cells of C15 naturally have such different sizes that distorting
them to make the volumes equal ruins whatever advantage pentagons give.
(If we considered a modified Kelvin problem where one-third of the cells
were to have somewhat larger volume, then presumably the C15 structure
would do very well.) The author has also made computer experiments with
equal-volume foams generated from other chemical structures; the foam
from ')'-brass, for instance, is better than that from C15, despite having
even some triangular faces. Perhaps pentagons are not as desirable in foams
as has been assumed.

References
1. Alexandrov, A. D.: 1958, 'Uniqueness Theorems for Surfaces in the Large, I'. Vestnik
Leningrad Univ. Math. 19(13), 5-8. English transl. in Amer. Math. Soc. Transl.
(Ser. 2) 21 (1962), 412-416.
2. Almgren, Jr., F. J.: 1976, 'Existence and Regularity Almost Everywhere of Solutions
to Elliptic Variational Problems with Constraints'. Mem. Amer. Math. Soc. 4(165).
3. Aste, T., D. Boose, and N. llivier: 1996, 'From One Cell to the Whole Froth: A
Dynamical Map'. Phys. Rev. 53, 6181-6191.
4. Brakke, K. and F. Morgan: 1996, 'Instability of the Wet X Soap Film'. Preprint.
5. Brakke, K. A.: 1992, 'The Surface Evolver'. Exper. Math. 1(2), 141-165.
6. Brakke, K. A. and J. M. Sullivan: 1997, 'Using Symmetry Features of the Surface
Evolver to Study Foams'. In: K. Polthier and H.-C. Hege (eds.): Visualization and
401

Mathematics. Heidelberg, pp. 95-117.


7. Choe, J.: 1989, 'On the Existence and Regularity of Fundamental Domains with
Least Boundary Area'. J. Diff. Geom. 29, 623-663.
8. Coxeter, H. S. M.: 1958, 'Close-Packing and Froth'. Ill. J. Math. 2(4B), 746-758.
Reprinted in (52].
9. Delaunay, C.: 1841, 'Sur la surface de revolution, dont la courbure moyenne est
constante'. Journal de mathematiques 6, 309-320.
10. Dierkes, U., S. Hildebrandt, A. Kiister, and 0. Wohlrab: 1992, Minimal Surfaces
I, Vol. 295 of Grundlehren der Mathematischen Wissenschaften. Berlin: Springer-
Verlag.
11. Frank, F. C. and J. S. Kasper: 1958, 'Complex Alloy Structures Regarded as Sphere
Packings. I. Definitions and Basic Principles'. Acta Crystall. 11, 184-190.
12. Frank, F. C. and J. S. Kasper: 1959, 'Complex Alloy Structures Regarded as Sphere
Packings. II. Analysis and Classification of Representative Structures'. Acta Crys-
tall. 12, 483-499.
13. Grof3e-Brauckmann, K., R. Kusner, and J. M. Sullivan: 1997, 'Classification of
Embedded Constant Mean Curvature Surfaces with Genus Zero and Three Ends'.
GANG Preprint IV.29, UMass.
14. Grof3e-Brauckmann, K., R. Kusner, and J. M. Sullivan: 1998, 'Constant Mean Cur-
vature Surfaces with Cylindrical Ends'. To appear in the Springer proceedings of
VisMath'97.
15. Hass, J., M. Hutchings, and R. Schlafl.y: 1995, 'The Double Bubble Conjecture'.
Electron. Res. Announc. Amer. Math. Soc. 1(3), 98-102.
16. Hildebrandt, S.: 1970, 'On the Plateau problem for surfaces of constant mean cur-
vature'. Comm. Pure Appl. Math. 23, 97-114.
17. Hildebrandt, S. and A. Tromba: 1996, The Parsimonious Universe. New York:
Copernicus.
18. Hoffman, D. and W. H. Meeks, III: 1990, 'Embedded Minimal Surfaces of Finite
Topology'. Ann. of Math. 131(1), 1-34.
19. Hsu, L., R. Kusner, and J. M. Sullivan: 1992, 'Minimizing the Squared Mean Curva-
ture Integral for Surfaces in Space Forms'. Experimental Mathematics 1(3), 191-207.
20. Jiilicher, F., U. Seifert, and R. Lipowsky: 1993, 'Conformal Degeneracy and Con-
formal Diffusion of Vesicles'. Phys. Rev. Lett. 11, 452-455.
21. Korevaar, N., R. Kusner, and B. Solomon: 1989, 'The Structure of Complete Em-
bedded Surfaces with Constant Mean Curvature'. J. Diff. Geom. 30, 465-503.
22. Kraynik, A. M., R. Kusner, R. Phelan, and J. M. Sullivan, 'TCP Structures as
Equal-Volume Foams'. In preparation.
23. Kraynik, A.M. and D. A. Reinelt: 1996, 'Elastic-Plastic Behavior of a Kelvin Foam'.
Forma 11(3), 255-270. Reprinted in (52].
24. Kusner, R.: 1992, 'The Number of Faces in a Minimal Foam'. Proc. R. Soc. Lond.
439, 683-686.
25. Kusner, R. and N. Schmitt: 1996, 'On the Spinor Representation of Minimal Sur-
faces'. GANG preprint III.27, UMass.
26. Kusner, R. and J. M. Sullivan: 1996, 'Comparing the Weaire-Phelan Equal-Volume
Foam to Kelvin's Foam'. Forma 11(3), 233-242. Reprinted in [52].
27. Luo, F. and R. Stong: 1993, 'Combinatorics of Triangulations of 3-Manifolds'. Trans.
Amer. Math. Soc. 337(2), 891-906.
28. Matzke, E. B.: 1946, 'The Three-Dimensional Shape of Bubbles in Foam'. Amer.
J. Botany 33, 58-80.
29. Meier, W. M. and D. H. Olson: 1992, Atlas of Zeolite Structure Types. Butterworths,
3rd edition.
30. Michalet, X. and D. Bensimon: 1995, 'Observations of Stable Shapes and Conformal
Diffusion in Genus 2 Vesicles'. Science 269, 666-668.
31. Morgan, F.: 1994, 'Clusters Minimizing Area Plus Length of Singular Curves'. Math.
Ann. 299(4), 697-714.
402

32. Morgan, F.: 1995a, A Beginner's Guide to Geometric Measure Theory. Academic
Press, 2nd edition.
33. Morgan, F.: 1995b, 'The Double Soap Bubble Conjecture'. MAA FOCUS pp. 6-7.
Dec. 1995.
34. Morgan, F.: 1996, 'The Hexagonal Honeycomb Conjecture'. Preprint.
35. Okabe, A., B. Boots, and K. Sugihara: 1992, Spatial Tessellations: Concepts and
Applications of Voronoi Diagrams. Wiley & Sons.
36. O'Keeffe, M.: 1997, 'Crystal Structures as Periodic Foams and vice versa'. Appearing
in this volume.
37. O'Keeffe, M. and B. G. Hyde: 1996, Crystal Structures I: Patterns and Symmetry.
Washington: Mineral Soc. Amer.
38. Osserman, R.: 1986, A Survey of Minimal Surfaces. New York: Dover Publications,
2nd edition.
39. Plateau, J.: 1873, Statique Experimentale et Theorique des Liquides Soumis aux
Seules Forces Moleculaires. Paris: Gauthier-villars.
40. Reinelt, D. A. and A. M. Kraynik: 1993, 'Large Elastic Deformations of Three-
Dimensional Foams and Highly Concentrated Emulsions'. J. of Colloid and Interface
Science 159, 46Q-470.
41. Rivier, N.: 1994, 'Kelvin's Conjecture on Minimal Froths and the Counter-Example
of Weaire and Phelan'. Europhys. Lett. 7(6), 523-528.
42. Rogers, C. A.: 1958, 'The Packing of Equal Spheres'. Proc. London Math. Soc. 8,
609-620.
43. Sadoc, J.-F. and R. Mosseri: 1982, 'Order and Disorder in Amorphous Tetrahedrally
Coordintaed Semiconductors: A Curved-Space Description'. Philos. Mag. 45, 467.
44. Schwarz, H. A.: 1884, 'Beweis des Satzes, dass die Kugel kleinere Oberflache besitzt,
als jeder andere Korper gleichen Volumnes'. Nach. Ges. Wiss. Gottingen pp. 1-13.
Reprinted in 1972 in Gesammelte mathematische Abhandlungen, pp. II.327-340,
New York: Chelsea.
45. Senechal, M.: 1990, Crystalline Symmetries. Adam Hilger.
46. Shoemaker, D. P. and C. B. Shoemaker: 1986, 'Concerning the Relative Numbers of
Atomic Coordination Types in Tetrahedrally Close Packed Metal Structures'. Acta
Crystall. 42, 3-11.
47. Sullivan, J. M.: 1988, 'The vcs Software for Computing Voronoi Diagrams'. Avail-
able by email from jmsCimath. uiuc. edu.
48. Sullivan, J. M. and F. Morgan (Editors): 1996, 'Open Problems in Soap Bubble
Geometry: Posed at the Burlington Mathfest in August 1995'. International J. of
Math. 7(6), 833-842.
49. Taylor, J. E.: 1976, 'The Structure of Singularities in Soap-Bubble-Like and Soap-
Film-Like Minimal Surfaces'. Ann. of Math. 103, 489-539.
50. Thompson, Sir W. (Lord Kelvin): 1887, 'On the Division of Space with Minimum
Partitional Area'. Philos. Mag. 24, 503-514. Also published in Acta Math. 11,
121-134, and reprinted in (52).
51. Thurston, W. P.: 1997, Three-Dimensional Geometry and Topology, Vol. 1. Prince-
ton. Edited by Silvio Levy.
52. Weaire, D. (ed.): 1997, The Kelvin Problem. Taylor & Francis.
53. Weaire, D. and R. Phelan: 1994, 'A Counter-Example to Kelvin's Conjecture on
Minimal Surfaces'. Phil. Mag. Lett. 69(2), 107-110. Reprinted in (52).
54. Williams, R. E.: 1968, 'Space Filling Polyhedron: Its Relation to Aggregates of Soap
Bubbles, Plant Cells, and Metal Crystallites'. Science 161, 276-277.
55. Willmore, T. J.: 1992, 'A Survey on Willmore Immersions'. In: Geometry and
Topology of Submanifolds, IV (Leuven, 1991}. pp. 11-16.
56. Wintz, W., H.-G. Dobereiner, and U. Seifert: 1996, 'Starfish Vesicles'. Europhys.
Lett. 33, 403-408.
57. Yarmolyuk, Y. P. and P. I. Kripyakevich: 1974. Kristallographiya 19, 539-545.
Translated in Sov. Phys. Crystallogr. 19, 334-337.
CRYSTAL STRUCTURES AS PERIODIC FOAMS AND VICE VERSA

M. O'KEEFFE
Department of Chemistry
Arizona State University
Tempe AZ 85287, USA

One of the great lessons of condensed matter physics is


that nature is more fertile than the human imagination in
devising ways for matter to organize itself into coherent
structures. -D. C. Wright & N. D. Mermin [1]

1. Introduction

There has been a remarkable symbiosis between crystal chemistry and the study of cellular
structures such as that of foams. This was already foreshadowed by Kelvin in that the
structure, originally proposed by him as the division of space with minimal surface area
(and hence as the structure of a minimal energy foam of equal bubbles), also plays a
prominent role in his discourse On the Molecular Tactics of a Crystal [2]. The Kelvin
structure is, of course, derived from a space-filling packing of truncated octahedra and is
better known to the crystal chemist as the framework of the sodalite structure. In this
context it is considered the prototype of a series of framework structures built up of
packings of simple polyhedra (those with three faces meeting at each vertex) generically
known in the case of silicates as clathrasils, but of wide occurrence in crystal chemistry in
other contexts as well.
The body-centered cubic sphere packing corresponds to the division of space into
congruent tetrahedra [3] and the vertices of the Kelvin structure are at the centers of these
tetrahedra. Conversely the centers of the polyhedra of the Kelvin structure are at the
centers of the spheres in the body-centered sphere packing, and it is appropriate to refer to
each of the structures as the dual of the other (the vertices of one structure correspond to
the centers of the polyhedra of the other). More generally there is a dual relationship
between sphere packings in which space is divided into tetrahedra and 4-connected
structures derived as a packing of simple polyhedra whose vertices are in the tetrahedral
holes of the sphere packing. These play a central role in crystal chemistry and in this
article I mention just a few of the more important, and attempt to provide a concordance
between the different terms used in various disciplines.

2. Homogeneous sphere packings with tetrahedral holes and their duals

In homogeneous sphere packings all spheres are related by symmetry; as a consequence


the structure is periodic. If further the sphere centers divide space into tetrahedra, the dual
structure is a packing of congruent polyhedra. Such structures are described first.

2.1 THE BODY-CENTERED CUBIC AND SODALITE STRUCTURES

Most readers will be entirely familiar with the body-centered cubic (bee) sphere packing,
nevertheless some aspects are illustrated in Fig. 1. The figure illustrates on the left, one
403
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 403-422.
© 1999 Kluwer Academic Publishers.
404

tetrahedron around a tetrahedral hole, in the center how four tetrahedra combine to form an
octahedron, and on the right how six octahedra (24 tetrahedra) combine to form the
coordination polyhedron which is a rhombic dodecahedron with 14 vertices. Notice that
the coordination number of bee is considered to be Z = 14. In any sphere packing in
which space is divided up into tetrahedra it is easy to show that the number of tetrahedra
per sphere is equal to <Z>/2-1 where <Z> is the average coordination number in the
sphere packing, so in the case of bee there are six tetrahedral holes per sphere.
The bee structure is that of some metallic elements such as Fe (at room temperature
and pressure) and W (often cited as the typical example). Some hundreds of binary
compounds have the /3-brass (CuZn) structure in which different atoms are at the cell
comers and center.

Fig. 1. Body-centered cubic packing showing, at left, a tetrahedral hole and, at right, the coordination
polyhedron (rhombic dodecahedron) consisting of 24 tetrahedra.

Fig. 2 shows a fragment the dual structure; on the left as a packing of truncated
octahedra and on the right as a 4-connected net. Notice that the polyhedra have six square
and eight hexagonal faces; this is registered in the face symbol [46.6 8 ]. The net is
accordingly made up of 4- and 6-rings and this is reflected in the vertex symbol [3]
4-4·6·6·6-6. A quantity of interest in the dual structure is the average number of edges of
the faces of the polyhedra (average ring size), <n> = 6- 12/<Z>. Clearly this corresponds
to the average number of tetrahedra meeting with a common edge in the sphere packing.
In the bee structure <n> = 3617 = 5.14. It is interesting to compare this number with the
number obtained by dividing a circle by the dihedral angle of a regular tetrahedron, i.e.
2n/cos- 1(113) =5.10.

Fig. 2. The sodalite net. Left: as a packing of truncated octahedra, and right: as a 4-connected net ..

The mineral sodalite has ideal formula Na4[Al3Si3]012CI. The atoms in brackets are
at the positions of the vertices of the polyhedron packing and links -0- correspond to the
polyhedron edges; the Na and Cl atoms fill the cavities in the cage. The oxygen atoms of
405

the water molecules in the hydrate HPF6·6H20 are again in the positions of the vertices
of the Kelvin structure and now -H- bonds correspond to the edges. Another example of a
crystal structure built around the same theme is that of BaPd2P4 [4] in which the Pd and P
atoms are at the polyhedron vertices and Ba atoms are at the center of the polyhedra.

2.2 OTHER SPACE FILLING PACKINGS OF CONGRUENT POLYHEDRA

2.2.1 Packings of "twisted Kelvin" polyhedra


If the body-centered cubic sphere packing is distorted to body-centered tetragonal (bet) with
cia= ...f(2/3) =0.816 the (110) layers now become close packed (cp) with each sphere in
contact with six others [3]. These cp layers are stacked so that each sphere has two
neighbors in the layer above and in the layer below so that each sphere is in contact with
10 neighbors. In the packing there are 4 additional neighbors so the coordination is
written 10+4, and space is still divided by the sphere centers into tetrahedra; the dual
structure is of course just a tetragonal distortion of the sodalite structure (just this
distortion is observed in BaPd2P4 mentioned above).
However, there is more than one way of packing adjacent cp layers to produce 10+4
coordination and a second way is found in chemistry as the orthorhombic (symmetry
Fddd) y.Pu structure [3]. The dual structure corresponds to a new 4-connected net that
also corresponds to a space-filling of congruent polyhedra (here "congruence" includes
being identical to the mirror image). In BaCu2P4 [5] the Cu and P atoms are in the
tetrahedral holes of the Fddd sphere packing of Ba atoms. The polyhedron in the new
structure is illustrated in Fig. 3. It has been identified as space-filling and discussed in the
context of foams by Weaire and Phelan [6] (who named it the "twisted Kelvin cell") and
by Aste, Boose and Rivier [7] . The polyhedron is chiral (symmetry D2 222) and in the =
Fddd structure both left- (l) and right-hand (r) forms of the polyhedron occur in equal
amounts with layers alternating lr .... It transpires [8] that there is a second packing llrr...
with the same symmetry in which again all polyhedra are congruent; and there is also a
packing (symmetry P~22 or P6422) in which all polyhedra are of the same hand (r or l).
Notice that the "twisted Kelvin" polyhedron still has 14 faces but now some are
pentagons: the face symbol is [44.54.66].

Fig. 3. Left a layer of the bet structure showing spheres close-packed in the layer. On the extreme left the
distorted truncated octahedron ("Kelvin polyhedron") of the dual structure is shown. On the right a similar
layer is shown for a structure derived from a packing of "twisted Kelvin polyhedra".

2.2.2 The Williams' structures.


Some years ago Williams [9] described two variations of the sodalite (Kelvin) structure
that correspond to space filling by congruent polyhedra. These are well known (see e.g.
[3]) and not described here except to remark that the polyhedra of the Williams' packings
both have 14 faces and face symbols [44.54.66] and [42.58.64]. The second polyhedron,
406
which Williams called the "/3-tetrakaidecahedron", has a distribution of face sizes more in
accord with observations on model foams and other cellular structures [9].

2.2.3 A fifth variant of the Kelvin polyhedron and intergrowth structures


Yet another 14-face polyhedron is known to fill space [8]. The symmetry is Cs m and, =
in common with the twisted Kelvin and the first of Williams' polyhedra, this has face
symbol [44 .5 4 .66] . Part of a layer of a space-filling by this polyhedron is shown in Fig.
4. The packing of spheres is orthorhombic (symmetry Pnma) and again the sphere
packing is ( 10+4)-coordinated. I have not yet found an example of this structure in crystal
chemistry.
There are many possibilities for the five space-filling polyhedra described above to
combine to fill space. I have described some of the simpler of these elsewhere [8].

Fig. 4. Part of a layer of the space filling described in § 2.2.3.

2.2.4 The ABR polyhedron packing


The space-filling polyhedra described so far all have 14 faces and 24 vertices and the
average number of edges/face is <n> =3617. It appears that a space-filling polyhedron that
produces a 4-connected structure will have at least 14 faces [10], but one would like to
know if a larger space-filling polyhedron is possible. It is of interest therefore that Aste,
Boose and Rivier (ABR) [7] have described a 16-face space-filling polyhedron with larger
average face side (<n> =2114 ). A configuration of the polyhedron is shown in Fig. 5; it
may be seen that this shape does not appear to be a likely for a bubble in a stable froth .
However, the network of the packing (symmetry P42/mcm) is plausible for a chemical
compound, and in fact it corresponds to the Cr,Si part of the ThCr2Si2 structure (the most
populous of all ternary structure types) with one quarter of the Cr positions empty [11].

Fig. 5. Left: two views of the ABR polyhedron. Center and right: the corresponding space-filling packing.

3. Tetrahedral packings involving two or more kinds of sphere

In the development of an arbitrary packing of spheres (not necessarily of equal size), in


407

general each added sphere will come into contact with three spheres of the existing
assembly, and their centers together will form a tetrahedron. In contrast, in some regular
periodic patterns, holes of larger size may occur in the packing. Conspicuous in this
context are the closest sphere packings in which separate octahedral as well as tetrahedral
holes are prominent [3] (contrast bee in which the "octahedral holes" are really aggregates
of four tetrahedral holes-Fig. 1). Closest packings and their relation to crystal structures
were already discussed by Barlow and others more than 100 years ago, and their
importance in crystal chemistry is well documented [3]; here I focus on periodic sphere
packings in which space is divided up into tetrahedra-these are referred to as
"tetrahedrally packed" (tp). A subset of these structures is referred to as "tetrahedrally close
packed" (tcp) and they serve as the basis for some of the more important intermetallic
structure types. The dual structures are equally of importance as the basis of structures of
low coordination such as framework aluminosilicates and covalent materials.

3.1 COORDINATION POLYHEDRA IN TP STRUCURES AND THEIR DUALS

In tp structures it is best to determine coordination numbers from a consideration of the


