International Journal of Shap e Mo deling, Vol. 0, No. 0 1998 000{000
f
cWorld Scienti c Publishing Company
FOAMS AND BUBBLES:
GEOMETRY AND SIMULATION
JOHN M. SULLIVAN
Department of Mathematics, University of Il linois
Urbana, IL, USA 61801-2975
Received July 1998
Revised April 1999
Communicatedby Michele Emmer
The geometry of soap lms, bubble clusters and foams has a nice mathematical descrip-
tion, though many op en problems remain. Recently, computer simulations, and their
graphical output, have led to some new insights.
Keywords :Foams, soap lms, minimal surfaces, computer simulation.
1. Intro duction
Soap lms, bubble clusters, and foams and froths can b e mo deled mathematically
as collections of surfaces which minimize their surface area sub ject to volume con-
straints. Each surface will have constant mean curvature, and is thus called a cmc
surface. Section 2 intro duces cmc surfaces, and describ es the balancing of forces
within them. Section 3 lo oks at existence results for bubble clusters, and at the
singularities in such clusters and in foams. These are describ ed by Plateau's rules,
which imply that a foam is combinatorially dual to some triangulation of space.
Among the most interesting triangulations, from the p oint of view of equal-volume
foams, are those of the tetrahedrally close-packed tcp structures from chemistry,
describ ed in Section 4. Finally, Section 5 discusses the application of these ideas to
Kelvin's problem of partitioning space.
2. CMC Surfaces and Forces
Wewant to examine the geometry of a surface like one in a foam which minimizes
its surface area, given constraints on the enclosed volume. Rememb er that a surface
in space has at each p oint two principal curvatures k and k ; these are the
1 2
minimum and maximum over di erent tangent directions of the normal curvature.
The mean curvature is their sum H = k + k twice the average normal curvature.
1 2
A p erturbation of a surface M is given byavector-valued function ~v on M ,
telling how each p ointistomove. Let us consider the rst variations of area A and
~
volume V under this p erturbation. If ~n is the unit normal to M and H = H~n is 1
2 John M. Sul livan
the mean curvature vector, then
Z Z
~
V = ~v ~n dA; A = ~v H dA;
M M
~
so wesay that ~n is the gradientofvolume and H is the gradient of area. If we write
H = A= V ,we obtain a second, variational, de nition of mean curvature.
These Euler-Lagrange equations show that a surface is critical for the area func-
tional exactly if H 0. Such a surface is called a minimal surface, and indeed any
small patch of it is then area-minimizing for its b oundary.Any minimal surface
is saddle shap ed, as in Fig. 1. Therefore, a plane broughttowards a patchofa
minimal surface must touch rst at the b oundary of the patch, not in the interior.
More generally, the maximum principle says that the same happ ens whenever any
two minimal surfaces are brought together.
Fig. 1. On the left, the least-area surface spanning xed edges of a cub e. As for any minimal
surface, near each p oint it is saddle shap ed, with equal and opp osite principal curvatures. On the
right, three minimal surfaces meet along a Plateau b order at 120 . This surface would b e achieved
by soap lm spanning three wire Vs; when extended bytwo-fold rotation ab out these b oundary
lines, it would form the Kelvin foam.
A soap lm spanning a wire b oundary, with no pressure di erence across it,
will b e a minimal surface, but in general we need to consider the e ect of a volume
constraint. In this case, we nd that the surface has constant mean curvature H c,
so it is called a cmc surface. Here the constant c is the Lagrange multiplier for the
volume constraint, which is exactly the pressure di erence across the surface; this
relation H =p is known as Laplace's law.
In physical bubble clusters or foams, each cell contains a certain xed amount
of air or other gas neglecting slow di usion e ects. Of course, air is not an
incompressible uid. But for our mathematical theory,we will usually pretend it
is, xing a volume rather than a mass of gas in each cell. This approximation is
no problem: no matter what the relation b etween pressure and volume is for the
real gas, the equilibrium structures will still b e cmc surfaces.
