Stationary Partitioning of Convex Bodies

JOHANNES C . C . N I T S C H E

A specific version of the general partitioning problem is considered here.

Let ~ be a convex body in R 3 of volume m3(~) < <x~. A disc-type surface
is spanned across the interior of ~ dividing this interior into two simply-connected
cells ~ ' and ~ " of volumes m 3 ( ~ ' ) = am3(~), m 3 ( ~ " ) - - - - ( 1 - a ) m 3 ( ~ ) ,
0 < tr < 1. The fraction ~ is considered as fixed, and we are interested in the
physically relevant case where Sr has smallest possible area, or at least stationary
area. Both conditions imply that the rate of change (the first variation) of the
area of 6a vanishes for all slight modifications of 60. It then follows that the sur-
face 6e is regular up to its boundary (the boundary regularity depending on the
properties of 8~), that its mean curvature H must be constant, and that 6,' must
meet ~ at a right angle. 1 A surface with these properties will be called stationary
or, more precisely, a stationary solution surface for the partitioning problem. We
may visualize 5~ as a membrane separating two imcompressible gases or liquids
in ~ .
I f ~ is a ball, any plane disc through its center (for tr ---- 1/2), or any spherical
cap orthogonal to ~ (for tr @ 1/2), satisfies the above conditions. The same is
true of the ellipses in the principal planes of a three-axes ellipsoid. Only one of
these ellipses is area minimizing (for fixed a = 1/2). For a specific tetrahedron,
H. A. SCHWARZ ([29], pp. 149-150) has determined and depicted a qualifying
minimal surface--see figure 12; and the case of general simplices has been discussed

x For area minimizing surfaces, this is well known. For stationary minimal surfaces,
it has recently been proved in [13], [11]. The general case of stationary surfaces with
H :~ 0 is the subject of a current investigation. The crucial step is always the continuity
proof for the trace of 5e on 8~'; see [19].
2 From [29]: "Hr. Kummer zeigte ein von Hrn. Professor Schwarz in Ztirich an-
gefertigtes Gypsmodell einer Minimalfl~iche vor, deren Begrenzung durch eine Reihe
von vier Ebenen gebildet wird, auf denen sie tiberall senkrecht stehen muss. Die von
2 J . C . C . NITSCHE

by B. SMYTH3. Generally, speaking, however, little information seems to be avail-

able concerning the general situation. Until recently it was even u n k n o w n whether
plane discs and spherical caps exhaust all possibilities for a ball 9~~. The p r o o f o f
this fact is one o f the purposes o f this paper.

Fig. 1

It should be emphasized that other stationary solutions exist if the topological

conditions on 6a and ~ ' , ~ " are altered. F o r instance, the catenoid 6a: t/x 2 + y2
= ~ eosh(z/o O, ]z] ~ 0.552, 0~ = 0.460 is the unique (up to rotations) stationary
minimal surface o f ring-type [1, 2, 0] s if ~ is the unit ball. The b o u n d a r y o f 6e
consists o f two components. Here the cells & ' and & " have genus 0 and 1, respec-
tively, and a = 0.544. The unique doubly-connected stationary (also area mini-
mizing) surface for the case tr = 1/2 is a portion ~ o f a Delaunay surface with
m e a n curvature H = 0.108. The area A[S/'] ---- 5.222 o f this surface is larger
than the area ~ o f a plane disc t h r o u g h the center o f ~ which also divides the
volume o f ~ equally. This fact is in agreement with the results o f [4].

Hrn. Prof. Schwarz zuerst allgemein gestellte und behandelte Aufgabe Minimalfl~.chen zu
finden, ... bietet namentlich in dem Falle, wo die Begrenzung durch eine Kette yon Ebe-
hen allein gegeben ist, einige Schwierigkeiten fiir die geometrische Anschauung dar, da
es scheint, als ob die so zu begrenzenden Fl~ichen jedes gegebene Maass der Kleinheit
iiberschreiten k~nnten." It is, of course, the volume constraint which mandates the ortho-
gonality condition at the boundary and, through it, the rigidity of the minimal surface.
a The minimal surfaces in p~n meeting a given simplex at right angles. Lecture at
Bonn University, July 3, 1981; see also [32].
4 This question arose during SMYTH'S talk; see [25], [26].
s The topological type Z[~] = [e, r, Z] of a compact connected surface ~ is
described by its orientability character e ( = :El), the number r of its boundary com-
ponents and its Euler characteristic Z. The genus g of an orientable surface is g = 1 --
(r + Z)/2. See [20] and [24], w42.
 Partitioning of Convex Bodies 3

More complex shapes for if', of higher topological type (and equally interesting)
are possible as well. Figures 2, 3 contain sketches for stationary surfaces of types
[1, 4, --2] and [1, 6, --4]. The first has tetrahedral symmetry, the second cubic
symmetry. Unlimited reflection across the plane boundary faces of &, made
possible in view of the orthogonality properties, leads to complete embedded sur-
faces effecting a peculiar division of space into two disjoint but highly linked do-
mains of infinite connectivity. In the case of minimal surfaces, similar structures
have been discovered and discussed extensively by H. A. SCHWARZ, E. R. NEO-
VlUS and others; for references see [24], ~ 2 7 6 - 2 7 9 , 369, 818, 882 and [22].
G. DO,NAY and D. L. PAWSON [10] have made the interesting observation that
the interface between inorganic crystalline and organic amorphous matter in the
skeletal elements of echinoderms have a striking similarity to such periodic mini-
mal surfaces. The work of these authors, as well as H. U. NISSEN [23], contains
convincing pictures. For H =t= 0, the subject of "bicontinuous" space dividing
structures is still in its infancy. It is of interest in chemical engineering applications
to micro-emusions and liquid crystals; see [30], [31], [21].