dual structures. The dual structure will be a packing of simple polyhedra (now in general
of more than one kind) whose vertices are in the tetrahedral holes of the structure. Each
polyhedron will enclose one sphere center and all points closer to that center than to any
other; it is the Voronoi polyhedron associated with that point. A simple and logical
definition [12] of coordination number is now the number of faces of the Voronoi
polyhedron. (Notice that this definition leads to the assignment of 14 = 8 + 6 as the
coordination number in the bee sphere packing). S~heres whose centers share a face of a
Voronoi polyhedron are considered to be neighbors.

~~~$··
€?®@~$~
9 (Z) 10 (Y) 12 (X) 14 (R) 15 (Q) 16 (P)

Fig. 6. Top row: Some coordination polyhedra in tp structures (numbers are coordination numbers = numbers
of vertices). Those for coordination numbers 12-16 are the Frank-Kasper polyhedra. Bottom row: the
polyhedra dual to those on the top row. The numbers are now the numbers of faces.

It follows from the above discussion that the neighbors of an atom in a tp sphere
packing form coordination polyhedra with triangular faces and their duals are simple
polyhedra. The most likely coordination polyhedra are those in which either m or m+ 1
triangles meet at each vertex (their duals are simple polyhedra with faces that are m- or

1 In some packings of low coordination number (which do not concern us here), one may wish to
distinguish "direct" and "indirect" neighbors . A pair of neighboring atoms are direct neighbors if the straight
line joining them goes through a face common to their two Voronoi polyhedra [13].
408

m+ 1-gons ). The cases for m = 3 or 4 can be realized as convex polyhedra with equilateral
triangular faces (often called deltahedra) and exist for any number between 4-12 vertices
with the exception of 11 [3]. The polyhedra with 9, 10 and 12 vertices and their duals are
shown in Fig. 6. The labeling Z, Y and X is used to identify the polyhedra below.
The polyhedra for m = 5 are of special interest; if it is required that vertices at which
six triangles meet are adjacent only to vertices at which five triangles meet, there are only
four possibilities: those with 12 (again), 14, 15 or 16 vertices. These are often referred as
the X, R, P and Q polyhedra respectively. I use the same letters below to refer to the
corresponding dual polyhedra with 20, 24, 26 and 28 vertices. The polyhedra and their
duals are also shown in Fig. 6. They are, of course, the celebrated Frank-Kasper
coordination polyhedra [12]. Their duals are all the simple polyhedra with pentagonal
faces (there are exactly twelve such faces) and hexagonal faces, in which hexagonal faces
share edges only with pentagonal faces.2

3.2 THE FRANK-KASPER (TCP) STRUCTURES AND THEIR DUALS

The tp structures with coordination numbers 12-16 are especially important in crystal
chemistry, they are often also named for Frank and Kasper (FK) whose analyzed their
geometries in a classic series of papers [12]. Many aspects of these structures, in
particular their generation from packings of regular tetrahedra in curved space, have been
discussed at length elsewhere [e.g. 15-17]. The standard reference for intermetallic
structures in general remains Pearson's book [18]. A more recent review of the
compositions of known phases is that of Shoemaker and Shoemaker [19] for an update
see e.g. [17]. Here there is just room for some general comments.
In the illustrations of the FK polyhedra in Fig. 6, heavy lines are drawn joining the
centers of the R, P and Q polyhedra to the 6-coordinated vertices. It transpires that these
vertices are always in turn at the centers of other R, P or Q polyhedra so that in all FK
structures the network of such lines form what FK call the major skeleton of the structure
[12]. In the terminology of Sadoc [15] this is the network of disclination lines of the
structure. As indicated below, the skeleton is sufficient to specify the topology of the
structure.

IS
14
F K tetrahedron YK triangle

Fig. 7. Left: the tetrahedral space corresponding to all possible combinations of Frank-Kasper (FK)
polyhedra for coordination numbers 12, 14, IS and 16. Right: the triangular cross section corresponding to all
known structures according to Yarmolyuk and Kripyakevich (YK). The symbols are explained in the text..

Unnoticed by FK was the fact that all the known structures have a remarkably small
range of average coordination number, and it was probably Yarmolyuk and Kripyakevich

2The julie rene polyhedra are the infinite class of simple polyhedra with pentagonal faces (again exactly
twelve of them) and hexagonal faces but now in which the pentagons shares edges only with hexagons (the
isolated pentagon rule) [3, 14].
409

(YK) who first [20] called attention to the fact that the coordination polyhedra occur in
fixed ratios which may be derived as linear combinations of three basic compositions
which in fact occur in the three simplest FK structures (see Fig. 7) here called I, II and III
and described briefly next. An implication of this observation is that the average
coordination number in all phases is in the narrow range of 13.33-13.50 [19].
The labels I, II and III are those applied3 to the clathrates4 dual to the FK structures.
Very many clathrate hydrates have the cubic type I structure, including chlorine hydrate
(discovered by Davy in 1811) and xenon hydrate with ideal formulas XR3·23H20 in which
X refers to a cavity formed by 20 water molecules and R refers to a cavity formed by 24
water molecules (see Fig. 6). There are also many Type II hydrates (also cubic) such as
that of chloroform, the ideal composition is X2P·l7H20 where now Prefers to a cavity
formed by 28 water molecules. Jeffrey [22] lists almost 100 molecules known to form
type I and/or type II hydrates. Known compounds with the type III structure are salt
hydrates and are not strictly clathrates, because the anion (in the simplest case OH-)
forms part of the enclosing network. The ideal formula is now X3R2Q2AOT where T
refers to a combination of water and anion. Some type I and type II salt hydrates are also
known.
A fourth hydrate, type IV, has an ideal formula intermediate between those of types I
and III, and is shown on the right of Fig. 7 as "cr". The ideal formula is now X5R8Q2·86T
(= 2xl + III). Bromine hydrate has been found to have this structure.
The type I structure is also found in nature as the Si structure of an impure form of
silica known as melanophlogite, and the type II structure occurs in the silica framework
structure known as dodecasil-3C and ZSM-39. The codes for these two structures in the
Atlas of Zeolite Structure Types [24] are MEP and MTN respectively.
These structures have also been known for many years as the silicon frameworks of
alkali silicides, particularly due to the work of the Bordeaux group [25]. The ideal formula
for the type I compound is Na4Si 23 and isostructural compounds (with atoms in bold
forming the clathrate framework) include [26]: ~Si23• ~Ge23• ~Sn23• Cs4Sn23• and
Ge23-xl4+x· Fewer compounds have been reported with the type II structure which has
ideal formula Na3Si 17 • There has been a recent resurgence of interest in the silicon
clathrate structures due in part to the fact that they (and their C and Ge analogs) have been
shown to be only slightly metastable (ca. 0.1 eV/atom) with respect to the diamond form,
and that the metal-free material is predicted to have a larger band gap than the normal
(diamond-structure) form [27].
The type I structure is of course, the structure that provided the famous counter-
example ofWeaire and Phelan [28] that disproved Kelvin's conjecture as to the structure
of the lowest energy foam of equal size bubbles, and is discussed in several places in this
volume in that context.
The FK structures dual to the clathrates types I, II and III have formulas XR 3, X 2P
and X 3R2 Q2 where now X, R, Q and Prefer to 12-, 14-, 15- and 16-coordinated atoms

3The labels I, II, III and IV are used by O'Keeffe and Hyde [3) following Wells [21] [who actually used
the symbols (i), (ii), (iii) and (iv)]. On the other hand Jeffrey [22], to whom we owe much of our knowledge
of the clathrate hydrates, labels these four structures I, II, IV and III (i.e. III and IV interchanged).
4The term "clathrate" was introduced by Powell [23) who discussed 'cage structures of suitable form
[which] imprison molecules of a second kind .. .It is suggested that they be named "clathrate" compounds.'
[the word "clathrate" itself come from the Greek word for bars (as in a grating)]. It is an amusing comment
on the modem penchant for euphemism that the "imprisoned" molecules are now often referred to as
"guests", although the term "guest" is more properly reserved for molecules that are free to come and go (as
in true zeolites). Purists do not consider clathrasils to be zeolites.
410

(notice that the numbers of H20 molecules, or more generally T species, in the formulas
for the clathrates is now the number of tetrahedral holes per formula unit).
The type compound for the XR3 structure is Cr3Si and several hundred intermetallic
compositions are listed under this heading in Pearson's Handbook [26]. Niobium
compounds (Nb3Ge, Nb3Sn etc.) with this structure are the basis of the most important
practical superconducting materials. The structure was once ascribed to a form of
elemental W (the so-called /3-W) but it was shown many years ago [29] that this
assignment was highly unlikely, and that "/3-W" was much more likely to be w 3o.
However the error led to the structure being ascribed a Strukturbericht symbol
appropriate for an element, viz. A 15 (the "A" signifying an elemental structure) and this
has become very common usage. It is unfortunate however that this obscures the fact that
the structure is only formed for a combination of elements of significantly different size. 5
Several aspects of the structure and its dual are shown in Fig. 8. Notice that in e.g.
Cr3Si, there are non-intersecting rods of Cr atoms running in the three <100> directions.
These rods are also the lines joining the 6-coordinated vertices of the FK coordination
polyhedra and thus form the major skeleton of the structure. If one wanted to make a
"spaghetti" model of the dual structure all that is necessary to know is this major
skeleton. As the vertices on the skeleton are 2-connected the polyhedra are all type R (14
faces, see Fig. 6) sharing hexagonal faces. When such a model is completed (it is easily
made using tetrahedral vertices and equal connecting links [3]) it will be found that the
space not occupied by the R polyhedra are pentagonal dodecahedra (duals of the X
polyhedra). Notice that the polyhedra of the dual structure cover space just once. In
contrast, the polyhedra of the original tp structure are interpenetrating and it is not easy to
illustrate the structure while simultaneously illustrating the coordination polyhedra of
both constituents. A convenient structural formula is XR3·T23 signifying that, per
formula unit of the sphere packing, there are one 12-coordinated (X) sphere, three 14-
coordinated (R) spheres and 23 tetrahedral holes (1).

r~
6.-::: .... ,.....~·.:::0 m~
~-······~

Fig. 8. The type I FK structure and its dual clathrate. Left: a unit cell of Cr3Si showing Si (open circles) and
the rods of Cr (filled circles). The next two sketches show a 12-coordination polyhedron around one Si and a
14-coordination polyhedron around one Cr. The next figure to the right shows some of the 14-face polyhedra
of the dual structure, and the last sketch shows the corresponding 12-face polyhedra. The polyhedra of the
last two parts combine to fill space.

The corresponding type compound for the X2P structure (dual to the type II hydrate)
is MgCu2; Pearson's Handbook [26] lists many hundreds of compounds under this
heading (actually under "Cu2Mg") and this is one of the most populous of all binary
crystal structure types. It is also known as the C15 structure ("C' is the Strukturbericht
designation for binary compounds with elements in the ratio 2:1). The X2P combination
of coordination polyhedra is also known for a number of polytypes, the simplest of which

5 Although relative size is important, a chemist will find it significant that in all A 15 compounds A38, the
element A is an early transition element (i.e. from one of columns 4, 5 or 6 of the periodic table).
411

is the hexagonal Cl4 structure of MgZn 2 (the type compound of another large family). A
third polytype is known as MgNi2. Collectively the family of various polytypes with
X2P coordination is known as the Laves phases.
The major skeleton of MgCu2 is the diamond structure, and again this allows a
simple construction of the dual structure. Each 4-connected vertex of the diamond
structure corresponds to the center of a dual P polyhedra (Fig. 6) and again the polyhedra
are joined by sharing hexagonal faces. Once a model of this sort is constructed the space
corresponding to the X polyhedra is apparent (one of the seventeen tetrahedra vertices is
not on the P polyhedra and will have to be added).6 The polytypes of diamond [3]
correspond to the skeletons of the polytypes of type II. In particular the lonsdaleite net is
the skeleton of MgZn2.
It is difficult to omit the observation [30] that important crystal structures often
result as a combination of the metal structure with anions in some (rather than all) of the
tetrahedral holes. One of the most striking such correspondence is provided by the
structure of spinel, MgAl204 (the prototype of a large class of mainly oxide materials).
The MgAl2 part corresponds to MgCu2 and the 0 atoms fill all the MgAl3 tetrahedra.7
Fig 9. shows the diamond structure and also aspects of the MgCu2 structure-notice how
the primitive unit cell contain just two vertices of the diamond structure, and this unit
contains all the information necessary to construct the apparently complicated type II
clathrate and spinel structures (see e.g. [3] for illustrations of these).

Fig. 9. Left: a primitive cell of the diamond structure. Center: a primitive cell of MgCu 2 (shaded circles are
Mg). Right: a primitive cell of MgAJ 20 4 showing a Mg04 tetrahedron and AI atoms.

The intermetallic compound corresponding to the type ill structure was not known to
Frank and Kasper, but was been reported later for ZqAl3 (so far the only intermetallic
compound with this structure); it is symbolized Z. The major skeleton is shown in Fig.
I 0 and again a model of the clathrate is readily constructed from it. 8
An infinite family of structures can be constructed as intergrowths of the type I and
type III compositions. The simplest is the important alloy structure known as a-phase
and its dual the type IV clathrate. A form of uranium (/3-U) also has this structure, but it

61t is emphasized that such models are very cheaply and simply made and are far superior to drawings if
one wants to appreciate these structures. Plastic tetrahedra (calthrops) are sold cheaply by chemical supply
houses as carbon atoms and these are readily linked by plastic tubing of the appropriate diameter. A good
model ensues if the tubing is cut into approximately equal lengths (readily measured by eye).
?The spinel structure was solved in the very early days of X-ray crystallography by W. H. Bragg (1915).
The structure of MgCu 2 was later determined by J. B. Friauf (1927) who recognized the relationship to spinel.
8The coordinates given by O'Keeffe and Hyde [3, p. 427] for Type III clathrate are not correct. If the z
coordinate for the vertices in I 2 o is changed from 0. I 386 to 0.1836 the four shortest distances (not all exactly
equal) from each vertex will then correspond to the edges of the net.
412
should be emphasized that there are five crystallographically-distinct kind of atoms (three
different coordination polyhedra) in the structure. The major skeleton of all these
compounds consists of two layers (of 2-connected and 3-connected vertices) with rods of
2-connected vertices running normal to the layers (Fig. 10, cf. [15]). One sometimes sees
the statements either that these are the only structures discussed by Frank and Kasper, or
that they did not discuss this particular family at all. Neither is correct.

Fig. 10. Top left: The major skeleton of the type I structure (cf. Fig. 8). Filled circles represent rods of 2-
connected vertices normal to the plane of the figure. Bottom left: the skeleton for type III. Right: the skeleton
for type IV. Notice the two layers of 2- and 3-connected vertices with columns of 2-connected vertices
normal to them ..

The skeleton of the ,u-phase structure is an intergrowth of those of types II and III as
shown in Fig. 11. Again it should be clear that there are many possibilities including the
possibility of polytype formation.

Fig. 11. Left: two layers of the diamond structure (the skeleton of the type II structure). Right: part of the
skeleton of ,u-phase as an intergrowth of the diamond structure and the skeleton of the type III structure.
Open, shaded and filled circles are 4-, 3- and 2-connected vertices respectively.

Brief mention should be made of the famous T phase (which I label T1; also known
as the "Bergman" phase) which has been described in detail in many places, e.g. [3]. The
basic unit of the intermetallic structure is a truncated icosahedron ("soccer ball") composed
of twenty truncated tetrahedra around a central icosahedron. Atoms also center the
icosahedron and each truncated tetrahedron. As shown in Fig. 12 the truncated icosahedra
fit inside truncated octahedra. In the T1 structure, the truncated octahedra fill space in a bee
packing. The building unit of the dual structure consists of a central pentagonal
413
dodecahedron (dual of the X polyhedron) each face of which is shared with another
pentagonal dodecahedron. The opposite face of each of these outer dodecahedra is also
shared with a dodecahedron of a third shell, and twenty 28-vertex polyhedra (duals of the P
FK polyhedron) also fit in this third (outer) shell of polyhedra. The centers of the
28-vertex polyhedra are on the vertices of a large pentagonal dodecahedron which forms a
building unit of the major skeleton of the structure.

Fig. 12. The build-up of the T structure. From the left: An icosahedron sharing a face with a truncated
tetrahedron. Five (slightly irregular) truncated tetrahedra with a common edge. Twenty (again slightly
irregular) truncated tetrahedra enclosing an icosahedron and forming a larger truncated icosahedron. Finally
right: the soccer ball enclosed in a truncated octahedron.

The truncated icosahedron building unit of the T structure has symmetry m3 and in
the bee packing is translated by vectors < 112, 112, 112> to produce a structure with
symmetry 1m3. As pointed out by Sadoc and Mosseri [31] a second structure, often called
"SM" but here labeled Tp, is generated by translation of <112,112,112> followed by a 90•
rotation. This is equivalent ton glide in {110} and produces a structure with symmetry
Pm3n. The two structures are discussed at length by Rivier and Sadoc [32] and I mention
just a few points of special interest here.
The major skeleton of the T1 and Tp structures [32] is built around an 1m3 or Pm3n
packing of pentagonal dodecahedra linked together by further tetrahedral vertices (these
partial structures have been discussed elsewhere [33] in the context of possible carbon
structures). The Pm3n structure is that shown on the far right in Fig. 8; this is of course
the framework of the type I clathrate structure. To complete the major skeleton of Tp all
that is needed is to combine this with a Cr3Si-like pattern of rods of 2-connected vertices
(such as shown on the extreme left in Fig. 8). Restated slightly differently: the major
skeleton of Tp is an intergrowth of type I and the major skeleton of type I.
A second point of interest about the Tp structure is that it consists only of P, Rand
X polyhedra; i.e. it is an intergrowth of types I and II (see Fig. 7). It is the only known
such example, and the interest arises in part from the fact that it was supposed in fact that
such intergrowths were impossible [34].
Interest in the Tp structure also stems from the fact that in experimental work on the
preparation of NaxSi clathrates (see e.g. [35]) the only clathrates found (so far!) are types I
and II. One possible reason for this might have been that for some reason structures with
the Q polyhedron (type III, and intergrowths with type III) are less stable as forms of
silicon. However theoretical calculations [36] indicate just the opposite: to a first
approximation, all clathrates along the type I - type III (including T1) and the type I - type
II joins have the same energy per atom, but the Tp structure is significantly higher in
energy (about twice as high in energy/atom above diamond silicon as the other
structures).
Table I list some of the basic FK phases and their structural formulas. Notice that the
average ring size in the dual structure is related to the average coordination number by
<n> = 6- 12/<Z>.
414
Table I. Some Frank-Kasper and dual clathrate phases. In the structural formula X, P, Q and R
refer to 12-, 14-, 15- and 16-coordinated metal atoms and T refers to tetrahedral holes.

clathrate FKphase structural formula <Z> <n>


I Cr3Si (A15) XR3·T23 13.50 5.111
n MgCu2 (CIS) X2P·T11 13.33 5.100
ill Zr~l3(Z) X3R2Q2·T4o 13.43 5.106
N=2XI+ill CrFe (CT) XsRsQ2·Ts6 13.47 5.109
2xll+ill W~e7 (J.t) X7R2Q2P2·T74 13.39 5.104
20XII+ 3xill T[=Bergman X49R6Q6P2o· T460 13.36 5.102
6xi + 46xll Tp=SM X9sR1sP46"T92o 13.36 5.102

3.3 SOME OTHER INTERMETALLIC STRUCTURES AND THEIR DUALS

There are many tp structures other than the FK structures in which coordination numbers
other than 12, 14, 15 and 16 occur. These have not been systematically enumerated in the
same way as the FK structures have, but it should be rewarding to do so. An example is
provided by the CuAl2 structure [3] which is that of yet another large family-Pearson's
Handbook [26] lists about 100 isostructural phases. The atoms are now 15- and 10-
coordinated with the AI atoms in the 10-coordinated polyhedron Y of Fig. 6. The structure
types named for CuMg2 and NiMg2 are closely related-they have the same coordination
polyhedra. In such structures, as well as the part of the major skeleton joining the vertices
with surface coordination 6 on the coordination polyhedra, and corresponding to lines of
positive disclination, there is a network joining 4-coordinated vertices and corresponding
to lines of negative disclination. In the CuAl2 and related families every atom falls on
this extended skeleton and there appears to be little advantage in considering such
"skeletons" rather than more conventional representations of the structure.
The structures of CuMg2 and Cu2Mg do, however, emphasize the important fact that
these structures (and by extension the rest of the FK phases) are structures of atoms of
different sizes. In this particular pair of Cu, Mg compounds the coordinations are 10 +
2x15 and 2x12 + 16 respectively, in both cases the average is the same: 40/3, but with
Mg in higher coordination than Cu. By most measures of size, Mg is considerably larger
than Cu (see especially the discussion in Pearson's book [18] on this point).9
Another large group of intermetallic compounds has a structure named for CaCus.
This structure may be derived from that of Zr4Al3 (type ill) by replacing two of the Zr
atoms by a single atom. CaCus is better written CaCu(1)2Cu(2)J and the Cu(2) atoms are
in 12-coordination just like the AI in Zr4Al3, but theCa atom is in a 20-coordination and
Cu(1) is in a non-FK 12-coordination polyhedron. Notice that the average coordination
number is again 40/3. If one allows substitution of two Zr by one large atom in different
ratios a variety of related structures are obtained; the most common are named for
ThMn 12, Th2Ni17 and Th2Zn17 (in these structures the average coordination numbers are
13.23, 13.26 and 13.26). The combined membership of this extended family is surpassed
only by that of the Laves phases among binary intermetallic compounds. In the context
of foams, the dual structures might be considered as possibilities for foams consisting of
large and small bubbles.

91t is, I think, a red herring that Mn and some of the actinide metals (notably U, Np and Pu) have
complicated structures with atoms in more than one coordination. The other electropositive (one needs to
exclude elements such as Ga and Sn) elemental metals (including W!) are all bee and/or close packed, and
when they have more than one structure at ambient pressure (e.g. Fe), the molar volumes of the different
polymorphs (e.g. fcc and bee) are very nearly the same.
415

The polyhedron dual to the 20-vertex Ca coordination polyhedron in CaCus has face
symbol [512.68] and the Cu(l) coordination polyhedra have dual [43.56.63]. The dual
structure built of these polyhedra is the clathrate known as dodecasil lH with structure
symbol DOH [24].
The 20-face (36-vertex) polyhedron of the previous garagraph is known as the
hexagonal barrel (see e.g. [3], appendix 4). It has an isomerl known as the "tennis ball"
which is also of interest in clathrasil-like compounds. Thus the clathrasil sigma-2 (SGT)
is constructed of tennis balls and the 9-face Z polyhedron of Fig. 6 as shown in Fig. 13.
The complementary structure is the well-known ThSi2 [3]. Combining the hexagonal
barrel with Z polyhedra in the ratio of I :2 produces the structure dual to yet another very
common structure type: that of AIB2. In these structures the average coordination number
is a little lower: <Z> = 38/3 and the average ring size in the dual structures is <n> =
96/15 = 5.05.

fiJI
0
Fig. 13. Left : the hexagonal barrel (top) and tennis ball icosahedra. Right: the packing of tennis balls and Z
polyhedra (far right) in the clathrasil SGT.

Table II (below) lists the more popular binary crystal structure types according to
Pearson's Handbook [26]. I have included structures (like that of CaCus) for which most
representatives are chemically binary although there are more than two kinds of site
crystallographically. I have also excluded structures listed under a binary headings such as
BaA1 4 and Cu2 Sb as these are definitely ternary structure types (better named for
ThCrzSiz and BaMgSi or PbFCl respectively [3]). It is striking how sphere packing
principles appear to dominate the crystal chemistry of intermetallic compounds. II
This section is closed with a personal favorite which shows the influence of
geometry not only on the structure but on the composition of chemical compounds. In
the NaZn13 structure, Na atoms are surrounded by 24 Zn atoms at the vertices of a snub
cube-a configuration which is well known to be an especially favorable solution to the
Tammes' problem (that of arranging points on a sphere such as to maximize the
minimum distance between them- see e.g. [3]). The snub cubes, which have symmetry
432 (0) share all their square faces to make a large cubic array with stoichiometry NaZn 12
in which left- and right-handed polyhedra alternate. Notice that the centers of the square
faces are centers of octahedral holes so that this is not a tp structure. The packing of snub
cubes also leaves tetrahedral and icosahedral holes as shown in Fig. 14. Filling the
icosahedral holes with Zn leads to the stoichiometry NaZnr3· The average coordination

I 0 As the polyhedra get larger, the number of different possibilities even with the same combination of
faces rapidly gets large. There are for example IS distinct polyhedra with 12 pentagonal and eight hexagonal
faces (such as the hexagonal barrel and the tennis ball) [14].
11 I can't resist the reminder that most of the elements are metals.
416

number is now 12 (notice the lower value for non-tp structures). Members of this family
include KCd13, YBe13, BaCu13 and LaCo13 whose formulas have very little to do with
chemical ideas of valence, but a lot to do with geometry. No doubt this is why they are
rarely mentioned in chemistry text books despite being known for 60 years (Zintl &
Hauke, 1938). That the structure really is a sphere packing is borne out by the fact that it
is also found [37] in opals, which are packings of typically micron-sized silica spheres.

Fig. 14. The NaZn 13 structure. Left: showing four snub cubes of 24 Zn atoms. Right: Zn-centered icosahedra
of Zn atoms around a central snub cube of Zn atoms.

Table II. The most populous binary intermetallic structure types with (very) approximate number of
representatives [26]. The entry "*" under "type" signifies that the type has been discussed in the text.

structures type number


MgCuz, MgZnz C15 =Laves, Type II 1400
CaCu5 , ThMn12,
Th2Zn 17 , Th2 Ni 17 related to type m 1200
Cu3Au, CuAu orderedccp 570
CuZn (CsCI) ordered lx:c 440
AIB2 * 320
Cr3Si A15, Type I 280
CrB 190
CeCu2 (SrAI2) 145
FeB 120
CuAI2 * 110
NaZn13 * 90

4. Geometric problems associated with variations on the sodalite theme


4.1 RING SIZES IN POLYHEDRON PACKINGS

28 of the 78 4-connected zeolite nets listed in the Atlas of Zeolite Structure Types [24]
are derived from packings of finite simple polyhedra and for these the average ring size,
<n>, ranges from 4.97 to 5.14 [38] suggesting that in zeolites at least, there is a fairly
strong constraint on this quantity. This is important as one wants to have at least some
large rings in a zeolite to generate large pores that allow reversible sorption of molecules,
and the constraint on average ring size means that one must "pay" for large ring size by
having a large number of small rings. It turns out, somewhat paradoxically, that because
of the large angle in the T-0-T (T = Si, AI, P etc.) links that correspond to the edges of
the framework, configurations involving large numbers of small rings (especially 3-rings)
417
are unfavorable in aluminosilicates and related materials [3].
The following argument was suggested to me by John Sullivan (see the contribution of J.
M. Sullivan to this volume) and is used with his permission. Consider a structure such as
the sodalite (Kelvin) structure with/0 rings (faces) and e0 edges per repeat unit. As each
edge is common to three rings, the average ring size is 3eoff0 Imagine now the structure
to be inflated by converting some of the rings to double rings (prisms). For each k-ring
per repeat unit converted to a prism, the total number of rings is increased by k 4-rings
and one k-ring, and the number of edges has increased by 2k. The average ring size when s
prisms are formed is now 3(e0 +2sk)l(j0 +sk+s). Ass becomes large (repeated formation of
prisms from k rings) the average ring size, <n> approaches the limit of 6k/(k+1). This
already gives the correct bounds <n> > 9/2 fork= 3 and <n> < 6 fork---? oo [39].
If we start with the sodalite structure as the canonical polyhedron packing, we can
only inflate hexagons (leaving <n> unchanged) and squares (decreasing <n>). Nevertheless
it is instructive to follow the procedure and generate new structures starting from sodalite.
Inflating half the hexagons of sodalite produces the faujasite framework (symbol FAU)
shown in Fig. 15. (It should be clear from the figure that converting all the hexagons to
hexagonal prisms would result in vertices too close together.) The average ring size
(average number of edges of the polygon faces) is 36/7 as in sodalite. This is readily
verified from the vertex symbol 4·4·4·6·6·12 (notice that this tells us that per vertex
there are 3/4 4-ring, 2/6 6-ring and 1112 12-ring).

Fig. 15. A layer of the faujasite framework. The black hexagons are hexagonal prisms seen in projection.

Fig. 16 shows the structure obtained by converting the squares of sodalite to cubes
(see especially the drawing on the left in the figure) . The new structure is the framework
of the zeolite Linde type A (LTA). On the right in the figure the new framework is shown
as an assembly of truncated cuboctahedra (4.6.8)-the structures shown in the two parts
of the figure combine to fill space. The vertex figure is now 4·6·4·6·4·8 and the average
ring size is 144/29 =4.97 (i.e. less than that of sodalite).
Replacing the octagons in Fig. 16 by octagonal prisms will increase ring size again.
The new structure is shown on the left in Fig. 17. It is the framework of the zeolite rho
(RHO). The vertex figure is 4·4-4·6·8·8 and <n> = 36/7 (i.e. now again the same as in
sodalite). The rho structure is interesting as the surface shown in the figure divides space
into identical halves (contrast type A in Fig. 16).
Continuing the process of converting octagons to octagonal prisms on one of the
halves of rho will produce the structure on the right of Fig. 17. Now there are two kinds
of vertex: 4·4·4·6·8·12 and 4·4·4·4·8·12 which occur in the ratio 2:1. The average ring
418

size is now 432/83 =5.20 (i.e. now larger than in sodalite).

Fig. 16. Two views of the type A structure. Left as an assembly of cubes and truncated octahedra. Right: as
an assembly of truncated cuboctahedra.

Fig. 17. Left: the framework of zeolite rho. Right: a framework derived from rho by replacing octagons by
octagonal prisms.

4.2 PACKINGS WITH TETRAHEDRA AND RAREST(?) SPHERE PACKING

Clearly the process of the previous section could be continued indefinitely, but is clear
that the average ring size changes only slowly from that in sodalite. If 3-rings (triangles)
are excluded from the structures, the average ring size is 2! 24/5 =4.8.
Another way of generating new structures is to replace vertices with tetrahedra of
vertices-a process called decorating. Fig. 18 shows the decorated sodalite net. This
structure is of special interest [40] as it very likely corresponds to the least dense stable
sphere packing. If the vertices are replaced by spheres in contact, the structure is stable
(not all contacts with contiguous spheres are on the same hemisphere) and all spheres are
related by symmetry. The density (fraction of space filled by spheres) is p =81li(2 + 3...J2)3
=0.102 ...---c.f. p =nt3...J2 =0.740... for closest packing. I do not know if there is a less
dense stable packing in which the spheres are not all related by symmetry. It is
sometimes stated that the decorated diamond net (p =0.123 ... ) is the least dense stable
sphere packing, e.g. by Hilbert [41], showing that even the very greatest can nod.
The structure has vertex symbol 3·8·3·12·3·12 and average ring size 144/31 =4.65.
Clearly one reduces ring size quickly by decoration. Repeated decorations leads to average
419
ring size of:

for the mth decoration. It may be confirmed that this goes to <n> =9/2 as m ~ oo. For
repeated decorations of the diamond net see [42].

Fig. 18. Left: the decorated sodalite net. The rarest (?) stable packing of equivalent spheres.

4.3 N-DIMENSIONAL SODALITES

One often gains insight into the geometry of our 3-dimensional world by generalizing to
other dimensions. The sodalite structure is rather nicely generalized [43] toN dimensions
as the structures dual (in the sense used in this paper!) to simplicially-packed (cf. tp)
lattice sphere packings. The sphere packings [44] correspond to the lattices known as
AN*· The primitive cell of the lattice is rhombotopal (all edges equal to a and all angles
equal to a) with a= cos- 1(-liN). The centers of the simplicial holes fall on a hyperplane
and have coordinates which are all permutations of 1/(N + 1), 2/(N +I), ... NI(N + 1). The
structure obtained by placing vertices in each of these holes has N + 1 nearest neighbors
at a distance a...J[2/N(N + 1)].

Fig. 19. The "2-dimensional sodalite" structure showing a unit cell

This structure is a space filling by polytopes ("permutohedra" [44]) with (N + 1)!


vertices and square and hexagonal faces. At each vertex N + 1 hexagons and (N + 1)(N-
2)/2 squares meet. For N = 2 the structure is just the familiar honeycomb 63 net and the
repeat unit (unit cell contents) is a line (Fig. 19). The three dimensional structure (Fig.
20) is likewise built up of the two-dimensional units ("bubbles") and each higher
dimensional structure has a primitive unit cell containing a (N- I)-dimensional bubble.
The average ring size is <n> = 12N/(3N- 2). This is very slightly larger than the
420
value expected [45] for a statistical froth, <n> = 27r/cos- 1(-l/N) (except for N = 2 and oo,
when they are equal). For N = 2 and oo, <n> = 6 and 4 respectively. For some other
properties of these structures see [43].

Fig. 20. The "3-dimensional sodalite" structure showing a primitive unit cell (left) and the 3-dimensional
bubble (right).

5. Concluding remarks and acknowledgments

In this paper I have indicated some commonality of interest between the study of crystal
structures and cellular structures such as foams. Much of it well documented, but I hope
to have imparted a personal slant and brought up some lesser-known results of crystal
chemistry. My belief is that crystal structures provide a rich source of unsolved problems
in geometry. Some of them have been alluded to here; others, such as those concerned
with the topology of nets, have been described elsewhere [3, especially App. 3]; This
paper will have served a useful purpose if it can stimulate the more mathematically
inclined to examine these problems.
I am greatly indebted to Jean-Fran~ois Sadoc and Nicolas Rivier for the opportunity
to participate in the school that served as the basis for this volume. Stephen Hyde and
John Sullivan have attempted to instruct me on some mathematical points relevant to
this paper. The errors that persist are the result of my invincible ignorance in such
matters. My own work described herein is supported by the U.S. National Science
Foundation (DMR 94 24445) and the A.S.U. Materials Research Science and Engineering
Center, also supported by NSF (DMR 96 32635).

6. References

[1] D. C. Wright & N. D. Mermin, Rev. Mod. Phys. 61, 385 (1989).
[2] Lord Kelvin, On the Molecular Tactics of a Crystal, Robert Boyle Lecture 1893,
reprinted in Baltimore Lectures, Appendix J. C. J. Clay and Sons, London, 1904.
[3] M. O'Keeffe & B. G. Hyde, Crystal Structures /: Patterns and Symmetry,
Mineralogical Society of America, Washington, D.C. 1996.
[4] D. Johrent & A. Mewis, J. Alloys Compounds, 205, 183-189 (1994).
[5] J. Diinner & A. Mewis, J. Alloys Compounds, 167, 127-134 (1990).
[6] D. Weaire & R. Phelan, Phil. Mag. Lett., 70, 345 (1994).
[7] T. Aste, D. Boose & N. Rivier, Phys. Rev. E53, 6181-6191 (1996).
[8] M. O'Keeffe, Acta Crystallogr. A, in press [1997).
[9] R. E. Williams, Science 161, 276-277 (1968).
[10] R. Kusner, Proc Roy. Soc. (London) A439, 683-686 (1992).
[11] M. O'Keeffe, Phil. Mag. Letts. in press (1997).
[12] F. C. Frank & J. S. Kasper, Acta Crystallogr. 11, 184-190 (1958) and 12, 483-
499 (1959).
421

[13] See the discussion by F. C. Frank and F. Laves on pp 521-2 of Phase Stability in
Metals and Alloys (P. S. Rudman, J. Stringer and R. I. Jaffee, eds.) McGraw-Hill,
New York (1967).
[14] P. W. Fowler & D. E. Manopolous, An Atlas of Fullerenes, Oxford (1995).
[15] J. F. Sadoc, J. de Phys. Lett. 44, L707-715 (1983).
[16] J. F. Sadoc & R. Mosseri, Frustration Geometrique, Editions Eyrolles, Paris
(1997).
[17] N. Rivier & T. Aste, Phil. Trans. Roy. Soc. Lond A 354, 2055-2069 (1996).
[18] W. B. Pearson, The Crystal Chemistry and Physics of Metals and Alloys, Wiley,
New York (1972).
[19] D.P. Shoemaker & C. B. Shoemaker, Acta Crystallogr. B42, 3-11 (1986).
[20] Ya. P. Yarmolyuk & P. I. Kripyakevich, Sov. Phys. Crystallogr.19, 334-337
(1974). [Translation of Kristallografiya 19, 539-545 (1974)]
[21) A. F. Wells, Structural Inorganic Chemistry, 5th Edition, pp 659-666, Oxford
(1984).
[22] G. A. Jeffrey, in Inclusion Compounds I (J. L. Atwood, ed) Ch 5. Academic Press,
New York (1984).
[23] H. M. Powell, J. Chem. Soc. 1948, 61-73.
[24] W. M. Meier & D. H. Olson, Atlas of Zeolite Structure Types, 3rd ed.
Butterworth-Heinemann, London (1992). Issue 5 of Zeolites 12 (1992).
[25] J. S. Kasper, P. Hagenmuller, M. Pouchard & C. Cros, Science, 150, 1713-
(1965). C. Cros et al. Bull. Soc. Chim. France 2, 379-386 (1971) give other
references.
[26] P. Villars & L. D. Calvert, Pearson's Handbook of Crystallographic Data for
Intermetallic Phases, 2nd Ed. ASM International, Materials Park, Ohio (1991).
[27] G. B. Adams, M. O'Keeffe, A. Demkov, 0. F. Sankey and Y.-M. Huang, Phys.
Rev. B49, 8084-8053 (1994).
[28] D. Weaire & R. Phelan, Phil. Mag. Lett. 69, 107 (1994).
[29] G. Hiigg & N. Schonberg, Acta Crystallogr. 7, 351-352 (1954).
[30] M. O'Keeffe & B. G. Hyde, Structure and Bonding, 61, 77-144 (1985).
[31] J. F. Sadoc & R. Mosseri, J. de Phys. 46, 1809-1826 (1985).
[32] N. Rivier & J. F. Sadoc, Europhys. Lett. 7, 523-528 (1988).
[33] G. B. Adams, 0. F. Sankey, J. B. Page & M. O'Keeffe, Chem. Phys. 176, 1792-
1795 (1992).
[34] E. Hellner & W. B. Pearson, J. Solid State Chem. 70, 241-248 (1987).
[35] S. B. Roy, K. E. Sim & A. D. Caplin, Phil. Mag. B65, 1445 (1992); J. Gryko, P.
F. McMillan & 0. F. Sankey, Phys. Rev. B 54, 3037-3039 (1996).
[36] G. B. Adams, M. O'Keeffe & 0. F. Sankey, to be published.
[37] J. V. Sanders & M. J. Murray, Nature 275, 201 (1975); Phil. Mag. 42, 721
(1980).
[38] M. O'Keeffe & S. T. Hyde, Zeolites in press (1997).
[39] F. Leo & R. Stong, Trans. Amer. Math Soc. 337, 891-906 (1993).
[40] M. O'Keeffe, Zeits. Kristallogr. 196, 21-37 (1991).
[41] D. Hilbert & S. Cohn-Vossen, Geometry and the Imagination, Chelsea, New York
(1952).
[42] M. O'Keeffe & S. T. Hyde, Zeits. Kristallogr. 211, 73-78 (1996).
[43] M. O'Keeffe, Acta Crystallogr. A47, 748-753 (1991).
[44] J. H. Conway & N. J. A. Sloane, Sphere Packings, Lattices and Groups,
Springer-Verlag, New York (1988).
[45] T. Aste & N. Rivier, J. Phys. A Math. Gen. 28, 1381-1398 (1995).
422
INVERSE MICELLAR LYOTROPIC CUBIC PHASES

JOHNSEDDONANDJOHNROBINS
Department of Chemistry
Imperial College
LONDON SW7 2AY, U.K.

1 . Introduction

Many amphiphilic molecules (surfactants, biological lipids, etc) can form


three-dimensionally ordered liquid-crystalline structures in the presence of
water [1, 2]. The most common of such complex lyotropic mesophases are
the bicontinuous cubic phases [3], which appear to be based upon
underlying periodic minimal surfaces [4]. For the inverse versions of these
phases, the structures consist of two interwoven networks of water
channels, separated by a single, continuous fluid lipid bilayer, typically 40
A thick. Such cubic phases typically have lattice parameters in the range 80
- 200 A, although in certain circumstances they can swell to much larger
dimensions [5, 6].
However, in this article we wish to focus attention on another family of
lyotropic cubic phases, which bear striking structural analogies with certain
TCP (tetrahedrally-close-packed) packings such as the Laves structure
Cl5, which is relevant to ordered foam, metal alloy, clathrate hydrate and
natural opal packings [7, 8]. This analogy with liquid crystals was first
explored by Charvolin and Sadoc in a theoretical study of the possible
ordered, space-filling configurations of frustrated fluid films [9]. Imposing
the constraint that the dihedral and edge angles should lie close to 120° and
109°28', so that the tensions of the fluid films are balanced, led to two
cubic structures. One type, closely analogous to the so-called 12 A cubic
clathrate hydrate structure, and the Weaire-Phelan (A15) structure of foams
[10], consisted of 2 dodecahedral and 6 tetrakaidecahedral cells, both
slightly distorted, packed in the unit cell with spacegroup Pm3n (No. 223)
(note that the newer editions of the International Tables for Crystallography
[11] replace the symmetry label3 by the more formally correct 3). They
suggested that this provided a description of the structure of the well-
known lyotropic liquid-crystalline phase of the same spacegroup, with the
polyhedral "film" located at the middle of the aqueous region, separating
423
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 423--436.
© 1999 Kluwer Academic Publishers.
424

two types of micelle. The second structure they found was closely
analogous to the so-called 17 A cubic clathrate hydrate, consisting of a C15
packing of 8 hexakaidecahedra and 16 dodecahedra, both slightly
distorted. They predicted that a micellar cubic phase having the C 15
structure should occur - a prediction which turns out to be fully vindicated,
experimentally.

2. Lyotropic Phase Diagrams

Lyotropic liquid crystalline phases may be categorized [2, 3] according to


their interfacial mean and Gaussian curvatures

(1)

where q and c2 are the principal curvatures at each point of the interface.
Depending on the mesophase structure, Hand K may either be uniform, or
non-uniform, along the interface. The sign of H is arbitrary: we take H
negative to correspond to curvature of the interface away from the
hydrocarbon chain region, and towards the water region. The sign of K
determines the form of the interface, and is directly related to the topology
of the phase. Positive K corresponds to elliptic surfaces, and hence phases
consisting of packings of discontinuous molecular aggregegates (micelles
or inverse micelles); zero K corresponds to parabolic surfaces, and hence
phases with flat or cylindrical interfaces; negative K corresponds to
hyperbolic surfaces, and hence to phases based upon saddle surfaces,
which are bicontinuous in either two- or three-dimensions. Note that for
simplicity we are ignoring any effects such as thermally-induced
deformations of the layers, the Landau-Peierls instability, which may
destroy true long range order, and defects (eg. pores or channels). For
many surfactant systems, where the layer bending modulus is of the order
of kT, such effects become very important, whereas for many biological
lipids the bending modulus is an order of magnitude larger and the effects
become small or even negligible.
It is useful to consider the natural sequence of phases with varying
average interfacial mean curvature <.H> (Fig. 1). Experimentally, it is
usually found that there is a systematic increase of the preferred
(spontaneous) interfacial mean curvature <.H0 > with increasing water
content. Similarly, decreasing temperature, or increasing hydrostatic
pressure, tends to increase <.H0 >.
425

0 •
Inverse
Micellar
Micellar a b c d
Solution
Solution

Mean interlacial curvature


---·
Figure 1. The sequence of lyotropic liquid-crystalline phases versus the
average interfacial mean curvature <H>. The experimental parameter
controlling the preferred value <.H0 > can be any thermodynamic variable
such as hydration, temperature, pressure, etc. From [3].

The fluid lamellar La. phase, consisting of a one-dimensional periodic


stacking of flat lipid bilayers, occupies a central location in this diagram,
with <H>=O (and <K>=O). On moving left in the diagram to the inverse or
type II phases, <H> becomes increasingly negative, corresponding to
increasingly strong mean curvature towards the water region. Conversely,
moving right towards the normal or type I phases corresponds to
increasingly positive <H>. Although there is an apparent mirror symmetry
in this diagram, in reality the molecular packing constraints and the
frustration within "mirror-image" pairs of phases are not equivalent.
Indeed, some mesophases, such as the hexagonal H1 and Hn phases, are
readily found in both normal and inverse versions, whereas some phases
have so far only been observed in one type or the other.
426
The intermediate regions labelled a, b, c and din Fig. 1 contain more
complex, intermediate phases, many with three-dimensional periodicities.
Usually these phases have cubic symmetry, and the most commonly
observed spacegroups, along with their allowed Bragg reflections [11] are
listed in Table 1.

Table 1. Lyotropic cubic phases and their characteristic spacings ratios.

Spacegroup Spacings ratios of allowed Bragg reflections


name (No.)
Pm3n (223) ~2. ~4. ~5. ~6. ~8. ~10. ~12, ~13, ~14, ~16 ......
Pn3m (224) ~2.~3.~4.~6.~8.~9.~10,~11.~12.~14,~16.
Fm3m (225) ~3. ~4. ~8. ~11, ~12, ~16, ~19, ~20, ~24, ~27 ...
Fd3m (227) ~3. -../8, ~11. ~12 7 -../16 7 ~19, ~24, -../21, ~32, ~43 ..
Im3m (229) ~2. ~4. ~6. ~8. 'VlO, '/12, ~14, ~16, ~18, ~20 .....
/a3d (230) ~6.~8.~14.~20,~22.~24.~26,~30,~32,~38 ..

The cubic phases found in regions b and c, lying between the lamellar and
hexagonal phases, have bicontinuous structures consisting of interwoven
networks of water channels, or sufactant channels, respectively. The
average interfacial Gaussian curvature <K> is thus negative for these
phases. The observed spacegroups are Pn3m (No. 224), Im3m (No. 229)
and Ia3d (No. 230), and these structures are based on underlying D-, P-
and G- minimal surfaces, respectively. So far Pn3m and Im3m have only
been reliably observed in inverse (type II) versions, whereas Ia3d is
commonly found in both normal and inverse versions.

3 . Micellar and Inverse Micellar Mesophases

The phases found in regions d and a of Fig. 1, adjacent to the micellar and
inverse micellar solutions, have structures based upon discontinuous
packings of micelles or inverse micelles, and are hence characterised by
positive interfacial <K>. Three different micellar cubic phases, of space
groups Pm3n (No. 223), Fm3m (No. 225) and Im3m (No. 229) have so
far been found in binary systems in region d [12, 13]. Note that there are
thus two common cubic phases having the same spacegroup of Im3m, but
with entirely different structures, one bicontinuous and the other
discontinuous. In addition, a micellar phase of 3-D hexagonal symmetry
(spacegroup P63/mmc) has recently been reported in a nonionic surfactant I
427
water system [14]. The Pm3n cubic phase was discovered many years ago
[15], and a structure based upon a packing of discrete micellar aggregates
was proposed a decade ago [16]. Subsequently, an alternative model
inspired by the A 15 space-filling packing of polyhedra was invoked [9]. In
this model (Fig. 2) there are spherical micelles at the centres of the two
dodecahedra, and disk-like micelles at the centres of the six
tetrakaidecahedra. The polyhedral faces thus define the middle of the
aqueous regions of the phase. The relationship between micellar cubic
phases and foam packings has recently been further extended to cover the
Fm3m and Im3m cubic phases [17].

Pm3n Fm3m 1m 3m

Figure 2. The type I micellar cubic phases, showing both the positions of
the micelles in the unit cells, and the polyhedral representations of the
structures. Each polyhedron represents a micelle with all of its associated
water. Adapted from [13, 17].

Amphiphilic molecules with relatively small, weakly hydrated headgroups


and/or bulky hydrocarbon chains might be expected to form inverse
micellar cubic phases, at least at low hydrations and/or high temperatures.
Surprisingly, it has only fairly recently been established that this is indeed
the case [ 18, 19], and the packing motif of the only such structure so far
characterised, of spacegroup Fd3m (No. 227) turns out to be quite
428

complicated [20]. Indeed, so far there is no evidence that simple (sc, bee or
fcc) cubic packings of inverse micelles of lipids can form.
The cubic phase of spacegroup Fd3m was first discovered in a lipid
extract from Pseudomonas fluorescens [21], but only recently has the
structure been solved by low-resolution crystallography [20], and
confirmed by freeze-fracture electron microscopy [22]. The Fd3m cubic
phase has now been observed in a diverse number of lyotropic systems,
ranging from mixtures of monoolein with oleic acid [23, 24], oleic acid
with sodium oleate [19], phosphatidylcholines with diacylglycerols [18,
25], fatty acids [20, 26], and fatty alcohols [27].
The low-angle X-ray powder pattern of the Fd3m cubic phase (Fig. 3)
shows that all of the allowed Bragg reflections of cubic aspect 15 (111,
220, 311, 222, 400, 331 , 422, 333/511, 440 and 533), up to the --./43
reflection are observed (the --./11 and--./ 12 peaks are very close together, and
are not clearly resolved in this image).

,::
~

Radiel ' " " ' "

Figure 3. The X-ray diffraction pattern obtained from an unaligned, fully


hydrated phosphatidylcholine I diacylglycerol sample using film detection,
along with a typical radial densitometer scan through an image obtained
from a similar sample, using an optoelectronic detector. The observed
peaks index as spacegroup Fd3m, with a lattice parameter of 153 A.

It should be noted that the intensities of the X-ray Bragg peaks from liquid-
crystalline phases invariably fall steeply with increasing hkl, due to the
short range disorder inherent in fluid phases.
429
4 . Structure of the Fd3m Cubic Phase

The structure of the Fd3m cubic phase, as deduced from X-ray diffraction
[20] and freeze-frac~re electron microscopy [22], is shown in Fig. 4, with
the origin chosen at 43m (origin choice 1).

Fd3m
r-\
.,..------r--~---:1 /

1=-------l~--fr:::'~----1( //-iE?
\
\
--®

b)

Figure 4. (a) The packing of the two types of inverse micelle in the Fd3m
cubic phase, with their polyhedral shapes indicated for each site (for clarity
the polyhedra are shown reduced in size, and are hence non-touching); (b)
the connection of a hexakaidecahedron with a neighbouring dodecahedron
via pentagonal faces, along with the location of the lipid molecules and the
water cores within the two types of polyhedra.
430

The freeze-fracture electron microscopy study [22] confirmed that the


spacegroup is indeed Fd3m rather than the less symmetrical Fd3, which
gives the same powder diffraction pattern (same cubic aspect) as Fd3m.
The unit cell contains two types of quasi-spherical inverse micelles, of
different size. The 8 larger ones sit inside hexak:aidecahedra at special
positions (a) within the unit cell ((0,0,0), (3/4,114,3/4), etc), of site
symmetry 43m. The 16 smaller ones occupy dodecahedra at special
positions (d) in the unit cell ((5/8,5/8,5/8), (3/8,7/8,1/8), (7/8,118,3/8),
(1/8,3/8,7/8), etc), of site symmetry 3m. The hexak:aidecahedra are
arranged on a diamond lattice, and meet tetrahedrally via their four
hexagonal faces. The smaller dodecahedra are grouped in tetrahedral
clusters, connected to neighbouring hexak:aidecahedra via three adjacent
pentagonal faces of the latter. The faces of the polyhedra define the average
loci of the methyl endgroups of the hydrocarbon chains of the amphiphilic
molecules. The volume of the hexak:aidecahedra and the dodecahedra are
0.0579 Vu and 0.03355 Vu. respectively, where Vu is the unit cell volume.
The total volume fractions occupied by the 8 larger and 16 smaller inverse
micelles are thus 0.4632 and 0.5368, respectively.
In accordance with the prediction of Charvolin and Sadoc [9], the
micellar packing in the Fd3m inverse micellar cubic phase is essentially the
same as the cubic Laves phase C15 found in binary metal alloys of
composition AB2, such as MgCu2, and has close analogy with the 17 A
cubic clathrate hydrates (e.g., CH2/H 2S/H20). In such crystals the water
molecules form space filling assemblies of hexakaidecahedral and
dodecahedral cages around suitable solute molecules such as CH2 and H2S
[28].

5 . Formation and Stability of Inverse Micellar Cubic Phases

Usually a minimum of two amphiphilic lipid components (one being more


strongly polar) are required for the formation of the Fd3m cubic phase.
This is easy to rationalize in terms of the structure of the phase, which
involves two different sizes of inverse micelle: the more polar lipid can
partially preferentially partition into the larger, less curved inverse micelles,
and vice versa for the more weakly polar amphiphile. However, this is not
an absolute prerequisite: we have recently reported that the Fd3m cubic
phase can be formed by a purely binary glycolipid I water system, where
of course the lipid composition has to be identical within each inverse
micelle [29].
As expected, self-diffusion measurements by pulsed field gradient NMR
give clear evidence of hindered long-range diffusion within the Fd3m
431

phase, and show that the less polar amphiphile is able to diffuse more
freely than the more polar one [30, 31].
To date, there has only been one report of a type I Fd3m cubic phase
(consisting of normal, rather than inverse micelles), in a pseudo four-
component surfactant I alcohol I oil I brine system which also exhibits a
Pm3n micellar cubic phase [32]. An Fd3m cubic phase has also recenly
been found in an amphiphilic diblock copolymer I water I p-xylene system
[33].
It is clear that the chain packing frustration in structures based upon
close packings of quasi-spherical inverse micelles will be quite large. The
reason for this is that spheres do not fill space very efficiently, but the fluid
hydrocarbon chains of the lipid molecules must fill all of the non-polar
volume of the phase. Some of the chains must therefore stretch away from
their preferred conformational state to achieve this, costing elastic energy
(we assume for simplicity that the polar I nonpolar interface between the
headgroups and the water remains spherical, so that all the molecules
within one micelle have the same curvature elastic energy). A simple
measure of the packing frustration is then given by the packing fraction of
the phase (Table 2). Such considerations lead to the conclusion that
increasing the chainlength of the amphiphilic molecules should facilitate
formation of inverse micellar cubic phases, and this is experimentally
found to be the case [29, 27].

Table 2. Packing fractions of inverse cylindrical and spherical phases.

Micellar Phase Packing fraction


p6m (2-D hexagonal) 0.907
Pm3m (sc) 0.524
Im3m (bee) 0.680
Fm3m (fcc) 0.740
Fd3m cubic 0.710

For a hexagonal Hn phase of inverse cylinders, the excess hydrophobic


volume to be filled is only 9.3 % of the total volume of the unit cell,
whereas for a sc packing of inverse spherical micelles it has a value as high
as 47.6 %. It is thus not surprising that such a Pm3m inverse cubic phase
appears not to form, at least in the absence of any added hydrocarbon
solvent which could partition into the void regions and hence relieve the
packing frustration. Even the most efficient fcc packing still leaves 26% of
