Академический Документы
Профессиональный Документы
Культура Документы
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
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
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.
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
A C.i.P. Catalogue record for this book is available from the Library of Congress.
Foreword xiii
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
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
Index 589
FOREWORD
xiii
xiv
Jean-Fran~ois Sadoc
Nicolas Rivier
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.
1. SURFACE ENERGY
+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)
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.
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
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)
(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
=
(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:
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)
2. SURFACE RHEOLOGY
(15)
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)
1 +
E =EO - - - - - - -
o (18)
1 + 20 + 202
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
(20)
2. FOAM STABILITY
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.
(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)
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
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)
(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
Conclusion
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
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
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.
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.
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.
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.
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
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
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.
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.
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.
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]
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].
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)
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
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.
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.
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.
(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.
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.
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.
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.
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
VANCE BERGERON
Rhone Poulenc Industrialisation
85, Av. Des Freres Perret- BP 62, Saint-Fons Cedex, France
1. Introduction
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.
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)).
(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.
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:
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
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.
(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)
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
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.
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,
(12)
55
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),
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).
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)
(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).
(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
where ~= kT/~ is the protrusion decay length, n is the density of protrusion sites, and ~ is
an interaction parameter (J /m).
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.
\
\ lldl- exp(-Kh)
+ \.
''
'
-
II '
0~~~~------~~~----~~
.,..,....... h
llsteric /
/
/ llyan- 11 ~
/
I
t t
Newton black films common black films
-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
-
"""'... ltll '
4
'
I I
~
Ill I I'
"\ ,
q I JLJ.,W
""-''' UJ I' I .UJU U I' I JJ,H£W
~
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).
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:
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).
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
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.
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),
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)
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
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
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/!.
-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.
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)
I··-~
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.
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
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
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
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
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
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
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
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].
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.
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.
->-<
Figure 4- Creation of new vertices in which only three films meet, as a result of bubble
disappearance.
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
$=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
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
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)
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.
2. Two-dimensional foams
3V =2E =<n>C ,
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.
3.1. 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
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],
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
(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,
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).
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)
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].
(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
The following procedure is standard maximum entropy inference [21]. One ma-
ximises the entropy
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.
(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
(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 ..
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,
(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)
(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
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)
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.
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.
Closed
Open
I
I
Time
Figure 6 . Phase diagram ofthe expanded foam Kerroc (N. Pittet [29])
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
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,
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 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
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
- 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
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)
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].
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
~··~ (b)
~ r""'
~·.
em
....
r"
f·
I~ -:7 · ·~I
I ~ ~ '~
@H
~
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
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.
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.
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
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
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),
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.
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
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
!.Introduction
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
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.
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
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.
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.
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(-)
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
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
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
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.
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.
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 '
'
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].
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
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
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.
Dislocations produced during plastic deformation are not randomly distributed. They
organize into cell structures as shown on figure 14.
Figure 14
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.
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.
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.
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
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).
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
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).
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]
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).
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
Figure 23
100
80
..
~ 60
!
40
zo
Compression Direction L
00 20 40 60 80 100
Deformation (%)
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].
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 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
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
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
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
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.
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)
l
the foam is initially atmospheric Pa· Using equation (3), the relative pressure p, is
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
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
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).
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].
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
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
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
TABLE I. Pore size of different structures obtained by injection moulding ofHMWPE using different amounts of
blowing agent (BA2 ).
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.
BAI-blowing agent trade name Hostalen P 9937 BA2 - blowing agent trade name Hostalen P 994 7
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
......-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
REFERENCES
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
2. BASIC RHEOLOGY
2.1. LINEAR VISCOELASTICITY
Go= (fPF)
812 v
(2)
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.
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
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.
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
(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
1 (1-
Fey!= 21ruR [ <Po¢- ~
¢0)1/2 (1-
-¢-
¢)1/2] (8)
(9)
Til( /R)
2
1.5
0.5
(10)
(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
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.
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
disc
unit
cell
> 2L'
2L
2L"
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:
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
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
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)
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.
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"
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).
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
4.4. A THEOREM
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.
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.
With the above ideas about water and surfactant flux patterns, we can
identify the following candidate modes:
224
!: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.
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.
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.
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:
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
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.
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
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.
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
/
/
Figure 10. Linear moduli G' (solid line) and G" (dashed) vs frequency w at various
noise temperatures x above the glass transition.
=
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.
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 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.
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
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
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
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.
(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
(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:
<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
(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].
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
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
\
rr i
R'T['.
\
\
-" .......... ,
---------·-------- --
1.0 I." L2 l.J I_L 1.~ 1.& U 1.9 1.9 2.(
-
1
X
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.
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
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
~~·:::::
_//
_____ __ _,_ .. _. _____ __....._,. ............... .-......... .....__ .... _., ...... __ , ____ ~ ... . . _..
, ,.~
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
Figure 6c. Marangoni stress profiles for dilute (x=O.l) and high (x=0.996) surface concentration.
249
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].
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
*
.~~
·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.
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
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'
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.
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
2. Theoretical Approach
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
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
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
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
Figure 2. Kelvin cell, three TCP structures, and two random foams.
4. Three Dimensions
4.1. LIQUID FOAMS IN THE DRY LIMIT
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
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
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
(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.
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
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.
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
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 .
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.
0.6
~
-> 0.5
i===
<I) 0.<4
<I)
~
U5 0.3
:;;
Q)
.t::. 0 .2
en
0 .1
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.
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
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.
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.
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.
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.
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
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
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.
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.
Figure 17. Shell element meshes for closed-cell Kelvin and random foam with N=lO.
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
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 •
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.
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.
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
Figure 3. An illustration of Plateau's equilibrium rules. Cell films meet at 120°and lines
meet at arccos(- t) ~ 109.5°.