"7

Fig. 2 Fig. 3

The minimal surfaces of the shapes suggested in figures 2, 3 are composed of congruent
parts representing stationary minimal surfaces inscribed in tetrahedra. For the case of
figure 3 where 48 copies of the fundamental domain PIP2P3P4 are involved, the tetra-
hedron has been indicated by dotted lines. Except for a similarity, the curved quadri-
lateral PxP2P3P4 is the conjugate--and thus an isometric deformation--of the unique
minimal surface through the sides of the skew quadrilateral with vertices (0, 0, 0),
(1, 0, 0), (1, 1, 1) and (0, 1, 0), in this order, formed by two edges and two face diagonals
of a cube. In fact, a wreath of four adjacent copies of PIP2P3P4 constitutes precisely the
stationary surface of Schwarz depicted in figure 1. The six closed contours on the faces
of the cube are very nearly circular (of course, only the catenoid has exact circles); their
4 J . C . C . NITSCHE

diameter is half the side of the cube. In a suitable coordinate system, the arc P~P,,,
one-eight of such a contour, has the equations
t t
~e(t) = f - - ~
6
1 -- ~2
at,
9
r/(,)=
0
f 2~

TC(-~ d~,
0 --<t< r
_ =
1,

where W(t)= 1/1 + 1 4 t * + t a.

We shall represent 5e in parametric form, b r = (~ = ~(u, v); (u, v)E/S},

where P is the closure of the unit disc P : {u, v; u 2 + v 2 < 1} and ~(u, v) =
{x(u, v), y(u, v), z(u, v)} denotes the position vector o f 5e. It is assumed that 5"
is an embedded surface and that the unit normal vector ~ = I ~.•215
points toward N'. The area o f Sg is given by

0) A[,~] = ffl~,,• ~,,I du dv.

P

The cubic functional 89f f (r, r,,, to) du dv usually employed in the study of H-
P
surfaces must be replaced by another functional which expresses the volume
of N '6. This functional has the form

(2) V[Sal = ff(~(~), r., r.) du dr,

P

where ~3(q0)= (Ql(x, y, z), Q2(x, y, z), Q3(x, y, z)} is a properly chosen vector
function. It is an intriguing task to determine such a vector function in an explicit
form. F o r a general discussion see [27].
Let us suppose that the boundary 0N is smooth and that the surface Se omits
at least one point on ON. Choosing this point as the "lowest" point (--uc, 0, --c),
e > 0, ~ ~ 0, in a suitably selected (x, y, z)-coordinate system, the boundary,
assumed here, for simplicity's sake, also to be strictly convex, can be given the
representation
(3) 0N : (x : uz + f(z, ~) cos % y : f(z, cp) sin ~0; [ z I ~ c, 0 =< ~0 ~ 2z~}.

Here f(-4-c, ~0) -----0, lira f~(z, ~) = q: oo. We then find

(4) ~ : Q(z, cp) {(x - uz) f,(z, ~) q- uf(z, ~), yf~(z, qJ),f(z, ~0)}
with
C
Y
(4') Q(z, ~) = f-3(z, ~0) f f 2 ( ( , q0) d(, ~ = ~p(x, y, z) -----arc tan - -
Z X-- ~Z

X2 y2 7,2
F o r the case of an ellipsoid N: ~ - -}- ~ - + -~- =< 1, (4), (4') reduces to

2c+ z
(5) r = ~ ~ ( - x T ` , -yT`, c 2 - z2}.

!
6 The functional ~ f f (~, ~,,, ~v) du dv is applicable for convex polyhedra.
 Partitioning of Convex Bodies 5

The following properties of s are easily verified by dia-ect calculation:

i) ~(~) becomes singular only at the point (--zc, O,--c) omitted by Se

and has, possibly, a b o u n d e d discontinuity along the line l ---- {x = m ,
y = O; lzl <= c}.
ii) d i v ~ ( ~ ) = - - I in & \ l .
iii) F o r ~ E O& \ / , the vector s is tangent to ~ .

T o illustrate the mechanism o f formula (2), let us consider the case where
is the unit ball and ~ is the part of the cone ~/y2 + z 2 = x tan o~ contained in
~ . This part can be parametrized in the f o r m 6" = (x = ]/u 2 + v2 cos 0~, y =
usino~, z = v s i n ~ ; u 2 + v 2 ~ 1}. Using polar coordinates u = r
v=~sin0, we find

~,,• ~ = sin o~{sin o~, - - c o s o~ cos O, - - c o s ~ sin O}

2 + ~ sin o~ sin 0
~(~) : ~r (1 + 0 sin or sin 0) 2 (--02 COS 0(, sin o~ sin O,

__ 02 sin 2 o~ sin 0 cos O, 1 - - 02 sin 2 o~ sin 2 O}

and
2 + O sin 0~ sin 0
(~(~), ~., ~v) = -- 89 sin o~ cos o~ sin 0
(1 + e sin o~ sin 0) 2 "

T h e volume V[Se] o f ,~' becomes

2n 1
2 + ~ sin o~ sin 0
V[S~] = --~- sin o~ cos o~f dO f do 9 sin 0
0 0 (1 + 9 sin 0r sin 0) 2

2~
f sin 0
= - - ~ sin o~ cos o~ 1 + sin o~ sin 0 dO
0

2~r
= -~- (1 - - cos 0r

as it should be.
In order to investigate the change o f the functionals A and V associated with a
variation of the surface 6 e, we consider a one-parameter family o f surfaces
~ ( t ) = {~ = ~(u, v; t); (u, v) E P}, ] t ] < to, spanning the interior o f ~ such that
g(u, v; 0) = g(u, v). Setting t~(u, v) = gt(u, v; 0), standard c o m p u t a t i o n s show
that
0
7 ; A[~~ = - - r (t~, ~. dE) - - 2 f f n(t~, ~., ~o) du de
OP P

