International Electronic Journal of Geometry

c
Volume 3 No. 2 pp. 67101 (2010) IEJG

SURFACES WITH CONSTANT MEAN CURVATURE IN

EUCLIDEAN SPACE

RAFAEL LOPEZ
(Communicated by H.Hilmi HACISALIHOGLU)

Abstract. This is an expository article. It discusses some topics on the theory of constant mean curvature (CMC) surfaces with non-empty boundary.
The paper starts with a simple introduction to the mean curvature of a surface giving different physical and mathematical motivations. Next we analyze
the mean curvature equation giving the Tangency Principle and the Alexandrov reflection method. The main part of the work focuses in surfaces with
non-empty boundary showing how the geometry of the boundary imposes geometrical restrictions to the surface. Finally we discuss the Dirichlet problem
associated with the mean curvature equation and some of the techniques employed in this context.

Contents
1. The mean curvature of a surface
2. CMC surfaces
3. Some special CMC surfaces
4. The constant mean curvature equation
5. The Alexandrov theorem
6. The effect of the boundary in the shape of a CMC surface
7. Some uniqueness results on CMC surfaces
8. Embedded CMC surfaces
9. CMC surfaces with circular boundary
10. Two equations for CMC surfaces
11. The Dirichlet problem of the CMC equation
References

68
72
76
78
81
82
87
89
92
94
97
100

2000 Mathematics Subject Classification. 53A10.

Key words and phrases. mean curvature; Maximum Principle; Alexandrov theorem; balancing
formula; a priori estimates.
Partially supported by MEC-FEDER grant no. MTM2007-61775 and Junta de Andaluca
grant no. P09-FQM-5088.
67

68

RAFAEL LOPEZ

1. The mean curvature of a surface

What is the curvature of a curve? What does it mean that a surface is curved?
The curvature of a curve is its acceleration. For this, we need to move throughout
the curve at constant speed. Consider a (regular) curve in Euclidean space R3 ,
that is, a differentiable map : I R R3 , = (s), such that 0 (s) 6= 0 for
any s I. We say that is parametrized by the arc length if |0 (s)| = 1, s I.
We point out that any regular
can reparametrize by the arc length. So,
R t curve
0
it suffices to define s(t) = t0 | (u)|du. Because s0 (t) 6= 0, s : I J R is a
diffeomorphism, where J R is an interval. Then taking (s) = ((s)), we have
1
1
0 (s) = 0
0 ((s)) = 0
0 ((s)) | 0 (s)| = 1.
|s (t)|
| (t)|
Definition 1.1. Given a curve : I R3 parametrized by the arc lenght, the
curvature of at s is defined by
(s) = |00 (s)|.
When the curve is defined in Euclidean plane R2 , : I R2 , we can assign a sign
on the curvature . We say that is positive if the curve goes to left and negative
if it goes to right. Exactly, because h0 (s), 0 (s)i = 1, then h00 (s), 0 (s)i = 0 and
this means that the acceleration 00 (s) is orthogonal to the velocity 0 (s). Let ~n(s)
be the /2-rotation of the vector 0 (s) in the counterclockwise sense. Then 00 (s)
is proportional to ~n(s): 00 (s) = (s)~n(s). We say that (s) is the signed curvature
of . We point out that in this case, |(s)| = |00 (s)|. Some elementary properties
and facts about are the following:
(1) If (s) = (s), then (s) = (s).
(2) If parametrizes a straight-line, (s) = sv + p, |v| = 1, then 0.
(3) If is a circle of radius r > 0, (s) = (r cos(s/r), r sin(s/r)), then 0 (s) =
(cos(s/r), sin(s/r)) and ~n(s) = ( sin(s/r), cos(s/r)). Thus (s) = 1/r.
(4) If is the graph of a function y = f (x) and we write (t) = (t, f (t)), then
(1.1)

(t) =

f 00 (t)
.
(1 + f 0 (t)2 )3/2

Now, we consider a surface M R3 , or X = X(u, v) : D R2 R3 a differentiable

map such that X(D) is a surface of R3 . For each p M there exists a tangent plane
Tp M formed by all velocity vectors at p: Tp M = {0 (0); : I M, (0) = p}.
Fix N (p) a unit vector orthogonal to Tp (s). Consider all planes P containing N (p),
which are transverse to M at p. Then P M is a planar curve containing p, called
a normal section. See Figure 1. We take the orientation on the curve such that the
normal vector to this curve is N (p). Each plane P is given by a tangent direction
v Tp M . We parametrize P = Pv , where v S1 (p) = {v Tp M ; |v| = 1}. Denote
v = Pv M and let
v (p) = v (0).
By the compactness of S1 (p), there exists v1 , v2 Tp M such that
1 (p) := v1 (0) = max{v (0); v S1 (p)}.
2 (p) := v2 (0) = min{v (0); v S1 (p)}.
Definition 1.2. The numbers i (p) are the principal curvatures of M at p, and vi
are the principal directions.

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

69

Figure 1. Normal sections; a plane, a sphere, a cylinder.

The principal directions at each point are orthogonal. Moreover, if we change N
by N , the signs of the principal curvatures change.
We are in position to define the curvature of M at p as a type of average of
the principal curvatures, for example, geometrical or arithmetic average:
Definition 1.3. The Gauss curvature K(p) and the mean curvature H(p) are
defined respectively by
1 (p) + 2 (p)
.
2
All concepts are invariant by rigid motions of space, except perhaps, by a sign.
In fact, the change of N by N implies that H changes of sign but do not K.
K(p) = 1 (p)2 (p),

H(p) =

Definition 1.4. An orientation (or a Gauss map) on a surface M is a differentiable

map N : M R3 such that |N (p)| = 1 and N (p)Tp M for each p M .
Any surface is locally orientable, that is, given a point p M , there exists a
neighborhood V of p at M such that V is an orientable surface: if X : U R3 is
a local parametrization of the surface around p, we define
Xu Xv
N (X(u, v)) = N (u, v) =
(u, v).
|Xu Xv |
Here is the vectorial product and the subscripts u and v denote the corresponding
derivatives. Therefore X(U ) M is an open set of M oriented by N X. We also
point out that closed surfaces (compact with not boundary) are orientable thanks
to the existence of a interior domain of the surface: see Figure 2

Figure 2. Local orientability in a surface; a closed surface is orientable.

We show some calculations of principal curvatures:


RAFAEL LOPEZ

70

(1) M is a plane. Because N is constant, the normal sections are straight-lines.

Then i = 0 and K = H = 0 on M .
(2) M is a sphere of radius r > 0. Take N pointing inside. The normal
sections are maximal circles on the sphere, that is, circles with radius r.
Then i = 1/r and K = 1/r2 and H = 1/r.
(3) We say that p is a umbilical point if 1 (p) = 2 (p). The surface is called
umbilical if all its points are umbilical. The only umbilical surfaces are
(open of) planes and spheres.
(4) M is a cylinder of radius r > 0. Consider N pointing inside. Then 1 = 1/r
and 2 = 0 corresponding to each circle of radius r and any straight-line
on the surface, respectively. Thus K = 0 and H = 1/(2r).
In all above cases, the principal curvatures are constant, and so, K and H.
We focus our attention on the mean curvature.
Facts:
(1) The inequality (1 2 )2 0 writes as H 2 K. Moreover p is a umbilical
point iff (H 2 K)(p) = 0.
(2) Let p M and v, w S1 (p) two tangent vectors such that hv, wi = 0. Then
H(p) =

1
(v (0) + w (0)).
2

As consequence, one can compute the mean curvature H along two any
orthogonal directions.
(3) As an application, consider a surface given as a graph z = f (x, y) and
we calculate the mean curvature at a point p. After a rigid motion of the
space, we assume that p = (0, 0, 0) and that the tangent plane is horizontal.
Consider the parametrization X(x, y) = (x, y, f (x, y)), with f (0, 0) = 0.
The tangent plane at p is generated by the vectors x X and y X, that is,
(1, 0, fx ) and (0, 1, fy ). Because the tangent plane is horizontal, fx (0, 0) =
fy (0, 0) = 0. Thus the intersection of M wih each one of the coordinate
planes {y = 0} and {x = 0} determine orthogonal curves at p. These curves
are (x, 0, f (x, 0)) and (0, y, f (0, y)) and the curvatures as curves are given
by (1.1):
fxx
(0, 0) = fxx (0, 0),
(1 + fx2 )3/2

fyy
(0, 0) = fyy (0, 0).
(1 + fy2 )3/2

Finally
H(p) =

 1
1
fxx (0, 0) + fyy (0, 0) = f (0, 0).
2
2

Here is the Laplacian operator on R2 . This equation shows that the

theory of surfaces with constant mean curvature are stronger related with
the study of elliptic PDEs.
Consider a Gauss map N : M S2 and the differentiable map dNp : Tp M
TN (p) S2 Tp M . This map is defined by

d 
dNp (v) =  N ((t)),
dt 
t=0

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

71

where : I M is curve on S such that (0) = p and 0 (0) = v. Then dNp is an

endomorphism, which is self-adjoint, that is,
hdNp (u), vi = hu, dNp (v)i, u, v Tp M,
or equivalently, the bilinear form p : Tp M Tp M R given by
p (u, v) = hdNp (u), vi
is symmetric. In particular, both dNp and p are diagonalizable. The proof of this
fact is as follows. Take a local parametrization X = X(s, t). A basis of the tangent
plane is Xs , Xt . Then we have to show
hdNp (Xs ), Xt i = hXs , dNp (Xt )i.
On the other hand, N is orthogonal to Xs and Xt . Then
hN (X(s, t)), Xs i = 0 hNt , Xs i + hN, Xst i = 0
hN (X(s, t)), Xt i = 0 hNs , Xt i + hN, Xts i = 0
and this concludes the proof.
Definition 1.5. The map Ap : dNp : Tp M Tp M is the Weingarten map and
p : Tp S Tp M R given by p (u, v) = hdNp (u), vi is the second fundamental
form. Moreover, the Weingarten map is diagonalizable.
Theorem 1.1. The eigenvalues of Ap are the principal curvatures.
Thus
H(p) =