potentially void hydrophobic volume to be filled. In fact, for Fd3m, which
432

has a packing fraction of 0.71, the situation appears worse, and so at first
sight it seems surprising that this should be the strongly preferred structure
for packings of inverse micelles, particularly when it requires the formation
of two types of inverse micelle with different sizes, and different interfacial
mean curvatures (note that the diameters of the two water cores are difficult
to measure accurately, but are significatly smaller at the dodecahedral
sites). However, calculations of the packing frustration energy in various
inverse lyotropic mesophases confirm that an Fd3m packing of two
different sized inverse micelles can indeed have a lower energy than sc,fcc
or bee packings of uniform inverse micelles [34].
Studying the effects of hydrostatic pressure should allow the molecular
factors controlling mesophase structure and stability to be disentangled.
Our measurements so far on a glycolipid system show that pressure
increases the Hrr-Fd3m transition temperature by 25 °C I kbar, a value that
is similar to the corresponding shifts in the gel-fluid (chain-melting)
transition [35].
Although so far all observed examples of inverse micellar cubic phases
tum out to have spacegroup Fd3m, in our laboratory we are studying what
will probably tum out to be a second example of an inverse micellar cubic
phase.

Acknowledgements. This work is supported by EPSRC grant


GR/K20309. JMS would like to dedicate this article to Aquarius, the spirit
ofCargese!

6 • References

[1] Lindblom, G. and Rilfors, L. (1989) Cubic phases and isotropic


structures formed by membrane lipids - possible biological relevance.
Biochim. Biophys. Acta 988, 221-256.
[2] Seddon, J. M. (1990) Structure of the inverted hexagonal (Hn)
phase, and non-lamellar phase transitions of lipids. Biochim. Biophys.
Acta 1031, 1-69.
[3] Seddon, J. M. and Templer, R. H. (1993) Cubic phases of self-
assembled amphiphilic aggregates. Phil. Trans. R. Soc. Lond. A 344,
377-401.
[4] Andersson, S., Hyde, S. T., Larsson, K. and Lidin, S. (1988)
Minimal-surfaces and structures - from inorganic and metal crystals to cell
membranes and biopolymers. Chem. Rev. 88, 221-242.
433
[5] Templer, R. H., Madan, K. H., Warrender, N. A. and Seddon, J.
M. (1992) Swollen lyotropic cubic phases in fully hydrated mixtures of
monoolein, dioleoylphosphatidylcholine and dioleoylphosphatidylethanol-
amine. In The Structure and Conformation of Amphiphilic Membranes,
(eds. R. Lipowsky, D. Richter and K. Kremer), pp. 262-265, Springer
Verlag, Berlin.
[6] Peter, U., Konig, S., Roux, D. and Bellocq, A. M. (1996)
Extremely dilute lyotropic 3D crystalline phase in a water I oil I surfactant I
cosurfactant mixture. Phys. Rev. Lett. 76, 3866-3869.
[7] Rivier, N. and Aste, T. (1996) Curvature and frustration in
cellular-systems. Phil. Trans. R. Soc. Lond. A 354, 2055-2069.
[8] Weaire, D. and Phelan, R. (1996) Cellular structures in 3
dimensions. Phil. Trans. R. Soc. Lond. A 354, 1989-1997.
[9] Charvolin, J. and Sadoc, J. F. (1988) Periodic-systems of
frustrated fluid films and micellar cubic structures in liquid-crystals. J.
Phys. France 49, 521-526.
[10] Weaire, D. and Phelan, R. (1994) A counterexample to Kelvin's
conjecture on minimal-surfaces. Phil. Mag. Lett. 69, 107-110.
[11] Hahn, T. (1983) International Tables for Crystallography, Vol. A,
D. Reidel, Dordrecht.
[12] Gulik, A., Delacroix, H., Kirschner, G. and Luzzati, V. (1995)
Polymorphism of ganglioside-water systems - a new class of micellar
cubic phases -freeze-fracture electron-microscopy and X-ray- scattering
studies. J. Phys. II France 5, 445-464.
[13] Sakya, P., Seddon, J. M., Templer, R. H., Mirkin, R. J. and
Tiddy, G. J. T. ( 1997) Micellar cubic phases and their structural
relationships: The nonionic surfactant system C12E012Iwater. Langmuir
13, 3706-3714.
[14] Clerc, M. (1996) A new symmetry for the packing of amphiphilic
direct micelles. J. Phys. II France 6, 961-968.
[15] Tardieu, A. and Luzzati, V. (1970) Polymorphism of lipids. A
novel cubic phase: a cage-like network of rods with enclosed spherical
micelles. Biochim. Biophys. Acta 219, 11-17.
[16] Fontell, K., Fox, K. K. and Hansson, E. (1985) On the structure
of the cubic phase It in some lipid-water systems. Mol. Cryst. Liq. Cryst.
Lett. 1, 9- 17.
[ 17] Luzzati, V., Delacroix, H. and Gulik, A. ( 1996) The micellar cubic
phases of lipid-containing systems - analogies with foams, relations with
the infinite periodic minimal-surfaces, sharpness of the polar apolar
partition. J. Phys. II France 6, 405-418.
[18] Seddon, J. M. (1990) An inverse face-centered cubic phase formed
by diacylglycerol-phosphatidylcholine mixtures. Biochemistry 29, 7997 -
8002.
434
[19] Seddon, J. M., Bartle, E. A. and Mingins, J. (1990) Inverse cubic
liquid-crystalline phases of phospholipids and related lyotropic systems.
J. Phys: Condens. Matter 2, SA285 - SA290.
[20] Luzzati, V., Vargas, R., Gulik, A., Mariani, P., Seddon, J. M.
and Rivas, E. ( 1992) Lipid polymorphism - a correction - the structure of
the cubic phase of extinction symbol Fd-- consists of 2 types of disjointed
reverse micelles embedded in a 3-dimensional hydrocarbon matrix.
Biochemistry 31, 279-285.
[21] Tardieu, A. (1972) Etudes Cristallographique de Systemes Lipides-
Eau, Universite de Paris-Sud.
[22] Delacroix, H., Gulik-Krzywicki, T. and Seddon, J. M. (1996)
Freeze-fracture electron-microscopy of lyotropic lipid systems -
quantitative-analysis of the inverse micellar cubic phase of space group
Fd3m (Q(227)). J. Mol. Bioi. 258, 88-103.
[23] Mariani, P., Luzzati, V. and Delacroix, H. (1988) Cubic phases of
lipid-containing systems - structure-analysis and biological implications. J.
Mol. Bioi. 204, 165-188.
[24] Mariani, P., Rivas, E., Luzzati, V. and Delacroix, H. (1990)
Polymorphism of a lipid extract from Pseudomonas fluorescens: structure
analysis of a hexagonal phase and of a novel cubic phase of extinction
symbol Fd--. Biochemistry 29, 6799 - 6810.
[25] Takahashi, H., Hatta, I. and Quinn, P. J. (1996) Cubic phases in
hydrated 1/1 and 1/2 dipalmitoylphosphatidylcholine-dipalmitoylglycerol
mixtures. Biophys. J. 70, 1407-1411.
[26] Koynova, R., Tenchov, B. and Rapp, G. (1997) Mixing behaviour
of saturated short-chain phosphatidylcholines and fatty acids. Eutectic
points, liquid and solid phase immiscibility, non-lamellar phases. Chem.
Phys. Lipids, in press.
[27] Huang, Z., Seddon, J. M. and Templer, R. H. (1996) An inverse
micellar Fd3m cubic phase formed by hydrated phosphatidylcholine/fatty
alcohol mixtures. Chem. Phys. Lipids 82, 53-61.
[28] Jeffrey, G. A. and McMullen, R. K. (1967) The clathrate hydrates.
Progr. Inorg. Chem. 8, 43-108.
[29] Seddon, J. M., Zeb, N., Templer, R. H., McElhaney, R. N. and
Mannock, D. A. ( 1996) An Fd3m lyotropic cubic phase in a binary
glycolipid/water system. Langmuir 12, 5250-5253.
[30] Hendrikx, Y., Sotta, P., Seddon, J. M., Dutheillet, Y. and Bartle,
E. A. ( 1994) NMR self-diffusion measurements in inverse micellar cubic
phases. Liq. Crystals 16, 893-903.
435

[31] Oradd, G. , Lindblom, G. , Fontell, K. and Ljusberg-Wahren, H.


(1995) Phase-diagram of soybean phosphatidylcholine-diacylglycerol-
water studied by X-ray diffraction and 3lp NMR and pulsed-field gradient
I H NMR: evidence for reversed micelles in the cubic phase. Biophys. J.
68, 1856-1863.
[32] de Geyer, A. (1993) Phase behaviour of surfactant-alcohol-oil-
water cubic liquid crystals. Progr. Colloid Polym. Sci. 93, 76-80.
[33] Alexandridis, P., Olsson, U. and Lindman, B. (1996) A reverse
micellar cubic phase. Langmuir 12, 1419-1422.
[34] Duesing, P. M., Templer, R. H. and Seddon, J. M. (1997)
Quantifying packing frustration energy in inverse lyotropic mesophases.
Langmuir 13, 351-359.
[35] Duesing, P. M., Seddon, J. M., Templer, R. H. and Mannock, D. A.
( 1997) Pressure effects on lamellar and inverse curved phases of fully
hydrated dialkyl phosphatidylethanolamines and ~ - D-xylopyranosyl-sn­
glycerols. Langmuir 13, 2655-2664.
11~11. iii r.J,,.Io~s.

,
It hos ~. 5U~)ft•<• . li.. ...~.J .~
lr'• all •~•"'· We ~eve t• •lf•r
It'• molt~• if ro b~ <•lllptlitlve
(•)f•••rs upt<l ·,t.
'"4alff"· 7~4,., . ~ ...
I Joh/ .,..,, ,1
)t.
.
~. tf..tl ..,..,., oYtr
!~ON; S~t's Q fr~t~o(.,ill
4"J s$.~'s ..r;tr;,3 4
tJ.ei~ 4~0Wf
ft.t~. ~hJ .~.t
: i,ah .fttrt 14 h<r,·-·

I~
Sponges
S.T. HYDE,
Applied Mathematics Dept.,
Research School of Physical Sciences,
Australian National University,
Canberra, 0200, AUSTRAliA.

Abstract

The geometric features of sponges, which describe polycontinuous


morphologies, are discussed. The focus is on 3-periodic surfaces, which have an
underlying 3-d lattice. Definitions and explanations of a range of concepts from
non-euclidean geometry, topology and differential geometry are used to derive
local and global features of smoothly curved sponges, facetted polyhedral
sponges and reticulating networks of edges on these forms.

Introduction

The structure of foams in 3-d euclidean space is dependent on the geometry of


the dividing surfaces that partition space into closed cells. Foams offer one
possible morphology for the dissection of space by surfaces. In addition, space
can be partitioned into open honeycombs (infinite prisms), layers (sheets), and
more convoluted interwoven domains.
The simplest examples of the last class are formed by an infinite, single-
sheeted surface, which winds throughout space - without ever touching itself -
cleaving space into two sub-volumes, each continuously connected. The
resulting saddle-shaped surface is a sponge". This surface may be
translationally ordered, as in the 3-periodic minimal surfaces, or their plane-
faced analogues, such as regular sponges (defined later). Alternatively, sponges
may be random and disordered. Such simple sponges (ordered or disordered)
are said to be embedded- free of self-intersections- and carve space into two
intertwined infinite sub-spaces (fig. 1). The resulting morphology has been
called bicontinuous [ 1].
Possible forms of self-intersecting (non-embedded) sponges remain largely
unexplored. However, some results for self-intersecting 3-periodic minimal

,. The term is a useful general one, used originally by Coxeter in a similar sense [62].

437
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 437-470.
© 1999 Kluwer Academic Publishers.
438

surfaces suggest the morphological wealth of these structures. Multi-continuous


morphologies are possible (so far, up to eight interwoven labyrinths have been
found), as well as closed-cell forms and honeycombs [2, 3].

Figure 1: Some bicontinuous morphologies: a random sponge (a sea-sponge, drawn by Robert


Hooke using one of the earliest optical microscopes); and a translational unit cell of a triply
periodic hyperbolic surface, the "gyroid".

Translationally ordered (crystalline) sponges have been identified in a


variety of condensed materials:
•solid ionic electrolytes, e.g. a-Agl (ca. sA unit cells) [4]
•synthetic and natural zeolite frameworks, e.g. sodalite, Linde A, analcime
00-15A) [5, 61
•metallic alloy structures, e.g. WAl12, NaZn 13 (ca. lOA) [7]
•glassy, mesocrystalline inorganics, e.g. silicates, oxides (20-1 OOA) [8-16]
•lipids or surfactant lyotropic liquid crystalline "bicontinuous" mesophases
[17-231 ooo- 200A)
•block copolymer melts [24-28] (lOOOA)
•lipid membranes in biological cells [18, 29-31] (ca. 1 J.lm.)
•biomineralised skeletons, e.g. magnesian calcite in echinoderms [32] (5
J.lm.)

Disordered "molten" sponges are less well characterised to date, found in:
•surfactant-water "sponge" (L3) mesophases (tunnel radii ca. 50A) [33-35]
•late-stage spinodal decomposition domains in alloys, immiscible
copolymers ... (length scale dependent on time evolution, scale invariant
structures) [36]
•geological materials, e.g. limestones [37]
•wood, termite nests[38,39]
This paper is devoted to an explication of the pure geometry of these forms,
rather than a detailed analysis of the origins of these structures in physical
439

systems. The geometric survey is, however, skewed toward those shapes of
most relevance to condensed atomic and molecular materials.

Some differential geometry of elliptic, parabolic and hyperbolic surfaces

The characteristic feature of a sponge is its local saddle-shape. In contrast to flat


(parabolic) or convex (elliptic) surfaces, sponges - which are hyperbolic -
exhibit curvature towards the "inside" and the "outside" of the surface.
The curvature of a planar curve at a point is the inverse of the radius of
curvature of the osculating circle that best fits the curve at the point P, plus
neighbouring points, Pt and P2, as Pt, P2 -> P:
k(P)=R(Pr1
so that dimensions of curvature are L -l. There is a sign associated with the
curvature, dependent on the concave or convex nature of the curve at P. By
convention, if the curve is locally convex, k(P) is positive, if it is concave, k(P)
is negative (and if the curve is straight, k(P) is equal to zero).

Figure 2: (Left) The curvature of a curve is equal to the reciprocal of the radius of the circle fitting
a curve in the vicinity of points P (negative, say) and Q (positive). (Right) The curvature of a
surface at a point P is dependent on the curvature of the planar curves described by the planes
containing the normal at P.

A surface contains a one-parameter family of planar curves; formed by the


intersection of an arbitrary plane containing the surface normal vector with the
surface (fig. 2). These normal curves vary continuously with orientation of the
plane, so there exist principal directions whose normal curves achieve the
maximum and minimum curvatures k1 , k2 •
440
The curvatures of normal curves which subtend an angle of 8 with either
principal direction is given by Euler's relation:
kn =k1 cos 2 (8)+k2 sin 2 (8).
Along asymptotic directions the surface arcs are straight, so that kn = 0. But
asymptotic arcs contain the surface curvatures encoded in their geodesic
torsion:
tg =(k1 - k2 )cos(a)sin(a)
If the surface is locally flat (a flat point) or equally curved in all directions
(an umbilic), the asymptotic directions are undefined. Along other directions
this equation dictates the twist associated with stacking on the surface, and is
relevant to chiral molecular liquid crystals.
The curvatures of principal curves determine the surface curvatures at P;
these are commonly described by the Gaussian and mean curvatures:
Gaussian curvature: K = k1 .k2 (dimensions L-2)
Mean curvature: H = ( k1 + k2 ) (dimensions L-1)
If the mean curvature vanishes at all points on the surface, it is a minimal
surface. This term is misleading in many contexts of physical interest, and often
leads to confusion surrounding such structures in real systems. While it is true
that a surface of locally minimal area for a given boundary has constant,
vanishing mean curvature, minimal surfaces which are not constrained to lie on
fixed boundaries need not have least area. Indeed, it is easy to construct
minimal surfaces of arbitrarily large surface area within a given volume.*
The radii of curvature of a surface at a point, Rt and R2, are related to the
mean and Gaussian curvatures:
2 --K- ; R =H -~H 2 -K
R1 =H +...fr-H- 2
Gaussian curvature is an intrinsic quantity, and is dependent only on the metric
of the surface. (The metric is a local differential form that determines the line
element of curves on the surface. This is described in standard texts on
differential geometry). In fact, lengths, areas, angles and Gaussian curvature on
a surface are intrinsic quantities, and depend only on the metric. So distinct
surfaces in 3-d space may have the same (2-d) metric. Examples will be met
later.
Non-euclidean geometry links the Gaussian curvature of a surface element to
its intrinsic geometry:
•K > 0 =>Elliptic geometry
•K = 0 =>Parabolic (euclidean) geometry
•K < 0 =>Hyperbolic geometry
Models for elliptic and hyperbolic non-euclidean geometries can thus be
developed by analysing geometry on surfaces of constant positive and negative
curvature respectively. It follows from the saddle shape of sponges that the

• Once this paper is digested, examples will, I hope, spring to the reader's mind immediately.
441

natural setting for a non-euclidean analysis of sponge geometries is a surface of


constant negative Gaussian curvature - the hyperbolic plane. One final tool is
essential for the analysis of the global structure of sponges: topology.

Differential geometry and topology

Topology and surface geometry are inextricably linked in two dimensions by


the Gauss-Bonnet Theorem:
If Kda = 21tX (1)
surface
valid for a boundary-free surface. X is the Euler-Poincare characteristic, related
to the surface genus, g, by:
x=2-2g (2)
assuming the surface is orientable (so that two distinct sides can be traced on
the surface, unlike e.g. the Mobius band). As the genus of a surface increases,
so does the number of handles in the surface (or doughnut-like holes).

Figure 3: A genus g surface can be unfolded into a 4g-gon. If the genus is 1, the polygon is
euclidean, once g exceeds 1, the polygon is necessarily hyperbolic.

Multi-handled surfaces can be sliced open to form a disc, whose form


depends on the topology (and geometry) of the surface. For example, a 1-torus
can be folded out into a 4-gon by making circular cuts around equatorial and
longtitudinal loops of the torus. Since the internal vertex angles of the 4-gon
necessarily sum to 21t, the polygon lies in the euclidean plane, '£2. More
generally, a genus g surface can be unfolded using 2g cuts in to a 4g-gon, where
the internal vertex angles of the polygon sum to 21t [40]. Once g>l, the polygon
is hyperbolic (the vertex angles, 1t/(2g), are less than those required for a
euclidean polygon), and its natural embedding space is now the hyperbolic
plane, J/2 (fig. 3).
442

Alternatively, once g>l, X is negative, so the average Gaussian curvature,


fJ Kda
(K) =surface = 27tX (3)
IJdd A
surface
is negative.
In the case of ordered sponges, it is convenient to characterise topology by
the genus (or Euler-Poincare characteristic) of a finite surface module, that
forms the complete surface under a lattice of translations. As for any lattice,
there is no unique choice of unit cell. However, the mathematical
parametrisation of these structures is governed by a well-defined topological
unit cell (also called a "lattice fundamental region" [41]). This domain is the
primitive crystallographic unit cell of the oriented sponge, whose "inner"
surface is painted white (say) and outer black. The crystallography of the
ordered sponge is determined by its 3-d space group, and its "black/white" sub-
group. The sub-group excludes all symmetry operations of the complete
symmetry group that exchange orientation of the surface (black-> white), such
as 2-fold axes in the surface and inversion centres on the surface.
A unit cell of the surface contains boundary loops. We remove their
contribution to the surface topology by "gluing" points on the cell boundary
related by a lattice vector (as in Born von Karman boundary conditions in
quantum mechanics), yielding a compactified unit cell. Since the (ordered)
sponge is hyperbolic, its genus (per compactified topological unit cell) must
exceed unity. It turns out that genus two surfaces cannot be realised as sponges
in '£3, rather, they form meshes, whose channels form 2-d networks, embedded
in a single continuum. These surfaces can stack into smectic arrays and can
resemble sponges, despite their topological differences (fig. 4).

Figure 4: Example of a tetragonal stack of square mesh surfaces.

The minimum genus (per compactified topological unit cell) of an ordered


sponge is three. A number of examples of genus three sponges will be described
later in the paper.
443

Homogeneous sponges

Define a homogeneous sponge to be a hyperbolic surface with a fixed radius of


curvature, R, so that the sponge has zero mean curvature (ki=-k2=R-1) and
constant (and negative) Gaussian curvature (R-2) . This is the hyperbolic
analogue of the better-known homogeneous elliptic surface, the sphere, and the
homogeneous parabolic surfaces, the cylinder or plane.
There is a crucial difference between hyperbolic and the other geometries,
that goes some way towards an explanation of the importance of 3-periodic
sponges in physical systems. While the homogeneous sponge is analytically
convenient, it is a fiction in 3-d euclidean space, as unbounded hyperbolic
surfaces must have variations of the (albeit non-positive) Gaussian curvature
over the surface [42]. For now we persist with that fiction in order to derive
estimates of the global packing efficiency of sponges. A system that seeks to
maintain constant hyperbolic curvature is frustrated in '£3 ,and a number of
competing structures can form. It looks as if the least frustrated examples are
spatially ordered in '£3. That claim is explored further on in this paper. Before
that, the features of ideal homogeneous sponges are discussed.

1 2 3
VOLUMES:

Figure 5: Archimedes' Theorem (top), and its generalisation to non-euclidean geometries.


(bottom). The volumes of euclidean polyhedra of equal height scale as I :2:3, volumes of the ideal
polyhedra capped by elliptic, parabolic and hyperbolic patches of constant radii of curvature, R,
scale as 2:3:4.

Consider first the volumes on either side of a sponge. If the sponge is


homogeneous, these volumes can be swept out by displacement of the surface
by its normal, forming a one-parameter family of parallel surfaces. Parallel
surfaces are generated from a surface by transport along the surface normal
444
vectors by a constant distance (x) everywhere, so that the one-parameter family
of parallel surfaces share identical normal vectors.
The area of a patch on a surface is related to the area of the related patch on
a parallel surface by:
dAx =dA0 (1+2H0 x+K0 x 2 )
(This equation offers the simplest criterion for deciding whether the
displacement x is positive or negative. If the area shrinks under the parallel
displacement, x<O, if it grows, x>O.)
The volume of a homogeneous film of thickness x, consisting of a stack of
parallel surfaces to a homogeneous surface is given by integrating the previous
equation,

V x =A 0 • x( 1+ Hx + ~2 )

For example, the interior volume of a sphere of radius R (H=R-1, K=R-2,

x=-R) is related to its surface area by:


-R R2 ) A 0 R
VR =Ao.R ( 1+R+ 3R2 =3.
Similarly, the inner volume of a cylinder (H=(2R)- 1, K=O, x=-R) is:
( -R)
VR =A0 .R 1+- =A-0 R
2R
-
2
The volume to one side of a homogeneous sponge - from the surface to the
centre of curvature on that side- (H=O, K=R-2, x=R) is:

V =A .R(1+_E_)= 2AoR
R 0 -3R2 3
The total volume containing the sponge is twice this volume, so that
V= 4A0 R.
3
We can summarise these results by a differential geometric analogue of
Archimedes' celebrated theorem, which states that the volumes of a cone, a
hemisphere and a cylinder of equal height are in the ratio 1:2:3. The parallel
surfaces to spherical, cylindrical and homogeneous sponge elements foliate
elemental cones, pyramids and tetrahedra, whose volumes are in the ratio of
2:3:4 for fixed radius of curvature (fig. 5). (Archimedes was apparently proud
of his discovery he had it engraved on his tombstone. One wonders what he
would have thought of the non-euclidean generalisation?)
The homogeneous sponge radius, R, can be expressed in terms of the sponge
topology, since, from ( 1),
1/2
K= 21tX =>R= ( ~ )
Ao -2nx
445
or
A 3/2 3
=-
- (-21tx)112 V 4
h= 0 ~)

introducing the homogeneity index, h, which turns out to be a useful


dimensionless measure of the surface to volume ratio of a sponge. Note that we
have established only that this index is equal to 3/4 in the ideal limit of a
homogeneous sponge! For generic sponges, the value of h is calculated
explicitly from the sponge area, volume and topology, and it rarely equals this
ideal value.

Em beddings of sponges in euclidean 3-space, '£}

The area of an ideal hyperbolic disc (embedded in a hyperbolic plane of


Gaussian curvature R-2) grows exponentially with radius:

~1f =21t(cosh(;)-1) =41tsinh (;R) 2

(5)
2
=> A( r) .?lz =1tr2(1 +_!_ (!...) + - 1 (!...) +...)
4
12 R 360 R
so that significant portions of J/l must be excised (and the remainder reglued),
since the area of a disc in '£3 grows -Tl. This "surgery" on J12 induces
variations in the surface curvatures, so that the homogeneous embedding is
impossible in '£3 [43].
The departure from ideal hyperbolic geometry can be gauged by the
moments of the distribution of Gaussian curvature over the surface:

(:? = fJ Ki da/[ fJ Kdali ~


( } surface suiface
1.

It follows from (5) that the ratio of average area of a disc in an


inhomogeneous hyperbolic surface to that of a disc in the hyperbolic plane is:

(A) =1+ {(K)r2)2 ((K2f -1] +i[{(K)r2t ((K2~~ -1]]~1


AH2 360 (K) i= 2 ci (K) 1

This approach allows the generalisation of hyperbolic geometry to average


geometry on inhomogeneous sponges, given the Gaussian curvature
distribution, and resulting curvature moments.
Arbitrarily inhomogeneous sponges can be constructed. A more interesting
issue is that of quasi-homogeneous sponges: just how near the homogeneous
limit can smooth hyperbolic surfaces in '£3 approach? That question remains
446
unresolved, but some progress towards its resolution is to be found in the study
of the simplest hyperbolic surfaces in '£3, minimal surfaces.

Minimal surfaces in '£3

The investigation of minimal surfaces in '£3 has been a very fertile area for
mathematicians, crystallographers and materials scientists in the past decade.
Numerous new minimal surfaces have been discovered, and undoubtedly, many
more await discovery. So a complete survey of the field is impossible here (see
[44-46]).

Figure 6: Examples of generalised Enneper surface; left to right: 1-, 2-, 3-Enneper surfaces.

Consider for now only embedded , oriented minimal surfaces, free of self-
intersections. The simplest examples of these surfaces are those containing one
asymptotically flat "end", forming the family of generalised Enneper surfaces.
A generic n-Enneper member of this family has n + 1 hills and n + 1 valleys
about its centre (fig. 6). Notice that once n exceeds unity, the centre of the
surface also has zero Gaussian curvature, and the curvature inhomogeneity
increases with n. This notion will be canvassed in more detail later, using the
Gauss map.
The classic examples of minimal surfaces are the catenoid and the helicoid,
(fig. 7).
These examples illustrate two important relations between symmetry and
local geometry on minimal surfaces.

Figure 7: The catenoid and the helicoid.


447

The helicoid is "ruled" - it can be constructed entirely of straight lines. The


horizontal rulings lie along the asymptotic directions of the surface. The
presence of lines - linear asymptotes - in a minimal surface signals a 2-fold axis
along that line, and any minimal surface containing a 2-fold axis contains a
straight line. (That is the crystallographic technique used by Koch and Fischer
to derive new examples of 3-periodic minimal surfaces [47-49]. Following
them, we call minimal surfaces containing straight lines balanced.)
The curve tracing a principal direction of the catenoid around its waist is a
planar curve, and lies in the mirror plane relating top and bottom halves of the
surface. In general, a planar principal curve - a plane line of curvature - lies in a
mirror plane, and vice versa. These symmetries on minimal surfaces allow the
construction of ordered surfaces, forming 1-, 2- and 3-periodic forms in '£3 (e.g.
the 1-periodic helicoid).
Enneper-like minimal surfaces can be stacked to give 1-periodic "saddle
towers" which smoothly connect 2(n+ 1) planes along a central axis (fig. 8).

Figure 8: (Left) Saddle tower constructed from stacked J-Enneper surfaces. (Right) Scherk's
surface, a 2-periodic minimal surface.

A simple saddle bounded by four vertical edges can be extended periodically


to give the 2-periodic Scherk minimal surface (fig. 8). This structure offers an
interesting model for lamellar grain-boundary structures, actually observed in
domain walls of block copolymers forming lamellar microdomains [27, 28].
(Generalised 2-periodic Scherk (minimal) surfaces exist for other twist angles.)
All 0-, 1- and 2-periodic structures have "ends", which are asymptotically flat.
Their topology, x,
remains finite per unit cell, while their surface area per unit
cell is unbounded, so (K) -t 0. For example, introduction of periodicity along
the z-axis of the Scherk surface leads to the 3-periodic (tetragonal) tD minimal
surface. tD surfaces can thus be viewed a periodic arrays of twist boundaries (or
grids of screw dislocations), currently of interest in understanding novel
tetragonal mesophases in chiral smectic thermotropic systems [50].)
448
The sole remaining candidates for quasi-homogeneous sponges - assuming
that disordered minimal surfaces free of ends cannot form in ~3 - are the 3-
periodic minimal surfaces. Embedded examples are called "embedded triply
periodic minimal suifaces", or ETPMS [51, 52]. These surfaces are the simplest
examples of ordered sponges, carving space into two interconnected 3-d
complementary "labyrinths". The labyrinths are conveniently represented by
their labyrinth graphs, which describe the channel axes and connectivity within
the sponge (formally defined by Kusner [53]). The labyrinth graphs are fully
interpenetrating, in the sense (introduced by Wells [54]) that any loop of one
graph surrounds an edge of the other. Further, the graphs must be related by
inversion centres, which are located at the flat points (zero Gaussian curvature)
of the ETPMS. If the ETPMS contains straight lines, these lines are 2-fold
symmetry axes which exchange one graph for the other, so that both labyrinths
of the balanced surface are identical. Nodes of the labyrinth graphs must join at
least three edges.

Figure 9: Labyrinth graphs of the genus 3 gyroid. viewed (right) almost down the [100] (four-
fold) axis and (right) the [Ill] (three-fold) axis.

Trivalent labyrinths are a feature of the topologically simplest, most


symmetric ETPMS (genus 3, space group Ia~d.. ), known as the gyroid [51].
Ironically, this surface is one of the most difficult to visualise, as it contains
neither mirror planes nor in-surface 2-fold axes. Each labyrinth graph of the
gyroid is a uniform 3-connected 3-periodic graph, containing equal edges, and
equal vertex angles of 21t/3, whose nodes are located at the Sa or 8c Wyckoff
sites in the chiral space group l4I~2 (which is thus the black-white subgroup of
the gyroid), forming theY* lattice complex [55]. The intergrown +y* and-y*
graphs (one left-handed, the other right-handed) are separated by the gyroid
(fig. 9) .

.. The overbaris replaced by an underbar here for typesetting convenience.


449
If pairs of nodes in a Y* graph are fused by shrinking edges along [211] axes
of the cubic lattice, interpenetrating four-valent diamond graphs result (nodes at
8a sites in Fdlm, forming the D lattice complex) [43]. These define the
labyrinths of the D surface (fig. 10, genus 3, symmetry Pn.3.m). One further
stage of node condensation results in six-valent nodes, linking orthogonal edges
along [100] directions, forming intertwined J lattice complexes (at 3c Wyckoff
sites, symmetry Pm.3.m). These labyrinths define the P surface (fig. 10, genus 3,
full space group Im.3.m).

Figure 10: The D and P surfaces, viewed approximately along [110] and [Ill] diretions
respectively.

These three forms are both the simplest geometric realisations of a sponge in
'E3 and the most commonly encountered models for condensed materials (listed
in the introduction).
Other surfaces are less symmetric, and/or of higher genus. Other genus three
ETPMS include the H surface, the tetragonal CLP surface, and lower symmetry
surfaces of identical topology to the P, D and gyroid surfaces: the rPD
(rhombohedral), rG , tP (tetragonal), tD, tG ..... Among the higher genus cubic
ETPMS are the genus 4 1-WP surface, the genus 6 F-RO surface and the genus
9 Neovius (or C(P)) surface.
A number of ETPMS can share the same space group- for example the P, 1-
WP and Neovius surfaces all have symmetry lm.3.m. (An often-ignored corollary
is that structural identification of hyperbolic films, such as the amphiphilic
bilayer forms in bicontinuous "cubic phases" cannot be found from
measurement of symmetries alone.)
The genus (per compactified topological unit cell) of a triply periodic
hyperbolic surface can be determined from the form of either labyrinth graph,
depending on whether the tunnels pass through faces, edges or vertices [56]. An
alternative, intrinsic, approach to surface topology comes from analysis of the
differential geometry on the surface.
450

Table 1: Some intersection-free triply periodic minimal surfaces (ETPMS). The space group refers
to the symmetry of the surface including symmetry operations which exchange sides, the b/w
subgroup excludes those elements. The surface topology is listed by genus per lattice fundamental
region (g (lfr)) and genus per conventional crystallographic unit cell (g (cryst)). Lower symmetry
variants of many of these surface exist.

Surface space group blw subgroup cg (lfr) g(cryst.) homogeneity balanced?


Index

p Imlm Pmlm 3 3 0.7164 yes


D Pn.lm Fdlm 3 2 0.7498 yes
gyroid Iald 14132 3 5 0.7667 no
H P6)/mmc P2m2 3 3 variable yes
CLP P42/mcm P42/mmc(v) 3 3 variable yes
1-WP Imlm Imlm 4 4 0.7425 no
VAL emma Cmma(2c) 5 5 ? yes
F-RO Fmlm Fmlm 6 6 0.6577 no
Neovius Imlm Pmlm 9 9 0.664 yes
(C(P))
O,C-TO Pmlm Pmlm 10 10 ? no
s Iald 14ld 11 11 ? yes
C(D) Pn.lm Fdlm 19 19 ? yes

(Many other ETPMS are known, cf. H. Karcher, Manuscripta Math.,64, 291-357 (1989); W. Fischer and E.
Koch, Phil. Trans. R. Soc. Lond., A, 354, 2105-2142 (1996); A. Fogden and S.T. Hyde, Acta Crystallogr.,
A48, 575-591 (1992).)

The Gauss map

There is a strict connection between topology and integral curvature for 2-d
surfaces, demanded by the Gauss-Bonnet Theorem (eq. (8)). The integral
curvature of any surface patch can be "read off' the Gauss map: it is equal to
the solid angle swept out by the surface normal vectors within that patch (fig.
11). (If the Gaussian curvature varies from positive to negative over the surface,
a sign must be associated with the Gauss-mapped region accordingly, and the
sum of solid angle calculated.)
So the relative reorientations of the surface - gauged by the movements of
the endpoint to the unit normal along the surface - must in fact reflect the
surface topology.
Look, for example, at the Gauss map of a hyperbolic patch (fig. 11). Loops
around non-flat regions of the saddle map into loops on the Gauss sphere,
traversed in the opposite sense. Loops surrounding flat points are traversed a
number of times. For example, the Gauss map of paths enclosing the flat central
area on i-Enneper saddles (i>1) winds on the unit sphere i times. A monkey
saddle (i=2) is a double-cover (as two points sharing one normal vector can be
451

found - except at the origin), and so on. The map contains singularities of
various orders, and the structure of these singularities is a good measure of the
sponge form. The source of these singularities are the flat points at the centre.
The order of the (i-th say) flat point, {31, is given by the winding number, ({31+1).
The circle of connection between the surface topology and differential
geometry (via the Gauss map) is cemented by the Riemann-Hurwitz
requirement, that determines the sum of all flat point orders:
~); =-2x =4(g-1) =4s, (6)
i
summing over all i flat points within a single topological unit cell. For example,
the P surface contains eight ({3=1) flat points (so that g=3 and s=2).

Figure 11: Gauss map of a portion of a saddle.

These multiple windings impose a complex Riemann surface structure on the


Gauss map. If the map contains s sheets over the unit sphere, by the Gauss-
Bonnet theorem (eqs. (1), (2)):
s= -x =g-1.
2
For 3-periodic minimal surfaces, the range of the Gauss map of the complete
surface is exhausted by a single topological unit cell of the surface as domain.
The genus of the Gauss map, g (=s+ 1), is identical to that of the compactified
module formed from the topological unit cell.
Since the genus per topological unit cell must be an integer, the number and
order of flat points on an ETPMS is strongly constrained, and higher genus
ETPMS necessarily have a higher density of flat points, or higher order flat
points (which have larger flat domains - as in the higher order Enneper
examples). For example, the 1-WP surface contains 6 second-order flat points
per topological unit cell.
It is plausible then that the homogeneity of a sponge is related to the order
and distribution of flat points over the surface. That means that the variations in
curvature over a 3-periodic sponge is related to the underlying genus of the
sponge. Comparison of the second curvature moments for cubic genus 3, 4, 6
452
and 9 cubic ETPMS supports the contention of a link between inhomogeneity
and topology (fig, 12).

3 Neovius (C(P))

2.5

<K2>f<K>2 2
1-WP
1.5 • •
F-RD

P,D,gyr.

-x
4 6 8 10 12 14 16 18

Figure 12: Plot of variations of Gaussian curvature (second moment) vs. topology per lattice
fundamental region (X) for some cubic ETPMS (the P/D/gyroid, 1-WP, F-RD and Neovius
surfaces).

Recall that deformations of cubic ETPMS result in tetragonal and


rhombohedral, orthorhombic ... minimal surfaces. The variation of homogeneity
among these one-parameter families is complex, as are the domains of existence
of such surfaces [57]. Without exception however, the highest symmetry
members (cubic) are the most homogeneous.
The search for quasi-homogeneous sponges is thus narrowed to the most
tractable examples of ETPMS: low genus (3) and high symmetry
(rhombohedral, tetragonal and, principally, cubic) cases, viz, the P, D and
gyroid surfaces, and their lower-symmetry variants, the rG, rPD, tP, tD and tG.
To a crude (homogeneous) approximation, these surfaces are intrinsically
identical, provided they are scaled to have a common average radius of
curvature, (R). This radius can be written in the terms of the homogeneity index
and topology (cf. (4)) :

(R)=(-h-)3
1

a (7)
-21tX
(where d is the lattice parameter (of the lattice fundamental region) of the
sponge of homogeneity index h, and Euler-Poincare characteristic, z). Suppose
that two quasi-homogeneous sponges are locally identical. That intrinsic
hyperbolic equivalence is accompanied by an apparent rescaling in 3-d
euclidean space, according to (7). For cubic sponges we have the epitaxial
constraint for ratios of lattice parameters of surface sharing common ( R):
1

~-(h2 X 2 )3 (8)
d2 ~X1
453
It follows (using the data of Table I) that the lattice parameters of ETPMS of
the same average radius of curvature scale as:
( 1.576)gyroid:( 1)v:( 1.447)/- wp:( 1.279) p:(2.082)Neovius·

The Bonnet transformation

Equations (7,8) constrain the 3-d euclidean scaling of quasi-homogeneous 3-


periodic minimal surfaces sharing a common average hyperbolic geometry. It
turns out that the most homogeneous ETPMS, the D, P and gyroid are, in
addition, locally identical - not only is their average Gaussian curvature equal
(for appropriate scaling) but the curvature distributions are identical (i.e. the
surfaces are isometric). Isometric minimal surfaces are related by the Bonnet
transformation (named after the French geometer who first revealed this
possibility last century). This transformation is a remarkable example of the
flexibility of em beddings of hyperbolic structures in '£3, which can be realised
for minimal and constant mean curvature surfaces (plus "Willmore surfaces")
[58]. Its effect on minimal surfaces is to "bend" the surface (or, more accurately
shear it), without changing the Gaussian or mean curvatures at each point! Thus
the surface metric remains unchanged by the transformation, so that all lengths
and angles measured along the surface remain fixed.
Any minimal surface can be transformed according to this transformation.
For example, its effect on the catenoid is to unwind it to the helicoid (fig. 13).

Figure I 3: The Bonnet transformation bends the helicoid into the catenoid. All surfaces are
locally identical. only their global embedding in space changes.

The catenoid and helicoid are "adjoint". They are natural endpoints of the
transformation, with all lines of curvature on the catenoid transformed into
asymptotes on the helicoid. (For example, the central closed loop surrounding
the waist of the catenoid, which is a (mirror) plane line of curvature, becomes
the screw axis of the helicoid, a (2-fold) linear asymptote.) The adjoint surface
to a 3-periodic minimal surface is itself a 3-periodic minimal surface, however,
an adjoint surface to an embedded 3-periodic minimal surface need not itself be
embedded (and often isn't). Some ETPMS are self-adjoint (e.g. the 1-WP
surface).
454
The most important examples of adjoint surfaces, both 3-periodic and
embedded, are the P and D surfaces. (The gyroid is a rare example of an
embedded intermediate surface.) All mirror planes of the P-surface are
transformed into 2-fold axes on the D surface. In particular then, the skew
hexagon bounded by mirror planes on the P transforms to the Petrie polygon of
the cube on the D (fig. 14).

Figure 14: Bonnet related hyperbolic hexagons, made up of six equivalent triangles on the P (left)
and D (right) surfaces. Equivalent points on each surface share a common normal vector. The
centres, 0, are flat points.

Because the Bonnet transformation is an isometry (metric-preserving), the


ratio of lattice parameters of adjoint ETPMS is fixed. We can estimate the
metric relations between the P and D surfaces, assuming they are both ideally
homogeneous, from elementary hyperbolic geometry.
From fig. 14 we have:
OA ap
AB = {ia 0 •

The hyperbolic triangle(s) OAB obeys the standard identity from hyperbolic
geometry [59]:
cosh ( R
AB) =cosh (OA)
R cosh (OB) . (OA)
R - smh . (OB)
R smh R cos(AOB A )

where R is the radius of curvature of the (assumed) homogeneous triangle (of


Gaussian curvature R-2) and AOB is the internal angle of the triangle at 0.
Since OA=OB and A0B= 3 ,
"' 7t

R=
OA ( 2cosh (AB))2
R .
But AB is one quarter of a complete circle, which surrounds tunnels of the P
surface. Assuming (again) homogeneity, this implies that 4AB=21tR, so that
455

AB 1t
R 2
1

ap sinh- 1( 2cosh(~)Y
:.-=2v2 =1.188
ao 1t
The true value of this ratio is 1.22 (see above), a difference of less than 3%
from this homogeneous estimate. Clearly then, these simple sponges are close to
ideal.
Table II: Distinct phases of some liquid crystals believed to be related by the Bonnet
transformation of underlying two-dimensional hyperbolic surfaces. The final column indicates the
discrepancy between theoretical and measured lattice ratios.

-
Material phase symmetry a(A) related rallo Bonnet raUo dllrerenee
atoms/molee sponge
ules
perunlteell

lyotropic liquid
rystals:

monoolein-water bicontinuous Pnlm 84.5 (a) -800 0


cubic
(lipid) 1.66 1.5757 5%
lald 140.8 (a) -3600 gyroid

OOAB-eyclohexane- . inlm 116 (C) -600 p


Willer (b)
(surfactant) 1.26 1.2793 2%
Pnlm 92 (C) -800 0

lumnotropic liquid
rystals:

R-bipbenylcarboxylic SmecticO Iald 170.8 <0 1650 gyroid


acid (d)

" . Pnlm (h) 111.9 (0 3150 D (e)


1.53 1.5757 3%

CB15-E9 (g) Blue Phase I 14132 4970 (h) -3 X loB (i) gyroid (j)
1.59 1.5757 1%
Blue PhaseD 1'4232 3120 (h) -1 x to7 (i) D (j)

Refs. & Notes:


(a) S.T. Hyde. S. Andersson. B. Ericsson. and K. Larsson. Z Kristallogr. 1984,168: 213-219.
(b) DDAB= didodecyldimethyl ammonium bromide, a double-chained cationic surfactant.
(c) P. Barois, S.T. Hyde, B.W. Ninham, and T. Dowling,I.angmuir 1990, 6: 1136-1140.
(d) The molecules forming the lald and Pnlm phases differ slightly: X=N~ for the la3d phase, and CN for the Pn3m phase.
(e) Different indexing to that discussed in (1) ..
(I) A.-M. Levelut and M. Clerc, "Structural investigations on the "Smectic D" and related mesopbases"; preprint (1997).
(g) CB15 is a strongly polarisable cbiral molecule; E9 is a (nematic-forming) additive.
(b) Estimated from "Bragg wavelengths" reported in: G. Heppke, B. Jer6me, H.-S. Kitzerow, and P. Pieranski, J. Phys.
(France) 1989, 50: 2991-2998; assuming the refractive index is unity.
(i) Calculated assuming a specific volume of I cm 3g·l.
(j) Proposed by Pansu and Dubois-Violette, J. Phys. (Fr.), Col/oque. C·7. (1990), 281.

The epitaxial constraint (8) offers a useful probe of underlying 2-d


hyperbolic structures: if these ratios of lattice edges are detected in physical
systems, such as covalent frameworks, alloys and liquid crystals, they suggest
456

that the system may be inherently 2-d (and non-euclidean). The Bonnet
transformation - or at least the resulting "epitaxial" relation between associate
ETPMS - crops up in a range of real materials - solid and liquid crystalline -
subject to phase transformations. (Table II gives some examples among liquid
crystals. The earliest application of this transformation was to the austenite-
martensite transition in carbon-containing steels, for which the Bonnet
transformation explains the (Bain) latttice correspondence, and relative
orientations of lines in parent and product phases [6b]).
The isometric epitaxial ratios among these materials point to the existence of
a common underlying radius of curvature in parent and product phases and an
intrinsically 2-d structure. The transformation route need not follow that of the
Bonnet transformation. A continuum of ETPMS, including tetragonal or
rhombohedral genus 3 surfaces, links the gyroid D and P structures [60]. The
origins of this intrinsic curvature are various and are explained in detail
elsewhere.

Faceting of sponges: infinite polyhedra

Abandon, for now, the differential geometric approach, which distributes


curvature as evenly as possible over sponges, and concentrate the Gaussian
curvature at isolated vertices of the sponge, and mean curvature along edges
linking vertices. The resulting structure is an infinite (facetted) sheet, creased
into planar faces. As with smoothly-curved sponges, the sub-volumes defined
by the infinite polyhedron - 2 of them if it is embedded - tile E3. If the number
of edges in each face is n, and z of these faces meet at each vertex, the numbers
of faces and edges are F = Vz ; E = Vz (for V vertices). It follows from Euler's
n 2
Theorem (generalised to any polyhedron of Euler-Poincare characteristic, X)
that
(n-2){z-2)=4- 2 ;x. (9)
The sign of the product on the LHS of (9) sets the geometry of the
polyhedron. Conventional (convex) polyhedra are elliptic (X is positive, so that
(n-2)(z-2) < 4 ). Planar (or cylindrical, toroidal) polyhedra have (n-2)(z-2) =4.
Hyperbolic "infinite" polyhedra have (n-2)(z-2) > 4. These arise from tilings of
the hyperbolic plane. As for conventional convex polyhedra, Platonic,
Archimedean, (etc.) infinite polyhedra exist in '£3 (see earlier) [54, 61]. The
generalisation of Platonic polyhedra to the hyperbolic case has been explored
partly by Coxeter and Petrie; Coxeter calls the one class of the most symmetric
such polyhedra "regular sponges" [62, 63]. Just as regular polyhedra contain
rotation axes of symmetry through all vertices, and all face centres, regular
sponges contain rotary-inversion axes, such that faces are alternatively pointing
upward and downward around each vertex (so that, in particular, regular
sponges must have an even number of faces about each vertex). There are only
457

three regular sponges in '£3, found by Coxeter and Petrie in 1926 [62]. These
are the P-like packing of regular hexagons, four around each vertex, fonning
square holes, denoted by Coxeter {6,414} (the final entry denoting the tunnel
polygons, or "collar rings"), its dual, the P-like tiling of squares, {4,614 }, and
lastly, the D-like packing of regular hexagons, {6,613} (fig. 15).

Figure 15: The regular sponges (clockwise from top left: {6,414},{4,614},{ 6,613 }.

Triangular infinite polyhedra - close-packing on sponges

Consider tilings, with all faces assumed to be regular triangles. At least three
faces must meet at each vertex in a polyhedron: z=3 yields the tetrahedron, z=4
the octahedron, z=5 the icosahedron (all convex polyhedra). The planar case
results when z=6, and z>6 results in infinite polyhedra. The result is topological
only (and we can replace z and n by averages, although not all solutions of
Euler's relation can be realised in <£3).
These tilings define the centres of triangular close-packed discs in 2
dimensions, the lower-dimensional analogues of tetrahedrally close-packed
arrays described elsewhere in this volume. It is of some interest to explore the
behaviour of disc close-packings on the hyperbolic plane. Clearly, these
packings admit a higher coordination number than six (characteristic of
458

(euclidean) planar close packing). Indeed it follows from the discussion in the
previous paragraph that an arbitrarily high coordination number is available in
2-d hyperbolic space.

Figure 16: The arrangement of equivalent neighbouring discs in close-packing, In this hyperbolic
example, 7 discs surround each disc, and the disc centres lie at the nodes of a {3,7} tiling ofthe
hyperbolic plane. (The picture is necessarily distorted to lie in the page.)

The packing efficiency of hyperbolic discs follows at once from the


hyperbolic equilateral triangle whose vertices are located at the centres of three
touching discs (fig. 16).
Each triangle encloses an area of 3/z times the area of the Voronoi region
associated with a single disc. The integral curvature of each triangle is given by
its angular defect, 1t(6-z)/z, so the integral curvature of the regular hyperbolic
Voronoi z-gon (abcdefg) surrounding each disc,
KA _ Avoronoi _ 1t(z-6)
- Voronoi - R2 - 3
The z triangles around each disc form a regular z-gon (ABCDEFG), larger than
the Voronoi polygon, of edge length 2r. This polygon has outradius 2r, so that
(applying the standard formula relating edge length and outradius for hyperbolic
polygons):

or,
cosh(;)= tcosec(~).
This allows the area of each disc to be determined (5), viz.

~(;) =21t( tcosec(~) -1}


459

The packing efficiency of the discs is thus:

A(r) =
6(.!.cosec(~)
2 z
-1)
Avoronoi z-6
(The formula holds also for tiling of the elliptic plane -such as close packing on
the sphere - for appropriate z (less than 6), if the denominator is replaced by
(6-z)).
The resulting variation of packing efficiency with Gaussian curvature of the
underlying space (scaled by the disc radius) or, more concretely, coordination
number, is plotted in fig. 17.
It follows from the formulae above that the minimum close packing
efficiency is 2--J3("-./3-1) (ca. 85%), realised for the {3,3} packing on the sphere
(where the disc centres lie on the vertices of a regular tetrahedron), the planar
packing {3,6} has the (well-known) packing fraction of7ti--J12 ca. 91%), while
the most efficient packing is realised for the hyperbolic {3, oo} array, viz. 3/rc
(ca. 95%).
Euclidean embeddings of hyperbolic 2-d layers - such as sponges - thus offer
subtrates for dense 2-d (hyperbolic) disc packings. (Hyperbolic discs describe
well packed 3-d spheres whose centres lie in the sponge, since the resulting 2-d
sections in the hyperbolic sponge are close to ideal hyperbolic discs.) However,
the packing efficiency must be modified relative to the ideal hyperbolic fraction
by a factor accounting for the accompanying curvature variations in the sponge,
viz. (5).

IIS)'IV'tolie
0.94 linit
~
n
,;
.2 0.92
ts
.!! 0.9
l4Mr
rfsWr
J
~
,12
II

l>o 0.88
:H

coozdillllion number

Figure 17: Variation of packing efficiency of discs with coordination number on elliptic,
parabolic and hyperbolic planes.

Geometric realisations of these triangular tilings have been studied most


fully by solid state chemists, particularly by Alan Wells, who reported examples
of triangular polyhedra- all of whose faces are regular- for z=7, 8, 9, 10, and
12 in <£3 [54], see also [7]. They lead to a number of 3-d arrangements of
regular polyhedra in 3-d space that are found in inorganic crystals. The
460
"regular" (in a loose sense) polyhedra are triangulations of 3-periodic cubic
sponges whose labyrinths and symmetries mimic the simpler cubic ETPMS: the
gyroid, P, D and I-WP surfaces. All of these examples have 3-rings enclosing
the sponge channels (so that the "collar rings" are triangles), and are related to
face-sharing packings of conventional convex polyhedra, including tetrahedra,
octahedra, icosahedra and snub cubes. The prevalence of further examples of
triangulated sponges remains unknown, and deserves attention. One example,
{3,9} is illustrated in fig. 18.

Figure 18: (Left) Portion of a {3,7} sponge (Note the arrowed proto-tunnels). (Right) The infinite
polyhedron {3,9} defines all unshared faces in the arrangement of octahedra (transparent) and
icosahedra shown .

All of these {3,z} arrays represent sphere packings in '£3 [7, 64], usually z-
coordinated, whose 3-d packing efficiency (distinct from their 2-d efficiency!)
varies from very low ( {3, 7 }), to cubic close-packing, in the {3,9} polyhedron.
Their relevance to solid state structures lies in their realisation as atomic
arrangements (often somewhat distorted) in a variety of solids, particularly
metallic alloys. An incomplete list of triangular infinite polyhedra found to date,
and their related atomic structures, is in Table III.
Given that no regular sponges contain triangular faces, none of these
examples have zero mean curvature, and complementary labyrinths are distinct.
These triangulated sponges "close up" with increasing x, so that the {3,12} case
- which is topologically identical to the I-WP surface- has mean curvature so
far from zero that one labyrinth of the polyhedron almost fills space, to the
exclusion of the other (cf. [541 fig. 17.9).
That feature is worth noting, as it sheds some light on the nature of
embedded and non-embedded sponges. There is an upper bound in the integral
curvature of a hyperbolic surface patch containing a single umbilic, of order {3.
That bound follows at once from analysis of a growing hyperbolic disc, which
can, at most end up asymptotically parallel along the (/3+ 1) ridges and (/3+ 1)
valleys. The topological form of such a disc, which cannot gain any extra
461

integral curvature without forming self-intersections, leads to an upper bound of


-27t(,B+ 1) for the integral curvature of a patch enclosing an umbilic of order {3.
(That bound is evident from the vertex angles of the bounding frame, and the
Gauss-Bonnet formula).

Table 1/1: Some infinite Platonic polyhedra containing only (equilateral) triangular faces,

Connectivity Underlying xrv Space group Related atomic structure Ret- nodal poly. Complementary
(z) ETPMS (vertices marked (II polyhedral stacklna
tunnd poly.
7 D -1/6 Rll WAIJ2 substructure I it,.'OS. OCL self-complementary
8 p -1/3 Fml< NaZn[ZNt2 ). Brazilian opal I snub cubes distoned icosahedra
• gynna l;t_>U _>IHCp•
[OJ in garnet " OCt.. let seu-comp ementary


9 1-WP -ill
hl,m
ml
pyrocntore. 1w, we,c
W[Ailzl I
OCI.
icos .. oct.
seu-comp ementary
complex
9 D Fmlm let.. OCL complex
JOlin s~~::;,l ideal
10 p -2/3 unlikely (non-bonded vertices OCI. complex
closer than bonded vertices)
12 1-WP -I OCI. com lex

(*ref. I examples can be found in [54]; ref. 2 in [6].

Rescaling integral curvature in terms of the Euler-Poincare characteristic, x.


gives the result,
X ~-(~+1)
so that the number of faces sharing a common vertex in an embedded sponge
must obey the inequality:
2n(~+2)
z~ .
n-2
If the number of faces exceeds this maximum (for a given {3) the surface
necessarily self-intersects in E3. (These non-embedded sponges are discussed
very briefly later.Possible forms of the maximally curved embedded examples
for n=4 are shown in fig. 19.

14.81

Figure 19: Simplest embeddings of {4,z} sponges, whose connectivity z is at the upper limit for a
given umbilic order,~-
462
Generalised infinite polyhedra

The construction of infinite polyhedra is not, of course, confined to polyhedra


with triangular or square faces. Generic polyhedra are the result of any tiling of
J(2, coupled with appropriate collar rings, formed by identification of some
edges of that tiling. Quasi-regular polyhedra can be characterised in terms of
subgroups (due to formation of collar rings) of discrete groups on J/2.
Appropriate sub-groups for embeddings in 'E3 remains an open problem,
although some specific cases have been dealt with by Coxeter [62, 63]. Indeed,
construction of these polyhedra from the hyperbolic plane leads to examples
which are embedded in '1!1, for arbitrarily large n! For example, square tilings
{4,2n} forming 4n-2 collar rings result in n-d sponges whose vertices lie at the
nodes of the generalised simple cubic lattice, zn
(e.g. {4,4} gives the square
lattice, {4,614} the simple cubic lattice ... ). Once the embeddedness constraint is
abandoned, an even richer array of forms is revealed. In that case, polyhedral
tessellations can also be realised as sponges. For example, {4,4n}, can be folded
to give (non-embedded) sponges which are equivalent to the partitioning of 'En
into closed cells, that are n-d measure polytopes (the cube, the hypercube, .. ) (cf
fig. 20).

Figure 20: Vertex configuration of {4,6} and {4,12} sponges. The former yields an embedded
sponge {4,614} (see above), the latter results in a non-embedded sponge, partitioning '1!'- into
cubes.

4-connected nets as reticulations of sponges

Just as the edges of a polyhedron form a closed net, the edges of infinite
polyhedra, or facetted sponges, form 3-periodic networks. Analysis of these nets
as reticulations* of sponges allows for some useful, and counter-intuitive,
relations linking the net density and ring structure.

,. This term too has an impeccable pedigree. It was used in this sense by Newton.
463
We can estimate the geometric density of 3-periodic nets - the density of
nodes - assuming they tile quasi-homogeneous sponges. That assumption is a
reasonable one for many zeolite frameworks, discussed elsewhere [67]. The