~-7
-.v,-,jrv,"--,,,=o = -
OP
(t~, ~(r), d~) + f f (div s
P
(t), ~,,, ~ ) du ely.
6 J . C . C . NITSCHE

a
The stipulation that ~-(A[se(t)] + 2V[se(t)])lt=0 vanish with respect to all vectors
t~(u, v) tangent to 8 ~ for (u, v)E ~P, leads to the conditions
i) 2 H = 2 div s = --2 = const.
ii) ~g + ;L~3(~) is tangent to ~ for ~ E ~ , that is, in view of iii) above,
is tangent to 8& (and S# orthogonal to ~ ) for ~ E a ~ .
It is on the basis of these formulas and the properties of s that the proofs
for the boundary regularity of Se are designed; see footnote 1.
Various bounds from below for the area A[Se], in terms oftr and the properties
of the body ~ (for instance its diameter) can be found in the literature; see [2],
[3], [4], [5], [9], [12], [14], [15]. [16], [18], [25], [32]. These bounds are sharp for
specific bodies--sphere, cube, cone, ellipsoid7--and have been used for the deriva-
tion of Sobolev estimates, in the geometry of numbers and for diophantine equa-
tions. They say little about genuinely stationary surfaces for which one might wish
to develop upper bounds. In fact, certain stationary surfaces possess largest
possible area and are solutions of a maximum problem complementary to the
usual minimum problem. The situation for the three-axes ellipsoid illustrates
this point.
I f & is the oblate ellipsoid of revolution
X2 y2
~--q- ~-T-]- ~ ~ 1, O<c<a,

then our partitioning problem for the simplest case ~ = 89 allows as stationary
solutions the continuum of vertical ellipses s

Se(~p) = (x = a cos v/u, y ---- a sin ~pu, z ---- cv; u 2 -k v 2 ~ 1}, 0 ~ ~p ~ 2z~

of minimum area A[Se0p)] = :rrac and the stationary disc 6" = (x = au, y = av,
z = 0; u 2 -k v 2 =~ 1) of area A[Se] = zm2 > A[se0p)]. Depending on the aspect
~x = a/c (or, equivalently, on the numerical eccentricity e = I/1 -- c2/a2), there
may be further stationary solutions. For t E (0, z~/2), let Bt be the sphere with
center (0, 0, c/sin t) orthogonal to 8~. Denote by Set the cap o f Bt in ~ and by
~ the part of ~ contained in Bt. The volume V[set] of &'t is equal to

2~ 2 ~
V[set] = ~ a c + -~- c3F(t; o~) : 89 ) -}- -~- c 3 F(t; 0~),

where
F(t, 00 : 0~2(--3 sin t + sin 3 t)
+ 2 cotan 3 t((cos 2 t + ~2 sin 2 03/2 _ cos 3 t -- -320~2cos t sin 2 t).

7 Even for these bodies, the extremal sets ~ , for general values of a, are often elusive.
a Note that u and v are not isothermal parameters on Sa(W). An isothermal represen-
tation of Se0p) has the form (x ----cos v,, Ref(w), y = sin ~pRef(w), z = Imf(w);
w l =< 1}, where ~ = ~ + # / = f(w), w = u + iv, provides the 1 - 1 conformal map-
ping of P onto the ellipse (~2/a2) + (~/2/c2) =< 1.
 Partitioning of Convex Bodies 7

We have F(0; ~) = lim F(t; ~) ---- 0, F ;~ = --20~2, SO that

t~0

v[ifo] = 89 VEif~/2]-- 0.
Further,

F'(0;oc)=3o~2[~---11, F"(0;cr F"'(0;o~)------~[12o~2--7o~4--o~6].

Note that F'(0; 2) ---- F"(O; 2) = 0, F"'(0; 2) = --96. These formulas (and
a subsequent more careful discussion of the function F(t; o0) show the following:
1) For a/c =< 2, there are no further spherical stationary solutions.
2) For a/c > 2, that is, if the ellipsoid ~ is sufficiently flattened, there exist
two further stationary solutions: the spherical cap if'to and its reflection on the
(x, y)-plane, where to = to(O0 is the root of the transcendental equation F(t; ~) = O.
The area o f these caps satisfies the inequalities ~rac < A[ifto] < ~ a 2 . The table
contains a few numerical values for the case a = 1 (~ = l/c).

1
c = 1/~ to(S) -- ,4IS, o] n, o

0.49 0.1726 0.9994 0.3402

0.45 0.3808 0.9866 0.7159
0.40 0.5313 0.9513 0.9588
0.35 0.6442 0.8989 1.1336
0.30 0.7392 0.8322 1.2853
0.25 0.8247 0.7521 1.4354
0.20 0.9060 0.6590 1.6015
0.15 0.9881 0.5514 1.8084
0.10 1.0777 0.4253 2.1095
0.05 1.1905 0.2701 2.6938
0.01 1.3535 0.0928 4.6377

The preceding discourse indicates the scope of the problems associated with
the partitioning theme even in the specialized form considered here9:
1) Existence. P r o o f of the existence of stationary (or area minimizing, area
maximizing) connected embedded surfaces i f of prescribed topological type
Z [ i f ] = [1, r, Z] which divide ~ into two cells &', ~ " of given volume ratio.
Even in the case Z [ i f ] : [1, 1, 1] (surfaces of disc-type) discussed here, no
existence theorem seems presently available in the literature.
2) Number of solutions. In general, there is no uniqueness. For given o, there
may exist stationary minimal surfaces, or stationary surfaces of non-vanishing