1
trace (dNp ).
2

We rediscover the mean curvature

(1) If M is a plane, then N is constant, dNp = 0 and H = 0.
(2) If M is a sphere of radius r > 0, then N (p) = p/r. Then dNp (u) = u/r,
u Tp M and H = 1/r.
(3) If p is an umbilical point, then dNp = (p)Ip , where Ip is the identity
map on Tp M .
Theorem 1.2. In local coordinates X = X(u, v), the mean curvature H is given
by the formula
1 eG 2f G + gE
(1.2)
H=
,
2
EG F 2
where
E = hXu , Xu i, F = hXu , Xv i, G = hXv , Xv i.
det(Xuu , Xu , Xv )
det(Xuv , Xu , Xv )
det(Xvv , Xu , Xv )

e=
, f=
, g=
.
2
2
EG F
EG F
EG F 2
Proof. Consider X = X(u, v) a local parametrization of M with the Gauss map
N = (Xu Xv )/|Xu Xv |. If B = {Xu , Xv } is a basis of the tangent plane of M
at X(u, v), we denote the matrix of dNp with respect to B as


a b
dNp
:= A.
c d
On the other hand,
p (w1 , w2 ) = hdNp (w1 ), w2 i = hAw1 , w2 i.


RAFAEL LOPEZ

72

Denote the matricial expressions of h, i and as




E F
e
,
F G
f

f
g


,

respectively. For h, i is evident; for , we only do the proof for e:

hdNp Xu , Xu i = hN, xuu i =

hXu Xv , Xuu i
det(Xu , xv , Xuu )

= e.
=
|Xu Xv |
EG F 2

Then
w1T

e
f

f
g

w2 = (Aw1 )

E
F

F
G

w2 A =

e
f

f
g



E
F

Because trace(A) = trace(AT , we deduce (1.2).

F
G

1
.


We compute the mean curvature H for two spacial surfaces of Euclidean space.
(1) If the surface is the graph of a function z = f (x, y), we take X(x, y) =
(x, y, f (x, y)) as a parametrization. Then (1.2) gives
H(x, y, z)

=
=

1 (1 + fy2 )fxx 2fx fy fxy + (1 + fx2 )fyy

2
(1 + fx2 + fy2 )3/2
!
!
fx
1
fy
1
p
p
+
2
2
1 + |f |2
1 + |f |2
x

(1.3)

f
1
div p
.
2
1 + |f |2

(2) Consider a surface of revolution obtained by rotation with respect to the

z-axis the curve (r(s), 0, s), s I. With the parametrization X(s, ) =
(r(s) cos , r(s) sin , s), the expression of H is
(1.4)

H=

1 + r02 rr00
.
2r(1 + r02 )3/2

2. CMC surfaces
The simplest case of mean curvature function is that H is constant. Does have
any physical meaning the fact that the mean curvature is constant? Surfaces with
constant mean curvature are solutions of a variational problem. We consider a
compact surface M with possible non-empty boundary M . Let x : M R3 be a
isometric immersion. A variation of x is a differentiable map X : M (, ) R3
such that
(1) For each t, Xt : M R3 given by Xt (p) = X(p, t), is an immersion.
(2) X(p, 0) = x(p), that is, X0 = x.
(3) X(p, t) = x(p) for any t (, ) and p M . This means that the
variation preserves the boundary.
We define the area and volume functionals A, V : (, ) R, as
A(t) = area(Xt ),
or

Z
A(t) =

1 dMt ,
M

V (t) = volume(Xt ),
1
V (t) =
3

Z
hXt , Nt i dMt .
M

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

73

We focus in the variations that preserve the volume, that is, V (t) = V (0) for any
t. We ask for those immersions x, such that A0 (0) = 0.
Theorem 2.1. An immersion of a compact surface in Euclidean space has constant mean curvature (CMC) iff it is a critical point of the area functional for any
preserving volume variation.
General texts on CMC surfaces, with physical interpretation of these surfaces,
are [12, 14, 28]
We show Theorem 2.1 in the case that the surface is a graph. Consider M given
as a graph of a function f defined on D R2 . Consider a variation Mt as graphs
on D where
(1) g : D (, ) R, with Mt = g(M {t}) and g(x, y, 0) = f (x, y). This
means that at t = 0, we have the original graph M .
(2) g(x, y, t) = f (x, y) for any (x, y) D. With this condition, the variation
preserves the boundary of M .
The area of Mt is
Z
Z q
1 + gx2 + gy2 dxdy.
A(t) =
1 dMt =
M

We differentiate with respect to t, and let t = 0. We obtain

Z
fx gxt + fy gyt
p
A0 (0) =
(x, y, 0) dxdy.
1 + |f |2
D
p
Denote T (f ) = f / 1 + |f |2 . We integrate by parts:
!
! !
Z
gt fx
gt fy
0
p
A (0) =
+ p
(x, y, 0) dxdy
1 + |f |2 x
1 + |f |2 y
D
Z

gt divT (f )(x, y, 0) dxdy

D
Z
Z
=
gt hT (f ), ~nids
gt divT (f )(x, y, 0) dxdy.
D

In the last identity we have used the divergence theorem, where ~n is the outer unit
normal vector to D. The first integral vanishes because g(x, y, t) = f (x, y) for any
(x, y) D, and so, gt (x, y, 0) = 0 on D. Then
Z
Z
A0 (0) =
gt divT (f )(x, y, 0) dxdy = (2H)gr (x, y, 0)dxdy.
D

If H = 0, then A0 (0) = 0 for any g. Assume now that A0 (0) = 0 for any preserving
volume variation of M . We consider an appropriate variation g given by:
g(x, y, t) = f (x, y) + tHp(x, y)(x, y),
R
where p(x, y) > 0 on D and p = 0 on D. Then gt = Hp and so, M pH 2 dM = 0
and thus, H = 0. As conclusion we have proved that H = 0 on the surface if and
only if the surface is a critical point of the functional area.
If we Rassume that the variation preserves the volume, we have the constraint
V (t) = D g dxdy = constant. If t = 0 is a critical point of A(t), the theory of
Lagrange multipliers implies the existence of a constant R such that
A0 (0) + V 0 (0) = 0.


RAFAEL LOPEZ

74

Figure 3. A drop resting in a support surface.

Now
V 0 (0) =

Z
gt (x, y, 0) dxdy.
D

Then
A0 (0) =

Z
gt (2H + )(x, y, 0) dxdy.
D

If this happens for any g, for appropriate variations g, we have 2H + = 0, that

is, H is constant. On the other hand, if H is constant, and because V 0 (0) = 0, we
have
Z
Z
A0 (0) = (2H)gt (x, y, 0) dxdy = 2H
gt (x, y, 0) dxdy = 0
D

Definition 2.1. A minimal surface is a surface whose mean curvature vanishes on

the surface.
Physically, CMC surfaces correspond with the following physical setting. We
deposit an amount of liquid on a planar substrate, and assume that there are no
chemical and physical reactions between liquid, air and solid. Also, we delete the
gravitational forces. We denote by L, A and S the liquid, air and solid phases, and
by SIJ the interface between the I and J phases. In mechanical equilibrium, the
liquid drop attains its shape when the following equation holds:
PL (p) PA (p) = (2H)(p)

(Laplace)

for each p SLA . Here PL and PA are the pressures in the liquid and air. The
constant is the surface tension coefficient of the liquid and H is the mean curvature
of the interface SLA . The coefficient is determined by chemical and physical
properties of the liquid and it measures the intermolecular forces that exist in the
liquid which are necessary to move the molecules from inside to the SLA interface.
If the pressures in both sides of the interface are constant, then the interface is a
surface with constant mean curvature.
In our system, the only force acting on the interface is the surface tension. This
force is proportional to the area of this interface. Then the energy is proportional
to the area of SLA . We remark that the volume of the drop remains constant. If
we perturb the drop, the liquid tries to reduce its energy (proportional to the area
of SLA ) and when this occurs, this interface has constant mean curvature. Thus we
can say that the shapes of (small) liquid drops are modeled by CMC surfaces. We
suggest the amazing book of de Gennes [7], Nobel Prize in Physics, where shows

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

75

a number of physical settings where the mean curvature appears in equilibrium

shapes.
From the mathematical viewpoint, CMC surfaces can be introduced by the
isoperimetric problem: among all compact surfaces in Euclidean space with the
same volume, which is the one with smaller area? For minimal surfaces, the analogous problem is the so-called minimizing area: Characterize those surfaces which
have least area among all surfaces with the same boundary. In both cases, because
the area is a minimum and then, A0 (0) = 0.
We end this section showing the relation between the theory of minimal surfaces
and the Complex Analysis: see [29]. Consider X : D R3 an immersion, X =
X(u, v), where (u, v) are isothermal parameters, that is, E = G, F = 0. We
compute the Laplacian of X, X = Xuu + Xvv . We know that Xuu = aXu +
bXv + cN . Now
Eu
1
Ev
1
,
b = hXuu , Xv i =
Xv ,
c = e.
a = hXuu , Xu i =
E
2E
G
2G
Thus
Eu
Ev
Xuu =
Xu
Xv + eN.
2E
2G
Similarly,
Gu
Gv
Xvv =
Xu +
Xv + gN.
2G
2G
Using that E = G, we conclude
E
 E
Gu 
Gv 
u
v
X = Xuu + Xvv =

Xu +
+
Xv + 2HE 2 N = 2HE 2 N.
2E
2G
2G 2G
Thus we have proved
Theorem 2.2. Let X : D R3 an immersion in isothermal coordinates. Then
H = 0 if and only if X is a harmonic map.
This result says us the strong relation between minimal surfaces and harmonic
maps. We can continue as follows. Let write X(u, v) = (x1 (u, v), x2 (u, v), x3 (u, v))
the coordinates functions of a minimal surface and let z = u + iv. We define
: D C C, 1 i 3 as
1 = (x1 )u i(x1 )v ,

2 = (x2 )u i(x2 )v ,

3 = (x3 )u i(x3 )v .