geometric density is related to the surface area (A, per arbitrary unit cell) of the
sponge, the unit cell volume, V, and the area per vertex ( .Q):
A -2nx
Pg =QV =QV(K)
invoking the Gauss-Bonnet theorem. Introducing the homogeneity index gives
the scaling of geometric density with topology:
1
h ! or Pg =h- (-21tX)2
Pg =-(K)2 3 -- ,
Q - N
Q2
where N is the number of vertices within the volume containing the surface of
topology X· In terms of the surface rings and net connectivity:
1
2

(10)

For 4-connected nets then:


= (21t)l/2h(n2 -4)1/2 (11)
Pg g3/2 n
2
The area per vertex must lie between the area projected onto '£2, and that
projected onto 912. Assuming equal edge lengths in the net, the ratio of these
areas is close to unity provided (n2-4) is small- which is certainly the case for
low density nets - so that the .a-312 term in the expression for the geometric
density is approximately constant. Since h ... ~ for quasi-homogeneous sponges,
the geometric density should mirror the expectation of (10). If this approach is
correct, it offer a number of insights into the geometric and topological
requirements for low density chemical frameworks, be they microporous
catalysts, or mesoporous separation and storage materials.
For example, (1 0) states that density increases monotonically with increasing
(surface) ring-size (n2). This result is counter-intuitive from a 3-d perspective,
but has been observed in zeolite nets, where the lowest density frameworks are
those with smallest (3 and 4) rings [68, 69].
Alternatively, it follows from (10) that low density frameworks can be
achieved by reducing the net connectivity, z. That, in fact, is the basis of the
series of very large-pore "interrupted frameworks" (containing dangling bonds
which lower z) reported for aluminium phosphate frameworks. Families of
ultra-low density silica frameworks, whose pores vary from 20A-100A have
464

been recently synthesised [8-16]. These mesoporous zeolite-like solids are silica
multilayers, so the in-layer connectivity is less than that of a monolayer. Again,
the effective connectivity is reduced to 3 , as the fourth bond is used to stitch the
layers together [70, 71]. Some hypothetical hyperbolic double-layer frameworks
reticulating the P and gyroid ETPMS have been reported [72].
A number of the better-known zeolites can be recognised as simple 2-d nets,
reticulating sponges [6]. For example, the sodalite (4.4.6.6, P), faujasite
(4.4.4.6, D) rho (4.4.4.6, I-WP) analcime (4.4.6.6, gyroid) and Linde A (4.6.4.6,
P) frameworks contain only 4- and 6-rings on the surface.
The smallest average ring size (n2) is realised for the faujasite and rho nets
(n2=48/11). Since the homogeneity index is higher for the D than the I-WP, we
expect faujasite to be the rarest net, as found. Indeed, from eq. (11), we
expect 7 FAU = hFAu = 0·66 =0.92(accounting for the nonzero mean curvature of
rRHo hRHo 0.72
the faujasite framework [70]); the actual ratio for idealised nets is close to this,
viz 0.89 [7].
A further expectation of this hyperbolic analysis is that Bonnet epitaxial
constraint is expected for quasi-homogeneous nets with the same surface ring-
size (n2), since the area per vertex is conserved. That explains the near-ideal
lattice ratio of sodalite and analcime (Table II) (according to the Bonnet
relation), which tile isoareal D and gyroid ETPMS respectively.
This approach is particularly simple for the nets listed above.However, in
general, the correspondence between the net and an underlying hyperbolic
sponge is less readily determined. It is likely, for example, that nets containing
edges lying in 3-fold axes of the network symmetry cannot be a reticulation of
an intersection-free sponge. In those cases, the underlying sponge contains
simple surface branch points (in addition to the flat points, which form branch
points on the Gauss map) [48]. In order to generalise this analysis then, further
study of non-embedded sponges in needed. Indeed, the simplest TPMS found to
date containing branch points, the W1-00 surface of Koch and Fischer [48], can
be reticulated to yield the PmJ.n four-connected melanophlogite network (the
network of edges in the starting structure for the Weaire-Phelan foam).

3-connected nets and novel graphitic "schwarzites"

Low density 3-connected nets necessarily contain a substantial fraction of 6-


rings. Nets reticulating sponges, with an average surface ring-size exceeding 6,
have been proposed as model structures for graphite-like hyperbolic analogues
of the (elliptic) fullerenes [74-76], dubbed "schwarzites" [77]. The best-known
fullerene, C6o. is a (5.6.6) reticulation of the sphere (topologically, at least). The
analogous (7.6.6) schwarzite has an average ring-size of 63/10, so that, by (9),
V=-42z. Reticulations of this net on sponges with the topology of both the P
and D surfaces (196 atoms per unit cell) have been studied [60, 78]. Indeed,
465
already in the 40's, hypothetical graphitic frameworks were described as
possible models for amorphous graphitic carbon [79] (fig. 21).

Figure 21: Polybenzene frameworks, which tile the P and D surfaces.

These "polybenzene" frameworks were later rediscovered, and shown to be a


(6.8.8) reticulation of the D and P surfaces; the D tiling is estimated to be
particularly favourable [80, 81]. Polybenzene is an excellent example of the
duality of perspectives of 3-periodic nets: they are at once hyperbolic 2-d nets
and 3-d euclidean structures. The 2-d perspective allows for very simple
estimation of the relative energies of various curved graphitic frameworks, in
good agreement with more complex calculations [82], and equivalent to the
bending energy formalism used to model the relative stability of liquid
crystalline mesophases.

Cell-structures and hyperbolic surfaces

A brief discussion of the relation between 3-periodic hyperbolic surfaces and


(closed) cell structures is relevant to this volume. The focus of this paper has
been on embedded sponges, while cell structures are clearly not embedded. If
the condition of embeddedness is abandoned, 3-periodic minimal surfaces can
self-intersect such that crystalline arrays of discrete cells result, bounded by
finite portions of minimal surfaces. These "saddle polyhedra" - containing
straight edges and hyperbolic faces - are space-filling. The earliest published
example of such a structure is the TPMS found by Neovius to be adjoint to the
embedded Neovius surface [83].
Balanced ETPMS containing closed circuits of straight lines alone generate
families of space-filling saddle polyhedra. These polyhedra are the result of two
intergrown complementary ETPMS which share the same subset of straight
lines. In general, the polyhedra contain two types of faces, each characteristic of
one of the underlying ETPMS [3, 51, 84]. These can be readily generated by
successive rotation of the linear circuit: in some cases a single (unary space-
466
filling) polyhedron results (e.g. fig. 22), more generally more than one
polyhedron results [3].

Figure 22: Saddle polyhedron, consisting entirely of minimal surface faces (cf. fig. 14), derived
from a linear net on the D surface. The polyhedron contains the faces of two intergrown D
surfaces (shaded differently), so that the tunnels of one surface pass through the faces of the other.

These cells are far from optimal solutions to the celebrated Kelvin problem,
despite the fact that their faces are minimal surfaces. (For example, the saddle
polyhedron in fig. 22 has normalised isoperimetric quotient:
361tV2 =!(1...)-2 "'0.25
A3
4 4h
which is far below the recent best estimate, derived from the melanophlogite
network [85]).

Final comments
Given that generic sponges - derived from tilings of the hyperbolic plane - can
be embedded in arbitrary dimensional spaces, and form embedded bicontinuous
morphologies, closed discontinuous foams, 1-d honeycomb-like channels (as
well as hybrids of all of these), the generality of 2-d hyperbolic geometry, and
its embeddings in higher dimensional spaces, is clear. Much remains unknown,
however. In the shorter term, two important broad areas demand further
research. First, greater understanding of the geometric variety and homogeneity
anisotropic and intersecting sponges, and embedded sponges of nonzero mean
curvature is needed. Second, the entropic content of surfaces must be compared
in order to model melt phenomena in sponges.

Acknowledgments

I am very grateful to Drs. Rivier and Sadoc for organising a fascinating meeting
in Corsica as well as waiting patiently for this paper. I am also very grateful to
Andrew Fogden for calculations in fig. 12 and shared throughts.
467

References

I. L.E. Scriven, Nature 1976, 263: 123-125.


2. W. Fischer and E. Koch, Phil. Trans. Roy. Soc. Lond. A 1996, 354: 2105-2142.
3. S.T. Hyde and S. Andersson, Z. Kristallogr. 1984, 168: 221-254.
4. S. Andersson, S.T. Hyde, and J.-0. Bovin, Z. Kristallogr. 1985, 173: 97-99.
5. S. Andersson, Angew. Chern. Int. Ed. Engl. 1983, 22: 69-81.
6a. S. Andersson, S.T. Hyde and H.-G.von Schnering, Z. Kristallogr.1984, 168: 1-17.
6b S. T. Hyde and S. Andersson, ibid., 1986, 174: 225-236.
7. M. O'Keeffe and B.G. Hyde, "Crystal Structures. 1. Patterns and Symmetry."
Washington: Mineralogical Society of America; 1996.
8. T. Jiang, A.J. Lough, G.A. Ozin, D. Young, and R.L. Bedard, Chern. Mater. 1995,
7: 245-248.
9. G. Stucky, A. Monnier, F. Schuth, Q. Huo, D. Margolese, D. Kumar,
M. Krishnamurty, P. Petroff, A. Firouzi, M. Jamicke and B. Chmelka,
Mol. Cryst. Liq. Cryst. 1994, 240: 187-200.
10. S. Inagaki, A. Koiwai, N. Suzuki, Y. Fukushima and K. Kuroda, Bull. Chern. Soc.
Japan 1996, 96: 1449.
11. S. Inagaki, Y. Fukushima and K. Kuroda, Chern. Comm. 1993, 8: 680-682.
12. Q. Huo, R. Leon, P. Petroff and G. Stucky, Science 1995, 268: 1324-1327.
13. C.-Y. Chen, S.-Q. Xiao and M.E. Davis, Microporous Materials 1995,4: 1-20.
14. J.S. Beck, C.T.-W. Chu, I.D. Johnson, C.T. Kresge, M.E. Leonowicz, W.J. Roth
and J.C. Vartuli, U.S. Patent 5108725 (1992).
15. S.A. Bagshaw, E. Prouzet and T.J. Pinnavaia, Science 1995, 269: 1242.
16. P. Behrens, Adv. Mater. 1993, 5: 127-132.
17. K. Larsson, Nature 1983,304: 664.
18. K. Larsson, J. Phys. Chern. 1989, 93: 7304-7314.
19. W. Longley and T.J. Mackintosh, Nature 1983,303: 612-614.
20. V. Luzzati and P.A. Spegt, Nature 1967, 215: 701-704.
21. P. Strom and D.M. Anderson, Langmuir 1992, 8: 691-709.
22. J. Seddon and R. Templer, Phil. Trans. R. Soc. Lond. A 1993, 344: 377-401.
23. S.T. Hyde, S. Andersson, B. Ericsson and K. Larsson, Z. Kristallogr. 1984, 168:
213-219.
24. D.M. Anderson and E.L. Thomas, Macromolecules 1988, 21: 3221-3230.
25. D.A. Hajduk, P.E. Harper, S.M. Gruner, C.C. Honeker, G. Kim, E.L. Thomas and
L.J. Fetters, Macromolecules 1994, 27: 4063.
26. H. Hasegawa, H. Tanaka, K. Yamasaki and T. Hashimoto, Macromolecules 1987,
20: 1651-1662.
27. H. Hasegawa, Y. Nishikawa, S. Koizumi, T. Hashimoto and S. Hyde, in Proc.
Conference on Advanced Materials (JUMRS-ICAM-93). (1993): Elsevier S. P. &
MRS Japan.
28. E.L. Thomas, D.M. Anderson, C.S. Henkee and D. Hoffman, Nature 1988, 334:
598-601.
468

29. Y.J. Bouligand, J. Phys. (Fr.) Colloque 1990, C7: 35.


30. T. Landh, FEES Lett. 1995, 369(1): 13-17.
31. S. Hyde, B. Ninham, S. Andersson, Z. Blum, T. Landh, K. Larsson and S. Lidin,
The Language of Shape" Amsterdam: Elsevier; 1997.
32. G. Donnay and D.L. Pawson, Science 1969, 166, 1147; H.-U.Nissen, Science
1969, 166, 1150.
33. G. Porte, Curr. Rev. Colloid Sci. 1996, 1: 345-349.
34. G. Porte, R. Gomati, 0. El Haitamy, J. Appell and J. Marignan, J. Phys. Chern.
1986, 90: 5746-5751.
35. K. Fontell, ACS Symposium Series, 1975, (9): 270-277.
36. H. Jinnai, T. Koga, Y. Nishikawa, T. Hashimoto and S.T. Hyde, Phys. Rev. Lett.,
1997, 78, 2248-2251..
37. A.P. Roberts, D. Bentz and M.A. Knackstedt, Proc. Soc. Petrol. Eng. 37024,
(SPE Conference, Adelaide, 1996), 551-559.
38. M. Knackstedt, private communication
39. R. Corkery, T. Senden and S.T. Hyde, in preparation.
40. D. Hilbert and S. Cohn-Vossen, "Geometry and the Imagination" New York:
Chelsea Publishers; 1952.
41. D.M. Anderson, Studies in the microstructure ofmicroemulsions (1986), Ph.D.
thesis, University of Minnesota.
42. D. Hilbert, Trans. Am. Math. Soc. 1901, 87-99.
43. J.-F. Sadoc and J. Charvolin, Acta Cryst. 1989, A45: 10-20.
44. S.H. Tolbert, T.E. Schaffer, J. Feng, P.K. Hansma and G.D. Stucky, Chern.
Mater. 1997, 9: 1962-1967.
45. J.C.C. Nitsche, "Lectures on Minimal Surfaces", Vol. 1. Cambridge: Cambridge
University Press; 1989.
46. D. Hoffman, Math. Intell. 1987, 9: 8-21.
47. E. Koch and W. Fischer, Acta Cryst. 1990, A46: 33-40.
48. W. Fischer and E. Koch, Z. Kristallogr. 1996, 211: 1-3.
49. W. Fischer and E. Koch, Acta Cryst. 1989, A45: 726-732.
50. A.-M. Levelut, E. Hallouin, D. Bennemann, G. Heppke and D. Lotzsch, J. Phys. II
(France) 1997,7:981-1000.
51. A.H. Schoen, "Infinite periodic minimal surfaces without self-intersections",
(1970), N.A.S.A. Technical Report #D5541.
52. A. Schoen, in "Geometric Analysis and Computer Graphics", R. Finn, D.A.
Hoffman and P. Concus, eds. (1991), Springer-Verlag: New York. p. 147-158.
53. R. Kusner, in [52]. p. 103-108.
54. A.F. Wells, "Three-Dimensional Nets and Polyhedra" (1977). John Wiley and Sons:
New York. p. 160.
55. H.-G. von Schnering and R. Nesper, Angew. Chern. Int. Ed. Engl. 1987, 26:
1059- 1080.
56. S.T. Hyde, Z. Kristallogr. 1989, 187: 165-185.
469

57. S. Lidin, J. Phys. (France) 1988, 49: 421-427.


58. S.-S. Chern, in "Differential geometry and complex analysis", I.C. &. H.M.
Farkas, eds. (1985), Springer Verlag: Berlin. p. 155-163.
59. A. Ramsay and R.D. Richtmyer, Introduction to Hyperbolic Geometry", Springer
Verlag New York, Inc., 1995.
60. A. Fogden and S.T. Hyde, in preparation.
61. S.T. Hyde, Curr. Opinion Solid State Mat. Sci. 1996, 1: 653-662.
62. H.S.M. Coxeter, "Twelve Geometric Essays", Carbondale: Southern Illinois
University Press; 1968.
63. H.S.M. Coxeter, Proc. Lond. Math. Soc.,l937, Ser. 2, 43: 33-62.
64. W. Fischer, Z. Kristallogr. 1974, 140: 50-74.
65. P.A. Firby and C.F .. Gardiner, "Surface Topology", 2nd. ed., Chichester: Ellis
Horwood Limited; 1991.
66. M. O'Keeffe, Z. Kristallogr. 1991, 196: 21-37.
67. S. Andersson, S.T. Hyde, K. Larsson and S. Lidin, Che1~>. Rev. 1988, 88: 221-42.
68. G.O. Brunner and W.M. Meier, Nature I989, 337: 146-147.
69. L. Stixrude and M.S.T. Bukowinski, Am. Mineral. 1990, 75: 1159-1169.
70. S.T. Hyde, Acta Cryst. 1993, A50: 753-759.
71. S.T. Hyde, Z. Blum, and B.W. Ninham, Acta Cryst.1993, A49: 586-589.
72. A. Fogden and M. Jacob, Z. Kristallogr. 1995, 210: 398-406.
73. A.A. Barbosa and A.J. Palangana, Phys. Rev. E 1997, 56(2): 2295.
74. A.L. Mackay and J. Klinowski, Compu. & Math. Appl. 1986, 12B: 803-824.
75. A.L. Mackay and H. Terrones, Nature 1991,352: 762.
76. A.L. Mackay and H. Terrones, Phil. Trans. R. Soc. Lond. A 1993, 343: 113-127.
77. T. Lenosky, X. Gonze, M. Teter and V. Elser, Nature 1992, 355: 333-335.
78. D. Vanderbilt and J. Tersoff, Phys. Rev. Lett. 1991, 68: 511-513.
79. J. Gibson, M. Kolohan and H. Riley, J. Chern. Soc. 1946: 456-461.
80. G.B. Adams, O.F. Sankey, J.B. Page, M. O'Keeffe and D.A. Drabold, Science
1992, 256: 1792-1795.
81. M. O'Keeffe, G.B. Adams and O.F. Sankey, Phys. Rev. Lett. 1992, 68: 2325-8.
82. S.T. Hyde and M. O'Keeffe, Phil. Trans. R. Soc. Lond. A 1996, 354: 1999-2008.
83. E.R. Neovius, Bestimmung Zweier Speciellen Periodische MinimaL Flaechen"
Helsinki: Frenckel & Son; 1883.
84. P. Pearce, "Structure In Nature is a Strategy for Design", Cambridge, Mass.: MIT
Press; 1978.
85. D. Weaire, ed. "The Kelvin Problem", London: Taylor and Francis; 1996.
470
DEFORMATIONS OF PERIODIC MINIMAL SURFACES

C. OGUEY
LPTM, Universite de Cergy-Pontoise
2 av. A. Chauvin
F - 95302 Cergy-Pontoise

Abstract. Periodic minimal surfaces model a large variety of mesophases and


crystals of films in ternary liquid systems typically composed of water, soap and oil.
First I describe the Bonnet transformation with emphasis on the crystallographic
aspects which are specific to periodic minimal surfaces. Then I consider other
transformations which are non isometric but preserve minimality.

1. Introduction

Mesophases, or sponge phases, are liquid crystals where the ordering (lamel-
lar, hexagonal, cubic etc.) is due to the presence, in the core of the phase, of
a film made of either water (direct) or oil (reverse) sandwiched between two
layers of surfactant molecules with appropriate orientation. The space left
is filled, in majority, by the third component [1, 3, 21]. In binary systems
with similar structures, the film is just a bilayer of amphiphiles. I leave aside
the micellar phases which generally occur at extreme concentrations (high
or low water proportion) (see J. Seddon in this volume). At intermediate
concentrations of the mixture, the space is partitioned by a liquid film. For
the morphology of those films, the standard models are minimal surfaces.
This is a feature common with bubbles or foams.
There are however differences. 1) The length scale is typically smaller;
in the cubic phases, for example, the lattice parameter is of the order of
a few nanometers. 2) The relaxation times are reasonably short allowing
thermodynamic equilibrium to set in. In particular, the pressure is constant.
3) Except for the micellar phases, the structure is not a packing of closed
cells. Depending on the values of temperature and concentrations, the phase
may consist of tubes (hexagonal), lamellae or intertwined labyrinths as in
the cubic phases. 4) In the cubic phases, or their variants with similar
471
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 471-480.
© 1999 Kluwer Academic Publishers.
472

Figure 1. Example of a triply periodic minimal surface: the P surface.

geometries but lower symmetries, there are no singularities, no edges, no


Plateau borders etc. (fig. 1). Those are the ones which I will mostly consider,
at least when looking at non local features like topological changes.
When the pressure is equal on both sides, soap bubbles take the shape of
minimal surfaces. Locally minimizing the area with fixed boundaries (such
as wire frames) yields the equation H = 0 where H = (kt + k2)/2 is the
mean curvature, arithmetic mean of the principal curvatures k1 and k2. In
fact, after Laplace, the mean curvature H is precisely equal to the pressure
difference in situations close to equilibrium. So the surface energy is simply

E = uA = u !M dA, (1)

where M is the surface embedded in space and u is surface tension.


At a lower scale, when we want to describe the interfacial films in the
mesophases, Helfrich's form for the energy is a good approximation [7] :

(2)

The factors K, K, are elastic constants; LR denotes longer range interactions


such as those responsible for the spacing between lamellae [14] .
In mathematics, surfaces are often viewed as abstract objects of dimen-
sion 2 with intrinsic metrics and topology [22, 23]. This allows to describe,
with more precision, surfaces such as the Klein bottle or the projective
plane which cannot be embedded in our familiar Euclidean space R 3 . The
mean curvature H, however, hence minimality, is not an intrinsic prop-
erty; it relates to the way the surface is immersed in the Euclidean space
E = R 3 . Therefore one often speaks of minimal immersions or embedding
473

rather than minimal surfaces. On the other hand, the Gaussian curvature,
equal to the product of the principal curvatures K = k1 * k2, can be shown
to be intrinsic. The mathematical literature on minimal surfaces is abun-
dant. I can only mention textbooks [2, 11, 19, 17] and a selection of articles
[13, 9, 10, 15].
For modeling physical systems, the surface has to be taken as it is
embedded in space, this is clear. It is of physical relevance, nevertheless,
to know of what type the most important terms in the energy are. By
important I mean the ones determining the shape, the flexibility and low
excitations of the film. I will not solve that problem here but get a few
hints.

2. Immersion in the complex space C 3


The equation H = 0 is a non-linear second order partial differential equa-
tion for the coordinates x = (x 1, x2, x3) which is solved by the Weierstrass-
Enneper formula (WE)

un
z--+ x(z) Re (a~(z)), (3)

e(z) ~ 1z
zo
1- w2 )
( i(1 + w 2 )
2w
R(w) dw. (4)

Re denotes the real part. This formula gives the Cartesian coordinates of a
point of the surface x(z) as a function of the complex parameter z = u + iv
(equivalent to two real ones, as necessary for two dimensional objects). It
involves an analytic function R (so that R dw is a meromorphic differen-
tial). The integration runs along a path from zo (fixed) to z (running) in
the complex plane C. By Cauchy's theorem, the value of the integral is
independent of the path for all paths homotopic in the domain of definition
of R, which is a Riemann surface.
This formula is largely explained in the textbooks. Let me just mention
two points, setting "1 = ~ 1 = {)z~:
="'f
- "1 · "1 + TJ~ + "1~ = 0. That identity reflects the fact that the
coordinates z = u + iv are conformal (z--+ x(z) preserves angles).
- {)z*"l = 0. This is the minimality condition which turns out to be
equivalent to the Cauchy-Riemann condition for analyticity.
The module of the complex number a fixes the length-scale whereas
its phase provides, upon variation, different minimal surfaces. Changing
the phase (} = arg(a) in (3) is known as a Bonnet transformation. Two
surfaces related by such a transformation are in general different (they
cannot be brought to coincidence by any Euclidean motion) but they are
locally isometric as surfaces.
474
When the parameter z is restricted to vary within a suitable domain,
free from singularities of R, the WE formula provides an analytic 1-1 map
between the domain and a patch of the surface M. However such a re-
striction is not necessary. By analytic continuation, the integration can be
carried along any path avoiding the singularities (poles and branch points
in the cases under study). Of course non homotopic paths to some value
z of the parameter may give different values for the coordinates x( z), and
so, different points in R 3 . The vector joining two such points is a period,
in mathematical parlance, that is, a vector of the translation lattice of
the surface. This is the way periodic surfaces can be expressed in a single
formula.
The most well known triply periodic surfaces are Schwarz' P (primitive),
D (diamond or face centered F) and G (gyroid, body centered) [18, 20].
They all three have cubic symmetry and their WE differential is
R(w) = (1 + 14w4 + w8 )- 112 • (5)
Albeit different regarding their geometrical aspect (different space groups,
lattice parameters etc .. ), those three surfaces only differ, in formula (3-4),
by the Bonnet angle() which has value 0°, 90° and 51, 985 .. 0 respectively.
To study the Bonnet transformation, the most natural thing to do is to
investigate the surface in C 3 which is a lift (covering) of all the (three in
this case) members of a Bonnet family. What the WE formula (4) provides
is, first of all, an immersion of the surface into C 3 ~ R 6 : z -t ~(z). The
real surface is then obtained by linear orthogonal projection down to the
physical space: x = Re(a~). Note that, locally, this projection produces no
singularity on the surface since the metric tensors of the complex surface ~
and of the real projected one x are related by a constant factor 1/2.
The translation symmetries of the real surface are projections of the
translations of the complex one. Those translations are obtained by contour
integration along closed loops; they are thus tightly related to the topology
of the underlying Riemann surface. For example, the three surfaces P, D ,
G of genus g = 3 per unit cell, have a common covering, which I will call
PDG, and which is a periodic analytic surface embedded in C 3 . On the
Riemann surface R( w)- 2 = 1 + 14 w4 + w 8 the map (4) has six independent
(over the reals) complex periods generating a 6 dimensional lattice A, the
symmetry lattice of the complex surface PDG. A possible set of generators
for A is (see [18] for details)
r -is is 0 r- is -r -.is )
(t~, t2, ta, r~, r2, ra) = ( is r -is -r- is 0 r- ~s ,
-is is r r - is -r - is 0
where r = 2.156 .. and s = 1.686 .. are the cubic parameters of the P and F
surfaces respectively. The global scaling factor is, of course, arbitrary; here
475

Figure 2. The Bonnet transformation acting on a fundamental domain of the surface.


The angle has successive values 0 (P), 30, 52 (G), 70, 90° (D).

it is set to lal = 1. The ratio of the two parameters rands is, however, not
arbitrary because it has to fulfill the isometry condition : patches mapped
onto each other by the Bonnet transformation must have the same area.
This fixes r / s to be 1. 279 ...

3. ·The Bonnet transformation

As we have just seen, the Bonnet transformation is an isometry on bounded


patches (distances and angles within the surface are preserved). Now, in
mesophases, the interface or the film is believed to form an infinite bicon-
tinuous surface. So it seems natural to look at the transformation as acting
globally, or at least on larger patches.
Consider the example of the P surface which is typical in these respects.
Starting from the fundamental piece shown on fig. 3 (left), corresponding
to (} = 0, let us raise (} to a tiny non zero value. Then two requirements
476

Figure 3. Fundamental piece of the P surface at Bonnet angle () = 0 (left), and at () :::: 3°
(center and right) with cuts or self-intersections.

are possible. Either we roughly fix the total area of that piece (this would
correspond to conservation of the film material, which does not need to be
exactly conserved because of the liquid reservoir) and the piece tears itself
(fig. 3 center). Or we impose bicontinuity (no boundaries nor tearing), but
then the surface displays sequences of helicoidal convolutions with many
self intersections, the whole surface ending in densely filling space like a
bended millefeuille (fig. 3 right) . None of those features sounds physically
plausible.
The Bonnet transformation is, however, a suitable way to design other
examples of minimal surfaces, provided it is kept in mind that physically
reasonable candidates occur only for a finite subset of values of the Bonnet
angle: rational (s / r) tan 0 with small numerator and denominator.
Another issue of the Bonnet transformation is that it indicates the rele-
vance of the LR term in (2). Indeed an energy function depending only on
the local metric, as the first term in (2), does not depend on B, therefore it
has the same value on all the members of a Bonnet family. So if, at some
thermodynamic coordinates, the system adopts a definite structure with-
out degeneracy, choosing one (for example the G) rather than another (the
P or D) Bonnet equivalent conformation, it is necessarily due to non-local
interactions and, eventually, entropic contributions.

4. Non isometric transformations


Probably closer to real physical processes undergone by the film under
stress or low energy excitations, the continuous non isometric deformations
form a large class, even with suitable constraints (steric barriers, bounds on
curvatures etc.). A thermodynamic treatment would then be appropriate
[16].
477

Figure 4- Continuous rhombohedral distortion of the P (left) and the D (right) surfaces.

Figure 5. Movement of the branch points corresponding to fig. 4. In this representation,


zero and infinity are branch points and stay fixed. The other ones move radially.

Let me focus on one subclass which will be of interest for systems with
large elastic constant Kin (2) (in the elastic as well as non-linear regime).
For each minimal surface, there is a set of continuous deformations which
are not isometries but which preserve minimality: the deformed surfaces
still satisfy H = 0 everywhere. A number of examples have been studied
by the Australo-Swedish group [4, 5, 6, 12]. The salient features of these
deformations are, first, that they are bounded in some directions, reaching
instability beyond a certain threshold [8], second, that, given a minimal
surface with finite topology (finite genus for non periodic MS in R 3 , finite
genus per unit cell for periodic ones), the set of continuous deformations
preserving H = 0 is finite dimensional. Considering such a family of minimal
surfaces or immersions, it may happen that the Bonnet conjugates fall in
the same family. In that case, the family is said to be self adjoint.
A well known example is the rPD family. Uniaxial compression-dilation
along a 3-fold axis induces a continuous transition from the P surface to
the D and vice versa. The intermediate surfaces are all minimal with lower
symmetries (fig. 4). The G is not reached this way because the path followed
478

here is different from the Bonnet route even though the end points (P and
D) are the same.
To study those deformations, or classes, of minimal surfaces, the WE
representation is well suited. The parameters controlling the morphological
changes are those entering the WE function (13, 5]. For example, triply
periodic MS with genus 3 per unit cell form a family whose dimension
is, a priori, not larger than 18. They are characterized by a WE form
R(w) = (ao +a1w+ ... +asw8 )- 112 involving a polynomial of degree 8. The
parameters can be taken to be the complex coefficients of that polynomial.
Among these degrees of freedom, 4 are trivial (3 corresponding to global
rotations, 1 to change of length scale). A number of others will lead to non
embedded surfaces (with self-intersections and without minimal bound on
the distance between sheets). In order to get a regular embedded surface
with a true 3D lattice as part of the symmetry group, 3 basis vectors need
to cancel in the projection from 6D (where the complex PDG surface is).
Algebraically, this amounts to 9 scalar constraints leaving us with 5 de-
grees of freedom for the set of genuine deformations of regularly embedded,
triply periodic, minimal surfaces with g = 3. (This is assuming that the
9 constraints are independent; otherwise, this set might be larger). That
number 5 is also the dimension of the space of linear volume preserving
deformations in elasticity (symmetric 3 x 3 tensors with vanishing trace).
To my knowledge, a systematic investigation of the deformed MS is still
lacking, even for the case g = 3. Meeks [13] has a set V of dim. 5 which
includes many examples of interest but it is not complete because it does
not contain G. This is still under investigation.
Acknowledgments. I wish to thankS. Hyde, N. Rivier, J-F. Sadoc and
J. Seddon for discussions. Special thanks to the organizers of the school.

References
1. Andersson, S., Hyde, S.T., Larsson, K., Lidin, S. (1988) Minimal surfaces and struc-
tures: from inorganic and metal crystals to cell membranes and biopolymers Chem.
Rev. 88, 221-242.
2. Dierkes, U., Hildebrandt, S., Kuster, A., Wohlrab, 0. (1991) Minimal Surfaces I &
II Springer, Grundl. der math. Wiss. vol. 295 & 296.
3. Dubois-Violette, E., Pansu, B. eds (1990) Internationnal Workshop on Geometry
and Interfaces , J. Phys. Colloq. I+ance 51 C7.
4. Fogden, A. (1992) A systematic method for parametrising periodic minimal surfaces:
the F-RD surface J. Phys. I Jirance 2 233-239.
5. Fogden, A., Hyde, S. T. (1992) Parametrization of triply periodic minimal surfaces
Acta Cryst. A 48 442-451 and 575-591.
6. Fogden, A., Heaberlein, M., Lidin, S. (1992) Generalization of the gyroid surface J.
Phys. I Jirance 3 2371-2385.
7. Helfrich, W. (1973) Z. Naturforsh. 28C
8. Hoffman, D. A. Some basic facts, old and new, about triply periodic embedded
minimal surfaces , in [3].
479

9. Hoffman, D. A., Meeks, W. H. III (1990) Minimal surfaces based on the catenoid,
American Math. Monthly 97.
10. Hoffman, D. A., Osserman, R. (1985) The Gauss map of surfaces in R 3 and R 4
Proc. Land. Math. Soc. 50 27-56.
11. Lawson, H. B. (1971) Lectures on Minimal Submanifolds Berkeley, Publish or Perish.
12. Lidin, S., Hyde, S. T . (1987) A construction algorithm for minimal surfaces J. Phys.
France 48 1585-1590.
13. Meeks, W. H. III (1990) The theory of triply periodic minimal surfaces, Indiana U.
Math. J. 39 877-935.
14. Mosseri, R., Sadoc, J. F. (1997) Frustmtion Geometrique Alea Saclay, Eyrolles.
English version: (1997) Geometric Frustmtion Cambridge.
15. Nagano, T., Smyth, B. (1980) Periodic minimal surfaces and Weyl groups, Acta
Math. 145 1-27.
16. Nelson, D. R. et al. edts (1989) Mechanics of Membranes and Surfaces World Sci-
entific.
17. Nitsche, J. C. (1989) Lectures on Minimal Surfaces Cambridge.
18. Oguey, C., Sadoc, J. F. (1993) Crystallographic aspects of the Bonnet transforma-
tion for periodic minimal surfaces, J. Phys. I France 3 839-854.
19. Osserman, R. (1986) A Survey of Minimal Surfaces Dover.
20. Sadoc, J. F. , Charvolin, J. (1989) Infinite periodic minimal surfaces and their crys-
tallography in the hyperbolic plane Acta Cryst. A 45 10-20.
21. Seddon, J. (1996) Lyotropic phase behaviour of biological amphiphiles, Ber. Busen-
ges. Phys. Chem. 100 380-393.
22. Spivak, M. (1979) A Comprehensive Introduction to Differential Geometry Houston,
Publish or perish.
23. Willmore, T. J. (1993) Riemanian Geometry Oxford UP.
480
APERIODIC HIERARCHICAL TILINGS

CHAIM GOODMAN-STRAUSS
Unive1'sity of A1'kansas
Fayetteville, A1'kansas USA 72701

Abstract. A substitution tiling is a certain globally defined hierarchical


structure in a geometric space. In [6] we show that for any substitution tiling
in En, n > 1, subject to relatively mild conditions, one can construct local
rules that force the desired global structure to emerge. As an immediate
corollary, infinite collections of forced aperiodic tilings are constructed. Here
we give an expository account of the construction. In particular, we discuss
the use of hierarchical, algorithmic, geometrically sensitive coordinates-
"addresses", developed further in [9].

Figure 1. A substitution tiling

1. Introduction

Global structure arises constantly from local properties, in nature and in


abstract: trees, people and crystals are good examples of the former; groups,
automata and curvature provide good examples of the latter.
481
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 481-496.
© 1999 Kluwer Academic Publishers.
482

Figure 2. A matching rule tiling

Substitution rules generate global hierarchical structure in the plane,


such as spaces of aperiodic heirarchical tilings. For example, in figure 1, we
first see on the left a small L-tile and rules for "inflating and subdividing"
the tile. As this process is iterated, larger and larger regions of the plane
are tiled with L-tiles. On the right of figure 1, these L-tiles are arranged
hierarchically into larger and larger L-shaped "supertiles"- the images
of inflated and subdivided L-tiles. The thicker lines have been added to
emphasize this hierarchy.
We can thus define a global structure- the "substitution tiling" induced
by the inflation and division of the tiles. If a set of prototiles admits no tiling
that is invariant under a translation (or more generally, no tiling invariant
under any infinite cyclic group of isometries), the prototiles are aperiodic.
For example, the infinite L-tiling suggested at right on figure 1 cannot
be translated since any translation will place some giant L-shaped supertile
onto itself, and each tile is in only one L-shaped supertile of each size. But
the unmarked L-tiles by themselves are not aperiodic since they can tile
rectangles and hence can form periodic tilings.

By themselves L-tiles are not forced to tile aperiodically. They them-


selves cannot recreate the hierarchical structure of figure 1.
On the other hand, the tiles in figure 2 must make tilings in which
the original hierarchical L-tiling is clearly visible, when the simple rule of
matching black to black and white to white is followed [7] (The proof of
this is by different techniques than that of the theorem below) . That is,
local conditions force L-shaped supertiles of arbitrary size!
As the hierarchical structure of figure 1 is precisely reproduced, we
say the original substitution tiling has been "enforced" by these marked
"matching rule tiles". The tiles themselves are "aperiodic" and "hierarchi-
483

cal" as they can only produce aperiodic hierarchical tilings. Precise defini-
tions of these all of these terms are in [6] and elsewhere.

Beginning in 1964 with R. Berge's insightful paper [1], a number of


examples of aperiodic hierarchical tilings have appeared. S. Mozes produced
a method for producing matching rule tiles enforcing a certain class of
heirarchical tiling [14], but no truly general method had emerged until [6].
This theorem includes all examples known to the author:

Theorem [6] Every substitution tiling of En, n > 1, can be enforced with
finite matching rules, subject to a very mild condition:

the tiles are required to admit a set of "her·editar·y edges" such that the
substitution tiling is "sibling-edge-to-edge ". 1

The proof is somewhat technical but rests on simple principles. Here


we attempt only to discuss a few general principles that may have wider
application.
The primary structures are addresses generated by the substitution pro-
cess; these addresses can serve tilings, configurations, tiles and points and
encode a variety of information. In essence addresses are the geometric ex-
pression of regular languages, and form an extension of symmetry groups
to semigroups [9].
In a sense, addresses are a geometric lexicon of the regular language
given by the "substitution graph". The task is to break the space of ad-
dresses into fundamental domains and give rules for correctly piecing them
together. These addresses are hung upon skeletons of edges in the tiling,
which effectively parse the addresses into digits and piece digits into ad-
dresses.

Importantly, these are new combinatorial, geometric structures that


may be useful in describing natural phenomena.

l.l. APERIODIC TILINGS, HIERARCHICAL TILINGS

The possibility of an aperiodic set of prototiles, in the plane or elsewhere,


was essentially ignored until H. Wang and R. Berger began investigating
connections between tilings and undecidability. In the early 1960's Wang
1 The condition really is mild: we must be able to choose vertices and edges for our

tiles such that the vertices/edges of the parent coincide with those of the children and
such that the vertices/edges of sibling supertiles coincide.
484
encoded arbitrary Turing machines as tilings of the plane. However he re-
quired a somewhat awkward restriction- the placement of a "seed" tile to
begin the computation. In 1964 Berger removed this restriction by con-
structing an underlying hierarchical structure; this structure provided ar-
bitrarily large, enclosed domains for the run of the machine [1].
Berger thus could answer that the "Domino Question" is undecidable;
that is, there is no an algorithm to decide whether any given set of prototiles
admit a tiling. Berger's tiles admitted a tiling if and only if the underlying
Turing machine did not halt: If every set of prototiles admitting some tiling
admits a periodic tiling, one can produce an algorithm to check whether
any given set of prototiles actually admits a tiling. Thus Berger produced
an unexpected corollary- there are sets of prototiles that do tile the plane,
but cannot do so periodically! He gave such a set, of some 20 ,000 tiles.
By 1971 R. Robinson had produced a much smaller set of six prototiles
and streamlined Berger's general result as well in an especially lovely and
readable paper [20].

:<. . ~ -·...1L "'.;: ~~cJ~~-ea~


=:
J
,'-~ill~ .... ,
jL - ) ~(.::·~···········
\, ')
I

~he •.;'I: ./,


............

=tf . . ;r n ·r--./~~=
... .....
) ) I
~\.....

~ -\
)
j
~"· )
·+-f-'r-
LW. -- , L h.L
~'-
1
_j ___
~+-ft=
\ ...
f../ J1n I
....
:_+-
-,
' /-
i
~,L
)
~·)f(r ,L::J/
1 {
·c. r--,L v

-- :---t""""
\
-,...... .--
u )~~ Jr u'
u wr
;>
-- :--...,.:;_
1 l
"'( --.,l {
,L
'·· j
_Jl

/rr
. L~r-:

'-'-) ~./ l. J I_) (: :._/ \...!......i

Figure 3. An ape1iodic collection of square lattices and Robinson's tiling

Robinson's set of prototiles enforced aperiodicity by ensuring a specific


485

structure was formed: an infinite hierarchy of similar square lattices in the


plane (figure :3). The collection as a whole is aperiodic, since no translation
can leave every scaled lattice invariant.
R. Penrose found his celebrated tiles in 1972 [15]. The Penrose tiles have
become canonical examples of aperiodic tiles, for they are both aperiodic
hierarchical tiles but also are qua.<>iperiodic (see below).

Figure 4. The penrose rhombs

Many other examples of aperiodic heirarchical tilings have been found


since. In particular R. Amman, J. Socolar, and L. Danzer have constructed
many elegant examples [10],[24],[4). C. Radin ha.<> produced matching rules
for the Conway pinwheel [17]. S. Mozes gave a general method for construct-
ing specific cla.<>s of aperiodic hierarchical tilings [14], and a.<> described, a
very general method is given in [6).

Only three other methods of constructing aperiodic tiles have emerged:

N.G. De Bruijn pointed the way to the second interpretation in 1981:


aperiodic tilings could arise a.<> projections of slices through a higher dimen-
sional lattice [2]. That is, just a.<; the penrose tiles appear· to be projections
of cubes, they indeed are, in a nice manner: the rhombs of the tiling are
projections of the squares lying in a certain slab through a five dimensional
cubic lattice. This "slice and project" method of constructing aperiodicity
486
ha.'3 been very much studied. Recently Le T.T.Q. has given matching rules
for a wide cla.'3s of such tilings [12).

In 1988, Peter Schmitt gave a third cla.<>s of aperiodic tiling- tilings


arising from a single proto tile in Ea. These tilings are not isotropic and thus
Schmitt introduced a new symmetry of space. This example wa.'3 extended
upon by Conway and Danzer [:3).

In 1996 a fourth, completely new method of constructing aperiodic


tilings ha.'3 emerged: J. Kari has given a construction returning all the way
to Wang for inspiration. Essentially he avoids the need for a seed tile by
running infinite computations on infinite sequences [11). Kari and K. Culik
are beginning to extend this exciting work.

1.2. A VARIETY OF SUBSTITUTION TILINGS

We pause for a few examples of substitution tilings- heirarchical arrange-


ments of tiles in the plane. A more lengthy discussion of certain technical
issues can be found in Appendix A of [6]; a formal discussion is in Section
1.2 of [6].
The substitution tiling on the left of figure 5 is periodic and admits an
infinity of hierarchies. Generically, however, if we use marked squares the
resulting substitution tiling admits only one hierarchy! (That is, the tiling
on the right ha.'3 unique decomposition. Solmyak has proved that aperiodic
hierarchical tilings for which a certain natural condition is satisfied have
unique decomposition.)

. ......... ...... ...... r .,. - .. , . ~ ~· .., .,. ~ .,.""

D i' l '
' \,1,1
I
I\
'l_l_~"j
l'i
...,

EE
' \ 1111 I~ ~

Figure 5. Two quite distinct substitution tilings

In figure 6 are a number of examples, clockwise from upper-left: the


"Sphinx", discussed at length in [Godr] with matching rules provided by
E.A. Robinson [RobE]; the "Pinwheel", found by Conway with matching
487
rules provided by Radin [Rad]; a "Dimer" tiling, with matching rules found
by E.A. Robinson; a triangle tiling found by the author; a triangle tiling
found by Danzer and featured in Appendix B of [6]; and the "Half-hex" ,
with matching rules provided by Socolar [Soc]. These examples all general-
ize in a variety of ways.

IR'!
~
o-
00
-
Lb
~\L
QL -
~
1~
Figure 6. T he Sphinx, the Pinwheel, the Half-Hex and the Dimer, and t wo triangle
t ilings

Our last example, however points the way towards a far more general
setting. Here we are no longer restricted to Euclidean congruences. We
still might say t hat all the quadralaterals of a given level in the hierarchy
are congruent, and more or less leave the geometry at that. In this way,
488

substitution structures- addresses- may gain great utility, with widespread


applications.

Figure 7. A substitution tiling?

2. Addresses

The following is developed more fully in [9] and [6]. The reader is asked
to consider the useful properties of graph paper, or a ruler, a.'3 illustrated
at the top of figure 8. First, each point on the paper or ruler is located
to arbitrary precision through algorithmically produced strings of digits.
These strings of digits have a hierarchical structure.
Although points are identified by infinite strings of digits, these strings
are not necessarily unique. For example (in binary representation) .010 =
.OOI; nonetheless, there are simple algorithms that can identify when two
strings refer to the same point (indeed, we learn such an algorithm a.'3
children).
Finally, the geometry of the space is reliable with respect to the ad-
dressing given by the graph paper or ruler. A certain geometry is closely
bound to the particular coordinates imposed by the graph paper.
489

.,.. .1001010011 ...


/
I
I
f

l\

.0100 ..= .11 00 .. =


.0011 .. .1 011 ..
.100 ... = 1.00 .. =
.011 .. 0.11 ..

Figure 8. A ruler (base 2}

We will use the L-tiling to illustrate our addressing scheme. Consider


the following:
First, label each image of the L-tile in the substitution rule. There are
two games we might like to play: label points in an L-tile, and label L-tiles
in the tiling.
To be useful, this labeling must have a few properties: We would like
this labeling to be algorithmic, and algorithms to be available to tell us
when two labels describe adjacent L-tiles, or given a label what the labels
of adjacent L-tiles are. Similarly, we might allow points to labeled in more
than one way, but at least want an algorithm to determine when two labels
describe the same point; anda way to find, given some label, the other
labels describing the same point. Finally, we would like the addressing to
be sensitive to the geometry of the substitution scheme.

So begin with addresses for inflations of a single tile. We start with a


•, a decimal point. If our original tile is, in say, the inside corner of an
inflated L-tile, we label a• its position with respect to the larger tile. If this
larger tile is in, say, the outer corner of a twice inflated L-tile, the label
of the smallest tile with respect to this larger tile is bae. In this fashion
we can describe positions of tiles with respect to arbitrarily large L-shaped
su pertiles through a string Xn ••. x1 •, Xi E {a, b, c, d}.
In fact, these strings are exactly the words in the regular language given
by the "substitution graph" (a finite state automaton) in figure 9. (The
nodes of the graph correspond to the original tiles; the arrows to the images
of the tiles in the substitution. Addresses are directed paths in the graph.
490
substitution graph

Figure 9. Addresses

A more interesting example is m figure 7 of [6]). Hence these labels are


algorithmic.

Indeed we can describe positions in "infinitely large supertiles" ... x1 •.


These infinitely large supertiles might be tilings of the whole plane, but
could be a tiling of just a portion of the plane; again it is straightforward
to construct an automaton to test which case the address describes. For
example, it is clear that any infinite-to-the-left string .. .Xi···•, for which
there is some N with Xi = a for i > N, describes a tile in an infinitely large
supertile covering three-quarters of the plane, whereas a "generic" string
most likely corresponds to a tile in an infinitely large supertile th at covers
the entire plane.
Paradoxically, there are uncountably many distinct infinitely large su-
pertiles, even if we allow equivalence up to isometry: there are simply un-
countably many infinite-to-the-left-strings; moreover, two strings describe
equivalent (up to a isometry) infinitely large supertiles if and only if the
strings are equivalent to the left of some digit.

In [6] this labeling of hierarchies is precisely what we show is recreated


491

by matching rules.

Many interesting tricks can be performed (9] ; for example, how can one
detect that when strings describe adjacent L-tiles in some infinitely large
supertile? In figure 10 a portion of an automaton to carry out such a check
is illustrated. (A machine of this sort is a Mealy machine). Strings of digits
simply are paths that follow the arrows; infinite-to-the-left strings are thus
paths that have no beginning but do terminate. Two strings ... Xi···• and
···Yi···• describe adjacent L-tiles if there is a path with each arrow labeled
xdYi· The portion shown here illustrates the adjacency of the strings in
boldface in figure 9. The reader is encouraged to complete this finite state
automaton (there is one state in the automaton for each kind of edge that
appears as we inflate the original L-tile). 2

ala

(A)
·==·
A ~cA
• ._!C • ._.
dd

ft rk}'
dd

thus

~ ... bac6Ciillcbd
- ...baccdbbdbc

Figure 10. A Mealy machine describing adjacencies along the indicated edge

Similarly we can describe points in a tile as infinite-to-the right-strings


•xo ... , and of course points in infinitely large L-tiles as hi-infinite strings.
Again, points may be described by more than one string. Strikingly, the
same Mealy machine that descibes adjacency of tiles also describes when
strings label the same point, and can be used for identifying pairs of infinite
supertiles that can be matched to make a substitution tiling of the full
plane.

3. Keys

We now turn a technique for unambiguously encoding the hierarchy: local


and regional keys.
The goal here is to encode the labelings of all tiles in a substitution
tiling only using a finite amount of information in each tile. For now we
will assume that somehow finite amounts of information can be compared

2 Such an automaton can always be found up t o some condition on the substitution

tiling; the author believes that requiring vertex-to-vertex is more or less sharp.
492
between neighboring supertiles and between a supertile and its parent and
children.

The author imagines an infinite community piecing together a family


tree through the exchange of brief postcards.

We use a simple idea: certain tiles- keys- are to determine large amounts
of the tiling. Every tile is labeled with the last digit of its address: that is, its
intended position with respect to its parent. Keys, however, will be labeled
with one additional digit, giving the position of some ancestral supertile
with respect to its parent.
We choose all the tiles labeled, say, a to be local keys and all the tiles
labeled, say c to be regional keys. Thus a label ax indicates the tile is in
position a with respect to its parent, and its parent is in position x with
respect to its parent. Now if xis itself a key, this parent is supposed to carry
an additional label. This label is tucked into the regional key adjacent to
our original tile. That is, suppose the parent supertile is supposed to be
labeled xy. Then the local key within this parent is ax and the regional
key within this parent is cy. Of course if the parent is not itself a key, y is
vacant.
In this fashion, the entire hierarchy can be encoded with only one or
two digits in each tile - if information can be transmitted over arbitrary
distances from regional keys to the boundary of the supertile the key labels.
In figure 11 a portion of an L-tiling is encoded. The labels b and d, and
unfilled keys c have been left out of the illustration for clarity.
This simple mechanism is, forgive the author, the "key" to the whole
construction (although it is deeply disguised and virtually unidentifiable in
[6]).

4. Mechanisms

We prove the Theorem as follows: we first construct a new set of marked tiles
and matching rules by examining structures in supertiles fo some bounded
size. We define "well formed supertiles" as configurations of the new tiles
that are in some essential way equivalent to supertiles in the original tiling.
Then we proceed by induction:
We assume that every tile in the matching rule tiling lies in a well-
formed supertile of level n, and show this implies every tile lies in supertile
of level n + 1.
The key, of course, is selecting the right structures to encode in the new
marked tiles. Consider what properties we need from supertiles to force
them to organize into the next larger size supertile:
493

b cy

xy
ax
r

a is kx:al key
c is regiooal key

Thus ax cy indicate
next level up is
xy

labels b, d and
unfilled c_
have been re-
moved for darity

Figure 11 . Encoding addresses through local and regional keys

Supertiles should be the correct shape and size.


Supertiles should be labeled with their intended position with respect
to their parent.
Supertiles should be "combinatorially inert" except at a few special
points. {Since every neighborhood in the tiling is to contain only a
finite amount of information- and so can only serve a finite number of
supertiles- whereas a given point may be on the boundary of infinitely
many supertiles.)
At these combinatorially active sites, sibling supertiles are to compare
labels, fix their relative positions and orientations, and transmit in-
formation to the parent. "Sites" and "terminals" play this role in the
494
construction.
Finally there must be some structure to transmit information from
the sites connecting children to parent to the sites connecting parent
to grandparent. Moreover, every neighborhood in the tiling should only
lie in finitely many of these networks. (As a bonus, however, such a
network is the ideal place to store key labels). "Skeletons" and "wires"
play this role in the construction.

However, it is not hard to always construct such structures in any given


substitution tiling. The seemingly endless technical details in (6] arise be-
cause of numerous special situations.
We will only discuss one of the structures we exploit: the skeleton, a self-
organizing structure that simultanously contains and transmits information
about each supertile.
In the lower left of figure 12, the substitution for the pinwheel tiling
(Rad] is shown; above and to the right skeletons for three generations of
supertile are shown. Note that the skeletons are each connected sets, are
connected to each other at special points (sites), include all the "highest
level" edges in the supertile, every edge is in only finitely many (i.e. one)
skeletons, and that the skeletons are all similar.

Figure 12. Skeletons

Given the children supertiles are formed correctly, in order for the par-
ent's skeleton to form correctly, we must know the locations of the endpoints
("terminals") of each of its edges. If the end points lie on the skeletons of
lower level supertiles, no further work is needed. However, we sometimes
need an additional structure: we link certain terminals ("mesovertices") to
the skeleton through a series of lower level supertiles. Such a series is a
"vertex wire". A supertile may thus carry, for certain of its vertices, certain
information associated with some higher level supertile.
495

5. On the utility of substitution tilings


The title of this section admitedly overreaches its contents. But it is the
author's belief that, ultimately, this will make a fine title for a much longer
work. At the moment, this is only conjecture.
We will give two, simple, "theoretical applications."

1) Grids are used for any number and variety numerical simulations.
There is a seeming trade-off that must be faced by the designer of the
simulation:
One can choose a highly regular grid, in which one can ea.o;ily address
locations, adjacencies, etc. without the use of direct pointers between neigh-
boring cells. But such a grid may not ea.o;ily conform to the geometry of the
simulation, or worse, the geometry of the grid may introduce extraneous
results. For example, suppose we simulate the geometry of the circle on a
very fine square grid. Certain properties, such a.-; area, can be calculated to
arbitrary precision by subdividing the grid. But others, such a.-; perimeter
will never· be approximated, regardless of how fine the grid is. In fact, even
the length of a generic line segment is very badly estimated by a square
grid, no matter how fine.
To remedy this one can choose a grid that appears irregular, that is
transparent to the geometry of the simulation. But in such a grid, it is
difficult to "look up" a given location; the grid is constructed only through
local information: which cells are adjacent to which others. This also carries
a fairly large memory cost, since all adjacencies must be encoded.
Substitution tilings provide a middle road. They are clearly algorithmic
and have all the advantages of a rigid hierarchical structure. On the other-
hand, they admit an extraordinarily rich family of possible geometries. To
give one example, substitution tilings generically are isoparametric- that
is, one can "usually" approximate arbitrary smooth curves to arbitrary
precision.

2) Substitution tilings provide models of cell-division and growth, often


to great effect [16]. Whether matching rules or addressing are actually useful
is beyond this author's expertise. But it can be said that matching rules
encode local "well-formedness". Our main theorem can be interpreted a.-;:
every sufficiently regular hierarchical structure can be endowed with local
conditions such that if the structure is not faithfully reproduced, the local
conditions are not satisfied. One might imagine a growing embryo; local
conditions might exist which can detect a global defect a.-; soon a.-; it arises.
(Whether such a mechanism actually exists is of course quite far beyond
the scope of this discussion)
496
Finally, the proof of the theorem is an exercise in self-organization, a
fundamentally interesting phenomenon.

References

1. R. Berger, The undecidability of the domino problem, Memoirs Am. Math. Soc. 66
(1966) ..
2. N.G. de Bruijn, Algebraic theory of Penrose's non-periodic tilings, Nederl. Akad.
Wentensch Proc. Ser. A 84 (1981), 39-66.