9 The general problem of minimal, maximal, or stationary partitioning of a domain

in R,n (or all of Rn) into a given number of cells with prescribed volume leads to aggregates
of surfaces and to century old problems in the sciences; see e.g. [1]; [3]; [6]; [7], pp. 337-
338; [33], pp. 465-644; [34].
8 J . C . C . NITSOI-IE

mean curvature, or both. The example of pear shaped bodies shows that not all
stationary minimal surfaces divide the volume of ~ in equal parts.
3) Foliation of ~ . A discussion of the feasibility of foliations of the interior
of ~ , or parts of this interior, realized by families of stationary surfaces, possibly
depending on the parameter H or the parameter tr, would be desirable. For the
sphere, there is, in addition to the family of spherical caps, for which H is a single-
valued function of a, also a foliation with ring-type leaves (Dealaunay surfaces)
containing a diameter as singular leaf. The example of the ellipsoid of revolution
with a = b > 2c shows that the relation between H and tr need not be injective.
A discussion of the example indicated in Figure 3 suggests that a foliation of all
of int (&) may not be possible with leaves of a prescribed topological type.
4) Stability of solutions. This involves a discussion of the second variation of
area under the given volume and boundary constraints.
5) Area estimates.
6) Isolated character of solutions. If the body ~ possesses symmetries, there
will be continua of stationary solutions 1~ In general, however, stationary solu-
tions appear to be isolated. The surface 5e _-- {~ = ~(u, v); (u, v) E if} is isolated
if there is a constant z > 0 such that a suitable norm ]]g(u, v) -- ~(u, v)[[ >
for all other stationary surfaces 57 = (~ = ~(u, v); (u, v) E if}. It is assumed here
that the volume ratio a is kept fixed, but that the mean curvature H of 57 may
differ from that of 5a. A relaxation of the requirement regarding a, or the imposi-
tion of further conditions on H, leads to other types of isolatedness. For instance,
it is the author's conjecture that the three ellipses in the principal planes of a three-
axes ellipsoid ~ are the only stationary minimal surfaces in ~ , even in the absence
of a volume constraint.
7) Rigidity of solutions. This is a weaker concept than isolatedness. Generically,
a stationary solution surface 5~ = {~ = ~(u, v); (u, v) E P} is rigid in the sense
that no one-parameter family of stationary surfaces 5 e ( t ) = (~ = ~(u, v; t);
(u, v) E P}, ]tl <Z to, exists for which ~(u, v; 0) = ~(u, v). Here a is kept fixed,
but it is assumed that the mean curvature of the surfaces or is a function
H---- H(t) of the parameter t as well. We shall say that 5e is a-rigid.
The rigidity of stationary surfaces lends suspense to the understanding of
the physical model underlying the partitioning problem. Intuitively, one might
think that a stationary non-area minimizing surface 5a which separates two
incompressible liquids in ~ should move through a sequence of stationary states
into an area minimizing position. Of course, such a volume preserving change of
shape is always possible under violation of the transversality condition at the
boundary.
We shall say that 5e is H-rigid if there is no one-parameter family of surfaces
~ ( t ) as above, all having the same mean curvature as 5".
8) Geometric properties of stationary solutions. Are disc-type solution sur-
faces, area minimizing or stationary, actually graphs ?

lo Conversely, the presence of such continua allows conclusions regarding the sym-
metries of & (if ~ or parts of 0~ are analytic). Spheres, ellipsoids, cubes lead to in-
structive examples, particularly for the cases cr :6 112.
 Partitioning of Convex Bodies 9

9) Properties of the trace / ' = (~ : ~(cos 0, sin 0); 0 g 0 --< 2~} of .9' on
~ . F o r a tetrahedron ~ , the stationary minimal surfaces determined by B. SMYTH
[32] have connected intersections with all faces of ~ . The stationary spherical
caps do not have this property. On a minimal surface, the integral f ~ •
is p a t h independent; see [24], w 71. We can introduce the arc length s o n / ' and set
(cos O(s), sin O(s)) : tg(s). Since Idol : ds and since the vector ~ • dE is
parallel to the o u t w a r d unit n o r m a l vector 9~(~) of 8 ~ at ~, we find Smyth's
identity

(6) f 9Z(O(s))d s = O.

SMYTH calls curves on O& having this property balanced curves. F o r such curve
there exist length estimates; see [8].
A discussion of the above p r o b l e m areas in the simpler but equally fascinating
setting o f R 2 is instructive. Also, the restriction o f one's consideration to stationary
minimal surfaces, discarding any requirement a b o u t a, often leads to concrete
conclusions. In the following, we shall elaborate on the questions 7) and 8) for
specific bodies (sphere, ellipsoid). Other bodies (cylinder, torus) and other topo-
logical types are amenable to the methods developed as well.

Theorem 1. For a ball ~ , all stationary surfaces of type [1, 1, 1] are plane
discs through the center (a : 1/2) and spherical caps orthogonal to ~ (tr fi= 1/2).

O u r second t h e o r e m deals with the oblate ellipsoid o f revolution

X2 y2 Z2
~ : - -a+2- - r + a- -~< l , c 2 = 0<c<a.

T h e disc 6e = (x = au, y : av, z : 0; u 2 + v 2 "< 1} is stationary, but not area

minimizing for the case a : 1/2. 6e is a m e m b e r of a one-parameter family o f
stationary surfaces Se(t) : (~ ---- ~(u, v; t); u 2 + v 2 ~ 1}. These surfaces are
caps of spheres with center on the z-axis, with the isothermal representation
(w = u + iv)

2a cos tW(t) [W(t) + c cos t]

W
x + iy : [W(t) + c cos t] 2 + a 2 sin 2 tw-ff
(7)
2a 2 sin t cos tW(t) w~ c -- cos tW(t)
Z = +
[W(t) + c cos t] 2 + a 2 sin 2 tw'w sin t

Here W(t) = ]/a 2 sin 2 t + C 2 COS 2 t. Note that the additive part in the z-coordinate
is regular for t = 0. The mean curvature of oq~ is

(8) H(t) : W-l(t) tan t.

