Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
Surfaces
in Euclidean Spaces
Contents
Part I Theory of Surfaces
1 Immersions in R
n+2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Regular surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Remarks on branch points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Orthonormal normal frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Rotation of normal frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 The fundamental forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Parameter systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.7 Parameter transformations. Tensors. . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.8 Vector ﬁelds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Differential equations. Curvatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1 The Gauss equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 The mean curvature vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 The mean curvature system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 The Gauss curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5 The normal mean and the normal Gauss curvature . . . . . . . . . . . . . . . 28
2.6 The Weingarten equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.7 Weingarten forms. Principal curvatures . . . . . . . . . . . . . . . . . . . . . . . . 32
2.8 Application I: Geometry of the Gaussian curvature. . . . . . . . . . . . . . . 33
2.9 Application II: Surfaces with parallel mean curvature vector . . . . . . . 34
3 Integrability conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1 Integrability conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 The CodazziMainardi equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 Gauss integrability conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 The curvature tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.5 theorema egregium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.6 The Ricci equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.7 The curvature tensor of the normal bundle . . . . . . . . . . . . . . . . . . . . . . 45
i
ii Contents
3.8 The curvature of the normal bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.9 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.10 The fundamental theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4 Weighted differential geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2 The weighted fundamental forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3 Differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4 The weighted mean curvature system . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.5 The CodazziMainardi equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5 The Hopf vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.1 Linear dependence of the weighted fundamental forms . . . . . . . . . . . 71
5.2 Minimal surfaces and weighted minimal surfaces . . . . . . . . . . . . . . . . 72
5.3 The torsion of orthonormal normal frames . . . . . . . . . . . . . . . . . . . . . . 73
5.4 The functional of total torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.5 The curvatura integra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.6 Hopf functions and Hopf vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.7 A Pascali system for the Hopf functions . . . . . . . . . . . . . . . . . . . . . . . . 81
5.8 An example: Weighted minimal surfaces in R
3
. . . . . . . . . . . . . . . . . . 83
6 The GaussOsserman map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.1 The exterior product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2 The Grassmann normal space. Grassmann forms . . . . . . . . . . . . . . . . 89
6.3 Curvature vector and curvature matrix of the normal bundle . . . . . . . 91
6.4 Grassmann manifolds and GaussOsserman map . . . . . . . . . . . . . . . . 94
6.5 FubiniStudy metric and the total curvature . . . . . . . . . . . . . . . . . . . . . 96
6.6 Minimal surfaces with constant curvature
´
K . . . . . . . . . . . . . . . . . . . . 101
Part II Variational Problems
7 Normal Coulomb frames in R
4
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.1 The total torsion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.2 Curves in R
3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.3 Coulomb ONF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.4 Construction of normal Coulomb frames . . . . . . . . . . . . . . . . . . . . . . . 110
7.5 Minimality of normal Coulomb frames . . . . . . . . . . . . . . . . . . . . . . . . 111
7.6 A torsion estimates via the maximum principle . . . . . . . . . . . . . . . . . . 111
7.7 A torsion estimate via a RiemannHilbert problem . . . . . . . . . . . . . . . 114
7.8 Estimates for the total torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.9 An example: Holomorphic graphs in C
2
. . . . . . . . . . . . . . . . . . . . . . . . 119
7.10 Application to the mean curvature ﬂow in R
4
. . . . . . . . . . . . . . . . . . . 120
Contents iii
8 Normal Coulomb frames in R
n+2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.1 Problem formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.2 The EulerLagrange equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
8.4 Quadratic growth in the gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.5 Torsion free normal frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
8.6 Nonﬂat normal bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
8.7 Bounds for the total torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.8 Existence and regularity of weak normal Coulomb frames . . . . . . . . 147
8.9 Classical regularity of normal Coulomb frames . . . . . . . . . . . . . . . . . . 157
9 Minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
9.1 Minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
9.2 The ﬁrst variation of nonparametric functionals. Minimal graphs . . 163
9.3 Geometry of minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
9.4 A priori estimates for elliptic systems with quadratic growth . . . . . . 168
9.5 A curvature estimate for minimal graphs with subquadratic growth . 170
9.6 A theorem of Bernsteintype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
9.7 Osserman’s curvature estimate for minimal surfaces . . . . . . . . . . . . . . 173
9.8 A theorem of Bernstein type due to Osserman . . . . . . . . . . . . . . . . . . . 181
9.9 Gradient estimates for minimal graphs . . . . . . . . . . . . . . . . . . . . . . . . . 182
10 Immersions with prescribed mean curvature . . . . . . . . . . . . . . . . . . . . . . 187
10.1 Critical points of parametric Gullivertype functionals . . . . . . . . . . . . 189
10.2 A geometric maximum principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
10.3 A curvature estimate for surfaces with prescribed mean curvature . . 196
10.4 A theorem of Bernstein type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
11 Crystalline functionals in R
3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
11.1 Examples of parametric variational problems . . . . . . . . . . . . . . . . . . . 211
11.2 Further regularity assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
11.3 The ﬁrst variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
11.4 Principal curvatures and weighted mean curvature . . . . . . . . . . . . . . . 227
11.5 Nonparametric differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . 231
11.6 Quasilinear elliptic systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
11.7 Quadratic growth in the gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
11.8 The geometry of immersions of mean curvature type . . . . . . . . . . . . . 253
11.9 A curvature estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
12 Crystalline functionals in R
n+2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
12.1 Crystalline functionals. Remarks on Plateau’s problem . . . . . . . . . . . 267
12.2 Parameter invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
12.3 Euler’s homogeneity relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
12.4 The ﬁrst variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
12.5 Minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
12.6 The problem for weighted minimal surfaces . . . . . . . . . . . . . . . . . . . . 276
iv Contents
13 The second variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
13.1 Minimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
13.2 Immersions with a special mean curvature ﬁeld . . . . . . . . . . . . . . . . . 283
13.3 Stability and µstability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
13.4 µstability due to Schwarz for minimal graphs . . . . . . . . . . . . . . . . . . 289
13.5 Eigenvalue problems on manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
13.6 µstability due to Ruchert, Barbosa and do Carmo . . . . . . . . . . . . . . . 304
13.7 Calibration forms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
14 Energy estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
14.1 Geodesic discs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
14.2 The area of µstable geodesic discs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
14.3 An area estimate for minimal graphs. . . . . . . . . . . . . . . . . . . . . . . . . . . 315
14.4 Area estimates via the curvatura integra . . . . . . . . . . . . . . . . . . . . . . . . 317
14.5 The area of graphs with prescribed mean curvature . . . . . . . . . . . . . . 318
14.6 The isoperimetric inequality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
14.7 The spherical energy of µstable geodesic discs . . . . . . . . . . . . . . . . . 327
15 Fminimal surfaces in R
3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
15.1 Fminimal surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
15.2 The geometry of the spherical mapping . . . . . . . . . . . . . . . . . . . . . . . . 331
15.3 Stability and µstability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
15.4 Eigenvalue problems for the spherical mapping . . . . . . . . . . . . . . . . . 341
15.5 A curvature estimate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
15.6 Theorems of Bernstein type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
List of Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
Part I
Theory of Surfaces
Chapter 1
Immersions in R
n+2
1.1 Regular surfaces
1.2 Remarks on branch points
1.3 Orthonormal normal frames
1.4 Rotation of normal frames
1.5 The fundamental forms
1.6 Parameter systems
1.7 Parameter transformations. Tensors
1.8 Vector ﬁelds
4 1 Immersions in R
n+2
This ﬁrst chapter introduces the reader into basic important concepts of differential geometry of
twodimensional immersions in R
n+2
, particularly including tangential and normal spaces along
with the concept of orthonormal normal frames and orthogonal transformations between them.
Furthermore we introduce the concept of conformal parameters which provide a surface represen
tation most suitable for our purposes. Next we deﬁne the three fundamental forms, and a discussion
about geometrical nature tensors and its transformation behaviour concludes this ﬁrst chapter.
1.1 Regular surfaces
The objects of all our present investigations are twodimensional immersions of disc
type in parametric form with trace in Euclidean spaces R
n+2
with a natural number
n ≥1. More precisely, we consider twodimensional vectorvalued mappings
X = X(u, v) =
_
x
1
(u, v), . . . , x
n+2
(u, v)
_
∈C
4
(B, R
n+2
),
deﬁned on the closed unit disc
B :=
_
(u, v) ∈ R
2
: u
2
+v
2
≤1
_
∈ R
2
.
Additionally we set
˚
B :=
_
(u, v) ∈ R
2
: u
2
+v
2
< 1
_
for the open unit disc, i.e.
˚
B denotes the interior of B, and we write
∂B :=
_
(u, v) ∈ R
2
: u
2
+v
2
= 1
_
for its boundary. Many of our following considerations require simply connected and
smoothly bounded domains of deﬁnition. On the other hand, restricting to surfaces
deﬁned only on the closed unit disc B can justiﬁed on account of Riemann’s mapping
theorem.
1
1
If Ω ⊂Ris a simply connected and open domain which is not all of Cthen there exists a bijective
and holomorphic mapping from Ω onto the open unit disc.
6 1 Immersions in R
n+2
Furthermore we assume that all these mappings X are regular in the differential
geometric sense, i.e. we require they represent immersions in the following sense
rankDX ≡rank
_
_
_
_
x
1
u
x
1
v
.
.
.
.
.
.
x
n+2
u
x
n+2
v
_
_
_
_
= 2 in B (1.1)
for the rank of the Jacobian matrix DX ∈ R
n+2,2
, where the lower indices u and
v denote the partial derivatives w.r.t. to the respective parameters. This condition
ensures that at each point w ∈ B there exist two linearly independent tangential
vectors X
u
and X
v
of the mapping X (simultaneously its partial derivatives w.r.t. u
and v) spanning the twodimensional tangential space at this point w ∈ B, i.e.
T
X
(w) := Span
_
X
u
(w), X
v
(w)
_
∼
=R
2
, w ∈ B. (1.2)
In particular, this leads us to the decomposition
R
n
=T
X
(w) ⊕N
X
(w)
with the ndimensional normal space
N
X
(w) :=
_
Z ∈ R
n+2
: Z X
u
(w) = Z X
v
(w) = 0
_
(1.3)
at each point w ∈ B, where
X Y =
n+2
∑
i=1
x
i
y
i
denotes the inner product between two vectors X and Y.
Finally, the regularity condition (1.1) ensures that at each point w ∈ B the area
element W of the surface is positive, i.e. it holds
W :=
_
(X
u
X
u
)(X
v
X
v
) −(X
u
X
v
)
2
> 0 in B. (1.4)
Example 1.1. The twodimensional mapping
X(u, v) = (u, v, u
2
−v
2
, 2uv): B −→R
4
represents an immersion since the tangential vectors
X
u
= (1, 0, 2u, 2v), X
v
= (0, 1, −2v, 2u)
are always nonparallel and nonvanishing. In particular, it represents a surface
graph deﬁned over the [x, y]plane with coordinates x = u and y = v. As we will
see later, X is also a minimal graph, and it will serve as a typical example in various
situations of our analysis.
1.3 Orthonormal normal frames 7
1.2 Remarks on branch points
We want to exclude irregular points w ∈ B where the regularity condition (1.1) is
violated. An example of such a nonimmersed surface is given by the complexiﬁed
Neil’s parabola
X(w) = (w
2
, w
3
), w = u +iv ∈ C,
representing a minimal surface with a socalled branch point at w = 0 where
2
X
u
(0, 0) = 0 = X
v
(0, 0).
Using a calibration argument we will show in chapter 12 that X has smallest
area compared with all surfaces spanning the same boundary as X. In contrast
to this example, a minimizing surface in threedimensional Euclidean space R
3
is
indeed immersed (see e.g. Osserman’s monograph [128]), at least in the interior,
which shows that the geometry of surfaces in R
n+2
with arbitrary dimension turns
out to be more intricate than the classical differential geometry.
One word about Neil’s parabola
c(t) = (t
2
, t
3
)
found by William Neil (16371670) in 1657: It represents that curve along which
a particle, descending under gravity, describes equal vertical spacings within equal
times (see e.g. Eric Weinstein’s resource http://mathworld.wolfram.com [163]).
Finally, the Henneberg surface, parametrically given by
x
1
(u, v)=2sinhucosv −
2
3
sinh(3u)cos(3v),
x
2
(u, v)=2sinhusinv −
2
3
sinh(3u)sin(3v),
x
3
(u, v)=2cosh(2u)cos(2v),
is an example of a nonimmersed minimal surface in R
3
with a branch point at the
origin (u, v) = (0, 0).
1.3 Orthonormal normal frames
Our regularity assumptions ensure also that for any w ∈ B we can choose n linearly
independent, orthogonal unit normal vectors N
σ
, σ = 1, . . . , n, spanning the normal
space N
X
(w) at w ∈ B.
2
Identifying R
2
with C, we will eventually write X(w) instead of X(u, v) etc.
8 1 Immersions in R
n+2
Due to the contractibility of the domain of deﬁnition B we can extend this set to
a C
3
regular orthonormal normal frame
N(w) =
_
N
1
(w), . . . , N
n
(w)
_
∈ R
n(n+2)
moving smoothly along the whole surface and satisfying
N
σ
(w) X
u
(w) = 0 = N
σ
(w) X
v
(w),
N
σ
(w) N
ϑ
(w) =δ
σϑ
:=
_
1, if σ =ϑ
0, if σ ,=ϑ
for all w ∈ B
(1.5)
with the Kronecker symbol δ
σϑ
.
Deﬁnition 1.1. A matrix N ∈ R
n(n+2)
with these properties is called an orthonor
mal normal frame of X, shortly: ONF.
Example 1.2. The unit normal vector N for an immersion X : B →R
3
is deﬁned as
N(w) :=
X
u
(w) X
v
(w)
[X
u
(w) X
v
(w)[
, w ∈ B,
with the usual vector product in threedimensional Euclidean space. In this case
we simply identify the ONF with the vector N(w).
Example 1.3. The Euler unit normal vectors
´
N
1
:=
1
_
1 +[∇ϕ[
2
(−ϕ
x
, −ϕ
y
, 1, 0),
´
N
2
:=
1
_
1 +[∇ψ[
2
(−ψ
x
, −ψ
y
, 0, 1)
(1.6)
are orthogonal to the surface graph
X(x, y) =
_
x, y, ϕ(x, y), ψ(x, y)
_
∈ R
4
, (x, y) ∈ Ω ⊂R
2
,
with two smooth functions ϕ and ψ. Namely, for σ =1, 2 we immediately compute
´
N
σ
X
x
=
´
N
σ
(1, 0, ϕ
x
, ψ
x
) = 0,
´
N
σ
X
y
=
´
N
σ
(0, 1, ϕ
y
, ψ
y
) = 0.
But note that in general
´
N
1
and
´
N
2
are not orthogonal to each other:
´
N
1
´
N
2
=
∇ϕ ∇ψ
_
(1 +[∇ϕ[
2
)(1 +[∇ψ[
2
)
,= 0.
1.4 Rotation of normal frames 9
Thus we deﬁne new unit vectors
¯
N
1
:=
´
N
1
,
¯
N
2
:=
´
N
2
−(
´
N
1
´
N
2
)
´
N
1
[
´
N
2
−(
´
N
1
´
N
2
)
´
N
1
[
taking into account that
[
´
N
2
−(
´
N
1
´
N
2
)
´
N
1
[
2
= 1 −
(∇ϕ ∇ψ)
2
(1 +[∇ϕ[
2
)(1 +[∇ψ[
2
)
> 0.
This new system (
¯
N
1
,
¯
N
2
) now forms an ONF for the graph X.
Example 1.4. Let
Φ(u, v) :=ϕ(u, v) +iψ(u, v) ∈ C
be a holomorphic function satisfying CauchyRiemann equations
ϕ
x
=ψ
y
, ϕ
y
=−ψ
x
in Ω ⊂R
2
.
Then the surface graph
_
x, y, ϕ(x, y), ψ(x, y)
_
fulﬁlls
X
2
x
= 1 +ϕ
2
x
+ψ
2
x
= 1 +ψ
2
y
+ϕ
2
y
= X
2
y
,
X
x
X
y
= ϕ
x
ϕ
y
+ψ
x
ψ
y
= 0.
We say X is a conformally parametrized surface; see section 1.6 below for an exact
deﬁnition of conformality. Furthermore, we compute
∇ϕ ∇ψ =ϕ
x
ψ
x
+ϕ
y
ψ
y
=−ϕ
x
ϕ
y
+ϕ
y
ϕ
x
= 0,
where ∇ϕ = (ϕ
x
, ϕ
y
) denotes the Euclidean gradient of ϕ. We conclude that the
Euler normals
´
N
1
and
´
N
2
from (1.6) form an orthonormal normal frame.
1.4 Rotation of normal frames
Quantities representing inner geometric properties of a surface do not depend on
the choice of an ONF. Rather we can say an ONF is just a mathematical tool to
describe the behaviour of a surface in terms of dynamical quantities. Hence we may
refer to it as a moving frame.
Thus in a concrete situation we are supposed to prove independence of the ge
ometric quantities from the special choice of an ONF. In particular, this concerns
all curvature quantities describing the inner geometry of a surface as well as its
embedding into the space.
For this purpose we now present a simple algorithm to transform any given ONF
N into a new ONF
¯
N conserving the orientation.
10 1 Immersions in R
n+2
Namely, let us consider matrixvalued orthogonal mappings
R = (r
σω
)
σ,ω=1,...,n
∈C
3
(B, SO
n
)
satisfying the properties
n
∑
σ=1
r
σω
(w)
2
= 1,
n
∑
σ=1
r
ωσ
(w)
2
= 1 for ω = 1, . . . , n,
n
∑
σ=1
r
σω
(w)r
σω
′ (w) = 0,
n
∑
σ=1
r
ωσ
(w)r
ω
′
σ
(w) = 0 for ω ,=ω
′
as well as
det R(w) = 1 for all w ∈ B.
This latter property characterizes the conservation of orientation. For simplicity,
we will only focus on rotations as transformation mappings. Now the promised
transformation between the two ONF is given by
¯
N
σ
(w) =
n
∑
ω=1
r
σω
(w)N
ω
(w), σ = 1, . . . , n. (1.7)
Note that there hold
[
¯
N
σ
[
2
=
n
∑
ω=1
r
2
σω
= 1
as well as
¯
N
σ
¯
N
ϑ
=
n
∑
ω,ω
′
=1
r
σω
r
ϑω
′ δ
ωω
′ =
n
∑
ω=1
r
σω
r
ϑω
=δ
σϑ
in B for all σ, ϑ = 1, . . . , n.
Example 1.5. In case n = 2, i.e. if X is immersed into R
4
, we choose the orthogonal
mapping
R :=
_
cosϕ sinϕ
−sinϕ cosϕ
_
(1.8)
with a rotation angle ϕ ∈ R. In particular, we have
¯
N
1
= cosϕ N
1
+sinϕ N
2
,
¯
N
2
= −sinϕ N
1
+cosϕ N
2
.
Example 1.6. In case n = 3 we may introduce the Euler rotations generated by the
three matrices
R
1
:=
_
_
_
cosϕ sinϕ 0
−sinϕ cosϕ 0
0 0 1
_
_
_, R
2
:=
_
_
_
1 0 0
0 cosψ sinψ
0 −sinψ cosψ
_
_
_
1.5 The fundamental forms 11
as well as
R
3
:=
_
_
_
cosϑ sinϑ 0
−sinϑ cosϑ 0
0 0 1
_
_
_.
The transformation works as follows: First we rotate by ϕ about the zaxis, then by
ψ about the xaxis, and, ﬁnally, by ϑ about the zaxis again. The effective rotation
is then described by (see e.g. Funk [68])
R = R
3
◦ R
2
◦ R
1
. (1.9)
1.5 The fundamental forms
Differential geometry is essentially built up on the three fundamental forms. Already
Gauss [70] was aware of its special importance.
Let us start with the
Deﬁnition 1.2. The ﬁrst fundamental form I(X) ∈ R
22
of X is deﬁned by
I(X) = (g
i j
)
i, j=1,2
, g
i j
:= X
u
i X
u
j (1.10)
setting u
1
:= u and u
2
:= v.
Note that the deﬁnition of this quadratic form does not depend on the codimen
sion. Rather it is simply induced by the Euclidean metric of the embedding space
R
n+2
. To clarify, let us denote by
ds
2
= dx
2
1
+. . . +dx
2
n+2
(1.11)
the standard line element of the space R
n+2
with x
1
, . . . , x
n+2
being its Euclidean co
ordinates. Embedding the surface X =X(u, v) into the Euclidean space R
n+2
means,
from that formal point of view,
ds
2
= (x
1,u
du +x
1,v
dv)
2
+. . . +(x
n+2,u
du +x
n+2,v
dv)
2
= (X
u
X
u
)du
2
+2(X
u
X
v
)dudv +(X
v
X
v
)dv
2
= g
11
du
2
+2g
12
dudv +g
22
dv
2
=
2
∑
i, j=1
g
i j
du
i
du
j
.
Obviously it holds
W =
_
g
11
g
22
−g
2
12
=
_
det I(X)
for the area element W.
12 1 Immersions in R
n+2
Deﬁnition 1.3. The second and the third fundamental form w.r.t. any unit normal
vector N: B →R
n+2
are
II
N
(X) = (L
N,i j
)
i, j=1,2
, L
N,i j
:=−X
u
i N
u
j = X
u
i
u
j N,
III
N
(X) = (e
N,i j
)
i, j=1,2
, e
N,i j
:= N
u
i N
u
j .
(1.12)
Note that these forms now depend on the chosen unit normal vector N.
The ﬁrst and the second fundamental form for surfaces in R
3
were already in
troduced by Gauss [70]. In this special case of one codimension, the coefﬁcients of
the second fundamental form are also denoted by L, M and N, a terminology which
goes back to Reinhold Hoppe (compare Strubecker [152]).
The geometric meaning of the ﬁrst fundamental form is contained in the follow
ing result from elementary differential geometry which we do not prove here.
Theorem 1.1. The knowledge of the coefﬁcients g
i j
, i, j = 1, 2, 3, enables us to an
swer all questions about the metrical properties of a surface, in particular problems
concerning length of and angles between curves on it, or area of pieces of it.
Example 1.7. Consider a curve c: I ⊂R →B and its spatial image X ◦c(t) ∈ R
3
on
the surface X. Let c(t) = (u
1
(t), u
2
(t)). Then its length is given by the integral
L(c
i
) =
_
I
¸
¸
¸
_
2
∑
k,ℓ=1
g
kℓ
du
k
i
dt
du
ℓ
i
dt
dt .
Furthermore, the angle α between two curves X ◦ c
1
and X ◦ c
2
with c
1
= (u
1
, v
1
)
and c
2
= (u
2
, v
2
), and which intersect at c
1
(t) = c
2
(t) = w ∈ B, is
cosα =
c
′
1
c
′
2
[c
′
1
[[c
′
2
[
=
g
11
˙ u
1
˙ u
2
+g
12
¦ ˙ u
1
˙ v
2
+ ˙ u
2
˙ v
1
¦ +g
22
˙ v
1
˙ v
2
_
g
11
˙ u
2
1
+2g
12
˙ u
1
˙ v
1
+g
22
˙ v
2
1
_
g
11
˙ u
2
2
+2g
12
˙ u
2
˙ v
2
+g
22
˙ v
2
2
.
1.6 Parameter systems
In this section we want to present socalled conformal parameter systems and
geodesic polar coordinates. Other possible parametrization are discussed in later
chapters.
1.6 Parameter systems 13
Conformal parameters
Deﬁnition 1.4. The immersion X : B→R
n+2
is said to be conformally parametrized
if the conformality relations
X
u
X
u
=W = X
v
X
v
, X
u
X
v
= 0 in B (1.13)
with the area element W from (1.4) as conformal factor are satisﬁed.
The special meaning of these coordinates is that they diagonalize the Riemannian
line element such that we get
ds
2
:= X
2
u
du
2
+2X
u
X
t
v
dudv +X
2
v
du
2
=W(du
2
+dv
2
)
on the whole disc B. Here we say ds
2
is of Riemannian type if it holds
X
2
u
X
2
v
−(X
u
X
v
)
2
> 0 in B.
This is always true in view of our regularity assumptions.
Introducting conformal parameters into a Riemannian metric is justiﬁed by the
following results.
Proposition 1.1. (Sauvigny [142], 1999)
Assume that the coefﬁcients a, b and c of the Riemannian metric
ds
2
= adu
2
+2bdudv +cdv
2
are of class C
1+α
(B, R) with α ∈(0, 1). Then there is a conformal parameter system
(u, v) ∈ B.
The regularity condition required here is satisﬁed in our situation because we
suppose g
i j
∈C
3
(B, R). While Sauvigny’s result holds in the large, i.e. on the whole
closed disc B, further stronger results hold in the small. One example is the follow
ing.
Proposition 1.2. (Chern [28], 1955)
Assume that the coefﬁcients of the Riemannian metric
ds
2
= adu
2
+2bdudv +cdv
2
are H¨ older continuous in B. Then for every point w ∈
˚
B there exists an open neigh
borhood over which the surface can be pararmetrized conformally.
Example 1.8. As we already know, the mapping
X(w) = (w, w
2
) ∈ R
4
fulﬁlls the conformality relations.
14 1 Immersions in R
n+2
Example 1.9. The catenoid is a rotationally minimal surface which can be repre
sented using conformal parameters as
X(u, v) = (coshucosv, coshusinv, u) ∈ R
3
.
The conformally parametrized minimal helicoid
¯
X(u, v) = (−sinhusinv, sinhucosv, −v) ∈ R
3
is closely related to this catenoid: Namley, let λ ∈ [0,
π
2
] be a real parameter, then
catenoid and helicoid can be transformed isometrically into each other by means of
the mapping
X
λ
(u, v) := X(u, v)cosλ +
¯
X(u, v)sinλ.
In particular, X
0
represents the catenoid, while Xπ
2
equals the helicoid. This isometry
is a special case of a general theorem due to Bour on the isometrically deforming a
rotationally symmetric surface into a ruled surface (see e.g. Strubecker [152]).
Remark 1.1. Note that if only X ∈C
1
(B, R
n+2
) it is in general not true that X can be
parametrized conformally, see e.g. Chern, Hartman and Wintner [29] for a detailed
discussion along with various counterexamples.
Geodesic polar coordinates
For a detailed introduction of the socalled exponential map, geodesic discs and
geodesic polar coordinates we refer the reader to elementary textbooks on differen
tial geometry, e.g. Blaschke and Leichtweiss [15], Klingenberg [106], or Laugwitz
[111]. Here we just want to provide some important identities needed laters.
Assume that the immersion X : B →R
n+2
(or a part of it) is given as a geodesic
disc B
r
(X
0
) of geodesic radius r > 0 and with center X
0
∈ R
n+2
. Using geodesic
polar coordinates (ρ, ϕ) ∈ [0, r] [0, 2π] we can rewrite the mapping X = X(u, v)
into the new form
Z = Z(ρ, ϕ): [0, r] [0, 2π] −→R
n+2
.
Following e.g. Blaschke and Leichtweiss [15], §79, the new line element ds
2
P
reads
ds
2
P
= Z
2
ρ
dρ
2
+2Z
ρ
Z
ϕ
dρdϕ +Z
2
ϕ
dϕ
2
= dρ
2
+P(ρ, ϕ)dϕ
2
.
with a function P ∈C
1
((0, r] [0, 2π), R) satisfying
P(ρ, ϕ) > 0 for all (ρ, ϕ) ∈ (0, r] [0, 2π)
as well as
lim
ρ→0
+
P(ρ, ϕ) = 0, lim
ρ→0
+
∂
∂ρ
_
P(ρ, ϕ) = 1 for all ϕ ∈ [0, 2π).
1.7 Parameter transformations. Tensors 15
1.7 Parameter transformations. Tensors
We want to restrict our investigations to the following special class of parameter
transformations.
Deﬁnition 1.5. The C
4
regular parameter transformation
v
α
→u
i
(v
α
), i = 1, 2, α = 1, 2,
is called of regularity class P, shortly: u
i
(v
α
) ∈ P, if it is positively oriented, i.e.
∂(u
1
, u
2
)
∂(v
1
, v
2
)
> 0
for its Jacobian, and if it does not effect the orientation of the unit normal vectors of
the ONF N.
We compute the transformed derivatives of the surface vector
X
u
i =
2
∑
λ=1
X
v
λ
v
λ
u
i
=
2
∑
λ=1
Λ
λ
i
X
v
λ
setting Λ
λ
i
:= v
λ
u
i
for i =1, 2. The tangential vectors X
u
i of the surface X are inner geometrical objects.
From our point of view, they are parametrized vectorvalued mappings. Thus it is
important to compare the transformation behaviour of any of our quantities with that
of the X
u
i what ﬁnally leads us to the concept of tensors.
Let us start with the transformation behaviour of the g
i j
:
g
i j
= X
u
i X
u
j =
2
∑
κ,λ=1
Λ
κ
i
Λ
λ
j
X
v
κ X
v
λ =
2
∑
κ,λ=1
Λ
κ
i
Λ
λ
j
g
κλ
Next let
¯
Λ
i
κ
denote the coefﬁcients of the inverse transformation satisfying
2
∑
κ=1
Λ
κ
i
¯
Λ
j
κ
=δ
j
i
,
2
∑
i=1
Λ
κ
i
¯
Λ
i
λ
=δ
κ
λ
with the Kronecker symbols δ
j
i
and δ
κ
λ
. It follows the inverse transformation rule
2
∑
i, j=1
g
i j
¯
Λ
i
µ
¯
Λ
j
ν
=
2
∑
i, j=1
2
∑
κ,λ=1
(Λ
κ
i
¯
Λ
i
µ
)(Λ
λ
j
¯
Λ
j
ν
)g
κλ
=
2
∑
κ,λ=1
δ
κ
µ
δ
λ
ν
g
κλ
= g
µκ
.
As an immediate consequence we obtain the
16 1 Immersions in R
n+2
Corollary 1.1. Let v
α
→u
i
(v
α
) be a parameter transformation from class P, and
denote by W(u
1
, u
2
) and W(v
1
, v
2
) the area elements of the surface X : B →R
n+2
w.r.t. u
i
resp. v
α
. Then it holds
W(u
1
, u
2
) = (Λ
1
1
Λ
2
2
−Λ
2
1
Λ
1
2
)W(v
1
, v
2
).
Proof. This follows from evaluating the expression
W(u
1
, u
2
)
2
=
_
2
∑
κ,λ=1
Λ
κ
1
Λ
λ
1
g
κλ
__
2
∑
µ=1
Λ
µ
2
Λ
ν
2
g
µν
_
−
_
2
∑
κ,λ=1
Λ
κ
1
Λ
λ
2
g
κλ
__
2
∑
µ,ν=1
Λ
µ
1
Λ
ν
2
g
µν
_
which proves the statement. ⊓⊔
There will also appear tensors of higher rank, for example the curvature tensors
of surfaces.
Deﬁnition 1.6. Let r ∈ N, r ≥ 1. The function a
i
1
...i
r
is called a covariant tensor of
rank r, or shortly a rcovariant tensor if and only if it transforms in the way
a
i
1
...i
r
=
2
∑
µ
1
,...,µ
r
=1
Λ
µ
1
i
1
. . . Λ
µ
r
i
r
a
µ
1
...µ
r
.
Beside these covariant components we need contravariant tensors. These forms
are deﬁned in terms of the inverse ﬁrst fundamental form, i.e. in terms of the coefﬁ
cients g
i j
deﬁned by
2
∑
j=1
g
i j
g
jk
=δ
k
i
with the Kronecker symbol δ
k
i
.
Deﬁnition 1.7. Let r ∈ N, r ≥ 1. The contravariant components of the covariant
tensor a
i
1
...i
s
are deﬁned as
a
i
1
...i
r
:=
2
∑
j
1
,..., j
r
=1
g
i
1
j
1
. . . g
i
r
j
r
a
j
1
... j
r
.
To compute the transformation behaviour of the coefﬁcients g
kℓ
we ﬁrst notice
g
kℓ
=
2
∑
i, j=1
g
i j
g
ik
g
jℓ
.
1.7 Parameter transformations. Tensors 17
Then we calculate
g
kℓ
=
2
∑
i, j=1
g
i j
g
ik
g
jℓ
=
2
∑
i, j=1
2
∑
κ,λ=1
(Λ
κ
i
Λ
λ
j
g
κλ
)g
ik
g
jℓ
=
2
∑
i, j=1
2
∑
κ,λ=1
(Λ
κ
i
g
ik
)(Λ
λ
j
g
jℓ
)g
κλ
=
2
∑
i, j=1
2
∑
κ,λ=1
2
∑
µ,ν=1
(Λ
κ
i
g
ik
)(Λ
λ
j
g
jℓ
)g
κµ
g
λν
g
µν
=
_
2
∑
i=1
2
∑
κ=1
Λ
κ
i
g
ik
g
κµ
__
2
∑
j=1
2
∑
λ=1
Λ
λ
j
g
jℓ
g
λν
_
g
µν
=
2
∑
µ,ν=1
¯
Λ
k
µ
¯
Λ
ℓ
ν
g
µν
where we take
2
∑
i=1
2
∑
κ=1
Λ
κ
i
g
ik
g
κµ
=
2
∑
i=1
2
∑
κ=1
Λ
κ
i
g
ik
_
2
∑
m,n=1
g
mn
¯
Λ
m
κ
¯
Λ
n
µ
_
=
2
∑
i=1
2
∑
m,n=1
2
∑
κ=1
Λ
κ
i
¯
Λ
m
κ
¯
Λ
n
µ
g
ik
g
mn
=
2
∑
i=1
2
∑
m,n=1
δ
m
i
¯
Λ
n
µ
g
ik
g
mn
=
2
∑
m,n=1
¯
Λ
n
µ
g
mk
g
mn
=
2
∑
n=1
¯
Λ
n
µ
δ
k
n
=
¯
Λ
k
µ
etc. into account. This gives rise to our next
Deﬁnition 1.8. Let s ∈ N, s ≥1. The function a
i
1
...i
s
is called a contravariant tensor
of rank s, or shortly a scontravariant tensor if and only if it transforms as
a
i
1
...i
r
=
2
∑
µ
1
,...,µ
r
=1
¯
Λ
i
1
µ
1
. . .
¯
Λ
i
r
µ
s
a
µ
1
...µ
s
. (1.14)
Finally, we need covariant and contraviant tensor components.
Deﬁnition 1.9. Let r, s ∈ N, r, s ≥ 1. The function a
j
1
... j
s
i
1
...i
r
is called a (r, s)tensor if
and only if it transforms as
a
j
1
... j
s
i
1
...i
r
=
2
∑
µ
1
,...,µ
r
=1
2
∑
ν
1
,...,ν
s
=1
Λ
µ
1
i
1
. . . Λ
µ
r
i
r
¯
Λ
j
1
ν
1
. . .
¯
Λ
j
s
ν
s
a
ν
1
...ν
s
µ
1
...µ
r
. (1.15)
For detailed treatises we refer the reader to textbooks on differential geometry,
e.g. B¨ ar [4], do Carmo [21], [22], K¨ uhnel [109], or Raschewski [135].
18 1 Immersions in R
n+2
1.8 Vector ﬁelds
Geometric quantities can also be expressed using coordinateindependent vector
ﬁelds. Although we will mainly use special coordinate systems matching our spe
cial problems, we eventually formulate results also using geometric vector ﬁelds to
emphasize their geometric importance.
In particular, covariant tensors are sort of multivalued linear mappings on the
tangential space. Let us consider the following example for illustration: The coef
ﬁcients g
i j
of the ﬁrst fundamental form can be considered as the components of a
bilinear mapping
g: T
X
(w) T
X
(w) −→R
deﬁned as
X
u
i , X
u
j −→g
i j
:= g(X
u
i , X
u
j ) ∈ R.
The advantage of this approach is that we can deﬁne a metric g without referring
to a special parametrization. Namely, let ¸, ) be a positive deﬁnite quadratic form
on T
X
(w) T
X
(w). Then we simply set
g(X, Y )(w) :=¸X, Y )(w)
for two tangential vectors X, Y ∈ T
X
(w). The form ¸, , ) could result from pro
jecting the metric of the embedding space to the tangential space of the surface,
the former represented by the Euclidean inner product. Then g generates our ﬁrst
fundamental form I(X) from above.
Differential geometry can be formulated completely using this coordinatefree
vector ﬁeld technique, and such an approach takes real account of the geometrical
nature inherent in our surfaces and manifolds. But eventually it conceals subtle ana
lytical structures of the differential equations underlying the geometric phenomena
what would not be detected until we introduce suitable parameter systems.
Chapter 2
Differential equations. Curvatures
2.1 The Gauss equations
2.2 The mean curvature vector
2.3 The mean curvature system
2.4 The Gauss curvature
2.5 The normal mean and the normal Gauss curvature
2.6 The Weingarten equations
2.7 Weingarten forms. Principal curvatures
2.8 Application I: Geometry of the Gaussian curvature
2.9 Application II: Surfaces with parallel mean curvature vector
20 2 Differential equations. Curvatures
In this second chapter we introduce various curvature quantities of surfaces and discuss their geo
metric invariances. We derive the differential equations of Gauss and Weingarten which will lead
us to the meancurvaturesystem. Furthermore we introduce a concept of torsions as connection
coefﬁcients in the normal space. This is exactly what distinguishes the differential geometry in R
3
from the theory in spaces of higher dimensions.
2.1 The Gauss equations
The connection coefﬁcients of the metric (g
i j
)
i, j=1,2
are the Christoffel symbols.
1
Deﬁnition 2.1. The Christoffel symbols of the immersion X : B →R
n+2
are
Γ
k
i j
:=
1
2
2
∑
k=1
g
kℓ
(g
jℓ,u
i +g
ℓi,u
j −g
i j,u
ℓ ) (2.1)
for i, j, k = 1, 2, where the lower index u
j
denotes the partial derivative w.r.t. the
parameter u
j
.
Their conformal representations can be found in (2.6) below. For the moment
we admit arbitrary parametrizations and derive our ﬁrst set of partial differential
equations describing the dynamical behaviour of an immersion in terms of an ac
companying frame
_
X
u
, X
v
, N
1
, . . . , N
n
_
with a chosen ONF N =¦N
1
, . . . , N
n
¦. These equations are named after the German
mathematician Carl Friedrich Gauss (*1777 in Braunschweig; †1855 in G¨ ottingen).
Theorem 2.1. (Gauss equations)
Let the immersion X : B →R
n+2
together with an ONF N be given. Then there hold
X
u
i
u
j =
2
∑
k=1
Γ
k
i j
X
u
k +
n
∑
ϑ=1
L
ϑ,i j
N
ϑ
(2.2)
for i = 1, 2.
1
The name “connection coefﬁcients” has its origin in the fact that the Γ
k
i j
determine how to displace
parallely a vector on the surface, and thus how some vector of X is related or connected to another
vector in its neighbourhood; see e.g. Weyl [169], chapter II.
22 2 Differential equations. Curvatures
Here we use the abbreviation
L
ω,i j
:= L
N
ω
,i j
for the coefﬁcients of the second fundamental form w.r.t. the unit normal vector N
ω
.
Proof. We evaluate the ansatz
X
u
i
u
j =
2
∑
k=1
a
k
i j
X
u
k +
n
∑
ϑ=1
b
ϑ,i j
N
ϑ
with unknown functions a
k
i j
and b
ϑ,i j
. A ﬁrst multiplication by N
ω
yields the coefﬁ
cients b
ϑ,i j
in the form
L
ω,i j
=−X
u
i N
ω,u
j = X
u
i
u
j N
ω
=
n
∑
ϑ=1
b
ϑ,i j
N
ϑ
N
ω
= b
ω,i j
.
To compute the a
k
i j
we multiply by X
u
ℓ and arrive at
X
u
i
u
j X
u
ℓ =
2
∑
k=1
a
k
i j
X
u
k X
u
ℓ =
2
∑
k=1
a
k
i j
g
kℓ
=: a
iℓ j
. (2.3)
Note that a
iℓ j
= a
jℓi
due to X
u
i
u
j = X
u
j
u
i . We calculate
a
iℓ j
= (X
u
i X
u
ℓ )
u
j −X
u
i X
u
ℓ
u
j = g
iℓ,u
j −a
ℓi j
,
and therefore it holds g
iℓ,u
j = a
iℓ j
+a
ℓi j
. It follows that
g
jℓ,u
i +g
ℓi,u
j −g
i j,u
ℓ = a
jℓi
+a
ℓ ji
+a
ℓi j
+a
iℓ j
−a
i jℓ
−a
jiℓ
= 2a
iℓ j
,
and together with (2.3) we arrive at
2
∑
k=1
a
k
i j
g
kℓ
=
1
2
(g
jℓ,u
i +g
ℓi,u
j −g
i j,u
ℓ ).
Rearranging gives
a
m
i j
=
1
2
2
∑
ℓ=1
g
mℓ
(g
jℓ,u
i +g
ℓi,u
j −g
i j,u
ℓ )
which already proves the statement. ⊓⊔
With the proof of the Gauss equations we followed the lines of Blaschke and
Leichtweiss [15].
Note that the vector X
u
i
u
j has no geometric meaning since the Christoffel symbols
Γ
k
i j
do not transform like a (2, 1)tensor as the following result shows.
2.1 The Gauss equations 23
Lemma 2.1. Let u
i
(v
α
) ∈ P. Then there hold
Γ
k
i j
=
2
∑
α,β,γ=1
Γ
γ
αβ
¯
Λ
k
γ
Λ
α
i
Λ
β
j
+
2
∑
α=1
¯
Λ
k
α
Λ
α
i,u
j
for i, j, k = 1, 2 with the coefﬁcients Λ
α
i
and
¯
Λ
i
α
from section 1.7.
Proof. Recall the transformation formula for g
i j
and g
kℓ
from section 1.7. It follows
g
i j,u
k =
2
∑
α,β,γ=1
Λ
α
i
Λ
β
j
Λ
γ
k
g
αβ,v
γ +
2
∑
α,β=1
Λ
α
i,u
k
Λ
β
j
g
αβ
+
2
∑
α,β=1
Λ
α
i
Λ
β
j,u
k
g
αβ
.
We compute
Γ
k
i j
=
1
2
2
∑
ℓ=1
2
∑
α,β,γ,δ,ε=1
¯
Λ
k
α
¯
Λ
ℓ
β
g
αβ
. . .
. . .
_
Λ
γ
ℓ
Λ
δ
i
Λ
ε
j
g
γδ,u
ε +Λ
γ
j
Λ
δ
ℓ
Λ
ε
i
g
γδ,u
ε −Λ
γ
i
Λ
δ
j
Λ
ε
ℓ
g
γδ,u
ε
_
+
1
2
2
∑
ℓ=1
2
∑
α,β,γ,δ=1
¯
Λ
k
α
¯
Λ
ℓ
β
g
αβ
_
Λ
γ
ℓ,u
j
Λ
δ
i
+Λ
γ
j,u
i
Λ
δ
ℓ
−Λ
γ
i,u
ℓ
Λ
δ
j
_
g
γδ
+
1
2
2
∑
ℓ=1
2
∑
α,β,γ,δ=1
¯
Λ
k
α
¯
Λ
ℓ
β
g
αβ
_
Λ
γ
ℓ
Λ
δ
i,u
j
+Λ
γ
j
Λ
δ
ℓ,u
i
−Λ
γ
i
Λ
δ
j,u
ℓ
_
g
γδ
.
The ﬁrst row can be rearranged into the form
1
2
2
∑
α,β,δ,ε=1
g
αβ
_
g
βδ,u
ε +g
εβ,u
γ −g
γε,u
δ
_
Λ
δ
i
Λ
ε
j
¯
Λ
k
α
=
2
∑
α,δ,ε=1
Γ
α
δε
Λ
δ
i
Λ
ε
j
¯
Λ
k
α
.
For the second and the third line it follows
1
2
2
∑
α,β,γ,δ=1
¯
Λ
k
α
¯
Λ
ℓ
β
g
αβ
. . .
. . .
_
Λ
γ
ℓ,u
j
Λ
δ
i
+Λ
γ
i,u
j
Λ
δ
ℓ
−Λ
γ
i,u
ℓ
Λ
δ
j
+Λ
γ
i,u
j
Λ
δ
ℓ
+Λ
γ
i,u
ℓ
Λ
δ
j
−Λ
γ
ℓ,u
j
Λ
δ
i
_
g
γδ
=
1
2
2
∑
α,β,γ,δ=1
¯
Λ
k
α
¯
Λ
ℓ
β
g
αβ
_
Λ
γ
i,u
j
Λ
δ
ℓ
+Λ
γ
i,u
j
Λ
δ
ℓ
_
g
γδ
=
2
∑
α,β,γ=1
¯
Λ
k
α
Λ
γ
i,u
j
g
αβ
g
γβ
=
2
∑
α=1
¯
Λ
k
α
Λ
α
i,u
j
,
which proves the statement. ⊓⊔
24 2 Differential equations. Curvatures
2.2 The mean curvature vector
In the next section we will derive the mean curvature system from the Gauss equa
tions which will be of central interest in later considerations. In preparation for this
we start with introducing a normaldependent mean curvature quantity. Its geomet
ric meaning will be discussed shortly.
Deﬁnition 2.2. The mean curvature H
N
of an immersion X : B →R
n+2
w.r.t. the
unit normal vector N ∈ R
n+2
is deﬁned as
H
N
:=
1
2
2
∑
i, j=1
g
i j
L
N,i j
=
L
N,11
g
22
−2L
N,12
g
12
+L
N,22
g
11
2W
2
. (2.4)
Consider now an ONF N = (N
1
, . . . , N
n
), and set H
σ
:= H
N
σ
.
Deﬁnition 2.3. Let the immersion X : B →R
n+2
together with an ONF N be given.
Then its mean curvature vector H ∈ R
n
is deﬁned as
H :=
n
∑
σ=1
H
σ
N
σ
. (2.5)
Surfaces in R
3
possess, up to orientation, exactly one mean curvature and there
fore exactly one mean curvature vector. Its orientations agrees with the orientation
of the unit normal vector N ∈ R
3
. In this situation we will emphasize the special
nature of the reell number H and the vector N ∈ R
3
.
For the present, H represents vector in R
n
and N a matrix in R
n(n+2)
.
As we will elaborate later, the mean curvature vector H vanishes identically for
socalled minimal surfaces. Thus the vector H for immersions in R
n+2
hardly takes
on the role the unit normal vector N ∈ R
3
claims in case of one codimension. That
is one reason why we rather work with orthonormal normal frames instead of with
the mean curvature vector.
Nevertheless, the vector H possesses interesting analytical and geometric fea
tures.
Proposition 2.1. The curvatures H
N
and H are invariant w.r.t. parameter tranfor
mations of class P. Furthermore the mean curvature vector H does not depend on
the choice of the ONF N.
Proof. Let N ∈ R
n+2
be a unit normal vector. We compute
2
∑
i, j=1
g
i j
L
N,i j
=
2
∑
i, j=1
2
∑
α,β,µ,ν=1
(g
αβ
¯
Λ
i
α
¯
Λ
j
β
)(L
N,µν
Λ
µ
i
Λ
ν
j
)
=
2
∑
i, j=1
2
∑
α,β,µ,ν=1
g
αβ
L
N,µν
δ
µ
α
δ
ν
β
=
2
∑
µ,ν=1
g
µν
L
N,µν
from where we already infer the parameter invariance for H
N
and H.
2.3 The mean curvature system 25
Thus we are allowed to introduce conformal parameters. Then, for a frame
¯
N, gen
erated by a rotation mapping R = (r
σω
)
σ,ω=1,...,n
as described in the ﬁrst chapter,
we have
n
∑
σ=1
¯
H
σ
¯
N
σ
=
1
2W
n
∑
σ=1
(X
uu
¯
N
σ
+X
vv
¯
N
σ
)
¯
N
σ
=
1
2W
n
∑
σ=1
_
n
∑
ω=1
_
X
uu
(r
σω
N
ω
) +X
vv
(r
σω
N
ω
)
_
__
n
∑
ω
′
=1
r
σω
′ N
ω
′
_
=
1
2W
n
∑
ω,ω
′
=1
ω,=ω
′
_
n
∑
σ=1
r
σω
r
σω
′
_
(X
uu
N
ω
+X
vv
N
ω
)N
ω
′
+
1
2W
n
∑
ω=1
_
n
∑
σ=1
r
σω
r
σω
_
(X
uu
N
ω
+X
vv
N
ω
)N
ω
=
1
2W
n
∑
ω=1
(X
uu
N
ω
+X
vv
N
ω
)N
ω
=
n
∑
ω=1
H
ω
N
ω
.
All statements are proved. ⊓⊔
Let us come back to the normaldependent curvatures H
σ
≡H
N
σ
. Using the fore
going result we can prove they are uniquely deﬁned at one ﬁxed point w ∈ B if any
normal frame N is ﬁxed there.
Proposition 2.2. Let the immersion X : B →R
n+2
together with two ONF N and
¯
N
be given. Assume that
N
σ
(w
0
) =
¯
N
σ
(w
0
) for all σ = 1, . . . , n
at some point w
0
∈ B. Then there hold
H
σ
(w
0
) =
¯
H
σ
(w
0
) for all σ = 1, . . . , n.
Proof. Due to the preceding proposition it holds
H(w
0
) =
n
∑
σ=1
H
σ
(w
0
)N
σ
(w
0
) =
n
∑
σ=1
¯
H
σ
(w
0
)
¯
N
σ
(w
0
) =
n
∑
σ=1
¯
H
σ
(w
0
)N
σ
(w
0
).
Comparing the components of the basis vectors N
σ
(w
0
) proves the statement. ⊓⊔
2.3 The mean curvature system
From the Gauss equations (2.2) we want to derive an elliptic system for conformally
parametrized immersions with prescribed mean curvature vector H.
26 2 Differential equations. Curvatures
For this purpose we must rewrite the Christoffel symbols from (2.1) in these
special coordinates (the proof is again left to the reader):
Γ
1
11
=
W
u
2W
, Γ
1
12
=Γ
1
21
=
W
v
2W
, Γ
1
22
=−
W
u
2W
,
Γ
2
11
=−
W
v
2W
, Γ
2
12
=Γ
2
21
=
W
u
2W
, Γ
2
22
=
W
v
2W
.
(2.6)
Theorem 2.2. (Mean curvature system)
Let the conformally parametrized immersion X : B →R
n+2
with mean curvature
vector H be given. Then it holds
∆X := X
uu
+X
vv
= 2
n
∑
ϑ=1
H
ϑ
WN
ϑ
= 2HW in B. (2.7)
Proof. From the Gauss equations (2.2) we infer
∆X = (Γ
1
11
+Γ
1
22
)X
u
+(Γ
2
11
+Γ
2
22
)X
v
+
n
∑
ϑ=1
(L
ϑ,11
+L
ϑ,22
)N
ϑ
.
Note that Γ
1
11
+Γ
1
22
= 0, Γ
2
11
+Γ
2
22
= 0 as well as L
ϑ,11
+L
ϑ,22
= 2H
ϑ
W. The state
ment follows. ⊓⊔
This system (2.7) generalizes the classical mean curvature system
∆X = 2HWN in B (2.8)
from Hopf [94] valid case n = 1 of one codimension with the scalar mean curvature
H ∈ R and the unit normal vector N ∈ R
3
of an immersion X : B →R
3
.
Deﬁnition 2.4. A surfaces is called a minimal surface if and only if it holds H ≡0.
Minimal surfaces are the topic of a countless literature: Courant [38], Nitsche
[126], Lawson [112], Giusti [72], Osserman [130], Dierkes et al. [44], Colding and
Minicozzi [35], Eschenburg and Jost [52] to enumerate only some few signiﬁcant
distributions and to illustrate the importance of this special surface class in the ﬁeld
of geometric analysis and differential geometry.
2.4 The Gauss curvature
To introduce the Gaussian curvature of an immersion X, we start with the following
curvature quantities K
N
σ
along unit normal vectors N
σ
of a given ONF N.
Deﬁnition 2.5. The Gaussian curvature K of an immersion X : B →R
n+2
is
K :=
n
∑
σ=1
K
N
σ
, K
N
σ
:=
L
N
σ
,11
L
N
σ
,22
−L
2
N
σ
,12
g
11
g
22
−g
2
12
. (2.9)
2.4 The Gauss curvature 27
Note that due to its dependence on a unit normal vector N, the quantity K
N
has
no intrinsic nature in contrast to Gaussian curvature K.
Proposition 2.3. The Gaussian curvature K is invariant w.r.t. parameter transfor
mations of class P, and it does not depend on the choice of the ONF N.
Proof. First let N be some unit normal vector. Then the parameter invariance follows
from the identities
(N
u
1 X
u
1 )(N
u
2 X
u
2) −(N
u
1 X
u
2)
2
=
_
(N
v
1 X
v
1)(N
v
2 X
v
2 ) −(N
v
1 X
v
2 )
2
__
Λ
1
1
Λ
2
2
−Λ
2
1
Λ
1
2
_
2
,
(X
u
1 X
u
1)(X
u
2 X
u
2 ) −(X
u
1 X
u
2)
2
=
_
(X
v
1 X
v
1 )(X
v
2 X
v
2) −(X
v
1 X
v
2 )
2
__
Λ
1
1
Λ
2
2
−Λ
2
1
Λ
1
2
_
2
.
To complete this ﬁrst part of the proof, the reader should check a analogous transfor
mation formula for the determinant g
11
g
22
−g
2
12
! Let us now come to the invariance
w.r.t. to orthonormal normal frames. We have
¯
K
σ
W
2
=
¯
L
σ,11
¯
L
σ,22
−
¯
L
2
σ,12
= (X
uu
¯
N
σ
)(X
vv
¯
N
σ
) −(X
uv
¯
N
σ
)
2
,
and the right hand side of this equation can be written in the following form
_
n
∑
ω=1
r
σω
L
ω,11
__
n
∑
ω
′
=1
r
σω
′ L
ω
′
,22
_
−
_
n
∑
ω=1
r
σω
L
ω,12
_
2
=
n
∑
ω,ω
′
=1
r
σω
r
σω
′
_
L
ω,11
L
ω
′
,22
−L
ω,12
L
ω
′
,12
_
=
n
∑
ω=1
r
2
σω
_
L
ω,11
L
ω,22
−L
ω,12
L
ω,12
_
+
n
∑
ω,ω
′
=1
ω,=ω
′
r
σω
r
σω
′
_
L
ω,11
L
ω
′
,22
−L
ω,12
L
ω
′
,12
_
.
Taking the results of section 1.4 into account we have
n
∑
σ=1
n
∑
ω=1
r
2
σω
_
L
ω,11
L
ω,22
−L
2
ω,12
_
=
n
∑
ω=1
_
n
∑
σ=1
r
2
σω
_
_
L
ω,11
L
ω,22
−L
2
ω,12
_
=
n
∑
ω=1
_
L
ω,11
L
ω,22
−L
2
ω,12
_
.
28 2 Differential equations. Curvatures
Analogously it follows
n
∑
σ=1
n
∑
ω,ω
′
=1
ω,=ω
′
r
σω
r
σω
′
_
L
ω,11
L
ω
′
,22
−L
ω,12
L
ω
′
,12
_
= 0.
Thus we conclude
¯
KW
2
=
n
∑
σ=1
¯
K
σ
W
2
=
n
∑
ω=1
_
L
ω,11
L
ω,22
−L
2
ω,12
_
=
n
∑
ω=1
K
ω
W
2
= KW
2
proving the proposition. ⊓⊔
The Gaussian curvature K belongs to the inner geometry of an immersion and
neither depends on the parametrization nor its embedding in space.
In the next chapter we will prove that K can be expressed using the surface’s
ﬁrst fundamental form and its ﬁrst and second derivatives exclusively. This is the
contents of the wellknown theorema egregium.
No matter the geometric meaning of the normaldependent curvatures K
σ
actu
ally is, we next want to show
Proposition 2.4. Given the immersion X : B →R
n+2
together with two ONF N and
¯
N. Suppose that
N
σ
(w
0
) =
¯
N
σ
(w
0
) for all σ = 1, . . . , n
at some ﬁxed point w
0
∈
˚
B. Then there hold
K
σ
(w
0
) =
¯
K
σ
(w
0
) for all σ = 1, . . . , n.
Proof. We calculate
L
σ,11
(w
0
)L
σ,22
(w
0
) −L
σ,12
(w
0
)
2
=
_
X
uu
N
σ
(w
0
)
¸_
X
vv
N
σ
(w
0
)
¸
−
_
X
uv
N
σ
(w
0
)
¸
2
=
_
X
uu
¯
N
σ
(w
0
)
¸_
X
vv
¯
N
σ
(w
0
)
¸
−
_
X
uv
¯
N
σ
(w
0
)
¸
2
=
¯
L
σ,11
(w
0
)
¯
L
σ,22
(w
0
) −
¯
L
σ,12
(w
0
)
2
.
The statement is proved. ⊓⊔
2.5 The normal mean and the normal Gauss curvature
Beside the three fundamental forms introduced in section 1.5 there is a varity of
further quadratic forms reﬂecting the geometry of highercodimensional surfaces
and leading to new curvature concepts.
2.5 The normal mean and the normal Gauss curvature 29
Conveniently one deﬁnes the following quantities.
Deﬁnition 2.6. Let the immersion X : B →R
n+2
be given together with an ONF N.
Let furthermore
L
σϑ,11
:=
2
W
¸
¸
¸
¸
¸
L
σ,11
L
σ,12
L
ϑ,11
L
ϑ,12
¸
¸
¸
¸
¸
, L
σϑ,22
:=
2
W
¸
¸
¸
¸
¸
L
σ,12
L
σ,22
L
ϑ,12
L
ϑ,22
¸
¸
¸
¸
¸
,
L
σϑ,12
:=
1
W
¸
¸
¸
¸
¸
L
σ,11
L
σ,22
L
ϑ,11
L
ϑ,22
¸
¸
¸
¸
¸
for σ, ϑ = 1, . . . , n. Then the normal mean curvature matrix and the normal Gaus
sian curvature matrix of X w.r.t. N are deﬁned as
H := (H
σϑ
)
σ,ϑ=1,...,n
with H
σϑ
:=
2
∑
i, j=1
g
i j
L
σϑ,i j
,
K := (K
σϑ
)
σ,ϑ=1,...,n
with K
σϑ
:=
L
σϑ,11
L
σϑ,22
−L
2
σϑ,12
W
2
for σ, ϑ = 1, . . . , n.
The curvatures H
σϑ
and K
σϑ
are sectional curvatures in the following sense.
Proposition 2.5. For all σ, ϑ = 1, . . . , n there hold the following statements.
1. The curvatures H
σϑ
and K
σϑ
are invariant w.r.t. parameter transformations of
class P.
2. The quadratic forms L
σϑ,i j
, i, j = 1, 2, and therefore the curvatures H
σϑ
and
K
σϑ
do not depend on the choice of an orthonormal basis of Span¦N
σ
, N
ϑ
¦.
In other words, H and K are geometric quantities.
Proof. The parameter invariance can be proved using the methods fromour proof of
Proposition 2.3. Thus we only consider the second statement. Let us start with two
unit normal vectors N
σ
and N
ϑ
with σ ,=ϑ, taken from some ONF N, and evaluate
the SO(2)action on this frame, i.e. let
¯
N
σ
:= cosϕ N
σ
+sinϕ N
ϑ
,
¯
N
ϑ
:=−sinϕ N
σ
+cosϕ N
ϑ
be another orthonormal basis of Span¦N
σ
, N
ϑ
¦.
30 2 Differential equations. Curvatures
Then we exemplarily calculate
W
2
¯
L
σϑ,11
=
¯
L
σ,11
¯
L
ϑ,12
−
¯
L
σ,12
¯
L
ϑ,11
= −sinϕ cosϕ L
σ,11
L
σ,12
+cos
2
ϕ L
σ,11
L
ϑ,12
+sin
2
ϕ L
σ,12
L
ϑ,11
+ sinϕ cosϕ L
ϑ,11
L
ϑ,12
+sinϕ cosϕ L
σ,11
L
σ,12
−cos
2
ϕ L
σ,12
L
ϑ,11
+ sin
2
ϕ L
σ,12
L
ϑ,12
−sinϕ cosϕ L
ϑ,11
L
ϑ,12
= L
σ,11
L
ϑ,12
−L
σ,12
L
ϑ,11
=
W
2
L
σϑ,11
.
Analogously we prove the invariance of L
σϑ,12
and L
σϑ,22
. ⊓⊔
We will come back the curvatures quantities H
σϑ
in section 3.7 again.
2.6 The Weingarten equations
New aspects in the analysis of surfaces in spaces of higher codimensions are mainly
manifested in the connection coefﬁcients of the normal space or, as we prefer to say,
in the torsion coefﬁcients.
Deﬁnition 2.7. The torsion coefﬁcients of an ONF N are deﬁned as
T
ϑ
σ,i
:= N
σ,u
i N
ϑ
, T
ϑ
σ,i
=−T
σ
ϑ,i
, (2.10)
for i = 1, 2 and σ, ϑ = 1, . . . , n.
Most of our deﬁnitions and identities can be found in various textbooks on dif
ferential geometry, for example Chen [27], Brauner [18], or do Carmo [22]. The ter
minology “torsion” even essentially goes back to Weyl [168]: Aus einem normalen
Vektor n in P entsteht ein Vektor n
′
+dt (n
′
normal, dt tangential). Die inﬁnitesimale
lineare Abbildung n →n
′
von N
P
auf N
P
′ ist die Torsion.
2
In chapters 7 and 8 we will elaborately study special normal frames with spe
cial torsions. For the moment we concentrate on the basic differential equations of
surface theory.
Theorem 2.3. (Weingarten equations)
Let the immersion X : B →R
n+2
with an ONF N be given. Then there hold
N
σ,u
i =−
2
∑
j,k=1
L
σ,i j
g
jk
X
u
k +
n
∑
ϑ=1
T
ϑ
σ,i
N
ϑ
(2.11)
for i = 1, 2 and σ = 1, . . . , n.
2
From a normal vector n in P there arises a vector n
′
+dt (n
′
normal, dt tangential). The inﬁnites
imal linear mapping n →n
′
of N
P
to N
P
′ is the torsion.
2.6 The Weingarten equations 31
Proof. With unknown functions a
σ,i
and b
ϑ
σ,i
we evaluate the ansatz
N
σ,u
i =
2
∑
k=1
a
k
σ,i
X
u
k +
n
∑
ϑ=1
b
ϑ
σ,i
N
ϑ
.
Multiplication by X
u
ℓ gives
−L
σ,iℓ
= N
σ,u
i X
u
ℓ =
2
∑
k=1
a
k
σ,i
X
u
k X
u
ℓ =
2
∑
k=1
a
k
σ,i
g
kℓ
,
and rearranging yields
a
m
σ,i
=−
2
∑
ℓ=1
L
σ,iℓ
g
ℓm
.
A further multiplication by N
ω
shows
T
ω
σ,i
= N
σ,u
i N
ω
=
n
∑
ϑ=1
b
ϑ
σ,i
N
ϑ
N
ω
=
n
∑
ϑ=1
b
ϑ
σ,i
δ
ϑω
= b
ω
σ,i
,
which proves the statement. ⊓⊔
The system (2.11) generalizes the classical Weingarten equations
N
u
i =−
2
∑
j,k=1
L
i j
g
jk
X
u
k , i = 1, 2,
for the unit normal vector N of a surface X : B →R
n+2
in the case of one codi
mension n = 1, found by the German mathematician Julius Weingarten (*1836 in
Berlin; †1910 in Freiburg). All torsion coefﬁcients vanish identically here.
There are also immersions living in higher dimensional spaces and admitting
orthonormal normal frames which are globally free of torsion. We will discuss their
properties extensively in later sections.
Finally we want to mention that the T
ϑ
σ,i
transform like a (1, 0)tensor. The proof
is left to the reader.
Proposition 2.6. Let u
i
(v
α
) ∈ P. Then there hold
T
ϑ
σ,i
=
2
∑
µ=1
T
ϑ
σ,µ
Λ
µ
i
for i = 1, 2 and σ, ϑ = 1, . . . , n.
32 2 Differential equations. Curvatures
2.7 Weingarten forms. Principal curvatures
In this section we want to introduce the socalled Weingarten forms to deﬁne alge
braically the principal curvatures of an immersion.
Deﬁnition 2.8. The Weingarten forms (L
k
N,i
)
i,k=1,2
of an immersion X : B →R
n+2
w.r.t. some unit normal vector N ∈ R
n+2
are deﬁned by
L
k
N,i
:= L
N,i j
g
jk
for i, k = 1, 2.
Note that due to the deﬁnition of the curvature quantities H
N
from (2.4) and the
Gauss curvatures K
N
from (2.9) along the unit normal vector N we infer
L
1
N,1
+L
2
N,2
=
2
∑
j=1
L
N,1 j
g
j1
+
2
∑
j=1
L
N,2 j
g
j2
=
2
∑
i, j=1
L
N,i j
g
ji
= 2H
N
as well as
L
1
N,1
L
2
N,2
−(L
2
N,1
) =
2
∑
i, j=1
L
N,1i
L
N,2 j
g
i1
g
j2
−
2
∑
i, j=1
L
N,1i
g
i2
L
N,2 j
g
j1
= (L
N,11
L
N,22
−L
2
N,12
)(g
11
g
22
−g
12
g
21
) = K
N
.
The principal curvatures are now deﬁned for each particular unit normal vector
N as follows.
Deﬁnition 2.9. The principal curvatures κ
N,1
and κ
N,2
of an immersion X : B →
R
n+2
w.r.t. the unit normal vector N ∈ R
n+2
are deﬁned as the roots of the quadratic
eigenvalue equation
det (L
j
N,i
−λ
N
δ
j
i
)
i, j=1,2
= 0.
Note that H
2
N
−K
N
≥0 such that the corresponding eigenvalues follow from
κ
N,1
= H
N
−
_
H
2
N
−K
N
, κ
N,2
= H
N
+
_
H
2
N
−K
N
,
or equivalently
H
N
=
κ
N,1
+κ
N,2
2
, K
N
=κ
N,1
κ
N,2
.
Thus all the classical methods from the theory of surfaces in R
3
apply for each
single H
N
and K
N
.
2.8 Application I: Geometry of the Gaussian curvature 33
2.8 Application I: Geometry of the Gaussian curvature
Gauss himself approached to the curvature quantity K for surfaces in R
3
from a
purely geometrical point of view (we quote from Gauss [70], pp. 10–11):
. . . to each part of a curved surface inclosed within deﬁnite limits we assign a total or integral
curvature, which is represented by the area of the ﬁgure on the sphere corresponding to it.
From this integral curvature must be distinguished the somewhat more speciﬁc curvature
which we shall call the measure of curvature. The latter refers to a point of the surface and
shall denote the quotient obtained when the integral curvature of the surface element about
a point is divided by the area of the element itself; and hence it denotes the ratio of the
inﬁnitely small areas which correspond to one another on the curved surface and on the
sphere.
Or using modern mathematical notation:
K =
[N
u
N
v
[
[X
u
X
v
[
∈ S
2
:=
_
Z ∈ R
3
: [Z[ = 1
_
⊂R
3
with the unit normal vector N ⊂ R
3
of the immersion X : B →R
3
which maps into
the sphere S
2
⊂R
3
.
Let us now consider again the general situation of immersions X : B →R
n+2
.
Suppose the surface admits an ONF N which is free of torsion, i.e.
T
ϑ
σ,i
≡0 for all i = 1, 2, σ, ϑ = 1, . . . , n.
Existence of torsion free orthonormal normal frames N is strongly coupled with
a quantity representing the curvature of the normal bundle. We will introduce this
curvature tensor in the next chapter, and in chapters 8 and 9 we will attack the
problem of constructing smooth torsion free ONF.
Let (u, v) ∈ B be conformal parameters. The area of the spherical image of some
unit normal vector N
σ
can be computed from
Area(N
σ
)
2
= N
2
σ,u
N
2
σ,v
−(N
σ,u
N
σ,v
)
2
.
Inserting the Weingarten equations with zero torsion coefﬁcients yields
Area(N
σ
)
2
=
1
W
2
(L
2
σ,11
+L
2
σ,12
)(L
2
σ,12
+L
2
σ,22
) −
1
W
2
(L
σ,11
L
σ,12
+L
σ,12
L
σ,22
)
2
=
1
W
2
(L
2
σ,11
L
2
σ,22
+L
4
σ,12
−2L
σ,11
L
σ,22
L
2
σ,12
)
=
1
W
2
(L
σ,11
L
σ,22
−L
2
σ,12
)
2
.
34 2 Differential equations. Curvatures
Proposition 2.7. Let the immersion X : B→R
n+2
together with an ONF N be given.
Suppose that N is free of torsion. Then it holds
Area(N
σ
) =[K
N
σ
[W
2
for each N
σ
∈ N.
Of course, this is also true for immersions X : B →R
3
, but in general it fails in
case of higher codimensions.
2.9 Application II: Surfaces with parallel mean curvature vector
Using (2.11) together with the skewsymmetry of the torsion coefﬁcients we want
to compute the normal components of the derivative of the mean curvature vector
(denoted by the superscrip ⊥):
H
⊥
u
=
n
∑
σ=1
(H
σ,u
N
σ
+H
σ
N
σ,u
)
⊥
=
n
∑
σ=1
H
σ,u
N
σ
+
n
∑
σ,ϑ=1
H
σ
T
ϑ
σ,1
N
ϑ
=
n
∑
σ=1
_
H
σ,u
−
n
∑
ϑ=1
H
ϑ
T
ϑ
σ,1
_
N
σ
and analogously
H
⊥
v
=
n
∑
σ=1
_
H
σ,v
−
n
∑
ϑ=1
H
ϑ
T
ϑ
σ,2
_
N
σ
.
Deﬁnition 2.10. The mean curvature vector H is called parallel in the normal bun
dle if and only if
H
⊥
u
≡0 and H
⊥
v
≡0,
or equivalently
H
σ,u
i =
n
∑
ϑ=1
H
ϑ
T
ϑ
σ,i
for all i = 1, 2, σ = 1, . . . , n,
with respect to an arbitrary ONF N.
From this deﬁnition we immediately read elementary examples for surfaces with
mean curvature vector parallel in the normal bundle: Namely,
• minimal surfaces with the property H ≡0;
• surfaces which admit a torsion free ONF N satisfying T
ϑ
σ,i
≡0.
Let now H
σ
,= 0, and suppose that the mean curvature vector H is parallel in the
normal bundle.
2.9 Application II: Surfaces with parallel mean curvature vector 35
Multiplication of the ﬁrst order differential equations from the foregoing deﬁnition
by H
σ
gives
H
σ
H
σ,u
i −
n
∑
ϑ=1
H
σ
H
ϑ
T
ϑ
σ,i
= 0
for all σ = 1, . . . , n. Summation over σ shows
1
2
∂
∂u
i
[H[
2
=
n
∑
σ=1
H
σ
H
σ,u
i =
n
∑
σ=1
n
∑
ϑ=1
H
σ
H
ϑ
T
ϑ
σ,i
= 0
where the right hand side vanishes automatically due to the skewsymmetry of the
torsion coefﬁcients. Thus we conclude
[H[
2
= H
2
1
+H
2
2
+. . . +H
2
n
= const
if and only if H is parallel in the normal bundle.
In fact, skipping geometrical details, we have proved the following classiﬁcation
result due to Chen from [26].
Theorem 2.4. If H ,= 0 is parallel in the normal bundle, then either the immersion
is a minimal surface of a hypersphere of R
n+2
, or it has ﬂat normal bundle.
As it turns out in the course of our following considerations, ﬂat normal bundle is
just equivalent to the existence of a torsion free ONF N. But it is common practise
in differential geometry to speak of a ﬂat normal bundle as a geometric property
invariantly linked with the surface instead of stressing special properties of moving
normal frames.
Further Remarks
Surfaces with mean curvature vector parallel in the normal bundle in R
n+2
can be
compared with surface of constant mean curvature in R
3
. Since the theory of this
class of immersions has its own very interesting history, we want to refer some
cornerstones of this branch of differential geometry.
We abbreviatorily denote by M ⊂R
3
a twodimensional surface as a set in space.
3
Then we start with two results which go already back to Heinrich Liebmann (*1874
in Straßburg; †1939 in M¨ unchen) from 1899, see Liebmann [113], [114].
Theorem 2.5. 1. Let M ⊂R
3
be given with constant Gauss curvature K >0. Then
it holds K > 0, and the surface is a round sphere of radius
1
√
K
.
2. Let M ⊂ R
3
be given with Gauss curvature K > 0 and constant scalar mean
curvature H. Then the surface is a round sphere of radius
1
[H[
.
3
In fact, this point of view requires some theory of manifolds which we do not discuss here.
36 2 Differential equations. Curvatures
Heinz Hopf (*1894 near Breslau; †1971 in Zollikon) in [94] succeeded without
assuming the convexity K > 0.
Theorem 2.6. Let the surface M ⊂ R
3
be given with constant scalar mean curva
ture. Then M is a sphere.
The question arose whether beside the sphere there is a further compact immer
sion without boundary with constant mean curvature in R
3
. This problem is subject
to the socalled Hopf conjecture. In 1986, the American mathematician Henry Wente
in [167] proved the existence of such a constant mean curvature immersion, now of
the type of the torus.
Theorem 2.7. There is a conformal immersion of R
2
into R
3
with constant scalar
mean curvature H ,= 0 which is doublyperiodic with respect to a rectangle in R
2
.
A simpliﬁed but detailed proof of Wente’s result can be found in Abresch [1]. We
also want to refer to Glaeser and Polthier [73] for excellent numerical presentations
of compact constant mean curvature surfaces as well as references to mathematical
literature on this subject before Wente, see e.g. Dobriner [45].
There is an endless list of contemporary studies on constant mean curvature sur
faces. The reader ﬁnds various excellent contributions in the works of U. Abresch, B.
Ammann, C. Gerhardt, K. GrosseBrauckmann, F. Helein, J. Isenberg, H. Karcher,
M. Kilian, K. Kenmotsu, N. Kapouleas, R. Lopez, R. Kusner, F. Martin, W.H.
Meeks, F. Pedit, K. Polthier, N. Schmidt, J. Sullivan, M. Weber, H. Wente etc.
In 1972, David Hoffman in [92] considered the embedding problem for compact
surfaces with parallel mean curvature vector in fourdimensional Euclidean space.
Theorem 2.8. 1. Let M ⊂R
4
be given, and suppose that its mean curvature vector
H ∈ R
4
is parallel in the normal bundle. Then the surface is a round sphere of
radius
1
[H[
.
2. Let M ⊂ R
4
be given, and suppose that its Gaussian curvature K does not
change sign, and that its mean curvature vector H ∈ R
4
is parallel in the nor
mal bundle. Then M is a minimal surface, i.e. H ≡ 0, a sphere, a right circular
cylinder, or a product of circles S
1
(r) S
1
(ρ), where [H[ =
1
2
(
1
r
2
+
1
ρ
2
)
1
2
.
3. A piece of an immersed surface M ⊂R
4
, satisfying the conditions of point 2 as
well as H ,= 0, is either a piece of the round sphere, or it is ﬂat with K = 0 for its
Gaussian curvature.
Finally we want to quote the following generalization of Liebmann’s theorem to
the case of arbitrary codimension from 1985 due to Enmoto [51].
Theorem 2.9. Let M ⊂ R
n+2
be a regular, closed, simply connected, and compact
surface with Gauss curvature K > 0. Suppose that [H[ = const, and that its normal
bundle is ﬂat. Then the surface is a round sphere in a threedimensional afﬁne space
R
3
⊂R
n+2
.
Chapter 3
Integrability conditions
3.1 Integrability conditions
3.2 The CodazziMainardi equations
3.3 Gauss integrability conditions
3.4 The curvature tensor
3.5 theorema egregium
3.6 The Ricci equations
3.7 The curvature tensor of the normal bundle
3.8 The curvature of the normal bundle
3.9 Examples
3.10 The fundamental theorem
38 3 Integrability conditions
In this third chapter we derive the integrability conditions of CodazziMainardi, Ricci and the
theorema egregium from the equations of Gauss and Weingarten. In particular, we introduce the
Riemannian curvature tensor and the curvature tensor of the normal bundle. We conclude this
chapter with a general version of the fundamental theorem of surface theory.
3.1 Integrability conditions
To emphasize the special role certain partial derivative eventually are playing in our
analysis, we introduce the notation
∂
u
i X := X
u
i , i = 1, 2,
denoting the partial differentiation of X w.r.t. u
i
.
Then in view of the regularity assumptions X ∈ C
4
(B, R
n
) and N
σ
∈ C
3
(B, R
n
)
there hold necessarily
∂
uv
X
u
i −∂
vu
X
u
i ≡0, ∂
uv
N
σ
−∂
vu
N
σ
≡0 (3.1)
for all i = 1, 2 and all σ = 1, . . . , n due to an elementary theorem of H.A. Schwarz.
Evaluating these identities, taking the equations of Gauss (2.2) and Weingarten
(2.11) into account, gives the following three integrability conditions.
• Integrability conditions w.r.t. the derivatives of the surface vector X
∂
uv
X
u
i −∂
vu
X
u
i ≡0
(∂
uv
X
u
i −∂
vu
X
u
i )
norm
≡0 (∂
uv
X
u
i −∂
vu
X
u
i )
tang
≡0
CodazziMainardi equations
(Section 3.2)
theorema egregium
(Sections 3.3, 3.5)
40 3 Integrability conditions
To be precise, the Gauss equations (2.2) can be written in the form
∂
u
X
u
i =Γ
1
i1
X
u
+Γ
2
i1
X
v
+
n
∑
σ=1
L
σ,i1
N
σ
, ∂
v
X
u
i =Γ
1
i2
X
u
+Γ
2
i2
X
v
+
n
∑
σ=1
L
σ,i2
N
σ
for i =1, 2. Differentiation, while taking the Weingarten equations and the Gauss
equations into account, yields
∂
uv
X
u
i =
_
Γ
1
i1,v
+Γ
1
i1
Γ
1
12
+Γ
2
i1
Γ
1
22
−
n
∑
ω=1
L
ω,i1
L
ω,2ℓ
g
ℓ1
_
X
u
+
_
Γ
2
i1,v
+Γ
1
i1
Γ
2
12
+Γ
2
i1
Γ
2
22
−
n
∑
ω=1
L
ω,i1
L
ω,2ℓ
g
ℓ2
_
X
v
+
n
∑
σ=1
_
L
σ,i1,v
+Γ
1
i1
L
σ,12
+Γ
2
i1
L
σ,22
+
n
∑
ω=1
L
ω,i1
T
σ
ω,2
_
N
σ
,
(3.2)
as well as
∂
vu
X
u
i =
_
Γ
1
i2,u
+Γ
1
i2
Γ
1
11
+Γ
2
i2
Γ
1
12
−
n
∑
ω=1
L
ω,i2
L
ω,1ℓ
g
ℓ1
_
X
u
+
_
Γ
2
i2,u
+Γ
1
i2
Γ
2
11
+Γ
2
i2
Γ
2
12
−
n
∑
ω=1
L
ω,i2
L
ω,1ℓ
g
ℓ2
_
X
v
+
n
∑
σ=1
_
L
σ,i2,u
+Γ
1
i2
L
σ,11
+Γ
2
i2
L
σ,12
+
n
∑
ω=1
L
ω,i2
T
σ
ω,1
_
N
σ
.
(3.3)
From these identities we will derive the CodazziMainardi equations and the
theorema egregium.
• Integrability conditions w.r.t. the derivatives of unit normal vectors N
σ
∈ R
n+2
∂
uv
N
σ
−∂
vu
N
σ
≡0
(∂
uv
N
σ
−∂
vu
N
σ
)
tang
≡0 (∂
uv
N
σ
−∂
vu
N
σ
)
norm
≡0
CodazziMainardi equations
(Section 3.2)
Ricci equations
(Sections 3.6, 3.7)
3.2 The CodazziMainardi equations 41
Now we evaluate the integrability conditions for the unit normal vectors. To this
end, we ﬁrst compute
∂
uv
N
σ
= −
2
∑
j,k=1
L
σ,1 j,v
g
jk
X
u
k −
2
∑
j,k=1
L
σ,1 j
g
jk
,v
X
u
k −
2
∑
j,k=1
L
σ,1 j
g
jk
X
u
k
v
+
n
∑
ω=1
T
ω
σ,1,v
N
ω
+
n
∑
ω=1
T
ω
σ,1
N
ω,v
= −
2
∑
k=1
_
2
∑
j=1
L
σ,1 j,v
g
jk
+
2
∑
j=1
L
σ,1 j
g
jk
,v
+
2
∑
j,m=1
L
σ,1 j
g
jm
Γ
k
m2
+. . .
. . . +
2
∑
j=1
n
∑
ϑ=1
T
ϑ
σ,1
L
ϑ,2 j
g
jk
_
X
u
k
−
n
∑
ω=1
_
2
∑
j,k=1
L
σ,1 j
g
jk
L
ω,k2
−T
ω
σ,1,v
−
n
∑
ϑ=1
T
ϑ
σ,1
T
ω
ϑ,2
_
N
ω
(3.4)
and analogously
∂
vu
N
σ
= −
2
∑
k=1
_
2
∑
j=1
L
σ,2 j,u
g
jk
+L
σ,2 j
g
jk
,u
+
2
∑
j=1
2
∑
j,m=1
L
σ,2 j
g
jm
Γ
k
m1
+. . .
. . . +
2
∑
j=1
n
∑
ϑ=1
T
ϑ
σ,2
L
ϑ,1 j
g
jk
_
X
u
k
−
n
∑
ω=1
_
2
∑
j,k=1
L
σ,2 j
g
jk
L
ω,k1
−T
ω
σ,2,u
−
n
∑
ϑ=1
T
ϑ
σ,2
T
ω
ϑ,1
_
N
ω
(3.5)
using the Gauss and the Weingarten equations.
Comparing tangential and normal parts of these identities yields the socalled
CodazziMainardi equations resp. Ricci equations.
3.2 The CodazziMainardi equations
From (3.2) and (3.3) we immediately obtain the
Theorem 3.1. (CodazziMainardi integrability conditions)
Let the immersion X : B →R
n+2
together with an ONF N be given. Then the inte
grability conditions
(∂
uv
X
u
i −∂
vu
X
u
i )
norm
≡0
42 3 Integrability conditions
are the CodazziMainardi equations
∂
v
L
σ,i1
+Γ
1
i1
L
σ,12
+Γ
2
i1
L
σ,22
+
n
∑
ω=1
L
ω,i1
T
σ
ω,2
= ∂
u
L
σ,i2
+Γ
1
i2
L
σ,11
+Γ
2
i2
L
σ,12
+
n
∑
ω=1
L
ω,i2
T
σ
ω,1
(3.6)
for i = 1, 2 and σ = 1, 2, . . . , n.
The special case n = 1 of one codimension follows immediately with T
ϑ
σ,i
≡0 :
∂
v
L
i1
+Γ
1
i1
L
12
+Γ
2
i1
L
22
=∂
u
L
i2
+Γ
1
i2
L
11
+Γ
2
i2
L
12
for i = 1, 2.
These equations are named after the Italian mathematicians Gaspare Mainardi
(*1800 in Milano; †1879 in Lecco) and Delﬁno Codazzi (*1824 in Lodi; †1873 in
Pavia) who found them independently.
3.3 Gauss integrability conditions
Next, the equations (3.2) and (3.3) imply also the
Theorem 3.2. (Gauss integrability conditions)
Let the immersion X : B →R
n+2
together with an ONF N be given. Then the inte
grability conditions
(∂
uv
X
u
i −∂
vu
X
u
i )
tang
≡0
are the Gauss integrability equations
Γ
ℓ
i1,v
−Γ
ℓ
i2,u
+
2
∑
m=1
(Γ
m
i1
Γ
ℓ
m2
−Γ
m
i2
Γ
ℓ
m1
) =
2
∑
ℓ,m=1
n
∑
σ=1
(L
σ,i1
L
σ,2m
−L
σ,i2
L
σ,1m
)g
mℓ
(3.7)
for i, ℓ = 1, 2.
Note that these conditions do not involve the torsion coefﬁcients. We say they are
related to the inner geometry of a surface while the torsions are quantities arising
from the embedding of an immersion in space, i.e. they depend on the behaviour of
its normal components.
Thus the Gauss integrability conditions in case n = 1 of one codimension read
simply
Γ
ℓ
i1,v
−Γ
ℓ
i2,u
+
2
∑
m=1
(Γ
m
i1
Γ
ℓ
m2
−Γ
m
i2
Γ
ℓ
m1
) =
2
∑
ℓ,m=1
(L
i1
L
2m
−L
i2
L
1m
)g
mℓ
.
A good reference is Blaschke and Leichtweiss [15].
3.4 The curvature tensor 43
3.4 The curvature tensor
Now the left hand side of (3.7) gives rise to the following deﬁnition.
Deﬁnition 3.1. The curvature tensor R
ℓ
i jk
of the immersion X is deﬁned by
R
ℓ
i jk
:=∂
u
k Γ
ℓ
i j
−∂
u
j Γ
ℓ
ik
+
2
∑
m=1
(Γ
m
i j
Γ
ℓ
mk
−Γ
m
ik
Γ
ℓ
mj
) (3.8)
for i, j, k, ℓ = 1, 2. It is also called Riemann tensor or RiemannChristoffel tensor.
Then (3.7) can be written in the form
R
ℓ
i12
=
∑
ℓ,m=1
n
∑
σ=1
(L
σ,i1
L
σ,2m
−L
σ,i2
L
σ,1m
)g
mℓ
.
The covariant components of R
ℓ
i jk
are deﬁned as
R
ni jk
=
2
∑
ℓ=1
R
ℓ
i jk
g
ℓn
. (3.9)
Note ﬁnally that (for twodimensional surfaces) the components R
ni jk
reduce es
sentially to the one single quantity R
2112
. We particularly compute
R
1111
= 0, R
2222
= 0, R
1222
= 0, R
2111
= 0, R
2221
= 0, R
1112
= 0,
R
1122
= 0, R
2211
= 0, R
1121
= 0, R
1211
= 0, R
2212
= 0, R
2122
= 0,
R
2112
= R
1221
=−R
2121
=−R
1212
.
As we will see shortly, this component R
2112
represents exactly the Gauss curvature
K (modulo the area element). This is the contents of the theorema egregium.
Regarding higher dimensional manifolds we would eventually face a fully occu
pied curvature tensor, and commonly one deﬁnes various traces of R
ℓ
i jk
, for example
the Ricci curvature tensor
R
i j
:=
2
∑
m=1
R
m
imj
and the Ricci scalar curvature
R :=
2
∑
m,n=1
g
mn
R
mn
.
The German mathematician Bernhard Riemann (*1826 in Breselenz; †1866 in
Verbania) was the ﬁrst who introduced a conclusive concept of curvatures for geo
metric manifolds of arbitrary dimension without referring to their embedding in any
surrounding space.
44 3 Integrability conditions
3.5 theorema egregium
Consider the Riemannian curvature tensor with the special indices n = 2, i = 1,
j =1, k =2. Then from(3.7) and (3.9) we deduce the following fundamental result.
Theorem 3.3. (theorema egregium)
Let the immersion X : B →R
n+2
be given. Then it holds
R
2112
= KW
2
(3.10)
with the Gaussian curvature K of the immersion X and its area element W.
Proof. We compute
R
2112
=
2
∑
ℓ=1
R
ℓ
112
g
ℓ2
=
2
∑
ℓ,m=1
n
∑
σ=1
(L
σ,11
L
σ,2m
−L
σ,12
L
σ,1m
)g
mℓ
g
ℓ2
=
2
∑
ℓ=1
n
∑
σ=1
(L
σ,11
L
σ,22
−L
σ,12
L
σ,12
)g
2ℓ
g
ℓ2
=
n
∑
σ=1
K
σ
W
2
= KW
2
,
and the statement follows. ⊓⊔
This theorema egregium (“wonderful theorem”) states that the Gauss curvature
K can be expressed in terms of the coefﬁcients g
i j
of the surface’s ﬁrst fundamental
form and its ﬁrst and second derivatives, encoded by the Christoffel symbols.
Thus K does not depend on the embedding of X in space.
Compare this result with the deﬁnition of the Gauss curvature from section 2.4
using normal frames which actually represent the way of the surface’s embedding
in the surrounding Euclidean space!
In later applications we will use the following conformal representation of K.
Corollary 3.1. Using conformal parameters (u, v) ∈ B, the Gaussian curvature K
takes the form
K =−
1
W
∆ log
√
W (3.11)
with the area element W and the Euclidean Laplacian ∆.
Proof. Using (3.8) and (3.10) we compute
KW
2
= R
2112
=
2
∑
ℓ=1
R
ℓ
112
g
ℓ2
= R
2
112
W
=
_
Γ
2
11,v
−Γ
2
12,u
+
2
∑
m=1
Γ
m
11
Γ
2
m2
−
2
∑
m=1
Γ
m
12
Γ
2
m1
_
W.
3.7 The curvature tensor of the normal bundle 45
Thus together with the representations from (2.6) we arrive at
KW
2
=−
W
2
__
W
u
W
_
u
+
_
W
v
W
_
v
_
proving the statement. ⊓⊔
3.6 The Ricci equations