Foams and Bubbles: Geometry and Simulation 3
For instance, a single bubble will adopt the form of a round sphere. This isoperi-
metric theorem, that a sphere minimizes surface area for its enclosed 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 prop er mathematical pro of whichmust demonstrate the existence of a mini-
mizer rather than merely describing prop erties of one if it do es exist was only given
in the last century bySchwarz [1], using a symmetrization argument.
One of the most imp ortant ideas in the mathematical study of cmc surfaces
is that of forcebalancing, as describ ed by Korevaar, Kusner and Solomon [2]. In
general, No ether's theorem says that symmetries of an equation lead to conserved
quantities. The equation for a cmc surface is invariant under Euclidean motions;
the corresp onding conserved quantities can b e interpreted as forces and torques.
The physical interpretation is as follows. Start with a complete emb edded cmc
surface M , and cut it along a closed curve . The two halves of M pull on each other
with a force due to surface tension and to pressure exerted across an arbitrary disk
D spanning . If ~n is the normal to D , and ~ is the conormal to in M that is,
the unit vector tangenttoM but p erp endicular to , the force can b e written as
Z Z
~
F = ~ H ~n:
D
Here we think of pressure p = 0 outside and p = H inside. Torques around any
origin can b e de ned in a similar fashion.
The fact that M is in equilibrium, with surface tension balancing pressure dif-
0
ferences, means that no compact piece feels any net force or torque. If and are
homologous curves on M , their di erence b ounds such a compact piece, so their
~
forces must b e equal. That is, F actually dep ends only on the homology class
of .
The cmc surfaces of revolution, called unduloids,were rst studied by Delau-
nay [3] in the last century. An unduloid is a wavy cylinder as in Fig. 2; the force
vector, for a curve lo oping around the cylinder, necessarily p oints along the axis
of revolution. The family of unduloids for given H can b e parameterized by the
strength of this force, which is maximum for a straight cylinder. In fact, Delaunay
Fig. 2. Left, a typical unduloid a cmc surface of revolution, whose force vector p oints along its
axis. Right, a typical triunduloid emb edded cmc surface with three ends. Each end is asymptotic
to an unduloid; the three force vectors sum to zero.
4 John M. Sul livan
describ ed the generating curves for unduloids as solutions of a second-order ODE,
and the force is a rst integral for this equation. As the force decreases from that of
a cylinder, we see alternating necks and bulges. The necks get smaller and smaller,
like shrinking catenoids, and in the limit of zero force, the unduloid degenerates
into a chain of spheres, as in a b eaded necklace.
The idea of force balancing has b een very useful in the classi cation of complete
emb edded cmc surfaces. The only compact example is the round sphere [4], with
all forces equal to zero. In general, each end of such a surface is asymptotic to an
unduloid [2]. Thus it has a force vector asso ciated to it, directed along the axis of the
unduloid, pulling the \center" of the surface out towards in nity. These force vectors
on the ends must sum to zero. This balancing condition is the essential ingredient
in classifying such cmc surfaces, and combined with some spherical trigonometry
can lead to a complete classi cation of the three-ended surfaces [5, 6].
3. Bubble Clusters and Singularities in Soap Films
So far, wehave b een rather cavalier ab out assuming that solutions to our problem
of minimizing surface area with constraints exist and are smo oth surfaces hence
of constant mean curvature. Of course, the surfaces in a foam or bubble cluster
are only piecewise smo oth, and the pieces meet in the singularities observed by
Plateau [7].
Almgren formulated the bubble cluster problem as follows: given volumes V ,
1
.. ., V , enclose and separate regions in space with these volumes, using the least
k
total area. He then proved, using techniques of geometric measure theory, that a
minimizer exists and is a smo oth surface almost everywhere [8] see also [9].
The pro of do es not giveany control on the top ology of the regions; in fact we
are not allowed even to assume that each region is connected. For all we know,
it may b e that the b est waytomake a cluster with some four xed volumes is to
split one of them into two pieces with the correct total volume, and make what
lo oks like a cluster of ve bubbles. A reasonable conjecture, of course, is that in
any minimizer, each region is connected.
Almgren's result do es, however, say something ab out the shap e of the mini-
mizers: they are so-called M ; "; -minimizing sets. In the plane, an M ; "; -
minimizing set is a nite network of smo oth 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 b e improved at any small scale.