10 J . C . C . NITSCHE

An expansion for small values of t gives

[ a2 ]
x+iy=aw+aw -- 89 t2+...

(9) z=c 1--2--~-(1--ww-" ) t + . . .

1
H=--t+ ...
c

Theorem 2. (a) 6a is H-rigid, that &, ~ & no member of an analytic one-parameter

family o f stationary minimal surfaces.
(b) 6a is a-rigid, that is, 5e is no member of an analytic one-parameter family
of stationary surfaces dividing J# into parts of equal volume.

We shall assume that the stationary surfaces are before us in an isothermal

representation so that ~2 = ~2, ~.~ = 0. Since the bodies considered here have
analytic boundaries, ~(u, v) is analytic in if; since they are convex, and Se is em-
bedded, there can be no branch points on 5e o r / ' , so that ~ ~ E(u, v) ~> 0 in
_P. We denote by L, M, N the coefficients of the second fundamental form and by
K the Gaussian curvature of 6a. For isothermal parameters, the fundamental
equations which can be found in textbooks on differential geometry reduce to
the following
E. E~

E. Zo

L M
(10) x.= --~u - -~

M N
~o= --g~.---g~o
L v -- M,, = E v H , M~ -- N,, = - - E . H

H----
L + N
K--
LN -- M 2 1 A log E.
2E ' E2 2E
The equations (106) can be written in the form
a 8
(11) e-'-~ [ 8 9 N)- iM] : E-~w H
 Partitioning of Convex Bodies 11

and show that the complex valued combination 89 -- N) -- iM is an analytic

function on every stationary surface 6a11. We shall work here with the derived
function

(12) g(w) : [89 ,v) -- N(u, v)) -- iM(u, v)] (u + iv) 2 ~ A(w) + iB(w).

It will be convenient for us to rewrite formulas (10)using polar coordinates

oe i~ w = u + iv:

~Q= t~ Q 2E ~ t o + + EH

1 ' E o [ ' ~ ] 1 1

1 Eo 1 1 --EH~

(13)

1 1 1 (1 A )1

K = H2 [gl2 1
~4E2 -- 2E A log E.

Note that g(w) is analytic on surfaces of constant mean curvature, but that the
formulas (13) have general validity. We shall use the notations I(u, v) and I(w)
interchangeably.
F o r the p r o o f of Theorem 1, let us now assume that 9~ is a ball, say, the unit
ball [ E I =< 1, and that 5~ is a stationary surface meeting ~ at a right angle. This
condition is expressed by

~o(ei~ = 2(0) ~(~(ei~ = 2(0) ~(d~

where 2(0) -----El/2(ei~ > 0. Differentiation with respect to 0 (along coP) yields

L~o(ei~ = Z'(O) ~(d ~ + 2(0)~o(e '~

z'(0)
= ,~(0) ~(~;0) + 2(0) ~0(~;~

A comparison with (132) shows that B(e i~ = 0. Since B(w) is harmonic in P,

we have B(w) =-- 0 and thus A(w) =-- const. The constant must be zero because

11 This is a well known fact utilized in the study of minimal surfaces and surfaces
of constant mean curvature, notably by H. Hopf [17] in his proof that a compact orient-
ablesurface of constant mean curvature and genus zero must be a sphere. The function
vanishes at the umbilic points of 5e.
12 J . C . C . NITSCn~

A(0) = 0. As a consequence, g(w) =-O, that is, L(u, v) = N(u, v), M ( u , v) : 0

so that equations (134,5) are reduced to

~,~ = - - H ~ , ~.o = - - H r o or ~ = - - H r , , ~ = - - H r s .

I f H = 0, we see that ~(u, v) is a constant vector, and 5r must be the portion

o f a plane. I f H 4 0, integration yields H(~(u, v) -- ~o) = ~.(u, v), or
I~(u, v) -- ~ol = 1/[ H I , and 5r must be a spherical cap. This proves T h e o r e m 1.
It can be observed that H is uniquely determined by a (5 r is not); it is a de-
creasing function o f a satisfying

lira H = c~, H(89 = 0, lira H ---- -- cx~.

a-+0 a-+l

We turn next to the p r o o f of T h e o r e m 2 for which we can assume, without

loss o f generality, that a : 1. Let us suppose that there exists an analytic family
o f stationary surfaces ff'(t) : {~ : - ~(w; t); w E P} with position vectors ~(w, t) ----
{ x ( w ; t ) , y ( w ; t ) , z ( w ; t ) } such that 5" : i f ' (0) : (x : u , y : v, z : 0 ; u 2 q- v 2 "< 1}.
In view o f the possibility o f a preliminary conformal m a p p i n g (depending on the
p a r a m e t e r t) it can be arranged that u and v are isothermal parameters on all sur-
faces 5r Moreover, the m a p p i n g can be adjusted so that x(0; t) : y(0; t) ----
y(1; t) : 0 for all values o f t.
Since H(0)----0, we have
H(t) : H i t + 1"12t2 ~- ...
We shall set

x(w; t) = u + ~(w; t), y(w; t) : v -1- ~/(w; t), z(w; t) = ((u; t)

and
~(w; t) = t~<l)(w)+ t2~t2)(w) + . . .
~(w; t) = t~'(w) + t2~t2)(w) + . . .
((u; t) = t"(<p)(w)+ tp+l(~"+l)(w) + ...
where p ~ 1, (~ ~ 0. The real valued function 2(0; t) occurring in the bound-
ary condition

1 z(e io ; t)}
~e(e~ t) : 2(0; t)(x(ei~ t), y(ei~ t), ~-~

has the f o r m

2(0; t) = 1 + / ~ ( 0 ; t), #(0; t) ----#x(0) t -+-/t2(0) t 2 + ...