In case of one codimension, the normal components in (3.4) and (3.5), i.e.
2
∑
j,k=1
L
1 j
L
k2
g
jk
=
2
∑
j,k=1
L
2 j
L
k1
g
jk
,
are trivially satisﬁed taking the symmetry of the coefﬁcients g
i j
into account. But in
case of higher codimension we obtain the following new information.
Theorem 3.4. (Ricci integrability conditions)
Let the immersion X : B →R
n+2
together with an ONF N be given. Then the inte
grability conditions
(∂
uv
N
σ
−∂
vu
N
σ
)
norm
≡0
are the Ricci equations
T
ω
σ,2,u
−T
ω
σ,1,v
+
n
∑
ϑ=1
T
ϑ
σ,2
T
ω
ϑ,1
−
n
∑
ϑ=1
T
ϑ
σ,1
T
ω
ϑ,2
=
2
∑
j,k=1
(L
σ,2 j
L
ω,k1
−L
σ,1 j
L
ω,k2
)g
jk
(3.12)
for σ, ω = 1, . . . , n.
We immediately verify that the right hand side of this identity vanishes identically
in case n = 1, i.e.
2
∑
j,k=1
(L
2 j
L
k1
−L
1 j
L
k2
)g
jk
=
2
∑
j,k=1
(L
2 j
L
j1
−L
1 j
L
k2
)g
jk
= 0.
3.7 The curvature tensor of the normal bundle
The left hand side in (3.12) invites us to deﬁne a curvature of the normal bundle
analogously to our deﬁnition of the Riemannian curvature tensor in terms of the
Christoffel symbols
R
ℓ
i jk
=∂
u
k Γ
ℓ
i j
−∂
u
j Γ
ℓ
ik
+
2
∑
m=1
(Γ
m
i j
Γ
ℓ
mk
−Γ
m
ik
Γ
ℓ
mj
).
46 3 Integrability conditions
The normal space of a surface at w ∈ B was introduced as
N
X
(w) =
_
Z ∈ R
n+2
: Z X
u
(w) = Z X
v
(w) = 0
_
.
Deﬁnition 3.2. The curvature tensor S
ϑ
σ,i j
of the normal bundle
N
X
:=
_
w∈B
N
X
(w)
of the immersion X : B →R
n+2
is deﬁned as (notice the change of the signs)
S
ω
σ,i j
:= ∂
u
j T
ω
σ,i
−∂
u
i T
ω
σ, j
+
n
∑
ϑ=1
(T
ϑ
σ,i
T
ω
ϑ, j
−T
ϑ
σ, j
T
ω
ϑ,i
)
=
2
∑
m,n=1
(L
σ,im
L
ω, jn
−L
σ, jm
L
ω,in
)g
mn
(3.13)
for i, j = 1, 2 and σ, ω = 1, . . . , n.
The second identity in (3.13) is due to the Ricci equations.
We want to draw the reader’s attention to the skewsymmetry of the coefﬁcients
S
ω
σ,i j
w.r.t. interchanging i and j, i.e.
S
ω
σ,12
=−S
ω
σ,21
which enables us to concentrate on the components S
ω
σ,12
, and secondly to the skew
symmetry w.r.t. interchanging the indices σ and ω
S
ω
σ,12
=−S
σ
ω,12
.
It follows immediately that in case n = 2 of two codimensions there is just one
essential component, say S
2
1,12
. We faced a similar situation when we introduced the
Gaussian curvature as the only essential component of the Riemannian curvature
tensor. A detailled discussion of this fact follows in later chapters.
For the moment we want to present some elementary properties of the curvature
tensor of the normal bundle. We already know the
Corollary 3.2. The Ricci integrability conditions (3.12) from the last section can be
rewritten in the form
S
ω
σ,12
=
2
∑
m,n=1
(L
σ,1m
L
ω,n2
−L
σ,2m
L
ω,n1
)g
mn
(3.14)
for σ, ω = 1, . . . , n.
3.7 The curvature tensor of the normal bundle 47
We want to prove the tensorial transformation behaviour of the components S
ϑ
σ,12
.
Proposition 3.1. Let u
i
(v
α
) ∈ P. Then there hold
S
ω
σ,kℓ
=
2
∑
κ,λ=1
S
ω
σ,κλ
Λ
κ
k
Λ
λ
ℓ
for all k, ℓ = 1, 2 and σ, ω = 1, . . . , n.
Proof. We calculate
S
ω
σ,kℓ
=
2
∑
m,n=1
(L
σ,km
L
ω,ℓn
−L
σ,ℓm
L
ω,kn
)g
mn
=
2
∑
m,n=1
2
∑
κ,λ,µ,ν=1
(L
σ,κµ
L
ω,λν
−L
σ,λ µ
L
ω,κν
)g
µ
′
ν
′
Λ
κ
k
Λ
λ
ℓ
Λ
µ
m
Λ
ν
n
¯
Λ
m
µ
′ u
¯
Λ
n
ν
′
=
2
∑
κ,λ,µ,ν=1
(L
σ,κµ
L
ω,λν
−L
σ,λ µ
L
ω,κν
)g
µν
Λ
κ
k
Λ
λ
ℓ
,
which proves the statement. ⊓⊔
Although the components S
ω
σ,12
are not invariant w.r.t. SO(n)actions they can be
considered as sectional curvatures in the following sense.
Proposition 3.2. Let σ ,=ω. The curvature components S
ω
σ,12
do not depend on the
choice of an orthonormal basis of the sectional plane Span¦N
σ
, N
ω
¦.
We particularly infer if S
ω
σ,12
≡0 holds for one chosen ONF N then it is also true
for all ONF. This leads us to a central notion of our investigations.
Deﬁnition 3.3. The normal bundle N
X
is called ﬂat if and only there hold
S
ω
σ,12
≡0 for all σ, ω = 1, . . . , n
w.r.t. to some ONF N.
Now we come to the proof of the foregoing proposition.
Proof. We introduce conformal parameter (u, v) ∈ B. Let again
¯
N
σ
= cosϕ N
σ
+sinϕ N
ω
,
¯
N
ω
=−sinϕ N
σ
+cosϕ N
ω
,
and insert it into the representation of S
ω
σ,12
using Ricci’s integrability conditions.
48 3 Integrability conditions
Then we compute
W
¯
S
ω
σ,12
= (
¯
L
σ,11
¯
L
ω,12
−
¯
L
σ,21
¯
L
ω,11
) +(
¯
L
σ,12
¯
L
ω,22
−
¯
L
σ,22
¯
L
ω,21
)
= (cosϕ L
σ,11
+sinϕ L
ω,11
)(−sinϕ L
σ,12
+cosϕ L
ω,12
)
−(cosϕ L
σ,21
+sinϕ L
ω,21
)(−sinϕ L
σ,11
+cosϕ L
ω,11
)
+(cosϕ L
σ,12
+sinϕ L
ω,12
)(−sinϕ L
σ,22
+cosϕ L
ω,22
)
−(cosϕ L
σ,22
+sinϕ L
ω,22
)(−sinϕ L
σ,21
+cosϕ L
ω,21
)
= (L
σ,11
−L
σ,22
)L
ω,12
−(L
ω,11
−L
ω,22
)L
σ,12
= WS
ω
σ,12
,
which proves the statement. ⊓⊔
The next result reveals an interesting connection between the curvature compo
nents S
ϑ
σ,12
and normal mean curvature quantities H
σϑ
introduced in section 2.5 of
the previous chapter.
Theorem 3.5. There hold
S
ϑ
σ,12
≡
1
2
H
σϑ
W (3.15)
for σ, ϑ = 1, . . . , n.
Proof. Recall the deﬁnition of the coefﬁcients L
σϑ,i j
from Deﬁnition 2.6. Using
conformal parameters we calculate
H
σϑ
=
1
W
2
(L
σϑ,11
+L
σϑ,22
)
=
2
W
2
(L
σ,11
L
ϑ,12
−L
σ,12
L
ϑ,11
+L
σ,12
L
ϑ,22
−L
σ,22
L
ϑ,12
) =
2
W
S
ϑ
σ,12
proving the identity. ⊓⊔
3.8 The curvature of the normal bundle
Let u
i
= u
i
(v
α
) ∈ P. Then Proposition 3.1 yields
S
ϑ
σ,12
(u
1
, u
2
) = S
ϑ
σ,12
(v
1
, v
2
)(Λ
1
1
Λ
2
2
−Λ
2
1
Λ
1
2
).
Thus taking account of transformation behaviour of the area element W(u
1
, u
2
) from
Corollary 1.1 we conclude that the quantities
1
W(u
1
, u
2
)
S
ϑ
σ,12
(u
1
, u
2
)
are independent of the chosen parametrization.
3.9 Examples 49
This leads us to our next
Deﬁnition 3.4. Let the immersion X : B →R
4
be given. Then the scalar curvature
of its normal bundle is deﬁned as
S :=
1
W
S
2
1,12
.
Note, as mentioned above, that S
2
1,12
is the only essential component of the cur
vature tensor if n = 2. In case n > 2 of higher codimensions we would like to work
with a curvature vector instead of a single curvature scalar.
Deﬁnition 3.5. Let the immersion X : B→R
n+2
be given. Then the curvature vector
of its normal bundle is
S =
1
W
(S
2
1,12
, S
3
1,12
, . . . , S
n
1,12
, S
3
2,12
, . . . , S
n
n−1,12
) ∈ R
N
with N =
_
n
2
_
.
Let us conclude this section with two remarks.
1. Although all informations of the curvature tensor of the normal bundle are al
ready decoded in the normal mean curvature matrix H, we ﬁnd it more natural to
work with the vector S ∈ R
N
.
2. Worthwile for the future but omitted here is an elaborative investigation of im
mersions with prescribed curvature vector of the normal bundle.
3.9 Examples
We want to consider some interesting examples of surfaces with ﬂat and with non
vanishing normal bundles.
Spherical surfaces in S
3
Suppose [X(u, v)[ ≡1 for all (u, v) ∈ B. We immediately compute
X
u
X = 0, X
v
X = 0,
i.e. the surface vector X itself serves as a unit normal vector, say N
1
:= X. Choose
then N
2
to complete the system ¦X
u
, X
v
, N
1
¦ to a basis of the embedding space R
4
.
We immediately verify that the surface has ﬂat normal bundle since
T
2
1,1
= N
1,u
N
2
= X
u
N
2
= 0, T
2
1,2
= N
1,v
N
2
= X
v
N
2
= 0.
50 3 Integrability conditions
The Clifford torus
This is the surface
X =
1
√
2
(cosu, sinu, cosv, sinv)
resembling S
1
S
1
. We assign the moving 4frame ¦X
u
, X
v
, N
1
, N
2
¦ consisting of the
tangential vectors
X
u
=
1
√
2
(−sinu, cosu, 0, 0), X
v
=
1
√
2
(0, 0, −sinv, cosv)
as well as two unit normal vectors
N
1
:=
1
√
2
(cosu, sinu, cosv, sinv), N
2
:=
1
√
2
(−cosu, −sinu, cosv, sinv).
Obviously the torsion coefﬁcients of this ONF vanish identically, thus the normal
bundle of this immersion is ﬂat; see e.g. do Carmo [22] for this example.
For explicit constructions of surfaces with ﬂat normal bundles we want to refer
the reader to Ferapontov [57].
Parallel type surfaces
Consider now the normal transport
R(u, v) = X(u, v) + f (u, v)N
1
(u, v) +g(u, v)N
2
(u, v).
If the functions f and g are constant then we say R is the parallel surface of X
and vice versa, at least if the surfaces are immersed in R
3
. Parallelity in higher
codimensional space depends on the curvature S of the normal bundle.
Proposition 3.3. The normal transport R of an immersion X : B →R
4
is parallel,
i.e. there holds
R
u
i N
σ
= 0 for i = 1, 2, σ = 1, 2,
if and only if it holds S ≡0 for the scalar curvature S of the normal bundle.
Proof. For the proof we use the Weingarten equations and compute the normal parts
R
⊥
u
and R
⊥
v
of the tangential vectors R
u
resp. R
v
,
R
⊥
u
= f
u
N
1
+g
u
N
2
+ f N
⊥
1,u
+gN
⊥
2,u
= ( f
u
−gT
2
1,1
_
N
1
+
_
g
u
+ f T
2
1,1
_
N
2
,
R
⊥
v
= f
v
N
1
+g
v
N
2
+ f N
⊥
1,v
+gN
⊥
2,v
=
_
f
v
−gT
2
1,2
_
N
1
+
_
g
v
+ f T
2
1,2
_
N
2
.
The condition of parallelity leads us to the ﬁrst order system
f
u
−gT
2
1,1
= 0, f
v
−gT
2
1,2
= 0, g
u
+ f T
2
1,1
= 0, g
v
+ f T
2
1,2
= 0.
3.10 The fundamental theorem 51
Differenting the ﬁrst two equations and making use of the other two conditions
shows us
0= f
uv
−g
v
T
2
1,1
−gT
2
1,1,v
= f
uv
+ f T
2
1,1
T
2
1,2
−gT
2
1,1,v
,
0= f
vu
−g
u
T
2
1,2
−gT
2
1,2,u
= f
vu
+ f T
2
1,1
T
2
1,2
−gT
2
1,2,u
,
thus a comparison of the right hand sides brings
0 =−gT
2
1,1,v
+gT
2
1,2,u
=−g SW.
Similarly we ﬁnd 0 = f SW, which proves the statement. ⊓⊔
Parallel type surface are widely used in geometry and mathematical physics. We
want to refer the reader to da Costa [37] for an application in quantum mechanics in
curved spaces.
3.10 The fundamental theorem
We want to reconstruct an immersion X from given ﬁrst and second fundamental
forms, given torsion coefﬁcients, and a given (n +2)frame attached at some point
of the surface, say at (0, 0) ∈ B. The latter assumption is needed to construct an
initial (n +2)frame at an arbitrary point of the surface.
In particular, it is needed in the second point of our proof below: Assume we
know that the tangential planes at a certain point w ∈ B of two solutions X and
¯
X
coincide, and therefore the normal spaces are the same. If n =1, we are then allowed
to infer that the unit normal vectors also coincide (up to orientation).
But the situation is more involved in general: Even if the normal spaces of X and
¯
X agree at some point, we have no further information about the behaviour of the
respective normal frames. We will ﬁx this problem under this special assumption.
Theorem 3.6. (Fundamental theorem of surface theory)
Assumptions: Let us given
1. a quadratic, symmetric, and positive deﬁnite form ´ g
i j
∈C
3
(B, R), such that
´ g
i j
= ´ g
ji
, det (´ g
i j
)
i, j=1,2
> 0; (3.16)
2. n symmetric and quadratic forms
´
L
σ,i j
∈C
2
(B, R), such that
´
L
σ,i j
=
´
L
σ, ji
(3.17)
for i, j = 1, 2 and σ = 1, . . . , n;
52 3 Integrability conditions
3. and
_
n
2
_
forms
´
T
ϑ
σ,i
∈C
2
(B, R) such that
´
T
ϑ
σ,i
=−
´
T
ϑ
σ,i
(3.18)
for i = 1, 2 and σ, ϑ = 1, . . . , n.
Assume that these forms satisfy the integrability conditions (3.6), (3.7), and (3.12).
4. Finally, let n +2 vectors
˚
Z
(1)
,
˚
Z
(2)
, and
˚
N
1
, . . . ,
˚
N
n
be given with the properties
˚
Z
(i)
˚
Z
( j)
= ´ g
i j
(0, 0),
˚
Z
(i)
˚
N
σ
= 0,
˚
N
σ
˚
N
ϑ
=δ
σϑ
(3.19)
for i = 1, 2 and σ, ϑ = 1, . . . , n.
Statement: Then there is a unique immersion X ∈C
3
(B, R
n+2
) with an unique ONF
N = (N
1
, . . . , N
n
) such that
X(0, 0) =
˚
X, X
u
i (0, 0) =
˚
Z
(i)
, N
σ
(0, 0) =
˚
N
σ
.
The g
i j
≡´ g
i j
agrees with the ﬁrst fundamental formof the surface X, the L
σ,i j
≡
´
L
σ,i j
are the coefﬁcients of its second fundamental forms w.r.t. to the ONF N, and the
T
ϑ
σ,i
≡
´
T
ϑ
σ,i
represent the respective torsion coefﬁcients.
Our proof of this theorem follows the lines of Blaschke and Leichtweiss [15].
Preceded is the following lemma (see also ibidem, § 60).
Lemma 3.1. Consider the initial value problem
∂z
k
∂u
i
=
m
∑
ℓ=1
a
ℓ
ki
z
ℓ
, z
k
(u
0
, v
0
) = ˚ z
k
, i = 1, 2, k = 1, . . . , m, (3.20)
with m ≥ 1 linear partial differential equations for the unknowns z
ℓ
on a right
angled domain G : [u
i
−u
i
0
[ <b
i
, i =1, 2. Assume that a
ℓ
ki
∈C
2
(G, R) for all i =1, 2
and k, ℓ = 1, . . . , m. Then the following hold true:
1. There is at most one solution vector (z
1
, . . . , z
m
) ∈C
3
(G, R
m
) of the initial value
problem (3.20).
2. The system of the linear equations is solvable if and only if the integrability
conditions
∂
2
z
k
∂u
i
∂u
j
−
∂
2
z
k
∂u
j
∂u
i
≡0
are satisﬁed for all k = 1, . . . , m.
3.10 The fundamental theorem 53
We particularly compute
0 ≡
∂
∂u
j
m
∑
ℓ=1
a
j
ki
z
ℓ
−
∂
∂u
i
m
∑
ℓ=1
a
ℓ
k j
z
ℓ
=
m
∑
ℓ=1
_
∂a
ℓ
ki
∂u
j
−
∂a
ℓ
k j
∂u
i
+
m
∑
s=1
a
s
ki
a
ℓ
s j
−
m
∑
s=1
a
s
k j
a
ℓ
si
_
z
ℓ
.
Thus the system is solvable if there hold the integrability conditions
∂a
ℓ
ki
∂u
j
+
m
∑
s=1
a
s
ki
a
ℓ
s j
=
∂a
ℓ
k j
∂u
i
+
m
∑
s=1
a
s
k j
a
ℓ
si
for all k, ℓ = 1, . . . , m.
Proof of the theorem. 1. Rewriting the differential equations (2.2) and (2.11) of
Gauss and Weingarten we arrive at the linear system
X
u
i = Z
(i)
,
Z
(i),u
j =
2
∑
k=1
Γ
k
i j
Z
(k)
+
n
∑
ϑ=1
L
ϑ,i j
N
ϑ
,
N
σ,u
i = −
2
∑
j,k=1
L
σ,i j
g
jk
Z
(k)
+
n
∑
ϑ=1
T
ϑ
σ,i
N
ϑ
(3.21)
for the unknown vector functions X, Z
(1)
, Z
(2)
, N
1
, N
2
, . . . , N
n
.
2. Assume that there are two immersed solutions X and
¯
X. At (0, 0) ∈ B there hold
¯
X
u
i (0, 0)
¯
X
u
j (0, 0) = g
i j
(0, 0) = X
u
i (0, 0) X
u
j (0, 0) for i, j = 1, 2.
After a suitable translation and rotation of
¯
X we can arrange the geometry so that
¯
X(0, 0) = X(0, 0),
¯
X
u
i (0, 0) = X
u
i (0, 0) for i = 1, 2.
Together with (3.19) we conclude that in (0, 0) the tangential vectors, the ONF N
and ONF
¯
N as well as their orientations agree. Lemma 3.1 proves the uniqueness.
3. To prove that a solution X of (3.21) is indeed immersed with the prescribed forms
(3.16), (3.17), (3.18) and the given (n +2)frame (3.19) in (0, 0) ∈ B, we will
show that the following functions vanish identically in B (by assumption they
vanish at (0, 0) ∈ B)
f
(1)
i j
:=Z
(i)
Z
( j)
−´ g
i j
for i, j = 1, 2,
f
(2)
iσ
:=Z
(i)
N
σ
for i = 1, 2, σ = 1, . . . , n,
f
(3)
σ
:=N
2
σ
−1 for σ = 1, . . . , n,
f
(4)
σϑ
:=N
σ
N
ϑ
for σ, ϑ = 1, . . . , n, σ ,=ϑ.
54 3 Integrability conditions
4. For this purpose we establish a linear system for all these functions. First, (3.21)
yields
_
Z
(i)
Z
( j)
−´ g
i j
_
u
k
= Z
(i),u
k Z
( j)
+Z
( j),u
k Z
(i)
−´ g
i j,u
k
=
2
∑
m=1
´
Γ
m
ik
Z
(m)
Z
( j)
+
2
∑
m=1
´
Γ
m
jk
Z
(m)
Z
(i)
−´ g
i j,u
k
+
n
∑
ϑ=1
´
L
ϑ,ik
_
N
ϑ
Z
( j)
_
+
n
∑
ϑ=1
´
L
ϑ, jk
_
N
ϑ
Z
(i)
_
=
2
∑
m=1
´
Γ
m
ik
_
Z
(m)
Z
( j)
−´ g
mj
_
+
2
∑
m=1
´
Γ
m
jk
_
Z
(m)
Z
(i)
−´ g
mi
_
+
n
∑
ϑ=1
´
L
ϑ,ik
_
Z
( j)
N
ϑ
_
+
n
∑
ϑ=1
´
L
ϑ, jk
_
Z
(i)
N
ϑ
_
(3.22)
taking
2
∑
m=1
´
Γ
m
ik
´ g
mj
+
2
∑
m=1
´
Γ
m
jk
´ g
mi
=
1
2
2
∑
m,n=1
´ g
mn
(´ g
ni,u
k +´ g
kn,u
i −´ g
ik,u
n)´ g
mj
+
1
2
2
∑
m,n=1
´ g
mn
(´ g
n j,u
k +´ g
kn,u
j −´ g
jk,u
n)´ g
mi
=
1
2
(´ g
i j,u
k +´ g
k j,u
i −´ g
ik,u
j ) +
1
2
(´ g
i j,u
k +´ g
ki,u
j −´ g
jk,u
i )
= ´ g
i j,u
k
into account. The functions Z
(i)
N
σ
satisfy
_
Z
(i)
N
σ
_
u
k
= Z
(i),u
k N
σ
+Z
(i)
N
σ,u
k
=
´
Γ
m
ik
_
Z
(m)
N
σ
_
+
´
L
σ,ik
N
2
σ
−
´
L
σ,km
´ g
mn
_
Z
(i)
Z
(n)
_
+
n
∑
ϑ=1
σ,=ϑ
´
L
ϑ,ik
_
N
σ
N
ϑ
_
+
n
∑
ϑ=1
´
T
ϑ
σ,k
_
Z
(i)
N
ϑ
_
=
´
Γ
m
ik
_
Z
(m)
N
σ
_
+
´
L
σ,ik
_
N
2
σ
−1
_
−
´
L
σ,km
´ g
mn
_
Z
(i)
Z
(n)
−´ g
in
_
+
n
∑
ϑ=1
σ,=ϑ
´
L
ϑ,ik
_
N
σ
N
ϑ
_
+
n
∑
ϑ=1
´
T
ϑ
σ,k
_
Z
(i)
N
ϑ
_
(3.23)
3.10 The fundamental theorem 55
due to the property
2
∑
m,n=1
´
L
σ,km
´ g
mn
´ g
ni
=
2
∑
m=1
´
L
σ,km
δ
m
i
=
´
L
σ,ki
.
Next we have
_
N
2
σ
−1
_
u
k
= 2N
σ
N
σ,u
k
= −2
2
∑
m,n=1
´
L
σ,km
´ g
mn
_
Z
(n)
N
σ
_
+2
n
∑
ϑ=1
´
T
ϑ
σ,k
_
N
σ
N
ϑ
_
.
(3.24)
Finally we calculate
_
N
σ
N
ϑ
_
u
k
= N
σ,u
k N
ϑ
+N
σ
N
ϑ,u
k
= −
2
∑
m,n=1
´
L
σ,km
´ g
mn
_
Z
(n)
N
ϑ
_
−
2
∑
m,n=1
´
L
ϑ,km
´ g
mn
_
Z
(n)
N
σ
_
+
n
∑
ω=1
´
T
ω
σ,k
_
N
ω
N
ϑ
_
+
n
∑
ω=1
´
T
ω
ϑ,k
_
N
ω
N
σ
_
.
(3.25)
6. Summarizing (3.22), (3.23), (3.24), (3.25) we arrive at the linear system
f
(1)
i j,u
k
=
2
∑
m=1
´
Γ
m
ik
f
(1)
mj
+
2
∑
m=1
´
Γ
m
jk
f
(1)
mi
+
n
∑
ϑ=1
´
L
ϑ,ik
f
(2)
jϑ
+
n
∑
ϑ=1
´
L
ϑ, jk
f
(2)
iϑ
,
f
(2)
iσ,u
k
=
2
∑
m=1
´
Γ
m
ik
f
(2)
mσ
+
´
L
σ,ik
f
(3)
σ
−
2
∑
m,n=1
´
L
σ,km
´ g
mn
f
(1)
in
+
n
∑
ϑ=1
σ,=ϑ
´
L
ϑ,ik
f
(4)
σϑ
+
n
∑
ϑ=1
´
T
ϑ
σ,k
f
(2)
iϑ
,
f
(3)
σ,u
k
= −2
2
∑
m,n=1
´
L
σ,km
´ g
mn
f
(2)
nσ
+
n
∑
ϑ=1
σ,=ϑ
´
T
ϑ
σ,k
f
(4)
σϑ
,
f
(4)
σϑ,u
k
= −
2
∑
m,n=1
´
L
σ,km
´ g
mn
f
(2)
nϑ
−
2
∑
m,n=1
´
L
ϑ,km
´ g
mn
f
(2)
nσ
+
n
∑
ω=1
ω,=ϑ
´
T
ω
σ,k
f
(4)
ωϑ
+
n
∑
ω=1
ω,=ϑ
´
T
ω
ϑ,k
f
(4)
ωσ
with the same initial conditions as the trivial solution 0, . . . , 0. Thus the unique
ness stated in Lemma 3.1 proves that all the functions f
(1)
i j
, f
(2)
iσ
, f
(3)
σ
and f
(4)
σϑ
vanish identically in B.
56 3 Integrability conditions
7. We complete our proof of the fundamental theorem as follows:
• Due to (3.21), (3.22) and the property
2
∑
i=1
α
i
X
u
i = 0 implies α
1
=α
2
= 0
for arbitrary functions α
1
, α
2
since
_
2
∑
i=1
α
i
X
u
i
_
_
2
∑
j=1
α
j
X
u
j
_
=
2
∑
i, j=1
α
i
α
j
g
i j
= 0,
the mapping X ∈ C
3
(B, R
n+2
) represents a twodimensional immersion in
R
n+2
with ﬁrst fundamental form g
i j
≡ ´ g
i j
.
• Following (3.23), (3.24), (3.25), the moving frame N forms an orthonormal
normal frame which is orthogonal to the span of X
u
and X
v
.
• Therefore, X
u
i
u
j N
σ
=
´
L
ϑ,i j
for all i, j =1, 2 and all σ =1, . . . , n, and the
´
L
σ,i j
are detected as the coefﬁcients of the second fundamental forms w.r.t. N.
• Analogously we prove that the
´
T
ϑ
σ,i
agree with the torsion coefﬁcients w.r.t.
the ONF N.
The proof is complete. ⊓⊔
This result was already proved in Weyl [168]. We want to refer the reader also to
Brauner [18].
Chapter 4
Weighted differential geometry
4.1 Introduction
4.2 The weighted fundamental forms
4.3 Differential equations
4.4 The weighted mean curvature system
4.5 The CodazziMainardi equations
58 4 Weighted differential geometry
In this chapter we extend the previous investigations and consider weighted fundamental forms of
Finslerian type, the natural differential geometric setup for the calculus of variations.
4.1 Introduction
Consider an immersion X : B →R
3
in threedimensional Euclidean space with unit
normal vector N ∈ R
3
. We suppose that its ﬁrst fundamental form, now denoted by
(h
i j
)
i, j=1,2
∈ R
22
to distinguish it fromour investigations so far, results from the action of a symmetric
and positive deﬁnite matrix
W(X, Z) =
_
w
i j
(X, Z)
_
i, j=1,2,3
: R
3
R
3
¸ ¦0¦ −→R
33
in the following way:
h
i j
:= X
u
i ◦ W(X, N) ◦ X
u
j , i, j = 1, 2.
In particular, if W(X, Z) ≡ E
3
with the threedimensional unit matrix E
3
, then
h
i j
= g
i j
with the coefﬁcients
g
i j
= X
u
i X
u
j , i, j = 1, 2
of the classical ﬁrst fundamental from the previous chapters.
So far as we know, weighted fundamental forms h
i j
of this special kind where ﬁrst
introduced by Sauvigny [141] for analytical studies of critical points X : B →R
3
for
twodimensional parametric and anisotropic variational problems
F[X] :=
__
B
F(X
u
X
v
)dudv −→min!
Critical points of F[X] are called Fminimal surfaces. In many respects they
behave like ordinary minimal surfaces if certain classical geometric quantities are
replaced by their weighted counterparts as demonstrated in chapter 11.
60 4 Weighted differential geometry
As we expound later, such weighted geometric quantities, like the quadratic
form (h
i j
)
i, j=1,2
and various curvatures derived from it, furnish the natural setup
for anisotropic variational problems.
There is a strong coincidence with the socalled Finsler spaces from the cal
culus of variations, named after the Swiss mathematician Paul Finsler (*1894 in
Heilbronn; †1970 in Z¨ urich). Excellent introductions to this matter can be found in
Finsler’s thesis [59], or in the textbook of Funk [68].
In this chapter we want to develop an approach to a weighted differential geom
etry for surfaces immersed in R
3
and R
n+2
. For illustration consider the weighted
line element of the form
ds
2
W
:=
3
∑
i, j=1
w
i j
(X, Z)dx
i
dx
j
.
If the w
i j
depend only on the space point X then the inhomogeneous metric
ds
2
W
is called of Riemannian type. More generally, if ds
2
W
is inhomogeneous and
anisotropic, i.e. the w
i j
(X, Z) depend additionally on a direction vector Z attached
at the point X, we say ds
2
W
is of Finslerian type.
Consider now the space R
3
equipped with such an inhomogeneous and aniso
tropic metric. Inserting the surface’s representation X = X(u, v) and its unit normal
vector N = N(u, v) into the form ds
2
W
yields
ds
2
W
=
2
∑
i, j=1
w
i j
(X, N)dx
i
dx
j
=
2
∑
i, j=1
2
∑
k,ℓ=1
w
i j
(X, N)x
i
u
k
x
j
u
ℓ
du
k
du
ℓ
=
2
∑
k,ℓ=1
h
kℓ
du
k
du
ℓ
setting
h
kℓ
:=
2
∑
i, j=1
X
u
i ◦ W(X, N) ◦ X
u
j , k, ℓ = 1, 2,
and W(X, Z) = (w
i j
(X, Z))
i, j=1,2,3
.
But these h
i j
are exactly the coefﬁcients of the weighted ﬁrst fundamental form
from the beginning of this introduction!
It remains to remark that in the special case W(X, Z) ≡E
3
we obviously retrieve
the classical metrical form (g
i j
)
i, j=1,2
.
4.2 The weighted fundamental forms 61
4.2 The weighted fundamental forms
Consider an immersion X : B →R
n+2
with ONF N. We introduce a symmetric and
positive deﬁnite matrix
W(X, Z) = (w
i, j
)
i, j=1,...,n+2
of class C
3
(R
n+2
R
n+2
¸ ¦0¦ . . . R
n+2
¸ ¦0¦, R
n+2
R
n+2
)
(4.1)
satisfying the properties:
Let Z = (Z
1
, . . . , Z
n
) and
¯
Z = (
¯
Z
1
, . . . ,
¯
Z) be some orthogonal frames in the normal
space of X, not necessarily consisting of unit vectors, then suppose
(W1) W(X, Z) is invariant w.r.t. the choise of the normal frame, i.e. it holds
W(X, Z) = W(X,
¯
Z) for all ONF Z and
¯
Z;
(W2) W(X, Z) acts nontrivially only on the tangent space, i.e. there hold
W(X, Z)
¸
¸
T
X
(w)
: T
X
(w) −→T
X
(w), in particular,
rankW(X, Z)
¸
¸
T
X
(w)
= 2
as well as
W(X, Z) ◦ Z
σ
= Z
σ
for all σ = 1, 2, . . . , n;
(W3) W(X, Z) is positive deﬁnite, i.e. with a real constant ω
0
∈ [0, +∞) it holds
(1 +ω
0
)
−1
[ξ[
2
≤ξ ◦ W(X, Z) ◦ ξ ≤(1 +ω
0
)[ξ[
2
for all ξ ∈ R
n+2
;
(W4) W(X, Z) is normalized in the following sense
W(X, Z) = 1 for all X, Z.
Weight matrices were ﬁrst introduced in Sauvigny [141] for a new representation
of critical points for anisotropic variational problems in R
3
in parametric form.
Sauvigny’s methods were further developed and applied e.g. in Bergner und Dit
trich [12], Clarenz [30], [31], Clarenz and von der Mosel [32], Fr¨ ohlich [62], [64],
and Winklmann [172], [173] to inhomogeneous and anisotropic variational prob
lems. In chapter 11 we will present some of those results.
Deﬁnition 4.1. A matrix W(X, Z) with the properties (W1) to (W4) is called a
weight matrix.
By means of such a weight matrix we are able to create a differential geometry
of Finslerian type. For this purpose we start with the
62 4 Weighted differential geometry
Deﬁnition 4.2. The weighted ﬁrst fundamental form I
W
(X) ∈ R
2
R
2
of the im
mersion X : B →R
n+2
with ONF N is deﬁned as
I
W
(X) = (h
i j
)
i, j=1,2
, h
i j
:= X
u
i ◦ W(X, N) ◦ X
u
j . (4.2)
Note that h
i j
≡g
i j
with the coefﬁcients g
i j
of the classical ﬁrst fundamental form
if W(X, Z) ≡E
n+2
with the (n +2)dimensional unit matrix E
n+2
.
We furthermore remark that the weighted ﬁrst fundamental I
W
(X) of an immer
sion X : B →R
n+2
is symmetric and positive deﬁnite since the matrix W(X, Z) is
then symmetric and positive deﬁnite.
Deﬁnition 4.3. The weighted second and the weighted third fundamental forms of
the immersion X : B →R
n+2
are deﬁned as
II
W,N
σ
(X) = (L
σ,i j
)
i, j=1,2
, L
σ,i j
:=−X
u
i N
σ,u
j = X
u
i
u
j N
σ
,
III
W,N
σ
(X) = ( f
σ,i j
)
i, j=1,2
, f
N
σ
,i j
:= N
σ,u
i ◦ W(X, N)
−1
◦ N
σ,u
j .
(4.3)
with a unit normal vector N
σ
∈ R
n+2
of an ONF N.
With these deﬁnitions of the weighted second and third fundamental forms we
essentially follow Sauvigny [141] and Fr¨ ohlich [64]. Alternative ways were en
couraged by Clarenz [31] in case n = 1 and by Winklmann [172], [173] for m
dimensional manifolds in R
m+1
.
Consider that, in contrast to the case n = 1, the vector N
σ,u
i is not necessarily
tangential due to the presence of torsion coefﬁcients. A further deﬁnition thus also
commonly used only involves the tangential parts of N
σ,u
i .
To illustrate this fact consider an unit normal vector N
σ
= N with its derivatives
splitted up into tangential and normal parts
N
σ,u
i = N
⊤
σ,u
i
+N
⊥
σ,u
i
,
the latter vanishing identically if n = 1. We compute
N
σ,u
i ◦ W(X, N)
−1
◦ N
σ,u
j
= (N
⊥
σ,u
i
+N
⊤
σ,u
i
) ◦ W(X, N)
−1
◦ (N
⊥
σ,u
j
+N
⊤
σ,u
j
)
= N
⊥
σ,u
i
◦ W(X, N)
−1
◦ N
⊥
σ,u
j
+N
⊥
σ,u
i
◦ W(X, N)
−1
◦ N
⊤
σ,u
j
+N
⊤
σ,u
i
◦ W(X, N)
−1
◦ N
⊥
σ,u
j
+N
⊤
σ,u
i
◦ W(X, N)
−1
◦ N
⊤
σ,u
j
= N
⊤
σ,u
i
◦ W(X, N)
−1
◦ N
⊤
σ,u
j
+N
⊥
σ,u
i
N
⊥
σ,u
j
due to W(X, N)
−1
◦ N
⊤
σ,u
j
= N
⊤
σ,u
j
as well as
N
⊥
σ,u
i
◦ W(X, N)
−1
◦ N
⊤
σ,u
j
= 0, N
⊤
σ,u
i
◦ W(X, N)
−1
◦ N
⊥
σ,u
j
= 0.
4.3 Differential equations 63
In particular, if the ONF N is free of torsion then
III
W,N
σ
(X) ≡
´
III
W,N
σ
(X) = (
´
f
σ,i j
)
i, j=1,2
with
´
f
σ,i j
:= N
⊤
u
i
◦ W(X, N)
−1
◦ N
⊤
σ,u
j
.
(4.4)
It is the form
´
III
W
(X) which is often introduced as a third fundamental form, see
e.g. Brauner’s textbook [18] for an elaboration of the nonweighted case.
Finally we want to remark that property (W4) particularly ensures that the clas
sical area element W equals its weighted counterpart W
W
in the following sense
W
W
:= det (DX
T
◦ W(X, N) ◦ DX) = det (DX
T
◦ DX) =W (4.5)
with the Jacobian DX operating on the tangent plane of the surface.
Thus we do not need to distinguish between a “weighted” and a “nonweighted”
area element. The same holds true for the Gaussian curvature, but as we will see
later, it is necessary to introduce a weighted mean curvature.
4.3 Differential equations
The equations of Gauss and Weingarten reﬂect the way of representation of the
derivatives X
u
i
u
j and N
σ,u
i in terms of a moving frame
_
X
u
, X
v
, N
1
, . . . , N
n
_
.
In chapter 2 we derived these equations for the case W(X, Z) ≡ E
n+2
. Now we
want to present its counterparts in the weighted setup, but we will omit the proofs
which actually work as in the classical case.
Let that a weight matrix W(X, Z) ∈ R
(n+2)(n+2)
be given.
Theorem 4.1. (Gauss equations in weighted form)
Let the immersion X : B →R
n+2
together with an ONF N be given. Then it holds
X
u
i
u
j =
2
∑
k=1
(Γ
k
i j
+Ω
k
i j
)X
u
k +
n
∑
ϑ=1
L
ϑ,i j
N
ϑ
(4.6)
for all i, j = 1, 2, with the Christoffel symbols
Γ
k
i j
:=
2
∑
ℓ=1
1
2
h
kℓ
(h
ℓi,u
j +h
jℓ,u
i −h
i j,u
ℓ ) (4.7)
and the corrective terms
Ω
k
i j
:=−
1
2
2
∑
ℓ=1
h
kℓ
(ω
ℓi j
+ω
jℓi
−ω
i jℓ
), ω
i jℓ
:= X
u
i ◦ W(X, N)
u
ℓ ◦ X
u
j . (4.8)
64 4 Weighted differential geometry
We remark
W(X, N)
u
ℓ = W
X
(X, N) ◦ X
u
ℓ +W
Z
(X, N) ◦ N
u
ℓ
with the settings
W
X
(X, N) ◦ X
u
ℓ =
_
n+2
∑
k=1
w
i j,x
k x
k
u
ℓ
_
i, j=1,...,n+2
∈ R
(n+2)(n+2)
etc.
Theorem 4.2. (Weingarten equations in weighted form)
Let the immersion X : B →R
n+2
together with an ONF N be given. Then it holds
N
σ,u
i =−
2
∑
j,k=1
L
σ,i j
h
jk
W(X, N) ◦ X
u
k +
n
∑
ϑ=1
T
ϑ
σ,i
N
ϑ
(4.9)
for all i = 1, 2, σ = 1, . . . , n.
4.4 The weighted mean curvature system
Due to the positive deﬁniteness and symmetry of W(X, Z), the line element
ds
2
W
=
2
∑
i, j=1
h
i j
du
i
du
j
is of Riemannian type. Thus we may introduce weighted conformal parameters
(u, v) ∈ B satisfying
X
u
◦ W(X, N) ◦ X
v
= W = X
v
◦ W(X, N) ◦ X
v
,
X
u
◦ W(X, N) ◦ X
v
= 0 in B
(4.10)
with the area element W since W
W
= W following our discussion above. We also
refer to Sauvigny [143], chapter XII, §8.
Then the Christoffel symbols Γ
k
i j
from(4.7) w.r.t. the coefﬁcients h
i j
satisfy again
the relations (2.6). In particular, we infer
Γ
1
11
+Γ
1
22
= 0, Γ
2
11
+Γ
2
22
= 0.
Now we can evaluate the Gauss equations (4.6) to get the mean curvature system
in weighted form generalizing the classical system
∆X = 2HW in B
for the conformally parametrized immersion X.
4.4 The weighted mean curvature system 65
Corollary 4.1. Let the weighted conformally parametrized immersion X : B→R
n+2
together with an ONF N be given. Then it holds
∆X = (Ω
1
11
+Ω
1
22
)X
u
+(Ω
2
11
+Ω
2
22
)X
v
+2
n
∑
ϑ=1
H
W,ϑ
WN
ϑ
= (Ω
1
11
+Ω
1
22
)X
u
+(Ω
2
11
+Ω
2
22
)X
v
+2H
W
W
(4.11)
with the weighted mean curvature vector
H
W
:=
n
∑
ϑ=1
H
W,ϑ
N
ϑ
, H
W,ϑ
:=
2
∑
i, j=1
h
i j
L
i j
. (4.12)
Note that (4.11) is a coupled elliptic system for the surface vector X and the unit
normal vectors N
σ
of some ONF N and their derivatives.
Deﬁnition 4.4. The immersion X : B →R
n+2
is called a weighted minimal surface
w.r.t to the weight matrix W(X, Z) if it holds
H
W
≡0 in B.
Thus minimal surfaces with the property H ≡ 0 are special weighted minimal
surfaces in the classical case W(X, Z) ≡E
n+2
.
Deﬁnition 4.5. We deﬁne the weighted Laplacian ∆
W
w.r.t. a weight matrix W(X, Z)
as the elliptic operator
∆
W
:=∆ −(Ω
1
11
+Ω
1
22
)
∂
∂u
−(Ω
2
11
+Ω
2
22
)
∂
∂v
(4.13)
with the Euclidean Laplacian ∆.
Then the weighted conformally parametrized immersion X satisﬁes
∆
W
X = 2H
W
W
in analogy to the classical mean curvature system ∆X = 2HW.
Remark 4.1. Clarenz and von der Mosel in [32] proposed another Laplacetype op
erator to analyse critical points for anisotropic and inhomogeneous variational prob
lems, namely
∆
C
:=
1
W
2
∑
i=1
∂
∂u
i
_
2
∑
j,k,ℓ=1
g
ik
h
kℓ
g
jℓ
∂
∂u
j
_
; (4.14)
see also the references in [32]. Their formalism was picked up e.g. by Cluttberbuck
[33] and Bergner and Dittrich [12].
66 4 Weighted differential geometry
4.5 The CodazziMainardi equations
We want to evaluate the integrability conditions which particularly pour out from
∂
uv
X
u
i −∂
vu
X
u
i = 0 for i = 1, 2,
by comparing the respective coefﬁcients of the normal directions. For this purpose
we write explicitely
X
u
i
u
=(Γ
1
i1
+Ω
1
i1
)X
u
+(Γ
2
i1
+Ω
2
i1
)X
v
+
n
∑
σ=1
L
σ,i1
N
σ
,
X
u
i
v
=(Γ
1
i2
+Ω
1
i2
)X
u
+(Γ
2
i2
+Ω
2
i2
)X
v
+
n
∑
σ=1
L
σ,i2
N
σ
for i = 1, 2. A second differentiation gives
∂
uv
X
u
i =
_
Γ
1
i1,v
+Ω
1
i1,v
+(Γ
1
i1
+Ω
1
i1
)(Γ
1
12
+Ω
1
12
) +(Γ
2
i1
+Ω
2
i1
)(Γ
1
22
+Ω
1
22
)
_
X
u
+
_
Γ
2
i1,v
+Ω
2
i1,v
+(Γ
1
i1
+Ω
1
i1
)(Γ
2
12
+Ω
2
12
) +(Γ
2
i1
+Ω
2
i1
)(Γ
2
22
+Ω
2
22
)
_
X
v
+
n
∑
σ=1
_
L
σ,i1,v
+(Γ
1
i1
+Ω
1
i1
)L
σ,12
+(Γ
2
i1
+Ω
2
i1
)L
σ,22
+
n
∑
ω=1
L
ω,i1
T
σ
ω,2
_
N
σ
−
2
∑
ℓ,m=1
n
∑
σ=1
L
σ,i1
L
σ,2ℓ
h
ℓm
W(X, N) ◦ X
u
m
as well as, interchanging u and v,
∂
vu
X
u
i =
_
Γ
1
i2,u
+Ω
1
i2,u
+(Γ
1
i2
+Ω
1
i2
)(Γ
1
11
+Ω
1
11
) +(Γ
2
i2
+Ω
2
i2
)(Γ
1
12
+Ω
1
12
)
_
X
u
+
_
Γ
2
i2,u
+Ω
2
i2,u
+(Γ
1
i2
+Ω
1
i2
)(Γ
2
11
+Ω
2
11
) +(Γ
2
i2
+Ω
2
i2
)(Γ
2
12
+Ω
2
12
)
_
X
v
+
n
∑
σ=1
_
L
σ,i2,u
+(Γ
1
i2
+Ω
1
i2
)L
σ,11
+(Γ
2
i2
+Ω
2
i2
)L
σ,12
+
n
∑
ω=1
L
ω,i2
T
σ
ω,1
_
N
σ
−
2
∑
ℓ,m=1
n
∑
σ=1
L
σ,i2
L
σ,1ℓ
h
ℓm
W(X, N) ◦ X
u
m .
Focusing on the normal components of both identitites proves the following version
of the CodazziMainardi equations which completes this section.
4.5 The CodazziMainardi equations 67
Theorem 4.3. (CodazziMainardi integrability conditions in weighted form)
Let the immersion X : B →R
n+2
together with ONF N be given. Then there hold
L
σ,i1,v
+(Γ
1
i1
+Ω
1
i1
)L
σ,12
+(Γ
2
i1
+Ω
2
i1
)L
σ,22
+
n
∑
ω=1
L
ω,i1
T
σ
ω,2
= L
σ,i2,u
+(Γ
1
i2
+Ω
1
i2
)L
σ,11
+(Γ
2
i2
+Ω
2
i2
)L
σ,12
+
n
∑
ω=1
L
ω,i2
T
σ
ω,1
(4.15)
for all i = 1, 2 and σ = 1, 2, . . . , n.
¡ These equations differ from the classical integrability conditions (3.6) in the
appearence of the additional data Ω
k
i j
. We will make essential use them in the next
chapter.
Chapter 5
The Hopf vector
5.1 Linear dependence of the weighed fundamental forms
5.2 Minimal surfaces and weighted minimal surfaces
5.3 The torsion of normal frames
5.4 The functional of total torsion
5.5 The curvatura integra
5.6 Hopf functions and Hopf vector
5.7 A Pascali system for the Hopf functions
5.8 An example: Weighted minimal surfaces in R
3
70 5 The Hopf vector
We prove a linear connection between the three weighted fundamental forms what enables us to
compare the spherical energy of an immersion with its curvatures and torsions. Furthermore we
consider complexvalued Hopf ﬁelds and verify their analyticity in the sense of Bers and Vekua.
This allows us to characterize singular points of the spherical mapping of special surfaces.
5.1 Linear dependence of the weighted fundamental forms
The three weighted fundamental forms can not be chosen independently from each
other as the following result shows.
Theorem 5.1. (Linear dependence of the fundamental forms)
Let the immersion X : B →R
n+2
together with an ONF N be given. Then it holds
III
W,N
σ
(X) −2H
W,σ
II
W,σ
(X) +K
σ
I
W
(X) =
_
n
∑
ϑ=1
T
ϑ
σ,i
T
ϑ
σ, j
_
i, j=1,2
(5.1)
for σ = 1, . . . , n with the Gauss curvatures K
σ
from (2.9) and the weighted mean
curvatures H
W,σ
from (4.12).
Proof. We evaluate the coefﬁcients f
σ,i j
= N
σ,u
i N
σ,u
j of the weighted third funda
mental form to get
f
σ,i j
=
_
−
2
∑
m,n=1
L
σ,im
h
mn
W(X, N) ◦ X
u
n +
n
∑
ϑ=1
T
ϑ
σ,i
N
ϑ
_
◦ W(X, N)
−1
◦ . . .
. . . ◦
_
−
2
∑
r,s=1
L
σ, jr
h
rs
W(X, N) ◦ X
u
s +
n
∑
λ=1
T
λ
σ, j
N
λ
_
=
2
∑
m,n,r,s=1
L
σ,im
L
σ, jr
h
mn
h
rs
h
ns
+
n
∑
ϑ,λ=1
T
ϑ
σ,i
T
λ
σ, j
δ
ϑλ
=
2
∑
m,r=1
L
σ,im
L
σ, jr
h
mr
+
n
∑
ϑ=1
T
ϑ
σ,i
T
ϑ
σ, j
.
72 5 The Hopf vector
Now we exemplarily compute for i, j = 1
f
σ,11
−
n
∑
ϑ=1
(T
ϑ
σ,1
)
2
= L
2
σ,11
h
11
+2L
σ,11
L
σ,12
h
12
+L
2
σ,12
h
22
=
_
L
σ,11
h
11
+2L
σ,12
h
12
+L
σ,22
h
22
_
L
σ,11
−(L
σ,11
L
σ,22
−L
2
σ,12
)h
22
= 2H
W,σ
L
σ,11
−K
σ
h
11
.
The remaining coefﬁcients follow analogously. ⊓⊔
The nonweighted version of this identity runs
III
N
σ
(X) −2H
σ
II
N
σ
(X) +K
σ
I(X) =
_
n
∑
ϑ=1
T
ϑ
σ,i
T
ϑ
σ, j
_
i, j=1,2
.
Finally, using the coefﬁcients
´
f
σ,i j
= N
⊤
σ,u
i
◦ W(X, N)
−1
◦ N
⊤
σ,u
j
from (4.4) we
would arrive at the relation
´
III
W,N
σ
(X) −2H
W,σ
II
W,N
σ
(X) +K
σ
I
W
(X) = 0 (5.2)
where we take the identity L
N,i j
=−X
u
j N
u
j =−X
u
i N
⊤
u
j
into account.
5.2 Minimal surfaces and weighted minimal surfaces
Consider a conformally parametrized minimal surface X : B →R
3
with the charac
teristic property
H ≡0
for its scalar mean curvature H. Then it holds
III(X) −KWE
2
= 0
with the twodimensional unit matrix E
2
⊂ R
22
. Since I(X) is given in diago
nal form we infer that the spherical mapping N: B → R
3
is also conformally
parametrized.
The same is true for weighted minimal surfaces satisfying
H
W
≡0
as well as the relations (4.10) w.r.t. some weight matrix W(X, Z). Namely, we infer
III
W
(X) −KWE
2
= 0.
5.3 The torsion of orthonormal normal frames 73
5.3 The torsion of orthonormal normal frames
Let us consider again the nonweighted case W(X, Z) ≡ E
n+2
. We write the linear
dependence of the fundamental forms from (5.1) as
e
σ,i j
−2H
σ
L
σ,i j
+K
σ
g
i j
=
n
∑
ϑ=1
T
ϑ
σ,i
T
ϑ
σ, j
with the coefﬁcients e
σ,i j
= N
σ,u
i N
σ,u
j .
Corollary 5.1. Let the conformally parametrized immersion X : B →R
n+2
together
with an ONF N be given. Then
e
σ,11
= 2H
σ
L
σ,11
−K
σ
W +
n
∑
ϑ=1
(T
ϑ
σ,1
)
2
,
e
σ,12
= 2H
σ
L
σ,12
+
n
∑
ϑ=1
T
ϑ
σ,1
T
ϑ
σ,2
,
e
σ,22
= 2H
σ
L
σ,22
−K
σ
W +
n
∑
ϑ=1
(T
ϑ
σ,2
)
2
.
In particular, we infer
[∇N
σ
[
2
= e
σ,11
+e
σ,22
= 2(2H
2
σ
−K
σ
)W +
n
∑
ϑ=1
_
(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
(5.3)
for the square of the Euclidean gradient ∇N
σ
= (N
σ,u
, N
σ,v
).
The sumon the right hand side of the latter identity motivates us for the following
deﬁnition.
Deﬁnition 5.1. The torsion of the ONF N is deﬁned as
T :=
1
2
2
∑
i, j=1
n
∑
σ,ϑ=1
g
i j
T
ϑ
σ,i
T
ϑ
σ, j
. (5.4)
Remark 5.1. A torsion within our Finslertype setting could take the form
T
W
:=
1
2
2
∑
i, j=1
n
∑
σ,ϑ=1
h
i j
T
ϑ
σ,i
T
ϑ
σ, j
(5.5)
Note here that for an arbitrary vector z ∈ R
2
it holds
z ◦ (h
i j
)
i, j=1,2
◦ z = z ◦ (DX ◦ W(X, N) ◦ DX) ◦ z
= (z ◦ DX) ◦ W(X, N) ◦ (DX ◦ z).
74 5 The Hopf vector
Now apply property (W3), and we conclude that T and T
W
are equivalent in the
following sense
1
1 +ω
0
T ≤T
W
≤(1 +ω
0
)T. (5.6)
While the torsion T (and so T
W
) depends on the chosen ONF N, it is invariant
w.r.t. parameter transformations of class P.
Proposition 5.1. T is invariant w.r.t. parameter transformations of class P.
Proof. Using the results from section 1.7 we calculate
2
∑
i, j=1
n
∑
σ,ϑ=1
g
i j
T
ϑ
σ,i
T
ϑ
σ, j
=
2
∑
i, j=1
2
∑
α,β µ,ν=1
n
∑
σ,ϑ=1
¯
Λ
i
µ
¯
Λ
j
ν
Λ
α
i
Λ
β
j
g
µν
T
ϑ
σ,α
T
ϑ
σ,β
=
2
∑
α,β,µ,ν=1
n
∑
σ,ϑ=1
δ
α
µ
δ
β
ν
g
µν
T
ϑ
σ,α
T
ϑ
σ,β
=
2
∑
α,β=1
n
∑
σ,ϑ=1
g
αβ
T
ϑ
σ,α
T
ϑ
σ,β
proving the claim. ⊓⊔
Now taking the deﬁnition of H and K from (2.5) and (2.9) into account, a direct
calculation proves the
Theorem 5.2. (Energy density of the Gauss map)
Let the conformally parametrized immersion X : B →R
n+2
together with an ONF
N be given. Then it holds
n
∑
σ=1
[∇N
σ
[
2
= 2(2H
2
−K)W +2TW (5.7)
with the squared length H
2
of the mean curvature vector H, the Gauss curvature K
and the torsion T of the ONF N.
This identity is very useful for various energy and gradient estimates as we will
see later. For the moment we want to mention that an integration yields
n
∑
σ=1
__
B
[∇N
σ
[
2
dudv = 2
__
B
(2H
2
−K)W dudv +2
__
B
TW dudv
what leads us directly to our next concept.
5.5 The curvatura integra 75
5.4 The functional of total torsion
Deﬁnition 5.2. Let the immersion X : B →R
n+2
together with an ONF N be given.
Then the functional
T
X
[N] :=
__
B
TW dudv
is called the total torsion of N.
Proposition 5.2. The functional T
X
[N] of total torsion is invariant w.r.t. to parame
ter transformations of class P.
Proof. This follows immediately from Proposition 5.1.
Though T
X
[N] is independent of the choice of the parametrization, it depends
on the choice of the ONF N. In chapters 7 and 8 below we will present methods
to establish existence and regularity of orthonormal normal frames critical for the
functional of total torsion.
5.5 The curvatura integra
Deﬁnition 5.3. Let the immersion X : B →R
n+2
be given. Then the functional
K [X] :=
__
B
KW dudv
is called its curvatura integra.
We want to omit proving its independence of the parametrization as well as the
choice of the ONF. Both facts reliy strongly on Proposition 2.3 from chapter 2.
Proposition 5.3. The curvatura integra K [X] is invariant w.r.t. to parameter trans
formations of class P, and it does not depend on the choice of the ONF N.
Now we want to prove the famous theorem of Bonnet and Gauss which relates
the curvatura integra of an immersion with the geodesic curvature κ
g
of its boundary
curve.
The curvature κ
g
is roughly deﬁned as follows: Consider a point P of the surface’s
boundary curve Γ ⊂ R
n+2
and the tangential plane of the surface at P. Project Γ
locally around P onto this tangential plane to get a new planar curve. Then κ
g
(P) is
deﬁned as the usual curvature of the projected planar curve at P.
For proving the integral formula of Bonnet and Gauss we need the representation
K =−
1
W
∆ log
√
W
of the Gauss curvature using conformal parameters which we derived in chapter 3.
76 5 The Hopf vector
Theorem 5.3. (Integral formula of Bonnet and Gauss)
Let the immersion X : B →R
n+2
be given. Then it holds
__
B
KW dudv = 2π −
2π
_
0
κ
g
(s)ds.
Proof. We follow Dierkes et al. [44]. Suppose X = X(u, v) is given in conformal
parameters (u, v) ∈B. Let ν ∈R
2
denote the outer unit normal vector at the boundary
curve ∂B. Let this curve be given in parametric form (cost, sint), t ∈ [0, 2π). Now
we take Minding’s formula
κ
g
(t)
_
W(t) = 1 +
∂
∂ν
log
_
W(t) for all t ∈ [0, 2π)
into account which remains unproved here (see [44], chapter 1). Partial integration
then yields
−
__
B
KW dudv =
__
B
∆ log
_
W(t)dudv =
2π
_
0
∂
∂ν
log
_
W(t)dt
=
2π
_
0
_
κ
g
(t)
_
W(t) −1
_
dt =
_
∂B
κ
g
(s)ds −2π
proving the statement. ⊓⊔
5.6 Hopf functions and Hopf vector
Consider an immersion X : B →R
3
with second fundamental form (L
i j
)
i, j=1,2
.
In 1950, H. Hopf [94] (see also Jost’s textbook [102]) discovered that the
complexvalued function
H (w) := L
11
(w) −L
22
(w) −2iL
12
(w), w ∈ B,
is holomorphic if the scalar mean curvature H of the surface is constant.
1
In this section we introduce a generalized vectorvalued Hopf function adapted
to our Finslertype setting. We will make essential use of complexanalytical tools
of Bers [14] and Vekua [160], [161].
Let a weight matrix W(X, Z) ∈ R
(n+2)(n+2)
be given.
1
In particular, this fact turns out to be a key ingredient for proving that a compact embedding of
constant, nonvanishing mean curvature is indeed the standard sphere.
5.6 Hopf functions and Hopf vector 77
Lemma 5.1. Let the weighted conformally parametrized immersion X : B →R
n+2
together with an ONF N be given. Deﬁne the functions
a :=Ω
1
22
+Ω
2
21
, b :=Ω
2
22
−Ω
1
21
, c :=Ω
1
12
+Ω
2
11
, d :=Ω
2
12
−Ω
1
11
, (5.8)
with Ω
k
i j
being the coefﬁcients from Gauss equations in weighted form from section
4.3, as well as
r
σ
:= W
u
H
W,σ
+2W∂
u
H
W,σ
−2Ω
2
21
WH
W,σ
+
n
∑
ϑ=1
(L
ϑ,22
T
σ
ϑ,1
−L
ϑ,21
T
σ
ϑ,2
),
s
σ
:= W
v
H
W,σ
−2Ω
2
11
WH
W,σ
+
n
∑
ϑ=1
(L
ϑ,12
T
σ
ϑ,1
−L
ϑ,11
T
σ
ϑ,2
)
(5.9)
with the weighted mean curvature vector H
W
from section 4.4. Then there hold
L
σ,11,u
+L
σ,12,v
= aL
σ,11
+bL
σ,12
+r
σ
,
L
σ,11,v
−L
σ,12,u
= cL
σ,11
+dL
σ,12
+s
σ
(5.10)
for σ = 1, . . . , n.
Proof. First come the CodazziMainardi equations from section 4.5, i.e.
L
σ,11,v
+(Γ
1
11
+Ω
1
11
)L
σ,12
+(Γ
2
11
+Ω
2
11
)L
σ,22
+
n
∑
ϑ=1
L
ϑ,11
T
σ
ϑ,2
= L
σ,12,u
+(Γ
1
12
+Ω
1
12
)L
σ,11
+(Γ
2
12
+Ω
2
12
)L
σ,12
+
n
∑
ϑ=1
L
ϑ,12
T
σ
ϑ,1
,
L
σ,21,v
+(Γ
1
21
+Ω
1
21
)L
σ,12
+(Γ
2
21
+Ω
2
21
)L
σ,22
+
n
∑
ϑ=1
L
ϑ,21
T
σ
ϑ,2
= L
σ,22,u
+(Γ
1
22
+Ω
1
22
)L
σ,11
+(Γ
2
22
+Ω
2
22
)L
σ,12
+
n
∑
ϑ=1
L
ϑ,22
T
σ
ϑ,1
.
Thus, using weighted conformal parameters together with (4.10) and (2.6) we get
L
σ,11,v
−
W
v
2W
L
σ,22
+
2
∑
m=1
Ω
m
11
L
σ,m2
+
n
∑
ϑ=1
L
ϑ,11
T
σ
ϑ,2
= L
σ,12,u
+
W
v
2W
L
σ,11
+
2
∑
m=1
Ω
m
12
L
σ,m1
+
n
∑
ϑ=1
L
ϑ,12
T
σ
ϑ,1
,
L
σ,12,v
+
W
u
2W
L
σ,22
+
2
∑
m=1
Ω
m
21
L
σ,m2
+
n
∑
ϑ=1
L
ϑ,21
T
σ
ϑ,2
= L
σ,22,u
−
W
u
2W
L
σ,11
+
2
∑
m=1
Ω
m
22
L
σ,m1
+
n
∑
ϑ=1
L
ϑ,22
T
σ
ϑ,1
.
78 5 The Hopf vector
Rearranging gives the identities
L
σ,11,v
−L
σ,12,u
= (Ω
1
12
+Ω
2
11
)L
σ,11
+(Ω
2
12
−Ω
1
11
)L
σ,12
−2Ω
2
11
WH
W,σ
+W
v
H
W,σ
+
n
∑
ϑ=1
(L
ϑ,12
T
σ
ϑ,1
−L
ϑ,11
T
σ
ϑ,2
),
L
σ,11,u
+L
σ,12,v
= (Ω
1
22
+Ω
2
21
)L
σ,11
+(Ω
2
22
−Ω
1
21
)L
σ,12
−2Ω
2
21
WH
W,σ
+W
u
H
W,σ
+2W∂
u
H
W,σ
+
n
∑
ϑ=1
(L
ϑ,22
T
σ
ϑ,1
−L
ϑ,21
T
σ
ϑ,2
)
for σ = 1, . . . , n. This proves the statement. ⊓⊔
These identities enable us to prove generalized analyticity of the following
complexvalued functions.
Deﬁnition 5.4. The Hopf function H
N
∈ C of an immersion X w.r.t. some unit nor
mal vector N is deﬁned as
H
N
(w) := L
N,11
(w) −L
N,22
(w) −2iL
N,12
(w), w ∈ B. (5.11)
Note that it holds
H
N
= 2L
N,11
−(L
N,11
+L
N,22
) −2iL
N,12
=−2H
W,N
W +2L
N,11
−2iL
N,12
.
Next we introduce the socalled Wirtinger symbols (see e.g. Vekua [161]): Let
Φ: C →C
n
be continuously differentiable. Then we deﬁne its complex derivatives
Φ
w
≡∂
w
Φ :=
1
2
(Φ
u
−iΦ
v
), Φ
w
≡∂
w
:=
1
2
(Φ
u
+iΦ
v
).
Lemma 5.2. Let the weighted conformally parametrized immersion X : B →R
n+2
together with an ONF N be given. Then it holds
∂
w
H
σ
= AH
σ
+BH
σ
+2(A+B−Ω
2
21
−iΩ
2
11
)H
W,σ
W +2∂
w
H
W,σ
W
+
n
∑
ϑ=1
_
(L
ϑ,22
+iL
ϑ,12
)T
σ
ϑ,1
−(L
ϑ,21
+iL
ϑ,11
)T
σ
ϑ,2
_
for σ = 1, . . . , n, where
A :=
1
4
(a −d +ic +ib), B :=
1
4
(a +d +ic −ib),
and a, b, c, and d are taken from (5.8).
5.6 Hopf functions and Hopf vector 79
Before we come to the proof of this lemma we want to consider some interesting
special situations.
The nonweighted case
In the nonweighted case W(X, Z) ≡E
n+2
this identity reduces to
∂
w
H
σ
= 2∂
w
H
σ
W +
n
∑
ϑ=1
_
(L
ϑ,22
+iL
ϑ,12
)T
σ
ϑ,1
−(L
ϑ,21
+iL
ϑ,11
)T
σ
ϑ,2
_
.
Thus H
σ
satisﬁes a system of inhomogeneous CauchyRiemann equations. In
particular, H
σ
is holomorphic if X represents a minimal surface satisfying H
σ
≡ 0
with ﬂat normal bundle.
Surfaces in R
3
Consider an immersion X : B→R
3
with unit normal vector N, and as usual we write
H instead of H
σ
. The lemma states
∂
w
H = AH +BH +2(A+B−Ω
2
21
−iΩ
2
11
)H
W
W +2H
W,w
W.
In particular, two cases are of special interest.
1. Weighted minimal surface
Let X : B →R
3
be a weighted minimal surface satisfying H
W
≡ 0, then H is a
pseudoholomorphic function fulﬁlling
∂
w
H = AH +BH .
In the section after the next we will see that this fact allows us to draw some
important conclusions about the geometry of the spherical mapping of weighted
minimal surfaces in R
3
.
2. Surfaces with constant mean curvature
Let X : B →R
3
represent an immersion with constant scalar mean curvature H.
Then H is holomorphic with
∂
w
H = 2H
w
W ≡0.
As we will see shortly
[H [
2
= 4(H
2
−K),
and thus either H
2
−K ≡0, equivalent to κ
1
≡κ
2
, or κ
1
=κ
2
at most at isolated
points in every compact subset Ω ⊂⊂
˚
B (see e.g. Hopf [94], or Jost [102]). Such
surface points are called umbilical points.
80 5 The Hopf vector
Now let us come to the proof of the previous lemma.
Proof. We consider the auxiliary function
H
∗
σ
:= L
σ,11
−iL
σ,12
=
1
2
H
σ
+WH
W,σ
. (5.12)
On the one hand it holds
AH
∗
σ
+BH
∗
σ
=
1
4
(a −d +ic +ib)(L
σ,11
−iL
σ,12
)
+
1
4
(a +d +ic −ib)(L
σ,11
+iL
σ,12
)
=
1
2
(aL
σ,11
+bL
σ,12
) +
i
2
(cL
σ,11
+dL
σ,12
),
while on the other hand, on account of (5.10), we infer
∂
w
H
∗
σ
=
1
2
(L
σ,11,u
+L
σ,12,v
) +
i
2
(L
σ,11,v
−L
σ,12,u
)
=
1
2
(aL
σ,11
+bL
σ,12
) +
i
2
(cL
σ,11
+dL
σ,12
) +
1
2
r
σ
+
i
2
s
σ
.
Comparing both identities yields
∂
w
H
∗
σ
= AH
∗
+BH
∗
σ
+
1
2
r
σ
+
i
2
s
σ
,
or equivalently, together with (5.12),
1
2
∂
w
H
σ
+∂
w
(WH
W,σ
)
= ∂
w
H
∗
σ
=
A
2
H
σ
+
B
2
H
σ
+(A+B)WH
W,σ
+
1
2
r
σ
+
i
2
s
σ
.
Rearranging for ∂
w
H
σ
gives
∂wH
σ
= AH
σ
+BH
σ
+2(A+B)WH
W,σ
+r
σ
+is
σ
−2∂
w
(WH
W,σ
).
Finally notice that from (5.9) if follows that
r
σ
+is
σ
−2(WH
g,σ
)
w
= 2W∂
w
H
W,σ
−2Ω
2
21
WH
W,σ
−2iΩ
2
11
WH
W,σ
+
n
∑
ϑ=1
(L
ϑ,22
+iL
ϑ,21
)T
σ
ϑ,1
−
n
∑
ϑ=1
(L
ϑ,21
+iL
ϑ,11
)T
σ
ϑ,2
,
and the statement is proved. ⊓⊔
5.7 A Pascali system for the Hopf functions 81
5.7 A Pascali system for the Hopf functions
Let X : B →R
n+2
be a weighted minimal surface satisfying H
W
≡ 0 with given
weight matrix W(X, Z). We want to rearrange the above compexvalued differential
system into a socalled Pascali system.
Deﬁnition 5.5. We deﬁne the complexvalued torsion vector
T
ϑ
σ
:= T
ϑ
σ,1
+iT
ϑ
σ,2
∈ C for σ, ϑ = 1, . . . , n. (5.13)
Now using weighted conformally parameters (u, v) ∈ B we calculate
(L
ϑ,22
+iL
ϑ,12
)T
σ
ϑ,1
−(L
ϑ,12
+iL
ϑ,11
)T
σ
ϑ,2
= 2H
W,ϑ
WT
σ
ϑ,1
−(L
ϑ,11
−iL
ϑ,12
)T
σ
ϑ,1
−i(L
ϑ,11
−iL
ϑ,12
)T
σ
ϑ,2
= 2H
W,ϑ
WT
σ
ϑ,1
−H
∗
ϑ
(T
σ
ϑ,1
+iT
σ
ϑ,2
)
= 2H
W,ϑ
WT
σ
ϑ,1
−
1
2
H
ϑ
T
σ
ϑ
−H
W,ϑ
WT
σ
ϑ
recalling the identity
H
∗
ϑ
= L
ϑ,11
−iL
ϑ,12
=
1
2
H
ϑ
+H
W,ϑ
W
from the previous section. Now H
W
≡0 gives us
(L
ϑ,22
+iL
ϑ,12
)T
σ
ϑ,1
−(L
ϑ,12
+iL
ϑ,11
)T
σ
ϑ,2
=−
1
2
H
ϑ
T
σ
ϑ
.
Finally we need a suitable complexvalued Hopf vector. Our central deﬁnition of
this section is
Deﬁnition 5.6. The Hopf vector H ∈C
n
of an immersion X : B →R
n+2
w.r.t. some
ONF N is deﬁned as
H := (H
1
, . . . , H
n
) ∈ C
n
. (5.14)
Now we come to the main result of this section.
Theorem 5.4. Let the weighted conformally parametrized immersion X : B →R
n+2
together with an ONF N be given. Then it holds
∂
w
H
σ
= AH
σ
+BH
σ
−
1
2
n
∑
ϑ=1
T
σ
ϑ
H
ϑ
= AH
σ
+BH
σ
+
1
2
n
∑
ϑ=1
T
ϑ
σ
H
ϑ
(5.15)
for σ = 1, . . . , n.
82 5 The Hopf vector
The Hopf vector H ∈ C
n
satisﬁes the Pascali system
∂
w
H = AH +BH +
1
2
T◦ H (5.16)
with the complexvalued torsion matrix
T := (T
ϑ
σ
)
σ,ϑ=1,...,n
∈ C
nn
.
Here we take into account
T
σ
ϑ
=−T
ϑ
σ
.
But now notice that
[H
σ
[
2
= (L
σ,11
−L
σ,22
)
2
+4L
2
σ,12
= (L
σ,11
+L
σ,22
)
2
+4(L
σ,11
L
σ,22
−L
2
σ,12
)
= 4(H
2
W,σ
−K
σ
)W
2
= −4K
σ
W
2
.
Thus we arrive at the interesting identity
[H [
2
=
n
∑
σ=1
[H
σ
[
2
=−4
n
∑
σ=1
K
σ
W
2
=−4KW
2
,
i.e. the zeros of the Gauss curvature K of a weighted minimal surface are the zeros
of the complexvalued Hopf vector H !
By the way, we also infer
[H [
2
= 4
n
∑
σ=1
(H
2
σ
−K
σ
)W
2
= 4(H
2
−K)W
2
,
and identity used already in the previous section.
Actually the complexvalued system (5.16) is a special Pascali system for our
Hopf function H . Following Wendland [164], Theorem 5.3.3, H can locally be
represented in the form
E◦ Φ
with a C
α
regular matrix E ∈ C
nn
with the property det E ,= 0, and a holomorphic
vector Φ ∈ C
n
.
Consequently we can apply the similarity principle for generalized analytic
vectorvalued functions (see e.g. Wendland [164], Theorem 5.3.5) and arrive at the
following characterization of weighted minimal surfaces.
Corollary 5.2. Either it holds H ≡ 0 in B, or H has only isolated zeros of ﬁnite
order in every compact subset Ω ⊂⊂
˚
B. Therefore, either it holds K ≡ 0 in B and
the weighted minimal surface is a plane, or in every such Ω ⊂⊂
˚
B there are only
ﬁnitely many points with K = 0.
5.8 An example: Weighted minimal surfaces in R
3
83
5.8 An example: Weighted minimal surfaces in R
3
Consider a weighted minimal surface X : B →R
3
with weight W(X, Z) ∈ R
33
and
unit normal vector
N =
X
u
X
v
[X
u
X
v
[
.
The associated Hopf function H satisﬁes
H
w
= AH +BH .
Solutions of a such differential equation are called pseudoholomorphic functions.
Following the similarity principle for pseudoholomorphic functions of Bers and
Vekua (see e.g. Bers [14], Courant and Hilbert [40], Sauvigny [143], Vekua [161],
or Wendland [164]), the Hopf function H can be represented in the form
Φ(w)e
Ψ(w)
with a holomorphic function Φ and some integral function Ψ. Again the zeros of
the Gaussian curvature K are isolated in every compact subset Ω ⊂⊂
˚
B as long as
X : B →R
3
is not a plane.
In the following we want to show that the zeros of K coincide to the branch
points of the spherical mapping N which are characterized by the property
N
u
N
v
= 0 at points with K = 0.
This tells us the following calculation: Introduce weighted conformal parameters
(u, v) ∈ B. Then the Weingarten equations in the weighted form from section 4.3
can be written as
N
u
= −
L
11
W
W(X, N) ◦ X
u
−
L
12
W
W(X, N) ◦ X
v
,
N
v
= −
L
12
W
W(X, N) ◦ X
u
−
L
22
W
W(X, N) ◦ X
v
.
We need the following calculus rule from Sauvigny [141].
Lemma 5.3. Let M⊂R
33
be a nonsingular and symmetric matrix. Then
_
M◦ X
_
_
M◦Y
_
=
_
det M
_
M
−1
◦ (X Y)
for arbitrary X,Y ∈ R
3
.
84 5 The Hopf vector
Proof. For an arbitrary vector Z ∈ R
3
we calculate
_
(M◦ X) (M◦Y) −(detM)M
−1
◦ (X Y)
_
(M◦ Z)
=
_
(M◦ X) (M◦Y)
_
(M◦ Z) −(detM)(X Y) (M
−1
◦ M◦ Z)
= (det M)
_
(X Y) Z −(X Y) Z
_
= 0.
This proves the calculus rule. ⊓⊔
At last we obtain
N
u
N
v
=
_
L
11
W
W(X, N) ◦ X
u
+
L
12
W
W(X, N) ◦ X
v
_
. . .
. . .
_
L
12
W
W(X, N) ◦ X
u
+
L
22
W
W(X, N) ◦ X
v
_
=
L
11
L
22
−L
2
12
W
2
_
W(X, Z) ◦ X
u
_
_
W(X, Z) ◦ X
v
_
=
L
11
L
22
−L
2
12
W
2
(X
u
X
v
) = KWN
making use of W(X, N) ◦ N = N. Thus, as stated, surface points with the property
N
u
N
v
= 0 are exactly points with K = 0.
In chapter 15 we will investigate more closely weighted minimal surfaces as well
as critical points for general anisotropic and inhomogeneous functionals in R
3
.
Chapter 6
The GaussOsserman map
6.1 The exterior product
6.2 The Grassmann normal space. Grassmann forms
6.3 Curvature vector and curvature matrix of the normal bundle
6.4 Grassmann manifolds and GaussOsserman map
6.5 FubiniStudy metric and the total curvature
6.6 Minimal surfaces with constant curvature K
86 6 The GaussOsserman map
This chapter is devoted to Grassmann forms and the GaussOsserman mapping as a further method
to generalize the classical spherical mapping of surfaces. This particularly proves its strength in
minimal surface theory. The natural metrical background for these studies is the FubiniStudy
metric. Finally we discuss a new curvature vector of the normal bundle.
6.1 The exterior product
For the following algebraic concepts of the Grassmann geometry we refer to Cartan
[23] or Heil [78]; see also Grassmann [74].
Deﬁnition 6.1. Let n ≥1. The exterior product
∧: R
n
R
n
→R
N
, N =
_
n
2
_
=
n(n −1)
2
,
is deﬁned by means of the following rules:
(E1) The mapping R
n
∋ (v, w) →v ∧w ∈ R
N
is bilinear, i.e.
(α
1
v
1
+α
2
v
2
) ∧(β
1
w
1
+β
2
w
2
)
= α
1
β
2
v
1
∧w
1
+α
1
β
2
v
1
∧w
2
+α
2
β
1
v
2
∧w
1
+α
2
β
2
v
2
∧w
2
for all α
i
, β
i
∈ R, v
i
, w
i
∈ R
n
, and it is skewsymmetric,
v ∧w =−w∧v for all v, w ∈ R
n
;
in particular, it holds v ∧v = 0.
(E2) With e
1
= (1, 0, 0, . . . , 0) ∈ R
n
, e
2
= (0, 1, 0, . . . , 0) ∈ R
n
etc. we set
e
1
∧e
2
:= (1, 0, 0, . . . , 0, 0) ∈ R
N
,
e
1
∧e
3
:= (0, 1, 0, . . . , 0, 0) ∈ R
N
,
.
.
.
e
n−1
∧e
n
:= (0, 0, 0, . . . , 0, 1) ∈ R
N
.
88 6 The GaussOsserman map
From this settings we immediately obtain the
Proposition 6.1. The vectors e
k
∧e
ℓ
form an orthonormal basis of R
N
, i.e.
(e
i
∧e
j
) (e
k
∧e
ℓ
) =
_
1 if i = k and j = ℓ
0 if i ,= k or j ,= ℓ
. (6.1)
Proposition 6.2. For two vectors v = (v
1
, . . . , v
n
) and w = (w
1
, . . . , w
n
) it holds
v ∧w =
∑
1≤i<j≤n
(v
i
w
j
−v
j
w
i
)e
i
∧e
j
. (6.2)
Proof. We compute
v ∧w =
_
n
∑
i=1
v
i
e
i
_
∧
_
n
∑
j=1
w
j
e
j
_
=
n
∑
i, j=1
v
i
w
j
e
i
∧e
j
=
∑
1≤i<j≤n
(v
i
w
j
−v
j
w
i
)e
i
∧e
j
,
which already proves the statement. ⊓⊔
Example 6.1. For n = 3 we have N = 3 and
e
1
∧e
2
= (1, 0, 0), e
1
∧e
3
= (0, 1, 0), e
2
∧e
3
= (0, 0, 1).
With two vectors v = (v
1
, v
2
, v
3
) and w = (w
1
, w
2
, w
3
) we compute
v ∧w = v
1
w
2
e
1
∧e
2
−v
1
w
3
e
1
∧e
3
+v
2
w
1
e
2
∧e
1
+v
2
w
3
e
2
∧e
3
−v
3
w
1
e
3
∧e
1
+v
3
w
2
e
3
∧e
2
= (v
1
w
2
−v
2
w
1
)e
1
∧e
2
+(v
3
w
1
−v
1
w
3
)e
1
∧e
3
+(v
2
w
3
−v
3
w
2
)e
2
∧e
3
= (v
1
w
2
−v
2
w
1
, v
3
w
1
−v
1
w
3
, v
2
w
3
−v
3
w
2
).
Thus the usual vector product
v w = (v
2
w
3
−v
3
w
2
, v
3
w
1
−v
1
w
3
, v
1
w
2
−v
2
w
1
)
in R
3
does not coincide with the exterior product v ∧w.
We want to collect some further properties of the exterior product.
Corollary 6.1. For arbitrary vectors a, b ∈ R
n
there hold
◦ (λa) ∧b =λ(a ∧b);
◦ (a +b) ∧c = a ∧c +b ∧c;
◦ (a ∧b)
u
i = a
u
i ∧b +a ∧b
u
i .
6.2 The Grassmann normal space. Grassmann forms 89
Let two vectors v = (v
1
, v
2
, 0, . . . , 0) and w = (w
1
, w
2
, 0, . . . , 0) be given. Then
◦ v ∧w ⊥span
_
v ∧e
3
, . . . , v ∧e
n
, w∧e
3
, . . . , w∧e
n
, e
3
∧e
n
, . . . , e
n−1
∧e
n
_
.
6.2 The Grassmann normal space. Grassmann forms
Let X : B →R
n+2
be conformally parametrized, N an associated ONF. Then the set
_
X
u
√
W
,
X
v
√
W
, N
1
, N
2
, . . . , N
n
_
forms an orthonormal system spanning the embedding space R
n+2
at each point of
the surface X.
Deﬁnition 6.2. The Grassmann normal space of the conformally parametrized im
mersion X : B →R
n+2
with ONF N at the point w ∈ B is given by
G
w
[X] := Span
_
N , X
11
, . . . , X
1n
, X
21
, . . . , X
2n
, N
11
, . . . , N
1n
, N
23
, . . . , N
n−1,n
_
with the N =
_
n
2
_
unit vectors
N :=
X
u
∧X
v
W
, X
iσ
:=
X
u
i ∧N
σ
√
W
, N
σϑ
:= N
σ
∧N
ϑ
(6.3)
for i = 1, 2 and σ, ϑ = 1, . . . , n.
Let us consider some examples.
1. In case n = 1 of one codimension we have simply
N = N, X
11
=
X
v
√
W
, X
21
=−
X
u
√
W
.
Thus the Grassmann space is just Span¦X
u
, X
v
, N¦
∼
=R
3
.
2. For further illustration we consider the case n = 2 of two codimensions. The
Grassmann space G
w
[X] consist of
N =
X
u
∧X
v
W
,
X
11
=
X
u
∧N
1
√
W
, X
12
=
X
u
∧N
2
√
W
, X
21
=
X
v
∧N
1
√
W
, X
22
=
X
v
∧N
2
√
W
,
N
12
= N
1
∧N
2
.
90 6 The GaussOsserman map
3. Let a weight matrix W(X, Z) be given. We introduce weighted conformal param
eters. The system
_
1
√
W
X
u
,
1
√
W
X
v
, N
1
, . . . , N
n
_
must then be replaced by the following orthonormal moving frame
_
1
√
W
W(X, N)
1
2
◦ X
u
,
1
√
W
W(X, N)
1
2
◦ X
v
, N
1
, . . . , N
n
_
where W(X, Z)
1
2
is deﬁned via the spectral decomposition of W(X, Z).
From [N [
2
= 1 we immediately obtain N
u
i ⊥N for i = 1, 2. We want to prove
a representation formula for the derivatives N
u
i similar to the Weingarten equations.
Theorem 6.1. (GrassmannWeingarten equations)
Let the conformally parametrized immersion X : B →R
n+2
together with an ONF
N be given. Then there hold
N
u
i =−
2
∑
m=1
n
∑
ϑ=1
L
mϑ
i
X
mϑ
(6.4)
for i = 1, 2 with the 2nd (Grassmanntype) Gfundamental form
L
mϑ
i
:=−N
u
i X
mϑ
=
1
√
W
_
L
ϑ,i1
W
X
u
∧X
u
m +
L
ϑ,i2
W
X
v
∧X
u
m
_
N .
In particular, there hold
L
1ϑ
1
= −
L
ϑ,12
√
W
, L
2ϑ
1
=
L
ϑ,11
√
W
,
L
1ϑ
2
= −
L
ϑ,22
√
W
, L
2ϑ
2
=
L
ϑ,12
√
W
.
(6.5)
Proof. We already know N
u
i N = 0 for i = 1, 2. Thus we make the ansatz
N
u
i =
2
∑
m=1
n
∑
ϑ=1
a
mϑ
i
X
mϑ
+
n
∑
σ,ϑ=1
b
σϑ
i
N
σϑ
.
From N N
σϑ
= 0 we infer
b
σϑ
i
= N
u
i N
σϑ
= −N N
σϑ,u
i = −N (N
σ
∧N
ϑ
)
u
i
= −N (N
σ,u
i ∧N
ϑ
+N
σ
∧N
ϑ,u
i ).
6.3 Curvature vector and curvature matrix of the normal bundle 91
But the Weingarten equations (2.11) imply
N
σ,u
i ∧N
ϑ
= −
L
σ,i1
W
X
u
∧N
ϑ
−
L
σ,i2
W
X
v
∧N
ϑ
+
n
∑
ω=1
T
ω
σ,i
N
ω
∧N
ϑ
= −
L
σ,i1
W
X
1ϑ
−
L
σ,i2
W
X
2ϑ
+
n
∑
ω=1
T
ω
σ,i
N
ωϑ
such that N
σ,u
i ∧N
ϑ
⊥N , and b
σϑ
i
≡0. We now determine the coefﬁcients a
mϑ
i
:
a
mϑ
i
= N
u
i X
mϑ
= −N X
mϑ,u
i
= −
1
√
W
N (X
u
i
u
m ∧N
ϑ
+X
u
m ∧N
ϑ,u
i ) −
_
∂
∂u
i
1
√
W
_
N (X
u
m ∧N
ϑ
)
= −
1
√
W
N (X
u
m ∧N
ϑ,u
i )
=
1
√
W
N
_
L
ϑ,i1
W
X
u
m ∧X
u
+
L
ϑ,i2
W
X
u
m ∧X
v
_
.
Then we set L
mϑ
i
:=−a
mϑ
i
. Furthermore, we examplarily compute
L
1ϑ
1
= a
1ϑ
1
=
1
√
W
X
u
∧X
v
W
L
ϑ,12
W
X
u
∧X
v
=
L
ϑ,12
√
W
and analogously for the other coefﬁcients. This proves the theorem. ⊓⊔
We want to point out the similarity to the classical Weingarten equations
N
u
i =−
L
11
W
X
u
−
L
22
W
X
v
for a conformally parametrized immersion X : B →R
3
with unit normal vector N.
This motivates to consider N as a possible generalisation of N if the codimension
of the surface is greater the 1.
6.3 Curvature vector and curvature matrix of the normal bundle
We want to use this Grassmann formalism to reformulate Deﬁnition 3.5 using the
exterior product. We also take advantage of the opportunity to introduce further
useful quantities describing the curvature of the normal bundle.
Our aim is to prove the
92 6 The GaussOsserman map
Proposition 6.3. Let the immersion X : B →R
n+2
be given. Then the Grassmann
curvature vector S of its normal bundle
S :=
1
W
∑
1≤σ<ϑ≤n
S
ϑ
σ,12
N
σ
∧N
ϑ
does neither depend on the parametrization nor on the choice of the normal frame.
In particular, S and its length
[S[ =
√
S S =
√
S S =
1
W
_
∑
1≤σ<ϑ≤n
(S
ϑ
σ,12
)
2
represent a geometric quantities.
Here we recall the deﬁnition
S =
1
W
_
S
2
1,12
, S
3
1,12
, . . . , S
n
n−1,12
_
∈ R
N
of the curvature vector of the normal bundle.
The curvature vector S (or just S) is a third important curvature quantity for
twodimensional surfaces in R
n+2
beside their Gaussian curvature K and their mean
curvature vector H :
K =
n
∑
σ=1
L
σ,11
L
σ,22
−L
2
σ,12
g
11
g
22
−g
2
12
∈ R,
H =
n
∑
σ=1
L
σ,11
g
22
−2L
σ,12
g
12
+L
σ,22
g
11
g
11
g
22
−g
2
12
N
σ
∈ R
n
,
S =
1
W
∑
1≤σ<ϑ≤n
S
ϑ
σ,12
N
σ
∧N
ϑ
∈ R
N
.
In case n = 2 we have
S =
1
W
S
2
1,12
N
1
∧N
2
or S =
1
W
S
2
1,12
.
To prove at last the above proposition we are only left with verifying the length
invariance of S. For this purpose we make the following
Deﬁnition 6.3. Let the immersion X : B →R
n+2
together with an ONF N be given.
Then the torsion matrices of N are deﬁned as
T
i
= (T
ϑ
σ,i
)
σ,ϑ=1,...,n
∈ R
nn
for i = 1, 2.
Furthermore, the curvature matrix of its normal bundle is given as
S
12
= (S
ϑ
σ,12
)
σ,ϑ=1,...,n
∈ R
nn
.
6.3 Curvature vector and curvature matrix of the normal bundle 93
Recall the setting R = (r
σϑ
)
σ,ϑ=1,...,n
∈ R
nn
for the orthogonal transformation
between ONFs. Let R
T
denote its transposition such that R◦ R
T
= R
T
◦ R = E
n
,
and let
¯
S
12
be the curvature matrix after applying R to an ONF N.
Lemma 6.1. It holds the transformation rule
¯
S
12
= R◦ S
12
◦ R
T
.
Proof. For the proof we consider
¯
T
ϑ
σ,i
=
¯
N
σ,u
i
¯
N
ϑ
=
n
∑
α=1
_
r
σα,u
i N
α
+r
σα
N
α,u
i
_
n
∑
β=1
r
ϑβ
N
β
=
n
∑
α,β=1
_
r
σα,u
i r
ϑβ
δ
αβ
+r
σα
r
ϑβ
T
β
α,i
_
=
n
∑
α=1
r
σα,u
i r
αϑ
+
n
∑
α,β=1
r
σα
T
β
α,i
r
βϑ
.
Thus we arrive at the rule
¯
T
i
= R
u
i ◦ R
T
+R◦ T
i
◦ R
T
.
Using this formula we evaluate
¯
S
12
=
¯
T
1,v
−
¯
T
2,u
−
¯
T
1
◦
¯
T
T
2
+
¯
T
2
◦
¯
T
T
1
.
Namely, ﬁrst we have
¯
T
1,v
−
¯
T
2,u
= (R
u
◦ R
T
+R◦ T
1
◦ R
T
)
v
−(R
v
◦ R
T
+R◦ T
2
◦ R
T
)
u
= R
u
◦ R
T
v
−R
v
◦ R
T
u
+R◦ (T
1,v
−T
2,u
) ◦ R
T
+R
v
◦ T
1
◦ R
T
+R◦ T
1
◦ R
T
v
−R
u
◦ T
2
◦ R
T
−R◦ T
2
◦ R
T
u
,
and furthermore
¯
T
1
◦
¯
T
T
2
−
¯
T
2
◦
¯
T
T
1
= (R
u
◦ R
T
+R◦ T
1
◦ R
T
) ◦ (R◦ R
T
v
+R◦ T
T
2
◦ R
T
)
−(R
v
◦ R
T
+R◦ T
2
◦ R
T
) ◦ (R◦ R
T
u
+R◦ T
T
1
◦ R
T
)
= R
u
◦ R
T
v
+R
u
◦ T
T
2
◦ R
T
+R◦ T
1
◦ R
T
v
+R◦ T
1
◦ T
T
2
◦ R
T
−R
v
◦ R
T
u
−R
v
◦ T
T
1
◦ R
T
−R◦ T
2
◦ R
T
u
−R◦ T
2
◦ T
T
1
◦ R
T
since R◦ R
T
= R
T
◦ R =E
n
.
94 6 The GaussOsserman map
Taking both identities together gives
¯
T
1,v
−
¯
T
2,u
−
¯
T
1
◦
¯
T
T
2
+
¯
T
2
◦
¯
T
T
1
= R◦ (T
1,v
−T
2,u
−T
1
◦ T
T
2
+T
2
◦ T
T
1
) ◦ R
T
+R
v
◦ T
1
◦ R
T
+R◦ T
1
◦ R
T
v
−R
u
◦ T
2
◦ R
T
−R◦ T
2
◦ R
T
u
−R
u
◦ T
T
2
◦ R
T
−R◦ T
1
◦ R
T
v
+R
v
◦ T
T
1
◦ R
T
+R◦ T
2
◦ R
T
u
= R◦ (T
1,v
−T
2,u
−T
1
◦ T
T
2
+T
2
◦ T
T
1
) ◦ R
T
taking T
i
=−T
T
i
into account. This proves the statement. ⊓⊔
Now we continue with the proof of the previous proposition.
Proof of the proposition. In terms of our usual SO(n)action we compute
n
∑
σ,ϑ=1
¯
S
ϑ
σ,12
¯
N
σ
∧
¯
N
ϑ
=
n
∑
σ,ϑ=1
n
∑
α,β=1
¯
S
ϑ
σ,12
r
σα
r
ϑβ
N
α
∧N
β
=
n
∑
σ,ϑ=1
n
∑
α,β=1
r
ασ
¯
S
ϑ
σ,12
r
ϑβ
N
α
∧N
β
=
n
∑
α,β=1
S
β
α,12
N
α
∧N
β
with the orthogonal mapping R = (r
σω
)
σ,ω=1,...,n
. This proves the proposition. ⊓⊔
6.4 Grassmann manifolds and GaussOsserman map
We follow Hoffman and Osserman [93], see also Osserman [130].
Deﬁnition 6.4. The kdimensional Grassmann manifold Gr
k
(m) is the space of all
kdimensional subspaces of R
m
.
For example, the Grassmann manifold Gr
1
(2) is the set of all lines through the
origin in R
2
. Or the Grassmann manifold Gr
2
(3) is the set of all planes through
the origin in R
3
. Such a plane is uniquely determined by the line intersecting it
perpendicularly. Hence Gr
2
(3) is isomorphic to Gr
1
(3).
Consider now an oriented plane P ⊂ R
n+2
spanned by an ordered pair of or
thogonal and equally long vectors V ∈ R
n+2
and W ∈ R
n+2
. With P we associate
the complexvalued point
Z =V +iW ∈ C
n+2
, Z = (z
1
, . . . , z
n+2
).
6.4 Grassmann manifolds and GaussOsserman map 95
If ¦V
′
,W
′
¦ denotes a second basis of P, then Z
′
= cZ with some complex num
ber c. This means that each oriented plane corresponds to a uniquely determined
point of the projective space CP
n+1
.
We compute the complex square of Z
Z
2
=
n+2
∑
k=1
(z
k
)
2
=[V[
2
−[W[
2
+2iV W
t
= 0.
Thus Z is an isotropic vector, and it belongs to the complexvalued quadric
Q
n+1
:=
_
Z ∈ CP
n+1
: Z
2
= 0
_
⊂CP
n+1
.
Each oriented plane corresponds onetoone to a point of the quadric Q
n+1
.
To be more precise: Consider a conformally parametrized surface X : B →R
n+2
,
then we would set
V := X
u
, W := X
v
.
In the Finslertype case for surfaces equipped with a prescribed weight matrix
W(X, Z) on the other hand we let
V := W(X, N)
1
2
◦ X
u
and W := W(X, N)
1
2
◦ X
v
.
This leads us to our next
Deﬁnition 6.5. Let the conformally parametrized immersion X : B→R
n+2
be given.
Then its GaussOsserman map is deﬁned as the complexvalued mapping
N: B −→Q
n+1
, N(w) := X
u
(w) +iX
v
(w), w ∈ B. (6.6)
Proposition 6.4. Let the conformally parametrized immersion X : B → R
n+2
be
given. Then its GaussOsserman map N is antiholomorphic if and only if X is a
minimal surface.
Proof. We compute
N
w
=N
w
=
1
2
△X = HW in B (6.7)
with the mean curvature vector H. ⊓⊔
This special analytical behaviour of Nenabled Osserman in [130], Theorem12.1
to establish the following characterisitic of the GaussOsserman map of minimal
surfaces.
Theorem 6.2. A complete minimal surface is either a plane, or its image under the
GaussOsserman map Napproaches arbitrarily closely every hyperplane in CP
n+1
.
The normals to a complete regular minimal surface in R
n+2
are everywhere dense
unless the surface is a plane.
96 6 The GaussOsserman map
6.5 FubiniStudy metric and the total curvature
The socalled FubiniStudy metric is closely related to the GaussOsserman map N.
To illustrate this connection we consider a curve Z = Z(t) in CP
n+1
, and deﬁne the
FubiniStudy metric as
_
d ˆ s
dt
_
2
:= 2
[Z ∧Z
′
[
2
[Z[
4
= 2
∑
1≤j<k≤n+2
[z
j
z
′
k
−z
k
z
′
j
[
2
_
n+2
∑
j=1
[z
j
[
2
_
2
(6.8)
where Z
′
denotes the derivative w.r.t. the parameter t. This line element d ˆ s replaces
the spherical line element of a surfaces in R
3
as follows (see Osserman [130]).
Proposition 6.5. Let the conformally parametrized minimal surface X : B →R
n+2
be given. Then it holds
_
d ˆ s
dt
_
2
= (−K)W(u
′2
+v
′2
) = (−K)
_
ds
dt
_
2
(6.9)
for the curve N(t) = X
u
(t) +iX
v
(t).
In other words, the Gauss curvature K of the minimal surface X is the negative
of the area magniﬁcation under N, i.e. setting
´
W = (−K)W, we obtain
K =−
´
W
W
=−
d ˆ s
2
ds
2
in B. (6.10)
Proof. We must evaluate
N∧N
′
= (X
u
+iX
v
) ∧
_
X
uu
u
′
+X
uv
v
′
+i(X
uv
u
′
+X
vv
v
′
)
_
= X
u
∧X
uu
u
′
+X
u
∧X
uv
v
′
−X
v
∧X
uv
u
′
−X
v
∧X
vv
v
′
+i
_
X
u
∧X
uv
u
′
+X
u
∧X
vv
v
′
+X
v
∧X
uu
u
′
+X
v
∧X
uv
v
′
_
.
The corresponding real part follows from
Re(N∧N
′
) =
_
Γ
2
11
u
′
+Γ
2
12
v
′
+Γ
1
12
u
′
+Γ
1
22
v
′
_
X
u
∧X
v
+
n
∑
σ=1
(L
σ,11
u
′
+L
σ,12
v
′
)X
u
∧N
σ
−
n
∑
σ=1
(L
σ,12
u
′
+L
σ,22
v
′
)X
v
∧N
σ
=
n
∑
σ=1
(L
σ,11
u
′
+L
σ,12
v
′
)X
u
∧N
σ
−
n
∑
σ=1
(L
σ,12
u
′
+L
σ,22
v
′
)X
v
∧N
σ
taking the Gauss equations (2.2) together with (2.6) into account.
6.5 FubiniStudy metric and the total curvature 97
Thus, using L
σ,11
+L
σ,22
= 0 we have
[Re(N∧N
′
)[
2
= W
n
∑
σ=1
(L
2
σ,11
u
′2
+L
2
σ,12
u
′2
+L
2
σ,12
v
′2
+L
2
σ,22
v
′2
+. . .
. . . +2L
σ,11
L
σ,12
u
′
v
′
+2L
σ,12
L
σ,22
u
′
v
′
)
= W
n
∑
σ=1
(L
2
σ,11
u
′2
+L
2
σ,12
u
′2
+L
2
σ,12
v
′2
+L
2
σ,22
v
′2
).
Analogously, for the imaginary part we get
Im(N∧N
′
) =
n
∑
σ=1
(L
σ,12
u
′
+L
σ,22
v
′
)X
u
∧N
σ
+
n
∑
σ=1
(L
σ,11
u
′
+L
σ,22
u
′
)X
v
∧N
σ
,
and therefore it holds
[Im(N∧N
′
)[
2
=W
n
∑
σ=1
(L
2
σ,12
u
′2
+L
2
σ,22
v
′2
+L
2
σ,11
u
′2
+L
2
σ,12
v
′2
).
Adding up both terms we arrive at (use L
2
σ,11
= L
2
σ,22
)
[N∧N
′
[
2
= W
n
∑
σ=1
(L
2
σ,11
+2L
2
σ,12
+L
2
σ,22
)(u
′2
+v
′2
)
= W
n
∑
σ=1
_
(L
σ,11
+L
σ,22
)
2
−2(L
σ,11
L
σ,22
−L
2
σ,12
)
_
(u
′2
+v
′2
)
= 2(−K)W
3
(u
′2
+v
′2
).
Moreover we have
[N[
4
=[X
u
+iX
v
[
4
= (2W)
2
= 4W
2
which ﬁnally implies
_
d ˆ s
dt
_
2
= 2
2(−K)W
3
4W
2
(u
′2
+v
′2
) = (−K)W(u
′2
+v
′2
).
This proves the statement. ⊓⊔
From
´
W = (−K)W we immediately infer Osserman’s result from [130])
Corollary 6.2. The total curvature of the minimal surface X is the negative of the
area
´
A of the image w.r.t. the metric (6.8), i.e.
´
A =
__
B
(−K)W dudv. (6.11)
98 6 The GaussOsserman map
This precisely generalizes the situation in R
3
where the curvatura integra equals
the image of the spherical mapping if we count multiplicities which eventually arise
from spherical branch points.
This leads us to the next remark: The FubiniStudy metric d ˆ s
2
for minimal sur
faces is singular at points with K = 0. But due to Corollary 5.2, in every compact
set Ω ⊂⊂R there are only ﬁnitely many singularities unless X is ﬂat.
We want to prove an upper bound for the Gaussian curvature
´
K of the Fubini
Study metric d ˆ s
2
for a minimal surface if its normal bundle is ﬂat. Hoffman and
Osserman in [93] have already shown that it holds
´
K ≤2
independently of the geometry of the normal bundle of the minimal surface. For a
proof we want to refer the reader to this paper.
In case of ﬂat normal bundles we can prover stronger estimates, and that is the
purpose of our following considerations.
Lemma 6.2. Let X : B → R
n+2
be a conformally parametrized minimal surface.
With a positive number κ
0
> 0 suppose that
χ := κ
0
−K > 0.
Furthermore let
´
K denote the Gaussian curvature of the new metric with coefﬁcients
´ g
i j
:= χg
i j
. Then there hold
χ
´
K = K−
1
2W
∆ logχ,
and therefore, using Wirtinger symbols,
χ
3
´
K = χ
2
K+
2
W
(χ
w
χ
w
−χχ
ww
).
Proof. Note that (3.11) together with
´
W = χW proves the ﬁrst identity, namely
χ
´
K = −
1
W
∆ log
_
χW = −
1
W
∆ log
√
W −
1
W
∆ log
√
χ
= K −
1
2W
∆ logχ.
Next we compute
△logχ =
1
χ
△χ −
[∇χ[
2
χ
2
=
4
χ
χ
ww
−
4
χ
2
χ
w
χ
w
.
Inserting the right hand side from here into our ﬁrst identity shows the lemma. ⊓⊔
6.5 FubiniStudy metric and the total curvature 99
Lemma 6.3. It holds
K =
2
W
2
_
W
w
W
w
W
−W
ww
_
.
Proof. We compute
K =−
1
2W
△logW =−
1
2W
_
△W
W
−
[∇W[
2
W
2
_
=
2
W
2
_
W
w
W
w
W
−W
ww
_
proving the statement. ⊓⊔
Now we come to the announced estimate of
´
K.
Theorem 6.3. Let X : B →R
n+2
be a conformally parametrized minimal surface
with ﬂat normal bundle. Then the Gauss curvature
´
K of the FubiniStudy metric d ˆ s
2
satisﬁes
´
K ≤1 in B.
Proof. Let N be a torsionfree ONF. For the complexvalued Hopf functions H
σ
there hold (see also section 5.7)
[H
σ
[
2
= 4(−K
σ
)W
2
resp.
n
∑
σ=1
[H
σ
[
2
= 4(−K)W
2
with the classical Gauss curvature K. Therefore we have
χ =
1
4W
2
n
∑
σ=1
[H
σ
[
2
+κ
0
for the function χ = κ
0
−K. For complexvalued vectors v = (v
1
, . . . , v
n
) ∈ C
n
and
w = (w
1
, . . . , w
n
) ∈ C
n
we introduce the following abbreviation
w⋆ v = v ⋆ w :=
n
∑
σ=1
v
σ
w
σ
, v
2
:= v ⋆ v =
n
∑
σ=1
v
σ
v
σ
=
n
∑
σ=1
[v
σ
[
2
.
Thus, using the Hopf vector H = (H
1
, . . . , H
n
) ∈ C
n
we can rewrite the previous
formula to get
χ =
1
4W
2
H ⋆ H +κ
0
=
1
4W
2
H 
2
+κ
0
.
Together with H
w
= 0 from (5.15) we calculate
χ
w
=
1
4W
2
H ⋆ H
w
−
W
w
2W
3
H 
2
=
1
4W
2
H ⋆ H
w
−
2W
w
W
(χ −κ
0
),
χ
w
=
1
4W
2
H ⋆ H
w
−
W
w
2W
3
H
2
=
1
4W
2
H ⋆ H
w
−
2W
w
W
(χ −κ
0
)
for the ﬁrst complex derivatives.
100 6 The GaussOsserman map
From here we further induce
χ
ww
= −
1
2W
3
_
W
w
H ⋆ H
w
+W
w
H ⋆ H
w
_
+
1
4W
2
H
w