Of course, the mo dels we use here are meant to describ e dry foams; in a wet foam
with p ositive liquid fraction '>0, there are Plateau b orders near each singularity.
In this case one mightwonder if additional singularities might b e p ossible. In fact,
awet X is stable for large ', though Brakke and Morgan [10] have shown that it
is unstable for small p ositive '. The analogous question is three dimensions is still
op en.
In three-space, Taylor was able to show [11] that the singularities observed
Foams and Bubbles: Geometry and Simulation 5
Fig. 3. The X junction at the left is not an M ; "; -minimizing set in the plane b ecause 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.
empirically in soap lms by Plateau are the only ones p ossible in an M ; "; -
minimizing set. That is, such a minimizer consists of a nite union of smo oth
surfaces, meeting in threes at 120 dihedral angles along a nite numb er of smo oth
curves; these curves in turn meet at nitely many tetrahedral corners whichlook
like the soap lm obtained when dipping a tetrahedral frame into soapywater: six
sheets come together along four triple curves into the central singularity.
In addition to proving these Plateau rules,Taylor showed each smo oth piece
has constant mean curvature, and the mean curvatures satisfy a cocycle condition:
numb ers the pressures can b e assigned to each comp onent of the complement
such that the mean curvature of eachinterface is the di erence of these pressures.
Although we cannot guarantee that each region of xed volume is a single comp o-
nent, the di erent comp onents making up a single region will necessarily have the
same pressure.
Of course a minimizing cluster of just one region will b e a round sphere. It is
surprising, though, how little is known ab out other small clusters. The case of two
regions, a double bubble, is one of the geometrical problems where \many mathe-
a
maticians b elieve, and all physicists know," that the obvious solution is correct.
The b est double bubble should b e the one seen when playing with soap bubbles,
with three spherical caps meeting along a single circular triple junction. Surpris-
ingly, this is not yet a theorem. However, Hutchings showed following a suggestion
of White that the minimizer must have rotational symmetry, and gave further con-
ditions, which, in the case of equal volumes left only one family of comp etitors, as
in Fig. 4 right. Hass and Schla y were able to rule these out with a computer
search [14] see also [15], thus solving the equal-volume case.
We can de ne a foam mathematically as a lo cally nite collection of cmc
surfaces, meeting according to Plateau's rules, and satisfying the co cycle condition.
We will call the comp onents of the complement the cel ls of the foam, call the
interfaces b etween them simply the faces, call the triple junction lines where they
meet the Plateau borders, and call the tetrahedral singularities the corners.We
a
The quote is from Rogers [12], describing Kepler's sphere packing problem, which has recently
b een rigorously solved by Hales [13].
6 John M. Sul livan
Fig. 4. At the left we see the standard cluster of three soap bubbles, with all interfaces spherical
and singularities following the Plateau rules. Presumably it minimizes total area for its enclosed
volumes. At the rightwe see a p ossible comp etitor for the least-area double bubble, where one
region is a toroidal b elt around the other.
exp ect this de nition to describ e all physical foams whose shap e is determined
by surface tension, though there are not mathematical results like Almgren's that
would apply to the case of in nite clusters.
For many purp oses, we can ignore the geometric parts of Plateau's rules, and
only pay attention to the combinatorial asp ects of the foam's cell complex. The
combinatorial rules mean that the dual cel l 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 adjacent cells are connected byan
edge, which is dual to a face of the foam. Where three faces come together along
a b order, we span the three corresp onding edges with a triangle; and where four
b orders come together at a corner, we ll in a tetrahedron.
Combinatorially, the Plateau rules mean that a foam and its dual triangulation
are likeaVoronoi decomp osition of space and the dual Delone triangulation. Given
a set of sites in space, the Voronoi cell [16, 17] for each site is the convex p olyhedron
consisting of p oints in space closer to that site than to any other. The dual Delone
complex, whose vertices are the original sites, has the prop erty that each tetrahedron
has no other sites inside its circumsphere. An example in the plane is shown in
Fig. 5.
