GENERAL I ARTICLE

How to Stay Away from Each Other in a Spherical

Universe
1. Tammes' Problem

Bhaskar Bagchi
M a t h e m a t i c s is full o f i n n o c e n t l o o k i n g p r o b l e m s
w h i c h , w h e n p u r s u e d , s o o n g r o w t o m a j e s t i c pro-
p o r t i o n s a n d b e g i n to i m p i n g e u p o n t h e frontiers o f
research. O n e s u c h p r o b l e m is t h e s u b j e c t o f t h i s
t w o - p a r t article - t h e p r o b l e m o f p a c k i n g s p h e r i c a l
c a p s o n t h e surface o f a s p h e r e .

B h a s k a r Bagchi is with Twelve Men on a Sphere

the I n d i a n Statistical
I n s t i t u t e since 1971, first For most practical purposes, we humans are inhabitants of
as a s t u d e n t and t h e n as
a (two dimensional) spherical universe: we are constrained
a m e m b e r of the faculty.
to eke out our lives on the surface of the earth. Suppose
He is i n t e r e s t e d in
diverse a r e a s of
twelve denizens of this spherical universe hate each other
m a t h e m a t i c s like so much t h a t t h e y decide to build their houses as far away
combinatorics, from each other as possible. We assume that, except for
elementary, and analytic this morbid hatred, these guys have no other constraints:
number theory, t h e y may build their houses anywhere on the earth's surface
f u n c t i o n a l analysis,
(must be stinkingly rich to have such freedom!). The ques-
combinatorial topology
tion is: how should these twelve houses be positioned ? T h e
and s t a t i s t i c s .
answer depends, of course, on what we m e a n by the phrase
'as far away from each other as possible'. Let us take it to
m e a n t h a t the minimum of the pair-wise distances between
the twelve houses is to be maximised subjected to the only
Figure 1 Icosahedron. constraint t h a t they be situated on the spherical surface.
W i t h this understanding, it turns out t h a t the problem has
a unique solution: the houses must be built at the vertices
of an icosahedron inscribed in the sphere. (This is proved
in P a r t 2 of this article.) Now, the icosahedron is a highly
structured and symmetrical figure (see Figure 1). Its group
of rotational symmetries has order 60 - it is the smallest
non-abelian 'simple' group. Isn't it amazing t h a t so much
structure and s y m m e t r y result as the solution of such a sim-
ple minded maximisation problem ? Before we finish we

18 RESONANCE I September ]997

 GENERAL I ARTICLE

Box 1. Symmetry Group

Given a solid body in Euclidean space (of any dimension), its rotational symmetries
are those rotations of the ambient space which send the body (as a whole) onto
itself. Clearly, applying two such rotations successively one obtains a third rota-
tional symmetry of the body. It follows that all such rotations form a group with
composition as the group law. This is called the isometry group of the body. Other
names: symmetry group, automorphism group.

Box 2. The Smallest Non-Abelian Simple Group

Recall that a group is called non-abelian if its law is not commutative: i.e., ff the
product of two elements depends, in general, on the order in which the multiplication
is performed. The order of a finite group means the number of elements in the group.
A group is called simple ff it has no (proper, non-trivial) normal subgroup. Simple
groups are to general groups what prime numbers are to arbitrary integers: they
are supposed to be the building blocks of general groups. This is why they are so
important. Trivialish examples - in fact the only examples - of abelian simple groups
are the symmetry groups of regular p-gons for various primes p. The symmetry group
of the icosahedron (as we have defined it here - excluding reflections) is the non-
abelian simple group of the smallest possible order. Abstractly, this group may be
described as the group of all even permutations of a set of five objects. All finite
simple groups are known. With 26 exceptions (the so called sporadic simple groups)
they fall in n e a t infinite families.

shall see m a n y more examples of this sort of phenomenon. If,

however, we wished to build thirteen houses with the above
constraint, t h e n n o b o d y knows what is the optimal solution
Tammes proposed
or w h e t h e r the solution is unique or not!
that for the sake of
Pollen Grain maximum
biological
This problem (with an arbitrary number n in place of twelve) efficiency the
was first raised by the botanist P M L Tammes in 1930. He pores on a grain of
w a n t e d to explain the observed distribution of the pores on a pollen should be
grain of pollen. He proposed that, for the sake of m a x i m u m
placed as far away
biological efficiency, the n pores on the surface of the grain
from each other as
(which may be taken to be a sphere) are placed as far away
from each other as possible. Of course, to test this theory, possible.