2
−
W
ww
2W
3
H
2
+
3W
w
W
w
2W
4
H
2
.
It follows
χ
w
χ
w
−χχ
ww
=
1
16W
4
(H ⋆ H
w
)(H ⋆ H
w
)
−
1
8W
5
_
W
w
H ⋆ H
w
+W
w
H ⋆ H
w
_
H 
2
+
W
w
W
w
4W
6
H 
4
−
H
2
H
w

2
16W
4
−
κ
0
H
w

2
4W
2
+
_
W
w
H ⋆ H
w
+W
w
H ⋆ H
w
_
_
H 
2
8W
5
+
κ
0
2W
3
_
−
_
H 
2
4W
2
+κ
0
__
3W
w
W
w
2W
4
−
W
ww
2W
3
_
H 
2
=
1
16W
4
_
(H ⋆ H
w
)(H ⋆ H
w
) −H
2
H
w

2
_
−
W
w
W
w
8W
6
H 
4
+
W
ww
8W
5
H 
4
−
W
w
W
w
κ
0
H 
2
2W
4
+
W
ww
κ
0
H 
2
2W
3
−
κ
0
4W
2
_
H
w

2
−
2
W
W
w
H ⋆ H
w
−
2
W
W
w
H ⋆ H
w
+4
W
w
W
w
W
2
H 
2
_
=
1
16W
4
_
(H ⋆ H
w
)(H ⋆ H
w
) −H
2
H
w

2
_
−
H 
2
2W
3
_
W
w
W
w
W
−W
ww
__
κ
0
+
H
2
4W
2
_
−
κ
0
4W
2
_
_
_
_
H
w
−
2
W
W
w
H
_
_
_
_
2
.
6.6 Minimal surfaces with constant curvature
´
K 101
Thus we arrive at the identity
χ
w
χ
w
−χχ
ww
=
1
16W
4
_
(H ⋆ H
w
)(H ⋆ H
w
) −H
2
H
w

2
_
−
H 
2
K
4W
χ −
κ
0
4W
2
_
_
_
_
H
w
−
2
W
W
w
H
_
_
_
_
2
.
Notice that
−
H
2
K
4W
=
χ
3
2
W −
χ
2
2
KW +χ
_
κ
0
−
3
2
χ
_
κ
0
W
≤
χ
3
2
W −
χ
2
2
KW.
On the other hand, we have
(H ⋆ H
w
)(H ⋆ H
w
) −H
2
H
w

2
=
_
n
∑
σ=1
H
σ
H
σ,w
__
n
∑
σ=1
H
σ
H
σ,w
_
−
_
n
∑
σ=1
[H
σ
[
2
__
n
∑
σ=1
[H
σ,w
[
2
_
=
¸
¸
¸
¸
¸
n
∑
σ=1
H
σ
H
σ,w
¸
¸
¸
¸
¸
2
−
_
n
∑
σ=1
[H
σ
[
2
__
n
∑
σ=1
[H
σ,w
[
2
_
≤ 0.
Therefore we conclude
χ
3
´
K = χ
2
K+χ
3
−χ
2
K−
κ
0
2W
3
_
_
_
_
H
w
−
2
W
W
w
H
_
_
_
_
2
+
1
8W
5
_
(H ⋆ H
w
)(H ⋆ H
w
) −H
2
H
w

2
_
≤ χ
3
.
The statement follows. ⊓⊔
The proof presented here is motivated by Ruchert’s considerations from [137].
We will apply Ruchert’s ingenious methods again in section 13.6.
6.6 Minimal surfaces with constant curvature
´
K
We want to conclude this chapter with some interesting characterizations of the
shape of minimal surfaces in R
4
with constant curvature
´
K which we take again
from Hoffman and Osserman [93].
102 6 The GaussOsserman map
Theorem 6.4. There hold the following alternatives.
1. The generalized Gauss image of the minimal immersion X : B →R
4
has constant
Gauss curvature
´
K ≡c. Then, c = 1 or c = 2, and
−
´
K ≡1 if and only if X(B) lies fully in some R
3
;
−
´
K ≡2 if and only if X(B) is a complex curve lying fully in C
2
, represented by
R
4
endowed with some orthogonal complex structure.
2. If the generalized Gauss image of the minimal immersion X : B →R
n+2
has con
stant Gauss curvature
´
K ≡2, then X(B) lies in an afﬁne subspace R
4
⊂R
n+2
.
Part II
Variational Problems
Chapter 7
Normal Coulomb frames in R
4
7.1 The total torsion
7.2 Curves in R
3
7.3 Coulomb ONF
7.4 Construction of normal Coulomb frames
7.5 Minimality of normal Coulomb frames
7.6 A torsion estimate via the maximum principle
7.7 A torsion estimate via a RiemannHilbert problem
7.8 Estimates for the total torsion
7.9 An example: Holomorphic graphs in C
2
7.10 Application to the mean curvature ﬂow in R
4
106 7 Normal Coulomb frames in R
4
In this chapter we focus on orthonormal frames in the normal space of immersions in R
4
which
are critical for a functional of total torsion. This generalizes the concept of parallel normal frames
for immersions with nonﬂat normal bundles.
7.1 The total torsion
In sections 5.3 and 5.4 we have introduced the torsion T of an ONF N in the form
T =
1
2
2
∑
i, j=1
n
∑
σ,ϑ=1
g
i j
T
ϑ
σ,i
T
ϑ
σ, j
as well as the following functional of total torsion
T
X
[N] =
__
B
TW dudv =
1
2
2
∑
i, j=1
n
∑
σ,ϑ=1
__
B
g
i j
T
ϑ
σ,i
T
ϑ
σ, j
W dudv.
This functional does not depend on the choice of the parametrization. But it de
pends on the choice of the ONF N. In particular, if there exist a torsionfree ONF N
then obviously T
X
[N] = 0. Otherwise it is always greater then zero.
We want to construct orthonormal normal frames N which are minimal for T [N].
7.2 Curves in R
3
To illustrate the underlying idea we want to start with constructing torsionfree nor
mal frames for onedimensional curves in R
3
.
Let the regular and arclength parametrized curve c =c(s) be given together with
the orthogonal moving 3frame
_
t(s), n(s), b(s)
_
, s ∈ I ⊂R,
consisting of the unit tangential vector t(s), the unit normal vector n(s), and the unit
binormal vector b(s).
108 7 Normal Coulomb frames in R
4
These vectors are given as follows
t(s) = c
′
(s), n(s) =
c
′′
(s)
[c
′′
(s)[
, b(s) =t(s) n(s)
under the assumptions c
′
(s) ,= 0 and c
′′
(s) ,= 0 for the ﬁrst and second derivatives of
the mapping c(s). The curve’s curvature κ(s) and torsion τ(s) are
κ(s) =[t
′
(s)[
2
, τ(s) = n
′
(s) b(s) =−b
′
(s) n(s).
Consider now a new ONF (¯ n,
¯
b) resulting from
¯ n(s) := cosϕ n(s) +sinϕ b(s),
¯
b(s) :=−sinϕ n(s) +cosϕ b(s)
with a rotation angle ϕ. Again there hold
[¯ n(s)[ = 1, [
¯
b(s)[ = 1, ¯ n(s)
¯
b(s) = 0, ¯ n(s) t(s) = 0,
¯
b(s) t(s) = 0
for all s ∈ I. The torsion w.r.t. this new frame, i.e.
¯ τ(s) := ¯ n
′
(s)
¯
b(s),
can be computed explicitely as follows
¯ τ = (−ϕ
′
sinϕ n +cosϕ n
′
+ϕ
′
cosϕ b +sinϕ b
′
) (−sinϕ n +cosϕ b)
= ϕ
′
sin
2
ϕ +cos
2
ϕ n
′
b +ϕ
′
cos
2
ϕ −sin
2
ϕ b
′
n
= ϕ
′
+n
′
b.
Thus we arrive at the transformation formula for the two torsions
ϕ
′
(s) = ¯ τ(s) −τ(s).
In particular, solving the ordinary differential equation ϕ
′
(s) = −τ(s), express
ing the property ¯ τ(s) ≡ 0, yields a rotation angle function ϕ = ϕ(s) which trans
forms the given ONF (n, b) into a new torsionfree ONF (¯ n,
¯
b).
Such frames are also called parallel for their Fren´ et equations do not contain
normal components.
The following considerations are devoted to the construction of torsionfree or
thonormal normal frames for surfaces in space R
4
. This problem becomes more
intricate because it will turn out that such a construction reduces to solving a partial
differential equation instead of an ordinary differenial equation.
7.3 Coulomb ONF 109
7.3 Coulomb ONF
Let the immersion X : B →R
4
together with an ONF N = (N
1
, N
2
) be given. We set
¯
N
1
= cosϕ N
1
+sinϕ N
2
,
¯
N
2
=−sinϕ N
1
+cosϕN
2
(7.1)
for a new ONF (
¯
N
1
,
¯
N
2
) resulting from the action of the SO(2)group realized by a
rotation angle ϕ ∈ [0, 2π]. Then the situation now is comparable to the situation of
onedimensional curves in R
3
.
Lemma 7.1. The torsion coefﬁcients of the ONF ¦
¯
N
1
,
¯
N
2
¦ are given by
¯
T
2
1,1
= T
2
1,1
+ϕ
u
,
¯
T
2
1,2
= T
2
1,2
+ϕ
v
. (7.2)
Proof. Taking account of N
σ
N
σ,u
i = 0 we compute
¯
T
2
1,1
= (−ϕ
u
sinϕ N
1
+cosϕ N
1,u
+ϕ
u
cosϕ N
2
+sinϕ N
2,u
) . . .
. . . (−sinϕ N
1
+cosϕ N
2
)
= ϕ
u
sin
2
ϕ −sin
2
ϕ T
1
2,1
+cos
2
ϕ T
2
1,1
+ϕ
u
cos
2
ϕ
= T
2
1,1
+ϕ
u
,
and analoguously it follows
¯
T
2
1,2
= T
2
1,2
+ϕ
v
. ⊓⊔
The remaining components T
σ
σ,i
vanish identically.
We are particularly interested in the difference
△T
X
:=T
X
[
¯
N] −T
X
[N],
i.e. in critical points of the functional of total torsion. For this purpose we introduce
conformal parameters (u, v) ∈ B and calculate
△T
X
=
__
B
[∇ϕ[
2
dudv +2
__
B
(T
2
1,1
ϕ
u
+T
2
1,2
ϕ
v
)dudv
=
__
B
[∇ϕ[
2
dudv +2
_
∂B
(T
2
1,1
, T
2
1,2
) ν ϕ ds −2
__
B
div(T
2
1,1
, T
2
1,2
)ϕ dudv
(7.3)
by partial integration, ν being the outer unit normal vector ν at the boundary ∂B.
In general, the right hand side of this identity does not vanishes. In this way we
immediately obtain the EulerLagrange equation for a T
X
critical ONF.
Proposition 7.1. Let the ONF (N
1
, N
2
) of the conformally parametrized immersion
X : B →R
4
be critical for the functional of total torsion T
X
[N]. Then there hold
div(T
2
1,1
, T
2
1,2
) = 0 in
˚
B, (T
2
1,1
, T
2
1,2
) ν = 0 on ∂B. (7.4)
110 7 Normal Coulomb frames in R
4
This leads us directly to the following
Deﬁnition 7.1. An ONF N satisfying (7.4) is called a normal Coulomb frame.
The terminology Coulomb frame was suggested by F. Helein. It is motivated
by the similarity of (7.4) to differential equations from electromagnetism where a
vector ﬁeld A ∈ R
3
is called Coulomb gauged if it satisﬁes divA =0 with the spatial
divergence operator.
7.4 Construction of normal Coulomb frames
We want to follow up on the question of constructing a T
X
[N]critical frame
(N
1
, N
2
) from a given ONF (
¯
N
1
,
¯
N
2
).
If (N
1
, N
2
) is actually critical for T
X
[N] then we know
0 = div(T
2
1,1
, T
2
1,2
) = div(
¯
T
2
1,1
−ϕ
u
,
¯
T
2
1,2
−ϕ
v
) in
˚
B,
0 = (T
2
1,1
, T
2
1,2
) ν = (
¯
T
2
1,1
−ϕ
u
,
¯
T
2
1,2
−ϕ
v
) ν on ∂B,
by virtue of (7.2) and (7.4). This implies our next result.
Proposition 7.2. Let the conformally parametrized immersion X : B →R
4
be given.
Its ONF (
¯
N
1
,
¯
N
2
) transforms into a T
X
[N]critical ONF (N
1
, N
2
) by means of (7.2)
if and only if the new torsion coefﬁcients satisfy the following Neumann boundary
value problem
∆ϕ = div(
¯
T
2
1,1
,
¯
T
2
1,2
) in
˚
B,
∂ϕ
∂ν
= (
¯
T
2
1,1
,
¯
T
2
1,2
) ν on ∂B
(7.5)
with a critical rotation angle ϕ =ϕ(u, v).
It is well known that the Neumann problem
∆ϕ = f in
˚
B,
∂ϕ
∂ν
= g on ∂B
has a solution if and only if the integrability condition
__
B
f dudv =
_
∂B
gds
holds true. Obviously this condition is fulﬁlled in our situation, and thus there actu
ally exist a normal Coulomb frame N. Note that such a frame is not uniquely deter
mined because a rotation of the whole normal Coulomb frame N about an arbitrary
global constant angle ϕ
0
does not affect the EulerLagrange equation.
7.6 A torsion estimates via the maximum principle 111
7.5 Minimality of normal Coulomb frames
Although normal Coulomb frames are not uniquely determined we can prove that
they actually minimize the functional of total torsion.
For this purpose let N = (N
1
, N
2
) be a T
X
[N]critical ONF. Applying the above
SO(2)action with a rotation angle ϕ we infer
T [
¯
N] =T [N] +2
__
B
[∇ϕ[
2
dudv
taking (7.3) together with the EulerLagrange equation (7.4) into account.
Proposition 7.3. A normal Coulomb frame N of the immersion X : B →R
4
mini
mizes the functional T
X
[N] of total torsion, i.e. it holds
T [N] ≤T [
¯
N]
for all normal frames
¯
N resulting from (7.1).
7.6 A torsion estimates via the maximum principle
Let (u, v) ∈ B be conformal parameters. We want to establish upper bounds for the
torsion coefﬁcients of normal Coulomb frames of X : B →R
4
.
Assume ﬁrst that the surface X has ﬂat normal bundle. Then our considerations
from section 3.7 yield
S
2
1,12
=∂
v
T
2
1,1
−∂
u
T
2
1,2
= div(−T
2
1,2
, T
2
1,1
) ≡0 in B (7.6)
for the only nontrivial component of the curvature tensor of the normal bundle.
Furthermore, from the EulerLagrange equation (7.4) we infer that for a T
X
[N]
critical ONF N, the vector ﬁeld (−T
2
1,2
, T
2
1,1
) is parallel to the outer unit normal
vector ν along ∂B. Thus partial integration gives us
__
B
S
2
1,12
dudv =
_
∂B
(−T
2
1,2
, T
2
1,1
) ν ds =±
_
∂B
_
(T
2
1,1
)
2
+(T
2
1,2
)
2
ds.
In particular, in case of ﬂat normal bundles with S
2
1,12
≡0 we ﬁnd
T
ϑ
σ,i
≡0 on ∂B
for all i = 1, 2 and σ, ϑ = 1, 2.
112 7 Normal Coulomb frames in R
4
On the other hand, differentiating (7.6) and taking the EulerLagrange equation
(7.4) into account, we arrive at
∆T
2
1,1
=
∂
∂v
S
2
1,12
= 0, ∆T
2
1,2
=−
∂
∂u
S
2
1,12
= 0
for ﬂat normal bundles. Therefore T
2
1,1
and T
2
1,2
are harmonic functions, and the
maximum principle implies
T
ϑ
σ,i
≡0 in B for all i = 1, 2 and σ, ϑ = 1, 2. (7.7)
This now holds true for all parametrizations of class P.
Theorem 7.1. A normal Coulomb frame N of an immersion X : B →R
4
with ﬂat
normal bundle is free of torsion.
In summary we have proved existence of regular normal Coulomb frames, and
if additionally the normal bundle of the surface is ﬂat then those frames are free of
torsion. Existence results concerning such situations in case of manifolds immersed
in Euclidean spaces of higher dimension can be found e.g. in Chen [27].
Next we want to consider the case of nonﬂat normal bundles. Due to the Euler
Lagrange equation (7.4), the torsion vector (T
2
1,1
, T
2
1,2
) of a normal Coulomb frame
is divergencefree. Thus the differential 1form
ω :=−T
2
1,2
du +T
2
1,1
dv
is closed, i.e.
dω =∂
u
T
2
1,1
du ∧dv −∂
v
T
2
1,2
dv ∧du = div(T
2
1,1
, T
2
1,2
)du ∧dv = 0.
Now from Poincare’s lemma we infer the existence of a C
3
regular function τ
with the property
dτ =τ
u
du +τ
v
du =ω, i.e. ∇τ = (−T
2
1,2
, T
2
1,1
).
Differentiating, taking again S
2
1,12
=∂
v
T
2
1,1
−∂
u
T
2
1,2
into account, leads to the homo
geneous boundary value problem
△τ = S
2
1,12
in
˚
B, τ = 0 on ∂B. (7.8)
To justify the homogeneous boundary values note that ∇τ (−v, u) = 0 on ∂B
since (−v, u) ∈ ∂B represents a tangential vector. We conclude τ = const on ∂B.
But τ is only deﬁned up to a constant which can be chosen such that (7.8) is true.
Next we want to establish bounds for τ and its gradient.
7.6 A torsion estimates via the maximum principle 113
First use Poisson’s representation formula for the solution τ of the foregoing
Dirichlet boundary value problem, i.e.
τ(w) =
__
B
Φ(ζ; w)S
2
1,12
(ζ)dξdη, ζ = (ξ, η), (7.9)
with the nonpositive Green’s function
Φ(ζ; w) :=
1
2π
log
¸
¸
¸
¸
ζ −w
1 −wζ
¸
¸
¸
¸
, ζ ,= w, (7.10)
for the Laplace operator △(see e.g. Sauvigny [143], chapter VIII, §1). Note that
ψ(w) =
[w[
2
−1
4
solves △ψ = 1 in
˚
B and ψ = 0 on ∂B.
Therefore Poisson’s representation formula yields
__
B
[Φ(ζ; w)[ dξdη =
1 −[w[
2
4
≤
1
4
.
Lemma 7.2. The solution τ from (7.8), (7.9) satisﬁes
[τ(w)[ ≤
1
4
S
2
1,12

C
0
(B)
. (7.11)
This is nothing else but a maximumprinciple for the above homogeneous bound
ary value problem for the integral function τ.
Potential theoretic estimates for the Laplacian (see e.g. Sauvigny [143], chapter
IX, §4, Satz 1) ensure ﬁnally the existence of a real constant C =C(α) such that
τ
C
2+α
(B)
≤C(α)S
2
1,12

C
α
(B)
(7.12)
for all α ∈ (0, 1). But the gradient of τ codiﬁes the torsion coefﬁcients. Thus we
have proved the main result of this section.
Theorem 7.2. Let the conformally parametrized immersion X : B →R
4
with nor
mal bundle of curvature S
2
1,12
be given. Then the torsion coefﬁcients of a normal
Coulomb frame N = (N
1
, N
2
) satisfy
T
ϑ
σ,i

C
1+α
(B)
≤C(α)S
2
1,12

C
α
(B)
(7.13)
for all α ∈ (0, 1) with the constant C(α) from (7.12).
We particularly recover (7.7) for ﬂat normal bundles with S
2
1,12

C
α
(B)
= 0.
114 7 Normal Coulomb frames in R
4
7.7 A torsion estimate via a RiemannHilbert problem
Once again, let us consider (7.4) and (7.6) for normal Coulomb frames
∂
∂u
T
2
1,1
+
∂
∂v
T
2
1,2
= 0,
∂
∂v
T
2
1,1
−
∂
∂u
T
2
1,2
= S
2
1,12
.
We immediately obtain
Lemma 7.3. The complexvalued torsion
Ψ := T
2
1,1
−iT
2
1,2
∈ C
solves the nonhomogeneous CauchyRiemann equation
Ψ
w
=
1
2
(∂
u
Ψ +i∂
v
Ψ) =
1
2
(∂
u
T
2
1,1
+∂
v
T
2
1,2
) +
i
2
(∂
v
T
2
1,1
−∂
u
T
2
1,2
) =
i
2
S
2
1,12
(7.14)
in the open disc
˚
B.
Note that the right hand side here is purely imaginary. In addition we write the
boundary condition from (7.4) as
Re
_
wΨ(w)
¸
= Re
_
(T
2
1,1
−iT
2
1,2
)(u +iv)
¸
= T
2
1,1
u +T
2
1,2
v = (T
2
1,1
, T
2
1,2
) ν (7.15)
on the boundary curve ∂B with its outer unit normal vector ν = (u, v).
The relations (7.14) and (7.15) form what is called a linear RiemannHilbert
problem for Ψ. There is a big machinery to treat such mathematical problem using
tools from complex analysis. We want to refer the reader e.g. to Begehr and Wen
[10], Courant and Hilbert [40], Sauvigny [143], Vekua [161], or Wendland [164] for
a ﬁrst orientation as well as various detailed studies.
Below we construct an explicit solution of our RiemannHilbert problem. As it
already turns out now this solution is the only possible.
Lemma 7.4. The RiemannHilbert problem (7.14), (7.15) possesses at most one so
lution Ψ ∈C
1
(
˚
B, C) ∩C
0
(B, C).
Proof. Assume there are two such solutions Ψ
1
and Ψ
2
. Then we set
Φ(w) := w[Ψ
1
(w) −Ψ
2
(w)]
and compute
Φ
w
= 0 in
˚
B, ReΦ = 0 on ∂B.
Consequently, it holds Φ ≡ ic in B with some real constant c ∈ R. But Φ(w) is a
continuous function, and Φ(0) = 0 implies c = 0. Thus Ψ
1
≡Ψ
2
. ⊓⊔
Following Vekua [161] (see also Sauvigny [143]) we will solve our Riemann
Hilbert problem in terms of socalled generalized analytic functions. Here come
some facts about this important class of complexvalued functions.
7.7 A torsion estimate via a RiemannHilbert problem 115
For arbitrary f ∈C
1
(B, C) we deﬁne Cauchy’s integral operator by
T
B
[ f ](w) :=−
1
π
__
B
f (ζ)
ζ −w
dξdη, w ∈ C. (7.16)
Lemma 7.5. There hold T
B
[ f ] ∈C
1
(C¸ ∂B) ∩C
0
(C) as well as
∂
∂w
T
B
[ f ](w) =
_
f (w), w ∈
˚
B
0, w ∈ C¸ B
. (7.17)
Proof. We want to verify brieﬂy the validity of the stated complexvalued differ
ential equation, see Sauvigny [143], chapter IV, §5. Let ¦G
k
¦
k∈N
be a sequence of
open, simply connected and smoothly bounded domains contracting to some point
z
0
∈ B for k → ∞. Let [G
k
[ denote the respective areas. Making use of Cauchy’s
integral formula
_
G
k
g(w)
w−ζ
dw = 2πig(ζ)
we compute with the characteristic functions χ
k
1
2i[G
k
[
_
∂G
k
T
B
[ f ](w)dw =
1
2i[G
k
[
_
∂G
k
_
_
−
1
π
__
B
f (ζ)
ζ −w
dξdη
_
_
dw
=
1
2πi[G
k
[
__
B
_
_
f (ζ)
_
∂G
k
1
w−ζ
dw
_
_
dξdη
=
1
2πi[G
k
[
__
B
f (ζ) 2πiχ
G
k
(ζ)dξdη
=
1
[G
k
[
__
G
k
f (ζ)dξdη.
Now recalling the integration by parts rule in complex form
__
G
k
d
dw
f (w)dξdη =
1
2i
_
∂G
k
f (z)dz
we get in the limit
d
dw
T
B
[ f ](w) = lim
k→∞
1
2i[G
k
[
_
∂G
k
T
B
[ f ](w)dw = f (z
0
)
for all z
0
∈ B. The statement follows. ⊓⊔
116 7 Normal Coulomb frames in R
4
Next we set
P
B
[ f ](w) := −
1
π
__
B
_
f (ζ)
ζ −w
+
ζ f (ζ)
1 −wζ
_
dξ dη
= T
B
[ f ](w) +
1
w
T
B
[wf ]
_
1
w
_
.
Now Satz 1.24 from Vekua [161] states the following
Lemma 7.6. With the deﬁnitions above we have the uniform estimate
¸
¸
P
B
[ f ](w)
¸
¸
≤C(p) f 
L
p
(B)
, w ∈ B,
where p ∈ (2, +∞], and C(p) is a positive constant dependent only on p.
Using this result (which remains unproved here) we obtain the following main
result of this section.
Theorem 7.3. Let the conformally parametrized immersion X : B →R
4
be given.
Then the complexvalued torsion Ψ of a normal Coulomb frame N satisﬁes
[Ψ(w)[ ≤c(p)S
2
1,12