:3. L. Danzer, A family of 3D-spacefillers not permitting any periodic or quasiperiodic
tilirig, preprint.
4. L. Danzer, personal communication.
5. C. Godreche, The sphinx: a limit periodic tiling of the plane, J. Phys. A: Math. Gen.
221989, L1163-L1166.
6. C. Goodman-Strauss, Matching rules and substitution tilings, to appear in Annals of
Math.
7. C. Goodman-Strauss, A small aperiodic set of tiles, preprint.
8. C. Goodman-Strauss, An aperiodic tiling of E" for all n > 1, preprint.
9. C Goodman-Strauss, Addresses and substitution tilings, in preparation.
10. B. Grunbaum and G.C. Shepherd, Tilings and patterns, W.H. Freeman and Co.
(1989).
11. J. Kari, A small aperiodic set of Wang tiles, preprint.
12. , T.T.Q. Le, Local rules for quasiperiodic tilings, Proceedings of the Fields Institute
(1995).
n. J .M. Luck, A periodic structures: geometry, diffraction spectra, and physical proper-
ties, to appear in Fund. Prob. in Stat. Mech VIII.
14. S. Mozes, Tilings, substitution systems and dynamical systems generated by them,
J. D'Analyse Math. 53 {1989), 139-186.
15. R. Penrose, The role of aesthetics in pure and applied mathematical research, Bull.
lnst. of Math. and its Appl. 10 (1974) 266-71.
16. P. Prusinkiewicz and A. Lindenmeyer, The Algorithmic Beauty of Plants, Springer-
Verlag (1990).
17. C. Radin, The pinwheel tilings of the plane, Annals of Math. 139 (1994), 661-702.
18. C. Radin, personal communication.
19. E.A. Robinson, personal communication.
20. R. Robinson, Undecidability and nonperiodicity of tilings in the plane, lnv. Math.
12 (1971), 177.
21. L. Sadun, Some generalizations of the pinwheel tiling, preprint.
22. M. Senechal, Quasicrystals and geometry, Cambridge University Press (1995).
2:3. J .E.S Socolar, personal communication.
24. J .E.S. Socolar and P.J. Steinhardt , Quasicrystals II. Unit cell configurations, Phys-
ical Review B 34 (1986), 617-647.
25. B. Solomyak, Non-periodicity implies unique composition for self-similar
translationally-finite tilings, to appear in J .Disc. and Comp. Geometry.
26. P.J. Steinhardt and S. Ostlund, The physics of quasi crystals, World Scientific (1987).
27. W. Thurston, Groups, tilings and finite state automata: Summer 1989 AMS collo-
quim lectures, GCG 1, Geometry Center.
28. P. Walters, editor, Symbolic dynamics and its applications, Contemporary Mathe-
matics 135 (1992).
THE SHELL MAP
The structure of froths through a dynamical map

TOMASO ASTE
Laboratoire de Dynamique des Fluides Complexes,
Universite Louis Pasteur Strasbourg, 67084 France.
tomaso@ldfc. u-strasbg.fr

1. Introduction

The shell map is a very simple representation of the structure of foams, com-
bining the geometrical (random tiling) and dynamical (loss of information
from an arbitrary cell out) aspects of disorder. We will illustrate it and give
several examples, including a few arising from discussions in Cargese. This
chapter is written by following the main lines of two previously published
papers [1, 2].
In Nature, space-filling disordered patterns and cellular structures are
widespread [3, 4]. These structures (froths) are partitions of D-dimensional
space by convex cells. Disorder imposes that each vertex has minimal num-
ber of incident edges, faces and cells (D + 1 edges incident on a vertex, D
faces incident on a edge, D - 1 cells incident on a face, in D-dimensions,
Fig.1). In this respect, a froth is a regular graph, but the number of edges
bounding each face, the number of faces bounding a polyhedral cell, etc., are
random variables [5]. Minimal incidences implies also that the topological
dual of a froth is a triangulation (Fig.2), a useful representation of packings.
Indeed, for any given packing, or point set, one can construct the Vorono'i
tessellation [6], which is a space-filling assembly of polyhedral cells. When
the starting points are disordered (no special symmetries) the Vorono'i tes-
sellation is a froth. The space filled by the froth can be curved (Fig.1). This
is for instance the case in amphiphilic membranes [7], fullerenes, the basal
layer of the epidermis of mammals [8] or the ideal structure of amorphous
materials [9, 10]. Disorder does not necessarily imply inhomogeneity. On the
contrary, in many cases, disordered froths are very homogeneous (cells with
very similar sizes and regular shapes). This is -for instance- the case in the
epidermis, where the biological cells have homogeneous sizes and isotropic
shapes but the structure is disordered. Indeed, the disordered arrangement
497
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 497-510.
© 1999 Kluwer Academic Publishers.
498

Figure 1. An example of two dimensional disordered cellular structure (froth) generated


by cell division and coalescence transformations (left). A froth on a curved space (right).

is the one which best guarantee both the partition of the curved space into
similar cells and the invariance under mitosis and detachment.
The interplay between disorder and curvature is illustrated in this chap-
ter by representing the froth as organized in concentric layers of cells around
an arbitrary central cell [1] (Figs.3 and 5). The structure is built from the
central cell outward like an ever expanding jigsaw puzzle without boundary.
The radial map, from one spherical layer of cells to the next, is the logistic
map [11], and the geometrical tiling is expressed mathematically as a dy-
namical system [12). The isotropy of the disordered structure is expressed
locally by averaging over each layer. The over-all translational in variance is
manifest in the independence of the structure and properties on the choice
of the central cell. The radial map from one layer to the next includes both
effects of disorder and of space curvature.
In a two dimensional froth, given a cell with n edges, one can define its
topological charge as q = 6-n. The total charge of a froth with N cells is the
sum over the charges of each cell: QT = L:i qi = (6- (n) )N = 6x, with x the
Euler-Poincare characteristic of the manifold tiled by the froth [13, 14). The
charge is a topological invariant, it cannot be generated or destroyed and
the local topological transformations in the froth redistribute it between
adjacent cells. The total charge is equal to zero in Euclidean froths, it is 12
for froths on the surface of a sphere and it is negative in spaces with negative
Gaussian curvature. In two dimensions, it is therefore possible to define the
curvature of a surface tiled by a froth by analyzing the local topological
configuration of its tiles. In three dimensions, for N --t oo or in closed
499

Figure 2. Froths and triangulations are dual structures. Triangulations are useful rep-
resentations of packings.

elliptic froths, the Euler's relation is homogeneous (x = 0, independently


on the space-curvature) and from the Gauss-Bonnet formula [14] it is not
possible to distinguish the global curvature of a tiled manifold from the
local average properties of its tiles. On the other hand, we show that, also
in three dimensions, the map makes possible to define the curvature of
the space from local topological configurations [1]. This is done by simply
computing the number of cells in successive layers.
In this chapter, we describe the map which gives the number of cells
Kj in a layer distant j from the central cell as function of the average
topological properties of the cells in the previous layers (section 2) [1]. We
discuss the link between the map and the space curvature giving examples
in two and three dimensions (section 3) [1, 2]. By using the map we exploit
the freedom of constructing a froth with different local topological config-
urations and determine the three dimensional Euclidean structures which
maximize such freedom (section 4) [1, 2].

2. From a cell to the whole froth, a topological map


All froths can be studied as structured in concentric layers of cells which are
at the same topological distance from a given central cell. The topological
distance between two cells is the minimum number of edges crossed by a
path from one cell to the other [1]. The layers are closed rings of irregular
polygons in two dimensions and spherical caps of irregular polyhedra in
three dimensions. The cells making the layer j can be distinguished in two
categories. Some cells have simultaneously neighbours in the layers j - 1
and j + 1, these cells make themselves closed layers and constitute the
skeleton of the shell-structure. Other cells (or clusters of cells) are local
inclusions (topological defects) between the layers of the shell-skeleton (they
don't have neighbours in the layer j +1). The shell-skeleton is itself a space-
500

r
I
\
A
I~

I
\

I
\

' ..... ___ / AL_ , A, __ . .... ___ ...

Figure 3. An example of SSI froth. The number of cells in each layer can be calculated
in term of a simple map

filling froth, hierarchically organized around the germ cell. Once the germ
cell is chosen, the shell-structure and its skeleton are univocally defined, but
different germ cells generate different skeletons. We call shell-structured-
inflatable (SSI) a froth free of topological defects. In this case, the shell-
structure and its skeleton coincide. An example of a two dimensional SSI
disordered cellular structure is given in Fig.(3). A three-dimensional regular
SSI structure (the Kelvin froth [15]) is given in Fig.(4). In this paragraph,
we first study SSI froths and then generalize the results to generic froths.

2.1. A TOPOLOGICAL MAP FOR SSI FROTHS

For SSI structures, a recursive equation gives the number of cells I<j in a
layer at a distance j from the germ cell, as a function of the number of cells
in the previous two layers. In two dimensions, one can easely verify that
[1, 12, 16, 17] (see Fig.3)

(1)
where the inflation parameter Sj is related to the average topological prop-
erties of the cells in layer j,

Sj = (n)j- 4 (2)
with (n)j the average number of neighbours per cell in layer j. The map
starts with I<0 0 and I<1 n = = =
number of neighbours of the central cell.
For example in the hexagonal lattice one has n = 6, (n)j = 6 and therefore
Sj = 2. In this case the map (1) correctly gives I<j = 6j.
501

Figure 4. (a) An example of 3D cellular system: the "Kelvin froth" . A part of its
shell-network is shown in (b) , it is the 2D elliptic set of facets which are bounding
externally a layer. (c) Construction of the 3D "Kelvin cell" from the 2D shell network.
(d) The "twisted Kelvin cell".

In three dimensions, one can write a recursive equation similar to Eq.(1)


[1 , 2]
Kj+l = SjKj- bjKj- 1 + Cj (for j 2: 1) . (3)
with
1 N
sJ· 2((J)j- 6)((n )j - 4) - bj - 1
(nN)j- 4
bJ·
(nN)j-1 - 4
Cj = 2(bj- Sj + 6((nN)j- 4) + 1) (4)

Here (f) j is the average number of neighbours of the cells in layer j. The
quantity (nN)j is the average number of edges per face in the 2D elliptic
set of facets which are bounding externally layer j (the shell-network, see
Fig.4(a,b) and [1, 2] for details). The map starts with K 0 = 0 and K 1 = f =
number of neighbours of the central cell. For example in the Kelvin structure
one has f = 14, (J)j = 14 and (nN)j = 5 + (24j2 + 24j + 7)- 1. By using
the map (3) we obtain K2 =50, K3 = 110 and in general Kj = 12j2 + 2,
which is the correct answer [18] .
Equations (1) and (3) take a much simpler form if the average topolog-
ical properties of the cells in layer j are independent on the layer number
(i.e. (n)j = (n), (J)j =(f) and (nN)j = (nN)) . In general, these quantities
can vary as one goes from one layer to the next (as for the Kelvin case) .
However, in the limit j-+ oo they must converge towards the averages over
the whole structure (in the Kelvin case (nN)j -+ 5) . Moreover, since the
502
choice of the central cell is arbitrary, in disordered systems this asymptotic
behaviour is reached much faster if one averages over the central cell first.
Finally, the relations (1) and (3) are also valid when the "central cell" is a
cluster or even an infinite set of cells. For instance the starting configura-
tion (the cells with j = 0) can be an infinite linear strip of adjacent faces in
2D or a planar layer of adjacent bubbles in 3D. In this case the asymptotic
behaviour and the average over the system can be set from the beginning.
When the average topological properties of the cells in layer j are inde-
pendent on the layer number, Eqs.(1) and (3), give [1, 12]

( ki+~ ) = ( s -1 ) ( _ki ) (5)


}(j 1 0 }(j-l

This equation is the logistic map [11]. In two dimensions, Kj = }(j is the
number of polygonal cells which are making the layer j. In three dimensions,
for s =J 2, Kj is the number of polyhedral cells making the layer j plus the
additive constant c/(2- s). When s = 2 Eq.(5) cannot be used and Eq.(3)
must be used instead.

2.2. TOPOLOGICAL DEFECTS AND NON-SSI FROTHS

In almost all disordered froths, non-SSI inclusions (topological defects) are


present. An example of non-SSI froth is given in Fig.5. A defect in layer
j is a cell (or a cluster of cells) which have no neighbours in layer j + 1.
Examples of 2D topological defects are given in Fig.5. The fraction (8) of
defects in layer j respect to the total number of cells, tends to a constant
as j tends to infinity. Typically, in 2D physical froths one finds c5 ~ 0.1 [19],
whereas very disordered computer simulated froths have 0.15 < c5 < 0.6
[20].
Maps (1) and (3) apply to SSI froths only. But, for any froth, given
a central cell, one can always single out the topological defects. Then the
reduction to the shell-skeleton is made by eliminating the defects. The
shell-skeleton is itself a space-filling froth hierarchically organized around
the germ cell, and it is SSI by construction. Maps (1) and (3) are therefore
applicable on this structure.

3. Space curvature from the map


Let start from a given cell of the froth and count its number of neighbours
}(j at a distance j. We expect different behaviours for }(j v .s. the distance j
depending on the curvature of the manifold tiled by the froth, as schemat-
ically shown in Fig.6. Mathematically, these different behaviours can be
easely studied from Eq.(5) when Sj = s independent of j. In this case, one
503

j+l

~ ~
·1:1. .
j- 1 _,. J J

~ ·J.:;
~Vertex renormalizalion

,..
J•l

i"
I+ I

J- 1
Edge renormalization

Figure 5. Example of non SSI froths (left). Defects, in layer j, are cells which have
no neighbours in layer j + 1. They are topological inclusions between the layers that can
be considered as vertex- or edge-renormalizations (right). The topological map doesn't
applies to such configurations, but it applies on the shell-skeleton, which is the SSI
structure obtained by eliminating the defects.

has the following solution of the map (5) [1],

Kj ex ( exp( <pj) - exp( -<pj)) (6)

The parameter s separates the solutions into classes. The region lsi < 2
is associated with <p = i cos- 1 ( s/2). It has bounded, finite trajectories
in the (j, I<j) parameter space. This is the characteristic behaviour of
cellular tilings of elliptic manifolds (see Fig.(6c)). For lsi > 2 one has
<p = cosh- 1 (s/2), and unbounded exponential trajectories (typical of hy-
perbolic tilings, Fig.(6a)). The point s = 2 separates the elliptic from the
hyperbolic regions. It corresponds to Euclidean tilings and the solution of
Eqs.(l) and (3) is Kj ex jD- 1 , with D the space-dimension (this is the
behaviour expected from elementary geometry, Fig.(6b)). Note that this
connection between the space curvature and the inflation parameter s ap-
plies both in two and three dimensions.
In two dimensions, equation (6) is the topological version of the relation
(attributed to Gauss) between the circumference C of a circle and its ra-
dius p measured on a manifold of constant curvature R (where R Ricci =
curvature scalar= 2 x Gaussian curvature),

(7)
504

1000
K.
I 900

800

700

600

500

400

300

200

100

0
0 10 20 30 40 .so (c)
J

Figure 6. The curvature of the space tiled by the froth is obtained from the behaviour
of the number of cells ( J(i) in successive layers as function of the topological distance
(j) from the central cell. The flat, Euclidean 2D space corresponds to a linear increment
of Kj with j (b) . In elliptic spaces, J(j first increases, reaches a maximum and then
decreases (c). In hyperbolic tilings this number increases exponentially (a).

In our case, C represents the topological length (I<j) of the boundary


of the cluster with topological radius j and H
is the parameter <p of
Eq.(6), given here in terms of the curvature.
Here are some examples of regular 2D froths with different s. To s =
-1 corresponds an elliptic froth made with four triangles, i.e. the surface
of a tetrahedron. The value s = 0 corresponds to an elliptic froth made
with six squares, the surface of a cube. To s = 1 is associated an elliptic
froth with pentagonal faces, the surface of a dodecahedron. It is known
that the hexagonal lattice (s = 2) is the only regular froth which tiles
the Euclidean plane. Froths with heptagonal (s = 3) or octagonal faces
(s = 4) tile hyperbolic surfaces. Two-dimensional disordered tilings can be
generated by Voronoi' (6) tessellations around points at random on a surface
with arbitrary curvature. Here the map must be studied experimentally case
by case. In the Euclidean plane, the linear behaviour of I<j v .s. j predicted
by our map, has been experimentally confirmed by many measurements
on natural and computer generated froths (20, 19, 21). The behaviour of
I<j in random Voronoi' networks constructed on curved surfaces is under
investigation (22).
505
In two dimensions, the classification by the map (5) is identical to that
provided by the combination of the Gauss-Bonnet theorem [14] and Euler's
equation:

J R 7r 7r 7r
-da = 27rX = -QT = -(6- (n))N = -(2- s)N
2 3 3 3
(8)

where N is the total number of cells in the system. Equation (8) shows
clearly that the parameter s < 2 corresponds to global positive Gaussian
curvature, s > 2 to negative curvature and s = 2 to the Euclidean limit.
In 2D, the determination of the curvature by the map (5) or by the Gauss-
Bonnet theorem are identical, but the two methods are independent. In
three dimensions, the Euler's relation is homogeneous and from the Gauss-
Bonnet formula it is not possible to distinguish the global curvature of
a tiled manifold from the local average properties of its tiles. However,
the map (5) is still applicable. It provides therefore a general means of
describing the curvature from topological considerations, also in 3D.
Here are two example of 3D froths with different inflation parameter
s. The regular froth made by packing dodecahedras has (!) = 12 and
(nN) = 5, which substituted into Eq.(4) leads to s = 1, indicating therefore
that this froth tiles a positively curved space. Indeed, it is known that it is
a closed structure made with 120 dodecahedras. The Kelvin froth of Fig.4
has (!) = 14 and asymptotically (nN) = 5, corresponding to s = 2, which
correctly indicate that the Kelvin's cells fill the Euclidean 3D space.

4. Construction of three dimensional disordered structures from


the map
Three-dimensional space-filling cellular systems are highly correlated struc-
tures. Each cell has a different shape and there exist in general very few
configurations where all cells pack together and fill space without gaps or
overlaps. The space-filling condition is a very important constraint which
strongly determines the properties of disordered structures.
In froths, the incidence numbers (number of edges on a vertex, of faces
on a edge, etc.) are fixed at their minimal value. By contrast, the coordina-
tion number (number of faces of a 3D cell, number of edges of a 2D face) is
a random variable and its average characterize the structure topologically
[5]. In two dimensions, the Euler relation fixes the average number of edges
per face (see Eq.(8)). In three dimensions, the Euler relation associates the
average number of edges per face (n) with the average number of faces per
cell (!)
12
(!) = 6- (n) (9)
506

/=14

..
<f>=12

- ,

Figure 7. Three dimensional froths generated by using the map starting from the shell
network. The shell network is always the result of a superposition of two three connected
networks, in the examples (a), (d) it is the superposition of two hexagonal lattices.

So that the two average coordination numbers are related, and one of these
two quantities is not constrained.
The representation of the cellular system in concentric layer around a
given central germ cell can be also regarded as an iterative way of construct-
ing disordered packings. In the Euclidean space (and in the asymptotic limit
when (f)j and (nN)j are independent of j), the inflation parameter of the
map is fixed at s = 2. For this value of s Eq.(4) gives [1]
8
(!) = 6 + (nN)- 4 (10)

This equation relates the average coordination ( (nN)) of a two dimensional


structure (the shell-network) with the average coordination ( (!)) of the
three dimensional cellular system. The shell-network is the result of a su-
perposition of two elliptic 3-connected networks, the "incoming" and the
"outgoing" froths [1]. The pattern of edges constituting the shell-network
sets the value of (nN). Therefore, Eq.(10) allows us to construct systemat-
ically 3D Euclidean SSI froths starting from 2D shell-networks.
The simplest 2D froth is the hexagonal lattice. The examples displayed
in Fig.4 and Fig. 7 (a-e) illustrate the construction of ordered, monotiled 3D
froths from a shell-network generated by different superpositions of two
hexagonal lattices (see also [1, 23]). The cell in Fig.4(c) is topologically
507

Figure 8. The shell network for the Z and A15 phases (here the shells are planes).
In the Z and A15 shell networks there are respectively 12 and 14 pentagons per each
hexagon. Therefore, (nN} result 66/12 and 76/15 for Z and A15 respectively. In the Z
phase, successive shells have alternating configurations (j) and (j + 1).

equivalent to Kelvin's a-tetrakaidecahedron (it builds up the Kelvin froth


shown in Fig.4(a)), and the cell 4 (d), to its twisted variant [24, 25]. Both
structures have (!) = 14 and asymptotically (nN) = 5, and fill the Eu-
clidean space. In Fig.7(a) is shown part of a shell-networks with 5-sided
faces, generated by the superposition of two "squeezed" hexagonal lattices
and its unit cell (b) (see also [1, 23]). This structure has again (nN) = 5,
(!) = 14 and is an Euclidean (s = 2) space-filler. With another inter-
section of the two "squeezed" hexagonal lattices, one generates the shell-
network shown in Fig.7 (c), it has (nN) = 4.8 a 3D unit cell (7 (d)) with
(!) = 16 (8 quadrilaterals, 6 hexagons and 2 octagons). This unit cell is
a monotile Euclidean space-filler that, as far as we know, was reported
for the first time in [1]. Recently, this tilihg has been proposed [26] as
the structure of the ternary crystal ThCr3Si 4 • Fig.7 (e),(f) show an ex-
ample of an Euclidean shell-structured-inflatable froth made of two differ-
ent cells. The shell-network in Fig.7 (e) has also two different tiles with
(nN) = 32/6 = 5.33 .... The associated 3D unit cell (7 (f)) has (!) = 12
and (n) = 5. This structure is an Euclidean space-filler with s = 2. One can
in general show that any Euclidean shell-structured-inflatable froth made
with topologically identical cells can be constructed from a shell-network
generated by the superposition of two hexagonal lattices.
We can restrict our attention to a special class of structures where the
packed polyhedras have only pentagonal and hexagonal faces. In this case,
the range of variability of (!) is restricted between 12 and 14 by Eq.(9)
and (10). The lower value corresponds to a structure with only pentagonal
faces: a packing of dodecahedras, and it cannot be realized in Euclidean
spaces (it is an elliptic system of 120 dodecahedras). Whether the upper
value is realizable or not in Euclidean spaces is still an open question.
(The only known structure with (!) = 14 made only with hexagon and
pentagons is the Goldberg froth [27], but this packing is hyperbolic [1]). Two
508

5,35

5,30
decuNed
5,25 dodacah~s
Frank-Kasper,

~ :. t.c.',p. phases r::t'ffc)


5,20
A1~-~ J
i-
\
5,15

5,10

5,05

5,00
12 13 <f> <f> 14

Figure 9. Full lines: average number of edges per face in the whole froth ( (n)) and in
the shell-network ( (nN)) as function of the coordination number ((f)). The circles and
diamonds are some examples of known three dimensional cellular structures made by
packing polyhedra with hexagonal and pentagonal faces only. Several of them (diamonds)
occurs in nature (the TCP phases of metallic compounds).

structures with coordinations number equal to 13.2 and 13.333 ... have been
obtained by decurving iteratively the curved structure of 120 dodecahedras
by introducing hexagonal faces [9, 28]. A natural class of ordered structures
made only with pentagons and hexagons are the 24 known TCP. They are
Euclidean and have 13.333 ... ~ (!) ~ 13.5 [30, 31, 32, 33, 34]. The TCP
A15, with (!) = 13.5, is the known packing of polyhedral cells of equal
volumes with minimal interfacial area per cell [35]. In a previous paper [1],
we proved that all TCP structures could be described topologically by the
map (5). In figure 8 are given two examples of shell networks generated by
combinations of pentagonal and hexagonal tiles only. From these networks
the structure of the Z phase and of the A15 phase can be constructed. They
have coordination (nN) = 66/13 = 5.076 ... (Z) and (nN) = 76/15 = 5.066 ...
(A15), that substituted into Eq.(10) give correctly (!) = 13.42 ... for Z and
(!) = 13.5 for A15.
Figure 9 summarizes in a plot some of the previous considerations and
results. The two lines in Fig.(9) are plots of Eq.(9) and (10). As one can
clearly see, in a given structure, (n) and (nN) are different in general. The
packing is organized so that the faces of the shell-network have an average
509
number of sides ((nN)) different from that of the whole structure ((n)). But
there is a point [1] where these two values coincide:

(nN) = (n) = (n)* = 10 + 2v'7 (11)


3
and, from (9) or (10)

(!)* = 8 + 2v'7 = 13.29 ... (12)

When (f) = (!)* = 13.29 ... , an arbitrary cell has the freedom to
adhere to a layer by any subset of its faces, without adjustment. This
freedom gives therefore many more local possibilities for building up the
structure. A coordination number (f) = 13.29 ... characterizes therefore
an hypothetical SSI topological packing which fills the Euclidean three-
dimensional space with the largest possible number of local arrangements,
that is with maximum entropy. Coordination 13.29 ... lies in between the
two froth obtained by decurving dodecahedras and it is inside the interval
where (f) is ranging in natural structures and computer simulated froths
~5, 29, 30, 31, 32, 34, 36, 3~.
Note that, this structure has a coordination number which is irrational.
It cannot be therefore a periodic crystalline structure. It can only be ap-
proximated by disordered froths or quasi-crystalline structures.
The author acknowledge many discussions with N. Rivier. A special thanks toM. O'Keeffe,
who sent us the preprints of his latests works. This work was partially supported by EU,
HCM Program, "FOAMPHYS" network, contract ERBCHRXCT940542 and by TMR
contract ERBFMBICT950380.

References
1. Aste, T., Boose, D., and Rivier, N. (1996) From one cell to the whole froth: a
dynamical map, Phys. Rev. E 53 6181-91.
2. Aste, T., and Rivier, N. (1997) Topological Modeling of disordered cellular structures,
in Shape Modeling and Applications (IEEE Computer Society Press, Los Almitos) 2-9.
3. Thompson, D'A.W., (1917, 1942) On Growth and Form, Cambridge Univ. Press.,
ch.7.
4. Weaire, D., and Rivier, N. (1984), Soap; cells and statistics: Random pattems in two
dimensions, Contemp. Physics 25, 59-99.
5. Aste, T., and Rivier, N. (1995) Random cellular froths in spaces of any dimension
and curvature, J. Phys. A 28, 1381-98
6. Voronoi·, G. (1908) Recherches sur les paralleloedres primitifs, J. reine angew. Math.
134 198-287.
7. Charvolin, J., and Sadoc J.F. (1990) Structures built by amphiphiles and frustrated
fluid films, Colloque de Physique C 7 83-96.
8. Rivier, N., and Dubertret, B. (1995) Why does skin stay smooth? The dynamics of
tissues in statistical equilibrium, Phil. Mag. B 72 311-322.
9. Sadoc, J.F., and Mosseri, R. (1985) Hierachical interlaced networks of disclination
lines in non-periodic structures, J. Physique 46 1809-1826.
510

10. Sadoc J. F., and Rivier, N. (1987) Hierarchy and disorder in non-crystalline struc-
tures, Phil. Mag. B 55 537-573.
11. Schuster, H. G. (1984) Deterministic Chaos (Physik-Verlag, Weinheim).
12. Rivier, N. (1985) unpublished notes; (1986) seminar, Imperial College.
13. Henle, M. (1979) A Combinatorial Introduction to Topology, (Dover, New York).
14. Kreyszig, E. (1991) Differential Geometry (Dover, New York).
15. Thomson, W. (Lord Kelvin) (1887) On the Division of Space with Minimum Par-
titional Area, Phil. Mag. 24 503-514.
16. Troadec, J.P. (1985) unpublished notes.
17. Fortes, M.A., and Pina, P. (1993) Average topological properties of successive neigh-
bours of cells in random networks, Phil. Mag. B 67, 263-276.
18. O'Keeffe, M. and Hyde, S. T. (1996) Crystal Structures I: Patterns and Symmetry,
(Mineral. Soc. Amer., Washington D.C.).
19. Aste, T., Szeto, K. Y., and Tam, W. Y. (1996) Statistical properties and shell
analysis in random cellular structures, Phys. Rev. E 54 5482 -92.
20. Ohlenbusch, H. M., Aste, T., Dubertret, B., and Rivier, N. (1997) The topological
structure of 2D disordered cellular systems, European J. of Phys., submeted.
21. Dubertret, B., and Rivier, N. (1997) The Renewal of the Epidermis: A topological
Mechanism, Biophys. Journal73 38-44.
22. Demartino, M., Spagnuolo M., and Aste, T., work in progress.
23. Glazier, J. A., and Weaire, D. (1994) Construction of candidate minimal-area space-
filling partitions, Phil. Mag. Lett. 70 351-356.
24. Williams, W.M., (1967) Geometry, Structure, Enviroment, (Southern Illinois Uni-
versity) 20.
25. Williams, W.M., (1968) Space Filling Polyhedron: Its Relation to Aggregates of
Soap Bubbles, Plant Cells, and Metal Crystallites, Science 161 276-277 ..
26. O'Keeffe, M., (1997) On a space filling polyhedron of Aste, Boose and Rivier, Phil.
Mag. Lett., to appear.
27. Goldberg, M., (1934) Tohoku Math. J. 40 226.
28. Rivier N., and Sadoc, J. F. (1988) The Local Geometry of Disorder J. non-cryst.
Solids 106 282-85.
29. Oger, L.,Gervois, A., Troadec, J. P., and Rivier, N. (1996) Voronoi tessellation of
packings of spheres. Topological correlations and statistics, Phil. Mag. B 74 177-197.
30. Frank, F. C., and Kasper, J. S. (1958) Complex Alloy Structures Regarded as
Sphere Packings. I Definitions and Basic Principles, Acta crystallogr. 11 184-190.
31. Frank, F. C., and Kasper, J. S. (1959) Complex Alloy Structures Regarded as Sphere
Packings. II Analysis and Classification of Representative Structures, Acta crystallogr.
12 483-499.
32. Shoemaker D.P., and Shoemaker, C. B. (1986) Concerning the Relative Numbers
of Atomic Coordination Types in Tetrahedrally Close Packed Metal Structures, Acta
crystallogr. B 42 3-11.
33. Rivier, N. (1994) Kelvin's conjecture on minimal froths and the counter-example
of Weaire and Phelan, Phil. Mag. Lett. 69 297-303
34. M. O'Keeffe, (1997) Sphere Packings and Space Filling by Congruent Simple Poly-
hedra, Acta. Cryst. submetted.
35. Weaire, D. and Phelan, R. (1994) A counter-Example to Kelvin's Conjecture on
Minimal Surfaces, Phil. Mag. Lett. 69 107-110.
36. Matzke, E.B. (1946) The Three-Dimensional Shape of Bubbles in Foam, Am. J.
Botany 33 58-80.
37. Hales, S. (1727) Vegetable Staticks, (lnnys and Woodward, London).
CURVED SPACES AND GEOMETRICAL FRUSTRATION

JEAN-FRANCOIS SADOC
Laboratoire de Physique des Solides
Universite de Paris-Sud
91405 ORSAY cedex, France

1. Introduction

Regular structures are such that no contradiction exists between local and global
requirements in which case the global approach (with symmetry groups) reveals to be
very powerful. In less regular structures, the local configuration may be viewed in
some cases as the discrete analog of a quantity which is the local curvature. Defining
an ideal structure where the local configuration can propagate, is then equivalent to
finding a new geometry with the appropriate distribution of curvature. If such geometry
allows for a global description, this ideal model is again regular and can be studied on
its own. The relation between the initial structure and the ideal one is studied under
different types of mapping. This point of view is called the "curved space model" of
disordered systems and will be discussed here.
A main question that a condensed matter physicist faces is to explain the stability of
a solid. Already in molecules, the most precise quantum mechanical calculations often
shows large diversity for the low energy atomic geometries. Due to their size, solids
require that many approximations be done in order to compute their cohesive energy.
Nevertheless, it is often possible to establish some local rules, of chemical nature, which
leads to low energy configurations and therefore govern structural and chemical order.
It is then crucial to analyze to which extent a low energy local configuration can be
assigned to each atom or molecule; clearly, if it is possible, according to the additive
property for the energy in this type of classical models, this should give at once the
ground state of the solid. The subject of this lecture refers to the opposite case, called
"geometric frustration", when the local order cannot be propagated freely throughout the
space [1].
The simplest example is the case where the local rule consists in packing spheres as
densely as possible, a crude model for metallic atoms, but also by the way of a Voronol
decomposition, a good model for a cellular division of space.
In two dimensions, a local packing of discs, the flat analogs of spheres, is densely
organized if centres of discs are located on vertices of equilateral triangles, which can

511
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 511-526.
© 1999 Kluwer Academic Publishers.
512
tile the plane along the six-fold symmetric triangular lattice; this is an unfrustrated case.
In three dimensions, the local densest packing of spheres is achieved by placing their
centres at a regular tetrahedron vertices. The geometric frustration reveals immediately
in that the 3 dimensional Euclidean space cannot be filled completely by regular
tetrahedra.
Our definition of geometric frustration is general enough not to be restricted to the
atomic level, and even to discrete systems. For example, in liquid crystals, under a
continuous like approximation, it is possible to characterize some systems as being
geometrically frustrated: assemblies of amphiphilic bilayers are an interesting example.

2. Simple two dimensional examples

Two simple two dimensional examples are helpful in order to get some understanding
about the origin of the competition between local rules and geometry in the large.
Consider first an arrangement of identical discs (a model for an "hypothetical" two
dimensional metal) on a plane; we suppose that the interaction between discs is
isotropic and locally tends to arrange the disks in the densest way as possible. The best
arrangement for three disks is trivially an equilateral triangle with the disk centres
located at the triangle vertices. The study of the long range structure can therefore be
reduced to that of plane tilings with equilateral triangles. A well known solution is
provided by the triangular tiling: we find a total compatibility between the local and
global rules and the system is said to be "unfrustrated". In our second two dimensional
example, the interaction energy is supposed to be minimum when atoms sit on the
vertices of a regular pentagon. Or supose a two dimensional foam, whose cells are
regular pentagons. Trying to propagate in the long range a packing of these pentagons
sharing edges and vertices, one faces a difficulty due to the impossibility of tiling a
plane with regular pentagons, simply because the pentagon vertex angle does not divide
21t. Three such pentagons can easily fit at a common vertex, but a gap remains between
two edges. It is this kind of discrepancy which is called "geometric frustration".

Figure I: Tiling of a plane by pentagons is impossible but can be realized on a sphere in the form
of pentagonal dodecahedron.
513

There is one way to overcome this difficulty. Let the surface to be tiled be free of
any presupposed topology and metrics, and let us build the tiling with a strict
application of the local interaction rule. In this simple example, we observe that the
surface inherits the topology of a sphere and so receives a curvature (figure 1). The final
structure, here a pentagonal dodecahedron, allows for a perfect propagation of the
pentagonal order. It is called an "ideal" (defect-free) model for the considered structure.
This will be our constant strategy: to solve the frustration by allowing using new
underlying geometries.

3. Sphere packing

3.1 DENSE STRUCTURES AND TETRAHEDRAL PACKINGS

The stability of metals can be describe using a very simplified picture of metallic
bonding and only keeping an isotropic type of interactions, leading to structures which
can be represented as densely packed spheres. And indeed the crystalline simple metal
structures are often either close packed face centred cubic (f.c.c.) or hexagonal close
packing (h.c.p.) lattices. Amorphous metals can also be modelled by close packing of
spheres. But the local atomic order is well modelled by a close packing of tetrahedra,
leading to an imperfect icosahedral order [2].
A regular tetrahedron is the densest configuration for the packing of four equal
spheres. The dense random packing of hard spheres problem can thus be mapped on the
tetrahedral packing problem. It is a practical exercise to try to pack table tennis balls in
order to form only tetrahedral configurations. One starts with four balls arranged as a
perfect tetrahedron, and try to add new spheres, while forming new tetrahedra. The next
solution, with five balls, is trivially two tetrahedra sharing a common face; note that
already with this solution, the f.c.c. structure, which contains individual tetrahedral
holes, does not show such a configuration (the tetrahedra share edges, not faces). With
six balls, three regular tetrahedra are built, and the cluster is incompatible with all
compact crystalline structures (f.c.c. and h.c.p.). Note however that, in term of number
of contacts between spheres, an octahedral configuration would compete here, for N=6,
with a tetrahedral close packing; but adding more spheres will again favours the latter:
adding a seventh sphere gives a new cluster consisting in two "axial" balls touching
each other and five others touching the latter two balls, the outer shape being an almost
regular pentagonal hi-pyramid. However, we are facing now a real packing problem,
analogous to the one encountered above with the pentagonal tiling in two dimensions.
The dihedral angle of a tetrahedron is not commensurable with 27t; consequently, a hole
remains between two faces of neighbouring tetrahedra. As a consequence, a perfect
tiling of the Euclidean space R3 is impossible with regular tetrahedra. Note that, at this
local level, the deviation from perfectness is of a metrical nature (figure 2). So, in the
case of polytetrahedral structures the frustration is due to the impossibility to tile the
514

space with regular tetrahedra. Slightly softening the spheres will not help. Indeed, the
frustration has also a topological character: it is impossible to fill Euclidean space with
tetrahedra, even severely distorted, if we impose that a constant number of tetrahedra
(here five) share a common edge.

a)
c)

Figure 2: Tetrahedral packing:


-a) A tetrahedron.
-b) The dihedral angle of a tetrahedron is not commensurable with 27t; consequently, a hole
remains between two faces of a packing of five tetrahedra with a common edge.
-c) A packing of twenty tetrahedra with a common vertex in such a way that the twelve outer
vertices form an irregular icosahedron

The next step is crucial: we define an unfrustrated structure by allowing, as in the


two dimensions example, for curvature in the space, in order for the local configurations
to propagate identically and without defects throughout the whole space.

4. Regular packing of tetrahedra: the polytope { 3,3,5 }

Twenty tetrahedra pack with a common vertex in such a way that the twelve outer
vertices form an irregular icosahedron (figure 2-c). Indeed the icosahedron edge length I
is slightly longer than the circumsphere radius r (/, 1.05 r ). It is possible to make this
icosahedron regular by first shortening accordingly the internal tetrahedral edges. But
the price to pay is that now the tetrahedra are no more regular. In terms of ball packing,
this means that the central ball is given a smaller radius than its twelve neighbour.
515

The tetrahedral regularity can then be recovered by giving a small fourth coordinate
to the central site. One then tries to add new shells of sites while keeping the same
icosahedral environment for all the sites. It is possible if these new sites are also given at
fourth coordinate with opposite sign. This procedure can be continued for several shells
until one sees that no new sites are needed. Indeed a finite set of points has been
generated such that each vertex has twelve neighbours in perfect icosahedral
configuration. Let us describe this set:
- There are one hundred and twenty vertices which all belong to the hypersphere s3
with radius equal to the golden number ( -r = (1 + {5) /2 ) if the edges are of unit length.
- The six hundred cells are regular tetrahedra grouped by five around a common edge
and by twenty around a common vertex.
- This structure is called a polytope [3] which is the general name in higher dimension
in the series polygon, polyhedron, ...
Even if this structure is embedded in four dimensions, it can be considered as a
three dimensional figure in the following sense. The hypersphere s3 is defined by the
equation xo2 + x1 2 + x22 + x32 = r2 which shows that, among the four coordinates,
only three are independent. As far as one disregards its interior, the "hyper-ball" limited
by s3, the latter can be considered as a 3 dimensional (curved) manifold. This is
analogous to considering the usual sphere s2 as a curved surface.
This point is conceptually important for the following reason. The ideal models that
have been introduced in the Curved Space Approach are three dimensional curved
templates. They look locally as three dimensional Euclidean models.

4.1 THE { 3,3,5 } POLYTOPE: THE FIRST CURVED SPACE MODEL

The { 3,3,5 } polytope, which is a tiling by tetrahedra, provides a very dense


atomic structure if atoms are located on its vertices. It is therefore naturally used as a
template for amorphous metals, but one should not forget that it is at the price of
successive idealizations where parts of the reality are lost.
The equivalent of a Vorono'i decomposition consist to consider the dual of the
polytope. The dual of the { 3,3,5 } polytope is the { 5,3,3 } polytope, with six hundred
vertices, one hundred and twenty edges,seven hundred and twenty pentagonal faces, and
one hundred and twenty dodecahedral cells. It is an exellent model for an idealised
foams.
The richness of this approach will become obvious when we shall show that there
exists deep intrinsic properties of the materials which survive these simplifications. We
shall see further that it allows to generate a general picture of amorphous solids, with
qualitative and quantitative results, at geometrical and topological levels. The main
results is probably to split the structure into ordered regions and defects, as it will be
explained later.
516

4 .. 2 THE PACKING FRACTION OF THE POLYTOPE

In order to better understand this important question of compacity, it is interesting to


compare packing fraction of different structures. The packing fraction of the polytope
can be easily calculated [4]. One gets the filling factor /=0.774. This is a very high
value which exceeds the value 0.74 of the densest sphere packing in R3 (realized for
example by the f.c.c. or the h.c.p. packing). The reason is rather simple. The f.c.c.
structure is a three dimensional regular packing of tetrahedra and octahedra while
polytope {3,3,5} only contains tetrahedra. The f.c.c. packing deficit is therefore due to
the octahedra which are less efficient packing configurations than tetrahedra.
Now, amorphous metals are known to be less dense than their crystalline
counterpart. This apparent discrepancy between real systems and the curved space
model disappears when one realizes that a decurving procedure from s3 to R3 implies
lowering the packing efficiency of the mapped structure. A simple estimation of the
mapped structure density gives very reasonable values compared to both numerical
simulation of sphere packings and experimental trends in density variations.

5. Decurving and disclinations


5.1 DISCLINATIONS

A disclination is a defect involving a rotation operation, as opposed to the more familiar


dislocation, which is associated to a translation given by it Burgers vector.
A disclination can be generated by a so-called "Volterra" process, by cutting the
structure along a line and adding (or removing) a sector of material between the two
lips of the cut. In two dimensions, this defect is point-like, while it is linear in three
dimensions. The two lips of the sector should be equivalent under a rotation belonging
to the structure symmetry group in order to get a pure topological defect confined near
the apex of the cut.

5.2 A SIMPLE EXAMPLE OF DISCLINATIONS: WEDGE DISCLINATIONS IN


TWO DIMENSIONS

It is possible to describe this defect, and the induced deformation, as a concentration of


curvature (figure 3). Let us first do the Volterra construction with a sheet of paper. We
first cut it along a straight segment up to its centre. Then, upon rotating around this
centre, we can either add or remove a sector, and then glue again along the lips of the
cut. The angle of rotation is called the weight, or the angular deficit, of the disclination
(eventhough the value can be positive as well as negative). As in figure 1, one gets a non
flat sheet of paper with either a conical or a saddle point singularity at its centre. Can we
have a more quantitative measure of this curvature?
517

b)

Figure 3: Disclinations in a hexagonal structure:


a) Positive disclination.
b) Negative disclination.

Consider a non flat surface. Its Gaussian curvature can be detected and measured by
means of the parallel transport of a vector along a closed circuit. The simplest thing to
do is to draw, on the surface, a polygonal line with geodesic edges. Then, a vector is
transported in such a way that its angle with the side along which it is transported
remains constant. When, having circum-travelled around the circuit, the vector returns
to its starting point, it is found to have been rotated as compared to its initial
orientation. This rotation is related to the integrated Gaussian curvature 1( in the area
enclosed by the circuit:
8 =If K"dO'
D
where o is the rotation angle of the transported vector. Analogously, on a surface
punctured by disclinations, a parallel transport of a vector along a circuit which
encloses a disclination also results in a clear rotation of the vector: this rotation angle is
exactly the angle of the disclination. Therefore, a disclination can be considered as a
concentration of curvature.
518

5.3 WEDGE DISCLINATION IN THREE DIMENSIONS

A three dimensional disclination can also be generated via a Volterra process by cutting
the structure along a half plane and adding (or removing) a wedge of material between
the two lips of the cut. The defect is now linear. The two faces of the wedge should be
equivalent under a rotation belonging to the structure symmetry group in order to allow
a perfect matching between the lips and the added wedge. A pure topological defect is
then confined near the axis of the cut. In the case of wedge disclinations in three
dimensional space the axis of the rotation is located on the defective line.

a)

Figure 4: a) Procedure to insert a disclination.


b) Effect of a disclination on a icosahedral configuration: a Z14 coordination polyhedron has
been generated.

It is then clear that wedge disclinations are located along geodesic lines of the
structure. Indeed, if a disclination is introduced in an Euclidean space, the defective line
is a straight line, which is invariant under the rotation defining the disclination. In a
spherical three dimensional space, the line invariant under a rotation is a great circle; if
519
the sphere s3 is embedded in R4, this line is the intersection of s3 with the 2d-plane
invariant under the rotation. Then, in this case, the disclination line will also lie on a
great circle.
As in two dimensions, wedge disclinations can be viewed as loci of curvature
concentration in a three dimensional space. If the disclination is obtained by a Volterra
process in which matter has been removed (added), it is a positive (negative)
disclination. The sign of a disclination can also be determined by parallel transport: if
the transported vector rotates in the same (opposite) direction as the circuit, the
disclination is positive (negative). Therefore, introducing negative disclinations can be
used in order to decurve a positively curved space.

5.4 EFFECT OF WEDGE DISCLINATIONS

A two dimensional structure can be described as a set of edges, joining vertices and
separating polygonal cells. A disclination changes the coordination number when it
goes through a vertex; it changes the number of sides of a polygonal cell threaded by
the defect.
In three dimensions, disclinations also change the network topology. For example,
like in two dimensions, they change the coordination number when they go through a
vertex (figure 4). So, introduction of disclinations not only allows for decurving the
embedding space, but it also generates slight modifications of the local configurations.

5.5 DISCLINATIONS IN A POLYTOPE

Consider a 2n/5 disclination line in the {3,3,5} polytope. The local order around such a
line is shown on figure 4. In the {3,3,5} polytope, the coordination polyhedron of a
vertex is a regular icosahedron. The cut axis is aligned with a 5-fold icosahedral axis
and a 2n/5 wedge is inserted. The new coordination polyhedron is a 14-vertex
triangulated structure and the central vertex is called a Z 14 site in the standard notation
due to Frank and Kasper. The central site and the two opposite vertices on the cut axis
belong to a 2n/5 disclination line. It is possible to show that such line cannot stop in the
material and either ends at the surface, splits at crossing points or form closed lines. The
set of crossing points and disclination segments or lines form a so-called disclination
network. This kind of network play an important role in the description of complex
structures.

A pair of disclination lines in polytope { 3,3,5}


We first describe the generation of two defects lines in the {3,3,5} polytope. It is
possible to create two disclinations in a {3,3,5} polytope along two completely
orthogonal great circles.
The new polytope contains 168 sites, 144 12-fold coordinated vertices Z 12 sites and 24
Z14 sites.
520

There are procedures allowing to generate a large number of disclinations leading to


structure without curvature. In this case a hierarchy of disclination networks could be
the solution [5].

6. Frustration in amphiphiles liquid crystals

There exist, situations in which a fluid is periodically modulated as, for example, in the
case of lamellar and smectic phases of lyotropic and thermotropic liquid crystals, and in
chiral cholesteric phases [6].
Stratification can concern the molecular position, like in the smectic phases of
lyotropic liquid crystals formed from alternated layers of water and organic molecules.
The most observed situation and the easiest to understand concern organizations of
stratified fluids in flat layers, periodically stacked along one direction, which result
from the constant distance imposed by the interactions between interfaces.
However, when some thermodynamical parameter is varied, the systems often build
more complex organizations, periodic along two or three directions. These new
organizations, found in rather different systems, result from a frustration effect, arising
from the conflictual requirement of constant distance and the appearance of a new
symmetry in the stratifications, introduced by a lateral area difference between different
layers of the stratification. The frustration appears related to the flatness of the
embedding Euclidean space, but it can be relaxed if the embedding space is given an
adequate curvature.

6.1 FRUSTRATION IN CRYSTALS OF FILMS

The characteristic scale for this structuration is not related to the constituting molecules
themselves, but to the characteristic length scales of the continuous films arranged
periodically along directions normal to the layers. This kind of structures has been
extensively studied by V. Luzzati [7] who was one of the first to understand their
complexity: they behave like a liquid structure at the molecular scale, although the
order is crystalline at a larger scale.
As said before, in the case of lamellar and smectic systems the stratification
concerns the positions of the molecules. We analyse the problem in the case of systems
of amphiphiles and lamellar phases only, but similar considerations hold for meso genic
molecules and smectic phases.
The film, the bi-layer, is made of two identical layers and several forces fix the
inter- and intra-layers distances: Van der Walls forces, electrostatic interactions
between polar groups at the interfaces, forces created by water polarization and charge
distributions in aqueous layers, hydrophobic interactions which prevent the presence of
water within the layers and fluctuation-induced forces of entropic character.
The precise role of these different interactions, as well as their mutual interplay, are
not fully understood at the moment. However, it is reasonable to state that the main
521

effect of the components normal to the interfaces is to maintain a constant distance


between them, if the interfaces are homogeneous, while the parallel components
determine the interfacial curvatures. Owing to the symmetry of the film with respect to
its middle surface, it is clear that, imposing symmetric curvatures for the two facing
interfaces is not compatible with keeping a constant distances between the interfaces, if
a lamellar type of stacking is kept, as shown in figure 5. The flat interfaces case is
represented in figure 5-a, while figure 5-b shows one conflictual situation with curved
interfaces. This is a typical case of frustration which has no solution in the three
dimensional Euclidean space R 3. As a consequence, the system will adopt a more
complex structure, in order to minimize the energy, that is building the best
compromise between these conflictual forces.

Figure 5: Schematic representation of a periodic system of films with flat interfaces: (a) constant
interfacial distances and zero curvature are compatible in R3; (b) constant interfacial distances
and non zero curvatures are no longer compatible in R3 and the system becomes frustrated.

6.2 LAMELLAR STRUCTURES IN CURVED SPACES

Having recognized a new case of geometrical frustration, it is natural to try to solve the
problem, like in the case of tetrahedral packing, by allowing the embedding space to be
curved. This is indeed possible and such ideal models have been proposed for hi-layers
in three dimensional curved spaces [8].
We limit this first approach to curved spaces with homogeneous positive
curvatures, because we look for configurations which relax the frustrations equally
everywhere; the positive curvature allows to replace the condition of periodicity by a
cyclic condition.
Facing the difficulty of pictorial representations of three dimensional curved
spaces, embedded in R4, and which is only practically solved by the use of
stereographic projections into R3, we find useful to discuss first a simple two
522

dimensional problem of frustration in the Euclidean plane R2, which is relaxed on the
ordinary sphere s2.

6.3 A SIMPLE TWO DIMENSIONAL EXAMPLE

We just consider the intersection of a three dimensional system of layers by a plane R 2


normal to the layers. Doing so, we have a two dimensional model where a film becomes
a strip. As said above, there is no frustration when the interfaces limiting the film are
flat, their two dimensional images being two parallel straight lines bordering a strip.
This results from the fact that the interfacial and middle lengths are equal,
corresponding to equal equilibrium distances for molecules in the bulk of the film, and
molecule heads on interfaces. Frustration appears when the interfaces become
symmetrically curved, or when the interfacial and middle lengths become different, as a
consequence of two different equilibrium lengths between hydrophobic parts of the
molecules in the bulk and hydrophylic parts at the interface.

Ceo

(i.)
~
(m}

~(L)

Figure 6: Example in 2d of a frustrated strip, because its middle length is different from the edge
length, forbidding parallelism; the frustration is eliminated by transferring the strip onto s2.

In our two dimensional case, this situation can be homogeneously relaxed when the
flat space R2 is curved into the spherical space s2 of radius r if, as shown in figure 15,
the middle line is placed on the equator and the shorter interfacial lines are placed on
"parallels" at equal distances from the equator. The frustration is obviously suppressed,
as the interfacial lengths 21t r sin 6 are smaller than the middle length 21t r , while the
basic symmetry of the system is preserved, as the equator is a stationary line separating
the sphere in two identical hemispheres. Moreover, when moving along meridians,