RESONANCE I September 1997 19

 GENERAL [ ARTICLE

B o x 3. C o n v e x Polyhedra
Recall that a set of points in space is called a convex set if the straight line segment
joining any two points in the set is entirely contained in the set. Clearly the inter-
section of any family of convex sets is again convex. It follows that among all convex
sets containing a given point set A, there is a smallest one (namely the intersection
of all the convex sets containing A); it is called the convex hull of A. Among the
simplest examples of convex sets are the half spaces: the removal of any fixed plane
from space breaks up space into two parts; each of these two parts is called a half
space determined by the plane. The union of a half space with the plane bounding
it is called a closed half space. If a bounded set of points can be expressed as the
intersection of finitely many closed half spaces then it is called a polyhedron (plu-
ral: polyhedra). A point in a polyhedron X is called a vertex (plural: vertices) or
extreme point (intuitively, a corner) of X if it is not an interior point of any line
segment contained in X. It is not diFFicult to see that any polyhedron has only
finitely many vertices and the polyhedron is the convex hull of the set of its vertices.
The straight line segment joining two vertices of a polyhedron X is called an edge
of X if it is the intersection of X with a plane. A subset of X is a face of X ff it is
not a vertex or edge and yet is obtainable as the intersection of X with a plane.

one needs to know the theoretical o p t i m u m arrangement(s).

This is the 'Problem of Tammes.' T h e case n = 13 of this
problem can be traced back to a controversy between New-
ton and Gregory. Many m a t h e m a t i c i a n s have worked on
Tammes' problem, prominent among the early workers be-
ing Fejes Toth and B L van der Waerden. And yet, only the
cases n < 12 and (surprise!) n = 24 have been solved so far.
Figure 2 Tetrahedron. In all these cases, excepting the case n - 5, the optimal so-
lution turns out to be unique. Let us briefly examine atleast
some of these solutions. Note t h a t the convex hull of a n y n
points on t h e sphere is a convex polyhedron whose vertices
are precisely the n points t h a t we began with. So it is often
convenient to describe a solution to an instance of Tammes'
problem as (the set of vertices of) a convex polyhedron. All
the polygonal faces which occur in the next two paragraphs
are regular (i.e., have equal sides and equal vertical angles).
In particular, the triangles are equilateral.

20 RESONANCE J September 1997

 GENERAL [ARTICLE

B o x 4. P l a t o n i c Solids

A flag of a convex polyhedron X is a triplet (z, e, F) where z is a vertex of X, e is an

edge of X through z and F is a face of X contRinlng e. X is called a platonic solid
if, given any two flags ~ = ( z , e , F ) and ~5~ = (z~,e~,F ~) of X, there is an isometry
of X (rotation or reflection) which takes 0 to 0t (i.e., maps z to z l, 9 to e I and F to
Ft.) Since no non-identity isometry of space can fix a flag, it follows that the order
of the extended symmetry group (including refiections as well as rotations fixing the
body) of a platonic solid equals the n-rnher of flags of the solid. For instance, the
icosahedron has 12 vertices, 5 edges through each vertex and 2 faces through each
edge, so that it has 12 • 5 x 2 = 120 flags, and 120 is the order of its extended
symYnetry group.

The Platonic Solids

T h e s e are objects t h a t have fascinated m a n since the time

of the Greeks. Informally, the platonic solids are the convex
polyhedra which look t h e same from the point of view of each
vertex, edge and face. Given one such solid, the centers of its
faces form t h e vertices of a second platonic solid, called the
dual of the first. T h e r e are only five platonic solids. These
are : (i) the (regular) t e t r a h e d r o n (Figure ~) with 4 equidis-
t a n t vertices, 6 edges and 4 triangular faces, (ii) the cube
with 8 vertices, 12 edges and 6 square faces, (iii) the octao Figure 30ctahedron.
h e d r o n (Figure 3) w i t h 6 vertices, 12 edges and 8 triangular
faces, (iv) the icosahedron with 12 vertices, 30 edges and
20 triangular faces, and, finally, (v) the dodecahedron (Fig-
ure 4) with 20 vertices, 30 edges and 12 pentagonal faces. Of
these, the t e t r a h e d r o n (or regular simplex) is its own dual,
t h e cube and the o c t a h e d r o n are duals of each other, and Figure 4 Dodecahedron.
so are the icosa- and t h e dodecahedron. Everybody knows
t h e regular t e t r a h e d r o n and the cube: these are the obvious
three dimensional analogues of the equilateral triangle and
t h e square. T h e o c t a h e d r o n is another - and less obvious -
analogue of the square: it may be obtained by p u t t i n g to-
gether two right pyramids on a c o m m o n square base. There
is no such easy description of the remaining two platonic
solids. In coordinates, the vertices of the icosahedron m a y
be t a k e n to be the even coordinate permutations of the four