L
p
(B)
for all w ∈ B
with some positive constant c(p) and p ∈ (2, +∞].
For ﬂat normal bundles we again verify that a normal Coulomb frame is free of
torsion: T
ϑ
σ,i
≡0.
Proof. Let us write
f :=
i
2
S
2
1,12
∈C
1
(B, C)
to apply the foregoing results. We claim that the complexvalued torsion Ψ pos
sesses the integral representation
Ψ(w) = P
B
[ f ](w) =−
1
π
__
B
_
f (ζ)
ζ −w
+
ζ f (ζ)
1 −wζ
_
dξ dη, w ∈
˚
B.
Then the stated estimate follows at once from the above lemma. First we claim
wP
B
[ f ](w) =
1
π
__
B
f (ζ)dξ dη +T
B
[wf ](w) −T
B
[wf ]
_
1
w
_
.
7.7 A torsion estimate via a RiemannHilbert problem 117
Let us check this identity:
1
π
__
B
f (ζ)dξdη +T
B
[wf ](w) −T
B
[wf ](w
−1
)
=
1
π
__
B
f (ζ)dξdη −
1
π
__
B
ζ f (ζ)
ζ −w
dξdη −T
B
[wf ](w
−1
)
= −
w
π
__
B
f (ζ)
ζ −w
dξdη +
1
π
__
B
ζ f (ζ)
ζ −
1
w
dξdη
= −
w
π
__
B
f (ζ)
ζ −w
dξdη +
w
π
__
B
ζ f (ζ)
ζ w−1
dξdη
= −
w
π
__
B
_
f (ζ)
ζ −w
+
ζ f (ζ)
1 −ζ w
_
dξdη
which already shows the stated identity.
Next, taking f =
i
2
S
2
1,12
into account, we infer
Re
_
wP
B
[ f ](w)
_
= 0, w ∈ ∂B,
what follows from
T
B
[
1
2
iwS
2
1,12
](w) −T
B
[
1
2
iwS
2
1,12
](w
−1
)
= −
1
π
__
B
i
2
ζS
2
1,12
ζ −w
dξdη +
1
π
__
B
i
2
ζS
2
1,12
ζ −w
−1
= −
i
2π
__
B
_
ζ
ζ −w
+
ζ
ζ −
1
w
_
S
2
1,12
dξdη.
The entry in the brackets is a real number because
1
w
=
w
[w[
2
= w holds true on the
boundary ∂B. Investing additionally
∂
∂w
P
B
[ f ](w) = f (w)
due to (7.17) and our representation of P
B
[ f ](w) we conclude that P
B
[ f ](w) solves
the RiemannHilbert problem for Ψ. The previous uniqueness result proves the
stated representation. ⊓⊔
Note that our proof relies crucially on the fact that f =
i
2
S
2
1,12
is purely imagi
nary!
118 7 Normal Coulomb frames in R
4
7.8 Estimates for the total torsion
The previous results allow us immediately to establish lower and upper bounds for
the total torsion T
X
[N] for normal Coulomb frames N. Namely, let X : B →R
4
be
conformally parametrized such that
T
X
[N] =
__
B
_
(T
2
1,1
)
2
+(T
2
1,2
)
2
_
dudv.
Thus Theorem 7.2 gives
T
X
[N] ≤2C(α)
2
S
2
1,12

2
C
α
(B)
__
B
1dudv = 2πC(α)
2
S
2
1,12

2
C
α
(B)
.
Analogously, Theorem 7.3 yields
T
X
[N] ≤πc(p)
2
S
2
1,12

2
L
p
(B)
, p > 2.
We want to reformulate these two results using the invariant scalar curvature
S =W
−1
S
2
1,12
of the normal bundle.
Theorem 7.4. Let the conformally parametrized immersion X : B → R
4
together
with a normal Coulomb frame N be given. Then there hold
T
X
[N] ≤2πC(α)
2
SW
2
C
α
(B)
for all α ∈ (0, 1) with the real constant C =C(α) from Theorem 7.2, as well as
T
X
[N] ≤πc(p)
2
SW
2
L
p
(B)
for all p ∈ (2, +∞] with the real constant c = c(p) from Theorem 7.3.
The following lower bound for the total torsion of normal Coulomb frames N is
a special case of a general estimate which we will prove in the next chapter.
Theorem 7.5. Let the conformally parametrized immersion X : B → R
4
together
with a normal Coulomb frame N be given. Assume S
2
1,12
,≡0. Then it holds
T
X
[N] ≥
_
S
2
1,12

2
L
2
(B)
2(1 −ρ)
2
S
2
1,12

2
L
2
(B
ρ
)
+
∇S
2
1,12

2
L
2
(B)
S
2
1,12

2
L
2
(B
ρ
)
_
−1
S
2
1,12

2
L
2
(B
ρ
)
> 0
where ρ =ρ(S
2
1,12
) ∈ (0, 1) is chosen such that
S
2
1,12

2
L
2
(B
ρ
)
=
__
B
ρ
(0)
[S
2
1,12
[
2
dudv > 0.
7.9 An example: Holomorphic graphs in C
2
119
7.9 An example: Holomorphic graphs in C
2
Let us again consider minimal surface graphs
X(w) = (w, Φ(w)), w = u +iv ∈ B,
with a holomorphic function Φ =ϕ +iψ. Then the Euler unit normal vectors
´
N
1
=
1
√
W
(−ϕ
u
, −ϕ
v
, 1, 0),
´
N
2
=
1
√
W
(−ψ
u
, −ψ
v
, 0, 1) (7.18)
form an orthonormal normal frame with the area element
W = 1 +[∇ϕ[
2
= 1 +[Φ
w
[
2
.
Note here that
Φ
w
=ϕ
w
+iψ
w
=
1
2
_
ϕ
u
−iϕ
v
+iψ
u
+ψ
v
_
=ϕ
u
−iϕ
v
due to ϕ
u
=ψ
v
, ϕ
v
=−ψ
u
. We especially infer
∆ϕ =∆ψ = 0,
i.e. X represents a conformally parametrized minimal graph in R
4
.
For the torsion coefﬁcients we compute
T
2
1,1
=
´
N
1,u
´
N
2
=
1
W
(−ϕ
uu
ϕ
v
+ϕ
uv
ϕ
u
) =
1
2W
∂
∂v
([∇ϕ[
2
),
T
2
1,2
=
´
N
1,v
´
N
2
= −
1
2W
∂
∂u
([∇ϕ[
2
).
Consequently it holds
div(T
2
1,1
, T
2
1,2
) = 0 in B.
In order to check the boundary condition in (7.4) we introduce polar coordinates
u = r cosα, v = r sinα. Since
1
r
∂
∂α
= u
∂
∂v
−v
∂
∂u
we obtain
(T
2
1,1
, T
2
1,2
) ν =
1
2W
_
u
∂
∂v
−v
∂
∂u
_
[∇ϕ[
2
=
1
2W
∂
∂α
[Φ
w
[
2
on ∂B.
Proposition 7.4. Given the conformally parametrized minimal graph (w, Φ(w))
with a holomorphic function Φ =ϕ +iψ. Then the Euler ONF ¦N
1
, N
2
¦ from (7.18)
represents a normal Coulomb frame if and only if [Φ
w
[ is constant on ∂B.
In particular, this result applies to graphs X(w) = (w, w
n
) for arbitrary n ∈ N!
120 7 Normal Coulomb frames in R
4
7.10 Application to the mean curvature ﬂow in R
4
We consider a family of immersions X(u, v; τ) evolving by the mean curvature ﬂow
in direction of the mean curvature vector H as follows
X
τ
≡
∂X
∂τ
=−
2
∑
σ=1
H
σ
N
σ
=−H.
We want to compute the evolution of the scalar curvature S = W
−1
S
2
1,12
of the
normal bundle under this ﬂow. Let N = (N
1
, N
2
) be an initial ONF.
Lemma 7.7. The evolution of an unit normal vector N
σ
reads
N
σ,τ
=
2
∑
i, j=1
g
i j
_
H
σ,u
i +
2
∑
ε=1
H
ε
T
σ
ε,i
_
X
u
j +
2
∑
ε=1
T
ε
σ,τ
N
ε
(7.19)
for σ = 1, 2 with the τdirected torsion coefﬁcients
T
ε
σ,τ
= N
σ,τ
N
ε
=−N
σ
N
ε,τ
=−T
σ
ε,τ
.
Proof. We make the ansatz
N
σ,τ
=
2
∑
i, j=1
g
i j
(N
σ,τ
X
u
i )X
u
j +
2
∑
ε=1
(N
σ,τ
N
ε
)N
ε
=
2
∑
i, j=1
g
i j
(N
σ,τ
X
u
i )X
u
j +
2
∑
ε=1
T
ε
σ,τ
N
ε
.
On the other hand, due to the mean curvature ﬂow equation, differentiation of the
identity X
u
i N
σ
= 0 w.r.t. τ yields
N
σ,τ
X
u
i = −X
u
i
τ
N
σ
=
_
2
∑
ε=1
H
ε
N
ε
_
u
i
N
σ
= (H
ε,u
i )N
ε
N
σ
+
2
∑
ε=1
H
ε
T
ϑ
ε,i
N
ϑ
N
σ
proving the statement. ⊓⊔
Thus setting
a
i
1
:=
2
∑
j=1
g
i j
(H
1,u
j +H
2
T
1
2, j
), a
i
2
:=
2
∑
j=1
g
i j
(H
2,u
j +H
1
T
2
1, j
)
we ﬁnd
N
1,τ
=
2
∑
i=1
a
i
1
X
u
i +T
2
1,τ
N
2
, N
2,τ
=
2
∑
i=1
a
i
2
X
u
i +T
1
2,τ
N
1
.
7.10 Application to the mean curvature ﬂow in R
4
121
Theorem 7.6. The evolution of the scalar curvature S of the normal bundle under
the mean curvature ﬂow is given by divergence term
S
τ
=
2
∑
i=1
(−a
i
1
L
2,i2
+a
i
2
L
1,i2
)
u
+
2
∑
i=1
(a
i
1
L
2,i1
−a
i
2
L
1,i1
)
v
.
Proof. We calculate
∂T
2
1,1
∂τ
= (N
1,u
N
2
)
τ
= (N
1,τu
) N
2
+N
1,u
N
2,τ
=
_
2
∑
i=1
a
i
1
X
u
i +T
2
1,τ
N
2
_
u
N
2
+. . .
. . . +
_
−
2
∑
m,n=1
L
1,1m
g
mn
X
u
n +T
2
1,1
N
2
_
_
2
∑
i=1
a
i
2
X
u
i +T
1
2,τ
N
1
_
=
2
∑
i=1
a
i
1
X
u
i
u
N
2
+
∂T
2
1,τ
∂u
−
2
∑
i=1
2
∑
m,n=1
a
i
2
L
1,1m
g
mn
g
ni
=
2
∑
i=1
a
i
1
L
2,i1
+
∂T
2
1,τ
∂u
−
2
∑
i=1
a
i
2
L
1,1i
as well as
∂T
2
1,2
∂τ
= (N
1,v
N
2
)
τ
= (N
1,τv
) N
2
+N
1,v
N
2,τ
=
2
∑
i=1
a
i
1
X
u
i
v
N
2
+
∂T
2
1,τ
∂v
−
2
∑
i=1
2
∑
m,n=1
a
i
2
L
1,2m
g
mn
g
ni
=
2
∑
i=1
a
i
1
L
2,i2
+
∂T
2
1,τ
∂v
−
2
∑
i=1
a
i
2
L
1,2i
.
It follows that
S
τ
=
∂
2
T
2
1,1
∂τ∂v
−
∂
2
T
2
1,2
∂τ∂u
=
2
∑
i=1
div(−a
i
1
L
2,i2
+a
i
2
L
1,i2
, a
i
1
L
2,i1
−a
i
2
L
1,i1
) −
∂
2
T
2
1,τ
∂u∂v
+
∂
2
T
2
1,τ
∂v∂u
=
2
∑
i=1
div(−a
i
1
L
2,i2
+a
i
2
L
1,i2
, a
i
1
L
2,i1
−a
i
2
L
1,i1
)
as stated. ⊓⊔
122 7 Normal Coulomb frames in R
4
Evolution of the torsion coefﬁcients
We want to refer the reader’s attention to an interesting artefact arising with the
evolution of the torsion coefﬁcients T
ϑ
σ,i
. Consider for this purpose the general case
n ≥2 of arbitrary codimensions.
Lemma 7.8. It holds
N
σ,u
i ∂
τ
N
ϑ
=−
2
∑
j,k=1
L
σ,i j
g
jk
H
ϑ,u
k +
2
∑
i, j=1
n
∑
ω=1
H
ω
L
σ,i j
g
jk
T
ω
ϑ,k
−
n
∑
ω=1
T
ω
σ,i
T
ϑ
ω,τ
for all i = 1, 2 and σ, ϑ = 1, . . . , n.
Proof. We compute
N
σ,u
i ∂
τ
N
ϑ
=
2
∑
j,k=1
g
k j
H
ϑ,u
k X
u
j N
σ,u
i −
2
∑
j,k=1
n
∑
ω=1
g
k j
T
ω
ϑ,k
H
ω
X
u
j N
σ,u
i
+
n
∑
ω=1
T
ω
ϑ,τ
N
ω
N
σ,u
i
= −
n
∑
j,k=1
g
k j
H
ϑ,u
k L
σ,i j
+
2
∑
j,k=1
n
∑
ω=1
L
σ,i j
g
k j
H
ω
T
ω
ϑ,k
+
n
∑
ω=1
T
ω
ϑ,τ
T
ω
σ,i
proving the statement. ⊓⊔
Lemma 7.9. It holds
∂
τ
N
σ,u
i N
ϑ
=
2
∑
r,s=1
g
rs
H
σ,u
r L
ϑ,is
−
2
∑
r,s=1
n
∑
ω=1
g
rs
H
ω
T
ω
σ,r
L
ϑ,is
+∂
u
i T
ϑ
σ,τ
+
n
∑
ω=1
T
ω
σ,τ
T
ϑ
ω,i
for all i = 1, 2 and σ, ϑ = 1, . . . , n.
Proof. For we compute
∂
τ
N
σ,u
i N
ϑ
=
2
∑
r,s=1
g
rs
H
σ,u
r X
u
i
u
s N
ϑ
−
2
∑
r,s=1
n
∑
ω=1
g
rs
H
ω
T
ω
σ,r
X
u
i
u
s N
ϑ
+
n
∑
ω=1
∂
u
i T
ω
σ,τ
δ
ωϑ
+
n
∑
ω=1
T
ω
σ,τ
N
ω,u
i N
ϑ
=
2
∑
r,s=1
g
rs
H
σ,u
r L
σ,is
−
2
∑
r,s=1
n
∑
ω=1
g
rs
H
ω
T
ω
σ,r
L
ϑ,is
+∂
u
i T
ϑ
σ,τ
+
n
∑
ω=1
T
ω
σ,τ
T
ϑ
ω,i
proving the statement. ⊓⊔
7.10 Application to the mean curvature ﬂow in R
4
123
Using these two results we are able to compute the evolution of the torsion coef
ﬁcients.
Proposition 7.5. It holds
∂
τ
T
ϑ
σ,i
=
2
∑
j,k=1
g
jk
(L
ϑ,ik
H
σ,u
i −L
σ,i j
H
ϑ,u
k ) +
2
∑
j,k=1
n
∑
ω=1
g
jk
H
ω
(L
σ,i j
T
ω
ϑ,k
−L
ϑ,ik
T
ω
σ, j
)
+∂
u
i T
ϑ
σ,τ
+
n
∑
ω=1
(T
ω
σ,τ
T
ϑ
ω,i
−T
ω
σ,i
T
ϑ
ω,τ
)
Proof. This follows from
∂
τ
T
ϑ
σ,i
= ∂
τ
(N
σ,u
i N
ϑ
) = ∂
τ
N
σ,u
i N
ϑ
+N
σ,u
i ∂
τ
N
ϑ
=
2
∑
r,s=1
g
rs
H
σ,u
r L
ϑ,is
−
2
∑
r,s=1
n
∑
ω=1
g
rs
H
ω
T
ω
σ,r
L
ϑ,is
+∂
u
i T
ϑ
σ,τ
+
n
∑
ω=1
T
ω
σ,τ
T
ϑ
ω,i
= −
2
∑
j,k=1
L
σ,i j
g
jk
H
ϑ,u
k +
2
∑
j,k=1
n
∑
ω=1
L
σ,i j
g
jk
H
ω
T
ω
ϑ,k
−
n
∑
ω=1
T
ω
σ,i
T
ϑ
ω,τ
=
2
∑
j,k=1
g
jk
(L
ϑ,ik
H
σ,u
j −L
σ,i j
H
ϑ,u
k ) +
2
∑
j,k=1
n
∑
ω=1
g
jk
H
ω
(L
σ,i j
T
ω
ϑ,k
−L
ϑ,ik
T
ω
σ, j
)
+∂
u
i T
ϑ
σ,τ
+
n
∑
ω=1
(T
ω
σ,τ
T
ϑ
ω,i
−T
ω
σ,i
T
ϑ
ω,τ
)
as stated. ⊓⊔
Let us now consider the evolution of a minimal surface in R
4
under mean curva
ture ﬂow with the property H ≡0 for all times τ. We obtain
∂
τ
T
2
1,1
=∂
u
T
2
1,τ
, ∂
τ
T
2
1,2
=∂
v
T
2
1,τ
and even though ﬂow does not affect the geometry of the surfaces it actually has
consequences of the evolution of the ONF!
Gauge of the time torsion coefﬁcients
The idea to ensure that the whole ONF remains unaffected in this special case just
discussed is to introduce an additional gauge for the time torsion coefﬁcients T
ϑ
σ,τ
.
This involves various technical difﬁculties even in case n = 2. Hence we want
to present an algorithm which at least guarantees that a normal Coulomb frame
actually remains a normal Coulomb frame under the ﬂow.
124 7 Normal Coulomb frames in R
4
For this purpose we ﬁrst have to establish the EulerLagrange equation for normal
Coulomb frames in arbitrary parametizations. Using our SO(2)action
¯
N
1
= cosϕN
1
+sinϕN
2
,
¯
N
2
=−sinϕN
1
+cosϕN
2
we proceed as follows:
T
X
[
¯
N] =
2
∑
i, j=1
__
B
g
i j
(T
2
1,i
+ϕ
u
i )(T
2
1, j
+∂
u
j )W dudv
= T
X
[N] +
2
∑
i, j=1
__
B
g
i j
(T
2
1,i
ϕ
u
j +T
2
1, j
ϕ
u
i )W dudv +o(ϕ
u
, ϕ
v
)
= T
X
[N] +2
2
∑
j=1
__
B
(g
j1
WT
2
1, j
, g
j2
WT
2
1, j
) (ϕ
u
, ϕ
v
)dudv +o(ϕ
u
, ϕ
v
)
= T
X
[N] −2
2
∑
i, j=1
__
B
∂
u
i (g
i j
WT
2
1, j
)ϕ dudv −2
2
∑
i, j=1
_
∂B
g
i j
WT
2
1, j
ν
i
ds
+o(ϕ
u
, ϕ
v
)
with the outer unit normal vector ν = (ν
1
, ν
2
) at the boundary curve ∂B. Thus we
have proved the
Lemma 7.10. A normal Coulomb frame N = (N
1
, N
2
) for the immersion X : B →R
4
solves the Neumann boundary value problem
2
∑
i, j=1
∂
u
i (Wg
i j
T
ϑ
σ, j
) = 0 in B,
2
∑
i, j=1
(Wg
i j
T
ϑ
σ,i
) ν
j
= 0 in B.
In the special case of a conformally parametrized immersion X : B → R
4
we
recover our EulerLagrange equation from (7.4), namely
div(T
2
1,1
, T
2
1,2
) = 0 in B, (T
2
1,1
, T
2
1,2
) ν = 0 on ∂B.
Now back to our problem: Differentiation of the EulerLagrange equation taking
account of the identity
∂
τ
(Wg
i j
) = 0 for i, j = 1, 2
since the minimal surface remains ﬁxed under mean curvature ﬂow, we infer
0 =
2
∑
i, j=1
∂
u
i ∂
τ
(Wg
i j
T
ϑ
σ, j
) =
2
∑
i, j=1
∂
u
i (Wg
i j
∂
τ
T
ϑ
σ, j
)
=
2
∑
i, j=1
∂
u
i (Wg
i j
∂
u
j T
ϑ
σ,τ
) = W∆
ds
2T
ϑ
σ,τ
7.10 Application to the mean curvature ﬂow in R
4
125
with the invariant LaplaceBeltrami operator
∆
ds
2 ψ =
2
∑
i, j=1
∂
u
i (Wg
i j
∂
u
j ψ).
This leads us to the following characterization.
Proposition 7.6. Let n =2 and H ≡0. Then the mean curvature ﬂow does not affect
the Coulomb frame property if and only if
∆
ds
2 T
ϑ
σ,i
= 0 in B
for all i = 1, 2 and σ, ϑ = 1, 2, and for all times τ.
Parallel mean curvature vector
From chapter 2 we recall the deﬁnition of a mean curvature vector H parallel in the
normal bundle, i.e.
H
⊥
u
≡0, H
⊥
v
≡0,
or equivalently
H
1,u
−T
2
1,1
H
2
= 0, H
1,v
−T
2
1,2
H
2
= 0,
H
2,u
+T
2
1,1
H
1
= 0, H
2,v
+T
2
1,2
H
2
= 0
in case n = 2.
Proposition 7.7. If the mean curvature H is parallel in the normal bundle for all
times then the curvature S = S
2
1,12
is constant in time, i.e. it holds
∂
τ
S ≡0.
Proof. From Proposition 7.5 we infer the representations
∂
τ
T
2
1,1
=
2
∑
j,k=1
_
(L
2,1 j
H
1,u
k −L
1,1 j
T
2
1,k
H
1
) −(L
1,1 j
H
2,u
k +L
2,1 j
T
2
1,k
H
2
)
_
g
jk
+∂
u
T
2
1,τ
,
∂
τ
T
2
1,2
=
2
∑
j,k=1
_
(L
2,2 j
H
1,u
k −L
1,2 j
T
2
1,k
H
1
) −(L
1,2 j
H
2,u
k +L
2,2k
T
2
1, j
H
2
)
_
+∂
v
T
2
1,τ
.
126 7 Normal Coulomb frames in R
4
Now we evaluate the parallelity condition to obtain
2
∑
j,k=1
_
(L
2,1 j
H
1,u
k −L
1,1 j
T
2
1,k
H
1
) −(L
1,1 j
H
2,u
k +L
2,1 j
T
2
1,k
H
2
)
_
g
jk
=
2
∑
j,k=1
_
g
jk
L
2,1 j
T
2
1,k
H
2
−g
jk
L
1,1 j
T
2
1,k
H
1
+g
jk
L
1,1 j
T
2
1,k
H
1
−g
jk
L
2,1 j
T
2
1,k
H
2
_
= 0
as well as
2
∑
j,k=1
_
(L
2,2 j
H
1,u
k −L
1,2 j
T
2
1,k
H
1
) −(L
1,2 j
H
2,u
k +L
2,2k
T
2
1, j
H
2
)
_
g
jk
=
2
∑
j,k=1
_
g
jk
L
2,2 j
T
2
1,k
H
2
−g
jk
L
1,2 j
T
2
1,k
H
1
+g
jk
L
1,2 j
T
2
1,k
H
1
−g
jk
L
2,2k
T
2
1, j
H
2
_
= 0.
This implies
∂
τ
T
2
1,1
=∂
u
T
2
1,τ
, ∂
τ
T
2
1,2
=∂
v
T
2
1,τ
,
and therefore it follows that
∂
τ
S =∂
τ
(∂
v
T
2
1,1
−∂
u
T
2
1,2
) =∂
v
∂
u
T
2
1,τ
−∂
u
∂
v
T
2
1,τ
= 0
proving the statement. ⊓⊔
It remains open to identitfy geometric constellations where the mean curvature
vector is actually parallel in the normal bundle for all times.
For further discussions on the mean curvature ﬂow for special surface classes we
want to refer the reader to Terng [156], Liu and Terng [116], or Smoczyk, Wang and
Xin [148]. But a satisying theory seems to fail up to the present.
Chapter 8
Normal Coulomb frames in R
n+2
8.1 Problem formulation
8.2 The EulerLagrange equations
8.3 Examples
8.4 Quadratic growth in the gradient
8.5 Torsion free normal frames
8.6 Nonﬂat normal bundles
8.7 Bounds for the total torsion
8.8 Existence and regularity of weak normal Coulomb frames
8.9 Classical regularity of normal Coulomb frames
128 8 Normal Coulomb frames in R
n+2
In this chapter we consider surfaces in R
n+2
and their orthonormal normal frames critical for the
functional of total torsion. We extend the results of the previous chapter, but in this general case
now the analysis becomes more intricate for the fact that normal Coulomb frames are solutions of
nonlinear and inhomogeneous elliptic systems of partial differential equations.
8.1 Problem formulation
We want to generalize our results of the previous chapter to the case of arbitrary
codimension n ≥ 2. We start with computing the EulerLagrange equations of the
following parameter invariant functional of total torsion
T
X
[N] =
1
2
2
∑
i, j=1
n
∑
σ,ϑ=1
__
B
g
i j
T
ϑ
σ,i
T
ϑ
σ, j
W dudv
for a normal frame N = (N
1
, . . . , N
n
). It will turn out that these equations form a
nonlinear system of elliptic partial differential equations with quadratic growth in
the gradient.
We derive analytical and geometric properties of critical points, and we will es
tablish existence and regularity results for normal frames in case of vanishing cur
vature of the normal bundle as well as for normal Coulomb frames in the general
situation of nonﬂat normal bundles.
8.2 The EulerLagrange equations
Normal Coulomb frames
Due to do Carmo [22], chapter 3, section 2 we can construct a family R(w, ε) of
rotations from the Lie group SO(n) for an arbitrary skewsymmetric matrix
A(w) = (a
σϑ
(w))
σ,ϑ=1,...,n
∈C
∞
(B, so(n))
by means of the geodesic ﬂow in SO(n), with so(n) the associated Lie algebra.
130 8 Normal Coulomb frames in R
n+2
In terms of such a oneparameter family of rotations
R(w, ε) =
_
R
ϑ
σ
(w, ε)
_
σ,ϑ=1,...,n
∈C
∞
(B(−ε
0
, +ε
0
), SO(n)),
with sufﬁciently small ε > 0 such that
R(w, 0) =E
n
,
∂
∂ε
R(w, 0) = A(w) ∈C
∞
(B, so(n))
holds true with the ndimensional unit matrix E
n
, we consider variations
¯
N = (
¯
N
1
, . . . ,
¯
N
n
)
of a given orthonormal normal frame N = (N
1
, . . . , N
n
) by means of
¯
N
σ
(w, ε) :=
n
∑
ϑ=1
r
σϑ
(w, ε)N
ϑ
(w), σ = 1, . . . , n.
Such a matrix A(w) is the essential ingredient for the following deﬁnition of the
ﬁrst variation of total functional.
Deﬁnition 8.1. An orthonormal normal frame N is called critical for the functional
of total torsion or a normal Coulomb frame if and only if the ﬁrst variation
δT
X
[N; A] := lim
ε→0
1
ε
_
T
X
[
¯
N] −T
X
[N]
_
vanishes w.r.t. all skewsymmetric perturbations A(w) ∈C
∞
(B, so(n)).
Computation of the ﬁrst variation
Now we come to the computation of the ﬁrst variation of the functional T
X
[N] and
determine the according EulerLagrange equations.
Proposition 8.1. The ONF N is a normal Coulomb frame if and only if its torsion
coefﬁcients solve the following system of Neumann boundary value problems
div(T
ϑ
σ,1
, T
ϑ
σ,2
) = 0 in B, (T
ϑ
σ,1
, T
ϑ
σ,2
) ν = 0 on ∂B
for all σ, ϑ = 1, . . . , n, and where ν denotes the outer unit normal vector along the
boundary curve ∂B.
Compare it with the single EulerLagrange equation in case n = 2 of two codi
mensions
div(T
2
1,1
, T
2
1,2
) = 0 in B, (T
2
1,1
, T
2
1,2
) ν = 0 on ∂B.
8.2 The EulerLagrange equations 131
Proof. We consider the oneparameter family of rotations
R(w, ε) = (r
σϑ
(w, ε))
σ,ϑ=1,...,n
as in the introduction. Expanding around ε = 0 yields
R(w, ε) =E
n
+εA(w) +o(ε).
Now we apply the rotation R(w, ε) to the given ONF N. The resulting unit normal
vectors
¯
N
1
, . . . ,
¯
N
n
are then determined by
¯
N
σ
=
n
∑
ϑ=1
r
σϑ
N
ϑ
=
n
∑
ϑ=1
_
δ
σϑ
+εa
σϑ
+o(ε)
_
N
ϑ
= N
σ
+ε
n
∑
ϑ=1
a
σϑ
N
ϑ
+o(ε),
and for their derivatives we compute
¯
N
σ,u
ℓ = N
σ,u
ℓ +ε
n
∑
ϑ=1
_
a
σϑ,u
ℓ N
ϑ
+a
σϑ
N
ϑ,u
ℓ
_
+o(ε).
Consequently, the new torsion coefﬁcients can be expanded to
¯
T
ω
σ,ℓ
=
¯
N
σ,u
ℓ
¯
N
ω
= N
σ,u
ℓ N
ω
+ε
n
∑
ϑ=1
_
a
σϑ,u
ℓ N
ϑ
+a
σϑ
N
ϑ,u
ℓ
_
N
ω
+ε N
σ,u
ℓ
n
∑
ϑ=1
a
ωϑ
N
ϑ
+o(ε)
= T
ω
σ,ℓ
+εa
σω,u
ℓ +ε
n
∑
ϑ=1
_
a
σϑ
T
ω
ϑ,ℓ
+a
ωϑ
T
ϑ
σ,ℓ
_
+o(ε)
such that for their squares we infer
(
¯
T
ω
σ,ℓ
)
2
= (T
ω
σ,ℓ
)
2
+2ε
_
a
σω,u
ℓ T
ω
σ,ℓ
+
n
∑
ϑ=1
_
a
σϑ
T
ω
ϑ,ℓ
T
ω
σ,ℓ
+a
ωϑ
T
ϑ
σ,ℓ
T
ω
σ,ℓ
_
_
+o(ε).
Before we insert this result into the functional T
X
[N] of total torsion we observe
n
∑
σ,ω,ϑ=1
_
a
σϑ
T
ω
ϑ,ℓ
T
ω
σ,ℓ
+a
ωϑ
T
ϑ
σ,ℓ
T
ω
σ,ℓ
_
=
n
∑
σ,ω,ϑ=1
_
a
σϑ
T
ω
ϑ,ℓ
T
ω
σ,ℓ
+a
σϑ
T
ϑ
ω,ℓ
T
σ
ω,ℓ
_
= 2
n
∑
σ,ω,ϑ=1
a
σϑ
T
ω
ϑ,ℓ
T
ω
σ,ℓ
= 0
taking the skewsymmetry of A(w) into account.
132 8 Normal Coulomb frames in R
n+2
Thus the difference between T
X
[
¯
N] and T
X
[N] computes to (notice that it holds
a
ω
σ,u
ℓ
T
ω
σ,ℓ
= a
σ
ω,u
ℓ
T
σ
ω,ℓ
)
T
X
[
¯
N] −T
X
[N] = ε
2
∑
ℓ=1
n
∑
σ,ω=1
__
B
a
σω,u
ℓ T
ω
σ,ℓ
dudv +o(ε)
= 2ε
∑
1≤σ<ω≤n
__
B
_
a
σω,u
T
ω
σ,1
+a
σω,v
T
ω
σ,2
_
dudv +o(ε)
= 2ε
∑
1≤σ<ω≤n
_
∂B
a
σω
(T
ω
σ,1
, T
ω
σ,2
) ν ds
−2ε
n
∑
1≤σ<ω≤n
__
B
a
σω
div(T
ω
σ,1
, T
ω
σ,2
)dudv +o(ε).
But A(w) was chosen arbitrarily which proves the proposition. ⊓⊔
The integral functions
Interpreting the EulerLagrange equations as integrability conditions analogously to
the situation considered in section 7.6, Poincare’s lemma ensures the existence of
integral functions τ
(σϑ)
∈C
3
(B, R) satisfying
∇τ
(σϑ)
=
_
−T
ϑ
σ,2
, T
ϑ
σ,1
_
in B for all σ, ϑ = 1, . . . , n.
Furthermore, due to the Neumann boundary conditions (T
ϑ
σ,1
, T
ϑ
σ,2
) ν =0, which
imply ∇τ
(σϑ)
(−v, u) =0 on ∂B with the unit tangent vector (−v, u) ⊥ν at ∂B, we
may again choose τ
(σϑ)
so that
τ
(σϑ)
= 0 on ∂B for all σ, ϑ = 1, . . . , n.
Notice that the matrix (τ
(σϑ)
)
σ,ϑ=1,...,n
is skewsymmetric.
A nonlinear elliptic system for the integral functions τ
σϑ
Let us now deﬁne the skewsymmetric matrix
δτ
(σϑ)
:=
n
∑
ω=1
det
_
∇τ
(σω)
, ∇τ
(ωϑ)
_
, σ, ϑ = 1, . . . n.
Our aim now is to establish an elliptic system for τ
(σϑ)
with quadratic growth in
the gradient.
8.3 Examples 133
Proposition 8.2. Let the conformally parametrized immersion X : B → R
n+2
to
gether with normal Coulomb frame N be given. Then the integral functions τ
(σϑ)
are solutions of the boundary value problems
∆τ
(σϑ)
=−δτ
(σϑ)
+S
ϑ
σ,12
in B, τ
(σϑ)
= 0 on ∂B,
for σ, ϑ = 1, . . . , n, where δτ
(σϑ)
grows quadratically in the gradient ∇τ
(σϑ)
.
Proof. Choose any (σ, ϑ) ∈ ¦1, . . . , n¦ ¦1, . . . , n¦. The representation formula
S
ϑ
σ,12
=∂
v
T
ϑ
σ,1
−∂
u
T
ϑ
σ,2
+
n
∑
ω=1
_
T
ω
σ,1
T
ϑ
ω,2
−T
ω
σ,2
T
ϑ
ω,1
_
for the normal curvature tensor together with ∇τ
(σϑ)
= (−T
ϑ
σ,2
, T
ϑ
σ,1
) yields
∆τ
(σϑ)
= ∂
v
T
ϑ
σ,1
−∂
u
T
ϑ
σ,2
= −
n
∑
ω=1
_
T
ω
σ,1
T
ϑ
ω,2
−T
ω
σ,2
T
ϑ
ω,1
_
+S
ϑ
σ,12
=
n
∑
ω=1
_
τ
(σω)
v
τ
(ωϑ)
u
−τ
(σω)
u
τ
(ωϑ)
v
_
+S
ϑ
σ,12
proving the statement. ⊓⊔
8.3 Examples
We want to determine the special formof this nonlinear elliptic system in the special
cases n = 2 and n = 3.
The case n = 2
There is only one integral function τ
(12)
satisfying
∆τ
(12)
= S
2
1,12
in B, τ
(12)
= 0 on ∂B.
This is exactly the Poisson equation with homogeneous boundary data from sec
tion 7.6 with τ =τ
(12)
.
134 8 Normal Coulomb frames in R
n+2
The case n = 3
If n = 3 we have three relations
∆τ
(12)
= τ
(13)
v
τ
(32)
u
−τ
(13)
u
τ
(32)
v
+S
2
1,12
,
∆τ
(13)
= τ
(12)
v
τ
(23)
u
−τ
(12)
u
τ
(23)
v
+S
3
1,12
,
∆τ
(23)
= τ
(21)
v
τ
(13)
u
−τ
(21)
u
τ
(13)
v
+S
3
2,12
.
Recall the curvature vector
S =
1
W
(S
2
1,12
, S
3
1,12
, S
3
2,12
) ∈ R
3
of the normal bundle from section 3.8. Analogously we deﬁne
T := (τ
(12)
, τ
(13)
, τ
(23)
) ∈ R
3
and infer
∆T =T
u
T
v
+SW in B, T = 0 on ∂B (8.1)
with the usual vector product in R
3
.
In other words: If n = 3, then the vector T solves an inhomogeneous Hsurface
system with constant mean curvature H =
1
2
and vanishing boundary data.
Namely compare it with the mean curvature system
△X = 2HWN
from chapter 2 with the scalar mean curvature H, the area element W and the unit
normal vector N of the surface. If the surface X : B→R
5
would additionally satisﬁes
the conformality relations
X
u
X
u
=W = X
v
X
v
, X
u
X
v
= 0 in B,
then X actually represents an immersion with scalar mean curvature H.
We want to point out that S ≡ 0 is a consequence of T ≡ 0 by a result of
Wente [165] on systems of the form (8.1), i.e. normal Coulomb frames for surfaces
X : B →R
5
with ﬂat normal bundle are free of torsion.
We consider the general situation of higher codimensions as well as analytical
and geometric properties of normal Coulomb frames for surfaces with nonﬂat nor
mal bundles in the following sections.
8.4 Quadratic growth in the gradient 135
8.4 Quadratic growth in the gradient
The Grassmanntype vector T
The previous example gives rise to the deﬁnition of the following vector of Grass
mann type
T :=
_
τ
(σϑ)
_
1≤σ<ϑ≤n
∈ R
N
, N :=
n
2
(n −1).
In our examples this vector T works as follows:
T =τ
(12)
∈ R for n = 2,
T =
_
τ
(12)
, τ
(13)
, τ
(23)
_
∈ R
3
for n = 3.
Analogously we deﬁne
δT :=
_
δτ
(σϑ)
_
1≤σ<ϑ≤n
∈ R
N
.
Then Proposition 8.2 can be written in the following succint form
∆T =−δT +SW in B, T = 0 on ∂B.
From the deﬁnition of δG we next obtain
[∆T [ ≤c[∇T [
2
+[SW[ in B
with some real constant c > 0. The exact knowledge of this constant will become
important later.
A nonlinear system with quadratic growth for T
Proposition 8.3. Let the conformally parametrized immersion X : B → R
n+2
to
gether with a normal Coulomb frame N be given. Then it holds
[∆T [ ≤
√
n −2
2
[∇T [
2
+[SW[ in B.
Proof. We already know
[∆T [ ≤[δT [ +[SW[ in B.
136 8 Normal Coulomb frames in R
n+2
Thus it remains to estimate [δT [ appropriately. We begin with computing
[δT [
2
=
∑
1≤σ<ϑ≤n
_
n
∑
ω=1
det
_
∇τ
(σω)
, ∇τ
(ωϑ)
_
_
2
≤ (n −2)
∑
1≤σ<ϑ≤n
_
n
∑
ω=1
det
_
∇τ
(σω)
, ∇τ
(ωϑ)
_
2
_
.
Note that only derivatives of elements of T appear on the right hand side, say R, of
[δT [
2
. Moreover this right hand side can be estimated by [T
u
∧T
v
[
2
since T
u
∧T
v
has actually more elements than R.
1
Thus using Lagrange’s identity
[X ∧Y[
2
=[X[
2
[Y[
2
−(X Y)
2
≤[X[
2
[Y[
2
for two vectors X and Y we can estimate as follows
[δT [
2
≤ (n −2)[T
u
∧T
v
[
2
≤ (n −2)[T
u
[
2
[T
v
[
2
≤
_√
n −2
2
_
[T
u
[
2
+[T
v
[
2
_
_
2
.
This proves the statement. ⊓⊔
8.5 Torsion free normal frames
The case n = 3
As already mentioned above, in Wente [165] we ﬁnd an interesting uniqueness result
for solutions of the homogeneous system
∆T =T
u
T
v
in B
with vanishing boundary values, corresponding to the ﬂat normal bundle situation
S ≡0 in case n = 3. Wente’s results particularly states
Proposition 8.4. (Wente [165])
The only solution of the elliptic system
∆T =T
u
T
v
in B, T = 0 on ∂B,
is the trivial solution T ≡0.
1
In particular, elements of the form det (∇τ
(σω)
, ∇τ
(ω
′
ϑ)
)
2
appear in [T
u
∧T
v
[
2
, but they do not
appear in R.
8.5 Torsion free normal frames 137
Below we will give a new proof of Wente’s result, independent of the codimen
sion, which follows from asymptotic expansions of solutions T in the interior B
and on the boundary ∂B.
For the present we can conﬁrm
Corollary 8.1. Suppose that the immersion X : B →R
5
admits a normal Coulomb
frame. Then this frame is free of torsion if and only if the curvature vector S van
ishes identically.
An auxiliary function
For a complete treatment of the general case n > 3 we need some further prepara
tions. Let us start with the
Lemma 8.1. Let S ≡0. Then the function
Φ(w) =T
w
(w) T
w
(w) =
∑
1≤σ<ϑ≤n
τ
(σϑ)
w
τ
(σϑ)
w
,
using the complex notation φ
w
=
1
2
(φ
u
+iφ
v
), vanishes identically in B.
Proof. We will prove that Φ solves the boundary value problem
Φ
w
= 0 in B, Im(w
2
Φ) = 0 on ∂B.
Then the analytic function
Ψ(w) := w
2
Φ(w)
has vanishing imaginary part, and the CauchyRiemann equations imply Ψ(w) ≡ c
with some c ∈ R. The assertion follows fromΨ(0) = 0.
1. In order to deduce the stated boundary condition, recall that τ
(σϑ)
= 0 on ∂B.
Thus all tangential derivatives vanish identically because
−vτ
(σϑ)
u
+uτ
(σϑ)
v
=−2Im(wτ
(σϑ)
w
) = 0 on ∂B
for all σ, ϑ = 1, . . . , n. The boundary condition follows from
Im(w
2
Φ) = Im
_
w
2
T
w
T
w
_
= Im
_
w
2
∑
1≤σ<ϑ≤n
τ
(σϑ)
w
τ
(σϑ)
w
_
=
∑
1≤σ<ϑ≤n
Im
_
_
wτ
(σϑ)
w
__
wτ
(σϑ)
w
_
_
= 2
∑
1≤σ<ϑ≤n
Re
_
wτ
(σϑ)
w
_
Im
_
wτ
(σϑ)
w
_
= 0
on the boundary ∂B.
138 8 Normal Coulomb frames in R
n+2
2. Now we show the analyticity of Φ using
∆τ
(σϑ)
= 4τ
(σϑ)
ww
=−δτ
(σϑ)
.
Interchanging indices cyclically yields
2Φ
w
= 4T
w
T
ww
= 4
∑
1≤σ<ϑ≤n
τ
(σϑ)
w
τ
(σϑ)
ww
=
1
2
n
∑
σ,ϑ=1
τ
(σϑ)
w
∆τ
(σϑ)
=
1
4
n
∑
σ,ϑ,ω=1
_
τ
(σω)
v
τ
(ωϑ)
u
τ
(σϑ)
u
−τ
(σω)
u
τ
(ωϑ)
v
τ
(σϑ)
u
_
−
i
4
n
∑
σ,ϑ,ω=1
_
τ
(σω)
v
τ
(ωϑ)
u
τ
(σϑ)
v
−τ
(σω)
u
τ
(ωϑ)
v
τ
(σϑ)
v
_
=
1
4
n
∑
σ,ϑ,ω=1
_
τ
(ωϑ)
v
τ
(ϑσ)
u
τ
(ωσ)
u
−τ
(σω)
u
τ
(ωϑ)
v
τ
(σϑ)
u
_
−
i
4
n
∑
σ,ϑ,ω=1
_
τ
(ϑσ)
v
τ
(σω)
u
τ
(ϑω)
v
−τ
(σω)
u
τ
(ωϑ)
v
τ
(σϑ)
v
_
which shows Φ
w
= 0. The proof is complete. ⊓⊔
The case n ≥3
The main result of this section is the
Theorem 8.1. Suppose that the immersion X : B →R
n+2
admits a normal Coulomb
frame N. Then this frame is free of torsion if and only if the curvature vector S of its
normal bundle vanishes identically.
Proof. We introduce conformal parameters (u, v) ∈ B. Furthermore, let N be a nor
mal Coulomb frame. If N is free of torsion then the curvature vector S vanishes
identically. So assume conversely S ≡ 0 and let us show that N is free of torsion.
Consider for this aim the Grassmanntype vector T from above. Since
T
w
T
w
≡0 in B
holds true by the previous lemma, we ﬁnd
[T
u
[ =[T
v
[, T
u
T
v
= 0 in B.
Thus T is a conformally parametrized solution of
∆T =−δT in B, T = 0 on ∂B.
8.6 Nonﬂat normal bundles 139
According to the quadratic growth condition [δT [ ≤ c[∇T [
2
(see paragraph 8.4),
the arguments in Heinz [81] apply
2
: Assume T ,≡ const in B. Then the asymptotic
expansion stated in Satz of Heinz [81] implies that boundary branch points w
0
∈ ∂B
with the property
T
u
(w
0
) =T
v
(w
0
) = 0
are isolated. But this contradicts our boundary condition T [
∂B
= 0! We thus infer
T ≡const = 0
which implies τ
(σϑ)
≡ 0 for all σ, ϑ = 1, . . . , n, and the normal Coulomb frame is
free of torsion. The theorem is proved. ⊓⊔
8.6 Nonﬂat normal bundles
We establish upper bounds for the torsion coefﬁcients of normal Coulomb frames.
An upper bound for T via Wente’s L
∞
estimate
Our ﬁrst result is
Proposition 8.5. Let N be a normal Coulomb frame for the conformally parametrized
immersion X : B →R
n+2
. Then the Grassmanntype vector T satisﬁes
T 
L
∞
(B)
= sup
w∈B
_
∑
1≤σ<ϑ≤n
[τ
(σϑ)
(w)[
2
≤
n −2
2π
∇T 
2
L
2
(B)
+
n(n −1)
8
SW
L
∞
(B)
with the Lebesgue norms L
2
(B) and L
∞
(B).
Proof. 1. For 1 ≤ σ < ϑ ≤ n, ω ∈ ¦1, . . . , n¦ with ω ,∈ ¦σ, ϑ¦, and given integral
functions τ
(σϑ)
we deﬁne the functions y
(σϑω)
as the unique solutions of
∆y
(σϑω)
=−det
_
∇τ
(σω)
, ∇τ
(ωϑ)
_
in B, y
(σϑω)
= 0 on ∂B.
2
Let X : ∈C
2
(B, R
n
) solve the system △X = H f (X, X
u
, X
v
) together with X
2
u
= X
2
v
, X
u
X
v
= 0,
where [ f (x, p, q)[ ≤ µ([x[)([p[
2
+[q[
2
). Then if X
u
(w
0
) = 0 for some w
0
∈ ∂B, but X
u
,
≡
0, the
asymptotic expansion X
w
(w
0
) = a(w−w
0
)
ℓ
−o([w−w
0
[
ℓ
) holds for w →w
0
, where a ∈ C with
a
2
1
+. . . a
2
n
= 0, and ℓ ∈ N¸¦0¦.
140 8 Normal Coulomb frames in R
n+2
Wente’s L
∞
estimate fom Wente [166] together with Topping [159] then yields
the optimal inequalities
3
y
(σϑω)

L
∞
(B)
≤
1
4π
_
∇τ
(σω)

2
L
2
(B)
+∇τ
(ωϑ)

2
L
2
(B)
_
for 1 ≤ σ < ϑ ≤ n, ω ,∈ ¦σ, ϑ¦. In addition we introduce the Grassmanntype
vector Z = (z
(σϑ)
)
1≤σ<ϑ≤n
as the unique solution of
∆Z =SW in B, Z = 0 on ∂B.
Introduce new indices to write
Z = (Z
1
, . . . , Z
N
), SW = (S
1
, . . . , S
N
)
for the moment. We use Poisson’s representation formula to estimate as follows
[Z(w)[ ≤
N
∑
i=1
[Z
i
[ =
N
∑
i=1
¸
¸
¸
¸
¸
¸
__
B
φ(ζ; w)S
i
(ζ)dξdη
¸
¸
¸
¸
¸
¸
≤
N
∑
i=1
__
B
[φ(ζ; w)[[S
i
(ζ)[ dξdη
≤
√
N
__
B
[φ(ζ; w)[
¸
N
∑
i=1
[S
i
(ζ)[
2
dξdη
=
√
N
__
B
[φ(ζ; w)[[S(ζ)W(ζ)[ dξdη
with the Green function φ(ζ; w) for the Laplacian ∆ in B; ζ = (ξ, η). From
section 7.6 we already know
__
B
[φ(ζ; w)[ dξdη =
1 −[w[
2
4
≤
1
4
which enables us to continue with estimating [Z(w)[ to get
Z
L
∞
(B)
≤
√
NSW
L
∞
(B)
__
B
[φ(ζ; w)[ dξdη ≤
√
N
4
SW
L
∞
(B)
.
3
In 1980 H. Wente proved: Let Φ ∈ C
0
(B) ∩H
1,2
(B) be a solution of ∆Φ = −( f
u
g
v
− f
v
g
u
)
in B, Φ = 0 on ∂Ω, with f , g ∈ H
1,2
(B), then Φ
L
∞
(B)
+∇Φ
L
2
(B)
≤C∇f 
L
2
(B)
∇g
L
2
(B)
.
Following Topping [159] we may set
C
2
=
1
4π
after applying H¨ older’s inequality.
8.6 Nonﬂat normal bundles 141
2. Now recall that
∆τ
(σϑ)
=−
n
∑
ω=1
det (∇τ
(σω)
, ∇τ
(ωϑ)
) +S
ϑ
σ,12
W =
n
∑
ω=1
∆y
(σϑω)
+∆z
(σϑ)
.
Taking account of the unique solvability of the above introduced Dirichlet prob
lems with vanishing boundary data, the maximum principle yields
τ
(σϑ)
=
∑
ω,∈¦σ,ϑ¦
y
(σϑω)
+z
(σϑ)
, 1 ≤σ <ϑ ≤n.
Now we collect all the estimates obtained and get (rearrange the summations and
redeﬁne some indices)
T 
L
∞
(B)
≤
∑
1≤σ<ϑ≤n
∑
ω,∈¦σ,ϑ¦
y
(σϑω)

L
∞
(B)
+
∑
1≤σ<ϑ≤n
z
(σϑ)

L
∞
(B)
≤
∑
1≤σ<ϑ≤n
∑
ω,∈¦σ,ϑ¦
y
(σϑω)

L
∞
(B)
+
√
NZ
L
∞
(B)
≤
1
4π
∑
1≤σ<ϑ≤n
∑
ω,∈¦σ,ϑ¦
_
∇τ
(σω)

2
L
2
(B)
+∇τ
(ωϑ)

2
L
2
(B)
_
+
N
4
SW
L
∞
(B)
=
1
4π
_
∑
1≤ω<σ<ϑ≤n
_
∇τ
(ωσ)

2
L
2
(B)
+∇τ
(ωϑ)

2
L
2
(B)
_
+
∑
1≤σ<ω<ϑ≤n
_
∇τ
(σω)

2
L
2
(B)
+∇τ
(ωϑ)

2
L
2
(B)
_
+
∑
1≤σ<ϑ<ω≤n
_
∇τ
(σω)

2
L
2
(B)
+∇τ
(ϑω)

2
L
2
(B)
_
_
+
N
4
SW
L
∞
(B)
= . . .
142 8 Normal Coulomb frames in R
n+2
. . . =
1
4π
_
∑
1≤σ<ϑ<ω≤n
∇τ
(σϑ)

2
L
2
(B)
+
∑
1≤σ<ω<ϑ≤n
∇τ
(σϑ)

2
L
2
(B)
+
∑
1≤σ<ϑ<ω≤n
∇τ
(σϑ)

2
L
2
(B)
+
∑
1≤ω<σ<ϑ≤n
∇τ
(σϑ)

2
L
2
(B)
+
∑
1≤σ<ω<ϑ≤n
∇τ
(σϑ)

2
L
2
(B)
+
∑
1≤ω<σ<ϑ≤n
∇τ
(σϑ)

2
L
2
(B)
_
+
N
4
SW
L
∞
(B)
=
1
2π
∑
1≤σ<ϑ≤n
∑
ω,∈¦σ,ϑ¦
∇τ
(σϑ)

2
L
2
(B)
+
N
4
SW
L
∞
(B)
=
n −2
2π
∇T 
2
L
2
(B)
+
1
4
n(n −1)
2
SW
L
∞
(B)
.
This proves the statement. ⊓⊔
An upper bound for T via Poincar´ e’s inequality
We want to present an alternative way to establish an upper bound for the Grassmann
type vector T . In particular, we show
[z
(σϑ)
(w)[ ≤
_
2
π
S
ϑ
σ,12
W
L
2
(B)
in B for all 1 ≤σ <ϑ ≤n
for the vector Z = (z
(12)
, z
(13)
, . . .) from the previous paragraph. This would lead
us to
Z
L
∞
(B)
= sup
B
_
∑
1≤σ<ϑ≤n
[z
(σϑ)
(w)[
2
≤
_
2
π
_
∑
1≤σ<ϑ≤n
S
ϑ
σ,12
W
2
L
2
(B)
=
_
2
π
SW
L
2
(B)
≤
√
2SW
L
∞
(B)
which ﬁnally gives a smaller upper bound for T 
L
∞
(B)
at least for codimensions
n = 2, 3.
In order to prove the stated inequality we start with the Poisson representation
formula
z
(σϑ)
=
__
B
φ(ζ; w)S
ϑ
σ,12
(ζ)W(ζ)dξdη, z
(σϑ)
= 0 on ∂B.
8.6 Nonﬂat normal bundles 143
Applying the H¨ older and the Poincar´ e inequality gives
[z
(σϑ)
(w)[ ≤ φ( ; w)
L
2
(B)
S
ϑ
σ,12
W
L
2
(B)
≤
1
2
√
π
∇
ζ
φ( ; w)
L
1
(B)
S
ϑ
σ,12
W
L
2
(B)
.
(8.2)
For the optimal constant
1
2
√
π
in the Sobolev inequality we refer to Gilbarg and
Trudinger [71], paragraph 7.7 and the references therein. Furthermore
φ =φ(ζ; w) :=
1
2π
log[
ζ −w
1 −wζ
[
denotes again Green’s function for the Laplace operator ∆ in B, and it satisﬁes
φ( ; w) ∈ H
1
0
(B) as well as
2φ
ζ
(ζ; w) ≡φ
ξ
(ζ; w) −iφ
η
(ζ; w) =
1
2π
_
ζ −w
[ζ −w[
2
+w
1 −wζ
[1 −wζ[
2
_
, w ,=ζ.
A straightforward calculation shows
[∇
ζ
φ(ζ; w)[ ≡2[φ
ζ
(ζ; w)[ =
1
2π
1 −[w[
2
[ζ −w[ [1 −wζ[
≤
1
2π
1 +[w[
[ζ −w[
≤
1
π
1
[ζ −w[
.
for ζ ,= w. And since the right hand side in the inequality
__
B
[∇
ζ
φ(ζ; w)[ dξdη ≤
1
π
__
B
δ
(w)
1
[ζ −w[
dξ dη +
1
π
__
B¸B
δ
(w)
1
[ζ −w[
dξ dη
≤ 2δ +
1
δ
becomes minimal for δ =
1
√
2
, we arrive at
[z
(σϑ)
(w)[ ≤
1
2
√
π
_
2
1
√
2
+
√
2
_
S
ϑ
σ,12
W
L
2
(B)
=
√
2
√
π
S
ϑ
σ,12
W
L
2
(B)
proving the stated inequality. Rearranging gives
[Z(w)[ ≤
√
2
√
π
√
π S
ϑ
σ,12
W
L
∞
(B)
≤
√
2SW
L
∞
(B)
.
Thus we have the following proposition.
144 8 Normal Coulomb frames in R
n+2
Proposition 8.6. Let N be a normal Coulomb frame for the conformally parametrized
immersion X : B →R
n+2
. Then the Grassmanntype vector T satisﬁes
T 
L
∞
(B)
≤
n −2
2π
∇T 
2
L
2
(B)
+
√
2SW
L
∞
(B)
.
This provides us a better estimate at least for small n.
An estimate for the torsion coefﬁcients
We are now in the position to prove our main result of this section.
Theorem 8.2. Let N be a normal Coulomb frame for the conformally parametrized
immersion X : B →R
n+2
with total torsion T
X
[N] and given SW
L
∞
(B)
. Assume
that the smallness condition
√
n −2
2
_
n −2
4π
T
X
[N] +C(n)SW
L
∞
(B)
_
< 1
is satisﬁed with C(n) :=min
_
n(n−1)
8
,
√
2
_
. Then the torsion coefﬁcients of N can be
estimated by
T
ϑ
σ,i

L
∞
(B)
≤c, i = 1, 2, 1 ≤σ <ϑ ≤n,
with a nonnegative constant c = c(n, SW
L
∞
(B)
, T
X
[N]) < +∞.
Proof. We have the following elliptic system
[∆T [ ≤
_
n(n −1)
2
[∇T [
2
+[SW[ in B, T = 0 on ∂B,
T 
L
∞
(B)
≤
n −2
2π
∇T 
2
L
2
(B)
+C(n)SW
L
∞
(B)
≤M ∈ [0, +∞).
The smallness condition ensures that we can apply Heinz’s global gradient estimate
Theorem1 in Sauvigny [143], chapter XII, § 3, obtaining ∇T 
∞
≤c.
4
This in turn
yields the desired estimate. ⊓⊔
It remains open to prove global pointwise estimates for the torsion coefﬁcients
without the assumed smallness condition. In particular, we would like to get rid of
the a priori knowledge of T
X
[N].
4
This theorem states: Let X ∈C
2
(B, R
n
) be a solution of the elliptic system [∆X[ ≤a[∇X[
2
+b in
B with X = 0 on ∂B and X
L
∞
(B)
≤M. Assume aM < 1. Then there is a constant c = c(a, b, M, α)
such that X
C
1+α
(B)
≤c(a, b, M, α).
8.7 Bounds for the total torsion 145
8.7 Bounds for the total torsion
Upper bounds
From the torsion estimates above we can immediately infer various upper bounds
for the functional of total torsion.
For example, let us focus on small solutions T : Namely, multiply
∆T =−δT +SW
by T and integrate by parts yields to get
Proposition 8.7. For small solutions T 
L
∞
(B)
≤
2
√
n−2
it holds
T
X
[N] = 2∇T 
2
L
2
(B)
≤
4T 
L
∞
(B)
SW
L
1
(B)
2 −
√
n −2T 
L
∞
(B)
.
The reader is refered to Sauvigny [143] for the construction of such small so
lutions of nonlinear elliptic systems. Let us emphasize that the case n = 2 is much
easier to handle: The classical maximum principle controls T 
L
∞
(B)
in terms of
SW
L
∞
(B)
, and no smallness condition is needed to bound the functional of total
torsion.
A lower bound
Finally we complete this section with establishing a lower bound for the functional
of total torsion.
Proposition 8.8. Let N a normal Coulomb frame for the conformally parametrized
immersion X : B →R
n+2
. Assume that the curvature vector S of its normal bundle
satisﬁes S ,≡0 as well as
∇(SW)
L
2
(B)
> 0.
Then it holds
T
X
[N] ≥
1
2
_
√
n −2SW
L
∞
(B)
+
SW
2
L
2
(B)
(1 −ρ)
2
SW
2
L
2
(B
ρ
)
+. . .
. . . +
2∇(SW)
2
L
2
(B)
SW
2
L
2
(B
ρ
)
_
−1
SW
2
L
2
(B
ρ
)
> 0
with the radius ρ = ρ(S) ∈ (0, 1) constructed in the proof given below, and the
setting B
ρ
=¦(u, v) ∈ R
2
u
2
+v
2
≤ρ¦.
146 8 Normal Coulomb frames in R
n+2
Proof. 1. Because of S ,= 0 there exists (a ﬁrst) ρ =ρ(S) ∈ (0, 1) such that
SW
L
2
(B
ρ
)
=
_
_
_
__
B
ρ
(0)
[SW[
2
dudv
_
_
_
1
2
> 0,
and this is already our radius ρ from the theorem. Now we choose a test function
η ∈C
0
(B, R) ∩H
1,2
0
(B, R) such that
η ∈ [0, 1] in B, η = 1 in B
ρ
, [∇η[ ≤
1
1 −ρ
in B.
Multiplying ∆T =−δT +SW by (ηSW) and integrating by parts yields
__
B
∇T ∇(ηSW)dudv =
__
B
η δT SW dudv −
__
B
η [SW[
2
dudv.
Taking [δT [ ≤
√
n −2[T
u
[[T
v
[ (see section 8.4) into account, we can now esti
mate as follows
__
B
ρ
[SW[
2
dudv ≤
__
B
η [SW[
2
dudv
≤
__
B
η
¸
¸
δT SW
¸
¸
dudv +
__
B
¸
¸
∇T ∇(ηSW)
¸
¸
dudv
≤ SW
L
∞
(B)
__
B
η [δT [ dudv +
__
B
[∇η[ [SW[ [∇T [ dudv
+
__
B
η [∇(SW)[ [∇T [ dudv
≤
√
n −2
2
SW
L
∞
(B)
__
B
[∇T [
2
dudv +
ε
2
__
B
[SW[
2
dudv
+
1
2ε(1 −ρ)
2
__
B
[∇T [
2
dudv
+
δ
2
__
B
[∇(SW)[
2
dudv +
1
2δ
__
B
[∇T [
2
dudv
with arbitrary real numbers ε, δ > 0.
8.8 Existence and regularity of weak normal Coulomb frames 147
Summarizing we arrive at
SW
2
L
2
(B
ρ
)
≤
_√
n −2
2
SW
L
∞
(B)
+
1
2ε(1 −ρ)
2
+
1
2δ
_
∇T 
2
L
2
(B)
+
ε
2
SW
2
L
2
(B)
+
δ
2
∇(SW)
2
L
2
(B)
.
2. Now we choose ε : Inserting
ε =SW
−2
L
2
(B)
SW
2
L
2
(B
ρ
)
> 0
and rearranging for S
2
L
2
(B
ρ
)
gives
SW
2
L
2
(B
ρ
)
≤
_
√
n −2SW
L
∞
(B)
+
SW
2
L
2
(B)
(1 −ρ)
2
SW
2
L
2
(B
ρ
)
_
∇T 
2
L
2
(B)
+
1
δ
∇T [
2
L
2
(B)
+δ∇(SW)
2
L
2
(B)
.
And since ∇(SW)
L
2
(B)
> 0 we can insert
δ =
1
2
∇(SW)
−2
L
2
(B)
SW
2
L
2
(B
ρ
)
which implies
SW
2
L
2
(B
ρ
)
≤ 2
_
√
n −2SW
L
∞
(B)
+
SW
2
L
2
(B)
(1 −ρ)
2
SW
2
L
2
(B
ρ
)
_
∇T 
2
L
2
(B)
+2
2∇(SW)
2
L
2
(B)
SW
2
L
2
(B
ρ
)
∇T 
2
L
2
(B)
.
Having T
X
[N] =∇T 
2
L
2
(B)
in mind we arrive at the stated estimate. ⊓⊔
8.8 Existence and regularity of weak normal Coulomb frames
Regularity results for the homogeneous Poisson problem
To introduce the function spaces coming next into play we consider the Dirichlet
boundary value problem
∆φ(u, v) = r(u, v) in B, φ(u, v) = 0 on ∂B. (DP)
148 8 Normal Coulomb frames in R
n+2
We want to recall very brieﬂy some important techniques to establish existence
and regularity of solutions φ to (DP).
Schauder estimates
Let r ∈ C
α
(B, R) hold true for the right hand side r. Then there exist a classical
solution of (DP) satisfying
φ
C
2+α
(B)
≤C(α)r
C
α
(B)
,
see e.g Gilbarg and Trudinger [71].
L
p
estimates
Now let r ∈ L
2
(
˚
B, R). Then any solution φ ∈ H
1,2
(
˚
B, R) belongs to the class
H
2,2
(
˚
B, R), and it holds
φ
H
2,2
(
˚
B)
≤C
_
φ
H
1,2
(
˚
B)
+r
L
2
(
˚
B)
_
.
In particular, if r ∈ H
m−2,2
(
˚
B, R) holds true for the right hand side we have
φ
H
m,2
(
˚
B)
≤C
_
φ
H
1,2
(
˚
B)
+r
H
m−2,2
(
˚
B)
_
.
For a detailed analysis we refer the reader to Dobrowolski [46], chapter 7. Note
that we must require r ∈ L
2
(
˚
B, R) to infer higher regularity φ ∈ C
0
(
˚
B, R) because
H
2,2
(
˚
B, R) is continuously emedded in C
0
(
˚
B, R) by Sobolev’s embedding theorem.
If on the other hand r ∈ L
1
(
˚
B, R) then any weak solution φ ∈ H
1,2
(
˚
B, R) of (DP)
satisﬁes
φ
L
q
(
˚
B)
≤Cr
L
1
(
˚
B)
for all 1 ≤q <∞,
φ
H
1, p
(
˚
B)
≤Cr
L
1
(
˚
B)
for all 1 ≤ p < 2.
A function φ ∈ H
1,2
(
˚
B, R) is not necessarily continuous.
Wente’s L
∞
estimate
This situation changes dramatically if the right hand side r possesses certain alge
braic structures. In particular, assume that
r =
∂a
∂u
∂b
∂v
−
∂a
∂v
∂b
∂u
with functions a, b ∈ H
1,2
(
˚
B, R).
8.8 Existence and regularity of weak normal Coulomb frames 149
Then again r ∈L
1
(
˚
B, R), but any solution φ ∈H
1,2
(
˚
B, R) is of class C
0
(B, R) and
satisﬁes Wente’s L
∞
estimate
φ
L
∞
(
˚
B)
+∇φ
L
2
(
˚
B)
≤
1
4π
∇a
L
2
(
˚
B)
∇b
L
2
(
˚
B)
,
see Wente [166]. We already used this inequality in section 8.6 for establishing an
upper bound for the functional of total torsion of normal Coulomb frames.
Hardy spaces
Wente’s discovery is the starting point of the modern harmonic analysis. Its general
framework is the concept of Hardy spaces H
1
(R
m
, R).
From Helein [82] we quote two common deﬁnitions of Hardy spaces which ﬁ
nally lead to equivalent formulations of the Hardy space theory.
Deﬁnition 8.2. (Tempereddistribution deﬁnition)
Let Ψ ∈C
∞
0
(R
m
, R) such that
_
R
m
Ψ(x)dx = 1.
For each ε > 0 we set
Ψ
ε
(x) =
1
ε
m
Ψ(ε
−1
x),
and for φ ∈ L
1
(R
m
, R) deﬁne
φ
∗
(x) = sup
ε>0
[(Ψ
ε
⋆ φ)(x)[.
Then φ belongs to H
1
(R
m
, R) if and only if φ
∗
∈ L
1
(R
m
, R) with norm
φ
H
1
(R
m
,R)
≤φ
L
1
(R
m
)
+φ
∗