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.
II=_ (8F)
av v 9
(3)
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:
(0,1)
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.
8. Simulation
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.
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
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.
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.
Rl 5.30874
R2 5.30690 5.30608
R3 5.30643 5.30627
R4 5.30632 5.30627
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.
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
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).
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.
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.
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
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.
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.
u = ~ (pg - ope)
ox
' (3)
!'TJP.
0 0
ot A(x, t) + OX (A(x, t)u(x, t)) = 0 (4)
308
(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).
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
.. .
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
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.
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) .
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 ·
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
0 20 40 60 80 100
E
,.....,.....,1
- 0 0.00.1
g.
C1)
Q./)()2
-;;;
0.001
O L--4~
0 --~
W~~80~~~~QQ---1~
20
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.
0.08
0.06
0.04
0.02
0 0 10 20 30 40
segment# [~em]
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
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 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).
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
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.
{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
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
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
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.
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
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
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.
Nevertheless, the drifts only have a minor impact on surface decontamination and the
use of the CCD camara remains relevant.
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
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)
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
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.
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
;;.1.4
I·
'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
Bibliograpic references
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.
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
(1)
a) 1 b)
:;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,%
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)
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.
"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.
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
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
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].
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."
time - - - - - - - - - - - - - - - - - - - - -
Figure 7. Schematic of lamella division as
a lamella passes through a pore body.
liquid film
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.
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.
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
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
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].
Steel baU,R
numerical studies described in this paper deal only with the second regime
where the ball attains a steady state.
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).
! 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.
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).
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) .
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.
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.
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
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 :
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].
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
porous plate
reservoir
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)
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
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
2.3 PROCEDURE
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 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
drainage
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.
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]:
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.
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
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.
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).
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).
<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.
5.1.3 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
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.
(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
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.
JOHN M. SULLIVAN
University of Illinois
Department of Mathematics
Urbana, IL, USA 61801-2975
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.
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.
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.
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.
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.
>
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.
211'
n = no := 1 ~ 5.1043
arccos 3
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
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
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.
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.
•
•
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.
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
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
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.
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.
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".
Fig. 5. Left: two views of the ABR polyhedron. Center and right: the corresponding space-filling packing.
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.
~~~$··
€?®@~$~
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
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.
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.
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
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.
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.
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. 20. The "3-dimensional sodalite" structure showing a primitive unit cell (left) and the 3-dimensional
bubble (right).
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
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.
(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
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.
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].
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).
,::
~
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
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].
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.
6 • References
,
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
Introduction
,. 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
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.
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.
• Once this paper is digested, examples will, I hope, spring to the reader's mind immediately.
441
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.
Homogeneous sponges
1 2 3
VOLUMES:
V x =A 0 • x( 1+ Hx + ~2 )
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 ~)
(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:
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 8: (Left) Saddle tower constructed from stacked J-Enneper surfaces. (Right) Scherk's
surface, a 2-periodic minimal surface.
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.
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.
(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).)
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).
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).
(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·
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.
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 )
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:
lumnotropic liquid
rystals:
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)
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.
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 }.
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.)
or,
cosh(;)= tcosec(~).
This allows the area of each disc to be determined (5), viz.
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.
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
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
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
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.
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)
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).
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
C. OGUEY
LPTM, Universite de Cergy-Pontoise
2 av. A. Chauvin
F - 95302 Cergy-Pontoise
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
E = uA = u !M dA, (1)
(2)
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.
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
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 ...
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.
Figure 4- Continuous rhombohedral distortion of the P (left) and the D (right) surfaces.
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
1. Introduction
cal" as they can only produce aperiodic hierarchical tilings. Precise defini-
tions of these all of these terms are in [6] and elsewhere.
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
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].
=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-:
D i' l '
' \,1,1
I
I\
'l_l_~"j
l'i
...,
EE
' \ 1111 I~ ~
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
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
l\
Figure 9. Addresses
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
3. Keys
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.
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
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
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.
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
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.
r
I
\
A
I~
I
\
I
\
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.
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".
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]
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.
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.
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).
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.
/=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)
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).
5,35
5,30
decuNed
5,25 dodacah~s
Frank-Kasper,
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:
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.
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
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)
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.
b)
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
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)
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.
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.
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.
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.
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
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.
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.
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.
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
Abstract
1. Introduction
2. Computer simulations
... ..
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].
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.
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.
t
t.
1
(x;,Yi,S;) I I
0
Figure 3. Schematic representation of the contact between two particles.
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
--- {J(:x y
(t, x, y) = (ty M' D' D). (5)
(6)
By using Eqs. (1) and (3) with the dimensionless variables (5) and (6)
we can redefined all the equations of the system.
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.
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
and approximated as
j W(u,h)du= 1, (13)
and
lim W(u, h)= 8(u). (14)
h-+0
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)
(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.
3.1. 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
a)
b)
c)
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.
.'
IS.l
•
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.
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.
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
4. Conclusions
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
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
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 ... . ······• ...
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)
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).
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.
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.
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.
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.
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
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
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.
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.
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)
=
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.
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
.·
.... ...
•.• • '---~"'----.._~_......_ _ _..J
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
1. Introduction
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].
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
(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].
2.3. 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 .--~-~-~-~----,
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
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.
16.0 r-------,------.--~--.--~-----,
15.0 0
0
0
0
0
OC\:@
N 14.0
13.0
center of the space (sphere S3) to any sphere center i of the packing verify:
.. _
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
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.
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 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.
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.
U(r) = t:(~) 6
r
+ Ar4 + B for r < 3a (9a)
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.
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.
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
~--------------------~/// /
/
/
/
/
/
/
/
/
/
/
/
,/
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.
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
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