This similarity suggests that that we might lo ok for foams as relaxations of
Voronoi decomp ositions. For instance, we can give a mo dern interpretation to
Kelvin's construction [18] of his candidate for a least-area partition of space into
equal volume cells as follows. Start with sites in the b o dy-centered cubic bcc
lattice. Their Voronoi cells are truncated o ctahedra, packed to ll space. If we let
the lms in this packing relax until the geometric parts as well as the combinatorial
Foams and Bubbles: Geometry and Simulation 7
Fig. 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
adjacentVoronoi cells.
parts of Plateau's rules are satis ed we should get a p erio dic foam, the one prop osed
by Kelvin.
Starting from other p erio dic arrays of sites, we can carry out the same pro cedure
to generate a foam. Mathematically, there is no theory for the relaxation step,
whichwould follow a mean-curvature ow for surfaces with triple junctions and
volume constraints. But in practice this pro cedure is a useful way to compute
foam geometries. The author's vcs software for computing Voronoi diagrams [19]
in three dimensions can send its output to Brakke's Surface Evolver [20]. This latter
program can interactively manipulate, re ne and evolve triangulated surfaces into
energy-minimizing con gurations, mo deling the physical e ects of surface tension
and pressure. New symmetry features in the Evolver [21] allow computation of
just one fundamental domain in a foam of arbitrary symmetry,thus giving greater
accuracy for given computational e ort.
4. TCP structures
The tetrahedral angle arccos 1=3 109:47 , found at all foam corners, is b etween
the average angle of a p entagon 108 and that of a hexagon 120 . Thus foams
with fairly straight b orders are likely to have mostly p entagonal and hexagonal
faces.
Chemists have studied transition metal alloys in which the atoms are packed in
nearly regular tetrahedra. These tetrahedrally close-packed tcp structures were
rst describ ed byFrank and Kasp er [22, 23], and have b een studied extensively by
the Sho emakers [24] among others. In all of these structures, the Voronoi cell of
each atom has one of four combinatorial typ es; these are exactly the four p olyhedra
whichhave only p entagonal and hexagonal faces, with no adjacent hexagons. Math-
ematically,we can de ne of tcp structures in this way [25]: triangulations whose
8 John M. Sul livan
combinatorial duals called tcp foams have only these four typ es of cells. Since
this de nition deals only with the combinatorics of the triangulation, presumably it
includes some examples that havetoomuch geometric distortion to work chemically
as tcp structures.
Fig. 6. There are four typ es of cells found in tcp foams, each with 12 p entagonal faces like the
do decahedron at the left. The remaining three have in addition two, three or four hexagonal faces,
arranged antip o dally, equatorially, and tetrahedrally.
Each of the four typ es of cells found in tcp foams has 12 p entagonal faces, as
seen in Fig. 6. One typ e is the p entagonal do decahedron. Cells of the other three
typ es have additionally two, three, or four hexagonal faces, which are arranged
antip o dally, equatorially, or tetrahedrally resp ectively. The 14-hedron can b e
6
viewed as a \ -fold unwrapping" of a do decahedron, since it has six-fold symmetry
5
through the centers of the opp osite hexagons.
It is an interesting op en question just what foams are p ossible with these four
typ es of cells. Dually,we are asking for triangulations of space in which each
edge has valence ve or six, and no triangle has two edges of valence six. We
knowwe cannot use only do decahedra, since a foam made of these alone meeting
tetrahedrally will ll a spherical space, not Euclidean space.
Fig. 7. The three basic tcp structures. A15, left, has two 12-hedra one upp er-left and six
14-hedra in colums likelower right in a cubic cell. Z, center, has three 12s top, two 14s in
vertical columns but separated in the gure, 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.
Chemists have describ ed three basic p erio dic tcp structures, shown in Fig. 7.
A15, observed for instance in Cr Si, has one 12-hedron and three 14-hedra in a 3
Foams and Bubbles: Geometry and Simulation 9
fundamental domain. Z, observed in Zr Al , has three 12-hedra, two 14-hedra, and
4 3
two 15-hedra. And C15, observed byFriauf and Laves in MgCu , has two 12-hedra
2
and one 16-hedron.