The fact that the coordinates are isothermal writes as

3
X
2i = |Xu |2 |Xv |2 2ihXu , Xv i = 0.
i=1

By the armonicity of X, the functions i satisfy the Cauchy-Riemann relations,

that is,
(xi )u
(xi )v
=
(xi )uu + (xi )vv = 0.
u
v
(xi )u
(xi )v
=
(xi )uv = (xi )uv .
v
u
Thus i are holomorphic functions. Finally, because X is an immersion, we obtain
3
X
i=1

|1 |2 =

3
3
X
X
(xi )2u +
(xi )2v = |Xu |2 + |Xv |2 = 2E.
i=1

i=1

It is possible the converse in the following sense:


RAFAEL LOPEZ

76

Theorem 2.3. Let D C be a simply-connected

domain. Let
P
P i be holomorphic
functions on D, 1 i 3. Assume that i |i |2 6= 0 and i 2i = 0. Then the
map X : D R3 , X = (x1 , x2 , x3 ) given by
Z z
xi (z) = <
i (z)dz
z0

defines a minimal immersion in Euclidean space R3 . The integral is calculated

along any path on D joining a fix point z0 with z.
Because D is a simply-connected domain, the above integrals do not depend on
the chosen path joining z0 with z.
3. Some special CMC surfaces
In this section we present examples of surfaces with constant mean curvature
whose geometry is particular in some sense.
(1) Surfaces of revolution. Rotational surfaces in the Euclidean space with constant mean curvature are known as Delaunay surfaces. From the expression
of the mean curvature in (1.4), we have
rr00
r0
rr0 r00
1

= 2Hr

= 2Hrr0 .
02
3/2
(1 + r )
(1 + r02 )3/2
1 + r02
1 + r02
This writes as


d
r
Hr2
= 0,
ds
1 + r02
and a first integral is
r
Hr2
=c
1 + r02
for an integration constant c. The function r cannot be completely integrated because the ordinary differential equation involves elliptic integrals.
Delaunay discovered that the geometry of the generating curve (r(s), 0, s)
is focus of a conic that rolls on a straight-line, being this line the axis of
revolution: see Figure 4. See also [6].

(3.1)

Figure 4. The trace of a focus of a conic.

Besides planes and catenoids (H = 0), Delaunay surfaces are unduloids
and nodoids, and the limit cases, spheres and cylinders. See Figure 5.
Unduloids are embedded (Figure 6) and nodoids are non-embedded.
If H = 0, the equation (3.1) can completely integrated: we have

r0
1
r
s
r
=c
= arc cosh( ) = + , R.
02
2
2
c
c
c
1+r
r c

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

77

Figure 5. Profiles curves of Delaunay surfaces.

Then

s
r(s) = c cosh( + ).
c
The profile curve is a catenary and the corresponding surface is called a
catenoid. See Figure 6.
Theorem 3.1. Planes and catenoids are the only minimal rotational surfaces.

Figure 6. A catenoid and an unduloid.

(2) Ruled surfaces.
Theorem 3.2. The only ruled surfaces with constant mean curvature are
helicoids (H = 0) and right circular cylinders (H 6= 0).
A helicoid is the ruled surface generated by a helix and the rulings are
horizontal straight-line at each point of the helix. If we take the helix
(s) = (r cos(s), r sin(s), as), a > 0, a parametrization of the helicoid is
X(s, t) = (s) + t(1 (s), 2 (s), 0) = (r(1 + t) cos s, r(1 + t) sin s, as).
See Figure 7.
(3) Translation surfaces. A translation surface is a surface that is a graph of a
function z = f (x) + g(y).
Theorem 3.3. The only CMC surfaces that are translations surfaces are
planes, Scherks surfaces (H = 0) and right circular cylinders (H 6= 0).


RAFAEL LOPEZ

78

Proof. For H = 0, the result is classic and it is due to Scherk. When H is

a non-zero constant, the theorem is proved in [18]. Here we do the proof
for the minimal case. The mean curvature satisfies
(1 + g 02 )f 00 + (1 + f 02 )g 00 = 2H(1 + f 02 + g 02 )3/2 .
Then
f 00
g 00
(1 + f 02 + g 02 )3/2
+
=
2H
.
1 + f 02
1 + g 02
(1 + f 02 )(1 + g 02 )
If H = 0, then
g 00
f 00
=

.
1 + f 02
1 + g 02
Since one side is a function depends on x and the right one depends on y,
there exists a constant c R such that
g 00
f 00
=

= c.
1 + f 02
1 + g 02
A simple integration gives, up constants,




1
1
f (x) = log cos(cx) , g(y) = log cos(cx) ,
c
c
or
!
1
cos(cy)
z(x, y) = log
.
c
cos(cx)
This surface is called a Scherks surface and it appears in Figure 7.

Figure 7. A helicoid and a Scherk surface.

4. The constant mean curvature equation
Let M be a surface in R3 and we locally write as the graph of a smooth function
z = f (x, y). Consider N the orientation on M given by
(4.1)

(fx , fy , 1)
N (x, y, f (x, y)) = q
,
1 + fx2 + fy2

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

79

where the subscripts indicate the correspondent derivatives. We know from (1.3)
that the mean curvature H of S satisfies the following partial differential equation:
(4.2)

(1 + fy2 )fxx 2fx fy fxy + (1 + fx2 )fyy = 2H(1 + fx2 + fy2 ) 2 .

Theorem 4.1 (Comparison). Consider two tangent surfaces M1 and M2 at some

point p M1 M2 . We orient both surfaces so N1 (p) = N2 (p). If M1 M2 , then
H1 (p) H2 (p).
Proof. We consider that both surfaces are graphs of functions z = fi (x, y), whose
tangent planes are horizontal and that p is the origin. Then the normal vectors at
p agree if we take the usual parametrization of the surface as a graph of a function.
We know that Hi (p) = (fi xx + fi yy )(0, 0). If M1 M2 then f1 (x, y) f2 (x, y)
around the origin. In particular, the function f2 f1 attains a local minimum
at (0, 0) and thus, (f2 f1 )xx (0, 0) 0 and (f2 f1 )yy (0, 0) 0. This implies
H2 (p) H1 (p).

Corollary 4.1. Let M be a compact CMC graph on a domain D included in a
plane P . Assume that the boundary of M is the boundary curve D. If H 6= 0,
then M lies in one side of P . If H = 0, then M = D. Moreover, if the orientation
points downward, H > 0 (resp. H < 0), then M lies over P (resp. below P ).
Proof. Compare the surface with the tangent planes at the lowest and highest
points.

Corollary 4.2. Any minimal compact surface with boundary lies in the convex hull
of its boundary
Proof. Compare the surface with spheres with sufficiently big radius: see Figure
8.


Figure 8. Proof of Corollary 4.2.

Corollary 4.3. Let M1 and M2 be two graphs with constant mean curvature H1
and H2 with respect to the downward orientations. Assume that M1 = M2 . If
H1 < H2 , then M1 M2 .
Proof. On the contrary, that is, if M1 has points over M2 , we move down M1
until the last position. This occurs in some interior point, where both surfaces
are tangent. However the orientations agree at that point and M1 lies over M2 :
contradiction.



RAFAEL LOPEZ

80

The next two results say that CMC surfaces minimize area, at least locally.
Theorem 4.2. Let M be a minimal compact surface with boundary M . If M is
a graph, then M has least area among all graphs with the same boundary than M .
Proof. Let z = f (x, y) a minimal graph, where f is defined in some domain D R2 ,
and the boundary of the surface is the curve C = {(x, y, f (x, y)); (x, y) D}.
Let g other differentiable function on D such that f = g along D and denote
M 0 = graph(g). Consider the Gauss map N on M :
1
(f, 1).
N (x, y, f (x, y))) = p
1 + |f |2
From (1.3), we deduce divT (f ) = 0. Consider W the enclosed domain by M M 0
and define on W R3 the vector field
X(x, y, z) = N (x, y, f (x, y)).
See Figure 9. We compute the divergence (in R3 ) of X, that is,

Figure 9. The vector field X in the proof of Theorem 4.2.