523

which are great circles and therefore geodesics of s2, the interfaces of the strip are
periodically crossed, as when moving along the normals to the interfaces of a lamellar
system. This finite system, with one strip in the curved finite space s2, is therefore
equivalent to the infinite periodic system of strips, but here without frustration.

6.4 BICONTINUOUS AND MICELLAR STRUCTURES

Frustration arises from area differences between the middle surface and the interfaces of
the films. We proceed in a quite similar way as we did in the above two dimensional
example, but with the use of three dimensional curved spaces. The frustration can be
homogeneously relaxed by curving R3 into the spherical space s3. The middle surface
of the film is placed on a particular surface of the curved space which divides it in two
equal subspaces and is a stationary surface. In s3 there are two possible surfaces: the
great sphere s2 of s3 and the so called spherical torus T2.
The two interfaces are placed on equidistant tori or spheres. In these situations the
symmetry of the film is preserved, as the surfaces supporting its middle surface are
stationary surfaces separating the spaces in two equal subspaces; the frustration is
relaxed, as the interfacial area are smaller than the middle area, and the distance
between the interfaces is kept constant. The periodicity of the stacking is replaced by a
continuous cyclic displacement along geodesics normal to the surfaces.
Lamellar phases can be described as stackings of alternating layers of water and
amphiphiles with flat interfaces defined by the polar heads of the amphiphiles. Besides
these phases with periodicity along one dimension, the phase diagrams of these systems
may present other ordered phases with periodicities along two or three dimensions,
curved interfaces and complex topologies [9]. The chapter XXV by J. Seddon and J.
Robins on inverse micellar cubic phases gives more details on such structures.
We just present the case of cubic structures, observed with two kinds of topologies:
"bicontinuous" and "micellar".
A few phase diagrams of amphiphilic molecules in presence of water exhibit cubic
structures in two distinct concentration domains [ 10]. Some of them are observed in the
vicinity of the lamellar phase, they are now well characterized; they can be described as
being built by two interwoven labyrinths of one medium separated by a film without
self-intersection of the other, hence the characterization of their topology as
"bicontinuous". They are modeled using the toroidal surface in s3.
The other cubic phases are observed in the vicinity of the micellar phase [6]. Their
structures are built with finite micelles of one medium separated by a self-intersecting
film of the other, as shown in fig. 7 and 8. This structures are very close to the clathrate
structures.
To model such structures, we use, as a supporting surface for the film of
amphiphiles in s3' a surface separating s3 in two identical subspaces. This surface is a
great sphere s2, which can be considered as the exact equivalent in 3D of the equator
drawn on a 2D sphere and separating it in two hemispheres. As, on the 2D sphere, there
524

is a family of smaller circles parallel to the equator with their centres at the poles, there
is, in the 3D sphere s3, a family of smaller spheres parallel to s2 with their centres at
the poles of s3.

Figure 7: Periodic aggregation of slightly distorted 12-hedra and 14-hedra. On the right position
of these polyhedra in the cubic cell: they are at positions occupied by atoms in the ~tungsten
structure.

Figure 8: Periodic aggregation of slightly distorted 12-hedra and 16-hedra. they are at positions
occupied by atoms in the cubic Laves phase.

Thus, the great sphere s2 with two parallel spheres at equal distances on either
sides of it, can be considered as another representation without frustration of the the
periodic system of fluid films which is frustrated in the Euclidean space. The
periodicity in this space, which is the repetition of the films when moving along the
525
geodesics of the space normal to the films, in reproduced in the curved space by the
displacements along great circles of s3 normal to s2 moreover, as already said above,
the two subspaces separated by the great sphere are identical, as are the two space
separated by the film in the cell of the structure and, finally, the frustration is obviously
relaxed as the two spheres parallel to the great sphere have a smaller area than it.
Then a disclination prodedure would generate cells. An intermediate case could be
the {5,3,3} polytope formed by a packing of 120 dodecahedral cells. New disclinations
achieved a complete decurving but they change some dodecahedral cells into 14-hedra
or 16-hedra.

7. Conclusions

A common feature of all these systems is that, even with simple local rules, they present
a large set of, often complex, structural realizations.
Here, we have focus mainly on geometrical tools. We hope to convince the reader
that geometrical frustration is an unifying concept, which plays a crucial role in very
different fields of condensed matter, ranging from clusters and amorphous solids to
complex fluids.
The method of approach follows two steps. First, the constraint of perfect space-
filling is relaxed by allowing for space curvature. An ideal, unfrustrated, structure is
defined in this curved space. Then, specific distortions are applied to this ideal template
in order to embed it in the three dimensional Euclidean space. It allows to describe the
final structure as a mixture of ordered regions, where the local order is similar to that of
the template, and defects arising from the embedding. Among the possible defects,
disclinations will play an important role.
Complexity is not limited to purely theoretical examples. A student in
crystallography could already be quite puzzled by learning that, among possible
structures, some are relatively simple, like the f.c.c. structure for dense metals, but that,
even for pure monoatomic metals, rather intricate structures of a high complexity can
occur. As an example, manganese crystallizes in a structure with a unit cell containing
fifty eight atoms ! Here, the local rules, if the problem can be simplified to local
interactions, should explain such a structure. And indeed, the curved space approach
can shed new lights on these types of complex crystalline structures.
An important qualitative result on complex frustrated structures is that it is possible
to save one order of magnitude for their description at the microscopic level. Since it is
possible to define an "order" in these frustrated systems, it becomes possible to define
defects, whose characteristic scale depend on the degree of frustration. The complexity
of the structure is then encoded into the complexity of its defect network (which are
mainly disclination lines). One then meet:
- ideal systems, without defects and frustration.
- structures with periodic network of defects, as the large cell crystals.
- and finally amorphous materials, characterized by a disordered network of defects.
526

References

J.F. Sadoc, R. Mosseri, Frustration Geometrique, Editions Eyrolles (1997).


To appear in English: Cambridge University Press.
2 J.F. Sadoc, J. Non-Cryst Solids 44 (198I) I
3 H.M.S. Coxeter, Regular Polytopes, Dover (I973).
4 J.F. Sadoc, C.R. Acd. Sc. Paris 292 (198I) serie II 435.
5 J.F. Sadoc, N. Rivier, Phil.Mag. B 55 (1987) 537.
6 J. Charvolin , J.F. Sadoc, in Micelles, Membranes, Microemulsions and Mono/ayers,
W.M. Gelbart, A. Benshaul, D. Roux eds., Springer-Verlag (1994).
7 V. Luzzati, Biological Membranes, I (I968) 7I
8 J. Charvolin, J.F. Sadoc, J. Phys. (France) 48 (1987) I559 and 49 (I988) 521.
9 M. Clerc, A.M. Levelut, J.F. Sadoc, J. Phys. II (France), I (l99I) I263.
10 J. Charvolin, J.F. Sadoc, J. Phys. Chern., 92 (1988) 5787.
COMPUTER SIMULATIONS AND TESSELLATIONS OF
GRANULAR MATERIALS

L. OGER AND J.P. TROADEC


Groupe Matiere Condensee et Materiaux, Universit€ de Rennes
1, Campus de Beaulieu, F35042 Rennes Cedex
A. GERVOIS
Service de Physique Theorique, Direction des Sciences de la
Matiere, CEA/Saclay, F91191 Gif-sur Yvette Cedex
N.N. MEDVEDEV
Institute of Chemical Kinetics and Combustion
3./nstitutskja, 630090 Novosibirsk 90, Russia

Abstract

This chapter provides a brief review of some algorithms used to construct


model granular media (packings of spheres), together with a comparison of results
obtained on tessellations of those packings. The computational algorithms involve
several techniques, including: a Random Sequential Adsorption approach (RSA),
an Event-Driven model (ED), a Molecular Dynamics code (MD), a Smoothed
Particle Hydrodynamics approach (SPH) and the classical static algorithm for
dense packings of spheres. These techniques model static or dynamic interactions
between the particles and give different geometries even for the same mean solid
fraction. These numerical methods can apply to two or three-dimensional packings.
Different methods can generate full tessellation of the two and three-dimensional
spaces (a tessellation is a random geometrical foam). The classical Voronoi tessel-
lation is well known and deals only with points or the coordinates of the particle
centers. In this case, the particles have finite radii and generate a minimal exclu-
sion. area. The radical and "navigation" tessellations both account for particle size
and give the same result if the packing is monosize but give different results if the
particles are of different diameters.

1. Introduction

Granular media comprise distinct particles which move independently and


interact only at highly localized interparticle contacts. The problem is to
define exactly the contact and how to measure this contact. This problem
is especially hard in the three-dimensional space because it is impossible
527
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 521-546.
© 1999 Kluwer Academic Publishers.
528

to examine 3D structures. Only M.R.I. (Magnetic Resonance Imaging) or


optical matching techniques allow examination of a 3D packing, but even
in this case the definition of the contact and the neighbors of a given grain
is not so easy to obtain. Thus we use numerical approaches to analyse the
geometry of grain packing.
The numerical program defines input parameters and practical algo-
rithms to simulate a given mechanical, physical or geometrical interaction
and at each time step, identifies the coordinates of each particle. The code
simulates different physical processes such as: interactions between gas par-
ticles, fluid movement, hard sphere displacement, soft spheres interaction
mechanism. This lecture will analyse the similarities and differences of sev-
eral numerical methods for generating 2D and 3D monosize and polydis-
perse packings of discs and spheres and how the different tessellation algo-
rithms can deal with them.

2. Computer simulations

In this section, we shall describe mostly three-dimensional algorithms be-


cause the performances of the new generation of computers authorize the
creation and the analysis of a large amount of particles (several thousands).
Of course, all the approaches and techniques available for three-dimensional
studies are useful and easy to manage for a two-dimensional study.
For a 3D packing of monosize grains, the packing fractions are known to
be between zero and 0.7405. FCC and HCP permit the upper limit packing.
The lower limit is obtained by a generation of points inside a 3D space. This
distribution is generally referred as a Poisson distribution and is created
by choosing at random the three coordinates of the points inside a defined
3D box. For fractions between these two limits, several packing techniques
exist (Figure 1). The main interest here is in analysing the randomness of
a packing or the transition between a fluid-like and a solid-like packing.
We will describe several algorithms which permit a packing of spheres as
nearly disordered as possible.
The algorithms can be classified in two well-defined groups : the static
and the dynamic construction methods. In the static case, the grains are
placed at a given time step and therefore cannot move. The extension of
the Poisson distribution technique as a static model is the Random Se-
quential Adsorption model (RSA) where the particle is no longer a point.
Other static models are given by the classical methods used to build dense
disordered packings [1, 2]. For contrast, the dynamic approach uses the ini-
tial position of the grains as an input parameter, then short or long range
interactions between grains generate displacements of the particles and re-
organizations of the whole packing. The simple dynamic models described
529
I

Smoothed Particle Hyprodynamics

Event-Driven II Molecu1ar Dynamics

I Jodrey & Tory II Jullienl


1,. .. 1
I I I
Visscher &solsterfi

... ..
I I PoweU I
1Random Sequential Adsorption 1
..'I
I I I
Poi~son FdC-HCP
I
0 .35 .64 .74

Packing fraction
Figure 1. Schematic representation of the different kinds of simulation which can produce
a given packing fraction for a three-dimensional space. Between 0.64 and 0. 74 some
ordered areas exist in the packing.

here can be found in a more detailed version in the book "Molecular Dy-
namics Simulation, Elementary Methods" by J.M. Haile [3]. We especially
focus on a "Molecular Dynamics" model [4], a so-called Molecular Dynam-
ics code for granular media [5] and a computational Continuum Dynamics
technique named "Smoothed Particle Hydrodynamics" [6]. In the last part
of this chapter, we describe results using some of these techniques. In all
our simulations, we use approximately 16000 grains. Because of finite size
effects (periodic boundary conditions cannot be defined in all cases), we
consider only 12000 particles in the statistical analysis. Our samples are
large compared to the numerical packings of Clarke and Wiley [7].

2.1. STATIC ALGORITHMS

2.1.1. Random Sequential Adsorption model


The Random Sequential Adsorption algorithm is known to generate a dilute
packing [8, 9] in 2 or 3 dimensions. The building rule is the following :
the coordinates of the center of each grain (disc or sphere) are taken at
random in a given space (area or volume). The grain is definitively placed
if it does not overlap previously positioned grains, otherwise, its position
is rejected and another trial is performed. The process ends either when
no supplementary grain can be placed (jamming limit) or when a given
530

packing fraction is reached. This procedure does not provide very dense
packings, the jamming limits for monosize particles being close to 0.38 for
a three-dimensional packing of spheres and to 0.55 for a two-dimensional
packing of discs. If there is a binary or a polydisperse distribution of size,
the jamming limit can vary drastically according to the way the rejection
is made. After a bad trial, we can keep the previous choice of the radius or
choose a new one. If the size distribution is very large, it is obvious that,
as the packing density increases, it is more difficult to place a large grain
than a small one. So keeping the previous choice of the radius maintains
the original size distribution but decreases the jamming limit compared to
a monosize packing, otherwise the amount of small particles will increase
in the system. Another way of creating a packing with the RSA technique
is to calculate first the requested amount of spheres of each species and
then place all the big ones before placing the small ones. This procedure
creates another kind of disorder because the places of the small grains are
controlled by the already placed big grains.

2.1.2. Dense Packing algorithms


Few approaches have been developed in the last thirty years in order to
generate dense disordered packings of discs or spheres [1, 2, 10]. These
models follow some very basic ideas : the grains are placed according to a
"gravity" field and cannot move after placement. This technique leads to
an average of 4 contacts in 2D and of 6 in 3D. Differences remain in the
way the disorder is created or conserved.
Powell [1] defines a box in which growth of the packing is performed
in one direction and periodic boundary conditions are applied in the other
one(s). The packing grows layer by layer. A grain in a layer is randomly
chosen and then the possible neighbors supporting the new grain are chosen.
This operation is repeated several times and the new grain is positioned
in the lowest position. This algorithm gives a larger packing fraction than
the RSA algorithm (around 0.80 in 2D and between .57 and .62 in 3D for
monosize packings) but smaller than the maximum possible for disordered
grain packing (0.82 in 2D [11] and 0.64 in 3D [12]).
Visscher and Bolsterli [2] create a big box and choose at random the
coordinates of a point at the top of the box and let the particle fall down
from that point to the already placed grains in the box. The particle finds
the first contact in the packing and rolls down to obtain the other ones.
This algorithm also generates a dense homogeneous and disordered packing
with a packing fraction around 0.58 at 3D.
In the Bennett algorithm [10], the "applied field" is not a gravity (par-
allel orientation) but a central force (converging orientation). According to
some extra rules, the next particle can be placed as close as possible to the
531
original grain of the packing or its neighbors can be chosen at random. This
algorithm [13] creates a dense but not homogeneous packing: the packing
fraction decreases from the center to the outer part of the packing (11].

2.2. DYNAMIC ALGORITHMS AND LOCAL INTERACTIONS

The dynamic algorithms are more flexible but also less easy to understand
because of the diversity and the complexity of the local interactions. The
final positions of the particles depend strongly on the process which can be
collective or individual. The possible values of the final packing fraction also
depend on the process. If the collisions are assumed to be instantaneous,
only one collision has to be examined at a given time step. If the colli-
sions occur during longer time (typically 50 time steps), several particles
can interact simultaneously. The first case corresponds to "event-driven"
simulations and the second one to molecular dynamic simulations.

2.2.1. Event-Driven (ED) model


In event-driven simulations, the components of the system under analysis
(discs at 2D or spheres at 3D) evolve independently, unless an event takes
place. An event is a collision between two particles or the collision of one
particle with a wall : both situations are characterized by a sudden change
of particle momentum. Hence in ED simulation the time during which col-
liding particles are in contact is ideally zero (this is quite different from the
MD simulation described later in this paper). Numerous simulations are
performed for a 1D or 2D group of particles. In those simulations a simple
event-driven algorithm is used, which updates the whole system after each
event. Because of the small number of particles involved (N < 1000), the
procedure is successful. The situation changes in 3D where a large number
of particles (around 20000 particles) is used. A way to handle large N is
described by D. Lubachevsky (14]. The advantage of his improved algo-
rithm is that one does not have to update the state of the system after
each event. This is made possible through a double-buffering data struc-
ture where the "old" status (i.e. time, position, velocity, partner) as well
as the "new" status (i.e. new time, position, velocity, partner) of each par-
ticle is recorded. If an event happens, the new status gets the "old" one
and the subsequent "new" status has to be computed. This computation
is performed only for the particles involved in the collision, because only
their velocities changed. In the computation of the "new" status, the first
step is to find the presumable new colliding partner and to calculate the
"new" event time; the second step is to compute the positions and velocities
after this "new" event. A "new" status might be preempted several times
because of collisions of the partners with other components of the system.
532

To make the algorithm more efficient, it is possible to apply the so-called


"delayed update" method. This means to postpone examining and updat-
ing the position and the velocity of a particle until its next event. One can
also store the event times in an ordered heap tree which simplifies finding
the next event (Fig. 2).
In that case, a particle is represented by the (r, t) spatia-temporal co-
ordinates where r characterizes the qpatial coordinates. The lines schemat-
ically represent the trajectories of the particles. The circles represent posi-
tion and time stored in memory. The present time t is marked by a dotted
line. The next collision will concern the two particles i and j at the time
Tf = Tj with Tf = Ti + ti and Tj = Tj + ti. The two particles i and j are
moved to these new position and time and the new velocities are computed.
The time for the next collisions is updated and then the nearest collision
time is sought etc.

t
t.
1

Figu1·e 2. Schematic spatio-temporal representation of the event-driven simulation.

Practically, the system is built inside a periodic box in which N particles


are placed according to several possible rules. For dilute systems, the RSA
techniques described earlier or some dilation of a static disordered dense
packing can be used for the initial choice of the positions. For denser systems
(packing fraction larger than .55), the positions can be determined only by
a dilation of a dense disordered packing or a dilation of a FCC packing.
A random number generator gives the amplitude and the orientations of
533
the initial velocities of each particle. In order to improve the algorithm,
the complete box is divided in a series of small box and the localization
of the future two particle collision is made through one small box and its
neighbors. The typical running time for the simulation is taken according to
the number of particles inside the box and is of the order of 20,000 collisions
per particle.

2.2.2. Molecular dynamics model


In the MD model, the particles are modeled as spheres and a "soft-particle"
approach in which each sphere can have multiple contacts that can persist
for extended durations is used. The size of the time step is chosen so that
about 50 time steps elapse over a typical "rapid" collision. Both normal
and tangential forces develop at the contact between two grains. When
compressive forces only are allowed, the simulations represent dry, non-
cohesive particle assemblies. By allowing tensile forces at the contacts, wet
cohesive particles can be modeled. No long range interactions are included
in this model.
For the sake of simplicity, the algorithm is explained for a 2D system
of discs, with a formulation similar to that used by Savage [15]. The ith
particle is characterized by its radius Ri, the position of its center (xi, Yi)
and the angular rotation Oi around its center. When cohesive forces are
neglected, interparticle forces exist only when two particles overlap (Fig. 3)
and the normal and tangential contact forces increase as the centers of the
particles approach each other.

(x;,Yi,S;) I I
0
Figure 3. Schematic representation of the contact between two particles.

The normal force Fn at the contact is modeled as viscoelastic. It consists


of an elastic contribution (a linear spring) and viscous damping contribution
534

(a linear dashpot) described as follows :

compression Fn = Kn8- bnVn for 8 =a -lri- rjl > 0 (1)


tension Fn 0 for 8 < 0

where Kn is the spring constant for normal forces, 8 is the relative normal
displacement between the centers of the two particles in contact (Fig. 3),
a is the sum of the two particle radii, ri - rj is the distance vector between
the two particles i and j, Vn is the relative normal velocity, and bn is the
dashpot constant for normal forces.
The force in the tangential direction is also modeled as a viscoelastic
one; a linear spring and a linear dashpot are used to generate a tangential
contact force as follows

(2)
where Kt is the spring constant for the tangential forces, 8t is the relative
lateral displacement (Fig. 3) during all the duration of the contact, Vt is the
relative tangential velocity, and bt is the dash pot constant for the tangential
forces. The tangential force, Ft is also limited to a maximum value which
is chosen according to a Coulomb friction law which generates the slipping
effect
(3)
where p; is the coefficient of friction. The tangential force acts in a direction
opposite to that of the relative tangential velocity Vt. Each particle is in
static equilibrium under the action of contact and external forces. At two
dimensions, the constitutive mechanical equations of the ith particle are:

"'~i pj,i
L...J=l X
"'.::' pj,i (4)
L...J=l. y
R·1 "'.::'
L...J=l
(Fj,inj,i
X y
- pj,inj,i)
y X

where nj,i is the normal vector oriented from particle j to particle i, ci is


the number of contacts on the ith particle, Fi•i and Fg•i are the x and y
components of the jth contact force acting on the ith particle and f~ and
J; are the x and y components respectively of the external forces on the
ith particle and TJ
is the torque acting on the angular component.
The x and y coordinates, the Vx and vy velocity components, the rota-
tion angle () and the angular velocity w of the ith particle at time step N
are obtained using the Verlet algorithm [16]. The coordinates are obtained
using the informations collected at time step N - 1. The velocities and
accelerations are calculated for the time step N- 1/2 then the equilibrium
535
equations (4) are applied. After this operation a final update is performed
to end the time step N.
It is convenient to cast the governing Eqs. {1) and (2) in dimensionless
form and perform the computations based upon these dimensionless equa-
tions [15]. It is straightforward to revert back to physical variables if desired,
but in using the dimensionless form of the equations there is the advantage
of being able to reach results for different physical scales easily (fine pow-
ders of 10 {tm up to ice floes of 100 km if the appropriate effective spring
constant and the time scale are correctly defined for one collision). Hence,
all lengths are non-dimensionalized by D, the diameter of the largest parti-
cles used in the computations. Time is non-dimensionalized by dividing by
JM/ Kn, where M is the mass of the largest particle and Kn the effective
spring constant used in Eq. (1). Velocities are non-dimensionalized by di-
viding by D/JM/Kn. Thus, the following dimensionless time and spatial
coordinates are introduced

--- {J(:x y
(t, x, y) = (ty M' D' D). (5)

The dimensionless velocity components and the dimensionless angular ve-


locity are defined as

(6)

By using Eqs. (1) and (3) with the dimensionless variables (5) and (6)
we can redefined all the equations of the system.

2.3. DYNAMIC ALGORITHMS AND LONG-RANGE INTERACTIONS

2.3.1. Jodrey-Tory or Jullien algorithms


Another well known dynamic algorithm is based on the classical molecular
dynamics developed by Jodrey and Tory [4]. A large number of points are
generated randomly inside a 3D cube with periodic boundary conditions.
As defined by Jodrey and Tory [4] : "Each point is the center of an inner
and an outer sphere. The inner diameter is set, after each iteration, to the
minim·um center-to-center distance between any two spheres and defines the
true density. The outer diameter is set initially to a very high value dealing
with a density close to 1. The algorithm eliminates overlaps while slowly
reducing the outer diameter. Thus the two diameters approach each other
and the eventual coincidence of the true and nominal densities terminates
the procedure".
The nominal density is defined according to the fixed size of the box and
the number of spheres N with a diameter equal to the outer one (nominal
536

diameter) while the true density is defined from the inner diameter. Periodic
boundary conditions are applied on the box in order to avoid some finite
size effects.
In the original version of Jodrey and Tory [4], the growing process is
obtained according to the equation :

(7)

where A(i) = d(i) / d(o) is the ratio between the initial nominal diameter d(o)
and the diameter d(i) obtained at the iteration step i, the term j is defined
as the integer part of -log10(flry(i)), where Llry is the difference between the
nominal and the true packing fraction. The parameter k characterizes the
rate of contraction of the ensemble; k is of the order of 10- 2 •
A small modification of this algorithm was introduced by J ullien et al
[17]. The new growing process is obtained by

(8)

where ck and c~ correspond to the packing fractions for the outer and
inner diameters. The rate "R" and the exponent a are two inputs of the
algorithm. The term dk is the_mean initial distance between two spheres
using the outer diameter, and d:J 1 is the new calculated one. The exponent
a is always equal to 0.33 in the Jullien simulations. For R ranging between
0.1 and 0.00001, the final packing fraction reaches the steady state ranging
between 0.35 and 0.64.

2.3.2. Smoothed Particle Hydrodynamics algorithms


Smoothed Particle Hydrodynamics (SPH) is a technique for problem solv-
ing continuum dynamics. SPH was originally applied in astrophysics and
cosmology to study unbounded systems [18, 19]. Since 1991, SPH has been
extended to treat the dynamic response of solids and the method has found
many new and interesting applications. For instance, various problems char-
acterized by an elastic-plastic constitutive behavior have been solved with
SPH [20, 21, 22, 23]. Takeda et a]. [24] have applied SPH to the simulation
of viscous flows. Gutfraind and Savage [25] have implemented a Mohr-
Coulomb type of rheology in the framework of SPH, to study the ice be-
havior modeled as an assembly of disks on the water surface. This method
has become a useful tool for numerical simulation of granular flows, and for
handling large deformations and free surface problems.
SPH is a Lagrangian technique in which the continuum flow is modelled
by the motion of a collection of point masses. Field properties such as
537
velocities and stresses are evaluated at each particle position by summing
over the properties of neighbor particles.
The foundation of SPH is interpolation theory. The conservation laws
of continuum dynamics are transformed into integral equations through
the use of an interpolation function that gives the kernel estimates of the
field variables at a point. Computationally, information is known only at
discrete points so that the integrals are evaluated as sums over neighboring
particles.
For our purpose, we can define a simple assembly of particles flowing
inside a box with periodic boundary conditions. We can control the local
densities and the interactions between the points. The classical Lagrangian
partial differential equations for the flow are given by the conservation
equations (continuity and momentum).

dp
dt = -p'\l ·v' {9)
and
dv VP
(10)
dt p
with p the density, P the pressure and v the velocity. In SPH approach,
any field variable in the space can be expressed

f(r) = j f(r') 8(r- r') dr', {11)

and approximated as

(f(r)) = j f(r') W(r- r', h) dr', (12)

where W(u, h) is an interpolating kernel with the properties

j W(u,h)du= 1, (13)

and
lim W(u, h)= 8(u). (14)
h-+0

Monaghan [19] defined three types of kernel: a cubic-B spline, a gaussian


and a super-gaussian one which is chosen for some stability problems.
For a fluid of density p(r), the right-hand side of Eq. {12) can be ex-
pressed as
j [f(r')/p(r1J W(r- r', h) p(r') dr'. (15)
538
In order to evaluate the integral (15), the space is divided into volume
elements having a mass mk, and the integral is approximated as
N
(f(r) ) = '""' !k W(r- rk, h),
L....t mk- (16)
k=l Pk

where fk = f(rk)· The derivatives are obtained by differentiating the inter-


polation kernel which is analytic. By these means we avoid the use of grids
and finite differences. For example, the kernel estimate of V' f is

(V' f(r)) :::= j W(r- r') V' f(r')dr'. (17)

Integrating by parts and assuming that W approaches zero fast enough so


that the surface terms vanish, one obtains

(V f(r)) :::= j V'W(r- r', h) f(r') dr', (18)

where V'W(r- r', h) is the gradient with respect to r. The integral (18) is
evaluated using the approach of (15) and (16)

(19)

where Vi means the gradient with respect to the coordinates of particle i


and Wik::: W(ri- rk, h).
The terms of equations (9) and (10) are multiplied by the kernel and
integrated. Derivatives of the field variables are transformed as was done in
(18) and the integrals are evaluated as in (16) and (19). The density can be
evaluated either by using (16) or the continuity equation but the continuity
equation is usually preferred. The kernel estimate of the continuity equation
is
(20)

Eq. (20) is evaluated by using the method of (16)

(21)

With (21) the density varies only when there is relative movement between
particles.
539

For the case of inviscid fluids [19] the pressure gradient terms that ap-
pear in the momentum equation lead to the following form of the SPH

dv=
- i - LN mk
(-Pk+ P;,- ) Y'iWik· (22)
dt k=l
Pk2 p·2
I

All the basic equations are well defined at the points in this meshless
method, and, by iterating with the time, we can simulate any flow of discs
or spheres. The group working at Los Alamos National Laboratory uses a
SPH code named SPHINX which deals with one million spheres.

3. Tessellations and analysis of packings


As discussed above, different algorithms have been introduced in order to
build packings. We compare the results of those algorithms to quantities
relative to the cells of the grains defined by simple spatial tessellations.
We will first recall how such tessellations may be determined, derive some
related results in the monosize case using algorithms presented in Sec. 2
and compare them.

3.1. TESSELLATIONS
A basic issue related to sphere or disc packings is the realization and geo-
metrical characterization of disordered assemblies - mainly in the dense
case. The coordination number is fundamental for transport properties
(conductivity, transmission of stress, .. ) and is a natural quantity in dense
assemblies. This property appears to be meaningless in dilute assemblies
(suspensions); the notion of neighbor, then, becomes more appropriate. A
convenient tool, at least for properties in the grain space, which accounts
for both the hindrance of the grains and their local behavior consists in
building their Voronoi cells, and analysing the statistical and topological
properties of the cells in place of the position of the centers of the spheres.
The Voronoi tessellation was introduced by Finney [26] for monosize pack-
ings of spheres (or discs) as a generalization of the Voronoi tessellation for
points [27]. The influence zone of two spherical grains is limited by their
bisecting plane and the Voronoi cell of a grain is the smallest convex poly-
hedron made with the bisecting planes. This cell completely contains the
grain and only that grain. Two grains are (first) neighbors if their cells
have one common face. It is possible to build without ambiguity a com-
plete hierarchy of neighbors at any packing fraction. The Voronoi cell may
be compared to the corresponding Wigner-Seitz cell in the case of ordered
arrays. Of course, touching spheres are neighbors in this sense. The Voronoi
tessellation completely fills the space, and the volume of the cell is a mea-
540

sure of the hindrance of the grain. Statistical and topological properties of


the ensemble of cells suggest the degree of correlations and disorder of the
corresponding dense or dilute packing. In the case of binary or polydisperse
assemblies of spheres, the Voronoi tessellation is no longer adequate, be-
cause in dense packings, the cell can "cut" the larger grain and touching
spheres may not be neighbors in the Voronoi sense. Several generalizations
have been proposed, which all reduce to the usual tessellation in the mono-
size case (see Fig. 4).

a)

b)

c)

Figure 4. Schematic representation of the different Voronoi tessellations. a) Classical


Voronoi tessellation b) Radical Voronoi tessellation c) Navigation map tessellation

The simplest generalization is the radical tessellation [28], with the bi-
secting plane being replaced by the radical plane (all the points in the radi-
cal plane have the same tangency length- or power- for the two spheres) (Fig.
4b). The definition may seem somewhat artificial, but the following features
of the Voronoi tessellation are maintained:
the cells are convex polyhedrons with planar faces, each one containing
one grain,
two touching spheres have a common face (the tangent plane) and thus
may be considered as neighbors,
the incidence properties still hold : a face is generically common to two
cells, an edge to three cells and a vertex to four cells.
541

In general, big grains have larger cells than smaller ones, which allows for es-
timating relative hindrance. Another possible generalization ("navigation"
tessellation) consists in limiting the influence zone of one grain by the points
at equal distance of the borders of the particles [29](Fig. 4c). In 2D, the
separation curve is a half hyperbola and in 3D the separation surface is
an hyperboloid. This definition is more satisfactory but it is less tractable.
This technique can be applied to analyse the connectivity structure of 2D
or 3D packings as described by Lemaitre in the chapter XXI in this volume.

3.2. COMPARISON OF RESULTS FOR DIFFERENT MONOSIZE


PACKINGS

It is well known that in 2D the average number of edges of a cell is a


constant [(n) = 6], for any algorithm and any packing fraction. This identity
is imposed by the strong topological constraints (Euler identity). This is
not true in 3D and the corresponding quantity, i.e. the average number of
faces per cell, say (!), is no longer a constant [30]. Figure 5 represents a
general overview of the variation of the average number of faces defined by
Voronoi tessellation as a function of the packing fraction C for the methods
of generation of packings presented in Section 2 (static algorithms, RSA and
Powell, and dynamical algorithms, Event-Driven and Jodrey-Tory).
We note the following:
- (!) is a decreasing function of C. When C = 0, the average number
of neighbors 15.5 is close to the Random Voronoi Poisson (RVP) or
Johnson-Mehl models [31] values. At low packing fractions (C < 0.2),
all algorithms yield similar results.
(!) is always larger than 14. Note that it is 12 in the ordered lattices
(FCC and HCP). There is no convergence to 12 when the packing
fraction is close to 0. 74. Actually, this discrepancy disappears when
the degeneracy on the vertices of an ordered lattice is removed by
allowing an infinitesimal error in the position of the particles in the
FCC or HCP structures [34].
- For compact assemblies and at a given packing fraction C, (!)depends
strongly on the algorithm. For instance, for a packing fraction of the
order of 0.58-0.60, not very far from the densest disordered packing
(C = 0.64), (!) = 14.5 for a Powell packing, while ED processes lead
to a value close to 14. Jodrey-Tory yields an intermediate value. The
differences probably come from the possibility of local arrangements
of the particles as given by the different algorithms.
The Event-Driven algorithm is based on simultaneous displacements of
all the particles and allows some local arrangements because the particles
542

.'
IS.l

X Powel ' f& I Event~nven ' Jodrey· T01'1 • JUitn


JS.l I

t I I

'I •
IS ' I t

'•
I I I &a
& l
Ill I t
A I I &
v

.
.I • &

ll.6 II •
l
I
I ' X

IH
II

.
ll.l
••
Ia

ll I '• I

u OJ OJ u 15 t• l7
PICIJing FtiC1ion

Figure 5. Mean number of faces for a Voronoi tessellation of different mode of packings
(R.S.A., Dense Disorder Packing, Event-driven, Molecular Dynamics, Jodrey-Tory).

repulse one another and are not in contact. The packing behaves like a
denser one, and (f) is weaker. Local arrangements are more difficult with
the Jodrey-Tory algorithm where only two particles move at the same time.
They do not exist in packings built with the Powell algorithm, which, for
the same packing fraction, have the largest value of (f).
For various model we can also compares the width JL 2 of the distribution
of values of the number of faces of the polyhedra, defined as JL 2 = (j2)- (!) 2
[33), to characterize the structure. In the figure 6, we have plotted JL2 for the
algorithms used in the figure 5. For a given packing fraction, JL2 is nearly
independent of packing building rule. However, at high packing fraction, one
can see that a larger distribution width corresponds with a larger average
value. As we expect there is more order in the Event-Driven packings.

3.3. POSSIBLE EXTENSIONS OF THE TESSELLATION ANALYSIS

3.3.1. Binary mixtures


We have extended our analysis of the disordered packings to binary mix-
tures (Fig 7) . All the dynamical simulation procedures deal with some
weight or momentum equations, so it is not possible to avoid some seg-
regation effects when the thermal activation occurs. This is the reason why
543

Il r-----------------------------------------~

10

••

"'
~ . • • t
•••
; ' It •
··1 ..
' I • I •
•• •• •\.1
I
I

a-'..,
....... I

OJ QJ Ql l! 1.7
PIQJng FIICllon

Figure 6. variation of J..l2 for a Voronoi tessellation of different mode of packings (R.S.A.,
Dense Disorder Packing, Event-driven, Molecular Dynamics, J odrey-Tory).

only packings made with the RSA algorithm or with a Powell's algorithm
are studied. As already observed for monosize packings of spheres, the mean
value of (!) decreases with the increase of the packing fraction C when we
keep constant the relative percentage of small spheres. When we maintain
constant the packing fraction and change the percentage of small spheres
we observed that (!) is nearly constant but the mean value of f for each
species changes strongly(Fig 7) . For a radii ratio equal 3, when the amount
of small spheres increases from 0 to 100%, the mean value Us) for the
small spheres grows from 6 to 14.7 for a packing fraction C = 0.58, at the
same time, the mean value (h) for the large spheres increases from 14.9 to
25. This can be easily explained by the local arrangement of small spheres
which can be placed around a large sphere [35) .
Comparison of monosize packings of spheres and monosize packing of
discs suggests that results obtained for a binary packing of disks can be
extrapolated to the binary mixtures of spheres. However the mean value of
the number of faces changes with the packing fraction and generates several
problems.

3.3.2. Navigation map


As described in section 2, the tessellation of the space can provide inter-
esting information concerning the structure of the packing and it can also
give some information about the porous space. This can be very useful for
544

45~------------------------------------~
40
-Large
JS
JO - =-total
II
... 25 -small
v
20
IS 1_....~~~~~~--

J
---.::-=-=-=---=---=--=-~---=-

1:1:::::::::::::::::::::::::::::::=:=-__
0 0.2 0.4 0.6 0.8
Numerical concentration of small spheres

Figure 7. Mean number of faces for a Voronoi tessellation of different percentage of


small spheres for a binary mixture of ratio 3.

the simulation of transport phenomena inside a porous space. If a small


sphere percolates through the porous space made with an assembly of large
spheres, it has to follow one of the possible paths available inside the struc-
ture. As the "navigation map" tessellation is constructed according to be
at equal distance of the outer part of the grains, this is exactly the more
probable place for the small particle to pass in order to avoid capture.
This property has been described in more detail by Lemaitre et al. in this
volume.

4. Conclusions

We have briefly described some algorithms used to construct model granular


media in the two-d imensional or three-dimensional cases. These computa-
tional algorithms involve techniques of dynamic (ED, MD, SPH, etc ... . )
and static (RSA, Powell, ... ) models. These techniques give different ge-
ometrical arrangements and properties for even the same mean packing
fraction.
We have also described different techniques which exist in order to generate
a full tessellation of the two-dimensional and three-dimensional spaces. The
classical Voronoi tessellation is well known and deals only with points or
the coordinates of the particle centers. The radical and "navigation" tessel-
lations take into account the size of the particles and give the same result if
the packing is monosize but different results if the particles are of different
diameters.
This work was partially supported by the EUHCM European program
"FOAMPHYS" under No EBRCHRXCT940542.
545
References
1. M.J. Powell, Phys. Rev. B 20 4194 (1979).
2. W.H. Visscher and H. Bolsterli, Nature 239 504 (1972).
3. J.M. Haile, Molecular Dynamics Simulation, Elementary Methods, John Wiley and
Sons Inc. (New York) (1992).
4. W.S. Jodrey and E.M. Tory, Phys. Rev. A 32 2347 (1985).
5. S. Luding, Phys Rev E 52 4442 (1995).
6. R.A. Gingold and J. J. Monaghan, Mon. Not. R. Astron. Soc. 171 375-389 (1977).
7. A.S. Clarke and J.D. Wiley, Phys. Rev. D 35 7350 (1987).
8. J. Feder, J. Theor. Bioi. 87 237 (1980).
9. J. Feder and I. Giaever, J. Colloid. Interlace Sc. 78 144 (1980).
10. C.H. Bennett, J. Appl. Phys 43 2727 (1972).
11. D. Bideau, A. Gervois, L. Oger and J.P. Troadec, J. Physique France 471697 (1986).
P J.L. Finney, Nature 266 309 (1974).
13. D. Bideau, J.P. Troadec and L. Oger, C.R. Acad. Sci. 297 II 319 (1983).
14. D. Lubachevsky, J. Comput. Phys. 94 255 (1991).
15. S.B. Savage in Mobile Particulate Systems, E. Guazzelli and L. Oger eds, NATO-ASI
Series, Kluwer Academic Press, 308 ( 1995).
16. L. Verlet, Phys. Rev. 159 98 (1967).
17. R. Jullien, P. Jund, D. Caprion and D Quitman, Phys. Rev. E 54 6035 (1996).
18. L.B. Lucy, The Astronomical Journal 82 1013-1024 (1977).
19. J.J. Monaghan, Ann. Rev. Astron. Astrophys. 30 543-574 (1992).
20. L.D. Libersky and A.G. Petschek, in The Next Free Lagrange Conference, Jackson
Hole, WY 1990, H.E. Trease, J.W. Fritts and W.P. Crowley eds, Springer-Verlag,
242-257 (1992).
21. L.D. Libersky, A.G. Petschek, T.C. Carney, J.R. Hipp and F.A. Allahdadi, J. Com-
put. Phys. 109 67-75 (1993).
22. A.G. Petschek and L.D. Libersky, J. Comput. Phys. 109 76-83 (1993).
23. W. Benz and E. Asphaug, Icarus 107 98-116 (1994).
24. H. Takeda, S.M. Miyama and M. Sekeiya, Progr. Theor. Phys. 92 939-959 (1994).
25. R. Gutfraind and S.B. Savage, J. Geophys. Res. 102 12647-12661 (1997).
26. J.L. Finney, Proc. Roy. Soc. Lond. A 319 479 (1970).
27. B.N. Boots, Metallography 15 53 (1982).
28. B.J. Gellatly and J.L. Finney, J. Non Crystalline Solids 50 313 (1982).
29. N.N. Medvedev, V.P. Voloshin and Y.l. Naberukin, J. Phys. A: Math. Gen. 21 1247
(1988).
30. L. Oger, A. Gervois, J.P. Troadec and N. Rivier, Phil. Mag. B 74 177 (1996).
31. J.L. Meijering, Philips Res. Rep. 8 270 (1953).
32. C. Annie, J.P. Troadec, A. Gervois, J. Lemaitre, M. Ammi and L. Oger, J. Physique
I France 4 115 (1994).
33. J. Lemaitre, A. Gervois, J.P. Troadec, N. Rivier, M. Ammi, L. Oger and D. Bideau,
Phil. Mag. B 67 347 (1993).
34. J.P. Troadec, A. Gervois and L. Oger, "Statistics of Voronoi cells of slightly per-
turbed face-centered cubic and hexagonal closed-packed lattices" submitted to Eu-
rophysics Letters.
35. J.P. Troadec, A. Gervois, D. Bideau and L. Oger, J. Phys. C 20 993 (1987).
546
STUDY OF EXPERIMENTAL AND SIMULATED EVOLUTIONS OF 2D
FOAMS

V. PARFAIT-PIGNOL*, R. DELANNAY*\ A. MOCELLIN**


* Laboratoire d'Energetique et de Mecanique Theorique et Appliquee,
URM 7563, Ecole des Mines, Pare de Saurupt, 54042 Nancy Cedex,
France
** Laboratoire de Science et Genie des Materiaux Metalliques, UMR
7584, Ecole des Mines, Pare de Saurupt, 54042 Nancy Cedex, France

Abstract: We study the effect of parameters such as the system size on the evolution of soap
froth and cellular structures. Experiments show a significant effect of the size and the
composition of the foaming liquid. Simulations based on Telley's model (Laguerre
tessellation) are in good agreement with experiments and confirm the effect of size. We give
here a detailed description of the experiments and the image treatment (including bias
correction) as well as the basic concepts of the model for simulating 2D foam coarsening.

1. Introduction

Suppose that you have filled a Hele-Shaw cell with soap bubbles. If you are fortunate
enough, you may obtain an honeycomb-like structure where all bubbles have almost the
same size. Only a small number of them are not six sided bubbles and appear to be like
defects in the ordered structure. From Von Neumann's law, we know that the area of n-
sided bubbles changes with time, at a rate proportional to (n-6) ([1], see also chapters rf
J. Stavans and N. Rivier). It implies that hexagonal cells don't change their areas but
defects do; so that some large bubbles appear after some time, and other bubbles shrink
and disappear. When a bubble disappears, the number of sides of its neighbours

1 Permanent address : Groupe Matiere condensee et Materiaux, U.M.R. C6626, Universite


de Rennes /, Avenue du General Leclerc, 35042 Rennes Cedex, France.
547
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 547-562.
© 1999 Kluwer Academic Publishers.
548

changes, so that the neighbours begin themselves to grow or to shrink. The area affected
by the initial defect can be seen as a disordered cluster. These clusters grow as a
dispersed phase in the continuous ordered (hexagons) phase. At some stage, the clusters
percolate and the disordered phase becomes continuous. The ordered phase, now
dispersed, shrinks and finally disappears.
This evolution of foam has attracted considerable attention within the last decades
(see [1,2] for recent reviews). The transient regime described above, is generally
followed by an asymptotic one where the dimensionless statistical characteristics reach
some limiting distribution. During this regime the average bubble area grows linearly
with time. The statistical characteristics generally considered useful include :
*Distribution of the number n of sides of the cells (p(n), mean <n>, variance J.L2).
* Two-cell correlations (m(n) is the mean number of sides of the neighbours of n-
sided cells).
*Distribution of area a of the cells (P(a)/<a>, mean area <a>).
For structures with a large number of cells, <n> is not far from its theoretical value
6 (required by Euler's law). The variance 112 is a convenient measure of the topological
disorder of the structure. It never exceeds 5 in natural structures. The average total
number of sides of cells adjacent ton cells, nm(n), varies linearly with n (Aboav-Weaire
law).
Glazier and Stavans [3,4,5] studied the asymptotic regime obtained within a
rectangular box (of the size of a usual sheet of paper) for a soap solution of water, liquid
detergent and glycerol. They assumed that the asymptotic distributions obtained were
independent of the size (and shape) of the box, and of the composition of the solution.
We here investigate these assumptions by means of experiments and simulations.

2. Experimental devices

Its constituents are represented on Figure I (see also [6]). A luminous table, placed
under the "box" filled with the 2D soap froth, illuminates the structure without
increasing its temperature. The CCD video camera is connected to a video recorder and
an Image Analysis Device and analysed by a software (VISILOG). This camera can
move vertically to change the size of the structure under investigation.
Two types of foaming liquid have been used to produce the froth. These liquids are
composed of a foaming agent, some glycerol to stabilise the froth and some distilled
water. The proportions are the following :
* (i): 10% of dish-washing liquid, 5% of glycerol and 85% of distilled water. This
composition was also used by Glazier et Stavans [3],
* (ii) : 2% (mass fraction) of Sodium Dodecyle Sulphate powder (SDS) as foaming
agent, 0.05% (volume fraction) of glycerol and the rest is distilled water.
The froths have been generated inside three types of experimental containers.
549

v;dt
Images Analysis Software (VISILOG)
reconl<>

Liquid

Luminous table

Figure 1. Experimental device.

The first one has the same characteristics as those used by Glazier and Stavans [3].
It consists of two rectangular Plexiglas plates (23 em x 14 em) separated by a 3 rom-
thick spacer.
The second one is larger, and is composed of two rectangular glass plates (30 em x
40 em). The spacer is only 2 mm thick. With this box, we can study larger structures
and obtain better statistics.
The third one (Figure 2) is made up of two circular glass plates (diameter 40 em)
separated by 2 mm. This box has been built to allow the drainage of the froths. Indeed,
during ageing the number of cells of the froth decreases and the mean area of the cells
increases. If the total amount of liquid remains constant, the edges thickness increases.
To overcome this problem, a very small gap is left under the top plate. The liquid is
sucked into the gap by capillary pressure. It can be collected in a groove and extracted
from it by a syringe (Figure 2).
Figure 3 shows the typical evolution of Plateau borders' thickness during the ageing
of a soap froth. The vertical lines indicate times of manual extraction of fluid by the
syringe. The thickness increases during evolution and decreases after extraction. We can
see that drainage doesn't keep the edges thickness exactly constant, but it avoids large
variations.
550

groove

0-ring
Drainage system ... . ······• ...

Figure 2. vertical cut of the circular box, drainage system (zoom).

In order to obtain the froth the first step is to fill the box (the space between the
plates) with the foaming liquid. Then, we tilt the box and we inject some air bubbles at
the bottom. The initial structure depends on the size of these air bubbles. When the rate
of air injection is constant, the structure is composed of cells having all the same size.
These cells have 6 edges, the structure is initially ordered. On the contrary, when the
rate of air injection varies, the air bubbles give birth to cells with various number of
edges : the initial structure is disordered.

11,34 I I

0)2 r-

. •
.. •
f-

- •- .• -· •,•
0,3

• -
U,2S - I• I I

11.26 -• 1-

0.24 r-

I l
11,22
jll 1110 llO 2lNJ 2lll 300

time (h)

Figure 3. Evolution of Plateau borders' thickness (mm) during drainage.


551
Once the box is filled with bubbles, it is placed horizontally under the camera and
sealed. We can then study the experimental ageing of these 2D soap froths. (some 3D
cells can be present in the beginning but they quickly disappear).

3. Image treatment

Every three hours, an image is selected, digitised and recorded. The working domain is
the whole box, but there are generally too many cells in the box to perform efficient
analysis on it. The working domain is not exhaustively investigated, but is sampled.
The image covers a sampling zone Z in which the analysis is actually performed. A first
treatment is made automatically with the help of the Image Analysis Software Visilog.
The aim of this treatment is the location of the edges and vertices of the structure. A
threshold operation is first applied to the original grey level image. It leads to a binary
image which is itself submitted to morphological operations such as "skeleton",
"dilation" or "erosion" so that the edges thickness is reduced to one pixel.
This treatment is generally efficient and fast but generates some defects (non detected
edges, 4 sided cells reduced to one vertex). A manual adjustment is necessary to correct
them.
The CRYSTAL software [7] is then used to measure the cells size and shape and
their correlations. This software reconstructs a cellular structure after removing cells and
edges cut by the boundaries of Z, and straightening the edges.
Since a larger cell is more likely to be cut by the boundary of Z and thus rejected,
than a smaller one, a bias correction based on the Miles and Lantuejoul's method [8]
has been applied to the data.
Let X be a cell and Rx the smallest rectangle of sides parallel to those of Z,
containing the cell. Let x be the centre of this rectangle. The X cell is then totally
included in Z if and only if the centre x belong to the zone Zx, where Zx is obtained by
the erosion of Z by Rx (Figure 4). The probability Px for X to be totally included in Z
is then proportional to Zx.
The method consists in assigning to each cell X a weight inversely proportional to
Px· Knowing the experimental value w* (X) of some measurements, we can obtain the
corrected mean value by :

L w·cx)
_ x area(Zx)
We= 1 (1)

~ area(Zx)
552

I Lin z
Zx

-t ...
LW
Ll rKf.~
;..-.,;
L2

Figure 4. bias correction : erosion of the sampling zone Z by the rectangle Rx.

4. Experimental results

Evolution ofthe soap froth (i) (dish-washing liquid) in the small rectangular box
reaches a stationary asymptotic regime. It confmns the results of Glazier and Stavans
[3,4,5] : the average area increases linearly with time and the distribution of reduced
areas becomes constant. The stationary distribution p(n) has a variance ll2 equal to 1.5 .
The solution (ii) (SDS) behaves differently : there is no evidence for an asymptotic
regime (but a continuous increase of ll2). Similar results have been obtained by Han.
[9]. These observations are confmned in the larger rectangular box (Figure 5). The soap
froth (ii) does not reach an asymptotic regime. The (i) does. However the limiting value
of J.L 2 is slightly but significantly larger : J.12 = 1. 7. Because more bubbles (- 2000) are
taken into account the statistics is better.

-·····--··············...!

...
I.Z

' o~--~~----~~~--~
~~~~~
~--~.~~--4~----=~--~-
tirne (b)
Figure 5. Time evolution of~t2 forSDS (full line), and dish-washing liquid (dashed line). Large box.