L
1
(R
m
)
.
Deﬁnition 8.3. (RieszFouriertransformdeﬁnition)
For any function φ ∈ L
1
(R
m
, R) we denote by R
α
φ the function deﬁned by
F(R
α
φ) =
ξ
α
[ξ[
F(φ)(ξ)
with the φFourier transform
F(φ)(ξ) =
1
(2π)
m
2
_
R
m
e
−ixξ
φ(x)dx.
150 8 Normal Coulomb frames in R
n+2
Then φ belongs to H
1
(R
m
, R) if and only if
R
α
φ ∈ L
1
(R
m
, R) for all α = 1, . . . , m
with norm
φ
H
1
(R
m
)
=φ
L
1
(R
m
)
+
m
∑
α=1
R
α
φ
L
1
(R
m
)
.
For a comprehensive presentation of harmonic analysis we would also like to
refer the reader to Stein’s monograph [151].
Consider again our Dirichlet problem (DP). Let a, b ∈ H
1,2
(
˚
B, R), and consider
its extensions a →´ a and b →
´
b in H
1,2
(R
2
, R) to the whole space R
2
such that these
mappings are continuous from H
1,2
(
˚
B, R) to H
1,2
(R
2
, R). Then due to Helein [82]
it holds r ∈ H
1
(R
2
, R).
Lorentz interpolation spaces
This latter fact becomes especially important in the following. Let us start with
Deﬁnition 8.4. Let p ∈ (1, +∞) and q ∈ [1, +∞]. The Lorentz space L
(p,q)
(
˚
B, R) is
the set of measurable functions φ :
˚
B →R such that
 f 
L
(p,q)
(
˚
B)
:=
_
_
∞
_
0
_
t
1
p
φ
∗
(t)
_
q
dt
t
_
_
1
q
<∞ if q < +∞
or
 f 
L
(p,q)
(
˚
B)
:= sup
t>0
t
1
p
φ
∗
(t) <∞ if q = +∞.
Here φ
∗
denotes the unique nonincreasing rearrangement of [φ[ on [0, meas(
˚
B)].
Lorentz spaces are Banach spaces with a suitable norm. They may be considered
as a deformation of L
p
.
Notice that
L
(p,p)
(
˚
B, R) = L
p
(
˚
B, R),
L
(p,1)
(
˚
B, R) ⊂L
(p,q
′
)
(
˚
B, R) ⊂L
(p,q
′′
)
(
˚
B, R) ⊂L
(p,∞)
(
˚
B, R)
for 1 < q
′
< q
′′
. Then
(i) if φ ∈ H
1,2
(
˚
B, R) solves (DP) with r ∈ H
1
(
˚
B, R) then
∂φ
∂x
,
∂φ
∂y
∈ L
(2,1)
(
˚
B, R);
(i) if φ ∈ H
1,2
(
˚
B, R) with
∂φ
∂x
,
∂φ
∂y
∈ L
(2,1)
(
˚
B, R) then φ ∈C
0
(
˚
B, R).
8.8 Existence and regularity of weak normal Coulomb frames 151
The general regularity result
Summarizing the foregoing facts we can state the following regularity result from
Helein [82], chapter 3.
Proposition 8.9. Let a, b ∈ H
1,2
(
˚
B, R), and assume φ ∈ H
1,2
(
˚
B, R) solves (DP).
Then
∂φ
∂u
,
∂φ
∂v
∈ L
(2,1)
(
˚
B, R), and in particular φ ∈C
0
(B, R).
Existence of weak normal Coulomb frames
In case n =2 we constructed critical points N of the functional T
X
[N] of total torsion
by solving the EulerLagrange equation explicitely. In the general situation here we
construct critical points by means of direct methods of the calculus of variations.
We start with the following
Deﬁnition 8.5. Let m ∈ N, m ≥ 2. For two matrices A, B ∈ R
mm
we deﬁne their
inner product
¸A, B) = trace(A◦ B
T
) =
m
∑
σ,ϑ=1
a
σϑ
b
σϑ
and the associated norm
[A[ :=
_
¸A, A) =
_
m
∑
σ,ϑ=1
a
2
σϑ
_1
2
.
Helein proved in [82], Lemma 4.1.3, existence of weak Coulomb frames in the
tangent bundle of surfaces. We want to carry out his arguments and adapt his meth
ods to our present situation.
We consider conformally parametrized immersions
X = X(u, v) of regularity class C
k−1,α
(B, R
n+2
)
with some H¨ older exponent α ∈ (0, 1).
Proposition 8.10. Let the conformally parametrized immersion X ∈ C
k,α
(B, R
n+2
)
be given with some α ∈ (0, 1). Then there exists a weak normal Coulomb frame
N ∈ H
1,2
(
˚
B) ∩L
∞
(
˚
B) minimizing the functional T
X
[N] of total torsion in the set of
all weak normal frames of class H
1,2
(
˚
B) ∩L
∞
(
˚
B).
Proof. We ﬁx
5
some C
k−1,α
regular orthonormal normal frame
¯
N and interpret
T
X
[N] as a functional F
X
[R] of SO(n)regular rotations
R = (r
σϑ
)
σ,ϑ=1,...,n
∈ H
1,2
(
˚
B, SO(n))
5
Notice that now we start from
¯
N and transform into N.
152 8 Normal Coulomb frames in R
n+2
by setting
F
X
[R] :=
1
2
n
∑
σ,ϑ=1
2
∑
i=1
__
B
(T
ϑ
σ,i
)
2
dudv =
1
2
__
B
_
[T
1
[
2
+[T
2
[
2
_
dudv
where again
N
σ
:=
n
∑
ϑ=1
r
σϑ
¯
N
ϑ
as well as
T
i
= (T
ϑ
σ,i
)
σ,ϑ=1,...,n
∈ R
nn
.
Choose a minimizing sequence
ℓ
R = (
ℓ
r
σϑ
)
σ,ϑ=1,...,n
∈ H
1,2
(
˚
B, SO(n)) and deﬁne
ℓ
N
σ
:=
n
∑
ϑ=1
ℓ
r
σϑ
¯
N
ϑ
.
As in section 6.3 we ﬁnd
6
ℓ
T
i
=
ℓ
R
u
i ◦
ℓ
R
T
+
ℓ
R◦
¯
T
i
◦
ℓ
R
T
which implies
ℓ
T
i
◦
ℓ
T
T
i
= (
ℓ
R
u
i ◦
ℓ
R
T
+
ℓ
R◦
¯
T
i
◦
ℓ
R
T
) ◦ (
ℓ
R◦
ℓ
R
T
u
i
+
ℓ
R◦
¯
T
T
i
◦
ℓ
R
T
)
=
ℓ
R
u
i ◦
ℓ
R
T
u
i
+
ℓ
R◦
¯
T
i
◦
ℓ
R
T
u
i
+
ℓ
R
u
i ◦
¯
T
T
i
◦
ℓ
R
T
+
ℓ
R◦
¯
T
i
◦
¯
T
T
i
◦
ℓ
R
T
.
In particular, we conclude
trace(
ℓ
T
i
◦
ℓ
T
T
i
) = trace(
ℓ
R
u
i ◦
ℓ
R
T
u
i
) +2trace(
ℓ
R◦
¯
T
i
◦
ℓ
R
T
u
i
) +trace(
¯
T
i
◦
¯
T
T
i
)
or, using our notion of a matrix norm,
[
ℓ
T
i
[
2
=[
ℓ
R
u
i [
2
+2¸
ℓ
R◦
¯
T
i
,
ℓ
R
u
i ) +[
¯
T
i
[
2
. (8.3)
Furthermore, taking [
ℓ
R◦
¯
T
i
[ =[
¯
T
i
[ into account, we arrive at the estimate
[
ℓ
T
i
[
2
≥
_
[
¯
T
i
[ −[
ℓ
R
u
i [
_
2
a.e. on B, for all ℓ ∈ N.
Now because the
¯
T
i
are bounded in L
2
(
˚
B, R), and since
ℓ
R is minimizing for F
X
[R],
the sequences
ℓ
T
i
are also bounded in L
2
(
˚
B, R). Thus
ℓ
R
u
i are bounded sequences in
L
2
(
˚
B, SO(n)) in accordance with the last inequality.
6
Note that the proof of this identity remains true for R ∈ H
2,1
(
˚
B, SO(n)) ∩H
1,2
(
˚
B, SO(n)).
8.8 Existence and regularity of weak normal Coulomb frames 153
By Hilbert’s selection theorem and Rellich’s embedding theorem we ﬁnd a subse
quence, again denoted by
ℓ
R, which converges as follows:
ℓ
R
u
i ⇀ R
u
i weakly in L
2
(
˚
B, SO(n)),
ℓ
R →R strongly in L
2
(
˚
B, SO(n))
with some R ∈ H
1,2
(
˚
B, SO(n)). In particular, going if necessary to a subsequence,
we have
ℓ
R →R a.e. on
˚
B and
lim
ℓ→∞
__
B
[
ℓ
R◦
¯
T
i
−R◦
¯
T
i
[
2
dudv = 0
according to the dominated convergence theorem. Hence we can compute in the
limit
lim
ℓ→∞
__
B
¸
ℓ
R◦
¯
T
i
,
ℓ
R
u
i )dudv
= lim
ℓ→∞
_
__
B
¸
ℓ
R◦
¯
T
i
−R◦
¯
T
i
,
ℓ
R
u
i )dudv +
__
B
¸R◦
¯
T
i
,
ℓ
R
u
i )dudv
_
=
__
B
¸R◦
¯
T
i
, R
u
i )dudv.
In addition we obtain
lim
ℓ→∞
__
B
[
ℓ
R
u
i [
2
dudv ≥
__
B
[R
u
i [
2
dudv
due to the semicontinuity of the L
2
norm w.r.t. weak convergence. Putting the last
two relations into the identity
[
ℓ
T
i
[
2
=[
ℓ
R
u
i [
2
+2¸
ℓ
R◦
¯
T
i
,
ℓ
R
u
i ) +[
¯
T
i
[
2
we ﬁnally infer
lim
ℓ→∞
F
X
[
ℓ
R] =
1
2
lim
ℓ→∞
__
B
_
[
ℓ
T
1
[
2
+[
ℓ
T
2
[
2
_
dudv
≥
1
2
__
B
_
[R
u
[
2
+[R
v
[
2
_
dudv +
__
B
_
¸R◦
¯
T
1
, R
u
) +¸R◦
¯
T
2
, R
v
)
_
dudv
+
1
2
__
B
_
[
¯
T
1
[
2
+[
¯
T
2
[
2
_
dudv
=
1
2
__
B
_
[T
1
[
2
+[T
2
[
2
_
dudv = F
X
[R]
154 8 Normal Coulomb frames in R
n+2
where T
i
= (T
ϑ
σ,i
)
σ,ϑ=1,...,n
denote the torsion coefﬁcients of the frame N with en
tries N
σ
:=∑
ϑ
r
σϑ
¯
N
ϑ
. Consequently,
N ∈ H
1,2
(
˚
B) ∩L
∞
(
˚
B)
minimizes T
X
[N] and, in particular, it is a weak normal Coulomb frame. ⊓⊔
H
2,1
loc
regularity of weak normal Coulomb frames
To prove higher regularity of normal Coulomb frames we make essential use of
techniques from harmonic analysis. Consult eventually paragraph 8.8. We always
use conformal parameters (u, v) ∈ B.
Proposition 8.11. Any weak normal Coulomb frame N ∈ H
1,2
(
˚
B) ∩L
∞
(
˚
B) for the
conformally parametrized immersion X : B →R
n+2
belongs to the class H
2,1
loc
(
˚
B).
Proof. 1. Following section 8.2 the torsion coefﬁcients T
ϑ
σ,i
, σ, ϑ = 1, . . . , n, of the
normal Coulomb frame N are weak solutions of the EulerLagrange equations
div(T
ϑ
σ,1
, T
ϑ
σ,2
) = 0 in
˚
B.
Hence by a weak version of Poincare’s lemma (see e.g. Bourgain, Brezis and
Mironescu [17] Lemma 3), there exist integral functions ∂
u
τ
(σϑ)
∈ H
1,2
(B) sat
isfying
∂
u
τ
(σϑ)
=−T
ϑ
σ,2
, ∂
v
τ
(σϑ)
= T
ϑ
σ,1
in
˚
B
Thus the weak form of the EulerLagrange equations can be written as
0 =
__
B
_
ϕ
u
∂
v
τ
(σϑ)
−ϕ
v
∂
u
τ
(σϑ)
_
dudv =
_
∂B
τ
(σϑ)
∂ϕ
∂t
ds
for all ϕ ∈C
∞
(B, R) where
∂ϕ
∂t
denotes the tangential derivative of ϕ along ∂B.
Note that τ
(σϑ)
[
∂B
means the L
2
trace of τ
(σϑ)
on the boundary curve ∂B (see
e.g. Alt [3], chapter 6, appendix A6.6)
7
. Consequently, the lemma of DuBois
Reymond
8
yields τ
(σϑ)
≡const on ∂B, and by translation we arrive at the bound
ary conditions
τ
(σϑ)
= 0 on ∂B.
7
Let 1 ≤ p ≤ ∞. Then there is a uniquely determined map S: H
1, p
(
˚
B, R) →L
p
(
˚
B, R) such that
S(φ)
L
p
(∂B)
≤Cφ
H
1,2
(
˚
B)
. Additionally it holds S(φ) =φ
¸
¸
∂B
if φ ∈H
1,2
(
˚
B, R) ∩C
0
(B, R)). The
map S is called the trace mapping.
8
Its onedimensional version is the following: Let f ∈ L
1
((a, b), R) and assume that
_
b
a
f (x)ϕ(x)dx = 0 for all ϕ ∈C
∞
([a, b], R). Then it holds f ≡const almost everywhere.
8.8 Existence and regularity of weak normal Coulomb frames 155
2. Thus the integral functions τ
(σϑ)
are weak solutions of the secondorder system
∆τ
(σϑ)
=−T
ϑ
σ,2,u
+T
ϑ
σ,1,v
=−N
σ,v
N
ϑ,u
+N
σ,u
N
ϑ,v
in
˚
B
where the second identity follows by direct differentiation. By a result Coifman,
Lions, Meyer and Semmes [34],
9
the righthand side of divcurl type belongs
to the Hardy space H
1
loc
(
˚
B, R) and, hence, the τ
(σϑ)
belong to H
2,1
loc
(
˚
B, R) by
a result of Fefferman and Stein[name]Stein, E.M. [55].
10
Consequently we ﬁnd
T
ϑ
σ,i
∈ H
1,1
loc
(
˚
B, R) ∩L
2
(
˚
B, R). Next, we employ the Weingarten equations
N
σ,u
i =−
2
∑
j,k=1
L
σ,i j
g
jk
X
u
k +
n
∑
ϑ=1
T
ϑ
σ,i
N
ϑ
in a weak form. For the coefﬁcients L
σ,i j
of the second fundamental form
we have L
σ,i j
= N
σ
X
u
i
u
j leading to L
σ,i j
∈ H
1,2
(
˚
B, R) and taking account
of N ∈ H
1,2
(
˚
B) ∩L
∞
(
˚
B) and X
u
i
u
j ∈ L
∞
(
˚
B, R
n+2
). Hence we arrive at N
σ,u
i ∈
H
1,1
loc
(
˚
B, R
n+2
) and N ∈ H
2,1
loc
(
˚
B) for our weak normal Coulomb frame. Note
that T
ϑ
σ,i
∈ H
1,1
loc
(
˚
B, R) ∩L
2
(
˚
B, R) and N
ϑ
∈ H
1,2
(
˚
B, R
n+2
) ∩L
∞
(
˚
B, R
n+2
) imply
T
ϑ
σ,i
N
ϑ
∈ H
1,1
loc
(
˚
B, R
n+2
) by a careful adaption of the classical product rule in
Sobolev spaces which is explained in the lemma following immediately. We have
proved the statement. ⊓⊔
Lemma 8.2. Under the assumptions of the foregoing proposition it holds
n
∑
ϑ=1
T
ϑ
σ,i
N
ϑ
∈ H
1,1
(
˚
B, R
n+2
).
Proof. Note that T
ϑ
σ,i
N
ϑ
∈L
2
(
˚
B, R
n+2
) since T
ϑ
σ,i
∈L
2
(
˚
B, R) and N
ϑ
∈L
∞
(
˚
B, R
n+2
).
We show that T
ϑ
σ,i
N
ϑ
has a weak derivative, i.e. we prove that
T
ϑ
σ,i
N
ϑ,u
j +N
ϑ
∂
u
j T
ϑ
σ,i
∈ L
1
(
˚
B, R
n+2
)
is the weak derivative of T
ϑ
σ,i
N
ϑ
. In other words
__
B
(T
ϑ
σ,i
N
ϑ,u
j +N
ϑ
∂
u
j T
ϑ
σ,i
)ϕ dudv =−
__
B
(T
ϑ
σ,i
N
ϑ
)ϕ
u
j dudv
for all ϕ ∈C
∞
0
(B, R). For such a test function ϕ deﬁne
ψ := T
ϑ
σ,i
ϕ ∈W
1,1
0
(
˚
B, R) ∩L
2
(
˚
B, R).
9
Let φ : H
1,2
(R
2
, R). Then f := det (∇φ) ∈ H
1
(R
2
, R), and it holds  f 
H
1
(R
2
)
≤Cφ
H
1,2
(R
2
)
.
10
Let φ ∈ L
1
(R
2
, R) be a solution of −∆φ = f ∈ H
1
(R
2
, R). Then all second derivatives of φ
belong to L
1
(R
2
, R), and it holds φ
x
α
x
β 
L
1
(R
m
)
≤C f 
L
1
(R
m
)
for all α, β = 1, 2.
156 8 Normal Coulomb frames in R
n+2
We claim
__
B
N
u
j ψdudv =−
__
B
Nψ
u
j dudv.
For the proof of this relation we approximate ψ with smooth functions ψ
ε
∈
C
∞
0
(B, R) in the sense of Friedrichs. Then ψ
ε
→ ψ in H
1,1
(
˚
B, R) ∩L
2
(
˚
B, R), and
ψ = 0 outside some compact set K ⊂⊂
˚
B. We estimate as follows
¸
¸
¸
¸
¸
¸
__
B
(N
ϑ,u
j ψ +N
ϑ
ψ
u
j )dudv
¸
¸
¸
¸
¸
¸
=
¸
¸
¸
¸
¸
¸
__
B
(N
ϑ,u
j ψ
ε
+N
ϑ
ψ
ε
u
j
)dudv
¸
¸
¸
¸
¸
¸
+
¸
¸
¸
¸
¸
¸
__
B
N
ϑ,u
j (ψ −ψ
ε
)dudv
¸
¸
¸
¸
¸
¸
+
¸
¸
¸
¸
¸
¸
__
B
N
ϑ
(ψ
ε
u
j
−ψ
u
j )dudv
¸
¸
¸
¸
¸
¸
≤ N
ϑ,u
j 
L
2
(
˚
B)
ψ −ψ
ε

L
2
(
˚
B)
+N
ϑ

L
∞
(
˚
B)
ψ
ε
u
j
−ψ
u
j 
L
1
(
˚
B)
taking
__
B
N
u
j ψ
ε
dudv =−
__
B
Nψ
ε
u
j
dudv
into account. Since ψ −ψ
ε

L
2
(
˚
B)
→ 0 and ψ
ε
u
j
−ψ
u
j 
L
1
(
˚
B)
→ 0 for ε → 0 we
arrive at the stated identity. Now we use the product rule and calculate
__
B
(T
ϑ
σ,i
N
ϑ,u
j +N
ϑ
∂
u
j T
ϑ
σ,i
)ϕ dudv =
__
B
N
ϑ,u
j ψdudv +
__
B
N
ϑ
∂
u
j T
ϑ
σ,i
ϕ dudv
= −
__
B
(T
ϑ
σ,i
ϕ)
u
j N
ϑ
dudv +
__
B
N
ϑ
∂
u
j T
ϑ
σ,i
ϕ dudv
= −
__
B
∂
u
j T
ϑ
σ,i
N
ϑ
ϕ dudv −
__
B
T
ϑ
σ,i
ϕ
u
j Ndudv +
__
B
N
ϑ
∂
u
j T
ϑ
σ,i
ϕ dudv
= −
__
B
(T
ϑ
σ,i
N
ϑ
)ϕ
u
j dudv.
This proves the statement. ⊓⊔
8.9 Classical regularity of normal Coulomb frames 157
8.9 Classical regularity of normal Coulomb frames
Our main result of this chapter is a proof of classical regularity of normal Coulomb
frames. Up to now we only know regularity in the sense of Sobolev spaces. An
essential tool on our road to regularity are again the Weingarten equations.
Theorem 8.3. For any conformally parametrized immersion X ∈ C
k,α
(B, R
n+2
)
with k ≥ 3 and α ∈ (0, 1) there exists a normal Coulomb frame N ∈ C
k−1,α
(B)
minimizing the functional T
X
[N] of total torsion.
Proof. 1. We ﬁx some normal frame
¯
N ∈ C
k−1,α
(B), and as above we construct
a weak normal Coulomb frame N from the regularity class H
1,2
(
˚
B) ∩L
∞
(
˚
B).
Furthermore we then know N ∈ H
2,1
loc
(
˚
B). Deﬁning the orthogonal mapping R =
(r
σϑ
)
σ,ϑ=1,...,n
by r
σϑ
:= N
σ
¯
N
ϑ
we thus ﬁnd
N
σ
=
n
∑
ϑ=1
r
σϑ
¯
N
ϑ
and R ∈ H
2,1
loc
(
˚
B, SO(n)) ∩H
1,2
(
˚
B, SO(n)).
In particular, we can assign a normal curvature matrix S
12
= (S
ϑ
σ,12
)
σ,ϑ=1,...,n
∈
L
1
loc
(
˚
B, R
nn
) to N by the formula
S
ϑ
σ,12
=∂
v
T
2
σ,1
−∂
u
T
ϑ
σ,2
+
n
∑
ω=1
(T
ω
σ,1
T
ϑ
ω,2
−T
ω
σ,2
T
ϑ
ω,1
).
From the previous section we then infer S
12
∈ L
∞
(
˚
B, R
nn
).
2. Introduce τ = (τ
(σϑ)
)
σ,ϑ=1,...,n
∈ H
1,2
(
˚
B, R
nn
) by
∂
u
τ
(σϑ)
=−T
ϑ
σ,2
, ∂
v
τ
(σϑ)
= T
ϑ
σ,1
in
˚
B,
τ
(σϑ)
= 0 on ∂B.
The above deﬁnition of the normal curvature marix gives us the nonlinear elliptic
system
∆τ
(σϑ)
= −
n
∑
ω=1
(∂
u
τ
(σω
∂
v
τ
(ωϑ)
−∂
v
τ
(σω)
∂
u
τ
(ωϑ)
) +S
ϑ
σ,12
in
˚
B,
τ
(σϑ)
= 0 on ∂B.
On account of S
12
= (S
ϑ
σ,12
)
σ,ϑ=1,...,n
∈ L
∞
(
˚
B, R
nn
), a part of Wente’s inequal
ity yields τ ∈C
0
(B, R
nn
), see e.g. Brezis and Coron [19]; compare also Rivi` ere
[136] and the corresponding boundary regularity theorem in M¨ uller and Schiko
rra [124] for more general results. By appropriate reﬂection of τ and S
12
(the
reﬂected quantities are again denoted by τ and S
12
) we obtain a weak solution
158 8 Normal Coulomb frames in R
n+2
τ ∈ H
1,2
(
˚
B
1+d
, R
nn
) ∩C
0
(
˚
B
1+d
, R
nn
) of
∆τ = f (w, ∇τ) in
˚
B
1+d
:=¦w ∈ R
2
: [w[ < 1 +d¦
with some d > 0 and a righthand side f satisfying
[ f (w, p)[ ≤a[p[
2
+b for all p ∈ R
2n
2
, w ∈
˚
B
1+d
with some reals a, b > 0. Now applying Tomi’s regularity result from [157] for
weak solutions of this nonlinear system possessing small variation locally in
˚
B
1+d
, we ﬁnd τ ∈ C
1,ν
(B, R
nn
) for any ν ∈ (0, 1) (note that Tomi’s result ap
plies for such systems with b = 0, but his proof can easily be adapted to our
inhomogeneous case b > 0).
3. From the ﬁrstorder system for τ
(σϑ)
we infer T
i
∈ C
α
(B, R
nn
). Thus the
Weingarten equations for N
σ,u
i yield N ∈ W
1,∞
(
˚
B) on account of N ∈ L
∞
(
˚
B),
and we obtain N ∈ C
α
(B) by Sobolev’s embedding theorem. Inserting this
again into the Weingarten equations we ﬁnd N ∈ C
1,α
(B). Hence we conclude
R = (N
σ
¯
N
ϑ
)
σ,ϑ=1,...,n
∈C
1,α
(B, R
nn
), and the transformation rule S
12
= R◦
¯
S
12
◦ R
t
implies S
12
= (S
ϑ
σ,12
)
σ,ϑ=1,...,n
∈C
α
(B, R
nn
) (note
¯
S
12
∈C
α
(B, R
nn
)
for k = 3; in case k ≥ 4 we even get S
12
∈ C
1,α
(B, R
nn
)). Now the righthand
side of the equation for ∆τ
(σϑ)
belongs to C
α
(B, R), and potential theoretic
estimates ensure τ ∈ C
2,α
(B, R
nn
). Involving again our ﬁrstorder system for
the τ
(σϑ)
gives T
i
∈ C
1,α
(B, R
nn
) which proves N ∈ C
2,α
(B) using the Wein
garten equations once more. Finally, for k ≥ 4, we can bootstrap by concluding
R ∈C
2,α
(B, SO(n)) and S
12
∈C
1,α
(B) from the transformation rule for S
12
and
repeating the arguments above. This completes the proof. ⊓⊔
Chapter 9
Minimal surfaces
9.1 Minimal surfaces
9.2 The ﬁrst variation of nonparametric functionals. Minimal graphs
9.3 Geometry of minimal surfaces
9.4 A priori estimates for elliptic systems with quadratic growth
9.5 A curvature estimate for minimal graphs with subquadratic growth
9.6 A theorem of Bernstein type
9.7 Osserman’s curvature estimate for minimal surfaces
9.8 A theorem of Berntein type due to Osserman
9.9 Gradient estimates for minimal graphs
160 9 Minimal surfaces
In this chapter we consider minimal surfaces as critical points of the area functional in parametric
and nonparametric form. We investigate analytical and geometric properties of these immersions,
and we present various proofs for establishing curvature estimates and theorems of Bernstein type.
9.1 Minimal surfaces
Minimal surfaces are critical points of the area functional
A[X] :=
__
B
W dudv
which implies their characteristic property H ≡0 as the following theorem shows.
Theorem 9.1. The functional
δ
´
N
A[X; ϕ] :=
d
dε
A[X +εϕ
´
N]
¸
¸
¸
ε=0
of the ﬁrst variation of A[X] w.r.t. perturbations
¯
X(u, v) = X(u, v) +εϕ(u, v)
´
N(u, v)
with an arbitrary unit normal vector
´
N, a smooth function ϕ ∈ C
∞
0
(B, R) and a
parameter ε ∈ (−ε
0
, +ε
0
) is given by
δ
´
N
A[X; ϕ] =−2
__
B
H
´
N
Wϕ dudv
with the mean curvature H
´
N
along
´
N. Thus critical points X : B →R
n+2
of A[X]
necessarily satisfy
H ≡0 in B.
Proof. Let (u, v) ∈ B be conformal parameters. Choose an ONF N = (N
1
, . . . , N
n
)
and consider normal variations
¯
X(u, v) = X(u, v) +εϕ(u, v)N
´ γ
(u, v)
162 9 Minimal surfaces
with arbitrary ϕ ∈C
∞
0
(B, R) and ε ∈ (−ε
0
, ε
0
), where we set
N
´ γ
(u, v) :=
n
∑
σ=1
´ γ
σ
(u, v)N
σ
(u, v),
n
∑
σ=1
´ γ
σ
(u, v)
2
= 1.
We compute
¯
X
u
= X
u
+εϕ
u
N
´ γ
+εϕN
´ γ,u
,
¯
X
v
= X
v
+εϕ
v
N
´ γ
+εϕN
´ γ,v
as well as
¯
X
2
u
= W +2εϕX
u
N
´ γ,u
+ε
2
ϕ
2
u
+ε
2
ϕ
2
N
2
´ γ,u
,
¯
X
2
v
= W +2εϕX
v
N
´ γ,v
+ε
2
ϕ
2
v
+ε
2
ϕ
2
N
2
´ γ,v
,
¯
X
u
¯
X
v
= ε
_
X
u
N
´ γ,v
+X
v
N
´ γ,u
_
ϕ +ε
2
ϕ
u
ϕ
v
+ε
2
ϕ
2
N
´ γ,u
N
´ γ,v
.
(9.1)
Let us now introduce the following auxiliary forms
L
´ γ,i j
:= X
u
i
u
j N
´ γ
=−X
u
i N
´ γ,u
j =−X
u
j N
´ γ,u
i
taking
N
´ γ,u
i N
´ γ
= 0 (because N
2
´ γ
= 1)
X
u
i N
´ γ,u
j = −X
u
i
u
j N
´ γ
(because X
u
i N
´ γ
= 0)
into account as well as X
u
N
´ γ,v
=X
v
N
´ γ,u
due to the symmetrie of L
σ,i j
. In particular,
it holds
L
´ γ,i j
= −X
u
i N
´ γ,u
j = −X
u
i
_
n
∑
σ=1
´ γ
σ
u
j
N
σ
+
n
∑
σ=1
´ γ
σ
N
σ,u
j
_
= −
n
∑
σ=1
´ γ
σ
X
u
i N
σ,u
j =
n
∑
σ=1
´ γ
σ
L
σ,i j
for i, j = 1, 2 which shows L
´ γ,i j
= L
´ γ, ji
. We rewrite the equations (9.1) into the form
¯
X
2
u
= W −2εϕL
´ γ,11
+ε
2
ϕ
2
u
+ε
2
ϕ
2
N
2
´ γ,u
,
¯
X
2
v
= W −2εϕL
´ γ,22
+ε
2
ϕ
2
v
+ε
2
ϕ
2
N
2
´ γ,v
,
¯
X
u
¯
X
v
= −2εϕL
´ γ,12
+ε
2
ϕ
u
ϕ
v
+ε
2
ϕ
2
N
´ γ,u
N
´ γ,v
which implies
δg
11
=−2ϕL
´ γ,11
, δg
12
=−2ϕL
´ γ,12
, δg
22
=−2ϕL
´ γ,22
9.2 The ﬁrst variation of nonparametric functionals. Minimal graphs 163
with the settings
δg
11
:=
∂ ¯ g
11
∂ε
¸
¸
¸
ε=0
etc. Now in view of
2WδW = g
22
δg
11
−2g
12
δg
12
+g
11
δg
22
with g
11
= g
22
and g
12
≡0 we arrive at the ﬁrst variation formula
δW =
1
2
δg
11
+
1
2
δg
22
=−
_
L
´ γ,11
+L
´ γ,22
_
ϕ =−2H
´ γ
Wϕ (9.2)
where we set
H
´ γ
:=
L
´ γ,11
g
22
−L
´ γ,12
g
12
+L
´ γ,22
g
11
2W
2
.
This proves the statement. ⊓⊔
9.2 The ﬁrst variation of nonparametric functionals. Minimal
graphs
The general problem
In this section we want to consider surface graphs of the form
X(x, y) = (x, y, ζ
1
(x, y), . . . , ζ
n
(x, y)), Ω ⊂R
2
,
which are critical for the nonparametric area functional. We start with computing
the ﬁrst variation of a general nonparametric functional before we treat the special
case of minimal surfaces.
We introduce the following abbreviations
p
σ
:=
∂ζ
σ
∂x
, q
σ
:=
∂ζ
σ
∂y
, σ = 1, . . . , n,
as well as
ζ = (ζ
1
, . . . , ζ
n
), p = (p
1
, . . . , p
n
), q = (q
1
, . . . , q
n
),
r = (r
1
, . . . , r
2n
) = (p
1
, q
1
, . . . , p
n
, q
n
).
Let us now consider the general nonparametric functional
F[ζ] :=
__
Ω
F(x, y, ζ, ∇ζ)dxdy =
__
Ω
F(x, y, ζ, r)dxdy.
164 9 Minimal surfaces
For the next computations we refer the reader to Fomin and Gelfand [61], Funk
[68] or Evans [53].
Theorem 9.2. The system of n EulerLagrange equations for the functional F[ζ]
reads
dF
p
σ
(x, y, ζ, r)
dx
+
dF
q
σ
(x, y, ζ, r)
dy
= F
z
σ
(x, y, ζ, r) for σ = 1, . . . , n.
Proof. Let ε ∈ (−ε
0
, ε
0
). We consider perturbations
¯
ζ
σ
=ζ
σ
+εϕ
σ
with ϕ
σ
∈C
∞
0
(Ω, R) for σ = 1, . . . , n. Let ϕ = (ϕ
1
, . . . , ϕ
n
). Then it follows
F[
¯
ζ] −F[ζ]
=
__
Ω
F(x, y, ζ +εϕ, ∇ζ +ε∇ϕ)dxdy −
__
Ω
F(x, y, ζ, ∇ζ)dxdy
= ε
n
∑
σ=1
__
Ω
F
z
σ
(x, y, ζ, ∇ζ)ϕ
σ
dxdy
+ε
n
∑
σ=1
__
Ω
_
F
p
σ
(x, y, ζ, ∇ζ)ϕ
σ,x
+F
q
σ
(x, y, ζ, ∇ζ)ϕ
σ,y
_
dxdy +o(ε).
Notice that
F
p
σ
(x, y, ζ, ∇ζ)ϕ
σ,x
=
d
dx
_
F
p
σ
(x, y, ζ, ∇ζ)ϕ
σ
_
−
dF
p
σ
(x, y, ζ, ∇ζ)
dx
ϕ
σ
,
F
q
σ
(x, y, ζ, ∇ζ)ϕ
σ,y
=
d
dy
_
F
q
σ
(x, y, ζ, ∇ζ)ϕ
σ
_
−
dF
q
σ
(x, y, ζ, ∇ζ)
dy
ϕ
σ
as well as
__
Ω
div
_
F
p
σ
(x, y, ζ, ∇ζ)ϕ
σ
, F
q
σ
(x, y, ζ, ∇ζ)ϕ
σ
_
dxdy = 0 for all σ = 1, . . . , n
due to the choice of the test functions ϕ
σ
. Thus we arrive at
F[
¯
ζ] −F[ζ] = ε
n
∑
σ=1
__
Ω
F
z
σ
(x, y, ζ, ∇ζ)ϕ
σ
dxdy
−ε
n
∑
σ=1
__
Ω
_
dF
p
σ
(x, y, ζ, ∇ζ)
dx
+
dF
q
σ
(x, y, ζ, ∇ζ)
dy
_
ϕ
σ
dxdy +o(ε).
The fundamental lemma of the calculus of variations proves the statement. ⊓⊔
9.2 The ﬁrst variation of nonparametric functionals. Minimal graphs 165
Minimal graphs
We want to specify the foregoing considerations to the problem of establishing the
EulerLagrange equations for area functional. Let X(x, y) be a surface graph with
tangential vectors
X
x
= (1, 0, ζ
1,x
, . . . , ζ
n,x
), X
y
= (0, 1, ζ
1,y
, . . . , ζ
n,y
).
For the coefﬁcients g
i j
of its ﬁrst fundamental form we therefore compute
g
11
= 1 +
n
∑
σ=1
ζ
2
σ,x
, g
12
=
n
∑
σ=1
ζ
σ,x
ζ
σ,y
, g
22
= 1 +
n
∑
σ=1
ζ
2
σ,y
,
and the squared area element W(x, y) takes the form
W
2
= g
11
g
22
−g
2
12
= 1 +
n
∑
σ=1
ζ
2
σ,x
+
n
∑
σ=1
ζ
2
σ,y
+
n
∑
σ=1
ζ
2
σ,x
n
∑
σ=1
ζ
2
σ,y
−
_
n
∑
σ=1
ζ
σ,x
ζ
σ,y
_
2
.
Thus the nonparametric area functional reads
A[ζ] =
__
Ω
¸
¸
¸
_
1 +
n
∑
σ=1
ζ
2
σ,x
+
n
∑
σ=1
ζ
2
σ,y
+
n
∑
σ=1
ζ
2
σ,x
n
∑
σ=1
ζ
2
σ,y
−
_
n
∑
σ=1
ζ
σ,x
ζ
σ,y
_
2
dxdy.
(9.3)
In the special case n = 1 of one codimension there is only one function ζ(x, y)
generating the surface graph, and the area functional can be written in the simpler
form
A[ζ] =
__
Ω
_
1 +ζ
2
x
+ζ
2
y
dxdy.
The system of n EulerLagrange equations then reduces to one single equation.
Now back to the case of arbitrary codimension. Using our abbreviations p ∈ R
n
and q ∈ R
n
from above we infer
W ≡F(p, q) =
_
1 + p
2
+q
2
+ p
2
q
2
−(p q)
2
for the integrand of the functional A[ζ]. Differentiation gives
∂F(p, q)
∂ p
σ
=
p
σ
+ p
σ
q
2
−q
σ
(p q)
W
,
∂F(p, q)
∂q
σ
=
q
σ
+q
σ
p
2
−p
σ
(p q)
W
.
Note furthermore F
z
σ
(p, q) ≡0 for σ = 1, . . . , n.
166 9 Minimal surfaces
Corollary 9.1. The nonparametric minimal surface system reads
div
∇ζ
σ
W
=−div
_
ζ
σ,x
q
2
−ζ
σ,y
p q, ζ
σ,y
p
2
−ζ
σ,x
p q
_
W
in Ω (9.4)
for σ = 1, . . . , n.
Note that the right hand side of this EulerLagrange system vanishes identically
in case n = 1 of one codimension. Then we particularly get
div
∇ζ
W
= 0 in Ω. (9.5)
Let us ﬁnally note down (9.4) for graphs X(x, y) = (x, y, ϕ(x, y), ψ(x, y)) :
div
∇ϕ
W
= −div
_
(ϕ
x
ψ
y
−ϕ
y
ψ
x
)ψ
y
, (ϕ
y
ψ
x
−ϕ
x
ψ
y
)ψ
x
_
W
,
div
∇ψ
W
= −div
_
(ϕ
y
ψ
x
−ϕ
x
ψ
y
)ϕ
y
, (ϕ
x
ψ
y
−ϕ
y
ψ
x
)ϕ
x
_
W
.
9.3 Geometry of minimal surfaces
Following section 2.3, the characteristic property H ≡0 for a conformally parametrized
minimal immersion X : B →R
n+2
is equivalent to
∆X(u, v) = 0 in B.
From this Laplace equation we will derive some interesting geometric properties
of minimal surfaces.
Embedding into subspaces
A ﬁrst remarkable consequence is the following:
Let a minimal surface X : B →R
4
with X(∂B) ⊂ R
3
be given. Then the whole
surface X(B) is also contained in the subspace R
3
.
The general situation is the contents of the
Proposition 9.1. Let the minimal surface X : B →R
n+2
satisfy
X
u
(u, v) e = X
v
(u, v) e = 0 on ∂B
with a ﬁxed vector e ∈ R
n+2
. Then there hold
X
u
(u, v) e = X
v
(u, v) e = 0 in B.
9.3 Geometry of minimal surfaces 167
Proof. Introduce conformal parameters (u, v) ∈ B, and let Φ
i
:= X
u
i e for i = 1, 2.
Then we compute
∆Φ
i
= (X
u
i
uu
+X
u
i
vv
) e = (X
uu
+X
vv
)
u
i e = (∆X e)
u
i = 0.
Thus the Φ
i
are harmonic functions, and the maximum principle proves the state
ment. ⊓⊔
The convex hull property
Next let a minimal surface X : B →R
n+2
together with an ONF N be given. Then
H ≡0 immediately implies
κ
σ,1
+κ
σ,2
≡0
for the principal curvatures κ
σ,1
and κ
σ,2
along an unit normal vectors N
σ
of N.
Thus we obtain
K
σ
=κ
σ,1
κ
σ,2
≤0
for the Gauss curvatures along N
σ
, and we infer
K =
n
∑
σ=1
K
σ
≤0 in B
for the Gauss curvature of the minimal immersion.
But surfaces with nonpositive curvature satisfy the socalled convex hull prop
erty. In particular, using an idea which goes back to S. Hildebrandt we prove the
Proposition 9.2. Let the minimal surface X : B →R
n+2
be given. Then it holds
X(B) ⊂convX(∂B)
with convX(∂B) denoting the convex hull of the curve X(∂B).
Proof. Use conformal parameters (u, v) ∈ B. Choose a unit vector Z ∈ R
n+2
with
[Z[ = 1 and a real number h ∈ R such that
X(∂B) ⊂H
Z,h
:=
_
X ∈ R
n+2
: Z X −h ≤0
_
.
Along with the minimal immersion X(u, v) we now deﬁne the function
Φ(u, v) := Z X(u, v) −h, (u, v) ∈ B,
which satisﬁes
∆Φ(u, v) = 0 in B, Φ(u, v) ≤0 on ∂B.
168 9 Minimal surfaces
Thus the maximum principle implies Φ(u, v) ≤0 in B, therefore X(B) ⊂H
Z,h
. But
on the other it holds
convX(∂B) =
Z,h
H
Z,h
,
i.e. it follows X(B) ⊂H
Z,h
for all half spaces H
Z,h
proving the statement. ⊓⊔
Various geometric enclosure theorems for minimal surfaces can be found e.g.
in the monograph Burago and Zalgaller [20]. We also want to refer to Dierkes et
al. [44], Nitsche [126], and Osserman [130]. For further developments we refer to
Dierkes [42], [43].
In particular, enclosure theorems of this kind are useful for establishing curvature
estimates as we will elaborate shortly.
9.4 A priori estimates for elliptic systems with quadratic growth
For this purpose we need some additional analytical facts from the theory of quasi
linear elliptic systems of second order with quadratic growth in the gradient. We
refer the reader to the work of Heinz [80] as well as to Schulz [146], Dierkes et al.
[44], and Sauvigny [143].
Such elliptic systems played an important role for our considerations from chap
ter 8. In particular, the model problem
∆Φ(u, v) = r(u, v) in B
with the right hand side
r =
∂a
∂u
∂b
∂v
−
∂a
∂v
∂b
∂u
discussed there growth quadratically in the gradients of a and b, i.e.
[∆Φ[ ≤
1
2
([a
u
[
2
+[b
v
[
2
+[a
v
[
2
+[b
v
[
2
) =
1
2
([∇a[
2
+[∇b[
2
).
In the following we consider mappings X : B →R
n+2
satisfying
[∆X[ ≤λ[∇X[
2
+µ in B (9.6)
with nonnegative real λ, µ ∈ [0, ∞).
The following results are taken from Sauvigny [143], chapter 12.
9.4 A priori estimates for elliptic systems with quadratic growth 169
Interior gradient estimates
Assume that the smallness condition
λ sup
(u,v)∈B
[X(u, v)[ < 1 in B
holds true. Let
δ(w) := dist (w, ∂B), d := sup
w∈B
δ(w).
Then there exists a constant c
1
= c
1
(λ, µ, M, d) ∈ [0, ∞) such that
δ(w)[∇X(w)[ ≤c
1
(λ, µ, M, d), M := sup
(u,v)∈B
[X(u, v)[.
Likewise it holds
X
C
1+α
(K
ε
)
≤c
2
(λ, µ, M, d, ε, α) ∈ [0, ∞) for α ∈ (0, 1)
for all compact domains K
ε
⊂⊂
˚
B.
Global gradient estimates
If we otherwise assume additionally
X(u, v) = 0 on ∂B
we can made a statement about a gradient estimate up to ∂B. Namely it holds
X
C
1+α
(B)
≤c
3
(λ, µ, M, α) for α ∈ (0, 1)
with a real constant c
3
∈ [0, ∞). The proof uses a reﬂection principle for (9.6).
Interior estimates for the Jacobian
Now consider a plane mapping X : B → B solving system (9.6) with λ ∈ [0,
1
2
),
X : ∂B →∂B topologically and positively oriented, X(0, 0) = (0, 0), and
J
X
(u, v) =
∂(x, y)
∂(u, v)
> 0 in B
for its Jacobian. Then it holds
c
4
(λ, µ, r) ≤[∇X(w)[ ≤c
5
(λ, µ, r) for all compact B
r
⊂⊂
˚
B
with positive constants c
4
, c
5
∈ (0, ∞).
170 9 Minimal surfaces
If additionally
__
B
(X
2
u
+X
2
v
)dudv ≤N
with a real constant N ∈ (0, ∞), then the latter estimate is true for 0 ≤λ < 1, i.e.
c
∗
4
(λ, µ, r, N) ≤[∇X(w)[ ≤c
∗
5
(λ, µ, r, N) for all compact B
r
⊂⊂
˚
B
with constants c
∗
4
, c
∗
5
∈ (0, ∞).
9.5 A curvature estimate for minimal graphs with subquadratic
growth
Using the analytical tools for nonlinear elliptic systems just presented together with
a classical Schauder estimate fromGilbarg and Trudinger [71] we can proof our ﬁrst
curvature estimate.
Theorem 9.3. Let the minimal graph
X(x, y) = (x, y, ζ
1
(x, y), . . . , ζ
n
(x, y))
on the disc closed disc B
R
=¦(x, y) ∈ R
2
: x
2
+y
2
≤R¦ be given. Let
X = X(u, v): B −→R
n+2
be a reparametrization of this graph such that X(u, v) is conformally parametrized.
Then there exists a real constant Θ ∈ [0, +∞) such that for the Gaussian curvature
K
N
(0, 0) of X(u, v) along an arbitrary unit normal vector N it holds
[K
N
(0, 0)[ ≤
Θ
R
4
X
2
C
0
(B
R
)
.
Proof. 1. Consider the conformal representation X(u, v) of the graph. Then we can
ensure that
f (u, v) := (x(u, v), y(u, v)), (u, v) ∈ B,
maps the boundary ∂B topologically and positively orientied onto ∂B
R
, and it
satisﬁes f (0, 0) = (0, 0) as well as J
f
(w) > 0 in B for the Jacobian of f (u, v)
since X(u, v) represents a surface graph; see Sauvigny [143]. Due to
△f (u, v) = 0 in B
there exists a real constant C
1
∈ (0, +∞) such that
[∇f (0, 0)[ ≥C
1
R.
9.6 A theorem of Bernsteintype 171
We infer
W(0, 0) =
1
2
[∇X(0, 0)[
2
≥
1
2
[∇f (0, 0)[
2
≥
1
2
C
2
1
R
2
.
2. Using conformal parameters it holds
[K
N
(0, 0)[ ≤
[N(0, 0) X
uu
(0, 0)[[N X
vv
(0, 0)[ +[N X
uv
(0, 0)[
2
W(0, 0)
2
≤
[X
uu
(0, 0)[[X
vv
(0, 0)[ +[X
uv
(0, 0)[
2
W(0, 0)
2
.
Due to △X(u, v) = 0 in B there is a universal constant C
2
∈ (0, +∞) with the
property (see e.g. Gilbarg and Trudinger [71], Theorem 4.6)
[X
u
i
u
j (0, 0)[ ≤C
2
X
C
0
(B
R
)
for i, j = 1, 2. (9.7)
Note here that the supremum norm does not depend on the choice of the param
eters. Putting all estimates together we arrive at
[K
N
(0, 0)[ ≤
8C
2
2
C
4
1
R
4
X
2
C
0
(B
R
)
=:
Θ
R
4
X
2
C
0
(B
R
)
setting Θ := 8C
2
2
C
−4
1
. The statement is proved. ⊓⊔
9.6 A theorem of Bernsteintype
Complete geodesic discs
For our curvature estimates later we need some new terminology the reader can ﬁnd
in any textbook on Riemannian geometry or geometric analysis. Let us refer e.g. to
Klingenberg [106] or Sauvigny [140].
We consider twodimensional simply connected and complete immersions M
2
in
the following sense.
Deﬁnition 9.1. The regular surface M
2
⊂R
n+2
is called complete if for every point
p ∈ M
2
any geodesic curve γ : [0, ε) → M
2
with γ(0) = p can be extended to a
parametric geodesic curve ¯ γ : R →M
2
deﬁned on the whole real line R.
Elementary examples of complete surfaces are the plane R
2
, the cylinder, or the
sphere. The minimal surface (w, w
2
) is also complete.
Geodesic curves are deﬁned by means of the socalled exponential map
exp
p
: T
M
2 (p) −→M
2
on the twodimensional tangential space T
M
2 (p) of M
2
at p ∈ M
2
.
172 9 Minimal surfaces
For sufﬁciently small r > 0 satisfying
r
_
K
0
<π
with a real constant K
0
such that
K(p) ≤K
0
for all p ∈ M
2
,
the exponential map represents a local diffeomorphism of a closed disc B
r
(p) of
radius r >0 with center p ∈ M
2
into the surface M
2
⊂R
n+2
, i.e. it maps onetoone
onto the geodesic disc
B
r
(p) := exp
p
(B
r
(p)).
In particular, if it holds K ≤ 0 for the Gaussian curvature of M
2
⊂ R
n+2
then the
exponental map is injective for all r ∈ (0, +∞) by a theorem of Hadamard.
A Bernsteintype result
The following corollary generalizes a wellknown result of Liouville which states
that every entire and bounded holomorphic functions must be constant.
Corollary 9.2. Let X : R
2
→R
n+2
be a minimal graph of class C
2
(R
2
, R) satisfying
the growth condition
X
C
0
(B
R
)
≤Ω
∗
R
ε
, ε ∈ (0, 2), (9.8)
for all B
R
⊂R
2
with a real constant Ω
∗
∈ (0, +∞). Then it holds
lim
R→∞
[K
N
(0, 0)[ = 0 for all unit normal vectors N,
i.e. a complete minimal graph deﬁned on the whole plane R
2
represents a plane.
Proof. Since the minimal graph is deﬁned over the whole plane R
2
we may consider
discs B
R
with radius R > 0 arbitrarily large due to Hadamard’s theorem mentioned
above. The previous curvature estimate then implies
[K
N
(0, 0)[ ≤
ΘΩ
2
∗
R
4
R
2ε
−→0 for R →∞
for all unit normal vectors N proving the statement.
Remark 9.1.
1. This result is sharp in the following sense: The mapping (w, w
2
) for w ∈ C does
not satisfy the assumed growth condition, and obviously it is not a plane.
2. Since holomorphic mappings (x, y) →(ϕ(x, y), ψ(x, y)) generate minimal graphs
(x, y, ϕ, ψ) in R
4
we may considerer our corollary as a generalisation of Liou
ville’s theorem.
9.7 Osserman’s curvature estimate for minimal surfaces 173
We want to point out that it is not necessary to suppose the growth condition
(9.8) in case n = 1 of one codimension. Rather it holds the
Proposition 9.3. Ever minimal graph X ∈C
2
(R
2
, R
3
) represents a plane.
The nonlinearity of the minimal surface equation
(1 +ζ
2
y
)ζ
xx
−2ζ
x
ζ
y
ζ
xy
+(1 +ζ
2
x
)ζ
yy
= 0
in case n = 1 allows us to skip such growth or boundness conditions. But minimal
graphs in R
n+2
are solutions of systems of quasilinear elliptic equations forcing us
to additional assumptions.
For example, Kawai in [105] brings stability into play which will be our subject
in the next chapters.
Proposition 9.4. Given the complete and stable minimal graph X : R
2
→R
4
. Then
either it is a plane or it is a graph generated by a holomorphic or antiholomorphic
function.
9.7 Osserman’s curvature estimate for minimal surfaces
In this section we want to present Osserman’s curvature estimate for minimal sur
faces X : B →R
n+2
from [127].
Theorem 9.4. Let X : B →R
n+2
be a conformally parametrized minimal surface
representing a geodesic discs B
r
(X
0
) of radius r >0 and with center X
0
:= X(0, 0).
Assume that the unit normal vectors at each point of the immersion make an angle
ω > 0 with some ﬁxed direction in space. Then it holds the curvature estimate
[K(0, 0)[ ≤
1
r
2
32(n +1)
sin
4
ω
.
An estimate for the complex valued surface vector
Let the conformally parametrized minimal immersion X : B →R
n+2
be given. Set
ϕ
k
:= x
k
u
−ix
k
v
, k = 1, . . . , n +2,
and ϕ = (ϕ
1
, . . . , ϕ
n+2
) ∈ C
n+2
. In particular, it holds
[ϕ[
2
=
n+2
∑
k=1
[ϕ
k
[
2
=
n+2
∑
k=1
x
k
u
+
n+2
∑
k=1
x
k
v
= X
2
u
+X
2
v
= 2W
with the area element W > 0 of the surface.
174 9 Minimal surfaces
The following two results and their proofs are taken from Osserman [127],
Lemma 1.1 and Lemma 1.2. The ﬁrst lemma was formulated there for conformally
parametrized minimal surfaces, but it also holds true for all immersions.
Lemma 9.1. Let the conformally parametrized immersion X : B →R
n+2
be given.
Then every normal vector N(w) at w ∈ B makes an angle of at least ω > 0 with the
x
n
axis if and only if
[ϕ
n+2
(w)[
2
[ϕ(w)[
2
≥
1
2
sin
2
ω.
Proof. Let e
n+2
= (0, . . . , 0, 1) be the unit vector in the positive x
n+2
direction. The
condition that an arbitrary unit vector Z = (z
1
, . . . , z
n+2
) ∈ R
n+2
makes an angle of
at least ω > 0 with the x
n+2
axis is
[z
n+2
[ ≤cosω.
Assume that e
n+2
lies in the tangential plane at X(w). Then we set ω =
π
2
. Thus
suppose in the following that e
n+2
does not lie in the tangential plane at the point
X(w). Choose an orthonormal frame ¦V
1
,V
2
,V
3
¦ as a basis of the threedimensional
space Span¦X
u
, X
v
, e
n+2
¦ such that
V
3
:= e
n+2
.
Let furthermore Z ∈ span¦V
1
,V
2
,V
3
¦ be a unit vector with the property Z ⊥ X
1
, X
2
where we set
X
1
:=
X
u
√
W
, X
2
:=
X
v
√
W
, X
3
:= Z.
Now we deﬁne coefﬁcients v
ik
by
X
1
=
3
∑
k=1
v
1k
V
k
, X
2
=
3
∑
k=1
v
2k
V
k
, X
3
=
3
∑
k=1
v
3k
V
k
.
Then (v
ik
)
i,k=1,2,3
∈ R
33
is an orthogonal matrix with v
2
13
+v
2
23
+v
2
33
= 1 since
¦X
1
, X
2
, X
3
¦ and ¦V
1
,V
2
,V
3
¦ are orthonormal frames. But on the other hand
v
i3
= X
i
V
3
= X
i
e
n+2
= x
n+2
i
,
and therefore
[ϕ
n+2
[
2
= (x
n+2
u
)
2
+(x
n+2
v
)
2
=
_
(X
1
V
3
)
2
+(X
2
V
3
)
2
¸
W.
Thus we infer
[ϕ
n+2
[
2
[ϕ[
2
=
1
2
(v
2
13
+v
2
23
) =
1
2
(1 −v
2
33
).
But v
33
= z
3
= cosϑ with ϑ being the angle between the unit normal vector Z and
the e
n
axis.
9.7 Osserman’s curvature estimate for minimal surfaces 175
Now if all normals make an angle of at least ω then [v
33
[ ≤cosω, and it follows
[ϕ
n+2
[
2
[ϕ[
2
≥
1
2
(1 −cos
2
ω) =
1
2
sin
2
ω.
Conversely, if the inequality from the lemma holds true we ﬁnd [v
33
[ ≤ cosω. We
extend ¦X
1
, X
2
, X
3
¦ to an orthonormal basis ¦X
1
, . . . , X
n+2
¦ of R
n+2
such that
N =λ
3
X
3
+. . . +λ
n+2
X
n+2
for an arbitrary unit normal vector N of the surface with suitable coefﬁcients
λ
3
, . . . , λ
n+2
. We infer X
k
e
n+2
= 0 for all k = 4, . . . , n +2. Hence the angle ϑ be
tween N and e
n+2
satisﬁes
cosϑ = N e
n+2
=λ
3
X
3
e
n+2
=λ
3
cosω ≤cosω
since λ
3
= N X
3
≤1. The lemma is proved. ⊓⊔
Lemma 9.2. Let the conformally parametrized minimal surface X : B →R
n+2
rep
resent a geodesic disc of radius r > 0 with center X
0
= X(0, 0). Assume that every
normal vector makes an angle of at least ω > 0 with the x
n+2
axis. Then it holds
[ϕ
n+2
(0, 0)[ ≥r sinω.
Proof. Denote by Γ(B) the set of all continuously differentiable curves γ in B con
necting the origin (0, 0) ∈ B with the boundary ∂B.
1. Now consider a smooth embedded curve
c
∗
(t) =
_
u(t), v(t)
_
: I ⊂R −→B
from the set Γ(B) with c(0) = (0, 0). Its length on the surface is given by
L[c
∗
] =
_
c
∗
_
W(ζ) dζ
with the positive line element dζ =
√
˙ u
2
+ ˙ v
2
dt. Using the complexvalued func
tions ϕ and ϕ
n+2
from the previous lemma we infer
_
c
∗
_
W(ζ) dζ =
1
√
2
_
c
∗
[ϕ(ζ)[ dζ ≤
1
sinω
_
c
∗
[ϕ
n+2
(ζ)[ dζ.
2. We want to estimate [ϕ
n+2
(ζ)[. For this purpose consider the complexvalued
integral
F(w) :=
w
_
0
ϕ
n+2
(ζ)dζ, w ∈ B.
176 9 Minimal surfaces
Since ϕ
n+2
(ζ) ,= (0, 0) by the foregoing lemma, the complexvalued derivative
F
′
(w) does not vanish:
[F
′
(w)[ , = 0 in B, where F
′
(w) =∂
w
F(w) =
1
2
_
F
u
(w) −iF
v
(w)
¸
.
Thus σ = F(w) has an inverse w = G(σ) in some circle about σ = (0, 0). If
R > 0 denotes the radius of the largest of such circles, we ﬁnd a point σ
0
∈ C
with [σ
0
[ = R such that G(σ) can not be extended to a neighborhood of σ
0
.
Now let γ denote the line segment connecting σ = (0, 0) and σ
0
, and consider its
image c
∗
under the mapping G. In particular, c
∗
is a curve going from w = (0, 0)
to ∂B, and [σ[ <R is mapped into [w[ <1. Finally it holds G(0, 0) = (0, 0). Thus
Schwarz’ lemma
1
yields
[G
′
(0, 0)[ ≤
1
R
.
We infer
r ≤ inf
c∈Γ(B)
_
c
√
W dζ ≤
_
c
∗
√
W dζ ≤
1
sinω
_
c
∗
[ϕ
n+1
[ dζ ≤
1
sinω
_
γ
dσ
=
R
sinω
≤
1
[G
′
(0, 0)[ sinω
=
[F
′
(0, 0)[
sinω
=
[ϕ
n+2
(0, 0)[
sinω
from where the statement follows. ⊓⊔
A representation of the Gauss curvature
The following representation of the Gaussian curvature K of a conformally para
metrized minimal surface using the complexvalued functions ϕ
k
from above goes
also back to Osserman [127].
Lemma 9.3. Let the conformally parametrized minimal surface X : B →R
n+2
be
given. Then it holds
−2KW
2
=
1
2W
n+2
∑
k,ℓ=1
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
for its Gaussian curvature K.
Proof. From section 3.5 we recall the conformal representation
−2KW =
_
W
u
W
_
u
+
_
W
v
W
_
v
=
1
W
∆W −
1
W
2
(W
2
u
+W
2
v
).
1
Let the holomorphic function φ : B
R
→B with φ(0, 0) = (0, 0) be given. Then [φ
′
(0, 0)[ ≤
1
R
.
9.7 Osserman’s curvature estimate for minimal surfaces 177
Furthermore, the complexvalued vector ϕ = (ϕ
1
, . . . , ϕ
n+2
) with the holomorphic
functions ϕ
k
satisﬁes
[ϕ[
2
=
n+2
∑
k=1
[ϕ
k
[
2
= 2W.
We start with evaluating the Laplacian (let ϕ
′
=ϕ
w
and note ϕ
w
≡0)
1
4
∆W = W
ww
=
1
2
∂
ww
[ϕ[
2
=
1
2
∂
w
n+2
∑
k=1
_
ϕ
k
ϕ
k,w
+ϕ
k
ϕ
k,w
_
=
1
2
∂
w
n+2
∑
k=1
ϕ
k
ϕ
k,w
=
1
2
∂
w
n+2
∑
k=1
ϕ
k
ϕ
′
k
=
1
2
n+2
∑
k=1
ϕ
′
k
ϕ
′
k
.
Taking next
W
2
u
+W
2
v
= 4W
w
W
w
into account we compute
W
w
=
1
2
∂
w
[ϕ[
2
=
1
2
n+2
∑
k=1
_
ϕ
k,w
ϕ
k
+ϕ
k
ϕ
k,w
_
=
1
2
n+2
∑
k=1
ϕ
′
k
ϕ
k
,
and analogously
W
w
=
1
2
n+2
∑
k=1
ϕ
k
ϕ
′
k
.
It follows that
W
ww
−
1
W
W
w
W
w
=
1
2
n+2
∑
k=1
ϕ
′
k
ϕ
′
k
−
1
4W
n+2
∑
k,ℓ=1
ϕ
′
k
ϕ
k
ϕ
ℓ
ϕ
′
ℓ
=
1
2
n+2
∑
k=1
ϕ
′
k
ϕ
′
k
−
n+2
∑
k,ℓ=1
ϕ
′
k
ϕ
k
ϕ
ℓ
ϕ
′
ℓ
2
n+2
∑
ℓ=1
ϕ
ℓ
ϕ
ℓ
=
1
2
n+2
∑
k,ℓ=1
_
ϕ
′
k
ϕ
′
k
ϕ
ℓ
ϕ
ℓ
−ϕ
′
k
ϕ
k
ϕ
ℓ
ϕ
′
ℓ
_
n+2
∑
ℓ=1
ϕ
ℓ
ϕ
ℓ
.
On the other hand we calculate
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
= (ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
)(ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
)
= ϕ
k
ϕ
k
ϕ
′
ℓ
ϕ
′
ℓ
−ϕ
k
ϕ
′
ℓ
ϕ
′
k
ϕ
ℓ
−ϕ
k
ϕ
′
ℓ
ϕ
′
k
ϕ
ℓ
+ϕ
′
k
ϕ
ℓ
ϕ
′
k
ϕ
ℓ
.
178 9 Minimal surfaces
We thus infer
n+2
∑
k,ℓ=1
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
=
n+2
∑
k,ℓ=1
_
ϕ
k
ϕ
k
ϕ
′
ℓ
ϕ
′
ℓ
+ϕ
′
k
ϕ
ℓ
ϕ
′
k
ϕ
ℓ
_
−
n+2
∑
k,ℓ=1
_
ϕ
k
ϕ
′
ℓ
ϕ
′
k
ϕ
ℓ
+ϕ
k
ϕ
′
ℓ
ϕ
′
k
ϕ
ℓ
_
= 2
n+2
∑
k,ℓ=1
_
ϕ
′
k
ϕ
′
k
ϕ
ℓ
ϕ
ℓ
−ϕ
k
ϕ
′
k
ϕ
ℓ
ϕ
′
ℓ
_
.
Altogether we have derived
−2KW
2
= ∆W −
[∇W[
2
W
= 4W
ww
−
4
W
W
w
W
w
=
1
[ϕ[
2
n+2
∑
k,ℓ=1
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
=
1
2W
n+2
∑
k,ℓ=1
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
which proves the statement. ⊓⊔
Now introduce the meromorphic function
f
k
(w) :=
ϕ
k
(w)
ϕ
n+2
(w)
, k = 1, . . . , n +1.
For its derivatives we compute
f
′
k
=
ϕ
′
k
ϕ
n+2
−ϕ
k
ϕ
′
n+2
ϕ
2
n+2
, k = 1, . . . , n +1,
as well as
f
k
f
′
ℓ
− f
′
k
f
ℓ
=
(ϕ
′
ℓ
ϕ
n+2
−ϕ
ℓ
ϕ
′
n+2
)ϕ
k
−(ϕ
′
k
ϕ
n+2
−ϕ
k
ϕ
′
n+2
)ϕ
ℓ
ϕ
4
n+2
=
ϕ
′
ℓ
ϕ
k
−ϕ
′
k
ϕ
ℓ
ϕ
3
n+2
for k, ℓ = 1, . . . , n +1.
Lemma 9.4. The Gaussian curvature K of the conformally parametrized minimal
surface satisﬁes
K =−
2
[ϕ
n+2
[
2
n+1
∑
k,ℓ=1
[ f
k
f
′
ℓ
− f
′
k
f
ℓ
[
2
+2
n+1
∑
k=1
[ f
′
k
[
2
_
1 +
n+1
∑
k=1
[ f
k
[
2
_
3
.
9.7 Osserman’s curvature estimate for minimal surfaces 179
Proof. We compute
K = −
1
4W
3
n+2
∑
k,ℓ=1
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
= −
2
_
n+2
∑
k=1
[ϕ
k
[
2
_
3
n+2
∑
k,ℓ=1
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
= −2
n+1
∑
k,ℓ=1
[ϕ
k
ϕ
′
ℓ
−ϕ
′
k
ϕ
ℓ
[
2
+2
n+1
∑
k=1
[ϕ
k
ϕ
′
n+2
−ϕ
′
k
ϕ
n+2
[
2
_
[ϕ
n+2
[
2
+
n+1
∑
k=1
[ϕ
k
[
2
_
3
= −2[ϕ
n+2
[
4
n+1
∑
k,ℓ=1
[ f
k
f
′
ℓ
− f
′
k
f
ℓ
[
2
+2
n+1
∑
k=1
[ f
′
k
[
2
[ϕ
n+2
[
6
_
1 +
n+1
∑
k=1
[ϕ
k
[
2
[ϕ
n+2
[
2
_
3
= −
2
[ϕ
n+2
[
2
n+1
∑
k,ℓ=1
[ f
k
f
′
ℓ
− f
′
k
f
ℓ
[
2
+2
n+1
∑
k=1
[ f
′
k
[
2
_
1 +
n+1
∑
k=1
[ f
k
[
2
_
3
This is the stated formula. ⊓⊔
Proof of Osserman’s curvature estimate
Under the assumptions of Theorem 9.4 we infer from Lemma 9.1
[ϕ
n+2
(w)[
2
n+1
∑
k=1
[ f
k
(w)[
2
=
n+1
∑
k=1
[ϕ
k
(w)[
2
≤
n+2
∑
k=1
[ϕ
k
(w)[
2
=[ϕ(w)[
2
≤
2[ϕ
n+2
(w)[
2
sin
2
ω
showing the uniform estimate
n+1
∑
k=1
[ f
k
(w)[
2
≤
2
sin
2
ω
or
n+1
∑
k=1
 f
k

2
C
0
(B)
≤
2
sin
2
ω
.
Now Schwarz’ lemma implies
[ f
′
k
(0, 0)[ ≤f
k

C
0
(B)
−
[ f
k
(0, 0)[
2
 f
k

C
0
(B)
=: η
k
 f
k

C
0
(B)
180 9 Minimal surfaces
with the abbreviation
η
k
:= 1 −
[ f
k
(0, 0)[
2
 f
k

2
C
0
(B)
.
We estimate the Gauss curvature at the point w = (0, 0). For this purpose we use the
CauchySchwarz inequality to get
[ f
k
(0, 0) f
′
ℓ
(0, 0) − f
′
k
(0, 0) f
ℓ
(0, 0)[
2
≤
_
[ f
k
(0, 0)[[ f
′
ℓ
(0, 0)[ +[ f
′
k
(0, 0)[[ f
ℓ
(0, 0)[
_
2
≤ 2
_
[ f
k
(0, 0)[
2
[[ f
′
ℓ
(0, 0)[
2
+[ f
′
k
(0, 0)[
2
[ f
ℓ
(0, 0)[
2
_
which in turn gives
n+1
∑
k,ℓ=1
[ f
k
(0, 0) f
′
ℓ
(0, 0) − f
′
k
(0, 0) f
ℓ
(0, 0)[
2
≤ 4
n+1
∑
k,ℓ=1
[ f
k
(0, 0)[
2
[ f
ℓ
(0, 0)[
2
≤ 4
n+1
∑
k=1
[ f
k
(0, 0)[
2
n+1
∑
ℓ=1
[ f
′
ℓ
(0, 0)[
2
.
Thus we infer
n+1
∑
k,ℓ=1
[ f
k
(0, 0) f
′
ℓ
(0, 0) − f
′
k
(0, 0) f
ℓ
(0, 0)[
2
+
n+1
∑
ℓ=1
[ f
′
ℓ
(0, 0)[
2
≤ 4
n+1
∑
k=1
[ f
k
(0, 0)[
2
n+1
∑
ℓ=1
[ f
′
ℓ
(0, 0)[
2
+
n+1
∑
ℓ=1
[ f
′
ℓ
(0, 0)[
2
=
_
4
n+1
∑
k=1
[ f
k
(0, 0)[
2
+1
_
n+1
∑
ℓ=1
[ f
′
ℓ
(0, 0)[
2
≤
_
4
n+1
∑
k=1
[ f
k
(0, 0)[
2
+1
_
n+1
∑
ℓ=1
η
2
ℓ
 f
ℓ

2
C
0
(B)
≤4
_
n+1
∑
k=1
[ f
k
(0, 0)[
2
+1
_
n+1
∑
ℓ=1
 f
ℓ

2
C
0
(B)
n+1
∑
ℓ=1
η
2
ℓ
≤
8
sin
2
ω
n+1
∑
ℓ=1
η
2
ℓ
_
n+1
∑
ℓ=1
[ f
ℓ
(0, 0)[
2
+1
_
.
9.8 A theorem of Bernstein type due to Osserman 181
Inserting into the above repesentation for the Gaussian curvature yields
[K(0, 0)[ ≤ −
4
[ϕ
n+2
[
2
n+1
∑
k,ℓ=1
[ f
k
f
′
ℓ
− f
′
k
f
ℓ
[
2
+
n+1
∑
k=1
[ f
k
[
2
_
1 +
n+2
∑
k=1
[ f
k
[
2
_
3
¸
¸
¸
w=(0,0)
≤
32
sin
2
ω
n+1
∑
k=1
η
2
k
[ϕ
n+2
[
2
_
1 +
n+1
∑
k=1
[ f
k
[
2
_
2
¸
¸
¸
w=(0,0)
≤
32(n +1)
sin
2
ω[ϕ
n+2
(0, 0)[
2
since η
k
∈ [0, 1] for all k = 1, . . . , n +1. But from Lemma 9.2 we know
[ϕ
n+2
(0, 0)[
2
≥r
2
sin
2
ω
with the geodesic radius r > 0 of the minimal disc. The ﬁnally proves the curvature
estimate. ⊓⊔
9.8 A theorem of Bernstein type due to Osserman
From Osserman’s curvature estimate we want to derive our next theorem of Bern
stein type for twodimensional minimal surfaces in R
n+2
.
Corollary 9.3. Let the minimal graph X : R
2
→R
n+2
be given. Assume that its unit
normal vectors at each point make an angle ω > 0 with some ﬁxed direction in
space. Then X represents a plane.
Proof. Due to Hadamard’s theorem (see section 9.6) we may introduce geodesic
minimal discs B
r
(X
0
) for all r > 0 with X
0
:= X(0, 0). The curvature estimate
[K(0, 0)[ ≤
1
r
2
32(n +1)
sin
4
ω
implies
lim
r→∞
K(0, 0) = 0.
But X
0
= X(0, 0) is chosen arbitrarily. The statement follows. ⊓⊔
182 9 Minimal surfaces
9.9 Gradient estimates for minimal graphs
We conclude this chapter with elementary gradients estimates for minimal graphs.
Such apriori bounds are sufﬁcient for solving the Dirichlet problem for this class
of quasilinear equations. For we do not elaborate on this existence problem we want
to refer the reader e.g. to Taylor [155] for detailed studies.
Gradient estimates on the boundary
Let us restrict to minimal graphs deﬁned on the closed disc
B
R
:=
_
(x, y) ∈ R
2
: x
2
+y
2
≤R
2
_
.
We want to start with establishing gradient estimates on the boundary.
Theorem 9.5. Let ζ
σ
∈C
2
(B
R
, R), σ =1, . . . , n, be a solution of the minimal surface
system
_
1 +
n
∑
ω=1
ζ
2
ω,y
_
ζ
σ,xx
+2
n
∑
ω=1
ζ
ω,x
ζ
ω,y
ζ
σ,xy
+
_
1 +
n
∑
ω=1
ζ
2
ω,x
_
ζ
σ,yy
= 0 in B
R
for σ = 1, . . . , n. Then there exists a real constant
C =C(ζ
1