DIV(X) =

fx
q

1 + fx2 + fy2

+
x

fy
q

1 + fx2 + fy2

1
q

1 + fx2 + fy2

!
= 0.
z

The divergence theorem assures that

Z
Z
Z
Z
0 =
DIV(X) =
hX, NW i =
hX, NM i +
hX, NM 0 i
W
M
M0
ZW
Z
Z
=
1+
hX, NM 0 i = area(M ) +
hX, NM 0 i.
M

M0

M0

Then
Z
area(M ) =

Z
hNM , NM 0 i

M0

1 = area(M 0 ).

M0

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

81

Theorem 4.3. Let M be a graph with constant mean curvature H. If M 0 is other

graph on the same domain, with the same boundary and volume, then area(M )
area(M 0 ).
Proof. Because the boundaries and volumes agree, then the volume of W is zero.
Then
Z
Z
Z
DIV(X) =
(2H) = 2H
1 = 0,
W

and the proof is as in the above theorem.

5. The Alexandrov theorem

Round spheres are closed CMC in Euclidean 3-space R3 . A pioneering work in
the theory of CMC closed surfaces is the following result due to Hopf.
Theorem 5.1 (Hopf). Spheres are the only CMC closed surfaces with genus 0 [11].
The proof uses Complex Analysis and the idea is as follows. Given X : M R3
an immersion, and using isothermal coordinates, we define : D C C as
(z) =

eg
if,
2

where {e, f, g} are the coefficients of the second fundamental form. Then
|| =

E2
|1 2 |.
2

This means that the zeroes of agree with the umbilical points. Moreover,
(e g)v
fu = E 2 Hv ,
2

(e g)u
+ fv = E 2 Hu .
2

Then
Theorem 5.2. An immersion X has constant mean curvature iff is holomorphic.
In particular, umbilical points are isolated, or the immersion is umbilical.
The proof of Hopf theorem consists to consider a complex structure on M . Since
M is a topological sphere, M is conformally equivalente to C with its usual analytic
structure. The function defines a holomorphic function on C which is bounded.
Thus = 0, that is, X is umbilical, that is, X(M ) is a round sphere.
In order to study the closed surfaces with constant mean curvature in Euclidean
space, Alexandrov proved in 1956 that any embedded closed CMC surface in R3
must be a round sphere. An embedded surface in R3 is a surface without selfintersections. For longtime, it was an open question if spheres were the only closed
CMC surfaces in R3 . If a such surface would exist, it would be a surface with
self-intersections and higher genus. In 1986, Wente succeeded by constructing an
explicit immersed torus with constant mean curvature [34]. This discovery activated
a great work in the search of new examples of closed CMC surfaces.
We show the Alexandrov result.
Theorem 5.3 (Alexandrov). The sphere is the only CMC closed surface that is
embedded [1].

82

RAFAEL LOPEZ

Figure 10. (left) Wente torus; (right) CMC surfaces with higher
genus and topology.
The proof of this theorem is geometric and it is based on the Maximum Principle of linear elliptic equations. Equation (4.2) may written as
!
f
(5.1)
Q(f ) := div p
= 2H,
1 + |f |2
where div y stand for the divergence and gradient operator respectively. In PDE
theory, this equation is an elliptic equation of divergence type. If two functions f1
and f2 satisfy (5.1), then
Q(f1 ) Q(f2 ) = L(f1 f2 ) = 0,
where L is a linear elliptic operator. For linear equations, the Maximum principle
asserts that the maximum of the function occurs at boundary points. A general
reference in elliptic equations is [8]. This geometrically translates for CMC surfaces
as follows:
Theorem 5.4 (Tangency principle). Let M1 and M2 two surfaces with the same
(constant) mean curvature. Assume that M1 and M2 are tangent at some point p
and both orientations agree at p. If one surface lies in one side of the other one,
then M1 = M2 agree in an open set around p.
See Figure 11. The proof of the Alexandrov theorem uses a method of reflection
that allows us to compare the surface with itself and next, to apply the Tangency
principle. One concludes that the surface has symmetries in each direction: see
Figure 12

Figure 11. The Tangency principle.

6. The effect of the boundary in the shape of a CMC surface
We now consider compact CMC surfaces with non-empty boundary. The simplest case of boundary is a round circle. If C is a circle of radius r > 0, we consider
C included in a sphere S(R) of radius R, R r. The mean curvature of S(R) is
H = 1/R with the inward orientation. Then C splits S(R) in two spherical caps

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

83

Figure 12. The Alexandrov reflection method.

with the same boundary C and constant mean curvature H. If R = r, both caps
are hemispheres, and if R > r, there are two different caps named the small and
the big spherical cap. On the other hand, the planar disc bounded by C is a compact surface with constant mean curvature H = 0. All these surfaces are the only
umbilical examples of compact CMC surfaces bounded by C. See Figure 13.

Figure 13. Planar disc and spherical caps.

It is natural to ask if planar discs and spherical caps are the only CMC compact
surfaces bounded by a circle. If we compare with the closed case and with the
Hopf and Alexandrov theorems, the natural hypothesis to consider is that M is
a topological disc or that M is embedded. However, as yet it is unknown if the
following two conjectures are true:
Conjeture 1. Planar discs and spherical caps are the only compact CMC
surfaces bounded by a circle that are topological discs.
Conjeture 2. Planar discs and spherical caps are the only compact CMC
surfaces bounded by a circle that are embedded.
However, we could do the following experiment. We take a drinking straw with
circular cross-section and introduce one of its ends into a container with a soap
solution. When we extract, it is formed a soap film that coincides with the flat disk
whose boundary is the circular rim of the end. Now and carefully, we pump air
into the straw from the other end and we see that the flat disk changes into a soap
bubble attached to the border. On the end where we are putting air, we place a
finger so that the air can not escape from the straw. The soap bubbles we almost
always observe are spherical caps. Once that the spherical cap is formed, we can
do small displacements of the straw without take off the finger and non destroying
the soap bubble in such way that the bubble follows attached to the straw. Then


RAFAEL LOPEZ

84

the bubble perturbs and changes of shape, but when the soap bubble attains a new
position of equilibrium, the surface formed is the original spherical cap again.
After these considerations, it is natural to ask: what is the shape of a mathematical soap bubble with circular boundary?
Surprisedly, and besides the spherical caps, there exist non-spherical mathematical soap bubbles and spanning circular boundaries. These surfaces were obtained in
1991 by Kapouleas founding other examples of CMC surfaces bounded by a circle
[13].
This means that our knowledge about the structure of the space of CMC surfaces
bounded by a circle is really small and only a few particular situations have been
considered. Moreover, the proofs of Hopf and Alexandrov theorems can not carry
to our context of non-empty boundary. This fact, together the lack of examples of
CMC surfaces bounded by a circular circle, says that although the problems in the
non-empty boundary case have the same flavor as in the closed one, the proofs are
more difficult.
The theorems of Hopf and Alexandrov, even the Maximum principle, can see as
results of uniqueness in the family of CMC surfaces. In the case of CMC compact
surfaces with boundary, the simplest result is for graphs, that is, given a value
of H and a boundary curve C, there exists a unique graph bounded by C with
constant mean curvature H. We begin this section obtaining an balancing formula
for CMC compact surfaces of R3 . First, we precise the definition of boundary of
an immersion. Let x : M R3 be an immersion from a compact surface and let
C R3 a closed curve. We say that C is the boundary of x if x|M C is a
diffeomorphism.
We have seen that if C is a circle of radius r, the possible values of mean curvatures H for spherical caps bounded by C lies in the range [1/r, 1/r]. Thus, the
boundary C imposes restrictions to the possible values of mean curvature. This
occurs for a general curved boundary. Consider M a compact CMC surface with
boundary M = C and let Y be a variational field in R3 . The first variation formula
of the area |A| of the surface M along Y gives
Z
Z
Y |A| =
2HhN, Y i dM
hY, i ds,
M

where N is an orientation of M , H is the mean curvature according to N , represents the inward unit vector along M and ds is the arc length element of M . Let
us fix a vector a R3 and consider Y the vector field of translations in the direction
of a. As Y generates isometries of R3 , the first variation of A is 0. Because H is
constant, we have
Z
Z
(6.1)
2H
hN, ai dM +
h, ai ds = 0.
M

The first integral changes into an integral on the boundary as follows. The divergence of the vector field Zp = (p a) N , p M , has div(Z) = 2hN, ai. Here
denotes the cross product of R3 . The divergence theorem, together with (6.1),
yields
Z
Z
(6.2)
h, ai ds + H
h 0 , ai ds = 0,
M

where is a parametrization of M such that 0 = N . This equality is known

as the flux formula or balancing formula. This identity reflects the fact that

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

85

the area (the potential) is invariant under the group of translations of Euclidean
space. On the other hand, the formula can be viewed as the physical equilibrium
between the forces of the surface tension of M that act along its boundary with
the exterior pressure forces that act on the bounded domain D by C. See Figure
14. If the boundary C lies in the plane P = {x R3 ; hx, ai = 0}, for |a| = 1, then

Figure 14. Scheme of the flux formula.

h 0 , ai is the support function of the boundary. From (6.2),
Z
(6.3)
2H A =
h, ai ds,
M

where A is the algebraic area of C. Given a closed curve C R3 that bounds a

domain D, and doing h, ai 1, the value H of the possible mean curvature of M
satisfies
length(C)
(6.4)
|H|
.
2 area(D)
In particular, if C is a circle of radius r > 0, a necessary condition for the existence
of a surface spanning C with constant mean curvature H is that |H| 1/r. The
inequality 6.4 was proved in [9] for parametric surfaces.
Remark 6.1. It follows from the divergence theorem and from (5.1) that if M is the
graph of z = f (x, y) then

 Z
Z

 

f

hp
2|H|area(D) =  2H dD = 
, ~ni ds
 D

1 + |f |2
D
Z
Z
|f |
p

ds <
1 ds = length(C),
1 + |f |2
D
D
where ~n is the unit normal vector to D in P . Then,
|H| <

length(C)
.
2 area(D)

Remark 6.2.
(1) Assume that M is a compact surface and x : M R3 an
immersion with non-necessarily with constant mean curvature. In the proof
of balancing formula we have seen
Z
Z
2
hN, ai dM =
h 0 , ai ds.
M

If M 0 is compact surface in R3 with M = M 0 , then

Z
Z
hN, ai dM =
hNM 0 , ai dM 0 .
M

M0


RAFAEL LOPEZ

86

Thus
Z
(6.5)

hNM 0 , ai dM +

2H
M0

h, aids = 0,
M

where is the inner unit conormal along M .