RESONANCE I September 1997 21

 GENERAL I ARTICLE

B o x 5. Archimidean Solids (also called semi-regular solids)

These are convex polyhedra such that (i) all their faces are regular polygons, and
(ii) the symmetry group is transitive on the vertices (i.e., any vertex can be mapped
to any other vertex by a suitable isometry of the body). The square anti-prism is
an example. It may be described as follows. Begin with the cube and rotate one of
its faces (in its plane) around its center by haft a right angle. Then move this face
closer to the opposite face by tra_n.qlating it parallel to the original position. For the
correct choice of the distance between these two faces, the sides of the convex hull
of the eight vertices (in their new positions) are all equal and the square anti-prism
results.

1 ( 0 , +1, + r ) , where r = ( l + v ~ ) / 2 is the 'golden

vertices 7~.$.~
ratio'. (Exercise: from this information, can you find the co-
ordinates of the vertices of the dual polyhedron, namely t h e
dodecahedron?) You may try to build your own model of
these two solids by pasting together 20 equilateral triangles
/ 12 regular pentagons of the same size, preferably m a d e of
cardboard.

If you wanted to place n points on (the circumference of) a

circle so t h a t t h e y are as far away from each other as possi-
ble, t h e n clearly you would have to place t h e m at the vertices
of a regular n-gon. (Exercise : W h y ?) Since the platonic
solids are the three dimensional analogues of regular poly-
gons, it is natural to conjecture t h a t all five of t h e m are
Since the platonic (unique ?) solutions of Tammes' problem for t h e relevant
solids are the three values of n. In fact, the tetrahedron, octahedron and t h e
dimensional icosahedron are indeed the unique solutions for n = 4, 6
analogues of and 12 respectively. (Notice t h a t these are the platonic
regular polygons, it solids with triangular faces. We shall present a proof of
is natural to these three results in P a r t 2 of this article.) But the conjec-
ture is false for n = 8: the cube is not optimal in the sense
conjecture that all
of Tammes' problem ! The unique solution in this case is
five of them are
provided by the vertices of the square anti-prism with 8 ver-
solutions of tices, 16 edges, 2 square faces and 8 triangular faces. We do
Tammes' problem not know if the dodecahedron is a solution for n = 20, per-
for the relevant haps it is not! For n = 11 the unique solution is obtained by
values of n. deleting a vertex of the icosahedron. For n = 5 one solution

22 RESONANCE I September 1997

 GENERAL I ARTICLE

is the octahedron minus a single point, but a second solution

is obtained by taking three equidistant points on the equator
together with the two poles. In fact, one has, in this case,
infinitely many solutions 'in between' these two extreme so-
lutions. Any n < 3 points on the sphere lie on a circle, so
that the solutions in these cases are obvious. The solutions
for n = 7, 9 and 10, though unique, are too difficult to
describe here. (At least, I must so declare since I do not
understand t h e m myself!) In a remarkable paper in Mathe-
matische Annalen, 1961, R M l~obinson settled a conjecture Figure 5 Snub cube.
of van der Waerden in the affirmative by showing that the
unique solution for n = 24 is given by the vertices of the snub
cube (Figure 5). This is a so-called archimidean solid (the
square anti-prism (Figure 6) is another) with 24 vertices, 60
edges, 6 square faces and 32 triangular faces. If the reader
is particularly brave and curious, he or she m a y try to go
through the monumental paper by L Danzer which occupies
64 pages of the 60th volume (1986) of Discrete Mathemat-
ics. In this paper, the cases n < 12 of Tammes' problem are
given a uniform treatment. ( The cases n = 10 and 11 are
solved for the first time in this paper.) The wonderful book
on regular polytopes by Coxeter (see Suggested Reading) is
highly recommended to any one wishing to know more about
the regular and semi-regular solids alluded to above.