C
2
(∂B
R
)
, . . . , ζ
n

C
2
(∂B
R
)
, R)
such that
max
_
[∇ζ
1
(z)[, . . . , [∇ζ
n
(z)[
_
≤C(ζ
1

C
2
(∂B
R
)
, . . . , ζ
n

C
2
(∂B
R
)
, R) (9.9)
for all z ∈ ∂B
R
.
Proof. Choose a point z
0
= (x
0
, y
0
) ∈ ∂B
R
. For real parameters c
1
, c
2
∈ (0, +∞)
consider the auxiliary functions
S
−
ζ
σ
(z) := c
1
log
_
1 +c
2
d
Γ
(z)
_
−
_
ζ
σ
(z) −ζ
σ
(z
0
)
_
, z = (x, y) ∈ B
R
,
fo σ =1, . . . , n where d
Γ
=d
Γ
(z) means the distance of z ∈ B
R
to the tangential line
Γ ⊂R
2
on ∂B
R
at z
0
∈ ∂B
R
. We have
S
−
ζ
σ
(z
0
) = 0 for all σ = 1, . . . , n. (9.10)
Now we consider ζ :=ζ
1
. Let C
ζ
∈ [0, +∞) be deﬁned as
C
ζ
:=ζ
C
2
(∂B
R
)
. (9.11)
9.9 Gradient estimates for minimal graphs 183
Then we may choose c
1
= c
1
(C
ζ
, R) and c
2
= c
2
(C
ζ
, R) such that
S
−
ζ
> 0 for all z ∈ ∂B
R
¸ ¦z
0
¦. (9.12)
Let us now abbreviate
M[ζ
ω
] := a
11
ζ
ω,xx
+2a
12
ζ
ω,xy
+a
22
ζ
ω,yy
, ω = 1, . . . , n,
for the minimal surface operator with coefﬁcients
a
11
:= 1 +
n
∑
σ=1
ζ
2
σ,y
, a
12
:=
n
∑
σ=1
ζ
σ,x
ζ
σ,y
, a
22
:= 1 +
n
∑
σ=1
ζ
2
σ,x
.
There hold
M[ζ
σ
] = 0 in B
R
for σ = 1, . . . , n. We compute the partial derivatives
S
−
ζ,x
=
c
1
c
2
1 +c
2
d
Γ
d
Γ,x
−ζ
x
, S
−
ζ,y
=
c
1
c
2
1 +c
2
d
Γ
d
Γ,y
−ζ
y
as well as
S
−
ζ,xx
= −
c
1
c
2
2
(1 +c
2
d
Γ
)
2
d
2
Γ,x
−ζ
xx
,
S
−
ζ,xy
= −
c
1
c
2
2
(1 +c
2
d
Γ
)
2
d
Γ,x
d
Γ,y
−ζ
xy
,
S
−
ζ,yy
= −
c
1
c
2
2
(1 +c
2
d
Γ
)
2
d
2
Γ,y
−ζ
yy
.
Thus it follows
M[S
−
ζ
] = a
11
S
−
ζ,xx
+2a
12
S
−
ζ,xy
+a
22
S
−
ζ,yy
= −
c
1
c
2
2
(1 +c
2
d
Γ
)
2
(a
11
d
2
Γ,x
+2a
12
d
Γ,x
d
Γ,y
+a
22
d
2
Γ,y
) −M[ζ]
= −
c
1
c
2
2
(1 +c
2
d
Γ
)
2
(a
11
d
2
Γ,x
+2a
12
d
Γ,x
d
Γ,y
+a
22
d
2
Γ,y
).
Taking the ellipticity condition a
11
a
22
−a
2
12
> 0 together with a
11
> 0 into account
we conclude
M[S
−
ζ
] ≤0 in B
R
.
Along with (9.10) and (9.12), the maximum principle yields S
−
ζ
(z) ≥0 in B
R
, i.e.
ζ(z) −ζ(z
0
) ≤c
1
log
_
1 +c
2
d
Γ
(z)
_
for all z ∈ B
R
. (9.13)
184 9 Minimal surfaces
Analogously we proceed with the function
S
+
ζ
(z) := c
1
log
_
1 +c
2
d
Γ
(z)
_
+
_
ζ(z) −ζ(z
0
)
_
and arrive at
M[S
+
ζ
] ≤0 in B
R
, S
+
ζ
(Z) ≥0 for all z ∈ ∂B
R
,
i.e. S
+
ζ
≥0 in B
R
resp.
−c
1
log
_
1 +c
2
d
Γ
(z)
_
≤ζ(z) −ζ(z
0
) for all z ∈ B
R
. (9.14)
Taking the estimates (9.13) and (9.14) together gives
[ζ(z) −ζ(z
0
)[ ≤c
1
log
_
1 +c
2
d
Γ
(z)
_
for all z ∈ B
R
.
Finally, for the outer unit normal vector ν at z
0
∈ ∂B we choose a point z ∈ B
R
such
that
¸
¸
¸
¸
∂ζ
∂ν
(z
0
)
¸
¸
¸
¸
≤ lim
d
Γ
(z)→0
[ζ(z) −ζ(z
0
)[
d
Γ
(z)
≤ lim
d
Γ
(z)→0
c
1
log
_
1 +c
2
d
Γ
(z)
_
d
Γ
(z)
= c
1
c
2
.
Together with (9.11), the choice of c
1
and c
2
, and ﬁnally a similar calculation for
ζ
2
, ζ
3
, . . . , ζ
n
we get the statement. ⊓⊔
Barrier constructions for minimal immersions and surfaces with prescribed mean
curvature as presented here can be found e.g. in Dierkes [43], Jorge and Tomi [100],
or Bergner [11].
Global gradient estimates
Finally we discuss a global gradient estimate for minimal graphs on strictly convex
domains of deﬁnition. A suitable threepointcondition which already goes back to
Schauder [144] ensures that the minimal surface system decouples.
Theorem 9.6. Let ζ
σ
∈ C
2
(
˚
B
R
, R) ∩C
1
(B
R
, R), σ = 1, . . . , n, be a minimal graph
together with Dirichlet boundary conditions ζ
R,σ
∈C
1
(∂B
R
, R). Let
C
∗
ζ
:= max
σ=1,...,n
ζ
R,σ