During evolution, the average cell area increases (as the number of bubbles
decreases) but at decreasing rate (non-linear behaviour). This is an effect of the
553
thickening of the Plateau borders: the excess fluid accumulates in the regions where the
films wet the glass plates (Plateau borders), the area available for gas diffusion (thin
films) decreases, evolution is slowed. Drainage in the circular box suppresses this
slowing down, as could be anticipated. The linearity of the mean area with time is
recovered.

5. Simulation models

We are discussing here models which account for the structure of the froth and its
evolution, thereby omitting those describing only the evolution of distributions
(Novikov, Hunderi, see [10]) or of individual cells [11-14]. Two different approaches
can be used for the simulation of the evolution : a statistical (Potts model [ 15]) or a
deterministic one [16,17]. The main problem with the deterministic approach lies with
the amount of necessary computer resources. The solution consists in using idealised
mappings of the cellular structure, local dynamics (but presumably realistic) and an
efficient way of identifying and performing the elementary topological transformations
(ETT) which occur during the coarsening.
The model presented here has been developed at EPFL by Telley, Liebling and
Mocellin [18,19]. It is based on weighted Voronoi" complexes. They furnish satisfactory
approximations of cellular structures, simple geometrical idealisations where topology
is induced by the defmition. The other ingredient is an equation of motion which applies
to the variables describing the structure. The model allows a determination of the place,
time and type ofETT. Moreover it is efficient, robust and extensible to 3D [20].
A Laguerre froth is a tessellation (see also chapters ofN. Pittet et al. and L. Oger et al)
generated from a set of circles. Let {Ci} be a set of circles. Each Ci generates a convex
cell L(Cj) defined as the set of the points which have a smaller power with respect to Cj
than with respect to any other circle. The set {L(Ci)} is a cell decomposition of the
plane. It is the Laguerre froth associated to {Ci}. Note that the power (Laguerre
distance) of a point to a circle can be negative if the point is inside the circle. This
enables us to construct a tessellation even if the circles are overlapping. The edge
separating two cells consists of the points of equal power to both circles. It is a straight
line. Obviously, if the two circles are intersecting, the edge separating their cells passes
through the two points of intersection. Even though it is often the case, the centre of a
generating circle does not necessary lie inside its associated cell (Figure 6(a)).
Contrary to the Voronor tessellation (which corresponds to a Laguerre froth with
equal radii), a circle can generate an empty cell (see Figure 6(a) : the small circle
embedded in a large one). Such a circle is not active, it could be removed from the
generating set without changing the structure.
554

(a) (b)
Figure6. (a) Laguerre froth (dashed lines) and weighted Delaunay triangulation (full straight lines), (b)
classical Delaunay triangulation (in grey : the modified elements).

A dual structure named weighted Delaunay triangulation can be associated to the


Laguerre froth (Figure 6). Its edges join the centres of active neighbouring circles. The
weighted Delaunay triangulation T is not identical to the classical Delaunay
triangulation Tv associated to the Voronoi' tessellation. The centres of generating circles
corresponding to empty cells are vertices of Tv but not ofT. Moreover, according to the
value of the radii, some edges ofT can be switched in Tv (see Figure 6(b)).
To build the Laguerre complexes in a plane 1t, Telley has used a parabolic scheme :
to each circle C we associate a paraboloid 'I' (see Figure 7).

Figure 7. Associated paraboloid

The paraboloid is then defmed by :


'l'(q) = {x,y,z)E 9\ 3 !(x-xS +{y- Yc J+(z-r 2)= 0]
where (xc,Yc) are the co-ordinates of the centre ofC and r is its radius. Each paraboloid
can be specified by the co-ordinates (x0 y0 r2) of its apex q.
555

In a simple way, the weighted Laguerre complexes are defined as the projection on rr
of the intersection of the various paraboloids defined from the active generating circles
[ 18] (see Figure 8). The Delaunay triangulation can be defined in a similar way [ 18].

1t

Figure 8. The« mountain chain» of paraboloids, q 5 lies below the« horizon» (empty cell).

During coarsening, the paraboloids can move up and down and sideways. When a
paraboloid is pushed below the horizon, the corresponding cell disappears (topological
transformation T2). It is also easy to produce a Tl transformation (neighbour switching)
between 4 cells. ETT are therefore naturally induced by the motion of the generating
circles.
Having recalled the essentials of Telley's model, we will now spell out the
algorithm of the program associated with it.

5.1 DEFINITION OF THE INITIAL STRUCTURE

The complexes are defined on a torus with periodic boundary conditions to avoid
problems with borders. A set of circles {Cj} is chosen either at random or in a
deterministic way and determines the initial structure. The weighted Delaunay
triangulation is then built in the following manner.
Let {Ci, Cj, Ck, C1}, be four generating circles, (xi,Yi) being the co-ordinates of the
centre Pi of Ci and ri its radius. Denote by Aijkl the determinant :
X; Y; r/ -(xt + Yl) I
x YJ· r2 -(x2 + y2)
A;ikl = 1 1 .! .!
" xk Yk rk2 -(x~ + yi}
x, Yt r/ -(xi + Yl)
LetT be a triangulation whose vertices form a subset of {Ci}. T is the weighted
Delaunay triangulation associated to {Ci} if Aijkl > 0 for every edge Tik ofT and for
every non active circle Ci (see Figure 9 for notations).
556

(b)

Figure 9. (a) An edge Tik of the triangulation T, Pi (resp. Pk)is the origin (resp. the destination) ofTik and
Pj (resp. PI) is the vertex ofT on the left (resp. on the right) ofTij. (b) Pi is the center of a non active circle
Ci, (Pj, Pk, PJ) is the clockwise oriented triangle ofT where Pi lies.

This condition is local and easy to implement numerically. In practice, the classical
Delaunay triangulation is constructed first, the preceding criterion is then applied to
switch the required edges when the weights are taken into account.
The Laguerre complex can then be drawn.

5.2 EQUATIONS OF MOTION

The complex is characterised by its energy defined by E = yL, where L is the total
length of the edges and y is the specific energy per unit length. Let us move one
paraboloid 'P, while keeping fixed the others. During its motion, 'P displaces the edges
and vertices of its associated cell, changing thus the energy of the froth. This define a
driving force
F =-VE =-"fl L
on 'P and generates its dynamic. 'P can move up and down (r2 varies), and
sideways (co-ordinates x andy vary). For any of these variables (x, y or r), Telley
assumed that the motion equation relates the velocity of the variable to some pressure,
exerted on the cell edges depending on this variable, via a constant M called mobility :
dx F
-=MPx =M x , Fx =-VxE
dt Lx(Fx)
Lx(Fx) is the total length of the projection parallel to Fx of the edges depending on
X.
Physically, the time constants ofthe two kinds of motion (vertical and horizontal)
can be different [21]. To represent the relative importance of both types, a parameter f.l.
( 0 ~ J.1 ~ 1) is introduced, (the horizontal mobility becomes f.l.M and the vertical one
(1-f.l.)M). When f.l. is 0, the growth is static (the centres of the cells are fixed). Increasing
f.l. allows some drift of the cells during coarsening (when f..1. is 1, the cells move but the
radii are fixed).
557

Telley also gave analytical expressions for the forces [18, 19]. During the motion of
the paraboloid '¥, only edges separating '¥ and its neighbours and edges separating
two neighbours of'¥ are affected by the motion. The forces are therefore very local and
can be expressed as functions of the co-ordinates (x, y, r2) ofthe moving circle and of its
(active) neighbours ..
M and yare physical constants depending on the materials. They are enclosed in a
normalised time 8, 8 = 2Myt where t is the real time. To take into account the fact
that the probability of a topological event, during a time step ~8, is roughly
proportional to the number of cells in the structures, Telley introduced another
parameter : 8 = ~8Ne (where Ne is the number of cells in the structure at time 8) as
the unit of time.
For the model only two internal adjustable parameters are needed: IJ. and 8.

5.3 ELEMENTARY TOPOLOGICAL TRANSFORMATIONS.

As mentioned before, topological changes are consequences of the motion of the


generating circles. During a time step, the algorithm calculates the time of the first
potential topological change, performs it, changes the neighbourhood of the generating
circles affected by it, modifies consequently the equation of motion locally.
The first results have been obtained by Telley et al. [22]. The influence of the initial
structure and of the parameter !J.. has been studied. Evolution leads to asymptotic
regimes. These regimes are very sensitive to the value of IJ.. For instance, if IJ. is low
(between 0 and 0.25) the value of the topological disorder (for the limiting distribution
P(n)) is close to that ofpolycristal, large values of!J. (between 0.5 and 0.75) correspond
to values of IJ.2 observed in soap froths. The long time limiting values seem to depend
also on the initial structure, a finite size effect appears for initial structures which are too
small or« unrealistic» (too far from natural structures), when they become «realistic »
not enough cells are left.
The finite size effect has the same consequence with simulations as with
experiments : it decreases the variance of the limiting distribution p(n).

6. Experiment versus simulation

The parameter IJ. also acts upon the elementary topological transformations (ETT) rate.
We investigated the influence of!J. on the ratio: Tl/T2, where Tl denotes the number
of neighbour switching and T2 the number of disappearance of 3-sided cells. We
observed [2] that, the higher !J., the higher Tl/T2 and the greater the topological
disorder IJ.2· When IJ."" 1, the cells can move but cannot change their areas, so that the
structure evolves through Tl transformations, Tl/T2 is high. Moreover, from Von
Neumann's law, the area of a cell will not change much only if it has 6,5 or 7 sides. So
558

that the topological disorder is low. A similar explanation holds for low values of 11.
which induce low values of Tl/T2 and large values of 112.
We measured experimentally TI/T2 for a soap froth obtained from dish-washing
liquid (in the large box). We obtained a value of about 1.4 (which can be compared to
the numerical value of 1.5 given by Herdtle and Aref [ 17]). The value of 11 which leads
to such a ratio is 0 .5. Consequently a value of 11 equal to 0 .5 is taken in the
simulations for comparisons.

6. l EVOLUTION OF AN ISOLA TED DEFECT

This problem was first discussed by Levitan [23]. This paper ~as been criticized [24].
Jiang et al. [25] and Ruskin et al. [26] performed simulations which are in conflict with
Levitan's predictions. It is thus interesting here to test Telley's model.
An hexagonal structure is built by putting the centres of the generating circles (with
equal radii) on a regular, triangular network. A defect is created by removing one circle,
generating a cluster of6-cells with two pairs of 5-/7-cells (Figure IO(a)). During all its
evolution the defect cluster grows and absorbs its neighbours. Precisely, the defect
consists of 4large cells (looking like a flower, see Figure 10(b)) surrounded by a ring c{
cells with less than 6 sides. The latter's average area fluctuates around a value of the
order of the area of the hexagonal cells. The bubbles outside the cluster remains
unchanged.

(a) (b)

Figure 10. (a) Initial defective network, (b) typical shape of the defect during evolution.

As the defect grows the number of neighbours (number of sides) n of the large central
cells becomes very large (n >>6). It is roughly proportional to the diameter d of the
defect. Using Von Neumann's law we have:
559

dan 1:-
"""'d( - n - d - "V" n

where an is the area of the large cell. Thus an increases as t2 and d as t. Moreover, since
the number of large cells remains constant (4), the number of cells in the cluster Nc
increases as d, i. e. as n or as t.
Thus Nc grows linearly with time, in agreement with Jiang et al. Results [25]. The
variance ll2c of the topological distribution Pc(n) of the cluster's cells varies also
linearly with time. Clearly f.L2c depends mostly on the large cells which have large n :
J.lzc ~ n2 Pc (n)
Because n increases as t and Pc(n) behaves as 1/Nc (i.e. as 1/t),
f.12c - t

Consider now the topological distribution overall p(n) and its variance ll2. For
n :t 6,
N.
p(n) = Pc(n)-"
N

where N is the total number of cells. Thus,

and f.L2 increases as t2.


Our simulations are in good agreement with these predictions, but they are affected
by large fluctuations (disappearance of small cells decreases N which decreases ll2 for
instance). If we exclude the four large cells in the centre of the cluster, we observe, for
the remaining boundary cells a limiting distribution, with a limiting value of f.L2 of the
order of 0. 7. This value and the associated topological distribution are very close to
those obtained by Jiang et al [25].

6.2 OTHER EVOLUTIONS

We also studied the evolution of an ordered structure (all the cells are hexagons
apart from a few defects). Each defect grows as above (single defect) until it reaches some
other defect. After some time the defective zone percolates and invades the whole
structure. Experiments and simulations are in very good qualitative and quantitative
agreement. ll2, starting from low values increases up to about 4 (Figure ll) and then
decreases to its limiting value (from -1.5 to -1.7). The stationary, asymptotic
distribution is the same, whether the initial structure is ordered or disordered.
560

..

·~.~--~----~--~----~--~.~

Figure ll.IJ.2 as a function of time for an ordered initial structure (simulation with 11 = 0.5).

7. Conclusions

In this chapter we have studied the evolution of two-dimensional soap froth.


Experiments show that the nature of the foaming agent is an important parameter.
Specifically, a stationary asymptotic regime is not always reached. This scaling state
only appears for soap froth generated from high intrinsic surface viscosity solutions [9]
(as for dish-washing liquid). Moreover a finite size effect has been pointed out.
Simulations from Telley's model are in good agreement with experiments. Telley's
model is probably one of the best programs for simulation of large cellular structures.
Acknowledgments
We would like to thank G. Le Caer (LSG2M, Nancy), L. Oger and N. Pittet (Rennes),
N. Rivier for useful discussions. T. Liebling, H. Telley and F. Righetti have helped us
to improve the program.

References

[1] Stavans J. (1993) The evolution of cellular structures, Reports on Progress in Physics 56, n°6, 735-789.
[2] Pignol V. (1996) Evolution et caracterisation de structures cellulaires bidimensionnelles experimentales,
en particulier les mousses de savon, et simulees, These INPL, Nancy.
[3] Glazier J. A., Gross S. P. and Stavans J. (1987) Dynamics of two-dimensional soap froth, Phys. Rev. A.
36 n° 1,.306-312.
[4] Stavans J. and Glazier J. A. (1989) Soap froth revisited: dynamic scaling in the two dimensional froth,
Phys. Rev. Lett. 62, 1318-1321.
[5] Stavans J. (1990) Temporal evolution of two dimensional drained soap froths, Phys. Rev. A. 42, 5049-
5051.
[6] Pignol V, Oelannay R. and Le Caer G. (1993) Characterization of topological properties of 20 cellular
structures by image analysis, Acta Stereo/1212, 149-154.
[7] Righetti F., Telley H., Liebling T. M. and Mocellin A. (1992) 20-CELL: image processing software for
extraction and analysis of 2-dimensional cellular structures, Computer Phys. Comm. 67,509- 526.
561

[8] Coster M., Chermant J. L. (1989) Precis d'analyse d'images, Presses du CNRS, Paris.
[9] Han G. (1996) Influence des proprietes rheologiques de surface sur Ia morphologic et Ia dynamique des
mousses bidimensionnelles, These, Poitiers.
[10] Atkinson H. V. (1988) Theory of normal grain growth in pure single phase systems, Acta Metall. 36,
469-491.
[II] Fradkov V. E.,Shvindlerman L. S., Udler D. G. (1985) Computer simulation of grain growth in two
dimensions, Scripta Metall. 19, 1285-1290.
[12] Beenakker C. W. J. (1988) Numerical simulation of a coarsening two-dimensional network, Phys.
Rev.A 37, 1697-1702.
[13] Flyvbjerg H. (1993) A model of coarsening froths and foams, Phys. Rev. E 47, 4037-4053.
[14] Stavans J., Domany E., Mukamel D. (1991) Universality and pattern selection in two-dimensional
cellular structures, Europhysics Letters 15, 479-484
[15] Glazier J. A., Anderson M. P., Grest G. S. (1990) Corsening in two dimensional soap froths and the
large-Q Potts model: a detailed comparison, Phil. Mag. B 62, 615-645.
[16] Weaire D., Kermode J.P. (1983) Computer simulations of a two-dimensional soap froth. I. Method and
motivation, Phil. Mag B 48, 245-259.
[17] Herdtle T. and Aref H. (1992) Numerical experiments on two-dimensional foam, J.Fiuid Mech. 241,
233-260.
[18] Telley H.(1989) Modelisation et simulation bidimensionnelle de Ia croissance des polycristaux, These
EPFL n°780, Lausanne.
[19] Telley H., Liebling T.and Mocellin A (1996) The Laguerre model of grain growth in two dimensions I.
Cellular structures viewed as dynamical Laguerre tessellations, Phil. Mag. B 73, 395-408.
[20] Xue X., Righetti F., Telley H., Liebling T. and Mocellin A. (1997) The Laguerre model for grain
growth in three dimensions, Phil. Mag. B 75, 567-585.
[21] Rivier N. (1993) Order and disorder in packings and froths, DISORDER AND GRANULAR MEDIA,
Elsevier Science Publishers B.V., 55-102.
[22] Telley H., Liebling T. and Mocellin A.(1996) The Laguerre model of grain growth in two dimensions.
II. Examples of coarsening simulations, Phil. Mag. B 73, 409-427.
[23] Levitan B. (1994) Evolution of two dimensional soap froth with single defect, Phys. Rev. Lett. 72, 4057-
4061.
[24] D. Weaire (1995) Comments on« evolution of two dimensional soap froth with single defect», Phys.
Rev. Lett. 74, 3708-3710.
[25] Jiang Y., Mombach J.C.M., Glazier J.A. (1995) Grain growth from homogeneous initial
conditions :anomalous grain growth and special scaling states, Phys. Rev. E 52, R3333-R3336.
[26] Ruskin H. J., Feng Y. (1995) The evolution of a two dimensional soap froth with a single defect,
J.Phys. : Condens. Mater 7, L553-L559.
562
SIMULATION OF THE FOAMING PROCESS

N. PITIET 1
LDFC, Universite Louis Pasteur
67084 Strasbourg Cedex, France

Abstract

The Kerroc® glass foams can be made with different porosities, depending on two
thermodynamics parameters of the foaming process, temperature and duration of the
expansion stage. Notably, at lower temperatures, one can pass from total imperviousness to
permeability as the time of expansion increases. The increase in bubble connectivity with
time is simulated, through a generalisation of Voronoi tessellations, by the build-up of a
network of physical contacts between bubbles which ultimately percolates. The temperature
is modelled by different distributions of spheres, representing a snapshot of the foam at
different temperatures during its expansion.

1 . Introduction

The expanded glass foam Kerroc2 [1, 2, 3, 4] is a solid foam produced as small beads or
as bricks. The material can have a different porosity (close or open partitions between
cells), depending on the thermodynamics conditions (duration and temperature of the
expansion process, as described by Rivier in this volume) of their formation in the
oven. The structure and physical properties of the glass foam are macroscopic,
thermodynamic parameters, related to the temperature and the time of expansion by an
equation of state [3].
Kerroc is made of crushed glass, mixed with AlN powder. The mixture is heated at
about 1000 °C. At this temperature, the glass is melted and the formation of bubbles is
made possible through the chemical reaction between the AlN and the oxides present in
the molten glass [1, 2].
The percolation transition from impervious to permeable state is characterised by a
critical number of channels (holes) open between bubbles. A hole can be formed only
between bubbles in contact; a contact is defined as a thin interface (film of molten glass)
between two bubbles. From the nucleation of independent bubbles (zero contact), the
number of contacts increases as the bubbles grow in size, because more larger bubbles
1 Present address: GMCM, Universite Rennes 1, 35042 Rennes Cedex, France
2 Kerroc® is a trademark of Cemix, 35235 Thorigne-Fouillard, France.

563
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 563-570.
© 1999 Kluwer Academic Publishers.
564

Time
Figure 1. Schematic phase diagram of solid foam. The thermodynamic variables are the temperature and
the time (duration) of the foaming process. The phases, illustrated by SEM photographs of Kerroc, are
closed cell porosity (impervious material) and open cell porosity (porous material).

touch each other. As soon as there is contact, there is a finite probability for a hole to
be nucleated. This probability depends on the temperature, on the rate of growth of the
two bubbles, on shape and viscosity of the shared interface and on the difference of the
sizes of the two bubbles [3, 4].
The foaming process begins with independent nucleation (at random in space and
time) and growth of bubbles. Then, bubbles interact through physical contacts (see [3]
and the chapters by Rivier and by Brechet in this volume). The establishment of more
and more contacts modifies the physical properties (electrical, thermal, mechanical, ... )
of the bulk material. In solid foams, the expansion process is stopped after some time.
At that time, contacts between bubbles may or may not have percolated through the
system. The percolation threshold corresponds therefore to a critical time in the foaming
process, separating two (impervious and permeable) phases of the solid foam. We will
treat the foam formation as a dynamical bond percolation problem. The relevant
parameter is the number of contacts per cell (bubble),
n(t) = 2C(t)/ N, (I)
at time t, where C(t) is the total number of contacts of the N bubbles.
At the end of the growth process, when the volume ratio of the initial liquid (molten
glass in Kerroc) to gas phases is less than few percents, each bubble is in contact with
its neighbours and has a polyhedral shape. By drawing bonds between the centres of
connected bubbles, we obtain a bond network describing tetrahedral close-packing (see
chapters by Jullien et al, by Oger et al and by Sadoc in this volume). The final
connectivity of the bond network is also the final number of contacts per bubble n1 .
Topologically, there is a duality between the bond network and the polyhedral structure
(one bond<-> one contact<-> one film).
565

Figure 2. Voronoi tessellation (bold lines) and Delaunay triangulation (its dual, thin lines) of a set of points.

Thus, the mean connectivity n(t), or the mean number of contacts characterises the
global topology of the solid foam. The connectivity increases during the foaming
process, n(t0 ) = 0 < n(t) < n(t1 ) = n1 .
To model the dynamics of the foaming process, we use a new numerical technique
[5] which puts growth into Voronoi diagrams. We do not consider here the coarsening of
the foam driven by diffusion of gas between adjacent cells. Although films are perforated
as soon as a contact occurs, they never rupture and bubbles never coalesce. This ensures
that the foam topology does not evolve after the last contact is established. In this
simulation, the hole nucleation probability is set to unity (one contact = one hole),
because the physics of the hole nucleation in curved films is not our concern here. As
the number of contacts increases, it reaches a critical number at and beyond which air or
fluid can pass through the foam. This critical number is the percolation threshold.

2. Growth and Tessellations

The foam at the final stage of its evolution (polyhedral packing of bubbles) is a
partition of the space (tessellation) into cells. This partition can be modelled with the
Voronoi tessellation technique applied to a packing of spheres (or circles in 20) with
different radii. The shape of the cells depends on the definition of the distance between
the spheres (Euclid, Laguerre, non-Euclidean, ... ). The final tessellation has informations
on the growth mode of the spheres, and on their nucleation (initial conditions).
To illustrate the relation between growth of cells and Voronoi diagram, consider a
cell i, nucleated at time t; and a second cell}, nucleated at time tj, at distance d ij from
the first. The cells grow with the same law R;(t), which we take here to be:
R;(t) = g(t- f; )a, (2)
(it could be any monotonic function). Here R; the radius of the cell at time t, and a is
the growth exponent. The two bubbles touch at contact time tc, defined by
566

Figure 3. Two growing bubbles and their interface. The growth exponent is a = I (left, Johnson-Mehl
model) and a = II 2 (right, Laguerre model). At time I, the bubbles are separated. At time 2, they have
one point of contact ( t 2 = t c). Then, an interface begins to grow.

( 3)
When t > tc, the two spheres intersect on a circle C(t) . All these circles, from tc to
t, lie on a surface of revolution S(t), bounded by C(t). S(t) defines the growing
interface between the two growing bubbles. As t increases, S(t) keeps increasing for
0 <aS 1, whereas it closes if a> 1. Figure 3 illustrates two situations at different
times for the exponents a =112 and a =1. The growing interface is finite (elliptic) if
a>l.

Figure 4. Build-up of the bond network. Top left: Initial structure, before any contact between bubbles.
Tessellation of this structure yields successive contact times, and the network of bubbles in contact with its
connectivity n(t) as it develops in time: Before percolation threshold (n=l), at the threshold (n=2.04) and
after (n=3).
567
random times and growing according to the same law (Eq. 2). This set of nucleation
centres and times, with a growth law, defines entirely the tessellation, the
neighbourhoods, and the shape of the interfaces. It also defines the times of contact,
which determine the porosity of the Kerroc.
Generalised Voronoi tessellations have been used extensively in metallurgy,
geography or ecology. Growth in a= 1/2 produces the Laguerre tessellation [6,7]
(planar interfaces); growth in a= 1, the (Avrami-) Johnson-Mehl tessellation [6]
(hyperboloid interfaces).
These tessellations are easily obtained from geometrical arguments, with a faster
algorithm than the pixel growth technique[6]. The tessellation is a shortcut to obtain the
final structure, given the structure at some time. The final structure gives the final bond
network, as the different contact times is directly computed from the initial packing of
spheres by tessellation. This initial packing has to be chosen carefully (Section 4 in this
chapter), since the evolution of our forming foam (back- and forwards in time) is
deterministic.

3 . Sequence of contact times

The sequence of contact times can be computed by the tessellation technique, given the
structure of the foam at an early stage of expansion, before any contact is established
between the bubbles (i.e. an initial distribution of spheres centres and radii or of their
nucleation times), and a growth law (eq. 2). The clock is started at the time t =0 of the
first contact. (Bubbles to be nucleated after t =0 can be represented in the initial
distribution by negative radii). At each contact time, a new link is OOded to the bond
network (linking the centres of bubbles in contact, dual of the tessellation). The bond
network is built up in time, in a deterministic fashion, by the tessellation. Specifically,
one obtains the development on the connectivity of the network, from n(t = 0) = 0 to
n1 , and the time at which the bond network percolates. The percolation transition is
here a dynamical process.
The number of contacts or connectivity n(t) is critical for the properties of the
material (thermal, resistivity, porosity, ... ). As a function of time, it has generic
1o' 1!'"'""'1,-rr......r-r"'T'T'I'I"'III"--r'"TT'I'lrmr-......,~rwr-..,...,......_
6

21'-----------1-
100

n(t)

Fillure 5. Characteristic evolution of the connectivitY.


568

=
Figure 6. Examples of the two distributions modelling the growth at low viscosity (left, fmax 0) and at
high viscosity (middle, !max =I). The numerical viscosity will be the maximum percentage of circle
interpenetration. The typical distributions of size circle is shown on the right.

features, common to all foams, whatever their growth law (eq. 2). Hereafter, we will
discuss only two-dimensional foams, which can be visualised readily (Figure 4). The
same algorithm applies to three-dimensional foams, and above. The main characteristic
features of n(t) (Figure 5) are:
(i) The exponent f3 of the initial power law.
(ii) The final connectivity n1 of the bubbles.
(iii) The percolation time tP, of the contacts throughout the material.
The porosity is closed before tP ' open afterwards. We will discuss the dependence of
the exponent f3 and of tP on the growth conditions.

4 . The example of Kerroc

We model here the evolution of the porosity of Kerroc. Nucleation of the bubbles and
their expansion takes places in a molten glass phase. The viscosity of glass around
l000°C is high, and it varies rapidly with the temperature [1-4]. (See also Rivier's
chapter).
At the higher temperatures, the glass is fluid and a spherical bubble will recoil
instead of forming a contact, as long as it is allowed to do so by the other bubbles. A
suitable initial structure modelling the Kerroc in the fluid phase is a distribution of non-
overlapping circles, built sequentially (Figure 6). A new circle (centre, radius) is chosen
at random; it is accepted if it does not overlap the circles already there, rejected if it does,
or if it is fully contained inside a pre-existing circle (nucleation of a bubble can occur
only in the molten phase and never inside a bubble).
At the lower temperatures, the glass is like a paste (see Rivier' s chapter) and the
centre of a bubble is pinned at its nucleation spot. The typical situation of Kerroc in the
fluid -~-- viscosity - - - paste

0 0.5 f
Figure 7. The overlapping factor f describes the type of distribution of circles and is related to the viscosity
of the Kerroc: Bv increasin!! ffrom 0 to I. one oass from a fluid to a oaste model Isee text).
569
paste phase is modelled by a random distribution of circles, possibly overlapping
(Figure 6).
Here, the overlap factor f between two circles (Figure 7) is defined as the ratio of
the area of overlap to the area of the smaller circle. The overlap factor of a distribution
of circles (Figure 6), fdistr, is then that of the pair of circles with maximum overlap. In
a simulation, we first set a maximum overlap !max, select circles at random but
sequentially, and reject a new circle if its overlap with a pre-existing circle is larger than
!max. The process ends when the number of circles reaches N. Then, fdistr is equal to,
or under !max. Figure 6 exhibits two distributions, one for !max = 0, the other for
fmax =0.
The bubbles themselves are not all circular, their shape is determined by the
tessellation (Section 2 in this chapter). Expansion proceeds according to the growth law
(Eq. 2). Alternatively, from the typical distributions of Figure 6, one can obtain the
nucleation time distribution of bubbles by running the growth law backwards in time.
Two numerical parameters can be extracted from the distributions of Figure 6. Both
have a physical interpretation. The overlap factor fmax fixes our numerical viscosity.
When we have !max = 0, we model the fluid phase, as the bubbles can move and remain
spherical until they are squeezed by others. A random close-packing of non-overlapping
circles describes the Kerroc, from the time when the structure passes from spherical to
polyhedral bubbles until the end of the expansion. With fmax = 0, the nucleation
positions are not known, only the nucleation times.
By increasing continuously fmax until 1 (the paste phase), as it becomes more
viscous, or as the temperature is lowered, the centres of the circles and the true
nucleation centres are more correlated for being the same at !max =1. Another numerical
parameter is the maximum radius of our circles distribution. It fixes the time elapsed
between the first and the last nucleations.
With this technique, we can compute the slope f3 and the percolation time for
different modes of growth (exponent a in eq. 2), different distributions of times of
nucleation and different viscosities of the molten glass (i.e. to the temperature of the
expansion process).
The percolation time can be read directly from Figure 5, knowing that, in two
=
dimensions, an average connectivity of n(tp) 2 is necessary for the bond network to
percolate [8]. On the numerical samples we tested, we have verified this assumption

...······
fluid
f.u =O .·
l3guerre .· ·'
' ·'
:'viscous

'·fj··' .. ' .· Jolwon·Mehl


.... ...
•.• • '---~"'----.._~_......_ _ _..J

lill'll! (:u'i1rary units )


• ... u
·····
lime: (3rlrruy uniul

Figure 8. The viscosity of the molten glass (hence the temperature) (left) and the mode of growth (right)
affects the evolution of the bond network Inetwork of bubbles in contact).
570

within 2 percents of error. On Figure 8, we see that percolation is reached earlier in the
fluid than in the paste. The slope is smaller in the fluid ( {3 = 1) than in the paste
( {3"' 2 ).
In the paste, the viscosity can be varied continuously, through the maximum of
interpenetrability allowed for the circle packing construction.
The mode of growth also affects the evolution of the connectivity. In the Johnson-
Mehl model (a = 1), percolation is reached later and with a larger slope than in the
Laguerre model (a= 1I 2 ).

5. Conclusions

In this chapter, we have modelled the properties and the structure of expanded foams, by
simulating the foaming process [4,5]. The simulations have concentrated on the
topology of the network of physical contacts between bubbles, as they become
established during the expansion, i.e. on dynamical percolation. Bubble growth and
physical contacts have been modelled by Voronoi tessellation. We have described the
influence of the mode of growth, of the temperature and duration of the expansion
process on the final structure and porosity of the Kerroc glass foam. Some parameters
(viscosity in the fluid phase, and permeability of the interface in the paste) have been
fixed. To vary them is the next computational challenge.

Acknowledgements

We would like to thank S. Graf (Institut Charles Sadron, Strasbourg) for the SEM
photographs, A. Gervois and J.P. Troadec for useful discussion on the numerical
technique, and N. Rivier for his scientific and linguistic contribution to this chapter.
This work has been supported by the Swiss Office Federale de !'Education et de la
Science and by the EU FoamPhys Network (Contract ERBCHRXCT940542).

References
[1] Tasserie, M. (1991) Optimisationphysicochimique d'un materiau expanse, These, Univ. Rennes I.
[2] Garnier, C (1993) Verres oxyazotes de sialons monolithiques et composites particulaires a hauts
modules elastiques, These, Univ. Rennes I.
[3] Rivier, N., Pittet, N., Laurent, Y., Troadec, J.-P. (1996) Thermodynamic of solid foam, SFP-JMC5,
Orleans and in preparation.
[4] Pittet, N. (1997) Thermodynamique et structure de Ia mousse, These, Univ. Louis Pasteur, Strasbourg.
[5] Pittet, N. (1997) Dynamical percolation through the Voronoi tessellations, submitted to J. Phys A.
[6] Frost, H.J. and Thompson, C.V. (1987) The effect of nucleation conditions on the topology and geometry
of two-dimensional grain structures, Acta Metal!., 35, 529-540.
[7] Telley, H. (1989) Modelisation et simulation bidimensionnelle de Ia croissance des polycristaux, These
No 780, Ecole Polytechnique Federale de Lausanne.
[8] J.M. Ziman, Models of Disorder, Cambridge University Press, Cambridge (1979).
VORONOI TESSELLATION IN MODEL GLASS SYSTEMS

REMI JULLIEN, PHILIPPE JUND AND DIDIER CAPRJON


Laboratoire des Verres, Universite Montpellier II,
Place Eugene Bataillon, 34095 Montpellier, France
AND
JEAN-FRANQOIS SADOC
Laboratoire de Physique des Bolides, Bat. 510,
Universite Paris-Sud, Centre d'Orsay, 91405 Orsay, France

Abstract. Numerical Vorono1 tessellation is used to investigate the mecha-


nisms of frustration in some model glass systems. First, random packings of
8192 hard spheres of increasing volume fraction care built in the flat three
dimensional space using an efficient computer algorithm. Their Vorono1
statistics evolves with c as if the system would like to reach a pure icosahe-
dral order when extrapolating the volume fraction above the Bernal limit
Cb ~ 0.645. Second, this study is extended in curved space, the sphere 83.
When decurving the space by increasing the number N of spheres, the most
compact packings converge to the Bernal packing. For particular N values,
the volume fractions exhibit maxima corresponding to narrower histograms
for the number of edges of Vorono1 polyhedra faces. Third, super-cooled
liquid and glass samples of 1000 atoms are generated at different tempera-
tures T after a quench from the liquid state, using classical micro-canonical
molecular dynamics with a simple soft-sphere potential. When decreasing
T, the ideal icosahedral order appears again as an extrapolated situation
which cannot be realized due to geometrical frustration.

1. Introduction

In this paper, we would like to report on the mechanisms of geometrical frus-


tration in some model glassy systems analyzed numerically via the Vorono1
tessellation. We first consider random packings of hard spheres since they
were extensively used throughout the last decades to represent the struc-
571
J. F. Sadoc and N. Rivier (eds.), Foams and Emulsions, 571-588.
© 1999 Kluwer Academic Publishers.
572

ture of liquids, amorphous solids and glasses [1, 2, 3, 4, 5, 6]. One of the
most fascinating features of random packings, which was first evidenced
by Bernal[!], is that there exists an upper limit of the volume fraction
Cb ~ 0.645 which cannot be exceeded and which is significantly smaller
than the one Cm = 0. 7405 of the ordered close packings (hexagonal-closed-
packed and face-centered-cubic). It is generally believed that the existence
of such an upper limit is related to the degeneracy between hexagonal-
closed-packed and face-centered-cubic structures leading to the so-called
geometrical "frustration" [7] associated with the impossibility to tile the
space with perfect tetrahedra only. The similarities between the local tetra-
hedral order found in random packings and the one observed in many metal-
lic glasses has been stressed since a long time [8]. Here, using numerical
Voronol tesselation, we show that, when increasing the volume fraction up
to the Bernal threshold, the system evolves as if a perfect tetrahedral order
would like to take place. All the characteristics of the Voronol cells vary
with c in such a manner that, when extrapolated above cb, they converge
to those of a perfect dodecahedron circumscribed to a sphere.
Since it is known that the frustration can be easily varied, and some-
times eliminated, by working in curved space[7], we have extended our
numerical study by building compact packings on the sphere 83. Since, for
a given individual sphere diameter, the number of spheres N in the packing
fixes the radius of curvature of the space, we find that the Bernal packing
is recovered in the limit of very large N values. For some peculiar N values,
one obtains highly symmetric packings with larger packing fractions and
narrower histograms for their Voronol cell statistics. The largest packing
fraction is obtained for N = 120 which corresponds to the famous polytope
{ 3,3,5} (using the notations of Coxeter[lO]) whose Voronol cells are known
to be perfect dodecahedrons.
However, one conceptual problem when trying to use random packings
to simulate the structure of glasses is that one does not consider neither any
realistic potential nor the true thermodynamics, and therefore all the fea-
tures associated with the existence of a glass transition cannot be accounted
for. This is why, for comparison, we have performed in this work the same
kind of geometrical analysis on atomic configurations, but of more mod-
est size, generated by using classical molecular dynamics (MD) of a model
glass. By determining the histograms for the distribution of volumes and
total surface areas of the Voronol cells, we are able to analyze the general
trends for the evolution of the local order as a function of the temperature
T. The same tendency to develop a perfect tetrahedral order, with dodec-
ahedral Voronol cells and five-fold symmetry, is observed when decreasing
T from the supercooled liquid state. Our results strongly support the idea
that the perfect tetrahedral order would be an ideal situation which can not
573
be reached due to geometrical frustration, as postulated in a recent theory
for the glass transition[9].
In section II, we present the calculations for random packings in flat
three-dimensional space. In section III, we present their extensions to the
sphere 8 3 . In section IV we present the calculations for atomic configura-
tions obtained by molecular dynamics and in section V, we conclude. Some
accounts of the work presented here have been published elsewhere[ll, 12,
13].