Higher Dimensional "Analogues: Optimal

Spherical Codes
In the formulation of Tammes' problem it was implicit that
the sphere in question was in three dimensional euclidean
space; further, by a suitable choice of coordinates, it could
be taken to be the unit sphere (i.e., the sphere of radius 1 Figure 6 Square anti-
centered at the origin). More generally, let S d-1 denote the prism.
unit sphere in the d dimensional euclidean space ~ d . By
a spherical code of size n and rank d one means a set of n
points in S d-1. This terminology is fashionable because the
points in the spherical code are thought of as the words in
a code language. The minimum distance of a code X is by
definition the m i n i m u m of { [ I x - y l l : x ~ y, x , y E X~.
(Here Jl. IJ is the usual euclidean norm: JJxll2 ~ " ~-~i=ldXi2.
So JJx - yJ[ is the euclidean distance between the points x

RESONANCE I September 1997 23

 GENERAL I ARTICLE

To be able to and y. We have, S d-1 = {x-E ~ d : i1~11= 1}.) Clearly,

to be able to detect and correct errors committed during
detect and correct
transmission, it is desirable that the m i n i m u m distance of
errors committed
a spherical code should be as large as possible. Let p(n, d)
during
denote the m a x i m u m of the minimum distance of X where
transmission, it is the m a x i m u m is over all spherical codes X of size n and rank
desirable that the d. Also, by an optimal spherical code of size n and rank d
minimum distance we mean such a spherical code whose minimum distance
of a spherical code equals p(n, d). So the problem is to compute p(n, d) and
be as large as find the optimal spherical codes for various values of n and
possible. d. Exercise: i f n = e + l < d + l , then show that the spherical
code is 'actually' a code of rank (at most) e, so that one is
reduced to a case of lower rank. So, in the following, one
may as well assume n > d + 1.

The Regular Polytopes

Given the paucity of solutions to this problem in the case of

rank 3, it is not surprising that very little is known about
the higher ranks. We may as well begin with the regular
polytopes which generalize the platonic solids of three di-
mension. In d dimension there are three regular polytopes
with n = d + 1, 2d and 24 respectively. These generalize the
tetrahedron, octahedron and the cube; they are called the
(regular) simplex, the cross polytope (or orthoplex) and the
hypercube. (The regular simplex is the convex hull of d + 1
equidistant vertices. The hypercube is the convex hull of
the 2 d vertices with zero-one coordinates. The orthoplex is
the convex hull of the 2d vertices with one coordinate equal
to -4-1 and remaining coordinates zero.) Apart from these
three families, there are five sporadic regular polytopes, two
in three dimension (which we have already met) and three in
four dimension. The sporadic regular polytopes in ~ 4 are:
(i) the 24-cell with 24 vertices, 96 edges, 96 triangles and 24
octahedral faces; in co-ordinates, its vertices may be taken
to be the permutations of the points ( + l / v ~ , + I / V ~ , 0, 0);
(ii) the 600-ceU with 120 vertices, 720 edges, 1200 triangles
and 600 tetrahedral faces, (iii) the 120-cell with 600 vertices,
1200 edges, 720 triangles and 120 dodecahedral faces. The
24-ceU is self dual while the other two are duals of each other.
The vertices of the 600-cell may be obtained from the points

24 RESONANCE I Seplember 1997

 GENERAL I ARTICLE

(I, O, 0,0), ~(
1 1, 1, 1, 1) and ~(
1 7", 1, r -1, 0) by arbitrary sign
The kissing
changes and even permutations of the coordinates. (Here,
number K (d) in
as in the description of the icosahedron, r is the golden ra-
tio.) Other interesting details of the 600-cell and its dual dimension d is the
may be found in Coxeter's book. In certain quarters, the largest number of
24-cell and the 600-ceU are better known as the root sys- solid balls in R ~
tems 1)4 and H4. In this avatar, they may be located in all of the same
Reflection Groups and Coxeter Groups by J E Humphreys. size, which can
A polytope is called simplicial if its maximal faces are sim- touch a given ball
plexes. Thus, among the regular polytopes, the simplicial of the same size.
ones are the simplexes, the cross polytopes, the icosahedron
and the 600-cell. In a paper published in volume 32 of Acta
Mathematica Hungarica (1978), K Boroczky proved that all
these simplicial regular polytopes are unique optimal codes.
The new result here is the optimality of the 600-cell.