F o r compactness's sake, we shall use the complex notation:

8 ~ 8 8 ~
2 -- Ou i'z-, 2 -- Ou + i - z - , e~=w~+w~,
Ow OV tTV

92
4 - - A.
Ow O-~ --
 Partitioning of Convex Bodies 13

The conditions to be satisfied by the position vector ~(w; t) are the following:
1) u, v are isothermal parameters and x(0; t) = y(0; t) = y(1; t) = 0.

(14) (~ q- i~7)~ ---- - - ~ -- (t q- in); (~ q- in)w for Iwl ~ 1.

(15) ~(0; t) ---- n(0; t) ---- n(1; t) = 0.

2) 5"(t) is bounded by ~ :
1
(16) 2 Re [(~ + i~/)~] q- (~ q- in) (~ q- i~/) q- --r ~2 : 0 for I wl = 1 .

3) 5e(t) must meet 8 ~ at a right angle:

(17) [(~ q- #l)e -- (~ q- in)] = #w -k #(~ + in)
for I w l = 1
1 1
(17') (o = ~ - ( q- ~ - / ~ ,
so that in particular
(18) /z ----- [(~ -k i~) o -- (~ -k in)] ~ -- #(~ -b in) ~ for I w[ -----1.
4) 5a(t) is a surface of constant mean curvature H(t):

Ar = 2H(--r + (~.r -- ~or

(19) A~/= 2H{--r -4- (r -- r
A~ = 2H{1 + (~. + ~,) + (~.n. -- #o~/.)}9

We shall also have to refer to the expansion for the volume of the cell ~ ' .

v[~(t)] : f f(~(r(w; t)), r.(w; t), to(w; t)) du dv

P
(20)
2ygr
+ V l t + V2t z + ....
3

Here s is the vector function (5).

Consideration of the linear terms in (14) shows that (~o) + ino))~ = 0, so
that

~o) + ino) : ~ anWn

n=O
t ~
where a. = a. + ta. are complex coefficients. It follows from (16) that

2 Re [(~o) + ino)) ~] __ ~ (Re a.w "-1) = 0 for w ---- e i~

rt=0

,t
and as a consequence, ~u) + ino) = ia~'w. But al must vanish in view of (15).
We find that ~o) + ino) _~ 0, and also #1 : 0. A repetition of the same argu-
14 J. C. C. NITSCHE

merit shows that

(21) ~o)+i~o)=0 for j = 1,2 . . . . , 2 p - - 1
(21') /~j = 0 for j = 1, 2, ..., 2p -- 1,
and. because of (193).
(2r') //j=0 for j = 1,2 . . . . . p - - 1,
while
(22) Ar = 2Hp in P
and
(22') (?) 1 T'0')on aP.
~'C2~

Let us now first assume that Hp = 0. Then ~(P)(w) is a harmonic function

satisfying the boundary condition (~p)= __1 ~p). For 1 / c 2 ~ 2,3, this
C2 "''~
boundary value problem has no non-trivial solution. If 1/c 2 = N ( = 2, 3 . . . . ),
we look at the functions ~(2v),~/(2v), ~(p) and proceed as follows. ~(P) must have
the form
a w N q- -~~- -wN ,
t 0') ---- Re (aw N) = --~

so that, in view of (t4), (15)

N2 ~Z~2N- I ~]
~(2p) .31.- i9~(2p)= 4(2N -- 1) -k .=1 b,~w".

Condition (16) requires that

2 Re [(~(2p) q_ irl(2p)) ~] ._~ N[~(p)]2 _~_ 0 for !w/ ~ 1
and leads to
N2 ~2-~2N-1 N(N- 1) a2w2N+ 1
~(2p) + ir/(2p) = blw 4(2N -- 1) 4 ( 2 N - 1) '
as well as
N 2 ( N - - 1) [a2w2N eiO.
/z2p(0) ~ 2(2N -- 1) ~_ ~-~2u], w=

The function r satisfies the differential equation Ar ~ = 2H3p (a conse-

quence of (19a), since H. = 0) and the boundary condition
~teaP)= N~ <3p)+ N/t2p(O) ~(P>
N 3 N--1
= N~ (3p) 2 2 N - - 1 { l a l 2 R e ( a r d v ) + Re(a3w3N)} for ] w [ : l .

For N ~ 2, a ~= 0, this boundary value problem has no solution. Theorem 1 a

is proved also for the case 1/c 2 ----- N.
 Partitioning of Convex Bodies 15

To be applicable to the proof of Theorem 2 b, the preceding argument requires

some additions. In general, the assumption Hp ---- 0 is not permitted a priori and
has to be recovered from a consideration of the volume functional. A direct com-
putation for the stage (21), (21'), (21") shows that the coefficients Vj in (20)have
the following values
Vj = 0 for j = 1, 2, . . . , p -- 1

(23) Vj = -- ff r du dv for j = p, p + 1. . . . . 2p -- 1
P

2c r 2 2 }
v2e --- ff {_r + 3 "" + ~17")) + ~c [(((P)) + O((P)(g)] du dv etc.