2. Random packings in flat space


2.1. THE JODREY-TORY ALGORITHM

To generate sphere packings in a square box of edge length L, with periodic


boundary conditions {PBC), we have followed an efficient numerical recipe
which was introduced by Jodrey and Tory (JT) [14] more than ten years
ago. The JT algorithm proceeds by an iterative sequential resorption of
overlaps of -imagined- spheres which consists in successive displacements
of pairs of nearest neighboring points (sphere centers) starting at iteration
i = 0 from a set of N points randomly located in the simulation box. At
each iteration i the set of points (the sphere centers) is characterized by
the list of coordinates and also by a list of distances {between pairs of
points) in increasing order together with some other tables necessary to
identify the points in the list. To the minimum distance cfm corresponds
a minimum packing fraction c~ = N(1rj6)(cfm/L) 3 . Along the iterative
procedure, one also carries a maximum distance cfM, related to a maximum
packing fraction cit= N(7rj6)(cPM/L) 3 , which is set to cf!M = L(6j1rN) 1I 3
(i.e. c~ = 1) at i = 0. After the identification of the pair of points Mf and
M~ realizing the minimum distance cfm = M{ M~, these points are spread
apart symmetrically along the Ml M~ line to new positions Mf+l and M~+l
such that Mf+l M~+l = cfM. Then, before going to the next iteration, the
list of distances and the related tables are updated, the new minimum
distance cr,;t 1 is determined, and the maximum distance is set to a lower
value given by:
{1)

where the "rate" K and the exponent a are two input parameters of the
algorithm in addition to L, N and cf!M. Note that formula {1) is slightly
different (simpler) than the original one used by Jodrey and Tory [14] and
consequently our definition of the rate K is different. The process stops
at iteration n when one finds~<~- Then the final minimum distance
~ is taken as the particle diameter d for the resulting packing. The final
574

packing fraction is then given by:

(2)

Note that the value taken for the box edge length L does not play any
role as it only fixes the unit of length. In practice L has been set to an
integer value L = 20 because we have used an underlying cubic lattice of
8000 cells to label the spheres in order to accelerate the search for their
neighbors. The number N of spheres has been set to N = 8192, almost
an order of magnitude larger than the JT value N = 1000 [14). We have
checked on a few other N values that the results are size insensitive. While
the parameter d~ is not very important (it should be taken sufficiently
large however to guarantee that the spheres of diameter dflM do overlap in
the initial configuration), the remaining parameters Kanda are essential
not only to fix the final packing fraction, but also to determine the rate at
which it is reached. Here we have taken a = 0.33 and we have varied K to
obtain a set of samples of various volume fractions ranging from c = 0.4235
to c = 0.6430, a value close to the Bernal's threshold q. When extrapolating
these volume fractions to K = 0, we obtain cb ~ 0.645 in good agreement
with previous estimates. To build the most dense packing with K = w- 4 ,
of volume fraction c = 0.643, we used a few days of IBM RISC 6000/580
computer time.
As soon as they have been built, all the packings are rescaled to get a
particle diameter of 1 and, consequently, in all what follows, the distances
will be expressed in diameter units, A detailed analysis of the long range
correlations in these packings obtained from a calculation of the pair cor-
relation function g(r), up to distances r of about twenty sphere diameters,
have been reported elsewhere [11].

2.2. DETERMINATION OF THE VORONOI CELLS

We recall that a Vorono1 cell is the extension of a Wigner-Seitz cell for a


disordered structure: it is defined as the ensemble of points closer to a given
sphere center than to any other and it is characteristic of the local environ-
ment around this sphere. The positions of the sphere centers being known
at a given time, the Vorono1 tessellation has been numerically performed
by first determining the Delaunay tetrahedral simplicial cells which are,
among all the tetrahedra formed with four sphere centers, the ones such
that no other sphere center lies inside their circumscribed spheres. After
determining the simplicial tetrahedra, all the elements, faces and edges,
of the Vorono1 cell of a given sphere can be determined knowing that its
vertices are the centers of the circumscribed spheres of all the simplicial
575
tetrahedra sharing this sphere center. We have checked that the cell ver-
tices have always three edges in common, as expected in random structures.
Therefore for each cell, the total number of edges E and the total number
of vertices V are related by 2E = 3V. Knowing that these two quantities
are related to the total number of faces F through the well-known Euler
formula, V- E + F = 2 [15, 16] , one has also F = 2 + V/2 = 2 + E/3 .
These relations have been verified by our numerical results. Another for-
mula, < F >= 12/(6- < e > ), has been verified. It relates the mean
number of faces < F >, also called coordination number z, to the mean
number of edges per face < e >[16]. Not only V, E, F and their mean over
the N atoms have been calculated but also, the volume v of each cell and
its total areas, which is the sum of the face areas.

2.3. RESULTS
In figures 1a and 1b we report on the distribution functions hv and hs of
the Vorono'i cell volumes and surfaces for some of our sphere packings with
various volume fractions.
\5.0 .--~-~-~-~----,

(a) 3.0 (b)

2.0

1.0 tl\\
.,
,.
,:~'·"'··· ·\,•-.
...'.. \. ,__
~ ........ _.... --~-~'·-·- ·
e.o 1.0 e.o
s

Figure 1. Cell volume hv(v) (a) and cell surface hs(S) (b) distribution functions for
sphere packings in flat space. Dot-dashed, dashed, dotted, and solid lines correspond to
c = 0.424, 0.518, 0.585 and 0.643, respectively.

They are defined such that hvdv is the proportion of cell volumes lying
between v and v + dv (and similarly for h 5 ) . Both histograms become more
and more peaked as the volume fraction increases. As a measure of the
widths of these histograms we have calculated the standard deviations av
and a 5 of v and s which have been reported as a function of c in figure 2.
Both quantities extrapolate to zero at about the same value of the
packing fraction. The black dot indicated on the figure corresponds to co =
576
0.8

0.6

o:."'

.
-g 0.4 D

oO D

D
0.2 0

0 D~
0
0
0
OCOll>
0.0
0.40 0.50 0.60 0.70 0.80
c

Figure 2. Standard deviations Uv (open circles) and us (open squares) as a function of c


for sphere packings in flat space. The filled circle indicates their expected extrapolations
for c =eo.

0. 754 a value obtained by dividing the volume of a sphere by the volume of


the regular dodecahedron with all faces tangent to it. It is worth noticing
that co is larger than the packing fraction Cm = 0. 7405 of the FCC structure.
The results of figure 2 can be interpreted as follows: when increasing the
concentration, the mean local arrangement of the neighboring sphere cen-
ters around a given one evolves progressively towards a regular icosahedron
while the corresponding Voronol cell evolves towards a perfect dodecahe-
dron. However, due to geometrical frustration, one cannot tile the Euclidean
three dimensional space with perfect dodecahedra only. Therefore the vol-
ume fraction c cannot exceed the threshold q, at which the distribution
of volumes has a well defined non-zero standard deviation. This conclu-
sion is also supported by a quantitative analysis of the long range damped
oscillations of the correlation function g(r) [11].
It is worth noticing that the mean cell volume Vm is trivially related to
the packing fraction through the formula (in which the sphere diameter is
set to unity):
L3 rr
Vm=-=- (3)
N 6c

As a consequence the maximum of the distribution function hv is shifted


towards low values as c increases. However the evolution of the mean cell
577
surface sm is non trivial and can give some information on the evolution of
the mean cell shape.
2
In figure 3 we have reported the quantity 8mfvJ, as a function of c. It
extrapolates nicely at c = co to the corresponding quantity calculated for a
perfect dodecahedron. This provides another support to the above analysis.
5.65 .---~-----.-------r------.---.-----,

0
5.55
0
0
0
~ E 5.45 0
>l
Ul
00
%

5.35


5.25
0.40 0.50 0.60 0.70 0.80
c
Figure 9. Dimensionless ratio Smfv!{ 3 as a function of c for sphere packings in flat
space. The filled circle indicates the expected extrapolation for c =eo.

In figure 4 we report the mean coordination number z =< F > as a


function of c.
Unless one accepts a very slow convergence, there is no clear evidence
for an extrapolation to z = 12 (dodecahedron) for c ---t c0 • In fact this is not
inconsistent with the above reasoning because extra faces might persist, the
areas of which vanish only at c =co. In that case z might be discontinuous
at c =eo. The same kind of behavior occurs when disturbing any degenerate
regular structure, for example when including an infinitesimal temperature
in an FCC crystal.

3. Random packings in curved space


3.1. EXTENSION OF THE NUMERlCAL METHODS TO 8 3

We recall that the sphere 83 is the extension of the regular sphere 8 2 to


one dimension more. Therefore, working in an Euclidian four dimensional
space, the coordinates Xik (k=1,2,3,4) of the vector ri = joining the oAt
578

16.0 r-------,------.--~--.--~-----,

15.0 0
0
0
0
0
OC\:@
N 14.0

13.0

12.0 '---~---,-'-.,-----~--:-':-:--~--.J...._-__._ __J


0.40 0.50 0.60 0.70 0.80
c
Figure 4. Mean coordination number z as a function of c for sphere packings in flat
space. The filled circle indicates the expected extrapolation for c = co.

center of the space (sphere S3) to any sphere center i of the packing verify:

lril 2 = L:xrk = R 2 (4)


k

where R is the radius of the sphere S3 which is a measure of the size of


the space. Starting from a random set of N points on S 3 , the Jodrey-
Tory algorithm can be straightforwardly extended by replacing the Euclid-
ian distances dM and dm by "geodesic" distances. The expression for the
"geodesic" distance dij between two points Mi and Mj ( i. e. the shortest
distance counted along the curved space) is simply related to the scalar
product of ri and rj:

.. _
d tJ R -1 ri.rj _ R -1 Lk XikXjk (S)
- cos R2 - cos R2
The volume of a sphere of radius r can be expressed as a function of w =
r/R,
v(w) = 21r(w- si~2 w)R3 (6)
As a consequence, the volume fraction c, of a packing made of N spheres
of diameter d is equal to the total volume occupied by the spheres, which
is Nv(d/2R), divided by the total volume of S3, which is v(1r). Therefore:

c = N (w _ sin2w) (7 )
7r 2
579

with:
d
w= 2R (8)

It is worth noticing that one recovers the well known expressions for the
definition of the distance and for the packing fraction in the limit R/ d -t oo
of a flat space.
The input parameters of the computer code are the same as in flat space
except that the box size L is here replaced by R . In the following, all the
lengths will be counted in units of the sphere diameter d, and therefore
R will stand for R/ d. Consequently the limit N -t oo corresponds to the
limit R -t oo of an infinitely large, and flat, space. To be sure to get,
for any N value, the most compact packing compatible with a reasonable
computing time we have used the same values for the parameters K and
a than the one used to build the most compact packing in the preceding
study, i. e., K = 10- 4 and a= 1/3. As soon as a packing is constructed a
Voronol tessellation is performed by straightforwardly extending to S 3 the
procedure explained above.

3.2. RESULTS
The numerical results for the packing fraction c for N values up to 300 are
depicted in figure 5. We have run the code five times for each values of N
using independent initial random configurations.
0.80 .---..------r----.-----r------.-----,

0.75

" 0.70

0.65

0·60 O
L__ _5_,_0_ _ _1...J.00
- - - - - -15-:-
' 0---:-
200
'-:----:-2=:
50- - --:::300
N

Figure 5. Numerical results for the packing fraction c as a function of N uptoN= 300
for sphere packings on Sa . Five independent results have been recorded for each N value.
580

Despite some noise which can be attributed to the lack of efficiency


of the algorithm and the randomness of the initial configuration, one can
define roughly two different mean curves. The lower curve is almost flat
and corresponds to a c-value of about 0.655, close to (slightly larger than)
the Bernal's packing fraction, while the upper curve exhibits a series of
maxima. We will comment more on these two curves in the next section
but we would like to first focus on the maxima of the upper curve.
The maxima correspond to more regular structures with narrower dis-
tributions for the cell statistics. This is obviously the case of the packings
obtained for N = 5, 8 and 120, where the cell statistics is monodisperse.
The cells are perfect tetrahedra (F=4, e=3), cubes (F=6, e=4) and do-
decahedra (F=12, e=5) and the packing fractions are 0.6806, 0.7268, and
0.7741, respectively. They correspond to the polytopes {3,3,3}, {3,3,4} and
{3,3,5} using the notation of Coxeter[lO]. Note that the volume fraction
obtained for the remarkable polytope {3,3,5} (which exhibit five fold sym-
metry) is larger than the one of face-centered-cubic (FCC) and hexagonal-
close-packed (HCP) structures in flat space. Slightly richer structures, but
always with a narrow cell statistics, are obtained for all secondary peaks of
the curve. As an example, we have reported in figure 6 the histogram for
the fraction of faces fe having a number e of edges for the most compact
packings obtained with N = 75, 76, 77.
1.0 . - - - - . - - - - , - - - - - , - - - - - , - - - - - , - - - - - - ,

0.8 075
e76
on

0
0
0.6


0.4

0.2
0
0

§ ffl
0.0
2 3 4 5 6 7 8
e
Figure 6. Fraction of Voronoi cell faces withe-edges, /e, as a function of e for spheres
packings on S 3 with N = 75 (open circles), 76 {filled circles) and 77 (squares).

As one can see, these histograms are more peaked for N = 76 which
581
corresponds to a maximum of the c( N) curve.
It is also worth noticing that the peaks are dissymmetric in N since it
is more easy to create a vacancy than to add an extra hard sphere to a
compact structure. In particular, all the most compact structures obtained
from N = 111 toN= 119 turn out to be the polytope {3,3,5} with a given
number of vacancies. As a consequence the upper c(N) curve is linear in
that range of N values.
To investigate the large-N limit, we have performed a series of calcula-
tions for N values of the type 2n with n ranging from 8 (N = 256) to 13
(N=8192), Note that such N values have been chosen to get a set of reg-
ularly increasing values (geometric progression) rather than for technical
reasons. The results for the the volume fraction are reported as a plot of c
versus 1/N in figure 7.
0.67


0.66

(.)

• •
0.65

••

0.64 '---~---'---~--L._-~--'---~---'
0.000 0.001 0.002 0.003 0.004
1/N

Figure 7. Packing fraction as a function of 1/N for Iarge-N sphere packings on S3.

One obtains an infinite-N extrapolation of about 0.647 in good agree-


ment with the known estimates of the Bernal limit. Note that there is no
way to imagine a convergence to the FCC value.
The results of the Voronoi: tessellation also show a nice convergence to
the Bernal packing. In figure 8 we give the results for the coordination
number z, which is the mean number of faces of the Voronoi: cells, as a
function of 1/N.
One obtains a quasi-linear plot which gives z=14.22 as the coordination
number of the "ideal" (infinite) Bernal packing. Such a value is consistent
with values of z obtained from finite Bernal packings in flat space (see
582

14.5 ,---~---.---~--,---~---.---~-----.,

••
14.0 •

N

13.5


13 ·8.o'-oo-~--o-.o'-o1-~--o-.o'-02-~--o-.oo'-3-~--o--'.004

1/N

Figure 8. Coordination number z as a function of 1/N for Iarge-N sphere packings on


Sa.

figure 4). Note that other algorithms based on tetrahedra packings (which
do not deal with hard spheres) lead to z values of about 13.4 [17, 19, 20] in
close connection with the statistical honeycombs of Coxeter [18]. It would
be interesting to understand whether these two values (14.2 and 13.4) are
related.

3.3. INFLUENCE OF THE INITIAL CONFIGURATION

In this section we analyze more carefully the influence of the initial random
configuration of points considered in the JT algorithm. For three values
of N, namely N = 128,256 and N = 512, we have ran the code fifty
times with independent initial configurations and we have established the
histograms h(c) defined such that h(c)dc is the number of resulting values
for the packing fraction lying between c and c + de. These histograms are
reported in figure 9.
For N = 128, the histogram exhibits two peaks, the upper one, located
around c ~ 0.665-0.67, being stronger than the lower one, located around
c ~ 0.65 - 0.66. For N = 256, the histogram exhibits again two peaks,
but the lower one is now slightly more intense than the upper one, and
finally, for N = 512 it only remains a single narrow peak located close to
the Bernal value. This means that the chance to find the most compact "or-
dered" solution decreases dramatically as the size of the systems increases,
while the chance to obtain the "frustrated" Bernal-like solution increases.
583
400.--~---.--~--.----~---.--~-----,

300 -128
- - - 256
-512

* 200

100 /\
/ \
I \
\ I \
' ' - - - -../ I

oL----L~~==~~~~~~~-)
0.64 0.66 0.67 0.68
c
Figure 9. Histogram of the c values obtained when running the Jodrey-Tory code fifty
times for Iarge-N sphere packings on 83.

These results might be interpreted as follows: there always exists a most


compact solution for each N value, which tends to 0.74 as N tends to in-
finity (for practical realizations of such solutions see [21]), but the paths in
the configuration space to reach it with the JT algorithm become so rare
that there is no chance to get it when starting from a random configuration
of points.

4. Configurations generated by molecular dynamics


4.1. PRINCIPLES OF THE CALCULATIONS
In our molecular dynamics calculations we have used the repulsive poten-
tial of Laird and Schober[22] which has been proven to be appropriate to
describe the vibration spectrum of glasses [23, 24]:

U(r) = t:(~) 6
r
+ Ar4 + B for r < 3a (9a)

U (r) = 0 for r ~ 3a (9b)

In this expression a is a typical length fixing the range of the potential


and the constants A and B have been chosen so that both the potential and
the force become zero for r = 3a. In our simulations we have used N = 1000
atoms with a density such that (N/L 3 )a3 = 1. This would correspond to a
packing fraction of 52.4% for hard spheres of diameter a. A few simulations
584

have been performed on larger systems with the same density to verify that
our results are size independent. In order to give some physical meaning
to the simulations and to be able to give the temperatures in Kelvin, we
have chosen e7 = 3.405A and € = 0.0103eV, which are the Lennard Jones
(LJ) values of Argon[25] (of course we do not pretend to simulate real
Argon in this way). Similarly we took the atomic mass of Argon (m = 40
amu) and consequently we used a time step ./).t = 0.004r = w- 2 ps where
T = (me7 2/E) 112 is the standard LJ unit of time. The glass configurations
have been obtained by quenching a well equilibrated initial liquid sample
obtained by melting a simple cubic crystal at a temperature of about 50K,
well above the melting temperature. After full equilibration of the liquid
the system has been cooled down to zero temperature at a quench rate of
1012 K/s which was obtained by removing .!).E = 8.6 w- 7eV from the total
energy of the system at each iteration.

At several temperatures during the quenching process the configura-


tions (positions and velocities) were saved. Each configuration was used to
start a constant-energy molecular-dynamics calculation during which the
temperature was recorded as a function of time. In all cases, we have ob-
served a relaxation process typical of such a system (Fig.5 in [26]) during
which a slight increase of the temperature was observed before a satura-
tion regime was taking place. We found that typically 10000 iterations (i.e.
100 ps) were enough to insure a well defined constant temperature for each
sample. After these 10000 relaxation steps we have started to calculate
all the physical quantities reported below which have been monitored dur-
ing 20000 additional steps. It is worth mentioning that during the total run
time of 30000 steps (= 300 ps) we have never observed a strong tendency to
crystallization. We now know that this comes from the choice of the quench
rate which insures a maximum stability of the resulting glass samples[27].

Standard physical quantities such that the two-point correlation func-


tion g(r) and its Fourier transform S(q), as well as the diffusion coefficient
have been reported elsewhere [28]. They all exhibit the expected general
behavior for such amorphous systems [29, 30]. The glass transition was
determined as usual by following the temperature evolution of the diffu-
sion coefficient (calculated over the 30000 steps mentioned above). These
results show a clear change of behavior between a very small constant
(< 5.10- 3 A 2 .ps- 1 ) and a linearly temperature dependent diffusion coef-
ficient at a temperature T 9 = 10.5K (as estimated by extrapolating the
high-T linear regime) in perfect agreement with a previous study using the
same potential[22].
585
4.2. RESULTS
In figure lOa and lOb , we report on the cell-volume and cell-surface distri-
bution functions (in reduced units a= 1) hv and h 8 , respectively, resulting
from an average over 20000 timesteps after the relaxation period of 10000
timesteps at each temperature, as explained in section II-B.

15.0 .---~--~--~----, 5.0

4,0

10.0
3.0

.s::."
2.0
5.0

/,'
.1 1.0
//
.I,'
·" /
0 .0 --~-:::~,/
0.80 0.90 ' .00 1.10 1.20
v s

Figure 10. Cell volume hv(v) (a) and cell surface hs(S) (b) distribution functions for
MD atomic configurations. Dot-dashed, dashed, dotted, and solid lines correspond to
T = 29.7, 14.5, 9.0 and 0.3 K, respectively.

With decreasing temperature, the distribution hv is more and more


peaked around the mean cell volume Vm, which is here precisely equal to 1
(i.e. a 3 ), due to our particular choice of the density. The same qualitative
behavior is found for h 5 , except that, now, the mean surface Sm (close to
the location of the maximum) is slowly varying with temperature.
The corresponding standard deviations av and a 5 have been plotted as
a function of temperature in the log-log plot of figure 11.
It is worth noticing that both av and a 8 are varying like vfT in the liquid
phase, as expected for classical thermally activated density fluctuations.
But both quantities are much more slowly varying in the glass phase, and, as
T vanishes, they tend to non-zero values, characteristic of spatial disorder.
Such a low temperature saturation of the density fluctuations which occurs
around T9 is the signature of the glass transition [29, 30].
The temperature dependence of the mean cell surface sm is shown in
figure 12. For comparison, the surface sd of the regular dodecahedron of
volume 1 is indicated by the filled circle.
The data can be interpreted as if , when decreasing the temperature
from the liquid phase, the mean surface would like to reach the one of the
dodecahedron, but, since it is impossible to fill the space with dodecahedra
586

/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/

~--------------------~/// /
/
/
/
/
/
/
/
/
/
/
/
,/

10
T[K]

Figure 11. Standard deviations Uv (open circles) and us (open squares) as a function
ofT for MD atomic configurations. The dashed lines indicate T 112 behaviors.

T[K)

Figure 12. Mean cell surface area Sm as a function ofT for MD atomic configurations.
The filled circle at T = OK indicates the surface area for a perfect dodecahedron of unit
volume. The dashed line extrapolation is a guide for the eye.

only, a glass transition takes place and below T9 the mean surface saturates
towards a value higher than sd as T goes to zero.
587
The same qualitative behavior can be observed for the mean coordina-
tion number z =< F >in figure 13.

14.5

14.0
I

13.5
N

13.0

12.5

12.0 ....._-~---''--~----'--~---'---~--'---~---'
0.0 10.0 20.0 30.0 40.0 50.0
T[K]
Figure 19. Mean coordination number z as a function ofT for MD atomic configurations.
The filled circle at T = 0 indicates z = 12. The dashed line extrapolation is a guide for
the eye.

When decreasing the temperature from the liquid phase z decreases,


but saturates below T 9 to a value of about 14. As in figure 4, one observes
however that the convergence to z = 12 of the T > T9 data is quite slow.

5. Conclusion
In three examples of glassy systems, namely sphere packings in flat space,
sphere packings in curved space and atomic configurations generated by
molecular dynamics, we have shown that the Vorono1 cell tessellation is a
very useful tool to investigate the mechanisms of geometrical frustration.
We have found striking similitudes when increasing the volume fraction
in sphere packings and decreasing the temperature in molecular dynamics
samples: in both cases the intensity of the local density fluctuations ex-
trapolates to zero for a pure icosahedral structure with five-fold symmetry.
The impossibility to reach such a structure, as a consequence of geomet-
rical frustration in the regular flat space, is shown by the existence of a
limiting volume fraction (the Bernal's threshold) in case of sphere packings
and by the occurrence of a glass transition in the case of molecular dynam-
ics samples. These results support a recent theory for the glass transition
in which the perfect icosahedral order would be a hidden fixed point [9].
588
Moreover, by showing that the Bernal packing is nicely recovered when in-
creasing the number of spheres of random packings built on 8 3 , we provide
an alternative way to reach the thermodynamic limit in numerical meth-
ods dealing with large three-dimensional systems. Instead of working in a
fiat space within a cube subject to periodic boundary conditions {which is
topologically equivalent than working on a torus), and trying to reach the
thermodynamic limit by increasing the edge length of the cube. one can
instead work on a sphere 83 and increase its radius.
Part of the numerical calculations were done at CNUSC (Centre Na-
tional Universitaire Sud de Calcul), Montpellier, France.

References
1. J. Bernal, (1964) Proc. Roy. Soc., A 280, 299.
2. J. L. Finney, (1970) Proc. Roy. Soc., A 319, 479.
3. G. H. Bennett, (1972) J. Applied Phys., 43, 2727.
4. D. J. Adams and A. J. Matheson, (1972) J. Chern. Phys., 56, 1989.
5. G. S. Cargill, (1972) J. Applied Phys., 41, 12.
6. J. D. Bernal and J. Mason, (1990) Nature, 188, 910.
7. J. F. Sadoc and R. Mosseri, (1997) Frustration Geometrique, Eyrolles ed. and Alea
Collection, Saclay, France.
8. J. F. Sadoc, J. Dixmier and A. Guinier, (1973) JNCS, 12, 48.
9. D. Kivelson, S. A. Kivelson, X. Zhao, Z. Nussimov and G. Tarjus, (1995) Physica,
A219, 27.
10. H. S. M. Coxeter, (1973) Regular Polytopes, Dover.
11. R. Jullien, P. Jund, D. Caprion and D. Quitmann, (1996) Phys. Rev. E, 54, 6035.
12. P. Jund, D. Caprion and R. Jullien, (1997) Europhys. Lett., 37, 547.
13. R. Jullien, J. F. Sadoc and R. Mosseri, (1997) J. de Physique(France), in press.
14. W. S. Jodrey and E. M. Tory, (1985) Phys. Rev. A, 32, 2347.
15. F. Sadoc and R. Mosseri, (1989) Aperiodidicity and order 3, Extended icosahedral
structures, p. 163, Academic Press, New York.
16. N. Rivier, (1993) Disorder and granular media, p. 55, ed. by D. Bideau and A.
Hansen, Elsevier Science Publishers, North Holland.
17. D. R. Nelson and F. Spaepen, (1989) Sol. State Phys., 42, 1.
18. H. S. M. Coxeter, (1958) Illinois J. of Mathematics, 2, 746.
19. N. Rivier, (1982) J. de Physique {France}, 43, 91.
20. R. Mosseri and J. F. Sadoc, (1990) Geometry in condensed matter physics, p. 239,
ed. by J. F. Sadoc, World Scientific, Singapore.
21. R. Mosseri and J. F. Sadoc, (1990) International J. of Space Structures, 5, 325.
22. B. Laird and H. Schober, (1991) Phys. Rev. Lett., 66, 636.
23. H. R. Schober and C. Oligschleger, (1996) Phys. Rev. B, 53, 1.
24. D. Caprion, P. Jund and R. Jullien, (1996) Phys. Rev. Lett., 77, 675.
25. D. Gazzillo and R. Della Valle, (1993) J. Chern. Phys., 99, 6915.
26. P. J. Steinhardt, D. R. Nelson and M. Ronchetti, (1983) Phys. Rev. B, 28, 784.
27. P. Jund, D. Caprion and R. Jullien, (1997) Phys. Rev. Lett., 79, 91.
28. P. Jund, D. Caprion and R. Jullien, (1997) Mol. Sim., in press.
29. P. H. Gaskell, (1991) Materials science and technology, p. 175, ed. by R. W. Cahn,
P. Haasen and E. J. Kramer.
30. N. E. Cusak, (1987) The Physics of structurally disordered matter: an introduction,
Adam Hilger, Bristol and Philadelphia.
INDEX

(M,£,0)-minimizing sets 386 CIS structure (see also Laves phase)


3-periodic minimal surfaces 448 397,410
Al5 structure (see also Weaire-Phelan capillary number 240, 244-6
foam) 265-6, 269-70, 273-4, capillary pressure - in porous media:
278,280-4,397,410,427,508 cf. osmotic pressure
ABAQUS (program for finite element capillary pressure 361-2
analysis) 280, 283 capilllary suction 24, 26
Aboav law (or Aboav-Weaire law) 98, catenoid 446
108-9, 137, 146 cavity growth 165, 565-7
address 481, 488-9 cellular dislocations 165
adsorbed layer 25 cellular solid: cf. solid foam
adsorption energy 4 CFC 321
aging : cf. Ostwald ripening chemical reagents 324
alkali silicides 409 Choe foam 392
alkane 38, 39 clathrate hydrates 396, 409, 413, 423
amphiphiles 471 close-packing (see also TCP) 457
anharmonicity 217 closed cell foam 175, 283-4
antifoamer 15, 21, 37 coalescence (of droplets or bubbles)
aperiodic tiling 481, 483 35, 129, 132, 142, 149
Archimedes' Theorem 444 coarsening (non-Ostwald) 137, 139,
Aste-Boose-Rivier (ABS) polyhedron 141-4, 149
packing 406 coarsening: cf. Ostwald ripening
Avrami-Johnson-Mehl: see Johnson- cocycle condition 387
Mehl tessellation colloidal glasses 224
Bancroft rule 12, 27 colloids 22
BCC lattice 390, 394, 397, 399, 427 combinatorics of foams 388
bending 178 common black film (CBF) 15, 26, 60
Bernal packing (-limit, -threshold) conductance measurements 308, 310
572, 580-2, 587 conductive segmented probe 311
bias correction 551 conformal map 473
bicontinuous structures 437 constant mean curvature (CMC) surface
bicontinuous topology 523 382
bilayer 471, 521 constitutive equation 231
biomaterials 196 contact angle 29, 36
Biot number 241, 249 contact line 28
black film: see common - (CBF), control parameter 137-8, 145, 148
Newton - (NBF) Conway tiles 486 ·
bone 193-197, 199 coordination number 505-6, 508-9
Bonnet transformation 453, 474-6 coordination polyhedra 407
Brakke K. A.: cf. Surface Evolver correlation function 576, 584
breaking: cf. foam breaking, film correlations 91
rupture creaming 41
bridging (of films) 34, 36-7 critical micelle concentration (CMC)
bridging: dewetting mechanism 37 27, 250-4
bubble cluster 385 crystals of films 471
bubble disproportionation 23 cubic lattices 462
bubble size distribution 326-30 cubic phases 423, 462, 471
bulk modulus 304
589
590

curvature (Gaussian) (defect as a drainage 303, 305


concentration of-) 111-3, 516 drainage pulse 311-2
curvature 439-40, 511 drainage: cf. film-, foam-
curvature : see Gaussian curvature, drainage: vertex corrections 313
mean curvature drop entry 32, 39
curved space 511, 577 droplet (microemulsion) 27
D surface 449, 474-8 dry foam 22,261-2,265-74,290,360,
Darcy's law 336 362, 364, 366-70
Debye length 53 dual cell complex (see also Delaunay
decontamination 323-4, 326, 328, 330, simplicial tessellation) 388
334 duplex film 32
decoration theorem 292 dynamic surface tension 9
defoamer 37 dynamics 303
deformation 154, 471, 476-8 dynamics algorithm 531-3
deforming interfaces 246-50 effective temperature 226
Delaunay simplicial tessellation 387, elastic modulus 304
406,574 elastic response 304, 305
Delaunay triangulation 389, 554-6 elasticity (of foams) 259-60,262-5,
Delone triangulation: cf. Delaunay 268-84, 304-5
triangulation elasticity number 240
dense random packing of hard spheres electrical conductivity 308, 310-1, 314,
513 315
desorption energy 4 electrostatic force 15, 53
dewetting 127-8, 133 elementary topological tranformations
diffusion - in foam in porous media (see also Tl, T2) 105-7, 553,
343 557
diffusion 349, 351, 355, 357, 359 embedded triply periodic minimal
diffusion resistance 220-1 surfaces (ETPMS) 448,450
dihedral angle 391 embedding 472-4, 478
dimple 87, 89-90 emulsion, oil-in-water-: see inversion
dimpling instability 12 emulsion, water-in-oil -: see inversion
disclination 408, 516 Ennepersurfaces 446
disclination line 408 entry coefficient 32-3
disjoining pressure 13, 15, 25, 45-72, equation of motion 556-7
75,220-1 equations of state (see also Aboav,
disjoining pressure isotherm 26, 60 Lewis, Peshkin, Lemalitre laws)
dislocation 151, 162-3, 166, 173 105, 107
disorder (role of d. in foam elastiicity) equilibrium (statistical) 109
217 equilibrium 137, 140-6, 148-9,309,
dispersion forces 55 383,471
disproportionation: cf. bubble equilibrium: see also thermodynamic -,
disproportionation force balancing, Plateau rules
dissipation 218 equilibrium: static 309
DLVO (Derjaguin-Landau-Verwey- Euler's formula 96, 575
Overbeek) theory 52, 60, 62 Euler-Poincare characteristic 390,
dodecahedron (pentagonal) 395, 513, 441,498
585 event-driven simulation (ED) 531-2,
Domino Question 484 541-2,544
double bubble 387 evolution 137-8, 142-6, 148-9
double wave experiment 310 Evolver: cf. Surface Evolver
591

expanded glass 120-3, 563 freeze-fracture electron microscopy


expansion process 120-3 429
extensional flow 244 froth: cf. foam
faujasite 464 frustration 431, 443, 511-2, 520, 572,
Fd3m (space group) 427-32 582
ferrofluid 137-9 fundamental domain 392, 475-6
film drainage 24 G surface: cf. gyroid
film rupture 25, 37 gas-cooler 324
film tension 33 Gauss map 450
film tension: excess - 25, 32 Gauss theorem (topological) 111
film viscosities 221 Gauss(ian) curvature 380, 424, 426,
film: cf. common black -, duplex -, 440,473,476,498,502-3,505,
lamellae (films in porous media), 516-7
metastable-, Newton black-, Gauss-Bonnet theorem 110, 381, 391,
pseudoemulsion -, thin -, 441,499,505
thin liquid - (TLF) genus 441, 474, 477-8
finite element analysis 280-4 geometry of foams 287
flow curve 210, 225,231 Gibbs elasticity 220
foam breaking 21 Gibbs equation 3
foam conductivity 314 glass 225, 571
foam drainage 303, 305, 308, 311-2 glass transition 226, 584, 586-7
foam drainage equation 303, 305, 307- glassy systems: cf. glass
8,310 Goldberg cell, Goldberg froth 507
foam generation - in porous media 341- grain boundary 152, 153-61, 164
3 grain growth 152, 157-9, 565-7
foam microstructure 180 grains 152
foam stability - in porous media 343-5 granular flow 349
foam surface structure 300 granular materials (- media) 349, 528
foamability 37 great sphere 523
foamed materials (synthetic) 193, gyroid 448, 474-8
197-8, 200-5 helicoid 446
foaming process 117-21, 563, 570 Herschel-Bulkley model 225, 232
foaming transition 117-21 hexagonal (lattice, foam): cf.
foams, simple 389 honeycomb
force balancing (cf. equilibrium) 383 hierarchical structure 481-3
force: cf. dispersion-, electrostatic-, high molecular weight polyethylene
hydrophobic -, steric -, (HMWPE) 197-200, 202-4
supramolecular -, surface -, homogeneity index 445
stratification-, thin film-, van honeycomb (disordered): cf.
derWaals- polydisperse hexagonal
fracture 151, 161-2, 164, 167 structure, statistical -
Frank-Kasper coordination polyhedra honeycomb 167, 171
408 hydrodynamics, capillary slug flow
Frank-Kasper phases (see also TCP 252-3
structures) hydrodynamics, extensional flow and
Frank-Kasper structure (TCP) 408, drops 246-50
413 hydrodynamics, translating drop 237-8,
free drainage 311-2 250-51
free drainage: cf. foam - hydrophobic 17, 65
free energy (particle attachment) 31 hydrophobic forces 65
592
hydrostatic pressure, compression Lemaltre law 111-7
281-3, 432 Lemlich law (for foam conductivity)
hyperbolic surfaces 440, 454 314,315
icosahedron 514 Lemlich law - nonlinear corrections
ideal foam model 287 318
image analysis 323, 326-8, 330, 334 Leonard-Jones (U) potential 584
image treatment 551 Lifshitz-Slyozov model 101
infinite polyhedra 456, 461 Linde type A 464
inflation 489, 500, 503-4, 506 line tension 28, 30
instability 137-40, 477 linear viscoelasticity (1. rheology) 208,
intermetallic structures 414 229
inverse micellar cubic phases 427, lipid bilayer 423
429-31 lipids 423
inversion (in emulsions, for micelles, liquid crystals 423, 455, 471
etc.) 27,40 liquid fraction 306
inversion 292 liquid lens 31
isolated defect 558 liquid-crystalline phase: cf. liquid
isoperimetric quotient 369, 374 crystals
isoperimetric theorem 383 liquid-paste transition 122-3
isotherm: surface pressure-surface area logistic map 498, 502
31 low frequency behaviour (viscoelastic)
Jodrey-Tory algorithm 535-6,541-2, 218,229
573,578 low density poly ethylene (LOPE)
Johnson-Mehl tessellation 119, 134, 175, 177
567 lyotropic mesophases 423
Jullien algorithm 536 macroscopic 138, 149
Kelvin cell 175, 181, 183,390,500-1, magnetic dipolar repulsion 138, 144-5,
506 148-9
Kelvin equation 23 magnetic fluid (MF) 137-49
Kelvin froth 265,268-84, 316, 399, Major skeleton 408,410
403,405,409,427,500-1,506 Marangoni elasticity: cf. Marangoni
Kelvin's problem, Kelvin foam, Kelvin stresses
structure: cf. Kelvin froth Marangoni flow 222
Kerroc 105, 120-3, 563-4, 567-70 Marangoni stresses 25, 237-55
Kugelschaum (cf. wet foam) marginal regeneration 222
Laguerre tessellation 119, 540, 547, mass transfer (of surfactant) 237-55
553-6,565 matching rules 482
Laird-Schober potential 583 Matzke, E.B. 267-8
lamella/lamellae (films in porous maxent (maximum entropy) (see also
media) 335-45 variational principle)
Langmuir adsorption 238-9 maximum entropy inference 108-17
Langmuir trough 31 maximum principle (of minimal sur-
Laplace equation 23-4, 34, 238, 288, faces and harmonic maps) 381
307,472 mean curvature 288, 380, 424-5, 440,
Laplace pressure, Laplace law, see 472,476
Laplace equation mean-curvature flow 390, 400
Laves phases (see also C15) 411, 427, mean-field 226, 232
430 melanophlogite 464, 466
layer (of cells) 498-501 meshes 442
least dissipation 221 mesophases 471-2, 475
593

mesoporous solids 464 Pearson's handbook 410


metastability 32, 225 Peclet number 241, 249
metastable film 32 Penrose tiles 485
micellar cubic phases 424, 426 percolation 336-7, 341, 563-70
micellar topology 523 periodic minimal surfaces 423, 471- 3,
micelle (as surfactant) 250 476,478
micelle 27 periodic tiling 482
micromechanics (of foams) 259-84 Peshkin law 108-11
Miles-Lantuejoul's method 551 petroleum recovery 335-45
minimal path 341 Pickering emulsion 40
minimal surface (see also periodic m.s.) plastic deformation 151, 154, 161-3,
381, 472-3 303,305-6
minimality 471-3 plastic response 303, 305-6
modulus, bulk, or elastic- 210, 263-4, plasticity 260, 262, 270, 272-4, 277-8,
278-80,283-4,304 303,305-6
modulus, Plateau - 209 Plateau (equilibrium) rules 261-2,
modulus, shear - 259-60, 262-5, 268- 265-6,270-1,275,290,316,387
70, 275, 278-80, 283-4 Plateau border 10, 21, 24,263,274-7,
modulus, Young- 175, 185,264,278- 279-80, 338, 386
84 Plateau border geometry 317
modulus: see also Poisson ratio Plateau border section 178
molecular dynamics (MD) 529, 531, Poiseuille flow 221, 305, 307, 314
533,535,544,572,583 Poisson ratio 185, 187, 306
monolayer 31 Poisson-Boltzmann equation 53
monomer (as surfactant) 250 polybenzene 465
Nax Si clathrates 413 polycrystal 96
net density 463 polydisperse hexagonal structure 261-
network (of droplets, bubbles, bound- 4
aries ... ) 41-2, 127, 129, 165 polydispersity 260, 262-70
network (of polyelectrolyte chains) 76 Polyederschaum (cf. dry foam)
network (polygonal). cf. tessellation polyelectrolytes 73
Newton black film (NBF) 15, 26, 60-1 polyethylene 193, 197-8
Newtonian liquid 232 polymer concentration 180
no-slip boundary condition 25, 314 polypropylene (PP) 175
no-slip condition 314 polystyrene (PS) 175
nuclear waste 323 polystyrene latex particles 41
oil drops 21 polytope {3,3,3} 580
oil-particle synergy 36, 39 polytope {3,3,4} 580
open cell foam 168, 278-82 polytope {3,3,5} 515, 572, 580-1
optical tomography 359, 361, 363, 377 polytope {5,3,3} 515
osmotic pressure - in porous media polyurethane foam 321
336-7, 342-4 porosity 563-4, 567-8, 570
osmotic pressure 208, 291 porous media 305, 335-45
Ostwald de Waele fluid 333 porous medium theory 305
Ostwald ripening 10, 23, 95, 226, 229, Powell algorithm 541, 544
359-62,364,372,375-7 power low fuid (non-Newtonian) 225,
P surface 449, 474-8 232
packing (random) 506,513,572 pressure: see under qualifier
packing fraction 431,528-9,531 Princen model (2D foam viscoelastic-
paste 224 ity) 210, 262, 275,278
594

pseudoemulsion film 31, 39 space -filling packings 405


pulsed drainage: cf. foam - sphere packings 403, 418-9, 460, 511,
quasi-static 303 513, 575, 578-9
quasi-static flow 303, 309 spherical manifolds 392
radical tessellation : cf. Laguerre spherical torus 523
tessellation sponges 437, 456
random sequential adsorption (RSA) sponges, homogeneous 443
529-30, 541, 544 spreading coefficient 32
refinement construction 392 spreading kinetics 33
relative density 175, 186 stability (of foam, emulsion) 22, 37
relative permeability in porous media stagnant interface 242-3
335-45 Stamenovic (- relation between elastic
remobilization (of surfactant) 249-54 moduli) 305
Reynolds law 84, 87-8 starch 193, 197-8
rheology - foam in porous media 338- static equilibrium 309
41 stationary surface 523
rheology 259-63, 268-78, 303, 331, statistical equilibrium 109
334 statistical honeycomb (Coxeter) 582
rheology, non-linear 225,231 steady drainage: cf. foam -
Riemann surface 473-4 steady-state drainage 308-9
ring size 416, 418 steady-state flow: cf. quasistatic -
saddle towers 447 steric forces 58
Sadoc-Mosseri (SM) structure 413 stochastic geometry 127, 129
schwarzite 464 strain energy 260, 262-3
sedimentation 41 stratification 520
segregation 349,350 stratification force 73
selection mechanism 101 stress conditions at interfaces 237-55
self-organisation 496 stress tensor 385
shear 262, 272-4, 277-8, 305-6 stress-strain curve 189
shear thinning 225, 231 structure (foam-, cellular-) 137-9,
shell map 497 141, 147-8,259-61,265-70,274-
shell-network 501, 506-8 5,278,283,287,359-60,364,
shell-structured foam 109 373
shell-structured inflatable foam (SSI) substitution graph (finite state
109-110,500,502-4,506,519 automaton) 489
silicon oil 38 substitution rules 482
skeleton 499, 500, 502 substitution tiling 481-2, 495
slow modes (slow dynamics) 218,225 supercritical (bubble, crack) 119
slurries 224 supramolecular forces 62
Smoothed particle hydrodynamics Surface Evolver 266, 274, 295, 298,
(SPH) 536-9,544 317,341,364-5,378,390
soap film 382 surface elasticity 5
sodalite 397, 403, 404, 419, 464 surface energy 472, 473, 476
sodalite (N-dinensional) 419 surface excess concentration 3
soft glassy materals 224 surface or colloid forces 25, 47
solid foam 120-2, 259-60, 263-5, 278- surface pressure 2, 31
84,563-5 surface pressure: see also surface
solitary wave 309, 312 tension
solitary wave (in foams) 309, 310, 312
solubilisation (in films) 38
595

surface tension (see also interfacial ten- virial equation of state (for foams)
sion) 2, 32, 96, 221, 237-55, 113-4
472 viscosity 5, 122-3, 221, 254-5
surface tension - non-linearity 246 Volterra process 516
surface viscosities 254-5 von Neumann law 96, 107, 137, 144-5,
surfactant 24, 237-55, 345, 423, 471 149,361,375,377
surfactant adsorption on solids 345 Voronoi decomposition: cf. Voronoi
T phase (Bergman) structure 412 tessellation
Tl (elem. topological transformation) Voronoi polyhedron 407
142-3, 144, 147-9, 262-3, 266, Voronoi tessellation 134, 262, 266-8,
270-8, 359-60, 371 283-4, 349, 352-3, 356, 358,
T2 (elem. topological transformation) 389,527,539-40,544,563,565,
359-60 567, 571-2, 574
TCP (tetrahedrally close-packed) Weaire-Phelan foam: cf. A15 structure
structures 265-6, 268-70, 273, Weierstrass-Enneper (WE) formula
278,280-4,394,408,413,508, 473-4,478
519 wet foam 22, 263, 274-8, 290, 298,
TCP structures, basic 396 362,366-8,370-2,377
Telley's model 119, 547, 553, 555 wettability 21
tension gradient 25 Williams structures 405
tension: cf. surface tension, film Willmore energy 381
tension, etc. X-ray diffraction 428-9
tessellation 527, 539-40, 565-7, 569-70 yield stress (y. energy, y. strain) 210,
tessellation: cf. Delaunay simplicial -, 227,229,232,303
Johnson-Mehl-, Laguerre-, yield stress - foam in porous media
Voronoi- 339-41
tetrahedral junctions 313, 316 yield(ing) 187, 210, 260, 262, 270,
tetrahedral packings 406 272-4, 277-8
tetrakaidecahedral cell; cf. Kelvin cell Z structure 508
thermal conductivity 315, 318 zeolite 397, 416, 463
thermodynamic equilibrium 471
thermodynamics of foams 105
thin film balance 66
thin film forces 25, 33, 66
thin liquid films (TLF) 83, 85, 88-9,
127,289
topological charge 498
topological defect 499, 502-3
topological distance 499, 504
topological interaction 136, 148
triangular tiling 512
truncated octahedra: cf. Kelvin cell
twisted Kelvin 405
two-dimensional (2D) 105, 137-42,
144-49
unduloids 383
valence, average 390
Vander Waals force 15
velocity profile 25
Photographs and artwork

Frontispiece Weaire-Phelan foam, by J. Sullivan


Foreword Basal layer of human epidermis, by B. Dubertret
103 Giant's Causeway
104 N. Rivier
126 U. Thiele I N.Rivier, N. Mills
136 F. Bouchama I T.Aste
150 F. Elias
174 Granular media and geometry
236 P. Sollich
258 K. Stebe I C. Dionisio-Eggleton, S. Datwani, J. Ferri
286 A. Kraynik
322 R. Phelan, G. Verbist
348 G. Verbist, B. Gardiner, J. Uhomoibhi, J. Lemaitre,
A. Abdelkader, B. Dubertret, N. Pittet
378 C. Monnereau, N. Pittet
422 Kelvin foam, by J. Sullivan
435 Positive and negative curvature
436 Cartoon, by Michael Leunig
470 M. O'Keeffe, S. Hyde
480 wet foam, by G. Boissonnet
526 J.-F. Sadoc
546 D. Weaire I R. Phelan
562 N. Pittet

You might also like