If we describ e p otential tcp structures by the ratio of cells they have of the four
typ es, we could plot them all within an abstract tetrahedron, as convex combina-
tions of the four vertices. But the observation of Yarmolyuk and Kripyakevich [26]
is that all known tcp structures are, in fact, convex combinations of the three ba-
sic ones just mentioned, so they all get plotted within the triangle of Fig. 8. The
Z
J
P
F M
H
I
R
T
X
C15
A15 C14 SM
Fig. 8. The basic tcp structures A15, Z, and C15 lie at the corners of this triangle, where the
1 1
to 13 and the horizontal axis plots the average numb er of faces p er cell ranging from 13
3 2
2
vertical axis plots the fraction of cells which are 15-hedra ranging from 0 to . Every known
7
tcp structure, when describ ed by its numb ers of 12-, 14-, 15-, and 16-sided cells, is a convex
combination of these three. Distinct structures may app ear at the same p oint, but need not then
have the same prop erties.
Sho emakers [24] attempt to explain this observation by noting that for a tiling by
tetrahedra only slightly distorted from regular, we exp ect to have z 13:397. Cer-
3 1
for A15, z =13 for Z, and tainly the three basic tcp structures, with z =13
2 7
1
z =13 for C15, are close to this value.
3
It is interesting to note that there are other chemical structures which exhibit
the structure of the tcp Voronoi cells that is, the dual tcp foams more explicitly.
These are describ ed in detail by O'Kee e [27], but we outline the ideas here. In
clathrates, large gas molecules are trapp ed at the lo cation of the tcp sites inside
water cages: oxygen atoms sit at the Voronoi corners, b onded byhydrogen along
the b orders Voronoi edges. For instance, chlorine hydrate Cl 46H O, has the
2 6 2
structure of Voronoi cells for A15, called a Typ e I clathrate. Many salt hydrates
have instead the structure of the Z foam called Typ e I I I, while others use non-tcp
structures like the bcc cells of the Kelvin foam. Some zeolites have similar struc-
tures, with Si and p ossibly some Al atoms at the corners, b onded tetrahedrally to
oxygen atoms along the b orders. For instance so dalite Na Al Si O Cl has the bcc
4 3 3 12
foam structure, with Cl atoms at the Voronoi sites, and Na in the centers of the
hexagonal faces. So dium silicide, when heated prop erly, can generate silicon cages
in the A15 or C15 foam pattern Typ e I or Typ e I I clathrates with so dium trapp ed
10 John M. Sul livan
inside. It should b e noted, however, that in other zeolites the 4-connected silicate
nets are less dense and do not close up into cages. See [28] for more information
on manyinteresting crystal structures, and [29] for instance for the structures of
zeolites.
5. Kelvin's problem
Over one hundred years ago, Kelvin prop osed the problem of partitioning space
into equal-volume cells using the least interface area p er cell [18]. He suggested
that the solution might b e what wehave b een calling the Kelvin foam, a relaxation
of the Voronoi diagram for the bcc lattice. Although Weaire and Phelan nowhave
a b etter partition, Kelvin's is still conjectured [30] to b e the b est if the cells are
required to b e congruentortohave equal pressure. Mathematically, there is no
theory to suggest that such a b est in nite cluster should exist, but we exp ect that
it will, and will have the structure of a foam.
Weaire and Phelan [31] were the rst to consider using tcp structures as foams.
Although they initially thought of these as mo dels for wet foams, they quickly
discovered that the dry A15 foam was a more ecient partition of space than
Kelvin's candidate. Togobeyond their go o d numerical evidence from Brakke's
Evolver and give a rigorous pro of that their foam is b etter than Kelvin's, we need
to give a b ound on howmuch the Kelvin foam can relax.
Although, as wehave mentioned, there is no general mathematical theory 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 owona
single surface. In Fig. 1, right, wesaw a symmetric unit of Kelvin's foam, b ounded
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
xed b oundary lines and one free b oundary 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 givesalower b ound on its area [25]. This b ound, it turns out, suces
to prove that even the unrelaxed Weaire-Phelan A15 foam b eats Kelvin. The Color
Plate shows a view of the A15 foam from the inside, using the author's computer
graphics rendering of realistic soap lms [32].