We start with the c a s e 1/c2=t = 2, 3 . . . . Then the general solution of (22)

(22') is ((P) -----o~ + flw~, where
c~ = ( d - 89 Hp,
We must have

Vp = -- f / ((') du dv = - - ~ ~ + = -- T (4c2 - 1 ) Hp= O

and come back to the conclusion lip = 0 which leads to a contradiction as before.
If 1/c 2 = N then the general solution of (22), (22') is

r = ~x + flw-f + Ta wN + T~ -fiN, O~= (_~_ -- 89 H e, fl = 89Hp.

Again,

lip= - ff r du dv = -- ~r o~ +
~p

and H e = 0, unless N = 4. Note that the remaining case N----- 4 corresponds

to the borderline situation discussed earlier in which the aspect of the ellipsoid
has the value 2, and where F(t; 2) ----- --16t 3 + .... If 1/c 2 ---- 4, then
a a_
r = ~x + ~w-f + T w4 + T w4; ~x = -- 88He, fl = 89Hp.

Substitution into (14), (15) and integration yields

~(2p) _31_ i~(2p) = - - f l 2W2-f - - fl ~W-f4 - - T4 a2-f7 @- ~ bnw~.

n=l

The coefficients bn will be determined with the help of relation (16) and lead to
~(2p) _~ i~/(2p) : - bl w _ f12w2-~ _ flaw 5 _ fl~w-f4 _ 4 ~12w7 _ 3 a2w 9,

where bl = (fl2/2) -- [a[ 2) + ibm'. With the help of (18) we compute

me(o) = -2r 2 - 4Caw' - 4/3a-f' - ~7' a 2w8 - ~ a ~-f8.

16 J . C . C . NITSCHE

The function r satisfies the differential equation

A~ 0p) = 2H3p q- 2Hp(~ 2p) -[- ~/~2p))
(24)
= 2H3,, + 4H,, Re (b, - - 2/32w~ -- 5/3aw ~ - - / 3 ~ -- 2_:a~w s)
and the b o u n d a r y condition
(25) 9l-op)
,~@ = 4(op) + 4/~2p(0) r for w = e ;~9

As before, the p r o b l e m (24), (25) can have a solution only if the coefficients o f
w 4, ~4 in (25) coincide. This leads to the condition a(7/32 q- 5[al z) ----0. Since
it is assumed that r not vanish identically, it follows that a-----0, so
that, because of (15),

r = ~ q- flww, ~ ---- -- 88Hp, fl = 89Hp,

and
~e(2t') _.]_ it/(2,)
_~_ 89 w __/32W2 ~ : __/3W~(p).

We n o w rely on the expressions V2p, V2p+l . . . . f r o m (23) and discuss here the
case p : 1; the a r g u m e n t for p > 1 follows the same lines. A c o m p u t a t i o n
shows that I:2 ---- 0. The integrand of the variation I:3 is
__r + ~[~(3)+ 7~3)] __ r + 7](v2)]

+ ~8 [~(1)r - (~(1))3] + x4 e ~0 [r162 _ (r

and the integration of (14)-(19) gives

r : __ 88 __ 2ww-) q- aw 4 q- a~'~
~(3) _~_ i~(3) = __ 2/3w r

~-(3) = _ 88 1 _ 2ww) -}- 89 -~- 2ww -- 2w2w 2) -[- bw'* -q- b-w'.

We then find
2~ ~ 3
v~ = --g-/3~ - i~ H , .

Since Vj, must vanish, we conclude that Hp ---- 0; the desired contradiction has
been established also in the case 1/c 2 = 4.

Acknowledgement. The preceding research was supported by the Alexander yon Hum-
boldt Foundation. The author wishes to thank the Humboldt Foundation, as well as
the University of Heidelberg for their hospitality during the summers, 1982, 1983.

Bibliography

1. ALMGREN,F. J. : Existence and regularity almost everywhere of solutions to elliptic

variational problems with constraints. Mem. Amer. Math. Soc., vol. 4, 165 (1976).
2. BOKOWSKI,J., Ungleichungen f'tir den Inhalt von Trennfl~chen. Arch. d. Math., to
appear.
 Partitioning of Convex Bodies 17