(2) There is other way to obtain the balancing formula. Let x : M R3
an immersion from a compact surface M . Given a R3 , we define the
differential 1-form
p (v) = h(Hx(p) + N ) v, ai.
Then
(d)p (u, v) = hH p , uix v hH p , vix u,
and if e1 , e2 Tp M is a positive orthonormal frame on the surface, then
(d)p (e1 , e2 ) = x [N (dx)p (H p )].
If HR is constant,
is closed and one follows from Stokes formula that
R
0 = M d = M w. Then we obtain
Z
Z
Z
0
0
(H + N ) = 0 H
h , aids +
h, aids = 0.
M

Theorem 6.1. The only CMC compact surfaces bounded by a circle making a
constant angle with the plane containing the boundary are planar discs and spherical
caps [21].
Proof. After a homothety, we assume the plane P is given by {(x, y, 0) R3 } and
C is a circle of radius 1 with center at the origin. Let M be a compact surface with
constant mean curvature H and with boundary C. If H = 0, M is a minimal surface
and the Maximum Principle concludes the surface is the planar disc bounded by
C. Next suppose that H 6= 0. Since the surface makes constant angle with P along
, the function h, ai is constant, where is the inner conormal along C and a
denotes the vector (0, 0, 1). We choose an orientation on C such that {0 , , N }
and {, 0 , a} are positively oriented orthonormal basis, where x is the immersion
of M in R3 , and 0 is a unit tangent field along C. The boundary M is a line of
curvature, because hN, ai is constant along M . Then
0 = hN 0 , ai (0 , 0 )h0 , ai (0 , )h, ai = (0 , )h, ai.
Therefore (0 , ) = 0 along M . Because the boundary is a circle and {, 0 , a}
is a positive oriented basis of R3 ,
hN, i = h0 , i = h 0 , i = h, ai.
Since 1 = h, i = hN, i2 + h, ai2 , and hN, ai is constant, the function h, ai is
constant too. From (6.2),
Z
Z
h, ai
1+H
1 = 0.
M

This implies that h, ai = H along M . We calculate the normal curvature along

M .
(0 , 0 ) = hN 0 , 0 i = hN, i = H.
Thus the boundary points are umbilical. Since the surface has constant mean
curvature and the boundary is a set of non-isolated umbilical points, then Theorem
5.2 says that the surface is totally umbilical. Hence, M is a spherical cap.


SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

87

Corollary 6.1. Let C be a circle of radius r. Then the only compact surfaces with
constant mean curvature |H| = 1/r bounded by C are the halfsphere of radius r [4].
Proof. The balancing formula (6.2) gives
Z
Z
h, ai = H
h 0 , ai = 2Hr2 = 2r.
M

Then

Z

2r = 

 Z

h, ai

1 = 2r.

Thus |h, ai| 1 along M and so the surface makes constant angle with this plane
along the boundary and the above theorem ends the proof.

7. Some uniqueness results on CMC surfaces
In this section we will obtain a type of control of the shape of a CMC surface
with boundary using the Tangency Principle and the flux formula. We begin with
CMC surfaces included in right-cylinders. The setting is the following. Let R2
be a planar domain included in a plane P . We ask for those CMC surfaces with
boundary and included in R. For example, if the surface is a graph G, we
know that G lies in R. Moreover, it lies in one side of P . We have the following
Theorem 7.1. Let M be an embedded CMC surface bounded by . If M R,
then M is a graph on [26].
Proof. The proof uses the Alexandrov reflection method with horizontal planes.
Let Pt = {x R3 ; hx, ai = t}, with P0 = P and a = (0, 0, 1). Let m > 0 sufficient
big so that the reflection of M with respect to P lies above the plane Pm . Consider
W the interior domain bounded by the closed, non-smooth, surface
M ( [m, 0]) ( {m}).
We begin with planes for t >> 0. After the first intersection point, we go reflecting
the part of M above Pt with respect to this plane until the first time that the
reflection touches the part of M below Pt . If this point is a tangent point, Maximum
Principle assures us that the plane is a plane of symmetry. Since M has boundary,
this occurs only when t = 0, and then M is closed surface: contradiction. Therefore
the Alexandrov method shows that we can arrive until t = 0 and not tangent points
exist, that is, M is a graph on .

Consider now CMC surfaces not necessarily embedded.
Theorem 7.2. Let be a bounded domain in a plane P and let G be a graph on
with constant mean curvature H and bounded by = . Then, up reflections,
G is the only compact surface immersed in Euclidean space with constant mean
curvature H bounded by and included in the cylinder R [26].
Proof. Let M be a compact CMC surface with the same boundary than G and such
that M R. Without loss of generality, assume that H is positive and that
G lies above P . Thus the orientation NG on G points downwards. Denote by NM
the orientation on M . First we prove that G lies above on M unless that M = G.
On the contrary case, if M has points above G, we move up G until it does not
intersect M . This is possible by the compactness of both surfaces. Next we move
down G until the first contact point p with M . Then p is not a boundary point


RAFAEL LOPEZ

88

and the Maximum Principle gets a contradiction because the mean curvature of G
is positive.
Thus we move down G until its original position. If at this moment there exists
a tangent (interior or boundary) point, the Tangency Principle says that M = G.
On the contrary, M lies strictly below G and the slope of M along is strictly
less than the one of G.
We do a similar reasoning with the the reflection G of G with respect to P . As
consequence, either M = G of M = G (and the proof finishes) or M lies in the
domain determined by GG . See Figure 15. Let G , M be the inner conormals of
G and M respectively. Balancing formula (6.2) gives in each one of both surfaces:

Z
Z



hM , ai .
(7.1)
2H area() =
hG , ai,
2H area() = 

But along we have the strict inequality |hM , ai| < hG , ai. An integration along
the boundary gives a contradiction with (7.1).


Figure 15. Scheme of the proof of Theorem .

It is known that given a domain , there exists H0 depending only on such
that there exist graphs on with constant mean curvature H and boundary
whenever H satisfies |H| < H0 . As conclusion, we obtain
Corollary 7.1. Given a planar bounded domain , there exists H0 > 0 such that
if |H| H0 , there is a unique compact surface with constant mean curvature H
included in R and with boundary . Moreover this surface is a graph on .
We are going to obtain other consequence of Theorem 7.2. First, we recall the
Plateau problem. Given a fixed Jordan curve C R3 , one asks for the existence of
a CMC surface spanning C. For this, one restricts to consider immersions X : D
R3 from the closed unit disc D R2 such that X : D C is a diffeomorphism.
Under certain conditions, and using techniques from the Functional Analysis, one
can obtain a solution of the problem [32, 33]. The classical result is the following
due to Hildebrandt [10]:
Let C be a Jordan curve included in a ball BR of radius R > 0. If
|H| R1 , there exists an immersion X : D R3 spanning C and
with mean curvature H. Moreover, X(D) BR .
We mention that the Hildebrandt result is the best possible as one can see if C
is a circle. As consequence of this result together Theorem 7.2, we have

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

89

Theorem 7.3. Let C be a convex curve and denote de bounded domain bounded
by C. Then there exists H1 > 0 depending only on C such that if |H| < H1 , any
Hildebrandt solution bounded by C and with mean curvature H is a graph.
Proof. Consider H0 the positive number such that for any H, |H| < H0 , there
exists a graph on spanning C. On the other hand, because C is a convex curve,
let R0 > 0 such that if R > R0 , any circle of radius R containing C in its interior,
can roll in such way that the circle touches every point of C. This follows if
R0 = 1/ min , with the curvature of C. Let H1 = min{H0 , R10 }.
Let R1 = 1/H1 . Then R1 R0 and so, C is included in the ball BR1 . Consider
|H| < H1 and R = 1/|H|. Then C is included in BR because BR1 BR . Let M be
the Hildebrandt solution corresponding to (H, BR ). Since the mean curvature of
the sphere BR is |H|, we can move BR in any direction being not possible to have
a tangent point with M , unless that the point lies in the boundary of the surface.
Moreover, the radius of BR is R > 1/H1 R0 . Thus we can place BR such that C
lies in the domain determined by an equator of BR . Then it is possible to roll BR
such that touches each point of C. This shows that M lies in R. Theorem 7.2
proves that M must be a graph.