C
1
(∂B
R
)
.
Then there exists a real constant C =C(C
∗
ζ
, R) ∈ (0, +∞) such that it holds
[ζ
σ
]
C
1
(
˚
B
R
)
≤C(ζ
σ
, R).
9.9 Gradient estimates for minimal graphs 185
Sketch of the proof. Insert (ζ
1
, . . . , ζ
n
) and its gradient into the minimal surface sys
tem to get the decoupled system
a
1
(x, y)ζ
1,xx
+2b
1
(x, y)ζ
1,xy
+c
1
(x, y)ζ
1,yy
= 0,
a
2
(x, y)ζ
2,xx
+2b
2
(x, y)ζ
2,xy
+c
2
(x, y)ζ
2,yy
= 0,
.
.
.
a
n
(x, y)ζ
n,xx
+2b
n
(x, y)ζ
n,xy
+c
n
(x, y)ζ
n,yy
= 0.
Here there hold a
σ
c
σ
−b
2
σ
> 0 for all σ = 1, . . . , n. The solutions ζ
σ
of this linear
system can be considered a saddle surfaces (x, y, ζ
σ
(x, y)) over B
R
⊂ R
2
. For such
surfaces we ﬁnd the following result: Let Γ
σ
⊂ R
3
be the spatial curves generated
by (x, y, ζ
R,σ
) ∈ R
3
. Each of these curves satisﬁes a threepointcondition with a
constant Λ ∈ R in the sense that any plane z = αx +βy +γ passing three different
points on Γ
σ
permits an uniform estimate of the form
_
α
2
+β
2
≤Λ.
Then by a theorem of Rad´ o there is a constant c
1
= c
1
(Λ) ∈ (0, +∞) with
[ζ
σ
]
C
1
(
˚
B
R
)
≤c
1
(Λ) for Λ = 1, . . . , n
for the halfnorms [ζ
σ
]
C
1
(
˚
B
R
)
, see e.g. Gilbarg and Trudinger [71]. On the other hand,
Schauder [144] ensures that Γ
σ
⊂ R
3
actually satisﬁes a threepointcondition over
the strictly convex domain B
R
⊂R
2
with a constant Λ =Λ(C
∗
ζ
, R) which leads us to
[ζ
σ
]
C
1
(
˚
B
R
)
≤c
2
(C
∗
ζ
, R) for all σ = 1, . . . , n.
This proves the statement. ⊓⊔
Chapter 10
Immersions with prescribed mean curvature
10.1 Critical points of parametric Gullivertype functionals
10.2 A geometric maximum principle
10.3 A curvature estimate for graphs with prescribed mean curvature
188 10 Immersions with prescribed mean curvature
In this chapter we consider critical points of functionals of Gulliver type in parametric and non
parametric form. Such surfaces represent special immersions with prescribed mean curvature
ﬁelds. We discuss various analytical and geometric properties, and we present a curvature esti
mate for graphs.
10.1 Critical points of parametric Gullivertype functionals
Preparations
Following Gulliver [75] and B¨ ohme, Hildebrandt and Tausch [16] we consider the
following functional of meancurvaturetype
G[X] :=
__
B
_
Γ(X)W +2X
u
◦ A(X) ◦ X
v
_
dudv (10.1)
which generalizes the usual area functional
A[X] =
__
B
W dudv,
and where
Γ ∈C
3+α
(R
3
, (0, ∞)), α ∈ (0, 1)
is a positive weight function and
A ∈C
3+α
(R
n
, R
nn
), α ∈ (0, 1)
a realvalued and skewsymmetric matrix.
Before we come to the computation of the EulerLagrange equations of critical
points for G[X] we want to convince ourself of its geometric character.
Proposition 10.1. The Gullivertype functional G[X] is invariant w.r.t. parameter
transformations of class P.
Proof. It is sufﬁcient to prove the invariance for the term X
u
◦ A(X) ◦ X
v
. Thus let
u
i
(v
α
) ∈ P be such a parameter transformation.
190 10 Immersions with prescribed mean curvature
Then we compute
X
v
1 ◦ A(X) ◦ X
v
2 =
_
X
u
1 u
1
v
1
+X
u
2u
2
v
1
¸
◦ A(X) ◦
_
X
u
1u
1
v
2
+X
u
2u
2
v
2
¸
= (u
1
v
1
u
2
v
2
−u
1
v
2
u
2
v
1
)X
u
1 ◦ A(X) ◦ X
u
2 .
The transformation formula for multiple integrals yields the claim. ⊓⊔
In particular, the form e
1
◦ A(X) ◦ e
2
does not depend on the choice of the or
thonormal frame (e
1
, e
2
) as the following lemma emphasizes.
Lemma 10.1. Let a point w∈B be given as well as two orthonormal frames ¦e
1
, e
2
¦
and (¯ e
1
, ¯ e
2
) spanning the tangential plane T
X
(w). Then it holds
e
1
◦ A(X) ◦ e
2
= ¯ e
1
◦ A(X) ◦ ¯ e
2
.
Proof. Consider the SO
2
action
¯ e
1
= cosϕ e
1
−sinϕ e
2
, ¯ e
2
= sinϕ e
1
+cosϕ e
2
with a rotation angle ϕ. Then we infer
¯ e
1
◦ A(X) ◦ ¯ e
2
= cos
2
ϕ e
1
◦ A(X) ◦ e
2
+sin
2
ϕ e
1
◦ A(X) ◦ e
2
due to the skewsymmetry of A(X), in particular e
k
◦A(X) ◦e
k
= 0 for k = 1, 2. ⊓⊔
Example: Hildebrandt’s functional
In case n = 1 of one codimension, Hildebrandt in [85] considered the functional
__
B
_
[X
u
X
v
[ +2Q(X) (X
u
X
v
)
_
dudv
with a smooth vector ﬁeld Q: R
3
→ R
3
. As it will turn out later critical points
X : B →R
3
have the scalar mean curvature
H(X) = 2div
X
Q(X) := 2
_
∂q
1
(X)
∂x
1
+
∂q
2
(X)
∂x
2
+
∂q
3
(X)
∂x
3
_
with the spatial divergence operator div
X
and satisfy
∆X = 2H(X)WN in B
using conformal parameters (u, v) ∈ B.
10.1 Critical points of parametric Gullivertype functionals 191
In terms of our skewsymmetric matrix A(X) we immediately verify
A(X) =
_
_
_
0 q
3
(X) −q
2
(X)
−q
3
(X) 0 q
1
(X)
q
2
(X) −q
1
(x) 0
_
_
_.
EulerLagrange equations
We come to the calculation of the ﬁrst variation of G[X].
Theorem 10.1. Let a conformally parametrized immersion X : B →R
n+2
together
with an ONF N be given. Assume that it is critical for the functional G[X]. Then it
satisﬁes the EulerLagrange system
△X = 2
n
∑
σ=1
H
σ
WN
σ
in B (10.2)
with the mean curvature ﬁeld
H
σ
=
1
Γ(X)W
n+2
∑
k,ℓ,m=1
a
kℓ,x
m(x
k
u
x
ℓ
v
n
m
σ
+x
m
u
x
k
v
n
ℓ
σ
+x
ℓ
u
x
m
v
n
k
σ
) +
Γ
X
(X) N
σ
2Γ(X)
(10.3)
for σ = 1, . . . , n and with the skewsymmetric matrix A(X) = (a
kℓ
(X))
k,ℓ=1,...,n
.
Proof. In view of our results from the previous chapter it remains to evaluate the
variations of Γ(X) and the form X
u
◦ A(X) ◦ X
v
. For this purpose we again consider
normal variations of the form
¯
X(u, v) = X(u, v) +εϕ(u, v)N
´ γ
(u, v)
where
N
¯ γ
=
n
∑
σ=1
´ γ(u, v)N
σ
(u, v),
n
∑
σ=1
(´ γ
σ
)
2
= 1
with arbitrary ϕ ∈C
∞
0
(B, R) and ε ∈ (−ε
0
, +ε
0
). First it holds
δΓ(X) :=
∂
∂ε
Γ(
¯
X)
¸
¸
¸
ε=0
=Γ
X
(X) N
´ γ
ϕ
such that, together with (9.2), we arrive at
δ [Γ(X)W] =−2Γ(X)H
´ γ
Wϕ +Γ
X
(X) N
´ γ
Wϕ. (10.4)
It remains to compute the ﬁrst variation of X
u
◦ A(X) ◦ X
v
.
192 10 Immersions with prescribed mean curvature
We calculate (note that X
u
◦ A(X) ◦ X
u
= X
v
◦ A(X) ◦ X
v
= 0)
δ [X
u
◦ A(X) ◦ X
v
] = δX
u
◦ A(X) ◦ X
v
+X
u
◦ A(X) ◦ δX
v
+X
u
◦ δA(X) ◦ X
v
=
_
ϕ
u
N
´ γ
+ϕN
´ γ,u
_
◦ A(X) ◦ X
v
+X
u
◦ A(X) ◦
_
ϕ
v
N
´ γ
+ϕN
´ γ,v
_
+X
u
◦ δA(X) ◦ X
v
= div
_
N
´ γ
◦ A(X) ◦ X
v
ϕ, X
u
◦ A(X) ◦ N
´ γ
ϕ
_
−N
´ γ
◦ A(X)
u
◦ X
v
ϕ −X
u
◦ A(X)
v
◦ N
´ γ
ϕ +X
u
◦ δA(X) ◦ X
v
.
Setting N
´ γ
= (n
1
´ γ
, . . . , n
n+2
´ γ
) we have
n+2
∑
k,ℓ,m=1
X
u
◦ δA(X) ◦ X
v
= a
kℓ,x
mn
m
´ γ
x
ℓ
v
x
k
u
ϕ ,
n+2
∑
k,ℓ,m=1
N
´ γ
◦ A(X)
u
◦ X
v
= a
kℓ,x
mx
m
u
x
ℓ
v
n
k
´ γ
,
n+2
∑
k,ℓ,m=1
X
u
◦ A(X)
v
◦ N
´ γ
= a
kℓ,x
mx
m
v
n
ℓ
´ γ
x
k
u
and we arrive at
δ [X
u
◦ A(X) ◦ X
v
] = div
_
N
´ γ
◦ A(X) ◦ X
v
ϕ, X
u
◦ A(X) ◦ N
´ γ
ϕ
_
+
n+2
∑
k,ℓ,m=1
a
kℓ,x
m(x
k
u
x
ℓ
v
n
m
´ γ
−x
m
u
x
ℓ
v
n
k
´ γ
−x
k
u
x
m
v
n
ℓ
´ γ
)ϕ
= div
_
N
´ γ
◦ A(X) ◦ X
v
ϕ, X
u
◦ A(X) ◦ N
´ γ
ϕ
_
+
n+2
∑
k,ℓ,m=1
a
kℓ,x
m(x
k
u
x
ℓ
v
n
m
´ γ
+x
m
u
x
k
v
n
ℓ
´ γ
+x
ℓ
u
x
m
v
n
k
´ γ
)ϕ.
Together with (10.4) we conclude
δ G[X] =
__
B
div
_
N
´ γ
◦ A(X) ◦ X
v
ϕ, X
u
◦ A(X) ◦ N
´ γ
ϕ
_
dudv
−2
__
B
Γ(X)H
´ γ
Wϕ dudv +
__
B
Γ
X
(X) N
´ γ
Wϕ dudv
+2
n+2
∑
k,ℓ,m=1
__
B
a
kℓ,x
m(x
k
u
x
ℓ
v
n
m
´ γ
+x
m
u
x
k
v
n
ℓ
´ γ
+x
ℓ
u
x
m
v
n
k
´ γ
)ϕ dudv
for all ϕ ∈C
∞
0
(B, R).
10.1 Critical points of parametric Gullivertype functionals 193
Vanishing of the ﬁrst variations now means
H
´ γ
=
1
Γ(X)W
n+2
∑
k,ℓ,m=1
a
kℓ,x
m(x
k
u
x
ℓ
v
n
m
´ γ
+x
m
u
x
k
v
n
ℓ
´ γ
+x
ℓ
u
x
m
v
n
k
´ γ
) +
Γ
X
(X) N
´ γ
2Γ(X)
.
Let ´ γ
σ
≡1 for some index σ ∈ ¦1, . . . , n¦ and ´ γ
ω
≡ 0 for all ω ,= σ. Then we have
N
´ γ
= N
σ
, n
ℓ
´ γ
= n
ℓ
σ
and H
´ γ
= H
σ
. We therefore conclude
H
σ
=
1
Γ(X)W
n+2
∑
k,ℓ,m=1
a
kℓ,x
m(x
k
u
x
ℓ
v
n
m
σ
+x
m
u
x
k
v
n
ℓ
σ
+x
ℓ
u
x
m
v
n
k
σ
) +
Γ
X
(X) N
σ
2Γ(X)
.
This holds for all σ = 1, . . . , n. ⊓⊔
Invariance of the EulerLagrange equations
As in Lemma 10.1 we immediately verify that this special meancurvature ﬁeld
H = (H
1
, . . . , H
n
) from the previous proof is invariant w.r.t. dilatations of the tangent
vectors X
u
and X
v
as well as w.r.t. rotations of the basis ¦X
u
, X
v
¦, i.e. there hold
H(X, λe
1
, µe
2
, N
σ
) = H(X, e
1
, e
2
, N
σ
) for real λ, µ ∈ (0, +∞)
as well as
H(X, ¯ e
1
, ¯ e
2
, N
σ
) = H(X, e
1
, e
2
, N
σ
).
Secondly, the reader is invited to verify the mean curvature H for critical points
X of Hildebrandt’s functional
__
B
_
[X
u
X
v
[ +2Q(X) (X
u
X
v
)
_
dudv
from above in case n = 1, i.e. to prove
H(X) =
∂
∂z
q
3
(X) +
∂
∂y
q
2
(X) +
∂
∂x
q
1
(X).
And ﬁnally we want to show explicitely the invariance of the mean curvature
system (10.2), (10.3) w.r.t. rotations of the ONF N. For this we write (10.3) as an
inner product in the form
H
σ
=
1
Γ(X)
_
1
W
n+2
∑
k,ℓ,m=1
(a
kℓ,x
m +a
ℓm,x
k +a
mk,ℓ
)x
k
u
x
ℓ
v
+
Γ
m
X
(X)
2
_
n
m
σ
=: Γ
∗
(X, X
u
, X
v
) N
σ
setting Γ
X
= (Γ
1
X
, . . . , Γ
n+2
X
).
194 10 Immersions with prescribed mean curvature
Applying then an orthogonal mapping R we arrive at
∆X = 2
n
∑
σ=1
H
σ
WN
σ
= 2W
n
∑
σ=1
_
Γ
∗
(X, X
u
, X
v
) N
σ
_
N
σ
= 2W
n
∑
σ=1
_
Γ
∗
(X, X
u
, X
v
)
_
n
∑
ω=1
r
σω
N
ω
__
n
∑
ω
′
=1
r
σω
′ N
ω
′
= 2W
n
∑
ω=1
_
n
∑
σ=1
r
2
σω
_
_
Γ
∗
(X, X
u
, X
v
) N
ω
_
N
ω
+2W
n
∑
ω,ω
′
=1
ω,=ω
′
_
n
∑
σ=1
r
σω
r
σω
′
_
_
Γ
∗
(X, X
u
, X
v
) N
ω
_
N
ω
′
= 2W
n
∑
ω=1
_
Γ
∗
(X
u
, X
u
, X
v
) N
ω
_
N
ω
=
n
∑
ω=1
H
ω
WN
ω
.
Proposition 10.2. The mean curvature system (10.2) together with the mean curva
ture ﬁeld (10.3) is invariant w.r.t. the choice of an ONF N.
Surfaces in R
3
In particular, critical points X : B →R
3
of the functional
__
B
_
Γ(X)[X
u
X
v
[ +2Q(X) (X
u
X
v
)
_
dudv
have the scalar mean curvature
H(X) =
1
Γ(X)
_
Γ
X
(X) N+2div
X
Q(X)
_
with the unit normal vector N. By the way, the special functional
__
B
Γ(X)W dudv
possesses an interesting geometric application.
10.2 A geometric maximum principle 195
Namely it can be considered as the area of a surface X : B →R
3
immersed in the
threedimensional space R
3
which is equipped with the nonEuclidean metric
d¯ s
2
=Γ(x
1
, x
2
, x
3
)
_
(dx
1
)
2
+(dx
2
)
2
+(dx
3
)
2
_
.
For example, the stereographic projection of the sphere S
2
⊂R
3
would yield the
special weight function
Γ(X) =
4
_
1 +[X[
2
_
2
,
see e.g. Bergner [11] or B¨ ohme, Hildebrandt and Tausch [16].
10.2 A geometric maximum principle
Following of Dierkes [42], [43] and Hildebrandt [84] we want to establish a ge
ometric maximum principle for immersions with prescribed mean curvature ﬁelds
generalizing our results for minimal surfaces from section 9.3.
Theorem 10.2. Let the conformally parametrized immersion X : B → R
n+2
with
prescribed mean curvature H = (H
1
, . . . , H
n
) w.r.t. to an ONF N be given. Assume
h
0
sup
(u,v)∈B
[X(u, v)[ ≤
1
n
is satisﬁed with
sup
(u,v)∈B
[H
σ
(u, v)[ ≤h
0
for all σ = 1, . . . , n.
Then it holds
△[X[
2
≥0 in B.
In particular, the function [X[
2
is subject to the maximum principle
sup
(u,v)∈B
[X(u, v)[
2
= sup
(u,v)∈∂B
[X(u, v)[
2
.
Proof. Namely, we compute
△[X[
2
= 2[∇X[
2
+2X △X = 2[∇X[
2
+4
n
∑
σ=1
H
σ
WX N
σ
≥ 2
_
1 −
n
∑
σ=1
[H
σ
[[X[
_
[∇X[
2
≥ 2
_
1 −nh
0
[X[
_
[∇X[
2
,
and the statement follows. ⊓⊔
196 10 Immersions with prescribed mean curvature
Thus in case n = 1 of one codimension we have to assure
h
0
sup
(u,v)∈B
[X(u, v)[ ≤1.
The limit case is represented by the half sphere with scalar mean curvature H ≡1.
10.3 A curvature estimate for surfaces with prescribed mean
curvature
We want to present a curvature estimate from Bergner and Fr¨ ohlich [13] for twodi
mensional graphs with prescribed mean curvature ﬁelds of the special form
H ∈C
0
(R
n
S
n+1
, R)
which additionally satisfy the following H¨ older and Lipschitz conditions:
[H(X, Z)[ ≤h
0
for all X ∈ R
n
, Z ∈ S
n+1
as well as
[H(X
1
, Z
1
) −H(X
2
, Z
2
)[ ≤h
1
[X
1
−X
2
[
α
+h
2
[Z
1
−Z
2
[
for all X
1
, X
2
∈ R
n
and Z
1
, Z
2
∈ S
n+1
.
(10.5)
Our proof essentially uses method from Osserman [127] already applied in sec
tion 9.7 above, and from Sauvigny [140], [141]; see also Fr¨ ohlich [62].
The curvature estimate
Theorem 10.3. Let the immersion X : B →R
n+2
with prescribed mean curvature
ﬁeld H =H(X, Z) satisfying the properties (10.5), together with an ONF N be given.
Assume furthermore
(A1) X represents a conformal reparametrization of a surface graph
(x, y, ζ
1
(x, y), . . . , ζ
n
(x, y)), ζ
σ
∈C
3+α
(Ω, R) for σ = 1, . . . , n
on a bounded, simply connected and regular domain Ω ⊂R
2
;
(A2) X represents a geodesic disc B
r
(X
0
) of geodesic radius r > 0 with center
X
0
:= X(0, . . . , 0), and with a real constant d
0
> 0, independent of the radius r,
it holds the growth estimate
A[X] ≤d
0
r
2
(10.6)
for the area of X;
10.3 A curvature estimate for surfaces with prescribed mean curvature 197
(A3) at each point w ∈ B every unit normal vector N of the surface makes an angle
of at least ω > 0 with the x
1
axis.
Then there exists a real constant
Θ =Θ(h
0
r, h
1
r
1+α
, h
2
r, d
0
, sinω, n, α) ∈ (0, +∞),
with the constant h
0
∈ [0, ∞) from the previous section, such that it holds the curva
ture estimate
κ
σ,1
(0, 0)
2
+κ
σ,2
(0, 0)
2
≤
1
r
2
_
(h
0
r)
2
+Θ
_
(10.7)
for the principal curvature κ
σ,1
und κ
σ,2
w.r.t. the ONF N and for all σ = 1, . . . , n.
Remarks
Before we come to the proof of our curvature estimate we brieﬂy want to discuss
the assumptions of the theorem.
1. The regularity conditions (10.5) can particularly be veriﬁed by means of the mean
curvature ﬁeld
H(X, Z) =
Γ
X
(X) Z
2Γ(X)
for critical points of the Gullivertype functional
G[X] =
__
B
Γ(X)W dudv,
at least if Γ
X
(X) and Γ(X) are regular enough.
2. Growth estimates for the area as assumed in (10.6) can be found in chapter 14.
For the purpose of a curvature estimate the quality of the constant d
0
is inessen
tial; in particular it could also depend on the radius r. But later when we apply
our curvature estimate to establish a theorem of Bernstein type we are forced to
require the growth estimate (10.6) in its sharp form.
3. The condition (A3) comes from Osserman’s curvature estimate presented in sec
tion 9.7. Let us verify it by means of the minimal graph (w, w
2
) for w ∈ B
R
. For
this purpose we introduce the two unit normal vectors
N
1
:=
1
√
1 +4u
2
+4v
2
(−2u, 2v, 1, 0),
N
2
:=
1
√
1 +4u
2
+4v
2
(−2v, −2u, 0, 1).
198 10 Immersions with prescribed mean curvature
Along the line ¦v = 0¦ we especially infer
[N
1
(1, 0, . . . , 0)[ =
2[u[
√
1 +4u
2
−→1 for [u[ →∞.
Thus (A3) is not fulﬁlled for R →∞ since according to this condition it should
hold contradictorily
[N
1
(1, 0, . . . , 0)[ ≤cosω < 1.
Unfortunately we can not apply our theorem.
Proof of the curvature estimate
First it holds
κ
σ,1
(0, 0)
2
+κ
σ,2
(0, 0)
2
= 4H
σ
(0, 0)
2
−2K
σ
(0, 0) ≤ 4h
2
0
+2[K
σ
(0, 0)[
=
1
r
2
_
(2h
0
r)
2
+r
2
[K
σ
(0, 0)[
_
for σ = 1, . . . , n. Then the desired estimate follows from the identity
K
σ
(0, 0) =
(X
uu
N
σ
)(X
vv
N
σ
) −(X
uv
N
σ
)
2
W
2
¸
¸
¸
(u,v)=(0,0)
if an upper and a lower bounds for the nominator resp. the denominator of the frac
tion on the right hand side are proved.
1. In the ﬁrst part of our proof we show
W(w)
r
2
≥C
1
for w ∈ B1
2
(0, 0)
with a constant C
1
=C
1
(h
0
r, d
0
, sinω, n) > 0.
1.1 As usual we construct a global orthonormal normal frame N as follows:
The basis vectors
e
3
:= (0, 0, 1, 0, . . . , 0), . . . , e
n+2
:= (0, . . . , 0, 1)
are not tangential to the surface. Its projections
N
∗
1
:= e
3
−
e
3
X
u
[X
u
[
2
X
u
−
e
3
X
v
[X
v
[
2
X
v
, N
∗
2
:= e
4
−
e
4
X
u
[X
u
[
2
X
u
−
e
4
X
v
[X
v
[
2
X
v
etc. can be transformed into such an ONF ¦N
1
, . . . , N
n
¦. We use this frame in
the ﬁrst part of the proof.
10.3 A curvature estimate for surfaces with prescribed mean curvature 199
1.2 Using conformal parameters (u, v) ∈ B we recall the estimate
[△X(u, v)[ ≤nh
0
[∇X(u, v)[
2
in B. (10.8)
1.3 Assumption (A2) yields: Let Γ(B) denotes the set of all continuous and
piecewise continuously differentiable curves γ : [0, 1] →B with γ(0) = (0, 0)
and γ(1) ∈ ∂B. Then it holds
inf
γ∈Γ(B)
1
_
0
¸
¸
¸
¸
d
dt
X ◦ γ(t)
¸
¸
¸
¸
dt ≥r.
1.4 Assumption (A3) ensures the estimate
[∇x
1
[
2
≥W sin
2
ω in B,
see the proof of Osserman’s curvature estimate fromsection 9.7. This inequal
ity actually is true for immersions which are not necessarily minimal for its
proof makes only use of the conformality relations.
To construct a lower bound for the area element we will employ various results
from Heinz [80], see section 9.4.
1.5 We deﬁne the plane mapping
F
∗
(u, v) := (x
1
(u, v), x
2
(u, v)): B −→R
2
.
(i) First we infer
[∆F
∗
(w)[ ≤
2nh
0
sin
2
ω
[∇F
∗
(w)[
2
for all w ∈ B
taking
[∆F
∗
[ ≤ [∆X[ ≤ nh
0
[∇X[
2
= 2nh
0
W ≤
2nh
0
sin
2
ω
[∇x
1
[
2
≤
2nh
0
sin
2
ω
[∇F
∗
[
2
into account;
(ii) Secondly it holds
[∇X(w)[
2
≤
2
sin
2
ω
[∇F
∗
(w)[
2
for all w ∈ B
what follows from
[∇X[
2
= 2W ≤
2
sin
2
ω
[∇x
1
[
2
≤
2
sin
2
ω
[∇F
∗
[
2
.
200 10 Immersions with prescribed mean curvature
1.6 Now let w
0
∈ B and ν ∈ (0, 1) such that
˚
B
2ν
(w
0
) :=
_
w ∈ B : [w−w
0
[ < 2ν
_
⊂
˚
B.
We consider
Y(w) :=
1
r
¦X(w
0
+2νw) −X(w
0
)¦, w ∈ B,
as well as
F(w) := (y
1
(w), y
2
(w)): B −→R
2
.
Again there hold
[Y
u
(w)[
2
=
4ν
2
r
2
W(w
0
+2νw) =[Y
v
(w)[
2
, Y
u
(w) Y
v
(w) = 0,
and due to (10.8) we infer
[△Y(w)[ ≤n(h
0
r)[∇Y(w)[
2
in
˚
B.
1.7 Together with point 1.5(ii) of the proof we conclude
[△F(w)[ ≤ [△Y(w)[ ≤ n(h
0
r)[∇Y(w)[
2
=
8nν
2
(h
0
r)
r
2
W(w
0
+2νw)
=
4nν
2
(h
0
r)
r
2
[∇X(w
0
+2νw)[
2
≤
8nν
2
(h
0
r)
r
2
sin
2
ω
[∇F
∗
(w
0
+2νw)[
2
=
8nν
2
(h
0
r)
r
2
sin
2
ω
r
2
4ν
2
[∇F(w)[
2
≤
2n(h
0
r)
sin
2
ω
[∇F(w)[
2
(10.9)
for all w∈
˚
B. Due to (A1) the mapping F =F(u, v) is onetoone with positive
Jacobian J
F
(w) > 0 in
˚
B. Furthermore, (A2) gives
D[F] ≤D[Y] ≤
1
r
2
D[X] ≤2d
0
for the Dirichlet energy of F given by
D[F] =
__
B
_
[F
u
[
2
+[F
v
[
2
_
dudv.
Following now Heinz [80], Theorem 6 (see section 9.4 above) there is a con
stant c
1
= c
1
(h
0
r, d
0
, sinω, n) ∈ (0, +∞) such that
[∇F(u, v)[ ≤c
1
(h
0
r, d
0
, sinω, n) for all (u, v) ∈
˚
B1
2
(0, 0).
10.3 A curvature estimate for surfaces with prescribed mean curvature 201
1.8 From (10.9) we infer
1
r
2
W(w
0
+2νw) ≤
1
4ν
2
sin
2
ω
c
1
(h
0
r, d
0
, sinω, n)
2
=: c
2
(h
0
r, d
0
, sinω, ν, n)
for all w ∈
˚
B1
2
(0, 0). In particular, it holds
1
r
2
W(w) ≤c
2
(h
0
r, d
0
, sinω, ν, n) for all w ∈ B
ν
(w
0
) (10.10)
and all w
0
∈
˚
B mit
˚
B
2ν
(w
0
) ⊂
˚
B. Later we will make use of this estimate.
1.9 Together with
J
F
(w) > 0 and D[F] ≤2d
0
we apply Heinz [80], Lemma 17
1
: There exists a constant c
3
∈ (0, ∞) such
that
[∇F(w)[
2
≤c
3
(h
0
r, d
0
, sinω, n)[∇F(0, 0)[
2
5
for all w ∈
˚
B1
2
(0, 0).
Together with (10.9) we thus have
4ν
2
r
2
W(w
0
+2νw) ≤
1
sin
2
ω
[∇F(w)[
2
≤
c
3
(h
0
r, d
0
, sinω, n)
sin
2
ω
[∇F(0, 0)[
2
5
≤
c
3
(h
0
r, d
0
, sinω, n)
sin
2
ω
[∇Y(0, 0)[
2
5
=
c
3
(h
0
r, d
0
, sinω, n)
sin
2
ω
_
8ν
2
r
2
W(w
0
)
_
1
5
,
and rearranging proves the following inequality of Harnack type
_
W(w
0
)
r
2
_1
5
≥
4 8
−
1
5
ν
8
5
sin
2
ω
c
3
(h
0
r, d
0
, sinω, n)
W(w
0
+2νw)
r
2
for all w ∈
˚
B1
2
(0, 0).
1
Let the mapping z(w) = (x(w), y(w)) ∈C
2
(
˚
B, R) satisfy the differential inequality
[∆z[ ≤a([z
u
[
2
+[z
v
[
2
) +b([z
u
[ +[z
v
[)
with positive constants a and b. Furthermore, let x
u
y
v
−x
v
y
u
,= 0 and D[z] ≤ N < ∞. Then there
hold
c
1
(a, b, N, r)
_
[z
u
(0, 0)[ +[z
v
(0, 0)[
_1+3r
1−r
≤[z
u
(w)[ +[z
v
(w)[ ≤c
2
(a, b, N, r)
_
[z
u
(0)[ +[z
v
(0)[
_ 1−r
1+3r
for all [w[ ≤r < 1.
202 10 Immersions with prescribed mean curvature
Equivalently we arrive at
c
4
(h
0
r, d
0
, sinω, ν, n)
_
W(w)
r
2
_
5
≤
W(w
0
)
r
2
for all w ∈
˚
B
ν
(w
0
)
with the real constant
c
4
(h
0
r, d
0
, sinω, ν, n) :=
2
7
ν
8
sin
10
ω
c
3
(h
0
r, d
0
, sinω, n)
5
∈ (0, +∞).
1.10 Now we prove the existence of a point w
∗
∈ B
1−ν
0
(0, 0), where we intro
duce ν
0
:= min(e
−4πd
0
,
1
2
), such that
W(w
∗
)
r
2
≥
1
4(1 −e
−4πd
0
)
=: c
5
(d
0
) > 0.
For this purpose we apply the lemma Courant and Lebesgue considering as
sumption (A2): For given δ := e
−4πd
0
we ﬁnd a δ
∗
∈ [δ,
√
δ] such that
_
[w−w
0
[=δ
∗
[dX(w)[ ≤2
¸
πd
0
r
2
−logδ
= 2
¸
πd
0
r
2
4πd
0
= r.
Recall point 1.3 of the proof and consider a curve γ
1
(t) in B from the origin
(0, 0) to w
1
:= (1 −δ
∗
)w ∈ B satisfying
γ
1
(t) := w
0
t, t ∈ [0, 1 −δ
∗
].
At the same time consider curves
γ
2
(t) := w
0
+(w−1 −w
0
)e
±it
, t ∈ [0, t
1
(δ
∗
)],
starting in w
1
, where t
1
(δ
∗
) is chosen such that γ
2
(t) ends on ∂B, and this
curve runs along the arc [w−w
0
[ = δ
∗
, w ∈ B. Furthermore, specify γ
2
(t) to
ensure
t
1
(δ
∗
)
_
0
¸
¸
¸
¸
d
dt
X(γ
2
(t))
¸
¸
¸
¸
dt ≤
r
2
.
Combine γ
1
(t) and γ
2
(t) to a continuous and piecewise continuously differen
tiable curve γ(t) and compute with a suitable t
0
∈ [0, 1 −δ
∗
]
r ≤
1−δ
∗
_
0
¸
¸
¸
¸
d
dt
X(γ
1
(t))
¸
¸
¸
¸
dt +
t
1
(δ
∗
)
_
0
¸
¸
¸
¸
d
dt
X(γ
2
(t))
¸
¸
¸
¸
dt
≤ (1 −δ)
¸
¸
¸
¸
d
dt
X(γ
1
(t
0
))
¸
¸
¸
¸
+
r
2
.
10.3 A curvature estimate for surfaces with prescribed mean curvature 203
Rearranging yields
¸
¸
¸
¸
d
dt
X(γ
1
(t
0
))
¸
¸
¸
¸
≥
r
2(1 −δ)
=
r
2(1 −e
−4πd
0
)
proving the existence of the desired point w
∗
∈ B
1−ν
0
(0, 0).
1.11 Using the above Harnack inequality (with the setting ν :=
1
2
ν
0
<
1
4
) we
arrive iteratively at
W(w
0
)
r
2
≥c
1+5+5
2
+...+5
m−1
4
c
5
(d
0
)
5
m
=: C
1
(h
0
r, d
0
, sinω, n) > 0
for all w
0
∈ B
1−ν
0
(0, 0). We infer
W(w)
r
2
≥C
1
(h
0
r, d
0
, sinω, n) for all w ∈ B1
2
(0, 0). (10.11)
This determines the ﬁrst part of our proof.
2. The second part of the proof is devoted to establishing an upper bound for the
second derivatives of X = X(u, v) making use of the elliptic system
△X = 2H(X, N
1
)WN
1
+2H(X, N
2
)WN
2
+. . . +2H(X, N
n
)WN
n
.
For this purpose we construct a suitable ONF N = (N
1
, . . . , N
n
).
2.1 First we deﬁne
Z(u, v) =
1
r
¦X(u, v) −X(0, 0)¦ =
1
r
X(u, v), (u, v) ∈ B.
Let W
Z
denote the area element of the mapping Z = Z(u, v). Then we have
[Z
u
[
2
=W
Z
=[Z
v
[
2
und Z
u
Z
v
= 0 in B.
Obviously it holds r
2
W
Z
=W
X
with W
X
:=[X
u
[
2
=[X
v
[
2
, and we compute
△Z =
2
r
H(X, N
1
)W
X
N
1
+. . . +
2
r
H(X, N
n
)W
X
N
n
= 2rH(rZ, N
1
)W
Z
N
1
+. . . +2rH(rZ, N
n
)W
Z
N
n
.
2.2 From the estimate (10.10) we infer
[△Z(w)[ ≤2(n −2)(rh
0
)c
2
(h
0
r, d
0
, sinω, n) (10.12)
for all w ∈ B1
2
(0, 0).
204 10 Immersions with prescribed mean curvature
Due to Z(0, 0) = (0, 0, 0) we have furthermore
[Z(u, v)[ = [Z(u, v) −Z(0, 0)[ ≤ 2 max
w∈B
1
2
(0,0)
[∇Z(w)[
≤ 2
_
2c
2
(h
0
r, d
0
, sinω, n) in B1
2
(0, 0).
Potential theoretic estimates (see e.g. Sauvigny [143], Kapitel XII, §2, Satz 2
2
yield a real constant c
6
(h
0
r, d
0
, sinω, n, α) with the property
[Z
u
i (w
1
) −Z
u
i (w
2
)[ ≤c
6
(h
0
r, d
0
, sinω, n, α)[w
1
−w
2
[
α
for w
1
, w
2
∈ B1
4
(0, 0)
(10.13)
and for all α ∈ (0, 1). Thus we arrive at
[W
Z
(w
1
) −W
Z
(w
2
)[ ≤c
7
(h
0
r, d
0
, sinω, n, α)[w
1
−w
2
[
α
for all w
1
, w
2
∈ B1
4
(0, 0)
with a constant c
7
:= 4
√
c
2
c
6
.
2.3 Using the mean value theoremwe arrive at the following Lipschitz estimate
[Z(w
1
) −Z(w
2
)[ ≤c
8
(h
0
r, d
0
, sinω, n, α)[w
1
−w
2
[ for w
1
, w
2
∈ B1
2
(0, 0).
In a neighborhood of the origin we now construct an ONF N with controlled
H¨ older norm.
2.4 For this purpose we choose unit normal vectors N
1
, . . . , N
n
∈ R
n+2
such
that
N
σ
Z
u
j (0, 0) = 0, N
σ
N
ω
=δ
σω
, j = 1, 2, σ, ω = 1, . . . , n. (10.14)
We deﬁne
N
∗
σ
(w) := N
σ
−
N
σ
Z
u
(w)
[Z
u
(w)[
2
Z
u
(w) −
N
σ
Z
v
(w)
[Z
v
(w)[
2
Z
v
(w) in
˚
B.
2.5 These vectors belong to normal space of Z(w) but eventually they are not
linearly independent.
2
Let X ∈C
2
(B, R
n
) solve [∆X[ ≤ a[∇X[
2
+b in B such that aM < 1 with M := sup
w∈B
[X(w)[. Let
α ∈ (0, 1). Then it holds
X
C
1+α
(B
1−ε
(0,0))
≤C(a, b, M, ε, α).
10.3 A curvature estimate for surfaces with prescribed mean curvature 205
We determine a ν
1
=ν
1
(h
0
r, d
0
, sinω, n, α) > 0 with the property
[N
∗
σ
(w)[
2
= 1 −
[N
σ
Z
u
(w)]
2
W
Z
(w)
−
[N
σ
Z
v
(w)]
2
W
Z
(w)
≥
1
2
in
˚
B
ν
1
(0, 0). (10.15)
Together with (10.14) and (10.13) we calculate
[N
σ
Z
u
ℓ (w)[
2
= [N
σ
¦Z
u
ℓ (w) −Z
u
ℓ (0, 0)¦[
2
≤ [Z
u
ℓ (w) −Z
u
ℓ (0, 0)[
2
≤ c
6
(h
0
r, d
0
, sinω, n, α)
2
[w[
2α
for ℓ = 1, 2, σ = 1, . . . , n, and from (10.11) we deduce
W
Z
(w) ≥C
1
(h
0
r, d
0
, sinω, n) in
˚
B1
2
(0, 0).
Thus it holds (10.15) if ν
2α
1
≤
C
1
4c
6
.
2.6 Note that for the vectors N
∗
σ
(w), σ = 1, . . . , n, the H¨ older estimate
[N
∗
σ
(w
1
) −N
∗
σ
(w
2
)[ ≤c
9
(h
0
r, d
0
, sinω, n, α)[w
1
−w
2
[
α
for w
1
, w
2
∈
˚
B
ν
1
(0, 0) with a constant c
9
(h
0
r, d
0
, sinω, n, α) are true. We infer
this estimate from the H¨ older estimates for Z
u
j and the lower bound for W
Z
.
2.7 For σ = 1, . . . , n we deﬁne
¯
N
σ
(w) :=
N
∗
σ
(w)
[N
∗
σ
(w)[
in
˚
B
ν
1
(0, 0).
These vectors are welldeﬁned since
[N
∗
σ
(w)[
2
≥
1
2
in B
ν
1
(0, 0)
but they are not necessarily orthogonal. But note that
N
∗
σ
N
∗
ω
=
(N
σ
Z
u
)(N
ω
Z
u
)
W
Z
+
(N
σ
Z
v
)(N
ω
Z
v
)
W
Z
for σ ,=ω
from where we conclude
[
¯
N
σ
¯
N
ω
[ =
[N
∗
σ
N
∗
ω
[
[N
∗
σ
[[N
∗
ω
[
≤
2
C
1
_
[N
σ
Z
u
[[N
ω
Z
u
[ +[N
σ
Z
v
[[N
ω
Z
v
[
_
≤
4c
2
6
C
1
[w[
2α
.
206 10 Immersions with prescribed mean curvature
2.8 Thus we ﬁnd a ν
2
= ν
2
(h
0
r, d
0
, sinω, n, α) with 0 < ν
2
≤ ν
1
such that the
following vectors are welldeﬁned in
˚
B
ν
2
(0, 0) :
N
1
:=
¯
N
1
,
N
2
:=
¯
N
2
−
_
N
1
¯
N
2
_
N
1
_
1 −
_
N
1
¯
N
2
_
2
,
N
n
:=
¯
N
n
−
_
N
1
¯
N
n
_
N
1
−. . . −
_
N
n−1
¯
N
n
_
N
n−1
_
1 −
_
N
1
¯
N
n
_
2
−. . . −
_
N
n−1
¯
N
n
_
2
.
(10.16)
Namely, choose ν
2
∈ (0, 1) sufﬁciently small to ensure that each denominator
in (10.16) is greater or equal to
1
2
. These vectors form an ONF
˚
B
ν
2
(0, 0).
Additionally there hold the H¨ older estimates
[N
σ
(w
1
) −N
σ
(w
2
)[ ≤ c
10
(h
0
r, d
0
, sinω, n, α)[w
1
−w
2
[
α
for w
1
, w
2
∈ B
ν
2
(0, 0)
for σ = 1, . . . , n with a constant c
10
(h
0
r, d
0
, sinω, n, α) which can be deduced
from the H¨ older estimates of the N
∗
σ
.
2.9 Now we make use of the differential system
△Z = 2rH(rZ, N
1
)W
Z
N
1
+. . . +2rH(rZ, N
n
)W
Z
N
n
in B
ν
2
(0, 0).
We already know
[△Z(w)[ ≤2n(h
0
r)c
2
in B
ν
2
(0, 0),
see (10.12). Together with (10.5) we get
[H(rZ(w
1
), N
σ
(w
1
)) −H(rZ(w
2
), N
σ
(w
2
))[
≤ h
1
r
α
[Z(w
1
) −Z(w
2
)[
α
+h
2
[N
σ
(w
1
) −N
σ
(w
2
)[
≤ h
1
4
α
r
α
c
α
8
[w
1
−w
2
[
α
+h
2
c
10
[w
1
−w
2
[
α
.
Therefore we ﬁnd a real constant c
11
= c
11
(h
0
r, h
1
r
1+α
, h
2
r, d
0
, sinω, n, α)
such that it holds
[△Z(w
1
) −△Z(w
2
)[ ≤c
10
[w
1
−w
2
[
α
for w
1
, w
2
∈ B
ν
2
(0, 0).
10.4 A theorem of Bernstein type 207
2.10 Let ν
3
:=
1
2
ν
2
. From the inner Schauder estimates (see e.g. Gilbarg and
Trudinger [71]) we infer a real constant C
2
∈ (0, +∞) with the property
[Z
uu
(w)[, [Z
uv
(w)[, [Z
vv
(w)[ ≤C
2
(h
0
r, h
1
r
1+α
, h
2
r, d
0
, sinω, n, α)
in B
ν
3
(0, 0).
2.11 Now from the beginning of our proof we recall
κ
Σ,1
(0, 0)
2
+κ
σ,2
(0, 0)
2
≤
1
r
2
_
(h
0
r)
2
+
[Z
uu
(0, 0)[[Z
vv
(0, 0)[ +[Z
uv
(0, 0)[
2
W
Z
(0, 0)
2
_
.
If we set
Θ(h
0
r, h
1
r
1+α
, h
2
r, d
0
, sinω, n, α) :=
2C
2
2
C
2
1
we ﬁnd
κ
σ,1
(0, 0)
2
+κ
σ,2
(0, 0)
2
≤
1
r
2
_
(h
0
r)
2
+Θ
_
.
The proof is complete. ⊓⊔
10.4 A theorem of Bernstein type
The Bernstein type result
This curvature estimate enables us to prove the following theoremof Bernstein type.
Consider a minimal graph
(x, y, ϕ
1
, (x, y), . . . , ϕ
n
(x, y)), (x, y) ∈ R
2
.
Because its Gaussian curvature is nonpositive, by Hadamard’s theorem we can in
troduce geodesic discs B
r
(X
0
) for all X
0
= (x
0
, y
0
, ϕ
1
(x
0
, y
0
), . . . , ϕ
n
(x
0
, y
0
)) and all
r > 0. Then the limit r →∞ yields the
Corollary 10.1. Let X = X(x, y), (x, y) ∈ R
2
, be a complete minimal graph with the
properties
(i) there exists X
0
= (x
0
, y
0
, ϕ
1
(x
0
, y
0
), . . . , ϕ
n
(x
0
, y
0
)) and a radius r
0
> 0 such
that all geodesic discs B
r
(X
0
) with center X
0
and radius r ≥r
0
satisfy
Area[B
r
(X
0
)] ≤d
0
r
2
for all r ≥r
0
(10.17)
for their area Area[B
r
(X
0
)] with a constant d
0
∈ (0, +∞) which does not depend
on the radius r;
208 10 Immersions with prescribed mean curvature
(ii) each normal vector of the graph makes an angle of at least ω > 0 with the
x
1
axis.
Then X = X(x, y) is a linear mapping.
Proof. For any point X
1
= (x
1
, y
1
, ϕ
1
(x
1
, y
1
), . . . , ϕ
n
(x
1
, y
1
)) on the graph we have
Area[B
r
(X
1
)] ≤4d
0
r
2
for all r ≥max¦r
0
, d(X
0
, X
1
)¦ (10.18)
where d(X
0
, X
1
) ≥ 0 is the inner distance between X
0
and X
1
on the surface. This
holds because of the inclusion
B
r
(X
1
) ⊂B
2r
(X
0
) for all r ≥max¦r
0
, d(X
0
, X
1
)¦ (10.19)
and assumption (i). Since K ≤0 for the Gaussian curvature we can consider geodesic
discs B
r
(X
1
) for all r ∈ (0, +∞) on account of Hadamard’s theorem. Introduce con
formal parameters into such a geodesic disc. Using our curvature estimate and let
ting r →∞ shows that all principal curvatures at X
1
vanish which proves the Corol
lary (note that Θ does not depend on r since h
0
, h
1
, h
2
= 0). ⊓⊔
Remarks
We want to conclude this chapter with the following remarks.
1. Osserman’s curvature estimate from the previous chapter does not need a growth
estimate on the surface area.
2. Jost and Xin [104] proved a curvature estimates for submanifolds with parallel
mean curvature ﬁelds. Due to the higher dimension of the manifolds itself the
authors assume apriori bounds for the gradients. Consult also the references
therein.
3. Curvature estimates and related Bernstein type result for minimal submanifolds
can also be found in Smoczyk, Wang and Xin [149] where the authors extend
methods from Schoen, Simon and Yau [145] for minimal immersions with van
ishing normal sectional curvature; see also Ecker [50] or Wang [162].
4. Our method of proof uses essentially results from Heinz [80], and follows Sauvi
gny [141] where curvature estimates for twodimensional immersions of mean
curvature type in R
3
where established.
Chapter 11
Crystalline functionals in R
3
11.1 Examples of parametric variational problems
11.2 Further regularity assumptions
11.3 The ﬁrst variation
11.4 Principal curvatures and weighted mean curvature
11.5 Nonparametric differential equations
11.6 Quasilinear elliptic systems
11.7 Quadratic growth in the gradient
11.8 The geometry of immersions of mean curvature type
11.9 A curvature estimate
210 11 Crystalline functionals in R
3
In this chapter we consider general elliptic functionals and its critical points in Euclidean space
R
3
. The Lagrangian densities of these functionals depend on the surface vector X as well as the
normal direction X
u
X
v
.
11.1 Examples of parametric variational problems
In the following we study critical points X : B →R
3
of parametric variational prob
lems of the form
B[X] :=
__
B
F(X, X
u
X
v
)dudv −→extr!
The case of general case of codimensions n ≥1 is the subject in chapter 12.
But before we go further into details we want to consider some special parametric
functionals to get a ﬁrst impression of its broad range of applications.
Minimal surfaces
An immersion X : B →R
3
is a minimal surface if its scalar mean curvature H van
ishes identically. As we have seen minimal surfaces are critical points of the area
functional
A[X] =
__
B
[X
u
X
v
[ dudv.
With the setting
F(X, Z) =[Z[
the area functional turns out to be a special parametric functional of the formB[X].
A famous minimal surface already found by the Swiss mathematician Leonhard
Paul Euler (*1707 in Basel; †1783 in St. Petersburg) is the catenoid
X(u, v) = (coshucosv, coshusinv, u) ∈ R
3
which is generated by rotating the hyperbolic cosine, the socalled catenary curve,
about the xaxis.
212 11 Crystalline functionals in R
3
Note that a full rotation would generate a surface of higher topological type, thus
we rather consider X restricted to a suitable simply connected domain in R
2
.
The catenoid representation X(u, v) is conformal, i.e. there hold
X
2
u
= sinh
2
u, X
2
v
= 1 +cosh
2
u = sinh
2
u, X
u
X
v
= 0.
Furthermore, it can be deformed continuously into the minimal helicoid
X
∗
(u, v) = (sinhusinv, −sinhucosv, v)
as follows
x
α
(u, v) := cosαsinhusinv +sinα coshucosv,
y
α
(u, v) := −cosαsinhucosv +sinαcoshusinv,
z
α
(u, v) := vcosα +usinα.
The helicoid is given for α = 0, and α =
π
2
represents the catenoid. The surface
parametrizations
X
α
(u, v) =
_
x
α
(u, v), y
α
(u, v), z
α
(u, v)
_
are conformal for all α. This deformation is a special case of a general result by
Bour (see Strubecker [152].
The helicoid was discovered by the French mathematician Jean Baptiste Maria
Charles Meusnier de la Place (*1754 in Tours; †1793 in le Pont de Cassel).
Other important examples of minimal surfaces are the Scherk surface
X(u, v) = (u, v, logcosu +logcosv)
found by H. Scherk in 1834, the Henneberg surface
x(u, v) = 2sinhucosv −
2
3
sinh(3u)cos(3v),
y(u, v) = 2sinhusinv −
2
3
sinh(3u)sin(3v),
z(u, v) = 2cosh(2u)cos(2v)
named after E.L. Henneberg who found it in 1875, E.C. Catalan’s surface
x(u, v) = u −sinucoshv, y(u, v) = 1 −cosucoshv, z(u, v) = 4sin
u
2
sinh
v
2
and A. Enneper’s surface
x(u, v) = u −
1
3
u
3
+uv
2
, y(u, v) =−v −u
2
v +
1
3
v
3
, z(u, v) = u
2
−v
2
.
11.1 Examples of parametric variational problems 213
By the way, the later one is an algebraic surface since it can be rewritten in the
following algebraic form
_
y
2
−x
2
+
4
3
z +
4
9
z
3
_
3
= 3z
_
y
2
−x
2
+
8
9
z −z
_
x
2
+y
2
+
8
9
z
2
__
,
see Nitsche [126].
An abundance of examples and simulations of minimal surfaces can be found in
various textbooks and in the world wide net.
Immersions with prescribed mean curvature
We also considered parametric functionals whose critical points possess nonvanishing
mean curvature ﬁelds, for example
__
B
_
[X
u
X
v
[ +
2h
0
3
X (X
u
X
v
)
_
dudv
with the volume constraint
2h
0
3
__
B
X (X
u
X
v
)dudv = 1.
Critical points have constant scalar mean curvature H ≡h
0
. We have
F(X, Z) =[Z[ +
2h
0
3
X Z.
Or, more generally, Hildebrandt’s functional
__
B
_
[X
u
X
v
[ +2Q(X) (X
u
X
v
)
_
dudv
with the constraint
2
__
B
Q(X) (X
u
X
v
)dudv = 1
for some given vector ﬁeld Q(X) = (q
1
(X), q
2
(X), q
3
(X)). Critical points are im
mersions with mean curvature
H(X) =
∂
∂x
1
q
1
(X) +
∂
∂x
2
q
2
(X) +
∂
∂x
3
q
3
(X).
214 11 Crystalline functionals in R
3
Fminimal surfaces
Of special interest are anisotropic parametric functionals of the form
F[X] :=
__
B
F(X
u
X
v
)dudv
with Lagrangians independent of the space point X. Critical points for such func
tionals are socalled Fminimal surfaces. Elaborating the imbedding of this class
of immersions into the classical calculus of variations is one topic of the following
considerations.
Critical points of F[X] are sort of analytical models for crystall growth processes
motivated by the work of Wulff in [174]. Spherical surfaces of this kind are com
monly called Wulff surfaces. We particularly refer the reader to Taylor [153], [154],
or Morgan [120].
Our analysis follows the ideas of Sauvigny [141] in essential points.
11.2 Further regularity assumptions
We consider the general parametric variational problem
B[X] =
__
B
F(X, X
u
X
v
)dudv −→extr!
with a Lagrangian density
F ∈C
5
(R
3
R
3
¸ ¦0¦, R).
We assume that F(X, Z) satisﬁes the following conditions.
(H) Homogeneity: For all λ > 0 it holds
F(X, λZ) =λF(X, Z).
(D) Positive deﬁniteness: There exist two real constants 0 < m
1
≤m
2
<∞ with
m
1
[Z[ ≤F(X, Z) ≤m
2
[Z[.
(E) Ellipticity: There exist two real constants 0 < M
1
≤M
2
<∞ such that
M
1
[ξ[
2
≤ξ ◦ F
ZZ
(X, Z) ◦ ξ ≤M
2
[ξ[
2
for all ξ ⊥Z
with the matrix
F
ZZ
(X, Z) = (F
z
i
z
j (X, Z))
i, j=1,2,3
∈ R
33
.
11.2 Further regularity assumptions 215
This latter ellipticity is equivalent to the convexity condition
F(X, λZ
1
+µZ
2
) ≤λF(X, Z
1
) +µF(X, Z
2
)
for all real λ, µ ∈> 0 and all Z
1
, Z
2
∈ R
3
¸ ¦0¦.
Deﬁnition 11.1. We additionally set
F
X
(X, Z) :=
_
F
x
1 (X, Z), F
x
2 (X, Z), F
x
3 (X, Z)
_
∈ R
3
,
F
Z
(X, Z) :=
_
F
z
1 (X, Z), F
z
2 (X, Z), F
z
3 (X, Z)
_
∈ R
3
,
F
XX
(X, Z) :=
_
F
x
i
x
j (X, Z)
_
∈ R
33
,
F
XZ
(X, Z) :=
_
F
x
i
z
j (X, Z)
_
∈ R
33
.
The Euler homogeneity conditions
Differentiation of the Lagrangian F(X, Z) yields the socalled Euler homogeneity
conditions as follows.
Lemma 11.1. The vector F
Z
(X, Z) is positive homogeneous of degree 0 w.r.t. Z, the
matrix F
ZZ
(X, Z) is positive homogeneous of degree −1 w.r.t. Z. Furthermore there
hold the relations
F
Z
(X, Z) Z = F(X, Z),
F
ZZ
(X, Z) ◦ Z = 0,
F
XZ
(X, Z) ◦ Z = F
X
(X, Z)
for all X ∈ R
3
and all Z ∈ R
3
¸ ¦0¦.
Proof. Differentiation of the homogeneity condition (H) w.r.t. λ yields
F
Z
(X, λZ) Z = F(X, Z).
Thus the ﬁrst statement follows if we set λ = 1. A further differentiation yields
F
ZZ
(X, Z)◦Z =0. Computing ﬁnally the derivative of F
X
(X, λZ) =λF
X
(X, Z) w.r.t.
λ shows the third relation. The lemma is proved. ⊓⊔
Parameter invariance
We show that the homogeneity condition ensures the parameter invariance of the
functional B[X].
Proposition 11.1. B[X] is invariant w.r.t. parameter transformations from class P
and does not depend on the chosen parameter domain iff it holds the homogeneity
condition (H).
216 11 Crystalline functionals in R
3
Proof. 1. Let ˜ u = ˜ u(u, v), ˜ v = ˜ v(u, v) be a parameter transformation of class P.
Assuming that (H) holds, we compute
__
B
F(X, X
u
X
v
)dudv =
__
B
F(X, ( ˜ u
u
˜ v
v
− ˜ u
v
˜ v
u
)X
˜ u
X
˜ v
)dudv
=
__
B
F(X, X
˜ u
X
˜ v
)( ˜ u
u
˜ v
v
− ˜ u
v
˜ v
u
)dudv =
__
¯
B
F(X, X
˜ u
X
˜ v
)d ˜ ud ˜ v
which proves the parameter invariance of B[X].
2. The value of the functional B[X] does not depend on the parametrization. We set
σ := ˜ u
u
˜ v
v
− ˜ u
v
˜ v
u
and infer
__
¯
B
F(X, X
˜ u
X
˜ v
)d ˜ ud ˜ v =
__
B
F(X, σ
−1
X
u
X
v
)σ dudv
=
!
__
B
F(X, X
u
X
v
)dudv.
Applying this formula to the special transformation ˜ u =
√
λ u, ˜ v =
√
λ v for real
λ > 0 gives the homogeneity condition in integral form
__
B
F(X, λX
u
X
v
)dudv =
__
B
λF(X, X
u
X
v
)dudv.
But this identity holds true on arbitrary discs B
r
(w), i.e.
__
B
r
(w)
F(X, λX
u
X
v
)dudv =
__
B
r
(w)
λF(X, X
u
X
v
)dudv.
This proves the homogeneity condition (H) and thus the proposition. ⊓⊔
Condition (D) for compactly immersed surfaces
Consider an immersion X : B →R
3
satisfying
X(u, v) ∈ K ⊂R
3
for all (u, v) ∈ B
with a compact subset K ⊂R
3
. We want to verify condition (D). For this set
m
1
:= inf
X⊂K
inf
E∈S
2
F(X, E), m
2
:= sup
X⊂K
sup
E∈S
2
[F
Z
(X, E)[
with the unit sphere S
2
:= ¦Z ∈ R
3
: [Z[ = 1¦.
11.3 The ﬁrst variation 217
Then the homogeneity relation (H) implies
m
1
[Z[ ≤F(X, Z) ≤m
2
[Z[ for all (X, Z) ∈ K R
3
¸ ¦0¦.
Namely, for ﬁxed Z ∈ R
3
¸ ¦0¦ and arbitrary X ∈ K we compute
F(X, Z) ≤ sup
X⊂K
F(X, Z) = sup
X⊂K
F
Z
(X, Z) Z
≤ sup
X⊂K
[F
Z
(X, Z)[[Z[ ≤ sup
X⊂K
sup
E∈S
2
[F
Z
(X, E)[[Z[ = m
2
[Z[
and analogously
m
1
≤F(X, Z/[Z[) =
1
[Z[
F(X, Z).
Thus we arrive at condition (D).
11.3 The ﬁrst variation
The EulerLagrange equations
In this section we want to derive the EulerLagrange equations for critical points
X : B →R
3
of the parametric functional B[X].
We use the following abbreviation for the triple product
[X,Y, Z] := X (Y Z), X,Y, Z ∈ R
3
.
Theorem 11.1. Let the immersion X : B →R
3
be critical for B[X]. Then it satisﬁes
[F
ZZ
(X, N)◦N
u
, N, X
v
] +[F
ZZ
(X, N)◦N
v
, X
u
, N] =W traceF
XZ
(X, N) in B (11.1)
with the trace term
traceF
XZ
(X, Z) = F
x
1
z
1 (X, Z) +F
x
2
z
2 (X, Z) +F
x
3
z
3 (X, Z).
Proof. Consider the variation
¯
X(u, v) := X(u, v) +εϕ(u, v)N(u, v), (u, v) ∈ B,
with a test function ϕ ∈C
∞
0
(B, R) and ε ∈ (−ε
0
, +ε
0
) small. First we have
¯
X
u
¯
X
v
= X
u
X
v
+ε (X
u
N
v
+N
u
X
v
)ϕ +ε (ϕ
v
X
u
N+ϕ
u
N X
v
) +o(ε).
218 11 Crystalline functionals in R
3
We compute
∂
∂ε
B[
¯
X] =
∂
∂ε
__
B
F(
¯
X,
¯
X
u
¯
X
v
)dudv
=
__
B
F
X
(
¯
X,
¯
X
u
¯
X
v
)
∂
∂ε
¯
X dudv
+
__
B
F
Z
(
¯
X,
¯
X
u
¯
X
v
)
∂
∂ε
(
¯
X
u
¯
X
v
)dudv.
Let as usual
δB[X] =
∂
∂ε
B[
¯
X]
¸
¸
ε=0
.
Using F
Z
(X, λZ) = F
Z
(X, Z) we arrive at
δB[X] =
__
B
F
X
(X, X
u
X
v
) Nϕdudv
+
__
B
F
Z
(X, X
u
X
v
) (X
u
N
v
+N
u
X
v
)ϕ dudv
+
__
B
F
Z
(X, X
u
X
v
) (X
u
Nϕ
v
+NX
v
ϕ
u
)dudv
=
__
B
[F
X
(X, N), X
u
, X
v
]ϕ dudv
+
__
B
_
[F
Z
(X, N), X
u
, N
v
] +[F
Z
(X, N), N
u
, X
v
]
_
ϕ dudv
+
__
B
_
[F
Z
(X, N), X
u
, N]ϕ
v
+[F
Z
(X, N), N, X
v
]ϕ
u
_
dudv.
The last two integrals are equal to
__
B
div([F
Z
(X, N), N, X
v
]ϕ, [F
Z
(X, N), X
u
, N]ϕ)dudv
−
__
B
_
[F
ZX
(X, N) ◦ X
u
, N, X
v
] +[F
ZX
(X, N) ◦ X
v
, X
u
, N]
_
ϕ dudv
−
__
B
_
[F
ZZ
(X, N) ◦ N
u
, N, X
v
] +[F
ZZ
(X, N) ◦ N
v
, X
u
, N]
_
ϕ dudv.
11.3 The ﬁrst variation 219
The integral over the divergence term vanishes due to ϕ[
∂B
= 0, i.e.
__
B
div([F
Z
(X, N), N, X
v
]ϕ, [F
Z
(X, N), X
u
, N]ϕ)dudv = 0.
Furthermore, F
X
(X, Z) = F
XZ
(X, Z) ◦ Z yields
[F
X
(X, N), X
u
, X
v
] = [F
XZ
(X, N) ◦ N, X
u
, X
v
] = [X
u
, X
v
, F
XZ
(X, N) ◦ N]
= [X
u
, X
v
, F
ZX
(X, N) ◦ N]
where we make use of
Z ◦ F
XZ
(X, Z) ◦ Z = Z ◦ F
ZX
(X, Z) ◦ Z.
Now we insert
[F
ZX
(X, N) ◦ X
u
, X
v
, N] +[X
u
, F
ZX
(X, N) ◦ X
v
, N] +[X
u
, X
v
, F
ZX
(X, N) ◦ N]
= [X
u
, X
v
, N] traceF
ZX
(X, N) = W traceF
XZ
(X, N).
into the δB[X] to get
δB[X] =
__
B
[F
X
(X, N), X
u
, X
v
]ϕ dudv
−
__
B
_
[F
ZX
(X, N) ◦ X
u
, N, X
v
] +[F
ZX
(X, N) ◦ X
v
, X
u
, N]
_
ϕ dudv
−
__
B
_
[F
ZZ
(X, N) ◦ N
u
, N, X
v
] +[F
ZZ
(X, N) ◦ N
v
, X
u
, N]
_
ϕ dudv
=
__
B
traceF
XZ
(X, N)Wϕ dudv
−
__
B
_
[F
ZZ
(X, N) ◦ N
u
, N, X
v
] +[F
ZZ
(X, N) ◦ N
v
, X
u
, N]
_
ϕ dudv.
The fundamental lemma of the calculus of variations yields
[F
ZZ
(X, N) ◦ N
u
, N, X
v
] +[F
ZZ
(X, N) ◦ N
v
, X
u
, N] =W traceF
XZ
(X, N).
The theorem is proved. ⊓⊔
220 11 Crystalline functionals in R
3
Critical points are immersions of mean curvature type
Following Sauvigny [141] we introduce a special weight matrix W(X, Z) to trans
form the EulerLagrange equations from the previous theorem into the socalled
system of meancurvature type.
Theorem 11.2. A critical point X : B →R
3
of B[X] solves the nonlinear elliptic
system
∇
ds
2
W
(X, N) :=
2
∑
i, j=1
h
i j
X
u
i N
u
j =−2H
W
(X, N) in B (11.2)
with the parameter invariant Beltrami operator ∇
ds
2
W
(, ) w.r.t.
− the metric
ds
2
W
:=
2
∑
i, j=1
h
i j
du
i
du
j
, h
i j
:= X
u
i ◦ W(X, N) ◦ X
u
j , (11.3)
− the weight matrix
W(X, Z) :=
_
1
_
det F
ZZ
(X, Z)
F
ZZ
(X, Z) +(z
i
z
j
)
i, j=1,2,3
_
−1
(11.4)
for X ∈ R
3
and Z ∈ S
2
,
− and the weighted mean curvature
H
W
(X, Z) =
1
2
_
det F
ZZ
(X, N)
traceF
XZ
(X, N).
Before we come to the proof of this theorem we want to verify ﬁrst that W(X, N)
from this deﬁnition actually represents a weight matrix in the sense of our deﬁnition
from chapter 4. Then we come to the proof the parameter invariance of the Beltrami
operator ∇
ds
2
W
(ϕ, ψ).
Properties of the weight matrix
In chapter 4 we introduced weight matrices W(X, Z) on the tangential space of sur
faces with arbitrary codimensions. The properties (W1) to (W4) given there reduce
to the following:
For all X ∈ R
3
and all Z ∈ R
3
¸ ¦0¦ there hold
(W1) W(X, Z) is invariant w.r.t. the choice of the normal vector, i.e.
W(X, λZ) = W(X, Z);
11.3 The ﬁrst variation 221
(W2) W(X, Z) acts nontrivially on the tangent space, i.e.
W(X, Z)
¸
¸
T
X
(w)
: T
X
(w) −→T
X
(w), in particular,
rankW(X, Z)
¸
¸
T
X
(w)
= 2, W(X, Z) ◦ Z = Z;
(W3) W(X, Z) is positive deﬁnite, i.e. with a real constant ω
0
∈ [0, ∞) we have
(1 +ω
0
)
−1
[ξ[
2
≤ξ ◦ W(X, Z) ◦ ξ ≤(1 +ω
0
)[ξ[
2
for all ξ ∈ R
3
;
(W4) W(X, Z) is normalized in the following sense
det W(X, Z) = 1.
Let us now consider the special weight matrix W(X, Z) from the previous theo
rem. We show that it satisﬁes these properties (W1) to (W4).
1. First we continue W(X, Z) constantly along lines ¦λZ¦ for λ ∈ R and Z ∈ S
2
such that it holds
W(X, λZ) = W(X, Z) for all λ ∈ R.
This proves (W1).
2. The weight matrix W(X, Z) consists of the mapping F
ZZ
(X, Z) which acts on the
tangential space, and of the mapping (z
i
z
j
)
i, j=1,2,3
. There hold
rankF
ZZ
(X, Z) = 2, rank(z
i
z
j
)
i, j=1,2,3
= 1.
The weight matrix is homogeneous of degree 0 in Z. We have
W(X, Z) ◦ Z = Z for all Z ∈ S
2
where we note
F
ZZ
(X, Z) ◦ Z = 0
as well as
(z
i
z
j
)
i, j=1,2,3
◦ Z =
_
_
_
(z
1
)
3
+z
1
(z
2
)
2
+z
1
(z
3
)
2
(z
1
)
2
z
2
+(z
2
)
3
+z
2
(z
3
)
2
(z
1
)
2
z
3
+(z
2
)
2
z
3
+(z
3
)
2
_
_
_= Z
for all Z ∈ S
2
. We infer (W4) from
det W(X, Z) = 1 for all X ∈ R
3
and Z ∈ S
2
.
222 11 Crystalline functionals in R
3
Property (W2) follows now from
W(X, Z)
−1
¸
¸
¸
T
Z
=
1
_
det F
ZZ
(X, Z)
F
ZZ
(X, Z) : T
Z
−→T
Z
for all Z ∈ S
2
.
3. Let ﬁnally ξ ⊥N be an arbitrary tangential vector. Then the ellipticity condition
(E) on F
ZZ
(X, Z) ensures
ξ ◦ W(X, N)
−1
◦ ξ =
1
_
det F
ZZ
(X, N)
ξ ◦ F
ZZ
(X, N) ◦ ξ ≤
M
2
M
1
[ξ[
2
as well as
ξ ◦ W(X, N)
−1
◦ ξ ≥
M
1
M
2
[ξ[
2
.
Thus (W3) follows with the setting
ω
0
:=
M
2
M
1
−1.
Note that ω
0
depends only on the ratio of the eigenvalue bounds on F
ZZ
(X, Z).
The invariant Beltrami operator
We come back to the operator ∇
ds
2
W
(ϕ, ψ), see e.g. Blaschke and Leichtweiß [15].
Deﬁnition 11.2. The ﬁrst Beltrami operator for continuously differentiable func
tions ϕ, ψ: B →R w.r.t. the line element ds
2
W
is deﬁned as
∇
ds
2
W
(ϕ, ψ) :=
2
∑
i, j=1
h
i j
ϕ
u
i ψ
u
j .
Analogously we deﬁne the ﬁrst Beltrami operator w.r.t. any other Riemannian
line element ds
2
, or even for smooth vectorvalued mappings X and Y, namely
∇
ds
2
W
(X,Y) =
2
∑
i, j=1
h
i j
X
u
i X
u
j .
To prove its parameter invariance, let u
1
(v
α
) ∈ P be parameter transformation
from class P. Then
2
∑
i, j=1
h
i j
L
i j
=
2
∑
i, j=1
2
∑
κ,λ=1
2
∑
µ,ν=1
¯
Λ
i
κ
¯
Λ
j
λ
Λ
µ
i
Λ
ν
j
h
κλ
L
µν
=
2
∑
κ,λ=1
2
∑
µ,ν=1
δ
µ
κ
δ
ν
λ
h
κλ
L
µν
=
2
∑
κ,λ=1
h
κλ
L
κλ
.
11.3 The ﬁrst variation 223
Proof of the mean curvature type representation
Now we come to the proof of Theorem 11.2. Together with lemma 5.3, i.e.
(M◦ X) (M◦Y) = (det M)M
−1
◦ (X Y)
for all nonsingular and symmetric matrices M∈ R
33
and all X,Y ∈ R
3
, we infer
_
W(X, N)
−
1
2
◦ N
u
_
_
W(X, N)
1
2
◦ X
v
_
=
_
W(X, N)
1
2
◦ W(X, N)
−1
◦ N
u
_
_
W(X, N)
1
2
◦ X
v
_
= W(X, N)
−
1
2
◦
_
(W(X, N)
−1
◦ N
u
) X
v
_
= (W(X, N)
−1
◦ N
u
) X
v
.
Analogously we obtain
_
W(X, N)
1
2
◦ X
u
_
_
W(X, N)
−
1
2
◦ N
v
_
= X
u
(W(X, N)
−1
◦ N
u
).
Now insert the deﬁnitions of the weighted mean curvature H
G
(X, Z) and the weight
matrix W(X, Z) from the theorem into the EulerLagrange system to get
−2H
W
(X, N)W =
1
_
det F
ZZ
(X, N)
[X
u
, F
ZZ
(X, N) ◦ N
v
, N]
+
1
_
det F
ZZ
(X, N)
[F
ZZ
(X, N) ◦ N
u
, X
v
, N]
= [X
u
, W(X, N)
−1
◦ N
v
, N] +[W(X, N)
−1
◦ N
u
, X
v
, N]
= [W(X, N)
1
2
◦ X
u
, W(X, N)
−
1
2
◦ N
v
, N]
+[W(X, N)
−
1
2
◦ N
u
, W(X, N)
1
2
◦ X
v
, N].
Multiplication by
[N, W(X, N)
1
2
◦ X
u
, W(X, N)
1
2
◦ X
v
] = [N, X
u
, X
v
] =W
yields
−2H
W
(X, N)W
2
=
¸
¸
¸
¸
¸
¸
¸
N
u
X
u
X
u
◦ W(X, N) ◦ X
v
0
N
u
X
v
X
v
◦ W(X, N) ◦ X
v
0
0 0 1
¸
¸
¸
¸
¸
¸
¸
+
¸
¸
¸
¸
¸
¸
¸
X
u
◦ W(X, N) ◦ X
u
N
v
X
u
0
X
v
◦ W(X, N) ◦ X
u
N
v
X
v
0
0 0 1
¸
¸
¸
¸
¸
¸
¸
.
224 11 Crystalline functionals in R
3
This is equivalent to
−2H
W
(X, N)W
2
= h
22
(X
u
N
u
) −2h
12
(X
v
N
u
) +h
11
(X
v
N
v
)
= W
2
2
∑
i, j=1
h
i j
(X
u
i N
u
j ).
This proves the stated representation. ⊓⊔
Example: Minimal surfaces
Consider again the area function
A[X] =
__
B
[X
u
X
v
[ dudv
with the Lagrangian F(Z) =[Z[. There hold
F
Z
(X, Z) =
Z
[Z[
, F
X
(X, Z) ≡0
as well as
F
ZZ
(X, Z) =
_
δ
i j
[Z[
−
z
i
z
j
[Z[
3
_
i, j=1,2,3
with the Kronecker symbol δ
i j
, thus W(X, Z) ≡E
3
with the threedimensional unit
matrix E
3
. Thus minimal surfaces X : B →R
3
as critical points of the area func
tional A[X] fulﬁll
∇
ds
2 (X, N) =
g
22
L
11
−2g
12
L
12
+g
11
L
22
W
2
= 0 in B
with the classical metric
ds
2
= g
11
du +2g
12
dudv +g
22
dv
2
.
It particularly holds H ≡0 for its scalar mean curvature.
Example: Immersions with prescribed mean curvature
We consider two variational problems from the introduction with critical points pos
sessing either constant mean curvature or mean curvature represented by the diver
gence of the discussed vector ﬁeld in space.
In both examples the weight matrix W(X, Z) simply equals the threedimensional
unit matrix E
3
.
11.3 The ﬁrst variation 225
1. Immersions with constant mean curvature
First consider the functional
__
B
_
[X
u
X
v
[ +
2h
0
3
X (X
u
X
v
)
_
dudv.
We immediately compute
F
X
(X, Z) =
2h
0
3
Z, F
Z
(X, Z) =
Z
[Z[
+
2h
0
3
X,
furthermore
F
XX
(X, Z) = 0, F
XZ
(X, Z) =
2h
0
3
E
3
, F
ZZ
(X, Z) =
_
δ
i j
[Z[
−
z
i
z
j
[Z[
3
_
i, j=1,2,3
as well as
F
XZ
(X, Z) = F
ZX
(X, Z).
Thus cricital points X : B →R
3
of this functional have constant mean curvature
H(u, v) ≡h
0
in B,
and they solve
∇
ds
2 (X, N) =−2h
0
w.r.t. the nonweighted line element ds
2
.
2. Immersions with prescribed mean curvature
Now we come to Hildebrandt’s functional
__
B
_
[X
u
X
v
[ +2Q(X) (X
u
X
v
)
_
dudv
with a prescribed vector ﬁeld Q(X) = (q
1
(X), q
2
(X), q
3
(X)). We have
F
XZ
(X) = (F
x
i
z
j )
i, j=1,2,3
= (2q
j,x
i )
i, j=1,2,3
∈ R
33
.
Thus critical points X : B →R
3
possess the mean curvature
H(X) =
3
∑
i=1
q
i,x
i (X) = div
X
Q(X) in B
with the spatial divergence operator div
X
. It holds
∇
ds
2 (X, N) =−2div
X
Q(X)
again w.r.t. the classical metric ds
2
.
226 11 Crystalline functionals in R
3
Example: Immersions of minimal surface type. Fminimal surfaces
For critical points of the anisotropic parametric functional
F[X] =
__
B
F(X
u
X
v
)dudv
we immediately infer
F
XZ
(X, Z) ≡0.
Now introducing the weight matrix W(X, Z) from Theorem 11.2 we arrive at
∇
ds
2
W
(X, N) = 0 in B
with the weighted line element
ds
2
W
= h
11
du +2h
12
dudv +h
22
dv
2
from (11.3). Thus critical points of F[X] are characterized by the property that
their weighted mean curvature H
W
from Theorem 11.2 vanishes identically,
H
W
≡0 in B.
These immersions of minimal surface type are called Fminimal surfaces.
Example: Immersions with prescribed weighted mean curvature
We consider again two examples.
1. Immersions of constant mean curvature
First let
__
B
_
F(X
u
X
v
) +
2h
0
3
X (X
u
X
v
)
_
dudv
with a real constant h
0
∈ R. It generalizes the variational problem for surfaces
with constant mean curvature by substituting [Z[ with a mapping F(Z). Now let
us denote the integrand by
¯
F(u, v). Then we calculate
¯
F
X
(X, Z) =
2h
0
3
Z,
¯
F
Z
(X, Z) = F
Z
(Z) +
2h
0
3
X
as well as
¯
F
XX
(X, Z) = 0,
¯
F
XZ
(X, Z) =
2h
0
3
E
3
,
¯
F
ZZ
(X, Z) = F
ZZ
(Z).
11.4 Principal curvatures and weighted mean curvature 227
Thus critical points X : B →R
3
have weighted mean curvature
H
W
(N) =
h
0
_
det F
ZZ
(N)
∈
_
h
0
M
2
,
h
0
M
1
_
w.r.t. the weight matrix W(Z) from Theorem 11.2. In particular, they satisfy
∇
ds
2
W
(X, N) =−
2h
0
_
det F
ZZ
(N)
in B.
2. Immersions of prescribed mean curvature type
In generalization, we now consider the Lagrangian density
¯
F(X, X
u
X
v
) = F(X
u
X
v
) +Q(X) (X
u
X
v
).
Analogously to the foregoing calculations we compute
H
W
(X, N) =
1
_
det F
ZZ
(N)
div
X
Q(X)
for the weighted mean curvature w.r.t. the above weight matrix W(Z) ∈ R
33
for
cricital points X : B →R
3
of the associated variational problem.
11.4 Principal curvatures and weighted mean curvature
From the elliptic mean curvature type system
∇
ds
2
W
(X, N) =−2H
W
(X, N) in B
w.r.t. some weight matrix W(X, Z) we want to deduce a relation between the
weighted mean curvature H
W
(X, Z) and the principal curvatures κ
1
and κ
2
of an
immersion X.
In the nonweighted case these curvatures satisfy
κ
1
+κ
2
= 2H(X, N)
with a prescribed mean curvature H(X, Z).
In the general case it turns that out that the principal curvatures have to be multi
plied with special weight factors to sufﬁce a curvature relation like this.
A mean curvature relation
Our next result does not only apply to critical points of variational problems. Rather
we formulate it for any immersion satisfying the above differential equation.
228 11 Crystalline functionals in R
3
Theorem 11.3. Let the immersion X : B → R
3
solve the nonlinear elliptic mean
curvature type system
∇
ds
2
W
(X, N) =−2H
W
(X, N)
with prescribed weighted mean curvature
H
W
: R
3
R
3
¸ ¦0¦ −→R
w.r.t. a weight matrix W(X, Z). Consider a point w ∈ B, and let Z
1
(w) and Z
2
(w)
denote the principal curvature directions of the surface X at this point. Set
ρ
1
(X, N):=Z
1
(w) ◦ W(X, N) ◦ Z
1
(w),
ρ
2
(X, N):=Z
2
(w) ◦ W(X, N) ◦ Z
2
(w)
using principal curvature parameters around this point. Then it holds
ρ
1
(X, N)κ
1
(w) +ρ
2
(X, N)κ
2
(w) = 2H
W
(X, N) for all w ∈ B. (11.5)
On the principal curvatures
Before we come to the proof of this result we want to recall some important proper
ties of principal curvature parametrizations.
For this purpose consider a point (u, v) ∈
˚
B and the circle S
1
(X(u, v)) of radius 1
and with center X(u, v) on the tangential plane T
X
(u, v).
Deﬁnition 11.3. The vector Z ∈S
1
(X(u, v)) is called a principal curvature direction
at X(u, v) if the mapping
Z =∂X(u, v) ◦
_
_
_
_
du
dt
dv
dt
_
_
_
_
−→L
11
_
du
dt
_
2
+2L
12
du
dt
dv
dt
+L
22
_
dv
dt
_
2
takes a stationary value. The smooth curve c(t) :=X(u(t), v(t)) is called a principal
curvature line if it holds
d
dt
c(t) ,= 0
and if
¸
¸
¸
¸
d
dt
c(t)
¸
¸
¸
¸
−1
d
dt
c(t)
represents a principal curvature direction.
A plane normal section of the surface at X(w
0
) along the tangential line of a
curve c = c(t) on that surface, attached at the point X(w
0
), is the projection of this
curve onto the intersecting plane. The curvature of this new plane curve is called
normal curvature at X(w
0
).
11.4 Principal curvatures and weighted mean curvature 229
There are always two linearly independent principal curvature directions where
the normal curvature takes its maximal and minimal value on the compact set
S
1
(X(w
0
)). These quantities are exactly the principal curvatures.
Now let κ
1
(u, v) ,= κ
2
(u, v) at some (u, v) ∈ B. Following Klingenberg [106],
Theorem 3.6.6, we can introduce new parameters in the neighborhood of (u, v) ∈ B
such that the new parameter lines agree with the principal curvature directions there.
Proposition 11.2. Under these conditions there hold the equations of Rodrigues
N
u
(u, v) =−κ
1
(u, v)X
u
(u, v), N
v
(u, v) =−κ
2
(u, v)X
v
(u, v) (11.6)
as well as
X
u
(u, v) N
v
(u, v) = 0, X
v
(u, v) N
u
(u, v) = 0.
In particular, the second fundamental form is diagonalised.
Now let κ
1
(u, v) = κ
2
(u, v) for a point (u, v) ∈ B. Then all directions on the sur
face, attached at that point, are principal curvature directions since there is only one
normal curvature. Such points are called umbilical points.
The next proposition can be found in Laugwitz [111].
Proposition 11.3. At each point there are two linearly independent and orthogo
nal principal curvature directions where the normal curvatures take their extremal
values.
If the parameter systemis chosen such that the parameter lines touch the principal
curvature lines at some point X(u, v) then there hold
L
11
(u, v) =κ
1
(u, v)g
11
(u, v), L
22
(u, v) =κ
2
(u, v)g
22
(u, v) (11.7)
as well as
g
12
(u, v) = 0, L
12
(u, v) = 0 (11.8)
with the coefﬁcients g
i j
and L
i j
of the ﬁrst resp. second fundamental form.
Proof of the mean curvature relation
Let us now come to the proof of the foregoing theorem.
Proof. Start with choosing principal curvature parameters (u, v) locally. We rewrite
the mean curvature type system
∇
ds
2
W
(X, N) =−2H
W
(X, N) in B
into the form
2
∑
i, j=1
L
i j
(u, v)h
i j
(u, v) = 2H
W
(u, v) in B,
h
i j
= X
u
i ◦ W(X, N) ◦ X
u
j the coefﬁcients of the weighted ﬁrst fundamental form.
230 11 Crystalline functionals in R
3
Then using (11.7) and (11.8) we ﬁnd
h
11
L
11
+h
22
L
22
= −
X
v
◦ W(X, N) ◦ X
v
W
2
(N
u
X
u
) −
X
u
◦ W(X, N) ◦ X
u
W
2
(N
v
X
v
)
=
X
v
◦ W(X, N) ◦ X
v
W
2
[X
u
[
2
κ
1
+
X
u
◦ W(X, N) ◦ X
u
W
2
[X
v
[
2
κ
2
=
_
[X
u
[
W
X
v
_
◦ W(X, N) ◦
_
[X
u
[
W
X
v
_
κ
1
+
_
[X
v
[
W
X
u
_
◦ W(X, N) ◦
_
[X
v
[
W
X
u
_
κ
2
with the area element W =[X
u
[[X
v
[. If follows
h
11
L
11
+h
22
L
22
= ([X
v
[
−1
X
v
) ◦ W(X, N) ◦ ([X
v
[
−1
X
v
)κ
1
+([X
u
[
−1
X
u
) ◦ W(X, N) ◦ ([X
u
[
−1
X
u
)κ
2
=: (Z
1
◦ W(X, N) ◦ Z
1
)κ
1
+(Z
2
◦ W(X, N) ◦ Z
2
)κ
2
= ρ
1
(X, N)κ
1
+ρ
2
(X, N)κ
2
with the orthonormal principal curvature directions
Z
1
:=
1
[X
v
[
X
v
, Z
2
:=
1
[X
u
[
X
u
and the weight factors ρ
1
(X, N) and ρ
2
(X, N) from the theorem. We arrive at
∇
ds
2
W
(X, N) =−
2
∑
i, j=1
h
i j
L
i j
=−ρ
1
(X, N)κ
1
−ρ
2
(X, N)κ
2
=−2H
W
(X, N)
proving the statement. ⊓⊔
Remarks on the weight factors ρ
1
and ρ
2
Together with (W3) from chapter 4 we have
1
1 +ω
0
≤ρ
1
(X, N), ρ
2
(X, N) ≤1 +ω
0
.
In the nonweighted case W(X, Z) ≡E
3
we particularly infer
ρ
1
(X, Z) ≡1, ρ
2
(X, Z) ≡1
and therefore κ
1
+κ
2
= 2H with the classical scalar mean curvature.
11.5 Nonparametric differential equations 231
We also infer
κ
1
(u, v)κ
2
(u, v) = K(u, v) ≤0 in B
for the Gaussian curvature K of a weighted minimal surface satisfying H
W
≡0.
We want to mention that the curvature relation
ρ
1
(X, N)κ
1
+ρ
2
(X, N)κ
2
= 0 in B
for Fminimal surfaces critical for the anisotropic but homogeneous functional
__
B
F(X
u
X
v
)dudv
can already be found in Sauvigny [141]. From there we also took the methods pre
sented before for deriving the general identities including also inhomogeneous vari
ational problems.
11.5 Nonparametric differential equations
Nonparametric variational problems
In section 9.2 we have already computet the EulerLagrange equations
dF
p
σ
(x, y, ζ, ∇ζ)
dx
+
dF
q
σ
(x, y, ζ, ∇ζ)
dy
= F
z
σ
(x, y, ζ, ∇ζ),
where σ = 1, . . . , n, for nonparametric functionals of the form
F[ζ] =
__
Ω
F(x, y, ζ, ∇ζ)dxdy
witht the setting
ζ = (ζ
1
, . . . , ζ
n
).
If F equals the area integrand, i.e.
A[ζ] =
__
Ω
_
1 + p
2
+q
2
+ p
2
q
2
−(p q)
2
dxdy
with p :=ζ
x
, q :=ζ
y
, we arrive at the nonparametric minimal surface system
div
(p
σ
, q
σ
)
W
=−div
_
p
σ
[q[
2
−q
σ
(p q), q
σ
[p[
2
−p
σ
(p q)
_
W
in Ω
for σ = 1, . . . , n.
232 11 Crystalline functionals in R
3
The right hand side in this system vanishes identically in case n = 1 and gives us
the nonparametric minimal surface equation in one codimension
div
∇ζ
W
= 0 in Ω.
Let us consider the more general integrand
F(x, y, z, p, q) =Γ(x, y, z)
_
1 + p
2
+q
2
+ p
2
q
2
−(p q)
2
where z = (z
1
, . . . , z
n
) etc.
Corollary 11.1. Critical points (x, y, ζ
1
, . . . , ζ
n
) of the functional
__
Ω
Γ(x, y, ζ)W dxdy
satisfy the EulerLagrange system
div
(p
σ
, q
σ
)
W
= 2H(X, N
σ
)
_
1 + p
2
σ
+q
2
σ
W
+
1
ΓW
_
_
p
2
+q
2
+ p
2
q
2
−(p q)
2
_
Γ
z
σ
−
n
∑
ω=1
(p
σ
p
ω
+q
σ
q
ω
)Γ
z
ω
_
−
1
Γ
div
_
p
σ
q
2
−q
σ
(p q)
W
Γ,
q
σ
p
2
−p
σ
(p q)
W
Γ
_
in Ω for σ = 1, . . . , n, with the mean curvature ﬁeld
H(X, N
σ
) :=
Γ
X
(x, y, z) N
σ
2Γ(x, y, z)
w.r.t. to the Euler unit normal vectors
N
1
:=
1
_
1 +[∇ζ
1
[
2
(−ζ
1,x
, −ζ
1,y
, 1, 0, 0, . . . , 0, 0),
N
2
:=
1
_
1 +[∇ζ
2
[
2
(−ζ
2,x
, −ζ
2,y
, 0, 1, 0, . . . , 0, 0),
.
.
.
N
n
:=
1
_
1 +[∇ζ
n
[
2
(−ζ
n,x
, −ζ
n,y
, 0, 0, 0, . . . , 0, 1).
11.5 Nonparametric differential equations 233
The second and the third rows in this system vanish in the case n = 1. Then we
arrive at the single nonlinear elliptic parametric meancurvature equation
div
∇ζ
W
= 2H(X, N) in Ω.
The proof follows from our considerations from chapter 10.
Representation of quasilinear equations in divergence form due to Bers
Now let again n = 1. Nonparametric variational problems
__
Ω
F(ζ
x
, ζ
y
)dxdy
lead to EulerLagrange equations in divergence form
d
dx
F
p
(ζ
x
, ζ
y
) +
d
dy
F
q
(ζ
x
, ζ
y
) = 0 in Ω.
We now follow Bers [14] and consider quasilinear equations
A(p, q)r +2B(p, q)s +C(p, q)t = 0 in Ω
with the additional settings
r :=ζ
xx
, s :=ζ
xy
, t :=ζ
yy
.
Assume that the coefﬁcents A, B and C as well as its ﬁrst derivatives are H¨ older
continuous in a [p, q]domain containing the origin (0, 0).
Now Bers proved that under these conditions we can always rewrite the above
quasiliner elliptic equation in divergence form
dλ(p, q)
dx
+
dµ(p, q)
dy
= 0.
Necessary and sufﬁcient for this is the existence of nonvanishing and positive func
tion χ = χ(p, q) with the properties
λ
p
= χA, λ
p
+µ
q
= 2χB, µ
q
= χC.
Then the complexvalued function κ :=λ +iµ is a solution of the system
λ
p
=
A
C
µ
q
, λ
q
=−µ
p
+
2B
C
µ
q
.
234 11 Crystalline functionals in R
3
Proposition 11.4. (Bers [14])
Under the above assumptions there exists a continuously differentiable homeomor
phism κ =λ +iµ solving the foregoing system such that
λ
p
> 0,
λ = µ =λ
p
= 0, λ
p
= 1 in p = q = 0,
pλ +qµ > 0 for p
2
+q
2
> 0,
λ
p
µ
q
−λ
q
µ
p
> 0.
On the size of quasilinear graphs
Consider now the divergence equation
d
dx
F
p
(x, y, ζ, ζ
x
, ζ
y
) +
d
dy
F
q
(x, y, ζ, ζ
x
, ζ
y
) = F
z
(x, y, ζ, ζ
x
, ζ
y
) (11.9)
on the closed disc
B
R
:=
_
(x, y) ∈ R
2
: x
2
+y
2
≤R
2
_
.
Theorem 11.4. Let ζ ∈C
2
(
˚
B
R
, R) ∩C
1
(B
R
, R) be a solution of (11.9). Suppose that
F
p
(x, y, z, p, q)
2
+F
q
(x, y, z, p, q)
2
≤k
2
0
,
F
z
(x, y, z, p, q) ≥F
min
for all (x, y, z, p, q) ∈ R
2
RRR
with real constants k
0
∈ [0, +∞) and F
min
∈ (0, +∞). Then it holds
R ≤
2k
0
F
min
.
Proof. Partial integration of the divergence equation yields
__
B
R
F
z
dxdy =
__
B
R
div(F
p
, F
q
)dxdy =
_
∂B
R
(−F
q
dx +F
p
dy).
Now on the one hand we know
F
min
πR
2
≤
__
B
R
F
z
dxdy,
11.5 Nonparametric differential equations 235
while on the other hand it holds
_
∂B
R
(−F
q
dx +F
q
dy) ≤
_
∂B
R
_
F
2
p
+F
2
q
_
dx
2
+dy
2
≤2πk
0
R.
Comparing both inequalities proves the statement. ⊓⊔
Our method follows the lines of Heinz [79] where graphs with prescribed scalar
mean curvature are considered.
Quasilinear elliptic equations and immersions of mean curvature type
Let ζ ∈C
3+α
(Ω, R) be a solution of the differential equation
A(x, y, ζ, ∇ζ)ζ
xx
+2B(x, y, ζ, ∇ζ)ζ
xy
+C(x, y, ζ, ∇ζ)ζ
yy
= D(x, y, ζ, ∇ζ) (11.10)
in Ω where w.l.o.g. we assume
Λ
2
:= AC−B
2
= 1 +[∇ζ[
2
.
The unit normal vector of such a graph X(x, y) = (x, y, ζ(x, y)) reads as
N(x, y) =
1
_
1 +[∇ζ[
2
(−ζ
x
, −ζ
y
, 1).
Following ideas of Sauvigny [141] we want to construct a weight matrix W(X, N)
which transforms (11.10) int a system of mean curvature type
∇
ds
2
W
(X, N) =−2H
W
(X, N).
Construction of the weight matrix
We start with
M(x, y) :=
_
_
_
_
_
_
1 +ζ
2
x
ζ
x
ζ
y
ζ
x
ζ
y
1 +ζ
2
y
_
−1
◦
_
X
x
X
y
_
N(x, y)
_
_
_
_
_
and deﬁne the weight matrix
W(X, N) := M(x, y)
T
◦
_
_
_
C −B 0
−B A 0
0 0 1
_
_
_◦ M(x, y). (11.11)
Let us call the matrix in the middle of the right hand side as K = K(x, y).
236 11 Crystalline functionals in R
3
Lemma 11.2. It holds
det W(X, N) = 1.
Proof. We compute
_
1 +ζ
2
x
ζ
x
ζ
y
ζ
x
ζ
y
1 +ζ
2
y
_
−1
◦
_
1 0 ζ
x
0 1 ζ
y
_
=
1
Λ
2
_
1 +ζ
2
y
−ζ
x
ζ
y
ζ
x
−ζ
x
ζ
y
1 +ζ
2
x
ζ
y
_
,
and therefore
M(x, y) =
1
Λ
2
_
_
_
1 +ζ
2
y
−ζ
x
ζ
y
ζ
x
−ζ
x
ζ
y
1 +ζ
2
x
ζ
y
−Λζ
x
−Λζ
y
Λ
_
_
_. (11.12)
Its determinant reads
det M(x, y) =
1
Λ
6
Λ(1 +ζ
2
x
+ζ
2
y
)
2
=
1
Λ
5
Λ Λ
4
=
1
Λ
.
Considering
det W(X, N) = det M(x, y) det K(x, y) det M(x, y) =Λ
−1
Λ
2
Λ
−1
= 1
the statement follows. ⊓⊔
Lemma 11.3. It holds
W(X, N) ◦ N = N.
Proof. Using (11.12) we compute
1
Λ
M(x, y) ◦ (−ζ
x
, −ζ
y
, 1) = (0, 0, 1)
as well as
1
Λ
K(x, y) ◦ M(x, y) ◦ (−ζ
x
, −ζ
y
, 1) = K(x, y) ◦ (0, 0, 1) = (0, 0, 1),
and the statement follows from (11.11). ⊓⊔
Lemma 11.4. It holds
_
C −B
−B A
_
=∂X
T
◦ W(X, N) ◦ ∂X. (11.13)
11.5 Nonparametric differential equations 237
Proof. Since
M(x, y) ◦ ∂X =
1
Λ
2
_
_
_
1 +ζ
2
y
−ζ
x
ζ
y
ζ
x
−ζ
x
ζ
y
1 +ζ
2
x
ζ
y
−Λζ
x
−Λζ
y
Λ
_
_
_◦
_
_
_
1 0
0 1
ζ
x
ζ
y
_
_
_=
_
_
_
1 0
0 1
0 0
_
_
_
we arrive at
_
1 0 0
0 1 0
_
◦
_
_
_
C −B 0
−B A 0
0 0 1
_
_
_◦
_
_
_
1 0
0 1
0 0
_
_
_=
_
C −B
−B A
_
proving the statement. ⊓⊔
Transformation into a weighted mean curvature system
We now deﬁne the weighted metrical coefﬁcients
h
11
(x, y) :=C, h
12
(x, y) :=−B, h
22
(x, y) := A (11.14)
and calculate
h
11
(x, y) =
A
Λ
2
, h
12
(x, y) =
B
Λ
2
, h
22
(x, y) =
C
Λ
2
.
Thus we can rewrite the differential equation in the form
h
11
(X
xx
N) +2h
12
(X
xy
N) +h
22
(X
yy
N) =
2
∑
i, j=1
h
i j
L
i j
=
_
1 +[∇ζ[
2
_
−
3
2
D.
For the weighted mean curvature we set
H
W
(X, N) :=
1
2
_
1 +[∇ζ[
2
_
−
3
2
D(x, y, ζ, ∇ζ). (11.15)
Theorem 11.5. With the settings (11.11), (11.14) and (11.15), any solution ζ ∈
C
3+α
(Ω, R) of the quasilinear elliptic equation (11.10) can be transformed into
the weighted mean curvature system
∇
ds
2
W
(X, N) =−2H
W
(X, N) in Ω.
An analogous connection was already established in Sauvigny [141] for the ho
mogeneous case of vanishing weighted mean curvature with D ≡0.
238 11 Crystalline functionals in R
3
Ellipticity of the weight matrix
Taking account of (11.13), the ellipticity constant ω
0
≥0 can be realized as follows
(again we refer to Sauvigny [141]):
1
1 +ω
0
≤
ξ ◦
_
C −B
−B A
_
◦ ξ
ξ ◦
_
1 +ζ
2
x
ζ
x
ζ
y
ζ
x
ζ
y
1 +ζ
2
y
_
◦ ξ
≤1 +ω
0
(11.16)
for all ξ ∈ R
2
¸ ¦0¦.
Agrowth estimate of the form(11.16) can already be found in works of Bernstein
and Finn (see e.g. the analysis in Nitsche [126], §575 ff.). Namely with
ξ = (ξ
1
, ξ
2
), p =ζ
x
and q =ζ
y
we consider the two metrical elements
ds
2
1
:= (1 + p
2
)dξ
2
1
+2pqdξ
1
dξ
2
+(1 +q
2
)dξ
2
2
,
ds
2
2
:= Cdξ
2
1
−2Bdξ
1
dξ
2
+Adξ
2
2
.
If (11.16) holds true, the elements ds
2
1
and ds
2
2
are said to be in quasiconformal
relation which means that for their dilatation δ satisﬁes
1
(1 +ω
0
)
2
≤δ :=
_
ds
2
ds
1
_
max
_
ds
1
ds
2
_
min
≤(1 +ω
0
)
2
.
Following Finn [58] we speak of a graph of minimal surface type if D ≡0, or of
graphs of mean curvature type in the nonhomogeneous case.
Let us ﬁnally consider the example of the mean curvature equation
(1 +ζ
2
y
)ζ
xx
−2ζ
x
ζ
y
ζ
xy
+(1 +ζ
2
x
)ζ
yy
= 2H(x, y, ζ, ζ
x
, ζ
y
)(1 +[∇ζ[
2
)
3
2
to illustrate the calculations so far.
11.6 Quasilinear elliptic systems 239
For the weight matrix from (11.11) we compute
1
Λ
4
_
_
_
1 +ζ
2
y
−ζ
x
ζ
y
−Λζ
x
−ζ
x
ζ
y
1 +ζ
2
x
−Λζ
y
ζ
x
ζ
y
Λ
_
_
_◦
_
_
_
1 +ζ
2
x
ζ
x
ζ
y
0
ζ
x
ζ
y
1 +ζ
2
y
0
0 0 1
_
_
_◦
_
_
_
1 +ζ
2
y
−ζ
x
ζ
y
ζ
x
−ζ
x
ζ
y
1 +ζ
2
x
ζ
y
−Λζ
x
−Λζ
y
Λ
_
_
_
=
1
Λ
4
_
_
_
1 +ζ
2
y
−ζ
x
ζ
y
−Λζ
x
−ζ
x
ζ
y
1 +ζ
2
x
−Λζ
y
ζ
x
ζ
y
Λ
_
_
_◦
_
_
_
Λ
4
0 Λ
2
ζ
x
0 Λ
4
Λ
2
ζ
y
−Λζ
x
−Λζ
y
Λ
_
_
_
=
1
Λ
4
_
_
_
Λ
4
0 0
0 Λ
4
0
0 0 Λ
4
_
_
_,
i.e. W(X, N) ≡E
3
, and we may choose ω
0
= 0.
11.6 Quasilinear elliptic systems
In this section we want to derive various elliptic systems for the surface vector X of
a weighted conformally parametrized immersion of mean curvature type and for
its spherical mapping N. Based on these identities we particularly elaborate the
quadratic growth of the gradients in the next section which becomes important for
the later curvature estimates.
An example: Minimal surfaces
For illustration we want to start with considering a minimal surface X : B →R
3
satisfying the conformality relation
X
2
u
=W = X
2
v
, X
u
X
v
= 0 in B.
Let N: B →R
3
denote the Gauss map of the minimal surface. From the invariant
representation
∇
ds
2 (X, N) = 0 in B
w.r.t. the nonweighted metric ds
2
= g
11
du
2
+2g
12
dudv +g
22
dv
2
we infer
∆X N = 0 in B.
240 11 Crystalline functionals in R
3
Furthermore, differentiating the conformality relations gives us
X
u
X
uu
= X
v
X
uv
, X
u
X
uv
= X
v
X
vv
,
X
u
X
uv
+X
v
X
uu
= 0, X
u
X
vv
+X
v
X
uv
= 0
from where we get
X
u
X
uu
=−X
u
X
vv
, X
v
X
uu
=−X
v
X
vv
.
Thus there also hold
∆X X
u
= 0, ∆X X
v
= 0,
and we arrive at the conformally minimal surface system
∆X = 0 in B.
In particular, the coordinate functions x
i
=x
i
(u, v), i =1, 2, 3, represent harmonic
functions.
Weighted minimal surfaces
From (4.11) we already know the representation
∆
W
X =∆X −(Ω
1
11
+Ω
1
22
)X
u
−(Ω
2
11
+Ω
2
22
)X
v
= 2H
W
(X, N)WN
using weighted conformal parameters (u, v) ∈ B setting
Ω
k
i j
=−
1
2
2
∑
ℓ=1
h
kℓ
(ω
ℓi j
+ω
jℓi
−ω
i jℓ
), ω
i jℓ
= X
u
i ◦ W(X, N)
u
ℓ ◦ X
u
j
with the coefﬁcients
h
i j
= X
u
i ◦ W(X, N) ◦ X
u
j
of the weighted ﬁrst fundamental form. In the nonweighted case W(X, N) = E
3
this system reduces to the classical minimal surface system ∆X = 0.
Elliptic systems for the spherical mapping
The differential system for the surface vector X from the last example contains the
spherical mapping N and its ﬁrst derivatives. Thus to arrive at a complete system of
equations we must establish suitable differential equations for the vector N.
Theorem 11.6. Let the weighted conformally parametrized immersion X : B →R
3
with prescribed weighted mean curvature H
W
(X, Z) be given.
11.6 Quasilinear elliptic systems 241
Then its spherical mapping N: B →R
3
satisﬁes
∆N = 2(N
u
N
v
) +W(X, N)
u
◦ (NN
v
) +W(X, N)
v
◦ (N
u
N)
−2
_
H
W,X
(X, N) X
u
+H
W,Z
(X, N) N
u
_
W(X, N) ◦ X
u
−2
_
H
W,X
(X, N) X
v
+H
W,Z
(X, N) N
v
_
W(X, N) ◦ X
v
−2H
W
(X, N)
_
W(X, N)
u
◦ X
u
+W(X, N)
v
◦ X
v
_
−2H
W
(X, N)W(X, N) ◦ ∆X .
(11.17)
For the proof we need the following calculus rule (see Sauvigny [141]).
Lemma 11.5. Using weigthed conformal parameters (u, v) ∈ B it holds
W(X, N) ◦ X
u
= (X
v
N), W(X, N) ◦ X
v
= (N X
u
). (11.18)
Proof of the Lemma. The vector triple
_
1
√
W
W(X, N)
1
2
◦ X
u
,
1
√
W
W(X, N)
1
2
◦ X
v
, N
_
forms an orthonormal moving frame of the surface. It particularly holds
W(X, N)
1
2
◦ X
v
= N W(X, N)
1
2
◦ X
u
.
Using the identity W(X, N) ◦ N = N and the rule from Lemma 5.3 from chapter 5
we compute
W(X, N)
1
2
◦ X
v
=
_
W(X, N)
1
2
◦ N
_
_
W(X, N)
1
2
◦ X
u
_
= W(X, N)
−
1
2
◦ (NX
u
)
proving the second identity in (11.18). Furthermore we have
X
u
= (N X
u
) N =
_
W(X, N) ◦ X
v
_
_
W(X, N) ◦ N
_
= W(X, N)
−1
◦ (X
v
N),
and that was stated. ⊓⊔
Now we come to the
242 11 Crystalline functionals in R
3
Proof of the Theorem. We write the Weingarten equations from section 4.3 using
weighted conformal parameters (u, v) ∈ B in the form
N
u
= −
L
11
W
W(X, N) ◦ X
u
−
L
12
W
W(X, N) ◦ X
v
,
N
v
= −
L
21
W
W(X, N) ◦ X
u
−
L
22
W
W(X, N) ◦ X
v
.
Together with (11.18) we calculate
N N
u
= −
L
11
W
N W(X, N) ◦ X
u
−
L
12
W
N W(X, N) ◦ X
v
= −
L
11
W
N (X
v
N) −
L
12
W
N (NX
u
)
= −
L
11
W
X
v
+
L
12
W
X
u
(11.19)
and analogously
N N
v
=−
L
12
W
X
v
+
L
22
W
X
u
. (11.20)
We conclude
N N
u
=
L
12
W
X
u
+
L
22
W
X
v
−
L
11
+L
22
W
X
v
= −W(X, N)
−1
◦ N
v
−2H
W
(X, N)X
v
and
N N
v
= −
L
12
W
X
v
−
L
11
W
X
u
+
L
11
+L
22
W
X
u
= W(X, N)
−1
◦ N
u
+2H
W
(X, N)X
u
.
Rearranging yields
N
u
= W(X, N) ◦ (NN
v
) −2H
W
(X, N)W(X, N) ◦ X
u
,
N
v
= −W(X, N) ◦ (NN
u
) −2H
W
(X, N)W(X, N) ◦ X
v
.
Now differentiate these relations to get
N
uu
= W(X, N)
u
◦ (N N
v
) +W(X, N) ◦ (N
u
N
v
)
+W(X, N) ◦ (NN
uv
) −2H
W
(X, N)W(X, N)
u
◦ X
u
−2H
W
(X, N)W(X, N) ◦ X