Given this example, it is natural to lo ok for go o d equal-volume partitions among
the other tcp foams. Rivier [33] prop osed that since C15 has among the known tcp
structures the lowest z and thus the highest prop ortion of p entagons, it might give
an even b etter partition. Instead, computer exp eriments [34] suggest that among
all tcp foams, A15 is the most ecient, and C15 among the least. Here, we are
lo oking 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 ecient partition in that pattern. Probably the 12-hedral and 16-
hedral cells of C15 naturally have such di erent sizes that distorting them to make
the volumes equal ruins whatever advantage p entagons give. If we considered a
mo di ed Kelvin problem where one-third of the cells were to have somewhat larger
Foams and Bubbles: Geometry and Simulation 11
volume, then presumably the C15 structure would do very well. The author has
also made computer exp eriments with equal-volume foams generated from other
chemical structures; the foam from -brass, for instance, is b etter than that from
C15, despite having even some triangular faces. Perhaps p entagons are not as
desirable in foams as has b een assumed.
Acknowledgements
The author is partially supp orted by NSF grant DMS-9727859 and NASA grant
NAG3-2122. He would like to thank Hassan Aref, Karsten Groe-Brauckmann,
Andy Kraynik, Rob Kusner, Frank Morgan, Mike O'Kee e and Nick Rivier for
helpful conversations on foams and related structures. Much of this pap er is based
on the author's rep ort [35] on lectures he delivered at the NATO scho ol on foams
in Cargese, and some of the gures originally app eared in [25].
References
1. H. A. Schwarz, Beweis des Satzes, dass die Kugel kleinere Ob er ache b esitzt, als jeder
andere Korp er gleichen Volumnes, Nach. Ges. Wiss. Gottingen 1884 1{13. Reprinted
in Gesammelte mathematische Abhand lungen, Chelsea, New York, 1972 I I.327{340.
2. N. Korevaar, R. Kusner, and B. Solomon, The structure of complete emb edded surfaces
with constant mean curvature, J. Di . Geom. 30 1989 465{503.
3. C. Delaunay, Sur la surface de r evolution, dont la courbure moyenne est constante,
Journal de math ematiques 6 1841 309{320.
4. A. D. Alexandrov, Uniqueness theorems for surfaces in the large, I, Vestnik Leningrad
Univ. Math. 19:13 1958 5{8. Translated in Amer. Math. So c. Transl. Ser. 2 21
1962, 412{416.
5. K. Groe-Brauckmann, R. Kusner, and J. M. Sullivan . Constant mean curvature
surfaces with three ends. ArXive e-print math.DG/9903101 1999.
6. K. Groe-Brauckmann, R. Kusner, and J. M. Sullivan , Constant mean curvature
surfaces with cylindrical ends, in Mathematical Visualization, eds. H.-C. Hege and
K. Polthier, Springer Verlag, Heidelb erg, 1998, 107{116.
7. J. Plateau, Statique Exp erimentale et Th eorique des Liquides Soumis aux Seules Forces
Mol eculaires, Gauthier-vill ars, Paris, 1873.
8. F. J. Almgren, Jr., Existence and regularity almost everywhere of solutions to elliptic
variational problems with constraints, Mem. Amer. Math. Soc. 4:165 1976.
9. F. Morgan, Clusters minimizin g area plus length of singular curves, Math. Ann. 299:4
1994 697{714.
10. K. Brakke and F. Morgan, Instability of the wet X soap lm, J. Geom. Anal. To app ear.
11. J. E. Taylor, The structure of singulariti es in soap-bubble-li ke and soap- lm-like mini-
mal surfaces, Ann. of Math. 103 1976 489{539.
12. C. A. Rogers, The packing of equal spheres, Proc. London Math. Soc. 8 1958 609{620.
13. T. C. Hales. An overview of the Kepler conjecture. ArXive e-print math.MG/9811071
1998.
14. J. Hass, M. Hutchings, and R. Schla y, The double bubble conjecture, Electron. Res.
Announc. Amer. Math. Soc. 1:3 1995 98{102.
15. F. Morgan, The double soap bubble conjecture, MAA FOCUS Decemb er 1995 6{7.
16. M. Senechal, Crystal line Symmetries, Adam Hilger, 1990.
12 John M. Sul livan
17. A. Okab e, B. Bo ots, and K. Sugihara, Spatial Tessel lations: Concepts and Applications
of Voronoi Diagrams, Wiley & Sons, 1992.