We end this section with the next result ([3]):
Theorem 7.4. Let M be a compact surface with constant mean curvature H and
bounded by a circle C. If M lies in a ball of radius 1/|H|, then M is a spherical
cap.
Proof. By basic geometry, the radius of C is r, with r 1/|H|. The mean curvature
of the ball B of radius 1/|H| is |H| with the unit normal pointing inside. The
Tangency Principle shows that if we move B in any direction, it can not exist a
tangent point with the surface, unless that M is a spherical cap.
On the contrary, M lies between two small spherical caps (and included en B)
with the same mean curvature as M . Now we apply Theorem 7.2 since there are
graphs (spherical caps): contradiction.

8. Embedded CMC surfaces
We consider embedded surfaces with constant mean curvature and with nonempty boundary C. Assume that C is a planar curve contained in a plane P . Now
we have
Theorem 8.1. Let M be a CMC embedded surface bounded by a circle. If M lies
in one side of P , then M is a spherical cap.
Proof. We use the same technique of Alexandrov as in Theorem 5.3. We attach to
the surface the disc bounded by C, obtaining a domain W . We have to show that M
is rotationally symmetric with respect to the line L orthogonal to P containing the
center of C. Now we take v a orthogonal direction to L. See Figure 16. We consider
the uniparametric family of planes orthogonal to v. Coming from t = , we arrive
to M and we begin with the reflection process until the first time of contact. If
this time is before to arrive to L, then the contact point must be interior, because
the surface lies over P . This is impossible by the Maximum Principle. Thus, we
arrive until the position that the plane contains L. If there is a contact point, then
the plane is a plane of symmetry. On the contrary, we begin from t = . Then

90

RAFAEL LOPEZ

Figure 16. The Alexandrov reflection method in Theorem 8.1.

it must be a contact (interior) point, which it is impossible. This shows that the
process finishes when the plane contains L and this plane is a plane of symmetry.
Because v is an arbitrary vector, we conclude that the surface is rotational.

Thus we ask for those hypothesis that assure that an embedded CMC surface
lies in one side of the boundary plane.
Corollary 8.1. Let C be a closed curve contained in a plane P . Assume that C
is symmetric with respect to a straight-line L P and that each piece of C that L
divides is a graph on L. If M is an embedded CMC surface with boundary C and M
lies in one side of P , then M is symmetric with respect to the plane orthogonal
to P with P = L.
One can extend Theorem 8.1 by considering that the boundary is formed by two
coaxial circles.
Theorem 8.2. Consider C1 C2 two coaxial circles in parallel planes Pi . Assume
that M is an embedded CMC surface bounded by C1 C2 . If M lies in the slab
determined by P1 and P2 , then M is a surface of revolution.
On the other hand, there exist pieces of nodoids spanning two coaxial circles,
but the surface is not included in the slab determined by the boundary. See Figure
17

Figure 17. Pieces of nodoids not included in the slab determined

by the boundary.
We present two results that inform us how the geometry of the boundary has
an effect on the whole surface. We give a quick proof of them without going into
details.

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

91

Theorem 8.3. Let C be a closed curve included in a plane P and denote D the
domain bounded by C. Assume that M is a CMC embedded surface bounded by C.
If M ext(D) = , then M lies in one side of P [15].

Figure 18. The contradiction in Theorem 8.3

Proof. We consider a sufficiently big hemisphere Q such that M together Q and
an annulus of P defines a closed embedded surface. Then M , Q and the annulus
determine an interior domain, called W R3 . If M has points in both sides of P ,
we will arrive to a contradiction: see Figure 18. We orient M so that the Gauss
map points towards W . In the highest point of M , the normal vector points down.
The Maximum Principle asserts that H > 0. On the other hand, in the lowest
point, the normal points down again (towards W ). But the Maximum Principle
comparing M with a horizontal plane at p yields a contradiction.

Theorem 8.4. Let C be a convex curve included in a plane P . Let M be a CMC
embedded surface spanning C. If M is transverse to P along C, then M lies in one
side of P [5].
Proof. We only present two cases that illustrate the proof of Theorem 8.4. In Figure
19 (left) the surface meets the plane P in a nullhomotopic closed curve G in P \ D.
We use the Alexandrov reflection method by vertical planes arriving from infinity.
Since C is convex, we would have an interior contact point, proving that M has a
symmetry by a vertical plane that does not intersect C, which it is impossible.
The second case that we analyze appears in Fig. 19 (right). Again, this surface is
impossible. In this situation, we use the balancing formula as follows. The surface
M together with D encloses a domain W and we orient M by N pointing towards
W . Here the mean curvature H is positive. Set a = (0, 0, 1). In the expression (6.2),
h, ai is positive, since the surface is transverse to P along M . Since 0 = N ,
the term h 0 , ai is also positive, which it is a contradiction in (6.2).

It is possible to add more information in Theorem 8.4.
Theorem 8.5. Let C be a Jordan curve included in a plane P and let D be the
domain that bounds in P . Let M be an embedded CMC surface spanning C. If M
is a graph over D around the boundary C, then the surface is a graph [20].
Proof. We use the Alexandrov reflection method but with horizontal planes. See
Figure 20. If the surface is not a graph on D, there would be a horizontal plane Q


RAFAEL LOPEZ

92

Figure 19. CMC surfaces that do no exist by Theorem 8.4

such that the reflection of the part of M over Q would contact with the below part.
Moreover, since M is a graph around C, this contact point must be an interior
point. Now the Maximum Principle implies that Q is a plane of symmetry of the
surface. Since M lies below Q, we get a contradiction.

Figure 20. CMC surface that does no exist by Theorem 8.5


9. CMC surfaces with circular boundary
Let M be a topological disc and let x : M R3 an immersion with constant
mean curvature and bounded by a circle C (for instance of radius 1). A result due
to Barbosa and do Carmo assures that the area of A satisfies A A or A A+ ,
where A and A+ denote the area of the small and big spherical cap with the same
mean curvature and boundary than M , respectively. An explicit computation of
A and A+ gives
p
p
2
2
A = 2 (1 1 H 2 ),
A+ = 2 (1 + 1 H 2 ).
H
H
Moreover the equality holds if and only if M is a spherical cap.
Theorem 9.1. If the A A , then M is a (small) spherical cap [25].
Proof. Using the Gauss-Bonnet theorem and that M is a topological disc, we have
Z
Z
Z
Z
Z
2
2
(9.1)
2 =
K dM +
g
H dM +
g = AH +
g ,
M

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

93

where g is the geodesic curvature along C. We have used that K H 2 , and

equality holds iff M is umbilical, that is, M is a spherical cap. We write the flux
formula (6.2) with a = (0, 0, 1):
Z
(9.2)
2H =
h, ai.
C

This equality together the Cauchy-Schwarz inequality imply

!2
Z
Z
1
(9.3)
h, ai ds
= 2H 2 .
h, ai2 ds
2
C
C
We compute the geodesic curvature. Since C is a circle of radius 1, 2n + 2g = 1,
where n is the normal curvature. In our case,
n = hN 0 , 0 i = hN, 00 i = hN, i.
Thus
p
|g | = 1 hN, i2 = |hN, ai|.
Using the hypothesis on the area, and the value of g , we have from (9.1)
Z
p
|hN, ai|.
2 2(1 1 H 2 ) +
C

Squaring,
4 2 (1 H 2 )

Z

hN, ai

Z
=
R
C

Z
2 hN, ai2
!C
2

1 h, ai

2
C

This implies
umbilical.

2

= 4 2

h, ai2 .

h, ai2 2H 2 and together (9.3) we obtain that the surface is



We report here recent results that characterize spherical caps in the family of
CMC surfaces with circular boundary.
Theorem 9.2. Let C be a circle of radius R and let M be a compact surface
spanning C and with constant mean curvature H. Then M is a spherical cap if one
of the following conditions holds:
(1) The surface is embedded and lies over the plane containing C.
(2) The surface is embedded and does not intersect the exterior domain of the
circle in the boundary plane.
(3) The mean curvature satisfies |H| = 1/R.
(4) The surface is included in a closed ball of radius 1/|H|.
(5) The surface is embedded and transverse to the boundary plane along the
boundary.
(6) The surface is a minimizer surface [16].
(7) The surface is stable with free boundary supported in a plane [17].
(8) The surface is a topological disc with area less than the area of the small
spherical cap with mean curvature H and boundary C.
(9) The volume is less than the volume of a hemisphere with the same mean
curvature [26].
(10) The surface is embedded and included in a slab of width 1/|H| [19].


RAFAEL LOPEZ

94

(11) The surface is a stable topological disc [2].

(12) The surface is a topological disc that makes a constant contact angle along
the boundary.
10. Two equations for CMC surfaces
3

Let x : M R be an immersion of an orientable surface in Euclidean space

R3 , let X (M ) be the set of vector fields on M and by N the Gauss map of the
immersion. Recall the fundamental equations for x:
(10.1)

0X Y = X Y + (X, Y )N, X, Y X(M ).

(10.2)

0X N = AN X = AX, (X, Y ) = hAX, Y i.