3. BOKOWSKI,J. : Schranken fiir Trennfl/icheninhalte in konvexen K6rpern. Preprint, 1982.

4. BOKOWSKI,J., • E. SPERNER: Zerlegung konvexer K6rper dutch minimale Trenn-
fl~ichen. J. Reine Angew. Math. 311/312, 80-100 (1979).
5. BOKOWSKI,J., & J. S. WILLS:Eine Ungleichung zwischen Volumen., Oberfl~iche und
Gitterpunktsanzahl konvexer Mengen in R 3. Acta Math. Acad. Sci. Hung. 25,
7-13 (1974).
6. BoYs, C. V.: Soap bubbles: Their colours and the forces which mold them. New York:
Dover 1959.
7. COLLATZ,L., & W. WEI"q"ERLING: Optimization problems. Berlin-Heidelberg-New
York: Springer 1971.
8. CROKE,C. B., & A. WEINSTEIN:Closed curves on convex hypersurfaces and periods
of nonlinear oscillations. Invent. math. 64, 199-202 (1981).
9. DEGIORGI, E., Sulla differenziabilit~t e l'analiticit~t delle estremali degli integrali
multipli regolari. Mem. Accad. Sr Torino (3) 3, 25-43 (1957).
10. DONNAY,G., & n . L. PAWSON: X-ray diffraction studies of echinoderm plates. Science
166, 1147-1150 (1969).
1 I. nZlUK, G. : Lrber die Stetigkeit teilweise freier Minimalfl/ichen. Manuscr. math. 36,
241-251 (1981).
12. FEDERER,H.: The area of a nonparametric surface. Proc. Amer. Math. Soc. 11,
436-439 (1960).
13. GROTER, M., S. HILDEBRANDT,• J. C. C. NITSCHE: On the boundary behavior of
minimal surfaces with a free boundary which are not maxima of the area. Manuscr.
math. 35, 387-410 (1981).
14. HADWIGER,H.: Gitterperiodische Punktmengen und Isoperimetrie. Monatsh. Math.
76, 410-418 (1972).
15. HADWIGER,H.: Lrber die Fl~icheninhalte ebener Schnitte konvexer K~Srper. Elem.
Math. 30, 97-102 (1975).
16. D. HENSLEY: Slicing the cube in t~n and probability. Proc. Amer. Math. Soc. 73,
95-100 (1979).
17. H. HOPF: Lrber F1/ichen mit einer Relation zwischen den Hauptkriimmungen. Math.
Nachr. 4, 232-249 (1950/51).
18. ISELI, M.: Llber die Fliicheninhalte ebener Schnitte konvexer K6rper. Elem. Math.
33, 129-134 (1978).
19. J.X.GER,W. : Behavior of minimal surfaces with free boundaries. Comm. P. Appl.
Math. 23, 803-818 (1970).
20. KERI~KJ.~RT6,B. v. : Vorlesungen fiber Topologie I. F1/ichentopologie. Berlin: Sprin-
ger 1923.
21. LONGLEY,W., c~; T. J. MCINTOSH: A bicontinuous tetrahedral structure in a liquid-
crystalline lipid. Nature, 303, 612-614 (1983).
22. NAGANO,T., & B. SMYTH: Periodic minimal surfaces. Comment. Math. Heir. 53,
29-55 (1978).
23. NISSEN,H. O." Crystal orientation and plate structure in echinoid skeletal units.
Science 166, 1150-1152 (1969).
24. NITSCHE,J. C. C.: Vorlesungen fiber Minimalfliichen. Berlin-Heidelberg-New York:
Springer 1975.
25, NITSCHE,J. C. C. : Stationary minimal surfaces in spheres. Preprint, 1981.
26. NITSCHE,J. C. C. : Partition of a convex body by a surface of stationary area. Ab-
stracts Amer. Math. Soc. 3, 248 (1982).
27. NITSCHE,J. C. C. : A volume formula. Analysis 3, 337-346 (1983).
28. SCHMIDT,W. M.: Volume, surface area and the number of integer points covered
by a convex set. Arch. d. Math. 23, 537-543 (1972).
18 J . C . C . N1TSCHE

29. SCHWARZ,H. A.: Gesammelte Mathematische Abhandlungen. Vol. I. Berlin: Sprin-

ger 1890.
30. SCRIVEN, L. E.: Equilibrium bicontinuous structure. Nature 263, 123-125 (1976).
31. SCRIVEN,L. E.: Equilibrium bicontinuous structures. In: Miscellization, solubiliza-
tion, and microemulsions. Vol. 2, pp. 877-893. K. L. Mittal ed. New York-London:
Plenum Press 1977.
32. SMYTH, B.: Stationary minimal surfaces with boundary on a simplex. Invent. math.
76, 411-420 (1984).
33. THIOMPSON,D. W.: On growth and form. New ed., Univ. Press, Cambridge, 1942.
34. THOMSON,Sir W.: On the division of space with minimum partitional area. Acta
Math. 11, 121-134 (1887-88).
35. ZINDLER,K.: Sehnen, die Umfang und Inhalt halbieren. Monatsh. Math. Phys. 31,
44-45 (1921).

Added in proof, October 4, 1984: Since the distribution of this paper as preprint,
summer 1982, progress has been achieved regarding the question of existence. F o r a
body M with C* boundary 0~, M. STRUWE proves the existence of a disc-type minimal
surface spanning the interior of ~ (Invent. math. 75, 547-560 (1984)). If ~ is strictly
convex, M. GROTER and J. JOST have now ascertained the existence of such an embedded
minimal surface. The regularity properties of the vector fields ~($) are discussed by
M. Ki)STER (SFB 72 preprint no. 638, Bonn, 1984). The author has extended Theorem 1
to ring-type surfaces in the unit sphere and in certain unbounded domains. Here all
stationary solutions are Delaunay surfaces. In connection with the study of macroscopic
and microstructured fluid interfaces D. M. ANDERSONand L. E. SCRIVENhave developed
a new numerical algorithm for the computation of surfaces with prescribed mean cur-
vature (Siam Summer Meeting 1984, Final Program, p. 26). Computer generated pic-
tures of the surface in the unit cube depicted in Figure 3, for the values H = 0.6, 1.2,
1.8, 2.1, were kindly made available to the author and are reproduced below. (Note
that the inscribed sphere which touches the six faces of the cube in their center has mean
curvature 2.) A more detailed discussion of such periodic surfaces is the subject of an
investigation by D. M. ANDERSON, J. C. C. NITSCHE, H . T . DAVIS and L. E. SCmVEN.

H = 0.6 H = 1.2
 Partitoning in Convex Bodies 19

H = 1.8 H = 2.1

Further pictures of the labyrinths alluded to on p. 3 which occur so abundantly in

nature, science and architecture can be found on p. 444 in the 1983 Brittanica Book of
the Year and on p. 1138 in Science 225 (1984). In this regard, also the following
works should be added to the bibliography:

[27a] PEARCE, P. : Structure in nature is a strategy for design. Cambridge, Mass. : The
M I T Press 1978.
[27b] ROLAND, C. : Frei Otto: Tension structures. New York-Washington: Praeger 1970

An inspection of the p r o o f shows that Theorem 1 remains valid for any constant
contact angle (capillary surfaces in the absence of gravity). The investigation mentioned
in footnote 1 will appear in Acta Mathematica.

School of Mathematic
University of Minnesota
Minneapolis

(Received December 3, 1982; revised May 25, 1983)