18. W. Thompson Lord Kelvin, On the division of space with minimum partitional area,
Philos. Mag. 24 1887 503{514. Also published in Acta Math. 11, 121{134, and
reprinted in [36].
19. J. M. Sullivan . The vcs software for computing voronoi diagrams. Available by email
from [email protected], 1988.
20. K. A. Brakke, The Surface Evolver, Exper. Math. 1:2 1992 141{165.
21. K. A. Brakke and J. M. Sullivan, Using symmetry features of the surface evolver to study
foams, in Visualization and Mathematics, eds. K. Polthier and H.-C. Hege, Springer
Verlag, Heidelb erg, 1997, 95{117.
22. F. C. Frank and J. S. Kasp er, Complex alloy structures regarded as sphere packings. I.
De nitions and basic principles , Acta Crystal l. 11 1958 184{190.
23. F. C. Frank and J. S. Kasp er, Complex alloy structures regarded as sphere packings.
I I. Analysis and classi cation of representative structures, Acta Crystal l. 12 1959
483{499.
24. D. P. Sho emaker and C. B. Sho emaker, Concerning the relativenumb ers of atomic
co ordination typ es in tetrahedrally close packed metal structures, Acta Crystal l. 42
1986 3{11.
25. R. Kusner and J. M. Sullivan, Comparing the Weaire-Phelan equal-volume foam to
Kelvin's foam, Forma 11:3 1996 233{242. Reprinted in [36].
26. Y. P.Yarmolyuk and P. I. Kripyakevich, Mean weighted co ordination numb ers and
the origin of close-packed structures with atoms of unequal size but normal co ordina-
tion p olyhedra, Sov. Phys., Crystal logr. 19 1974 334{337. Translated from Kristallo-
graphiya, 19, 539{545.
27. M. O'Kee e, Crystal structures as p erio dic foams and vice versa, in Foams and Emul-
sions, eds. N. Rivier and J.-F. Sado c, volume 354 of NATO Advanced Science Institute
Series E: Applied Sciences, Kluwer, 1998, 403{422.
28. M. O'Kee e and B. G. Hyde, Crystal Structures I: Patterns and Symmetry, Mineral
So c. Amer., Washington, 1996.
29. W. M. Meier and D. H. Olson, Atlas of Zeolite StructureTypes, Butterworths, 3rd
edition, 1992.
30. J. M. Sullivan and F. Morgan, eds., Op en problems in soap bubble geometry: Posed at
the Burlington Mathfest in August 1995, International J. of Math. 7:6 1996 833{842.
31. D. Weaire and R. Phelan, A counter-example to Kelvin's conjecture on minimal sur-
faces, Phil. Mag. Lett. 69:2 1994 107{110. Reprinted in [36].
32. F. J. Almgren, Jr. and J. M. Sullivan , Visualization of soap bubble geometries, Leonardo
24:3/4 1992 267{271 and Color Plate C. Reprinted in [37].
33. N. Rivier, Kelvin's conjecture on minimal froths and the counter-example of Weaire
and Phelan, Europhys. Lett. 7:6 1994 523{528.
34. A. M. Kraynik, R. Kusner, R. Phelan, and J. M. Sullivan . TCP structures as equal-
volume foams. In preparation.
35. J. M. Sullivan, The geometry of bubbles and foams, in Foams and Emulsions,
eds. N. Rivier and J.-F. Sado c, volume 354 of NATO Advanced Science Institute Series
E: Applied Sciences, Kluwer, 1998, 379{402.
36. D. Weaire, ed., The Kelvin Problem,Taylor & Francis, 1997.
37. M. Emmer, ed., The Visual Mind: Art and Mathematics, Cambridge, MIT Press, 1993.
oams and Bubbles: Geometry and Simulation 13
F
Color Plate John Sullivan: This computer-graphics image shows the Weaire-Phelan foam, ob-
tained by relaxing the tcp structure A15. This foam is, as far as we know, the b est partition of
space into equal-volume cells. It is rendered realistically as soap lms.