Here 0 is the usual connection on R3 , : Tp M Tp M R is the second fundamental form, A : Tp M Tp M , A = (dN ), is the Weingarten map.
Given a (smooth) function f on M , we define the gradient of f , f , as the vector
field given by
h(f )p , vi = (df )p (v),
v Tp M.
P2
If {e1 , e2 } is an orthonormal basis on Tp M , then (f )p = i=1 (df )p (ei )ei .
If X is a vector field on M , the divergence of X is the function div(X) defined
as


div(X)(p) = trace v 7 v X ,
where is the Levi-Civitta connection on M . Finally, the Laplacian of f is defined
by
f = div(f ).
If we fix M , there exists an orthonormal basis on Tp M , {e1 , e2 } such that
(ei ej )p = 0. Then (ei ej )p is orthogonal to M at p and then
X
X
X
f (p) =
hei , ei f i =
ei hei , f i
h0ei ei , f i
X
X
(10.3)
=
ei hei , f i =
ei (ei (f )).
In this section, we compute the Laplacian of two functions defined in a CMC
surface and we will obtain geometric consequences. For this, we use the Maximum
Principle to the Laplacian operator: if M is a compact surface with M 6= , and
if f 0, then f maxM f .
The next result holds for any immersion independently if H is constant.
Theorem 10.1. Let x : M R3 an immersion of an orientable surface M . If
a R3 , then
(10.4)

hx, ai = 2HhN, ai.

Proof. Fix p M and take {e1 , e2 } Tp M such that (ei ej )p = 0. Using (10.1)
and (10.3)
X
X
hx, ai =
ei ei hx, ai =
ei hei , ai
i

X
X
=
h0ei ei , ai =
(ei , ei )hN, ai = 2HhN, ai.
i

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

95

Theorem 10.2. Let x : M R3 an immersion with constant mean curvature.

Then
hN, ai + ||2 hN, ai = 0,

(10.5)

||2 = 4H 2 2K.

Proof. Using (10.1) and (10.2) and because 2H = trace(A), we know that
X
X
2H =
hAei , ei i =
(ei , ei ) = h0ei N, ei i
is constant. Thus for each j {1, 2},
X
X
X
0 = ej h
0ei N, ei i =
h0ej 0ei N, ei i +
h0ei N, 0ej ei i
i

(10.6)

X
X
h0ei 0ej N, ei i =
ei h0ej N, ei i.
i

Now we compute the Laplacian of hN, ai.

 X 

X
X 

hN, ai =
ei h0ei N, ai) =
ei h0ei N, ej iha, ej i =
ei h0ej N, ei iha, ej i
i,j

i,j



XX 
X
X
ei h0ej N, ei i ha, ej i +
h0ej N, ei i
ei ha, ej i

X

X

h0ej N, ei i
h0ej N, ei i


X
X
ha, 0ei ej i =
(ei , ej )2 hN, ai


ei ha, ej i

i,j

|| hN, ai.

In (*) we have used that Av = 0v N is a self-adjoint endomorphism: h0ei N, ej i =

h0ej N, ei i; in (**) we use (10.6).

The next result appears in [27] and generalizes one given in [30]. More generalizations, see [24].
Theorem 10.3. Let M be a graph with constant mean curvature H. Assume
that the boundary of M lies in a plane P . Then the height of M with respect to
P is less than 1/|H|. In the general case that the boundary is not planar, then
|hx, ai| maxM |hx, ai| + 1/|H|.
Proof. Because M is a graph, Corollary 4.1 says that M lies in one side of P .
Assume that M lies over P . We choose the orientation such that hN, ai 0, with
a = (0, 0, 1). This means that H > 0 (by Corollary 4.1 again). Using (10.4)-(10.5)
and the fact that ||2 = 4H 2 2K 2H 2 , we obtain
(Hhx, ai + hN, ai) = (2H 2 ||2 )hN, ai 0.
We are in position to apply the Maximum Principle. Since hx, ai = 0 on M , we
have
Hhx, ai + hN, ai

max(Hhx, ai + hN, ai)

max(hN, ai 0

Thus
hx, ai

1
hN, ai
.
H
H


RAFAEL LOPEZ

96


Corollary 10.1. Let M be a CMC embedded surface with boundary M included
in a plane P . Then the height of M with respect to P is 2/|H|.
Proof. We use the Alexandrov method with respect to horizontal planes. Because
there is not a contact interior point, in the reflection process we descend until a
height ho with ho h/2, being h the height of M . Moreover, the part of the
surface over the height ho is a graph and its height h ho is less than 1/|H|. Thus
h/2 < 1/|H|.

Using the same idea as in Theorem 9.1, we have the next result on uniqueness
of graphs without the use of the Maximum Principle.
Theorem 10.4. Let M be a compact CMC surface bounded by a round circle C.
If M is a graph, then M is a planar disc or a spherical cap.
Proof. We assume that the radius of the circle is 1 and we use the same notation
as in Theorem 9.1. We apply the divergence theorem in Equation (10.5) and we
get
Z
Z
2
(10.7)
|| hN, ai dM =
hdN , ai ds.
M

We study each side of (10.7) and we begin with the left-hand side. We know that
Z
Z
Z
2H
hN, ai = h, ai = 2H
hN, ai = .
M

Thus hN, ai is positive, since M is a graph. As K H 2 , then KhN, ai H 2 hN, ai.

Since ||2 = 4H 2 2K, we have
Z
Z
Z
||2 hN, ai dM = 4H 2
hN, ai dM 2
KhN, ai dM
M
M
M
Z
(10.8)
2H 2
hN, ai dM = 2H 2 .
M

Let the right-hand side of (10.7). First,

dN = (0 , )0 (, ).
Because hN, i = h, ai and the radius of C is 1 (00 = ), we have
(, )
(10.9)

2H (0 , 0 ) = 2H + hdN 0 , 0 i

2H hN, 00 i = 2H + hN, i = 2H + h, ai.

Because h0 , ai = 0, and using (9.2) and (10.9), we have

Z
Z
Z 

hdN , ai ds =
(, )h, ai ds =
2H + h, ai h, ai ds
C
C
C
Z
(10.10)
= 4H 2 h, ai2 ds.
C

Then (9.3) and (10.10) imply

Z
(10.11)

hdN , ai ds 2H 2 .

Finally, from (10.8) and (10.11), we obtain equalities in (10.7) and so ||2 = 2H 2
on M . Then M is an umbilical surface.


SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

97

11. The Dirichlet problem of the CMC equation

We look for CMC graphs with constant mean curvature. If z = u(x, y) is a graph
M on R2 with constant mean curvature H, then u satisfies
u
(11.1)
,
div (T u) = 2H
in ,
Tu = p
1 + |u|2
(11.2)
u =
along .
p
The orientation N assumed on the graph points upwards, that is, N = (f, 1)/ 1 + |f |2 .
If = 0, then the boundary of the surface is the curve .
The general technique employed in the solvability of the Dirichlet problem is the
method of continuity. We briefly explain this technique with some details (see [8]).
Consider H a fixed real number. For each t [0, 1], we pose the family of Dirichlet
problems

div (T ut ) = 2tH in
(Pt ) :
u =
along .
The solutions of (Pt ) are graphs on with constant mean curvature tH and all
these graphs have the same boundary. Define the set
J = {t [0, 1]; there exists a solution ut of (Pt )}.
In this setting, a solution of (11.1)-(11.2) exists provided one shows that 1 J.
For this purpose, we shall prove that J is a non-empty, open and closed subset of
[0, 1], and hence, J = [0, 1]. Let us prove first that 0 J, that is, there exists a
minimal surface with the same boundary value on . In general, this solution
is obtained from the theory of minimal surfaces and, at first, this difficulty is not
easily overcome. In our case if = 0, then u = 0 is an immediate solution. In
a second step, we prove that J is open in [0, 1]. Consider J and we will see
that the Dirichlet problem (Pt ) can be solved for each t in a certain interval around
. Denote by t the graph corresponding to ut and define a map h : C02, ( )
C0 ( ) taking each v onto the mean curvature function of the normal graph on
corresponding to the function v:
h(v) = mean curvature of (p 7 x(p) + v(p)N (p)),
where x : R3 is the inclusion map. The linearisation of h is the Jacobi
operator of , namely,
L(v) = v + ||2 v,
where is the Laplace-Beltrami operator in and is its second fundamental
form. Here L is a self-adjoint linear elliptic operator with trivial kernel, since
(11.3)

LhN, ai = 0

and

hN, ai < 0,

where N is the orientation on t and a = (0, 0, 1). Hence, and using the implicit
function theorem for Banach spaces, h is locally invertible. This shows that there
exists a solution ut of (Pt ) for values of t around .
Finally, it remains to be proved that J is closed in [0, 1]. The Schauder theory
reduces the question to establish a priori C 0 and C 1 estimates of each solution ut
of (Pt ) independent of t, that is, it suffices to prove that there exists a constant M
independent of t such that
sup |ut | M, sup |ut | M.


RAFAEL LOPEZ

98

The value of |ut |, that is, the height of t is controlled by a universal constant.
Exactly, the height of t is less than the one of because tH < H: see Corollary
4.3. Finally, the height of is given by Theorem 10.3. Then (putting = 0)
u0 < ut < u1 1/|H|.
Now we seek a priori estimates for |ut |. By the expression of N in terms of
ut , we know
1
.
(11.4)
hN, ai = p
1 + |ut |2
But Equation (11.3) tells us that hN, ai 0 and so, the minimum of hN, ai is
attained at a boundary point of . By combining with (11.4), we conclude
sup |ut | = sup |ut |.