The Kissing Number Problem

The known solutions to this problem constitute yet another

source of optimal spherical codes. The kissing number ~(d)
in dimension d is the largest number of solid balls in ~d~ all
of the same size, which can touch a given ball of the same
size. (Of course, solid balls are not allowed to penetrate each
other. Technically, these balls must have pairwise disjoint
interiors.) The problem is to determine this number. The
solution is known only for d = 2, 3, 8 and 24, and the cor-
responding values of ~ are 6 = 2(3), 12 = 2(~), 240 = 2(10)
and 196560 = 2(258). What is its relation to optimal spheri-
cal codes? Well, given ~(d) balls of radius half each touch-
ing a ball of radius half centered at the origin, the centers
of these balls form a spherical code of size ~(d) and rank
d with minimum distance _> 1. Clearly this construction
can be reversed, so that ~(d) is the largest value of n for
which p(n, d) _> 1. Since the regular hexagon is the unique
optimal code of size 6 and rank 2, and its minimum dis-
tance is exactly equal to 1, this proves ~(2) = 6. Since
the icosahedron is the optimal code of size 12 and rank 3,
and since its minimum distance is ~/2 - 2 / v ~ > 1, we see
that ~(3) > 12. It is not difficult to see that ~(3) <__ 13.
The Newton-Gregory controversy already alluded to was on

RESONANCE I Seplember 1997 25

 GENERAL J ARTICLE

Suggested Reading whether the true value in this case was twelve or thirteen.
Newton believed it was 12 but Gregory thought otherwise.
9 H S M Coxeter. Regular
Since p(12, 3) > 1, centering the balls at the vertices of an
Polytopes. Dover. New icosahedron is not the only way to place twelve balls touch-
York, 1973. ing a given ball. In this arrangement, no two of the 12 balls
9 JH ConwayandNJA touch each other, so that one may well think that they may
Sloane. Sphere Packings, be pushed around to make place for a thirteenth ball. In
Lattices and Groups. fact, a better way to place the twelve balls is to take t h e m
Springer Verlag. New
to be the balls of radius half centered at the permutations of
York, 1988.
the points ( • + l / v / 2 , 0) (an arrangement reminiscent
of the 24-ce11) in which case each of the twelve surrounding
balls touch four others apart from the central ball. It was
only in 1874 that R Hoppe proved that p(13, 3) < 1; thus
Newton was right. To date, no really accessible proof of
this fact is available. The case of four dimensions is very
analogous. It is known that ~(4) is either 24 or 25, b u t
nobody knows the correct value. It is very tempting to con-
jecture that the 24-cell is the (unique?) optimal spherical
code of size 24 and rank 4. If true, this will of course imply
~(4) = 24. T h o u g h one knows next to nothing about the in-
tervening dimensions, one knows everything about the cases
d = 8 and 24: not only is the kissing number known in
these two cases, but one also knows t h a t the arrangements
in these two cases are unique. One must take the balls cen-
tered at the roots of the E8 root system in one case, and at
the m i n i m u m norm vectors of the Leech lattice in the other
case. These are two objects which play very important roles
in other parts of mathematics. For one thing, they are aw-
fully symmetric structures : their automorphism groups are,
respectively, the group O~-(2) : 2 of order 348,364, 800 and
the largest of the sporadic simple groups of Conway, of order
4, 157, 776, 806,543,360,000. They provide us with unique
optimal codes with (n, d) = (240, 8), (196560, 24) and mini-
m u m distance 1. T h e usual proofs of these assertions involve
Address for Correspondence moderately heavy dozes of analytic number theory, invok-
Bhaskar Bagchi ing, as they do, Theta and modular functions. The reader
Statistics & Mathematics Unit wishing to see these proofs may consult the book by J H
Indian Statistical Institute
Conway and N J A Sloane (see Suggested Reading). Alter-
R.V. College Post
Bangalore 560 059, India
native proofs may be constructed following the outline to be
[email protected] presented in Part 2.

26 RESONANCE J September 1997