At this moment, and for each particular case of domain , we shall need suitable
surfaces as barriers to compare the slope of the graph of ut along its boundary.
Theorem 11.1. The Dirichlet problem (11.1)-(11.2) for = 0 has a solution if
one can establish C 1 -a priori estimates of a solution along the boundary .
When is a bounded convex domain, the classical result of the existence for
Equation (11.1) is due to Serrin [31]:
If the curvature of with respect to the inner orientation satisfies 0 < 2|H| < , then for an arbitrary smooth function on ,
there exists a unique solution of (11.1) with u = along .
However, if = 0 on , one expects that the range of possible H is bigger, as it
happens when is a round disc: in this case, if the radius is 1, = 1, and there
exist graphs for |H| < 1.
Theorem 11.2. Let be a bounded convex domain. If one of the following assumptions holds, then there is a solution of (11.1) for u = 0 along :
(1) 0 < |H| < , where is the curvature of [21].
(2) is included in a strip of width 1/|H| [22].
In this theorem, pieces of spheres and cylinders are used as barriers since they
fit well with the convexity of the domain .
We explicit the proof for the case (2) in Theorem 11.2. First, we claim: there
exists a number > 0, = (, H), such that if M is a graph on with M = ,
then the height h of M satisfies
1
(11.5)
h<
.
2|H|
Proof of the claim. Because is included in a strip of width 1/|H|, we can put
between two parallel lines L01 L02 such that dist(L01 , L02 ) < 1/|H|. Let M be a
graph on with constant mean curvature H > 0 with the downward orientation.
This means that M P + . Let SH be a half-cylinder of radius 1/(2|H|) such that
SH P + , SH P . Let remark that SH are two parallel lines L1 L2 whose
distance is 1/|H|. Let DH be strip bounded by SH . We place SH such that
L01 L02 DH (and then the four straight-lines are parallel) and dist(L1 , L01 ) =
dist(L2 , L02 ). We prove that it is possible to descend SH until to arrive the position
Li = L0i and the whole surface M lies below SH . For this, we move up SH until
it does not touch M . See Figure 21. Next, we move down. Because M and SH

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

99

have the same (constant) mean curvature with the downward orientation, it is not
possible to have a contact point until that SH arrives its original position. For the
same reason, we can move down SH until that SH P = L01 L02 . Thus M lies
below SH and so, the height h of M is less than the one of SH in the last position.
Now the height of SH is 1/(2|H|) , where depends only on L01 L02 , that is,
on , and the cylinder SH . We point out that we have not used that is a convex
domain.

Figure 21. Proof of Theorem 11.2: comparison between M and SH .

Once proved the estimate (11.5), we show the existence of a priori C 1 -estimates
along the boundary. We consider QH quarters of cylinders of radius 1/(2|H|) whose
boundary is formed by two parallel lines to the axis A of the cylinder, QH =
R1 R2 . We place QH such that QH P + , R1 A P . Thus the height of
QH is exactly 1/(2H). Consider a parallel direction v to P and we displace QH so
does not intersect M , A is orthogonal to v and M lies in the convex side of QH .
We descend QH until that the height of QH is 1/(2H) 2 , where is the number
given in (11.5). We call QH again the portion of QH in P + . We displace QH to
M orthogonally to v until that QH touch M : see Figure 22. Because both surfaces
have the same constant mean curvature (with the same downward orientation), it
is not possible a contact point between M and QH . Thus we can place QH until
that R1 touches and some point p . At this point of M , the slope of M is
less than the one of QH . But the slope of QH is constant and independently on M .
Finally, doing this with any parallel direction to P together the fact that is
convex, we can place QH at each point of M . This shows the existence of a priori
estimate of the slope of M along its boundary, proving the result.
Acknowledgements. This work contains a series of talks given by the author
in the Department of Mathematics of the Afyon Kocatepe University, Turkey, on
February 2010. This visit was supported by the the Scientific and Technological

Research Council of Turkey (TUBITAK).

I wish to express my gratitude and thanks
to all my colleagues in Turkey.
Final remark. The purpose of this paper has been to present some problems on
CMC surfaces related with the authors work. For this reason, this manuscript is
not a survey of the theory of CMC surfaces and this has not been the motivation of
this work. For example, the number of references in the text is small and we only
cite some recent papers related with enounced results. For the interested reader,
we refer to [14, 23]. For minimal surfaces, the classical text [29] is a good guide

100

RAFAEL LOPEZ

Figure 22. Proof of Theorem 11.2: comparison between M and QH .

References
[1] Alexandrov, A.D.: Uniqueness theorems for surfaces in the large, V. Vestnik Leningrad Univ.,
13, No. 19, A.M.S. (Series 2), 21 (1958), 412416.
[2] Alas, L.J., L
opez, R., Palmer, B.: Stable constant mean curvature surfaces with circular
boundary, Proc. A.M.S. 127 (1999), 11951200.
[3] Barbosa, J. L.: Constant mean curvature surfaces bounded by a planar curve, Matematica
Contemporanea, 1 (1991), 315.
[4] Brito, F., Earp, R.: Geometric configurations of constant mean curvature surfaces with
planar boundary, An. Acad. Bras. Ci. 63 (1991), 519.
[5] Brito, F., Earp, R., Meeks III, W. H., Rosenberg, H.: Structure theorems for constant mean
curvature surfaces bounded by a planar curve, Indiana Univ. Math. J. 40 (1991), 333343.
[6] Eells, J.: The surfaces of delaunay, Math. Intelligencer, 9 (1987), 5357.
[7] de Gennes P. G., Brochard-Wyart F., Qu
er
e D.: Capillarity and wetting phenomena: drops,
bubbles, pearls, waves, Springer Verlag, New York, 2004.
[8] Gilbarg, D., Trudinger, N. S.: Elliptic Partial Differential Equations of Second Order, 2nd
edition, Springer Verlag, Berlin, 1983.
[9] Heinz, H.: On the nonexistence of a surface of constant mean curvature with finite area and
prescribed rectificable boundary, Arch. Rat. Mec. Anal. 35 (1969), 249252.
[10] Hildebrandt, S.: On the Plateau problem for surfaces of constant mean curvature, Comm.
Pure Appl. Math. 23 (1970), 97114.
[11] Hopf, H.: Differential Geometry in the Large, Lecture Notes in Mathematics, 1000, SpringerVerlag, Berlin, 1983.
[12] Isenberg, C.: The Science of Soap Films and Soap Bubbles, Dover, New York, 1992.
[13] Kapouleas, N.: Compact constant mean curvature surfaces in Euclidean three-space, J. Diff.
Geom. 33 (1991), 683715.
[14] Kenmotsu, K.: Surfaces with constant mean curvature, American Math. Soc., Providence,
2003.
[15] Koiso, M.: Symmetry of hypersurfaces of constant mean curvature with symmetric boundary,
Math. Z. 191 (1986), 567574.
[16] Koiso, M.: A generalization of Steiner symmetrization for immersed surfaces and its applications, Manuscripta Math. 87 (1995), 311325.
[17] Koiso, M.: The uniqueness for stable surfaces of constant mean curvature with free boundary,
Bull Kyoto Univ Educ Ser B. 94 (1999), 17.
[18] Liu, H.: Translation surfaces with constant mean curvature in 3-dimensional spaces, J. Geom.
64 (1999), 141149.
[19] L
opez, R.: Surfaces of constant mean curvature bounded by convex curves, Geom. Dedicata
66 (1997), 255263.
[20] L
opez, R. A note on H-surfaces with boundary, J. Geom. 60 (1997), 8084.
[21] L
opez, R.: Constant mean curvature surfaces with boundary in Euclidean three-space,
Tsukuba J. Math., 23 (1999), 2736.
[22] L
opez, R.: Constant mean curvature graphs on unbounded convex domains, J. Diff. Eq., 171
(2001), 5462.

SURFACES WITH CONSTANT MEAN CURVATURE IN EUCLIDEAN SPACE

101

[23] L
opez, R.: Wetting phenomena and constant mean curvature surfaces with boundary, Reviews Math. Physics, 17 (2005), 769792.
[24] L
opez, R.: On uniqueness of graphs with constant mean curvature, J. Math. Kyoto Univ.,
46 (2007), 771787.
[25] L
opez, R., Montiel, S.: Constant mean curvature disc with boundary, Proc. Amer. Math.
Soc., 123 (1995), 15551558.
[26] L
opez, R., Montiel, S.: Constant mean curvature surfaces with planar boundary, Duke Math.
J., 85 (1996), 583604.
[27] Meeks III, W.: The topology and geometry of embedded surfaces of constant mean curvature,
J. Diff. Geom. 30 (1989), 465503.
[28] Oprea, J.: The Mathematics of Soap Films: Explorations with Maple, 2000 American Math.
Soc.
[29] Osserman, R.: Minimal Surfaces, Dover, 1969.
[30] Serrin, J.: On surfaces of constant mean curvature which span a given space curve, Math.
Z. 112 (1969), 7788.
[31] Serrin, J.: The problem of Dirichlet for quasilinear elliptics equations with many independent
variables, Phil. Trans. Roy. Soc. London A 264 (1969), 413496.
[32] Steffen, K.: Parametric surfaces of prescribed mean curvature, Lectures Note in Math. vol.
1713, 211265, Springer Verlag, Berlin, 1999.
[33] Struwe, M.: Plateaus Problem and the Calculus of Variations, Mathematical Notes, Princenton University Press, Princenton, 1988.
[34] Wente, H.C.: Counterexample to a conjecture of H. Hopf, Pacific J. Math., 121 (1986),
193243.
Departamento de Geometra y Topologa, Universidad de Granada, Granada, Spain
E-mail address: [email protected]