uu
−2H
W
(X, N)
u
W(X, N) ◦ X
u
11.6 Quasilinear elliptic systems 243
as well as
N
vv
= −W(X, N)
v
◦ (NN
u
) −W(X, N) ◦ (N
v
N
u
)
−W(X, N) ◦ (NN
uv
) −2H
W
(X, N)W(X, N)
v
◦ X
v
−2H
W
(X, N)W(X, N) ◦ X
vv
−2H
W
(X, N)
v
W(X, N) ◦ X
v
.
Summing up both identities proves the statement. ⊓⊔
We want to evaluate the normal part of the term
W(X, N)
u
◦ (NN
v
) +W(X, N)
v
◦ (N
u
N).
Let as usual [X,Y, Z] := X (Y Z). Then after differentiation of N = W(X, N) ◦ N
we compute
N ◦ W(X, N)
u
◦ (NN
v
) +N◦ W(X, N)
v
◦ (N
u
N)
= (N N
v
) ◦ W(X, N)
u
◦ N+(N
u
N) ◦ W(X, N)
v
◦ N
= (N N
v
) N
u
−(NN
v
) ◦ W(X, N) ◦ N
u
+(N
u
N) N
v
−(N
u
N) ◦ W(X, N) ◦ N
v
= −2[N, N
u
, N
v
]
−
_
W(X, N) ◦ N, N
u
, N
v
¸
+
_
W(X, N) ◦ N, N
u
, N
v
¸
+
_
N, W(X, N) ◦ N
u
, N
v
¸
+
_
N, N
u
, W(X, N) ◦ N
v
¸
=
_
traceW(X, N) −3
_
[N, N
u
, N
v
]
where [N, N
u
, N
v
] = KW. Thus this part vanishes identically in the nonweighted
case W(X, Z) ≡E
3
.
An alternative elliptic system for the spherical mapping
The identites from(11.18) enable us to derive an elliptic system for X : Differentiate
X
u
= W(X, N)
−1
◦ (X
v
N), X
v
= W(X, N)
−1
◦ (NX
u
)
to arrive at
X
uu
=
_
W(X, N)
−1
_
u
◦ (X
v
N) +W(X, N)
−1
◦ (X
uv
N+X
v
N
u
),
X
vv
=
_
W(X, N)
−1
_
v
◦ (NX
u
) +W(X, N)
−1
◦ (N
v
X
u
+NX
uv
).
244 11 Crystalline functionals in R
3
and summing up these identities shows that
∆X =
_
W(X, N)
−1
_
u
◦ (X
v
N) +
_
W(X, N)
−1
_
v
◦ (NX
u
)
+N
v
X
u
+X
v
N
u
.
The third term on the right hand side does not involves any tangential parts. Thus
from (11.19) and (11.20) we infer
N (X
u
N
v
+N
u
X
v
) = −X
u
(N N
v
) −X
v
(N
u
N)
= −
L
22
g
22
−2L
12
g
12
+L
11
g
11
W
= −2HW
with the nonweighted mean curvature H = H(u, v) – but using weighted conformal
parameters. Together with (11.18) we conclude
∆X =
_
W(X, N)
−1
_
u
◦ W(X, N) ◦ X
u
+
_
W(X, N)
−1
_
v
◦ W(X, N) ◦ X
v
+2HWN
= −W(X, N)
−1
◦ W(X, N)
u
◦ X
u
−W(X, N)
−1
◦ W(X, N)
v
◦ X
v
+2HWN
or after rearranging
W(X, N) ◦ ∆X =−W(X, N)
u
◦ X
u
−W(X, N)
v
◦ X
v
+2HWN.
This could eventually serve as an alternative systemfor the surface vector X. Fur
thermore, in the nonweighted case W(X, Z) ≡ E
3
we immediately read the mean
curvature system
∆X = 2HWN
in conformal parameters.
Inserting next this representation of W(X, N) ◦ ∆X into (11.17), taking account
of N
u
N
v
= KWN, proves the
Corollary 11.2. Let the weighted conformally parametrized immersion X : B →R
3
with prescribed weighted mean curvature H
W
(X, Z) be given. Then its spherical
mapping satisﬁes
∆N = −2
_
2HH
W
(X, N) −K
_
WN
+W(X, N)
u
◦ (NN
v
) +W(X, N)
v
◦ (N
u
N)
−2
_
H
W,X
(X, N) X
u
+H
W,Z
(X, N) N
u
_
W(X, N) ◦ X
u
−2
_
H
W,X
(X, N) X
v
+H
W,Z
(X, N) N
v
_
W(X, N) ◦ X
v
(11.21)
with the nonweighted mean curvature H.
11.6 Quasilinear elliptic systems 245
Examples
Let us exemplarily consider the case W(X, Z) = E
3
. Using conformal parameters
we infer
∆N = −2
_
2H(X, N)
2
−K
_
WN
−2
_
H
X
(X, N) X
u
+H
Z
(X, N) N
u
_
X
u
−2
_
H
X
(X, N) X
v
+H
Z
(X, N) N
v
_
X
v
.
Note the coupling with the gradient of the surface vector.
If X : B →R
3
represents a minimal surface then it holds
∆N = 2KWN = 2(N
u
N
v
).
We want to express ∆N in terms of the moving frame ¦N, N
u
, N
v
¦. Using confor
mal parameters we have
H
X
=
_
H
X
X
u
[X
u
[
_
X
u
[X
u
[
+
_
H
X
X
v
[X
v
[
_
X
v
[X
v
[
+
_
H
X
N
_
N
=
1
W
_
_
H
X
X
u
_
X
u
+
_
H
X
X
v
_
X
v
_
+
_
H
X
N
_
N
w.r.t. the basis ¦X
u
, X
v
, N¦ such that it follows
∆N = 2
_
K−2H(X, N)
2
+H
X
(X, N) N
_
WN −2H
X
(X, N)W
−2
_
H
Z
(X, N) N
u
_
X
u
−2
_
H
Z
(X, N) N
v
_
X
v
.
(11.22)
Now compare the terms
(H
Z
N
u
)X
u
+(H
Z
N
v
)X
v
= −
L
11
W
(H
Z
X
u
)X
u
−
L
12
W
(H
Z
X
v
)X
u
−
L
12
W
(H
Z
X
u
)X
v
−
L
22
W
(H
Z
X
v
)X
v
,
(H
Z
X
u
)N
u
+(H
Z
X
v
)N
v
= −
L
11
W
(H
Z
X
u
)X
u
−
L
12
W
(H
Z
X
u
)X
v
−
L
12
W
(H
Z
X
v
)X
u
−
L
22
W
(H
Z
X
v
)X
v
showing the identity
(H
Z
N
u
)X
u
+(H
Z
N
v
)X
v
= (H
Z
X
u
)N
u
+(H
Z
X
v
)N
v
.
246 11 Crystalline functionals in R
3
Thus we can rewrite (11.22) into the form
∆N = 2
_
K−2H(X, N)
2
+H
X
(X, N) N
_
WN−2H
X
(X, N)W
−2
_
H
Z
(X, N) X
u
_
N
u
−2
_
H
Z
(X, N) X
v
_
N
v
.
Finally we want to specify (11.21) for weighted minimal surfaces satisfying
H
W
(X, Z) ≡0 : We immediately infer
∆N = 2KWN +W(X, N)
u
◦ (NN
v
) +W(X, N)
v
◦ (N
u
N).
Note again the coupling with the gradient of the surface vector. This coupling van
ishes in the special case W = W(N).
11.7 Quadratic growth in the gradient
In this section we elaborate the quadratic growth in the gradient for the elliptic
systems from above.
Estimate of the Ω
k
i j
Beside the constant ω
0
∈ [0, +∞) we also need quantities controlling the derivatives
of the weight matrix.
Deﬁnition 11.4. Let ω
1
, ω
2
∈ [0, +∞) be real constants with the properties
¸
¸
¸
_
3
∑
i, j,k=1
w
i j,x
k (X, Z)
2
≤ω
1
as well as
¸
¸
¸
_
3
∑
i, j,k=1
w
i j,z
k (X, Z)
2
≤ω
2
with the components w
i j
(X, Z) of the weight matrix W(X, Z).
The coupling of the surface vector X and its spherical mapping N through the
weight matrix W(X, N) causes a coupling of the associated elliptic systems. Thus
our next deﬁnition.
Deﬁnition 11.5. With a real number r > 0 we set
X(u, v) :=
_
1
r
X(u, v), N(u, v)
_
, (u, v) ∈ B.
11.7 Quadratic growth in the gradient 247
In terms of this system we want to establish estimates for the weight matrix.
Lemma 11.6. Let some vector ξ ∈ R
3
be given. Then it holds
[ W(X, Z)
u
m ◦ ξ[ ≤[ξ[
_
2(rω
1
)
2
+2ω
2
2
[X
u
m[, m = 1, 2. (11.23)
Proof. For ξ = (ξ
1
, ξ
2
, ξ
3
) we compute
¸
¸
W(X, Z)
u
m ◦ ξ
¸
¸
2
=
3
∑
i, j,k=1
_
w
i j,x
k x
k
u
mξ
j
+w
i j,z
k z
k
u
mξ
j
_
2
≤ 2
3
∑
i, j,k=1
_
_
w
i j,x
k x
k
u
mξ
j
_
2
+
_
w
i j,z
k z
k
u
mξ
j
_
2
_
≤ 2[ξ[
2
3
∑
i, j,k=1
_
_
w
i j,x
k x
k
u
m
_
2
+
_
w
i j,z
k z
k
u
m
_
2
_
(11.24)
due to the H¨ older inequality, for example applied to
_
3
∑
j=1
w
i j,x
k x
k
u
mξ
j
_
2
≤
3
∑
j=1
_
w
i j,x
k x
k
u
m
_
2
3
∑
j=1
(ξ
j
)
2
=[ξ[
2
3
∑
j=1
_
w
i j,x
k x
k
u
m
_
2
.
In this way we arrive at
¸
¸
W(X, Z)
u
m ◦ ξ
¸
¸
2
≤ 2[ξ[
2
[X
u
m[
2
3
∑
i, j,k=1
(w
i j,x
k )
2
+2[ξ[
2
[Z
u
m[
2
3
∑
i, j,k=1
(w
i j,z
k )
2
≤ 2ω
2
1
[ξ[
2
[X
u
m[
2
+2ω
2
2
[ξ[
2
[Z
u
m[
2
≤ 2[ξ[
2
_
(rω
1
)
2
+ω
2
2
_
_
¸
¸
¸
1
r
X
u
m
¸
¸
¸
2
+[Z
u
m[
2
_
≤ 2[ξ[
2
_
(rω
1
)
2
+ω
2
2
_
[X
u
m[
2
(11.25)
proving the statement. ⊓⊔
Lemma 11.7. Using weighted conformal parameters (u, v) ∈ B there hold
[Ω
1
11
[ ≤
(1 +ω
0
)
_
(rω
1
)
2
+ω
2
2
√
2
[X
u
[, [Ω
2
22
[ ≤
(1 +ω
0
)
_
(rω
1
)
2
+ω
2
2
√
2
[X
v
[
248 11 Crystalline functionals in R
3
as well as
[Ω
2
11
[ ≤
(1 +ω
0
)
_
(rω
1
)
2
+ω
2
2
√
2
(2[X
u
[ +[X
v
[),
[Ω
1
22
[ ≤
(1 +ω
0
)
_
(rω
1
)
2
+ω
2
2
√
2
([X
u
[ +2[X
v
[).
Proof. We estimate as follows
[Ω
1
11
[ ≤
1
2W
[X
u
◦ W(X, N)
u
◦ X
u
[ =
[X
u
[
2
2W
¸
¸
¸
¸
X
u
[X
u
[
◦ W(X, N)
u
◦
X
u
[X
u
[
¸
¸
¸
¸
≤
[X
u
[
2
2W
¸
¸
¸
¸
W(X, N)
u
◦
X
u
[X
u
[
¸
¸
¸
¸
,
and together with (11.23) we infer
[Ω
1
11
[ ≤
_
(rω
1
)
2
+ω
2
2
[X
u
[
2
√
2W
[X
u
[ ≤
(1 +ω
0
)
_
(rω
1
)
2
+ω
2
2
√
2
[X
u
[.
Analogously we have
Ω
2
11
=−
1
W
X
u
◦ W(X, N)
u
◦ X
v
+
1
2W
X
u
◦ W(X, N)
v
◦ X
u
and compute
[Ω
2
11
[ ≤
[X
u
[[X
v
[
W
¸
¸
¸
¸
X
u
[X
u
[
◦ W(X, N)
u
◦
X
v
[X
v
[
¸
¸
¸
¸
+
[X
u
[
2
2W
¸
¸
¸
¸
X
u
[X
u
[
◦ W(X, N)
v
◦
X
u
[X
u
[
¸
¸
¸
¸
≤
[X
u
[
2
+[X
v
[
2
2W
_
2(rω
1
)
2
+2ω
2
2
[X
u
[ +
[X
u
[
2
2W
_
2(rω
1
)
2
+2ω
2
2
[X
v
[
≤ (1 +ω
0
)
X
u
◦ W(X, N) ◦ X
u
+X
v
◦ W(X, N) ◦ X
v
2W
_
2(rω
1
)
2
+2ω
2
2
[X
u
[
+(1 +ω
0
)
X
u
◦ W(X, N) ◦ X
u
2W
_
2(rω
1
)
2
+2ω
2
2
[X
v
[,
or summarized
[Ω
2
11
[ ≤
(1 +ω
0
)
_
(rω
1
)
2
+ω
2
2
√
2
_
2[X
u
[ +[X
v
[
_
.
The remaining inequalities are proved in the same way. ⊓⊔
11.7 Quadratic growth in the gradient 249
Estimate of ∆X
From the weighted mean curvature system we immediately infer
[∆X[ ≤([Ω
1
11
[ +[Ω
1
22
[)[X
u
[ +([Ω
2
11
[ +[Ω
2
22
[)[X
v
[ +h
0
[∇X[
2
with the constant
h
0
:= sup
X∈R
3
Z∈R
3
¸¦0¦
[H
W
(X, Z)[.
Here we note
W =
_
h
11
h
22
−h
2
12
=
_
g
11
g
22
−g
2
12
≤
√
g
11
g
22
≤
1
2
[∇X[
2
.
Furthermore we have
[∆(r
−1
X)[ ≤ (1 +ω
0
)
_
2(rω
1
)
2
+2ω
2
2
([X
u
[ +[X
v
[)([(r
−1
X)
u
[ +[(r
−1
X)
v
[)
+(h
0
r)[∇(r
−1
X)[
2
≤ (1 +ω
0
)
_
2(rω
1
)
2
+2ω
2
2
([X
u
[ +[X
v
[)
2
+r(h
0
r)[∇X[
2
.
This proves our next
Theorem 11.7. The weighted conformally parametrized immersion X : B →R
3
sat
isﬁes
[∆(r
−1
X)[ ≤Λ
X
[∇X[
2
in B (11.26)
with the constant
Λ
X
:=
_
2(1 +ω
0
)
_
2(rω
1
)
2
+2ω
2
2
+h
0
r
_
. (11.27)
Estimates of ∆N
First, system (11.17) yields
[∆N[ ≤ [∇N[
2
+
_
2(rω
1
)
2
+2ω
2
2
([X
u
[[N N
v
[ +[X
v
[[N
u
N[)
+2(1 +ω
0
)
_
[H
W,X
(X, N)[[X
u
[ +[H
W,Z
(X, N)[[N
u
[
_
[X
u
[
+2(1 +ω
0
)
_
[H
W,X
(X, N)[[X
v
[ +[H
W,Z
(X, N)[[N
v
[
_
[X
v
[
+2h
0
_
2(rω
1
)
2
+2ω
2
2
([X
u
[[X
u
[ +[X
v
[[X
v
[)
+2h
0
(1 +ω
0
)[∆X[.
(11.28)
250 11 Crystalline functionals in R
3
Next beside h
0
we also introduce real constants
h
1
:= sup
X∈R
3
Z∈R
3
¸¦0¦
[H
W,X
(X, Z)[, h
2
:= sup
X∈R
3
Z∈R
3
¸¦0¦
[H
W,Z
(X, Z)[.
Together with (11.26) we then conclude
[∆N[ ≤ [∇X[
2
+2
_
2(rω
1
)
2
+2ω
2
2
[X
u
[[X
v
[
+2(1 +ω
0
)
_
(h
1
r
2
)[X
u
[
2
+(h
2
r)[X
u
[
2
_
+2(1 +ω
0
)
_
(h
1
r
2
)[X
v
[
2
+(h
2
r)[X
v
[
2
_
+2(h
0
r)
_
2(rω
1
)
2
+2ω
2
2
([X
u
[
2
+[X
v
[
2
)
+2(h
0
r)(1 +ω
0
)
_
2(1 +ω
0
)
_
2(rω
1
)
2
+2ω
2
2
+h
0
r
_
[∇X[
2
.
Theorem 11.8. The spherical mapping N of a weighted conformally parametrized
immersion X : B →R
3
satisﬁes
[∆N[ ≤Λ
N
[∇X[
2
in B (11.29)
with the constant
Λ
N
:=1 +(1 +2h
0
r)
_
2(rω
1
)
2
+2ω
2
2
+2(1 +ω
0
)
_
(h
1
r
2
) +(h
2
r)
_
+2(h
0
r)(1 +ω
0
)
_
2(1 +ω
0
)
_
2(rω
1
)
2
+2ω
2
2
+h
0
r
_
.
(11.30)
Estimate of ∆X
The estimates (11.26) and (11.29) immediately imply
Theorem 11.9. Let the weighted conformally parametrized immersion X : B →R
3
be given. Then it holds
[∆X[ ≤Λ
X
[∇X[
2
=:
√
2(Λ
X
+Λ
N
)[∇X[
2
in B (11.31)
with the constants Λ
X
and Λ
N
from (11.27), (11.30).
Eventually these estimates are not strong enough. We should rather consider spe
cial classes of surfaces.
11.7 Quadratic growth in the gradient 251
Example: Minimal surfaces
From ∆X = 0 and ∆N = 2(N
u
N
v
) we immediately obtain
[∆X[ = 0, [∆N[ ≤[∇N[
2
in B
for a conformally parametrized minimal surface. In this case we have Λ
X
= 0 and
Λ
N
= 1, but we replace (11.29) by (11.28) with ω
1
, ω
2
= 0 and H
W
≡0.
Example: Surfaces with prescribed mean curvature
In this case there hold
[∆(r
−1
X)[ ≤(h
0
r)[∇(r
−1
X)[
2
,
[∆N[ ≤(1 +h
2
r)[∇N[
2
+
_
2(h
2
0
+h
1
)r
2
+h
2
r
_
[∇(r
−1
X)[
2
.
In particulary, for proving the second inequality we use (11.28) as follows
[∆N[ ≤ [∇N[
2
+2h
2
0
[∇X[
2
+2h
1
([X
u
[
2
+[X
v
[
2
) + 2h
2
([X
u
[[N
u
[ +[X
v
[[N
v
[)
= [∇N[
2
+2(h
0
r)
2
[∇(r
−1
X)[
2
+2(h
1
r
2
)[∇(r
−1
X)[
2
+2(h
2
r)([r
−1
X
u
[[N
u
[ +[r
−1
X
v
[[N
v
[).
We set Λ
X
= h
0
r and Λ
N
= 1 +2(h
0
r)
2
+2(h
1
r
2
) +2(h
2
r).
Special weighted minimal surfaces
Consider immersions with vanishing weighted mean curvature H
W
(X, Z) ≡ 0. The
estimates (11.26) and (11.29) imply
[∆(r
−1
X)[ ≤2(1 +ω
0
)
_
2(rω
1
)
2
+2ω
2
2
[∇X[
2
,
[∆N[ ≤
_
1 +
_
2(rω
1
)
2
+2ω
2
2
_
[∇X[
2
.
Let us now assume that the weight matrix has the special structure
W = W(Z).
Examples are Fminimal surfaces which are critical for the anisotropic parametric
functional F[X].
252 11 Crystalline functionals in R
3
For arbitrary ξ ∈ R
3
we replace the estimates (11.24) to (11.25) by
¸
¸
W(Z)
u
m ◦ ξ
¸
¸
2
=
3
∑
i, j,k=1
_
w
i j,z
k z
k
u
mξ
j
_
2
≤ [ξ[
2
3
∑
i, j,k=1
_
w
i j,z
k z
k
u
m
_
2
≤ [ξ[
2
[[Z
u
m[
2
3
∑
i, j,k=1
w
2
i j,z
k
≤ ω
2
2
[ξ[
2
[Z
u
m[
2
.
Furthermore we have
[Ω
1
11
[ ≤
(1 +ω
0
)ω
2
2
[N
u
[, [Ω
2
22
[ ≤
(1 +ω
0
)ω
2
2
[N
v
[,
[Ω
2
11
[ ≤ (1 +ω
0
)ω
2
[N
u
[ +
(1 +ω
0
)ω
2
2
[N
v
[,
[Ω
1
22
[ ≤
(1 +ω
0
)ω
2
2
[N
u
[ +(1 +ω
0
)ω
2
[N
v
[.
(11.32)
We will prove that in this special case W = W(Z) the weighted minimal surface
satisﬁes
[∆(r
−1
X)[ ≤ 2(1 +ω
0
)ω
2
[∇(r
−1
X)[[∇N[,
[∆N[ ≤ (1 +ω
2
)[∇N[
2
(11.33)
using weighted conformal parameters. Namely, ﬁrst we have
[∆X[ ≤ ([Ω
1
11
[ +[Ω
1
22
[)[X
u
[ +([Ω
2
11
[ +[Ω
2
22
[)[X
v
[
≤ (1 +ω
0
)ω
2
([N
u
[ +[N
v
[)([X
u
[ +[X
v
[)
showing the ﬁrst inequality. Next we compute
∆N = 2(N
u
N
v
) +W(N)
u
◦ (NN
v
) +W(N)
v
◦ (N
u
N)
which leads us to
[∆N[ ≤2[N
u
[[N
v
[ +ω
2
[N
u
[[N
v
[ +ω
2
[N
v
[[N
u
[ ≤(1 +ω
2
)[∇N[
2
,
and this proves (11.33). Note that X and N in (11.33) are not coupled.
Example: Crystalline variational problems
We want to consider critical points X : B →R
3
of the parametric functional
__
B
_
F(X
u
X
v
) +
2γ
0
3
X (X
u
X
v
)
_
dudv.
11.8 The geometry of immersions of mean curvature type 253
Note that the weight matrix has the structure W = W(Z), thus H
W
= H
W
(Z) for
the weighted mean curvature. Weighted conformally parameterized critical points
therefore fulﬁll
∆X = (Ω
1
11
+Ω
1
22
)X
u
+(Ω
2
11
+Ω
2
22
)X
v
+2H
W
(N)WN,
∆N = 2(N
u
N
v
) +W(N)
u
◦ (NN
v
) +W(N)
v
◦ (N
u
N)
−2
_
H
W,Z
(N) N
u
_
W(N) ◦ X
u
−2
_
H
W,Z
(N) N
v
_
W(N) ◦ X
v
−2H
W
(N)
_
W(N)
u
◦ X
u
+W(N)
v
◦ X
v
_
−2H
W
(N)W(N) ◦ ∆X
with the weighted mean curvature
H
W
(Z) =
γ
0
_
det F
ZZ
(Z)
.
Furthermore there hold
[∆(r
−1
X)[ ≤ 2(1 +ω
0
)ω
2
[∇(r
−1
X)[[∇N[ +(h
0
r)[∇(r
−1
X)[
2
[∆N[ ≤ 2(1 +ω
0
)(h
0
r)
2
[∇(r
−1
X)[
2
+(1 +ω
2
)[∇N[
2
+2
_
(h
0
r)ω
2
+(1 +ω
0
)(h
2
r) +2(1 +ω
0
)
2
ω
2
(h
0
r)
_
[∇(r
−1
X)[[∇N[.
The second estimate particularly follows from
[∆N[ ≤ 2[N
u
[[N
v
[ +2ω
2
[N
u
[[N
v
[ +2(1 +ω
0
)h
2
([X
u
[[N
u
[ +[X
v
[[N
v
[)
+2h
0
ω
2
([X
u
[[N
u
[ +[X
v
[[N
v
[) +4(1 +ω
0
)
2
ω
2
h
0
[∇X[[∇N[
+2(1 +ω
0
)h
2
0
[∇X[
2
.
11.8 The geometry of immersions of mean curvature type
Recall the elliptic differential operator of second order
∆
W
:=∆ −(Ω
1
11
+Ω
1
22
)
∂
∂u
−(Ω
2
11
+Ω
2
22
)
∂
∂v
with the classical Laplacian ∆. A weighted conformally parametrized immersion
X : B →R
3
with prescribed weighted mean curvature H
W
(X, Z) satisﬁes
∆
W
X = 2H
W
(X, N)WN in B.
254 11 Crystalline functionals in R
3
A geometric maximum principle
This system is the basis of the following
Theorem 11.10. Let the weighted conformally parametrized immersion X : B →R
3
with prescribed weighted mean curvature H
W
(X, Z) be given. Suppose that the
smallness condition
h
0
sup
(u,v)∈B
[X(u, v)[ ≤1
holds true with the constant
h
0
= sup
X∈R
3
Z∈R
3
¸¦0¦
[H
W
(X, Z)[.
Then it holds
∆
W
[X(u, v)[
2
≥0 in B.
In particular, [X(u, v)[ satisﬁes the geometric maximum principle
sup
(u,v)∈B
[X(u, v)[ = sup
(u,v)∈∂B
[X(u, v)[.
Proof. We compute
∆
W
[X[
2
= 2[∇X[
2
+2X ∆X −2(Ω
1
11
+Ω
2
11
)X X
u
−2(Ω
2
11
+Ω
2
22
)X X
v
= 2[∇X[
2
+4H
W
(X, N)WX N
≥ 2
_
1 −h
0
[X[
_
[∇X[
2
.
proving the statement. ⊓⊔
An enclosure theorem for weighted minimal surfaces
As we have seen above, the curvature relation
ρ
1
(X, N)κ
1
+ρ
2
(X, N)κ
2
= 0
with ρ
1
, ρ
2
>0 implies K ≤0 for the Gauß curvature of a weighted minimal surface.
This already implies (see e.g. Sauvigny [141])
Theorem 11.11. Let the weighted minimal surface X : B →R
3
be given. Let fur
thermore K
r
⊂ R
3
be the smallest closed ball containing the boundary curve
X(∂B) ⊂R
3
. Then it holds
X(u, v) ⊂K
r
for all (u, v) ∈ B.
11.9 A curvature estimate 255
Proof. Assume that there is an interior point w
0
∈
˚
B such that X(w
0
) ,∈K
r
. Then due
to compactness reasons there is a second point w
1
∈
˚
B where the surface touches
a sphere S
2
R
⊂ R
3
of radius R > r. But then it holds K(w
1
) > 0 contradicting the
property K ≤0 in B.
Convexhull property for weighted minimal surfaces
From K ≤0 we also infer the much stronger
Theorem 11.12. Let the weighted minimal surface X : B →R
3
be given. Then it
holds
X(B) ⊂convX(∂B)
with convX(∂B) denoting the convex hull of the boundary curve X(∂B) ⊂R
3
.
Proof. Choose an unit normal vector Z ∈ R
3
with [Z[ = 1 and a real number h ∈ R
3
such that
X(∂B) ⊂H
Z,h
:=
_
X ∈ R
3
: Z X −h ≤0
_
.
Now deﬁne the function
Φ(u, v) := Z X(u, v) −h, (u, v) ∈ B.
Then there hold
∆
W
Φ(u, v) = 0 in B, Φ(u, v) ≤0 on ∂B.
The maximum principle says Φ(u, v) ≤0 in B, thus X(B) ⊂H
Z,h
. But note
convX(∂B) =
Z,h
H
Z,h
,
and it follows X(B) ⊂H
Z,h
for all half spaces H
Z,h
. ⊓⊔
These and further maximumprinciples and enclosure theorems for critical points
of anisotropic and inhomogeneous variational problems can be found e.g. in Dierkes
[42], [43], or Clarenz [30].
11.9 A curvature estimate
The previous elliptic system together with Heinz’ and Sauvigny’s methods as brieﬂy
described in section 9.4 enable us to prove the following curvature estimate.
Theorem 11.13. Let the weighted conformally parametrized immersion X : B →R
3
with prescribed weighted mean curvature H
W
(X, Z) be given.
256 11 Crystalline functionals in R
3
We make the following assumptions:
(A1) Assume that X =X(u, v) represents a geodesic disc B
r
(X
0
) of geodesic radius
r > 0 with center X
0
= X(0, 0).
(A2) Assume that the Dirichlet energy growths quadratically as follows
D[X] ≤d
0
r
2
(11.34)
with a real constant d
0
∈ (0, +∞) independent of r.
(A3) For real λ =λ(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r) >0 there exists δ =δ(λ) such that
[X(w) −X(w
1
)[ ≤λ and [N(w) −N(w
1
)[ ≤λ for all w ∈ B
δ
(w
0
)
for arbitrary w
1
∈ B
δ
(w
0
) and w
0
∈
˚
B with B
δ
(w
0
) ⊂⊂
˚
B.
Then it holds the curvature estimate
κ
1
(0, 0)
2
+κ
2
(0, 0)
2
≤
1
r
2
Θ(d
0
, ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r)
+4(1 +ω
0
)
2
(2ω
0
+ω
2
0
)h
0
for the principal curvatures κ
1
and κ
2
of the surface X at the origin (0, 0) ∈ B with
a real constant Θ ∈ (0, +∞).
Remarks
We have already discussed various situations where some of the assumption from
our theorem were already realized. Others are discussed in the following chapter on
the second variation. Let us make now some additional remarks.
1. The structure of the right hand side in (11.34) becomes important for our follow
ing Bernstein type theorems. In fact, our curvature estimate requires only some
positive constant d
0
> 0 bounding the Dirichlet energy D[X].
2. Moduli of continuity for the surface vector can be deduced from geometric max
imum principles as we discussed in this and previous chapters. This especially
includes
(i) the convexhull property for immersions with nonpositive Gauss curvature;
(ii) the enclosure theorems for immersions with nonvanishing weighted or non
weighted mean curvature ﬁelds which are small in the sense of
h
0
sup
(u,v)∈B
[X(u, v)[ < 1.
For these results we refer to the sections 9.3, 10.2 and 11.8.
3. Moduli of continuity for the spherical mapping are discussed in the following
chapter on the second variation by means of special classes of surfaces.
11.9 A curvature estimate 257
An auxiliary result
For proving our curvature estimate we need the following lemma from Sauvigny
[141] which characterizes the socalled plane mapping of a weighted minimal sur
face.
Lemma 11.8. Let the weighted conformally parametrized immersion X : B →R
3
be
given. Then there exists a constant c
∗
=c
∗
(ω
0
) ∈(0, 1) such that the plane mapping
F(u, v) :=
_
x
1
(u, v), x
2
(u, v)
_
, (u, v) ∈ B,
satisﬁes
c
∗
(ω
0
)[∇X(u, v)[
2
≤[∇F(u, v)[
2
≤[∇X(u, v)[
2
in B.
We would like to remark that it holds
1
2
[∇X[
2
≤[∇F[
2
≤[∇X[
2
using conformal parameters.
Proof. Let as usual W(X, Z) = (w
i j
(X, Z))
i, j=1,2,3
. We rewrite the weighted confor
mality relations
3
∑
k,ℓ=1
w
kℓ
x
k
u
x
ℓ
u
=
3
∑
k,ℓ=1
w
kℓ
x
k
v
x
ℓ
v
,
3
∑
k,ℓ=1
w
kℓ
x
k
u
x
ℓ
v
= 0 in B
into the complex form
3
∑
k,ℓ=1
w
kℓ
x
k
w
x
ℓ
w
= 0, w = u +iv, [w[ < 1
with x
w
=
1
2
(x
u
−ix
v
). Namely, we note that
3
∑
k,ℓ=1
w
kℓ
x
k
w
x
ℓ
w
=
1
4
3
∑
k,ℓ=1
w
kℓ
(x
k
u
−ix
k
v
)(x
ℓ
u
−ix
ℓ
v
)
=
1
4
3
∑
k,ℓ=1
(w
kℓ
x
k
u
x
ℓ
u
−w
kℓ
x
k
v
x
ℓ
v
) −
i
2
3
∑
k,ℓ=1
w
kℓ
x
k
u
x
ℓ
v
,
and the right hand side vanishes identically. From the condition
(1 +ω
0
)
−1
[ξ[
2
≤w
kℓ
ξ
k
ξ
ℓ
≤(1 +ω
0
)[ξ[
2
for all ξ ∈ R
3
on the weight matrix we infer
w
kk
≥(1 +ω
0
)
−1
and [w
kℓ
[ ≤c
1
(ω
0
), k, ℓ = 1, 2, 3,
with a real constant c
1
∈ [0, +∞).
258 11 Crystalline functionals in R
3
Now let ε > 0 be given. We compute
(1 +ω
0
)
−1
[x
3
w
[
2
≤[w
33
x
3
w
x
3
w
[ =
¸
¸
¸
¸
¸
3
∑
k,ℓ=1
w
kℓ
x
k
w
x
ℓ
w
−w
33
x
3
w
x
3
w
¸
¸
¸
¸
¸
≤c
1
(ω
0
)
_
[x
1
w
[
2
+[x
2
w
[
2
+2[x
1
w
[[x
2
w
[ +2[x
1
w
[[x
3
w
[ +2[x
2
w
[[x
3
w
[
_
≤
_
2 +
1
ε
_
c
1
(ω
0
)
_
[x
1
w
[
2
+[x
2
w
[
2
_
+2εc
1
(ω
0
)[x
3
w
[
2
.
Thus for arbitrary small ε > 0 we ﬁnd c
2
= c
2
(ω
0
) with the property
[x
3
w
[
2
≤c
2
(ω
0
)([x
1
w
[
2
+[x
2
w
[
2
)
from where we conclude
[∇X[
2
=[∇F[
2
+4[x
3
w
[
2
≤
_
1 +c
2
(ω
0
)
_
[∇F[
2
in B.
The statement follows with c
∗
:= (1 +c
2
)
−1
. ⊓⊔
Proof of the curvature estimate
Now we come to the proof of the theorem.
1. From the linear dependence of the three weighted fundamental forms
N
u
i ◦ W(X, N)
−1
◦ N
u
j −2H
W
X
u
i N
u
j +KX
u
i ◦ W(X, N) ◦ X
u
j = 0
for i, j = 1, 2 from section 5.1 we infer
N
u
◦ W(X, N)
−1
◦ N
u
+N
v
◦ W(X, N)
−1
◦ N
v
= 2
_
2H
W
(X, N)
2
−K
_
W
using weighted conformal parameters and making use of
−X
u
N
u
−X
v
N
v
= L
11
+L
22
= 2H
W
W.
With ρ
1
κ
1
+ρ
2
κ
2
= 2H
W
we compute
ρ
2
1
κ
2
1
+2ρ
1
ρ
2
κ
1
κ
2
+ρ
2
2
κ
2
2
= 4H
2
W
or after rearranging
ρ
1
ρ
2
κ
2
1
+
ρ
2
ρ
1
κ
2
2
=
4
ρ
1
ρ
2
H
2
W
−2κ
1
κ
2
=
4
ρ
1
ρ
2
H
2
W
−2K.
11.9 A curvature estimate 259
Now we can estimate as follows
(1 +ω
0
)[∇N[
2
≥ 4H
2
W
W −2KW
= 4
_
1 −
1
ρ
1
ρ
2
_
H
2
W
W +
_
ρ
1
ρ
2
κ
2
1
+
ρ
2
ρ
1
κ
2
2
_
W
≥ −4(2ω
0
+ω
2
0
)h
2
0
W +
1
(1 +ω
0
)
2
(κ
2
1
+κ
2
2
)W
where we make us of
1
1 +ω
0
≤ρ
1
(X, N), ρ
2
(X, N) ≤1 +ω
0
.
Rearranging for the principal curvatures at (0, 0) gives
κ
1
(0, 0)
2
+κ
2
(0, 0)
2
≤ (1 +ω
0
)
3
[∇N(0, 0)[
2
W(0, 0)
+4(1 +ω
0
)
2
(2ω
0
+ω
2
0
)h
2
0
.
(11.35)
Thus we must establish an upper bound for the gradient [∇N(0, 0)[ and a lower
bound for the area element W(0, 0).
For this we essentially use the assumed moduli of continuity: First for an upper
gradient bound of the coupled mapping X = (X, N), then for a local modulus of
projection of the spherical mapping to obtain a lower bound for the area element.
2. The assumed moduli of continuity immediately yield gradient bounds for X and
N as follows: Let as usual
X(w) :=
_
r
−1
X(w), N(w)
_
, w ∈ B,
then for given λ > 0 we ﬁnd δ =δ(λ) such that
[X(w) −X(w
1
)[ ≤2λ for all w ∈ B
δ
(w
0
)
where w
1
∈ B
δ
(w
0
) chosen arbitrarily within the disc B
δ
(w
0
) ⊂⊂
˚
B with center
w
0
∈ B
1−ν
0
(0, 0). The parameter ν
0
> 0 will be ﬁxed in point 6 of the proof. The
new mapping
´
X(w) :=
1
2λ
_
X(w) −X(w
1
)
_
: B
δ
(w
0
) −→R
6
thus satisﬁes
[
´
X(w)[ ≤1 and [△
´
X(w)[ ≤2λΛ
X
[∇
´
X[
2
in B
δ
(w
0
)
with Λ
X
=Λ
X
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r) from (11.31).
260 11 Crystalline functionals in R
3
Now choose λ =λ(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r) such that 2λΛ
X
≤
1
2
. It follows
[∆
´
X(w)[ ≤
1
2
[∇X[
2
in B
δ
(w
0
)
with suitable δ = δ(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r). Heinz’ gradient estimates from
section 9.4 ensure the existence of a constant c
1
∈ (0, +∞) with the property
[∇X(w
0
)[ ≤c
1
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r)
for all w
0
∈
˚
B
1−ν
0
(0, 0).
(11.36)
3. This gradient estimate allows us to linearize the system for r
−1
X as follows
[∆(r
−1
X)(w
0
)[ ≤c
2
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r)[∇(r
−1
X)(w
0
)[
for all w
0
∈
˚
B
1−ν
0
(0, 0)
with the constant
c
2
:= 2
√
2c
1
(1 +ω
0
)
_
2(rω
1
)
2
+2ω
2
2
+c
1
(h
0
r) ∈ (0, +∞).
4. Secondly we arrive at the following modulus of projection for the surface:
Choose ν ∈ (0, ν
0
/4) such that 2νc
1
≤1 is true. Then it holds
[N(w) −N(w
0
)[ ≤2νc
1
≤1 for all
˚
B
2ν
(w
0
) and all w
0
∈
˚
B
1−2ν
0
(0, 0).
5. Now we establish an inequality of Harnacktype for the area element W : For this
we choose a ﬁx point w
0
∈ B
1−2ν
0
(0, 0). Assume furthermore X(w
0
) = (0, 0, 0)
and N(w
0
) = (0, 0, 0) which can be ensured by rotation and translation. Consider
now the plane mapping
F(w) :=
1
r
_
x
1
(w
0
+2νw), x
2
(w
0
+2νw)
_
: B −→R
2
with the number ν >0 from the previous point of the proof. Due to the foregoing
lemma we have
[△F(w)[ ≤ 4ν
2
[△(r
−1
X)(w
0
+2νw)[
≤ 4ν
2
c
2
[∇(r
−1
X)(w
0
+2νw)[
≤
2νc
2
√
c
∗
[∇F(w)[
=: c
3
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r, ν
0
)[∇F(w)[
(11.37)
for w ∈
˚
B with the constant c
3
:=
νc
2
√
c
∗
.
11.9 A curvature estimate 261
From the fourth point of our proof we know
J
F
(w) > 0 in
˚
B
for the Jacobian of the plane mapping F =F(w). As in the proof of our curvature
estimate for immersions with prescribed mean curvature ﬁelds from section 10.3
we ﬁnd a constant c
4
= c
4
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r, d
0
) such that the following
Harnack inequality is fulﬁlled
[∇F(w)[
2
≤
_
[F
u
(w)[ +[F
v
(w)[
_
2
≤ c
2
4
_
[F
u
(0, 0)[ +[F
v
(0, 0)[
_2
5
≤
5
√
2c
2
4
[∇F(0, 0)[
2
5
(11.38)
for all w ∈
˚
B1
2
(0, 0). Furthermore we obtain
[∇F(w)[
2
≥
16ν
4
c
2
2
c
∗
4ν
2
c
2
2
[∇(r
−1
X)(w
0
+2νw)[
2
=
4ν
2
c
∗
r
2
[∇X(w
0
+2νw)[
2
≥
8ν
2
c
∗
1 +ω
0
W(w
0
+2νw)
r
2
and analogously
[∇F(w)[
2
≤
4ν
2
r
2
[∇X(w
0
+2νw)[
2
≤8ν
2
(1 +ω
0
)
W(w
0
+2νw)
r
2
. (11.39)
Summarizing it follows
c
5
W(w
0
+2νw)
r
2
≤[∇F(w)[
2
≤c
6
W(w
0
+2νw)
r
2
with the constants
c
5
:=
8ν
2
c
∗
1 +ω
0
, c
6
:= 8ν
2
(1 +ω
0
).
Thus together with (11.38) and (11.39) we conclude
c
5
W(w
0
+2νw)
r
2
≤[∇F(w)[
2
≤
5
_
2c
6
c
2
4
_
W(w
0
)
r
2
_1
5
for all w ∈
˚
B1
2
(0, 0).
262 11 Crystalline functionals in R
3
Rearranging yields the desired Harnacktype inequality for the area element
c
7
_
W(w)
r
2
_
5
≤
W(w
0
)
r
2
for all w ∈ B
ν
(w) and all w
0
∈
˚
B
1−2ν
0
(0, 0) (11.40)
with the constant
c
7
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r, d
0
) :=
1
2
c
−1
6
c
5
5
c
−10
4
.
6. Let Γ(B) denote the set of all continuous and piecewise differentiable curves
γ : [0, 1] →B with the properties
γ(0) = (0, 0) and γ(1) ∈ ∂B.
Then it holds
inf
γ∈Γ(B)
1
_
0
¸
¸
¸
¸
d
dt
X(γ(t))
¸
¸
¸
¸
dt ≥r.
As in the proof of the curvature estimate from section 10.3 we can show that
there is a point w
∗
∈
˚
B
1−ν
0
setting ν
0
≥e
−4πd
0
∈ (0, 1) such that
W(w
∗
)
r
2
≥
1
16(1 +ω
0
)(1 −ν
0
)
2
=: c
8
(ω
0
, d
0
) (11.41)
with the constant c
8
= c
8
(ω
0
, d
0
) ∈ (0, +∞).
7. Now we want to establish a lower bound for the area element. For this purpose
we choose a natural number n ∈ N sufﬁciently large such that
1 −2ν ≤nν ≤1 −ν.
We deﬁne points
w
j
:=
j
n
w
∗
, j = 0, 1, 2, . . . , n, w
∗
∈
˚
B
1−ν
0
(0).
Then we arrive at
[w
j
[ =
j
n
[w
∗
[ ≤[w
∗
[ ≤1 −ν
0
for j = 0, 1, 2, . . . , n
which implies
[w
j+1
−w
j
[ =
1
n
[w
∗
[ ≤
1−ν
0
n
≤
1−2ν
n
≤ν
for j = 0, 1, 2, . . . , n −1.
(11.42)
11.9 A curvature estimate 263
On account of (11.40) and (11.41) it follows that
W(w
0
)
r
2
≥c
7
_
W(w
1
)
r
2
_
5
≥ c
1+5
7
_
W(w
2
)
r
2
_
5
2
≥ c
1+5+5
2
+...+5
n−1
7
_
W(w
n
)
r
2
_
5
n
=c
1+5+5
2
+...+5
n−1
7
_
W(w
∗
)
r
2
_
5
n
≥ c
1+5+5
2
+...+5
n−1
7
c
5
n
8
.
(11.43)
Together with
c
9
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r, d
0
) := c
1+5+5
2
+...+5
n−1
7
c
5
n
8
we also infer
W(w
0
)
r
2
≥c
9
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r, d
0
).
8. We collect (11.35), (11.36) and (11.43) to arrive at the estimate
κ
1
(0, 0)
2
+κ
2
(0, 0)
2
≤
1
r
2
(1 +ω
0
)
3
c
1
c
9
+4(1 +ω
0
)
2
(2ω
0
+ω
2
0
)h
2
0
where w
0
= (0, 0). If we ﬁnally set
Θ(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r, d
0
)
:= (1 +ω
0
)
3
c
1
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r)
c
9
(ω
0
, rω
1
, ω
2
, h
0
r, h
1
r
2
, h
2
r, d
0
)
the theorem follows. ⊓⊔
Chapter 12
Crystalline functionals in R
n+2
12.1 Crystalline functionals. Remarks on Plateau’s problem
12.2 Parameter invariance
12.3 Euler’s homogeneity relations
12.4 The ﬁrst variation
12.5 Minimal surfaces
12.6 The problem for weighted minimal surfaces
266 12 Crystalline functionals in R
n+2
This chapter is devoted to extend our considerations to the analysis of criticical points of crystalline
functionals in Euclidean spaces of higher dimensions. We put some important results on Plateau’s
problem for such parametric functionals together, elaborate on their ﬁrst variations and discuss
their relationships to our variational problems so far.
12.1 Crystalline functionals. Remarks on Plateau’s problem
In generalization of our earlier discussions from the foregoing chapters we now
consider inhomogeneous and anisotropic Lagrangian densities of the form
F ∈C
5+α
(R
n
R
N
¸ ¦0¦, R) ∈C
0
(R
n
R
N
, R)
and variational problems for the functional of crystalline type
B[X] :=
__
B
F(X, X
u
∧X
v
)dudv −→extr! (12.1)
As discussed in the foregoing chapter, such Lagrangians F(X, X
u
∧X
v
) arise from
the theory of crystall growth; see the original work of G. Wulff in [174], or the later
contributions of J.E. Taylor, e.g. [153], [154] in the context of geometric measure
theory.
Before we go into calculations we want to recall some history of the socalled
Plateau problem. In its original form it aks for ﬁnding a minimal surface spanning
a given curve in space which is critical for the area functional
A[X] =
__
B
[X
u
X
v
[ dudv.
Garnier [69], Douglas [47] and Rad´ o [133] independently succeeded in the con
struction of such surfaces minimizing the area among all admissable immersions
spanning the same boundary curve. Owing to his work, Jesse Douglas was granted
with the ﬁrst Fields Medal in 1936.
It was also Douglas in [48] and [49] who ﬁrst succeeded on Plateau’s problemon
multiple connected domains which particularly involves minimal surfaces spanning
more than one boundary curve.
268 12 Crystalline functionals in R
n+2
For Plateau’s problem in Riemannian or hyperbolic spaces we want to refer to
classical works of Lonseth [117] from 1942 and Morrey [121] from 1948.
The mathematician’s attention to anisotropic and inhomogenous variational prob
lems was already drawn in the early ﬁfties, see e.g. Cesari [24], Danskin [41], Jenk
ins [99], or Morrey [122], [123]. Using methods from geometric measure theory, B.
White in [170], [171] proved existence of smooth embeddings solving this “gener
alized” Plateau problem assuming the boundary curve is contained in the boundary
of a convex body. For a good introduction into the powerful methods of geometric
measure theory we want to refer the reader to Almgren [2].
In a series of papers, this general Plateau problems was revived by S. Hildebrandt
and H. von der Mosel; see [86], [87], [88], [89], [90], [91]. The Douglas problemfor
inhomogenous and anisotropic parametric functionals, asking for the construction
of minimizers of higher topological type, was considered by Kurzke and von der
Mosel [110].
Existence and regularity of minimizers
The following existence and regularity results for minimizers of the general problem
12.1 are taken from Hildebrandt and von der Mosel [87], Theorem 1.4 and 1.5.
Theorem 12.1. If the Lagrangian density F ∈C
0
(R
n
R
N
, R) is
• positive homogeneous F(X, λZ) =λF(X, Z) for all λ > 0;
• positive deﬁnite m
1
[Z[ ≤F(X, Z) ≤m
2
[Z[ with 0 < m
1
≤m
2
<∞;
• convex F(X, λZ
1
+µZ
2
) ≤λF(X, Z
1
) +µF(X, Z
2
)
for all λ, µ > 0, λ +µ = 1
for all X ∈ R
n+2
and all Z, Z
1
, Z
2
∈ R
N
, then there exits a minimizer X : B →R
n+2
of B[X] in the following class of immersions
C(Γ) :=
_
X ∈ H
1,2
(
˚
B, R
n+2
) :
X[
∂B
is a continuous, weakly monotonic mapping from ∂B to Γ
_
which is conformally parametrized, i.e. X ∈ C(Γ) satisﬁes
B[X] = inf
¯
X∈C(Γ)
B[
¯
X]
as well as the conformality relations
X
u
X
u
= X
v
X
v
, X
u
X
v
= 0 almost everywhere on B.
12.1 Crystalline functionals. Remarks on Plateau’s problem 269
Furthermore, every conformally parametrized minimizer X of B[X] in the class
C(Γ) satisﬁes
X ∈C
0
(B, R
n
) ∩C
0,γ
(
˚
B, R
n
) with γ =
m
1
m
2
as well as the Morrey growth condition
__
B
r
(w
0
)
[∇X[
2
dudv ≤
_
r
R
_
2γ
__
B
R
(w
0
)
[∇X[
2
dudv
for any w
0
∈ B and 0 < r ≤R ≤R
0
:= 1 −[w
0
[.
To establish boundary regularity we need the concept of a perfect dominance
function. For this purpose let
f (X, P) := F(X, P
1
∧P
2
)
denote the associated Lagrangian to the Lagrangian F(X, P).
Deﬁnition 12.1. (Hildebrandt, von der Mosel [89], Deﬁnition 1)
(i) Let F(X, Z) be given with the associated Lagrangian f (X, P). Then a function
G: R
n+2
R
2(n+2)
−→R
is said to be a dominance function for F if it is continous and satisﬁes the follow
ing two conditions
f (X, P) ≤G(X, Z) for any (X, Z) ∈ R
n+2
R
2(n+2)
(D1)
as well as
f (X, P) = G(X, P) if and only if [p
1
[
2
=[p
2
[
2
, p
1
p
2
= 0. (D2)
(ii) A dominance function G of F is called quadratic if
G(X, λP) =λ
2
G(X, P) for all λ > 0, (X, P) ∈ R
n+2
R
2(n+2)
, (D3)
and it is said to be positive deﬁnite if there are two numbers 0 <µ
1
≤µ <∞ with
µ
1
[P[
2
≤G(X, P) ≤ µ
2
[P[
2
for any (X, P) ∈ R
n+2
R
2(n+2)
. (D4)
(iii) A function
G ∈C
0
(R
n+2
R
2(n+2)
) ∩C
2
(R
n+2
R
2(n+2)
¸ ¦0¦)
is called a perfect dominance function for F if it satisﬁes the foregoing conditions
(D1) – (D4), and if for any R
0
> 0 there is a constant λ
G
(R
0
) > 0 such that
ξ ◦ G
PP
(X, P) ◦ ξ ≤λ
G
(R
0
)[ξ[
2
for [X[ ≤R
0
and P, ξ ∈ R
2(n+2)
, P ,= 0.
270 12 Crystalline functionals in R
n+2
Before we state the boundary regularity theorem we give two examples (see
Hildebrandt and von der Mosel [89]).
1. The integrand
D(P) :=
1
2
[P[
2
=
1
2
[P
1
[
2
+
1
2
[P
2
[
2
of the Dirichlet energy represents a perfect dominance function for the area inte
grand [P
1
∧P
2
[.
2. The integrand
G(X, P) :=
1
2
[P[
2
+Q(X) (P
1
∧P
2
)
is a perfect dominance function for E(X, Z) :=[Z[ +Q(X) Z.
The existence of perfect dominance functions for a large class of Lagrangians is
the contents of the next result from Hildebrandt and von der Mosel [89].
Theorem 12.2. Let F
∗
∈C
0
(R
n+2
R
N
) ∩C
2
(R
n+2
R
N
¸¦0¦) be positive homo
geneous, positive deﬁnite with constants m
∗
1
and m
∗
2
, and convex, and assume that
[Z[ ξ ◦ F
∗
ZZ
(X, Z) ◦ ξ ≥λ
F
∗ (R
0
)
¸
¸
ξ −[Z[
−2
(Z ξ)Z
¸
¸
2
with the ellipticity constant
λ
∗
:= inf
R
0
∈(0,∞]
λ
F
∗ (R
0
) > 0.
Then for
k > k
0
:= 2
_
m
∗
2
−min¦λ
∗
, m
∗
1
/2¦
¸
the parametric Lagrangian
F(X, Z) := kA(Z) +F
∗
(X, Z)
possesses a perfect dominance function.
Now comes Hildebrandt’s and von der Mosel’s boundary regularity result from
[88], Theorem 1.1.
Theorem 12.3. Suppose that F is of class C
2
(R
n+2
R
N
¸ ¦0¦, R) and that it is
positive homogeneous, positive deﬁnite and convex. Suppose also that F possesses
a perfect dominance function G, and that the boundary curve Γ is of class C
4
. Then
there is some α ∈ (0, 1) such that any conformally parametrized minimizer X of
B[X] is of class H
2,2
(
˚
B, R
n+2
) ∩C
1,α
(B, R
n+2
) and satisﬁes
X
H
2,2
(
˚
B,R
n+2
)
+X
C
1,α
(B,R
n+2
)
≤c(Γ, F)
where the number c(Γ, F) depends only on Γ and F.
12.3 Euler’s homogeneity relations 271
Further remarks on the literature
Sauvigny in [141] established curvature estimates for Fminimal surfaces which
may arise as critical points for the functional
F[X] =
__
B
F(X
u
X
v
)dudv
in case of one codimension; see also Fr¨ ohlich [62] for further developments.
Various geometric and analytical results for critical points of F[X] in one codi
mension were particularly established in Lin [115], R¨ awer [134], Clarenz [31],
Clarenz and von der Mosel [32]. Anisotropic variational problems with applications
to rotationally symmetric Delaunaytype surfaces in case n = 1 of one codimension
can be found in Koiso and Palmer [107], [108].
Moreover, Winklmann in [172], [173] established integral and pointwise curva
ture estimates for mdimensional manifolds in R
m+1
which are critical for function
als similar to (12.1). The methods he used there go essentially back to Ecker [50]
and Schoen, Simon, Yau [145].
For considerations on anisotropic mean curvature ﬂows we want to refer to Clut
terbuck [33], or Pozzi [132] for numerical investigations.
12.2 Parameter invariance
Before we are going to compute the ﬁrst variation of B[X] we will establish a nec
essary and a sufﬁcient condition for the functional to be independent of the choice
of the parameter system.
Proposition 12.1. B[X] is invariant w.r.t. a parameter transformations from class
Pand of the chosen parameter domain if and only it holds
F(X, λZ) =λF(X, Z) for all real λ > 0. (12.2)
We omit the proof for it works exactly as our calcululations from section 11.2.
12.3 Euler’s homogeneity relations
As in the previous chapter we introduce the following notations.
272 12 Crystalline functionals in R
n+2
Deﬁnition 12.2. We set
F
X
(X, Z) := (F
x
1 (X, Z), . . . , F
x
n (X, Z)) ∈ R
n
,
F
Z
(X, Z) := (F
z
1(X, Z), . . . , F
z
N (X, Z)) ∈ R
N
,
F
XX
(X, Z):= (F
x
i
x
j (X, Z))
i, j=1,...,n
∈ R
nn
,
F
XZ
(X, Z):= (F
x
i
z
j (X, Z))
i=1,...,n, j=1,...,N
∈ R
nN
,
F
ZZ
(X, Z):= (F
z
i
z
j (X, Z))
i, j=1,...,N
∈ R
NN
with the abbreviation N :=
_
n
2
_
.
Nowthe homogeneity condition (12.2) allows us to infer the following properties
in the same way we proceeded in the foregoing chapter.
Proposition 12.2. The vector F
Z
(X, Z) is positivehomogeneous of degree 0 w.r.t. Z,
the matrix F
ZZ
(X, Z) is positivehomogeneous of degree −1 w.r.t. Z. Furthermore
their hold the homogeneity relations
F
Z
(X, Z) Z = F(X, Z), F
ZZ
(X, Z) ◦ Z = 0, F
XZ
(X, Z) ◦ Z = F
X
(X, Z)
for all X ∈ R
n
, Z ∈ R
n
¸ ¦0¦.
12.4 The ﬁrst variation
We compute the EulerLagrange equations of the general parametric functional
B[X]. For this purpose it is necessary to recall the Grassmanntype formalism
X ∧Y =
∑
1≤i<j≤n
(x
i
y
j
−x
j
y
i
)e
i
∧e
j
introduced in chapter 6.
Theorem 12.4. The EulerLagrange equations of B[X] are
F
X
(X, N ) N
σ
W +(N
σ
∧X
v
) ◦ F
ZX
(X, N ) ◦ X
u
+(X
u
∧N
σ
) ◦ F
ZX
(X, N ) ◦ X
v
= −(N
σ
∧X
v
) ◦ F
ZZ
(X, N ) ◦ N
u
−(X
u
∧N
σ
) ◦ F
ZZ
(X, N ) ◦ N
v
(12.3)
in B for all σ = 1, . . . , n with the Grassmanntype vector
N :=
X
u
∧X
v
[X
u
∧X
v
[
from (6.3).
12.4 The ﬁrst variation 273
Proof. Consider the variation
¯
X = X +ε
n
∑
σ=1
ϕ
σ
N
σ
with test functions ϕ
i
∈C
∞
0
(B, R) and a real ε ∈ (−ε
0
, +ε
0
). First we calculate
¯
X
u
∧
¯
X
v
=
_
X
u
+ε
n
∑
σ=1
_
ϕ
σ,u
N
σ
+ϕN
σ,u
_
_
∧
_
X
v
+ε
n
∑
σ=1
_
ϕ
σ,v
N
σ
+ϕ
σ
N
σ,v
_
_
= X
u
∧X
v
+ε
n
∑
σ=1
_
ϕ
σ,u
X
u
∧N
σ
+ϕ
σ,v
N
σ
∧X
v
_
+ε
n
∑
σ=1
_
X
u
∧N
σ,v
+N
σ,u
∧X
v
_
ϕ
σ
+o(ε).
We compute the derivative
∂
∂ε
B[
¯
X] =
__
B
_
F
X
(
¯
X,
¯
X
u
∧
¯
X
v
)
∂
∂ε
¯
X +F
X
(
¯
X,
¯
X
u
∧
¯
X
v
)
∂
∂ε
(
¯
X
u
∧
¯
X
v
)
_
dudv
using
∂
∂ε
¯
X
¸
¸
ε=0
=
n
∑
σ=1
ϕ
σ
N
σ
as well as
∂
∂ε
(
¯
X
u
∧
¯
X
v
)
¸
¸
ε=0
=
n
∑
σ=1
_
ϕ
σ,u
X
u
∧N
σ
+ϕ
σ,v
N
σ
∧X
v
_
+
n
∑
σ=1
_
X
u
∧N
σ,v
+N
σ,u
∧X
v
_
ϕ
σ
.
Taking the Euler homogeneity conditions into account we infer at ε = 0
δB[X] =
n
∑
σ=1
__
B
F
X
(X, X
u
∧X
v
) N
σ
ϕ
σ
dudv
+
n
∑
σ=1
__
B
F
Z
(X, X
u
∧X
v
) (X
u
∧N
σ,v
+N
σ,u
∧X
v
)ϕ
σ
dudv
+
n
∑
σ=1
__
B
F
Z
(X, X
u
∧X
v
) (X
u
∧N
σ
ϕ
σ,v
+N
σ
∧X
v
ϕ
σ,u
)dudv
= . . .
274 12 Crystalline functionals in R
n+2
. . . =
n
∑
σ=1
__
B
F
X
(X, N ) N
σ
Wϕ
σ
dudv
+
n
∑
σ=1
__
B
div
_
F
Z
(X, N ) (N
σ
∧X
v
)ϕ
σ
, F
Z
(X, N ) (X
u
∧N
σ
)ϕ
σ
_
dudv
−
n
∑
σ=1
__
B
_
F
ZX
(X, N ) X
u
_
(N
σ
∧X
v
)ϕ
σ
dudv
−
n
∑
σ=1
__
B
_
F
ZX
(X, N ) X
v
_
(X
u
∧N
σ
)ϕ
σ
dudv
−
n
∑
σ=1
__
B
_
F
ZZ
(X, N ) N
u
_
(N
σ
∧X
v
)ϕ
σ
dudv
−
n
∑
σ=1
__
B
_
F
ZZ
(X, N ) N
v
_
(X
u
∧N
σ
)ϕ
σ
dudv.
The integral over the divergence vanishes due ϕ
σ
= 0. This proves the theorem. ⊓⊔
12.5 Minimal surfaces
In the special case F(X, Z) =[Z[ we have
F
X
(X, Z) ≡0, F
XZ
(X, Z) ≡0, F
ZZ
(X, Z) ≡E
N
with the Ndimensional unit matrix E
N
. Thus we must evaluate the system
(N
σ
∧X
v
) N
u
+(X
u
∧N
σ
) N
v
= 0 for all σ = 1, . . . , n.
To this end we introduce conformal parameters (u, v) ∈ B. Then
_
X
u
√
W
,
X
v
√
W
, N
1
, . . . , N
n
_
as well as the set
_
N , X
11
, . . . , X
1n
, X
21
, . . . , X
2n
, N
11
, . . . , N
1n
, N
23
, . . . , N
n−1,n
_
consisting of the N =
_
n
2
_
unit normal vectors
N :=
X
u
∧X
v
W
, X
i1
:=
X
u
i ∧N
1
√
W
, . . . , X
in
:=
X
u
i ∧N
n
√
W
, Z
i j
:= N
i
∧N
j
form orthornormal frames.
12.5 Minimal surfaces 275
The EulerLagrange equations (12.3) take the simpler form
X
1σ
N
v
−X
2σ
N
u
= 0 in B. (12.4)
Next we recall the GrassmannWeingarten equations (6.4)
N
u
i =−
2
∑
m=1
n
∑
ϑ=1
L
mϑ
i
X
mϑ
with the Gfundamental form L
mϑ
i
= −N
u
i X
mϑ
. In terms of these coefﬁcients,
equation (12.4) can be written as
L
1σ
2
−L
2σ
1
= 0 in B
for all σ = 1, . . . , n.
We want to prove that these n equations are equivalent to H
σ
≡ 0 for the mean
curvatures H
σ
, σ = 1, . . . , n, along the unit normal vectors N
σ
.
First, from (N
σ
∧X
v
) N = 0 and (X
u
∧N
σ
) N = 0 there follow after differen
tiation
(N
σ
∧X
v
) N
u
= (N
σ,u
∧X
v
+N
σ
∧X
uv
) N ,
(X
u
∧N
σ
) N
v
= (X
uv
∧N
σ
+X
u
∧N
σ,v
) N .
The Weingarten equations and the Gauss equations imply
(N
σ,u
∧X
v
) N =−L
σ,11
N N −
n
∑
ω=1
T
ω
σ,1
X
2ω
N =−L
σ,11
,
(X
u
∧N
σ,v
) N =−L
σ,22
N N +
n
∑
ω=1
T
ω
σ,2
X
1ω
N =−L
σ,22
as well as
(N
σ
∧X
uv
) N = −
W
v
2W
X
1σ
N −
W
u
2W
X
2σ
N +
n
∑
ω=1
L
ω,12
Z
σω
N = 0,
(X
uv
∧N
σ
) N = 0.
Thus it follows
0 = (X
u
∧N
σ
) N
v
+(N
σ
∧X
v
) N
u
= −(X
u
∧N
σ,v
) N −(N
σ,u
∧X
v
) N
= L
σ,22
+L
σ,22
= 2H
σ
W.
Critical points of the area functional have vanishing mean curvature vector.
276 12 Crystalline functionals in R
n+2
12.6 The problem for weighted minimal surfaces
Consider ﬁnally a critical point X : B →R
n+2
of the anisotropic variational problem
__
B
F(N )dudv −→extr!
Then X : B →R
n+2
satisﬁes the EulerLagrange system
(N
σ
∧X
v
) ◦ F
ZZ
(N ) ◦ N
u
+(X
u
∧N
σ
) ◦ F
ZZ
(N ) ◦ N
v
= 0 in B.
It remains open how we can express this system in terms of a weight matrix
W(X, Z) as discussed in chapter 4.
Chapter 13
The second variation
13.1 Minimal surfaces
13.2 Immersions with a special mean curvature ﬁeld
13.3 Stability and µstability
13.4 µstability due to Schwarz for minimal graphs
13.5 Eigenvalue problems on manifolds
13.6 µstability due to Ruchert, Barbosa and do Carmo
13.7 Calibration forms
278 13 The second variation
In this chapter we compute the second variation of some selected geometric functionals and deﬁne
stability and µstability. We also discuss various criteria which ensure µstability. For this purpose
we recall some elementary facts on eigenvalue methods on the sphere. We conclude this chapter
with considering the second variation for nonparametric problems.
13.1 Minimal surfaces
The second variation
We start with the following
Theorem 13.1. Let N = (N
1
, . . . , N
n
) be an ONF for the conformally parametrized
minimal surface X : B →R
n+2
. Then the second variation
δ
2
N
´ γ
A[X; ϕ] :=
∂
2
∂ε
2
A[X +εϕN
´ γ
]
¸
¸
¸
ε=0
of the area functional A[X] w.r.t. perturbations
¯
X(u, v) = X(u, v) +εϕ(u, v)N
´ γ
(u, v),
where
N
´ γ
(u, v) =
n
∑
σ=1
´ γ
σ
(u, v)N
σ
(u, v),
n
∑
σ=1
´ γ
σ
(u, v)
2
= 1,
´ γ = (´ γ
1
, . . . , ´ γ
n
), ϕ ∈C
∞
0
(B, R) and real ε ∈ (−ε
0
, ε
0
) is given by
δ
2
N
´ γ
A[X; ϕ] =
__
B
([∇ϕ[
2
+2K
´ γ
Wϕ
2
)dudv
+
__
B
n
∑
σ=1
_
_
_
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
2
+
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
2
_
_
_
ϕ
2
dudv
with the Gaussian curvature K
´ γ
:= K
N
´ γ
along N
´ γ
.
280 13 The second variation
In particular, if the ONF N is free of torsion then
δ
2
N
´ γ
A[X] =
__
B
([∇ϕ[
2
+[∇´ γ[
2
ϕ
2
+2K
´ γ
Wϕ
2
)dudv
with the setting [∇´ γ[
2
=
n
∑
σ=1
_
(´ γ
σ
u
)
2
+(´ γ
σ
v
)
2
_
.
Proof. From the proof of Theorem 9.1 we know
δ
N
´ γ
g
11
=−2ϕL
´ γ,11
, δ
N
´ γ
g
12
=−2ϕL
´ γ,12
, δ
N
´ γ
g
22
=−2ϕL
´ γ,22
with the coefﬁcients L
´ γ,i j
= N
´ γ
X
u
i
u
j . We additionally infer
δ
2
N
´ γ
g
11
= 2ϕ
2
u
+2ϕ
2
N
2
´ γ,u
, δ
2
N
´ γ
g
12
= 2ϕ
u
ϕ
v
+2ϕ
2
N
´ γ,u
N
´ γ,v
,
δ
2
N
´ γ
g
22
= 2ϕ
2
v
+2ϕN
2
´ γ,v
.
Using conformal parameters we rewrite these identities as follows
δ
2
N
´ γ
g
11
= 2ϕ
2
u
+
2
W
ϕ
2
_
L
2
´ γ,11
+L
2
´ γ,12
_
+2ϕ
2
n
∑
σ=1
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
2
,
δ
2
N
´ γ
g
12
= 2ϕ
u
ϕ
v
+
2
W
ϕ
2
(L
´ γ,11
+L
´ γ,22
)L
´ γ,12
+2ϕ
2
n
∑
σ=1
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
__
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
,
δ
2
N
´ γ
g
22
= 2ϕ
2
v
+
2
W
ϕ
2
_
L
2
´ γ,12
+L
2
´ γ,22
_
+2ϕ
2
n
∑
σ=1
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
2
.
Namely the Weingarten equations (2.11) imply
N
´ γ,u
= −
L
´ γ,11
W
X
u
−
L
´ γ,12
W
X
v
+
n
∑
σ=1
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
N
σ
,
N
´ γ,v
= −
L
´ γ,12
W
X
u
−
L
´ γ,22
W
X
v
+
n
∑
σ=1
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
N
σ
and therefore
N
2
´ γ,u
=
L
2
´ γ,11
+L
2
´ γ,12
W
+
n
∑
σ=1
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
2
,
13.1 Minimal surfaces 281
N
´ γ,u
N
´ γ,v
=
L
´ γ,11
+L
´ γ,22
W
L
´ γ,12
+
n
∑
σ=1
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
__
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
,
N
2
´ γ,v
=
L
2
´ γ,12
+L
2
´ γ,22
W
+
n
∑
σ=1
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
2
.
This shows the above representations for the δ
2
N
´ γ
g
i j
. Next we compute
2Wδ
2
N
´ γ
W +2(δ
N
´ γ
W)
2
= g
11
δ
2
N
´ γ
g
22
+2δ
N
´ γ
g
11
δ
N
´ γ
g
22
+g
22
δ
2
N
´ γ
g
11
−2g
12
δ
2
N
´ γ
g
12
−2(δ
N
´ γ
g
12
)
2
,
i.e. using conformal parameters we arrive at
δ
2
N
´ γ
W =
1
2
δ
2
N
´ γ
g
11
+
1
2
δ
2
g
22
+
1
W
δ
N
´ γ
g
11
δ
N
´ γ
g
22
−
1
W
(δ
N
´ γ
g
12
)
2
−
1
W
(δ
N
´ γ
W)
2
.
Inserting the identities for δ
N
´ γ
g
i j
and δ
2
N
´ γ
g
i j
previously obtained, together with
δ
N
´ γ
W =−2H
´ γ
Wϕ
from section 9.1, gives us
δ
2
N
´ γ
W = [∇ϕ[
2
−4H
2
´ γ
Wϕ
2
+
ϕ
2
W
(L
2
´ γ,11
+2L
2
´ γ,12
+L
2
´ γ,22
) +
4ϕ
2
W
(L
´ γ,11
L
´ γ,22
−L
2
´ γ,12
)
+ϕ
2
n
∑
σ=1
_
_
_
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
2
+
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
2
_
_
_
.
But it holds
L
2
´ γ,11
+2L
2
´ γ,12
+L
2
´ γ,22
= (L
´ γ,11
+L
´ γ,22
)
2
−2(L
´ γ,11
L
´ γ,22
−L
2
´ γ,12
)
= 4H
2
´ γ
W
2
−2K
´ γ
W
2
.
Therefore we obtain
δ
2
N
´ γ
W = [∇ϕ[
2
+2K
´ γ
Wϕ
2
+
n
∑
σ=1
_
_
_
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
2
+
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
2
_
_
_
ϕ
2
for arbitrary ϕ ∈C
∞
0
(B, R). The statement follows. ⊓⊔
282 13 The second variation
Invariant formulation
Fromthe previous proof we can infer the formof the second variation using arbitrary
parameters (u, v) ∈ B. For this purpose we need the following deﬁnition (see e.g.
Blaschke and Leichtweiss [15]).
Deﬁnition 13.1. Let two functions φ, ψ ∈C
1
(B, R) be given. Its Beltrami operator
of ﬁrst kind w.r.t. the metric
ds
2
= g
11
du
2
+2g
12
dudv +g
22
dv
2
is deﬁned as
∇
ds
2 (φ, ψ) :=
2
∑
i, j=1
g
i j
φ
u
i ψ
u
j (13.1)
In particular, if (u, v) ∈B are conformal parameters, i.e. g
11
=W =g
22
and g
12
=
0, then there hold
∇
ds
2 (φ, ψ) =
1
W
(φ
u
ψ
u
+φ
v
ψ
v
) =
1
W
∇φ ∇ψ
with the Euclidean gradient operator ∇φ = (φ
u
, φ
v
) etc., as well as
∇
ds
2 (φ, φ) =
1
W
[∇φ[
2
.
The Beltrami operator of ﬁrst kind is well deﬁned for arbitrary Riemannian met
rics dσ
2
with positive determinant.
Corollary 13.1. Using an arbitrary parameter system (u, v) ∈ B it holds
δ
2
N
´ γ
A[X; ϕ] =
__
B
∇
ds
2 (ϕ, ϕ)W dudv +2
__
B
K
´ γ
Wϕ
2
dudv
+
2
∑
i, j=1
n
∑
σ=1
__
B
g
i j
_
´ γ
u
i +
n
∑
ω=1
´ γ
ω
T
σ
ω,i
__
´ γ
u
j +
n
∑
ω=1
´ γ
ω
T
σ
ω, j
_
Wϕ
2
dudv.
Special cases
We want to exemplify the previous considerations.
Consider ﬁrst the case n = 1 of one codimension. Up to orientation there is only
one unit normal vector N for the minimal immersion X : B →R
3
, and the formula
for the second variation reduces to
δ
N
A[X; ϕ] =
__
B
_
[∇ϕ[
2
+2KWϕ
2
_
dudv.
13.2 Immersions with a special mean curvature ﬁeld 283
Furthermore, instead of
¯
X = X +εϕN
´ γ
we will often consider special normal
variations of the form
¯
X(u, v) = X(u, v) +εϕ(u, v)N
ω
(u, v).
Here the unit normal vector N
ω
is chosen from a given ONF N = (N
1
, . . . , N
n
), i.e.
we set
´ γ
ϑ
=
_
1 if ϑ =ω
0 if ϑ ,=ω
.
Then the second variations of the area functional reads
δ
2
N
ω
A[X; ϕ] =
__
B
_
[∇ϕ[
2
+2K
ω
Wϕ
2
_
dudv
+
n
∑
σ=1
__
B
_
(T
σ
ω,1
)
2
+(T
σ
ω,2
)
2
_
ϕ
2
dudv
(13.2)
using conformal parameters.
The functional of the total torsion which appears here on the right hand side in
combination with a test function ϕ ∈ C
∞
0
(B, R) can be controlled by means of our
estimates established in chapters 7 and 8 if we use normal Coulomb frames N of X
for the construction of
¯
X.
13.2 Immersions with a special mean curvature ﬁeld
Next we want to compute the second variation of the functional
J[X] :=
__
B
Γ(X)W dudv
which emerges from the general functional of Gullivertype
G[X] =
__
B
_
Γ(X)W +X
u
◦ A(X) ◦ X
v
_
dudv
in the special case A(X) ≡0.
From section 10.1 we ﬁrst infer that critical points of J[X] possess the mean
curvature ﬁeld
H
σ
(X, N
σ
) =
Γ
X
(X) N
σ
2Γ(X)
, σ = 1, . . . , n.
The case Γ(X) ≡ 1 corresponds to the area functional with minimal surfaces as
critical points.
284 13 The second variation
The second variation
Theorem 13.2. Using conformal parameters (u, v) ∈ B, the second variation of
J[X] w.r.t. to perturbations
¯
X(u, v) = X(u, v) +εϕ(u, v)N
´ γ
(u, v),
where
N
´ γ
(u, v) =
n
∑
σ=1
´ γ
σ
(u, v)N
σ
(u, v),
n
∑
σ=1
´ γ(u, v)
2
= 1,
´ γ = (´ γ
1
, . . . , ´ γ
n
), ϕ ∈C
∞
0
(B, R) and real ε ∈ (−ε
0
, ε
0
) is given by
δ
2
N
´ γ
J[X; ϕ] =
__
B
Γ(X)[∇ϕ[
2
dudv
+2
__
B
_
H
´ γ,X
(X, N
´ γ
) N
´ γ
−2H
´ γ
(X, N
´ γ
)
2
+K
´ γ
_
Γ(X)Wϕ
2
dudv
+
n
∑
σ=1
__
B
_
_
´ γ
σ
u
+´ γ
ω
T
σ
ω,1
_
2
+
_
´ γ
σ
v
+´ γ
ω
T
σ
ω,2
_
2
_
Γ(X)ϕ
2
dudv
with the mean curvature H
´ γ
= H
´ γ
(X, N
´ γ
) and the Gauss curvature K
´ γ
of the immer
sion along N
´ γ
. In particular, if the ONF N is free of torsion then the third integrand
on the right hand side vanishes identically.
Proof. First we compute
∂
∂ε
Γ(
¯
X)
¯
W = Γ
X
(
¯
X) N
´ γ
¯
Wϕ +Γ(
¯
X)
∂
∂ε
¯
W
= 2Γ(
¯
X)H
∗
´ γ
(
¯
X)
¯
Wϕ +Γ(
¯
X)
∂
∂ε
¯
W
setting
H
∗
σ
(X) =
Γ
X
(X) N
σ
2Γ(X)
with N
σ
ﬁxed. Note here that in the ﬁrst identity no variation
¯
N
´ γ
appears. Thus we
obtain
∂
∂ε
J[
¯
X] =
__
B
_
2Γ(
¯
X)H
∗
´ γ
(
¯
X)
¯
Wϕ +Γ(
¯
X)
∂
∂ε
¯
W
_
dudv.
A further differentiation w.r.t. ε at ε = 0 taking account of the identities
δ
N
´ γ
W =−2H
´ γ
(X, N
´ γ
)Wϕ =−2H
∗
´ γ
(X)Wϕ
13.2 Immersions with a special mean curvature ﬁeld 285
as well as
δ
2
N
´ γ
W = [∇ϕ[
2
+2K
´ γ
Wϕ
2
+
n
∑
σ=1
_
_
_
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
2
+
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
2
_
_
_
ϕ
2
from above yields
δ
N
´ γ
J[X; ϕ] = 2
__
B
_
Γ
X
(X) N
´ γ
H
∗
´ γ
(X) +Γ(X)H
∗
´ γ,X
(X) N
´ γ
_
Wϕ
2
dudv
−2
__
B
_
2Γ(X)H
∗
´ γ
(X)
2
+Γ
X
(X) N
´ γ
H
∗
´ γ
(X)
_
Wϕ
2
dudv
+
__
B
Γ(X)[∇ϕ[
2
dudv +2
__
B
Γ(X)K
´ γ
Wϕ
2
dudv
+
__
B
n
∑
σ=1
_
_
_
_
´ γ
σ
u
+
n
∑
ω=1
´ γ
ω
T
σ
ω,1
_
2
+
_
´ γ
σ
v
+
n
∑
ω=1
´ γ
ω
T
σ
ω,2
_
2
_
_
_
Γ(X)ϕ
2
dudv.
Due to H
∗
´ γ,X
(X) = H
´ γ,X
(X, N
´ γ
), rearranging proves the statement. ⊓⊔
Invariant formulation
Finally we want to reformulate the result of the previous theorem in terms of ar
bitrary parametrizations. For this purpose we use the invariant Beltrami operator
∇
ds
2 (, ) from the foregoing section.
Corollary 13.2. Using an arbitrary parameter system (u, v) ∈ B it holds
δ
2
N
´ γ
J[X; ϕ] =
__
B
∇
ds
2 (ϕ, ϕ)Γ(X)W dudv
+
__
B
_
H
´ γ,X
(X, N
´ γ
) N
´ γ
−2H
´ γ
(X, N
´ γ
)
2
+K
´ γ
_
Γ(X)Wϕ
2
dudv
+
2
∑
i, j=1
n
∑
σ=1
__
B
g
i j
_
´ γ
u
i +
n
∑
ω=1
´ γ
ω
T
σ
ω,i
__
´ γ
u
j +
n
∑
ω=1
´ γ
ω
T
σ
ω, j
_
Γ(X)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R).
286 13 The second variation
13.3 Stability and µstability
The most general geometric functional we have been considering so far is the fol
lowing functional of Gullivertype
G[X] =
__
B
_
Γ(X)W +2X
u
◦ A(X) ◦ X
v
_
dudv
whose critical points possess the mean curvature ﬁeld
H
σ
(X, N
σ
) =
1
Γ(X)W
n+2
∑
k,ℓ,m=1
a
kℓ,x
m(x
k
u
x
ℓ
v
n
m
σ
+x
m
u
x
k
v
n
ℓ
σ
+x
ℓ
u
x
m
v
n
k
σ
) +
Γ
X
(X) N
σ
2Γ(X)
.
Nonnegativity of the second variation of G[X] for critical points X leads us to
the concept of (weakly) stable immersions. We will consider stable minimal surfaces
and stable immersions with a special prescribed mean curvature H(X).
More general is the concept of µstability. In this section we will verify a certain
µstability condition for stable immersions. In the sections to follow as well as in
the next chapter we will establish µstability conditions without assuming stability
of the immersions under consideration.
Stable immersions
Deﬁnition 13.2. Let X : B →R
n+2
be critical for the functional G[X]. We say that
the immersion is additionally stable for G[X] if it holds
δ
2
N
´ γ
G[X; ϕ] ≥0
for all ϕ ∈C
∞
0
(B, R) and for all unit normal vectors N
´ γ
.
Examples
Let us come back to formula (13.2) for a minimal immersion X : B →R
n+2
with
ONF N. If the surface is stable then
__
B
[∇ϕ[
2
dudv ≥2
__
B
(−K
ω
)Wϕ
2
dudv −
n
∑
σ=1
__
B
_
(T
σ
ω,1
)
2
+(T
σ
ω,2
)
2
_
ϕ
2
dudv
for all ω = 1, . . . , n.
13.3 Stability and µstability 287
Notice that −K
ω
≥0. Summation implies
__
B
[∇ϕ[
2
dudv ≥
2
n
__
B
(−K)Wϕ
2
dudv
−
1
n
n
∑
σ,ω=1
__
B
_
(T
σ
ω,1
)
2
+(T
σ
ω,2
)
2
_
ϕ
2
dudv
with the Gaussian curvature
K =
n
∑
ω=1
K
ω
.
Consider now the second variation formula for the functional J[X]. Evaluating
it for N
ω
from the ONF N instead of the general ﬁeld N
´ γ
gives us
__
B
Γ(X)[∇ϕ[
2
dudv ≥ 2
__
B
_
2H
ω
(X, N
ω
)
2
−K
ω
_
Γ(X)Wϕ
2
dudv
−2
__
B
H
ω,X
(X, N
ω
) N
ω
Γ(X)Wϕ
2
dudv
−
n
∑
σ=1
__
B
_
(T
σ
ω,1
)
2
+(T
σ
ω,2
)
2
_
Γ(X)Wϕ
2
dudv.
Summation over ω = 1, . . . , n shows now
__
B
Γ(X)[∇ϕ[
2
dudv ≥
2
n
__
B
_
2H(X)
2
−K
_
Γ(X)Wϕ
2
dudv
−
2
n
n
∑
ω=1
__
B
H
ω,X
(X, N
ω
) N
ω
Γ(X)Wϕ
2
dudv
−
1
n
n
∑
σ,ω=1
__
B
_
(T
σ
ω,1
)
2
+(T
σ
ω,2
)
2
_
Γ(X)Wϕ
2
dudv
with the square H(X)
2
of the mean curvature vector
H(X) =
n
∑
ω=1
H
ω
(X, N
ω
)N
ω
.
µstable immersions
Motivated from these examples we introduce the following general concept.
288 13 The second variation
Deﬁnition 13.3. The immersion X is called µstable with a real number µ > 0 if
there is a smooth function q ∈C
2+α
(B, R) such that it holds
__
B
∇
ds
2 (ϕ, ϕ)W dudv ≥ µ
__
B
(q −K)Wϕ
2
dudv
−α(n)
2
∑
i, j=1
n
∑
σ,ϑ=1
__
B
g
i j
T
ϑ
σ,i
T
ϑ
σ, j
Wϕ
2
dudv
(13.3)
for all ϕ ∈ C
∞
0
(B, R), where α(n) is a nonnegative real number depending on the
codimension n ≥1.
Generalized stability conditions were studied in the literature. We especially want
to refer the reader to FischerColbrie and Schoen [60], Colding and Minicizzi [36]
and Fr¨ ohlich [62].
Examples
Let us ﬁrst examplarily consider the stability inequality
__
B
[∇ϕ[
2
dudv ≥
2
n
__
B
(−K)Wϕ
2
dudv
−
1
n
n
∑
σ,ω=1
__
B
_
(T
σ
ω,1
)
2
+(T
σ
ω,2
)
2
_
Wϕ
2
dudv
for a conformally parametrized minimal surface X : B →R
n+2
. Obviously this sur
face is µstable with
µ =
2
n
and α(n) =
1
n
, q ≡0.
The case n = 1 of one codimension leads us to µstable minimal surfaces with
µ = 2, i.e.
__
B
[∇ϕ[
2
dudv ≥2
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈ C
∞
0
(B, R). In this situation the usual stability inequality coincides with
the µstability with µ = 2.
Secondly we consider the stability inequality we previously derived for J
critical immersions X : B →R
n+2
. Let again two positive real numbers
0 <Γ
0
≤Γ(X) ≤Γ
1
<∞ for all X ∈ R
n+2
be given.
13.4 µstability due to Schwarz for minimal graphs 289
Assume that for all X ∈ R
n+2
the following smallness condition is satisﬁed
[H
ω,X
(X, N
ω
)[ ≤H
ω
(X, N
ω
)
2
for all ω = 1, . . . , n.
Then we conclude
__
B
[∇ϕ[
2
dudv ≥
2Γ
0
nΓ
1
__
B
_
_
2H(X)
2
−K
¸
−
n
∑
ω=1
H
ω,X
(X, N
ω
) N
ω
_
Wϕ
2
dudv
−
1
n
n
∑
σ,ω=1
__
B
_
(T
σ
ω,1
)
2
+(T
σ
ω,2
)
2
_
Wϕ
2
dudv
where we naturally assume the right hand side be nonnegative for all ϕ ∈C
∞
0
(B, R).
The following sections are devoted to the problem of establishing µstability
conditions from various geometric conditions. Further considerations regarding
weighted minimal surfaces can be found in the next chapter.
13.4 µstability due to Schwarz for minimal graphs
A classical result due to H.A. Schwarz is (see e.g. Nitsche [126], §104).
Given the conformally parametrized minimal immersion X in R
3
, and assume
that there is an everywhere positive solution χ of the differential equation
△χ −2KWχ = 0 in B.
Then X is stable in the sense that
__
B
[∇ϕ[
2
dudv ≥2
__
B
(−K)Wϕ
2
dudv for all ϕ ∈C
∞
0
(B, R).
Now we will prove this result in the general case of higher codimension but for
surfaces with ﬂat normal bundle.
An auxiliary function
For this purpose we introduce the function
χ :=
x
1
u
x
2
v
−x
1
v
x
2
u
W
with the Jacobian J
F
:= x
1
u
x
2
v
−x
2
u
x
1
v
(13.4)
of the plane mapping F = (x
1
, x
2
) for the surface vector X = (x
1
, x
2
, . . . , x
n+2
).
290 13 The second variation
It holds χ > 0 if X represents a graph over the [x, y]plane. Furthermore using
conformal parameters (u, v) ∈ B we have
∆χ −2KWχ = 0 in B
for the minimal graph if its normal bundle is ﬂat.
The general case is contained in our next
Proposition 13.1. Let X : B →R
n+2
be a conformally parametrized minimal sur
face with an ONF N be given. Then it holds
∆χ = 2KWχ +
n
∑
σ,ϑ=1
S
ϑ
σ,12
(n
1
σ
n
2
ϑ
−n
2
σ
n
1
ϑ
) in B
with the Gaussian curvature K of the surface and the components S
ϑ
σ,i j
of the cur
vature tensor of its normal bundle.
If n = 1 then χ is represents the third component of the unit normal vector N of
the immersion X : B →R
3
. In fact it holds
∆N = 2KWN in B
using conformal parameters for minimal surfaces.
An elliptic system for the vector N
The differential equation from the previous proposition follows at once from an
elliptic system for the unit vector
N =
X
u
∧X
v
W
which we introduced in chapter 6.
Theorem 13.3. Let the conformally parametrized minimal surface X : B →R
n+2
be
given. Then it holds
∆N = 2KWN +
n
∑
σ,ϑ=1
S
ϑ
σ,12
N
σ
∧N
ϑ
= 2KWN +2SW
with the Grassmann curvature vector
S =
1
W
∑
1≤σ<ϑ≤n
S
ϑ
σ,12
N
σ
∧ϑ
of the normal bundle from chapter 6.
13.4 µstability due to Schwarz for minimal graphs 291
Proof. Let us consider the function
¯ χ :=
x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
W
,
¯
F := x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
, k, ℓ = 1, . . . , n +2.
We will prove the identity
∆ ¯ χ = 2KW¯ χ +
n
∑
σ,ϑ=1
S
ϑ
σ,12
(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
).
We start with computing the ﬁrst and second derivatives of ¯ χ. First we have
¯ χ
u
=
1
W
n
∑
σ=1
_
L
σ,11
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,12
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
,
¯ χ
v
=
1
W
n
∑
σ=1
_
L
σ,12
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,22
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
.
For the proof we calculate
¯ χ
u
=
J
¯
F,u
W
−
W
u
W
2
J
¯
F
=
1
W
(x
k
uu
x
ℓ
v
+x
k
u
x
ℓ
uv
−x
ℓ
uu
x
k
v
−x
ℓ
u
x
k
uv
) −
W
u
W
2
J
¯
F
=
1
W
_
Γ
1
11
x
k
u
x
ℓ
v
+Γ
2
11
x
k
v
x
ℓ
v
+Γ
1
12
x
ℓ
u
x
k
u
+Γ
2
12
x
ℓ
v
x
k
u
−Γ
1
11
x
ℓ
u
x
k
v
−Γ
2
11
x
ℓ
v
x
k
v
−. . .
. . . −Γ
1
12
x
k
u
x
ℓ
u
−Γ
2
12
x
k
v
x
ℓ
u
_
−
W
u
W
2
J
¯
F
+
1
W
n
∑
σ=1
(L
σ,11
n
k
σ
x
ℓ
v
+L
σ,12
n
ℓ
σ
x
k
u
−L
σ,11
n
ℓ
σ
x
k
v
−L
σ,12
n
ℓ
σ
x
k
u
)
=
1
2W
2
_
W
u
x
k
u
x
ℓ
v
−W
v
x
k
v
x
ℓ
v
+W
v
x
ℓ
u
x
k
u
+W
u
x
ℓ
v
x
k
u
−W
u
x
ℓ
u
x
k
v
+W
v
x
ℓ
v
x
k
v
−. . .
. . . −W
v
x
k
u
x
ℓ
u
−W
u
x
k
v
x
ℓ
v
_
−
W
u
W
2
J
¯
F
+
1
W
n
∑
σ=1
_
L
σ,11
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,12
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
=
1
2W
2
_
W
u
x
k
u
x
ℓ
v
+W
u
x
ℓ
v
x
k
u
−W
u
x
ℓ
u
x
k
v
−W
u
x
k
v
x
ℓ
u
_
−
W
u
W
2
J
¯
F
+
1
W
n
∑
σ=1
L
σ,11
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,12
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
W
=
1
W
n
∑
σ=1
_
L
σ,11
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,12
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
.
292 13 The second variation
Analogously we continue with the derivative w.r.t. v. There follow
¯ χ
uu
= −
W
u
W
2
n
∑
σ=1
_
L
σ,12
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,22
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
+
1
W
n
∑
σ=1
_
L
σ,11,u
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,12,u
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
+
1
W
n
∑
σ=1
_
L
σ,11
(n
k
σ,u
x
ℓ
v
−n
ℓ
σ,u
x
k
v
) −L
σ,12
(n
k
σ,u
x
ℓ
u
−n
ℓ
σ,u
x
k
u
)
+
1
W
n
∑
σ=1
L
σ,11
(n
k
σ
x
ℓ
uv
−n
ℓ
σ
x
k
uv
) −L
σ,12
(n
k
σ
x
ℓ
uu
−n
ℓ
σ
x
k
uu
)
= −
W
u
W
2
n
∑
σ=1
_
L
σ,12
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,22
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
+
1
W
n
∑
σ=1
_
L
σ,11,u
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,12,u
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
−
1
W
2
n
∑
σ=1
L
σ,11
(L
σ,11
x
k
u
x
ℓ
v
+L
σ,12
x
k
v
x
ℓ
v
−L
σ,11
x
ℓ
u
x
k
v
−L
σ,12
x
ℓ
v
x
k
v
)
+L
σ,12
(L
σ,11
x
k
u
x
ℓ
u
+L
σ,12
x
k
v
x
ℓ
u
−L
σ,11
x
ℓ
u
x
k
u
−L
σ,12
x
ℓ
v
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
_
L
σ,11
T
ϑ
σ,1
(n
k
ϑ
x
ℓ
v
−n
ℓ
ϑ
x
k
v
) −L
σ,12
T
ϑ
σ,1
(n
k
ϑ
x
ℓ
u
−n
ℓ
ϑ
x
k
u
)
_
+
1
W
2
n
∑
σ=1
L
σ,11
(Γ
1
12
x
ℓ
u
n
k
σ
+Γ
2
12
x
ℓ
v
n
k
σ
−Γ
1
12
x
k
u
n
ℓ
σ
−Γ
2
12
x
k
v
n
ℓ
σ
)
−L
σ,12
(Γ
1
11
x
ℓ
u
n
k
σ
+Γ
2
11
x
ℓ
v
n
k
σ
−Γ
1
12
x
k
u
n
ℓ
σ
−Γ
2
12
x
k
v
n
ℓ
σ
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
_
L
σ,11
L
ϑ,12
(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
) −L
σ,12
L
ϑ,11
(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
_
13.4 µstability due to Schwarz for minimal graphs 293
= −
W
u
2W
2
n
∑
σ=1
_
L
σ,11
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,12
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
−
1
W
n
∑
σ=1
_
L
σ,22,u
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) +L
σ,12,u
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
+
1
2W
2
n
∑
σ=1
_
L
σ,11
W
v
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
) +L
σ,11
W
u
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
_
−
1
2W
2
n
∑
σ=1
_
L
σ,12
W
u
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
) −L
σ,12
W
v
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
_
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,11
L
ϑ,12
−L
σ,12
L
ϑ,11
)(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
−
1
W
2
n
∑
σ=1
(L
2
σ,11
+L
2
σ,12
)(x
k
u
x
ℓ
v
−x
k
v
x
ℓ
u
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
L
σ,11
T
ϑ
σ,1
(n
k
ϑ
x
ℓ
v
−n
ℓ
ϑ
x
k
v
)
−
1
W
n
∑
σ=1
n
∑
ϑ=1
L
σ,12
T
ϑ
σ,1
(n
k
ϑ
x
ℓ
u
−n
ℓ
ϑ
xu
k
)
as well as
¯ χ
vv
= −
1
W
2
n
∑
σ=1
_
L
σ,12
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) −L
σ,22
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
+
1
W
n
∑
σ=1
_
L
σ,12,v
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) +L
σ,11,v
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
−
1
2W
2
n
∑
σ=1
_
L
σ,12
W
u
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
) −L
σ,12
W
v
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
_
−
1
2W
2
n
∑
σ=1
_
L
σ,22
W
u
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
) +L
σ,22
W
v
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
_
294 13 The second variation
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,12
L
ϑ,22
−L
σ,22
L
ϑ,12
)(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
−
1
W
2
n
∑
σ=1
(L
2
σ,12
+L
2
σ,22
)(x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
L
σ,12
T
ϑ
σ,2
(n
k
ϑ
x
ℓ
v
−n
ℓ
ϑ
x
k
v
) −
1
W
n
∑
σ=1
n
∑
ϑ=1
L
σ,22
T
ϑ
σ,2
(n
k
ϑ
x
ℓ
u
−n
ℓ
ϑ
x
k
u
).
Now the Laplacian:
△¯ χ = −
1
W
2
n
∑
σ=1
_
L
σ,11
W
u
−
1
2
L
σ,11
W
u
−
1
2
L
σ,12
W
v
+L
σ,12
W
v
−. . .
. . . −
1
2
L
σ,12
W
v
+
1
2
L
σ,22
W
u
_
(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
−
1
W
2
n
∑
σ=1
_
−L
σ,12
W
u
−
1
2
L
σ,11
W
v
+
1
2
L
σ,12
W
u
−L
σ,22
W
v
+. . .
. . . +
1
2
L
σ,12
W
u
+
1
2
L
σ,22
W
v
_
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
+
1
W
n
∑
σ=1
(−L
σ,22,u
+L
σ,12,v
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
+
1
W
n
∑
σ=1
(−L
σ,12,u
+L
σ,11,v
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
−
1
W
2
n
∑
σ=1
(L
2
σ,11
+2L
2
σ,12
+L
2
σ,22
)(x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,12
L
ϑ,22
−L
σ,22
L
ϑ,12
+L
σ,11
L
ϑ,12
−. . .
. . . − L
σ,12
L
ϑ,11
)(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,12
T
ϑ
σ,2
+L
σ,11
T
ϑ
σ,1
)(n
k
ϑ
x
ℓ
v
−n
ℓ
ϑ
x
k
v
)
13.4 µstability due to Schwarz for minimal graphs 295
−
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,22
T
ϑ
σ,2
+L
σ,12
T
ϑ
σ,1
)(n
k
ϑ
x
ℓ
u
−n
ℓ
ϑ
x
k
u
)
= −
1
W
2
n
∑
σ=1
_
−L
σ,12
W
u
−
1
2
L
σ,11
W
v
+
1
2
L
σ,12
W
u
−L
σ,22
W
v
+. . .
. . . +
1
2
L
σ,12
W
u
+
1
2
L
σ,22
W
v
_
(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
+
1
W
n
∑
σ=1
(−L
σ,22,u
+L
σ,12,v
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
+
1
W
n
∑
σ=1
(−L
σ,12,u
+L
σ,11,v
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
−
1
W
2
n
∑
σ=1
(L
2
σ,11
+2L
2
σ,12
+L
2
σ,22
)(x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,12
L
ϑ,22
−L
σ,22
L
ϑ,12
+L
σ,11
L
ϑ,12
−. . .
. . . − L
σ,12
L
ϑ,11
)(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,12
T
ϑ
σ,2
+L
σ,11
T
ϑ
σ,1
)(n
k
ϑ
x
ℓ
v
−n
ℓ
ϑ
x
k
v
)
−
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,22
T
ϑ
σ,2
+L
σ,12
T
ϑ
σ,1
)(n
k
ϑ
x
ℓ
u
−n
ℓ
ϑ
x
k
u
)
=
1
W
n
∑
σ=1
(Γ
1
22
L
σ,11
+Γ
2
22
L
σ,12
−Γ
1
12
L
σ,12
−Γ
2
12
L
σ,22
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
+
1
W
n
∑
σ=1
(Γ
1
12
L
σ,11
+Γ
2
12
L
σ,12
−Γ
1
11
L
σ,12
−Γ
2
11
L
σ,22
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
−
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,22
T
σ
ϑ,1
−L
ϑ,12
T
σ
ϑ,2
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
296 13 The second variation
−
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,12
T
σ
ϑ,1
−L
ϑ,11
T
σ
ϑ,2
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
_
(L
σ,11
−L
σ,22
)L
ϑ,12
−(L
ϑ,11
−L
ϑ,22
)L
σ,12
_
(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
−
1
W
2
n
∑
σ=1
(L
2
σ,11
+2L
2
σ,12
+L
2
σ,22
)(x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
)
=
1
2W
2
n
∑
σ=1
(−W
u
L
σ,11
+W
v
L
σ,12
−W
v
L
σ,12
−W
u
L
σ,22
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
+
1
2W
2
n
∑
σ=1
(W
v
L
σ,11
+W
u
L
σ,12
−W
u
L
σ,12
+W
v
L
σ,22
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,22
T
σ
ϑ,1
−L
ϑ,12
T
σ
ϑ,2
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,12
T
σ
ϑ,1
−L
ϑ,11
T
σ
ϑ,2
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
+
n
∑
σ=1
n
∑
ϑ=1
S
ϑ
σ,12
(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
−
1
W
2
n
∑
σ=1
(L
2
σ,11
+2L
2
σ,12
+L
2
σ,22
)(x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
)
=
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,22
T
σ
ϑ,1
−L
ϑ,12
T
σ
ϑ,2
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,12
T
σ
ϑ,1
−L
ϑ,11
T
σ
ϑ,2
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
+
n
∑
σ=1
n
∑
ϑ=1
S
ϑ
σ,12
(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
)
13.4 µstability due to Schwarz for minimal graphs 297
−2
n
∑
σ=1
(−K
ε
)(x
k
u
x
ℓ
v
−x
ℓ
u
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,11
T
ϑ
σ,1
+L
σ,12
T
ϑ
σ,2
)(n
k
ϑ
x
ℓ
v
−n
ℓ
ϑ
x
k
v
)
−
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
σ,22
T
ϑ
σ,2
+L
σ,12
T
ϑ
σ,1
)(n
k
ϑ
x
ℓ
u
−n
ℓ
ϑ
x
k
u
)
=
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,22
T
σ
ϑ,1
−L
ϑ,12
T
σ
ϑ,2
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,12
T
σ
ϑ,1
−L
ϑ,11
T
σ
ϑ,2
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
+
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,11
T
σ
ϑ,1
+L
ϑ,12
T
σ
ϑ,2
)(n
k
σ
x
ℓ
v
−n
ℓ
σ
x
k
v
)
−
1
W
n
∑
σ=1
n
∑
ϑ=1
(L
ϑ,22
T
σ
ϑ,2
+L
ϑ,12
T
σ
ϑ,1
)(n
k
σ
x
ℓ
u
−n
ℓ
σ
x
k
u
)
+
n
∑
σ=1
n
∑
ϑ=1
S
ϑ
σ,12
(n
k
σ
n
ℓ
ϑ
−n
ℓ
σ
n
k
ϑ
) −2
n
∑
σ=1
(−K
σ
)W¯ χ
proving the statement. ⊓⊔
µstability for minimal graphs
We want to construct a real number µ > 0 such that the following µstability in
equality is true
__
B
[∇ϕ[
2
dudv ≥ µ
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈ C
∞
0
(B, R) using conformal parameters. For this purpose we recall the
Ricci integrability condition
S
ϑ
σ,12
=
1
W
(L
σ,11
−L
σ,22
)L
ϑ,12
−
1
W
(L
ϑ,11
−L
ϑ,22
)L
σ,12
from chapter 3, formula (3.14).
298 13 The second variation
We estimate as follows
[S
ϑ
σ,12
[ ≤
1
4W
(L
σ,11
−L
σ,22
)
2
+
1
W
L
2
σ,12
+
1
4W
(L
ϑ,11
−L
ϑ,22
)
2
+
1
W
L
2
ϑ,12
=
1
4W
(L
2
σ,11
+2L
σ,11
L
σ,22
+L
2
σ,22
) −
1
W
(L
σ,11
L
σ,22
−L
2
σ,12
)
+
1
4W
(L
2
ϑ,11
+2L
ϑ,11
L
ϑ,22
+L
2
ϑ,22
) −
1
W
(L
ϑ,11
L
ϑ,22
−L
2
ϑ,12
)
leading to
[S
ϑ
σ,12
[ ≤H
2
σ
W −K
σ
W +H
2
ϑ
W −K
ϑ
W =−K
σ
W −K
ϑ
W
where the right hand side is nonnegative due to H
σ
≡0 for all σ = 1, . . . , n.
Theorem 13.4. Let X : B → R
n+2
be conformally reparametrization of minimal
graph in R
n+2
together with an ONF N. Assume furtermore
0 < χ
0
≤ χ =
x
1
u
x
2
v
−x
2
u
x
1
v
W
holds true with a real number χ
0
> 0 as well as
[n
1
σ
n
2
ϑ
−n
2
σ
n
1
ϑ
[ ≤N
1
<∞ for all σ, ϑ = 1, . . . , n
with real N
1
∈ [0, ∞) such that additionally
χ
0
< nN
1
is satisﬁed. Then the graph is stable in the sense of
__
B
[∇ϕ[ dudv ≥ µ
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R) where the stability constant µ is chosen such that
0 < µ ≤2 −
2nN
1
χ
0
.
The following proof shows that we can replace the assumption 0 < χ
0
≤ χ by
0 < χ if n = 1. It is clear that in this case N
1
= 0.
Corollary 13.3. A minimal graph X : B →R
3
is µstable with µ = 2, i.e. it holds
__
B
[∇ϕ[
2
dudv ≥2
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R) using conformal parameters (u, v) ∈ B.
13.4 µstability due to Schwarz for minimal graphs 299
The same stability inequality holds true if the minimal graph X : B →R
n+2
has
ﬂat normal bundle.
Corollary 13.4. A minimal graph X : B →R
n+2
with ﬂat normal bundle is µstable
with µ = 2, i.e. it holds
__
B
[∇ϕ[
2
dudv ≥2
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R) using conformal parameters (u, v) ∈ B.
Both these results arise from the following proof of the Theorem.
Proof of the Theorem. We know that the Weingarten equations imply
n
∑
σ=1
[∇N
σ
[
2
= 2
n
∑
σ=1
(−K
σ
)W = 2(−K)W
for the Gauss curvature K. Let ϕ ∈C
∞
0
(B, R) be given. We compute [∇(ϕχ
−1
)[
2
and
arrive at
[∇ϕ[
2
= χ
2
[∇(ϕχ
−1
)[
2
+div
_
ϕ
2
χ
u
χ
,
ϕ
2
χ
v
χ
_
−
ϕ
2
χ
△χ
= χ
2
[∇(ϕχ
−1
)[
2
+div
_
ϕ
2
χ
u
χ
,
ϕ
2
χ
v
χ
_
+2(−K)Wϕ
2
−
1
χ
n
∑
σ,ϑ=1
S
ϑ
σ,12
(n
1
σ
n
2
ϑ
−n
2
σ
n
1
ϑ
)ϕ
2
Partial integration gives
__
B
_
[∇ϕ[
2
−µ(−K)Wϕ
2
_
dudv
≥
__
B
χ
2
[∇(ϕχ
−1
)[
2
dudv +(2 −µ)
__
B
(−K)Wϕ
2
dudv
−
n
∑
σ,ϑ=1
__
B
1
χ
[S
ϑ
σ,12
[[n
1
σ
n
2
ϑ
−n
2
σ
n
1
ϑ
[ϕ
2
dudv
≥ (2 −µ)
__
B
(−K)Wϕ
2
dudv −
N
1
χ
0
n
∑
σ,ϑ=1
__
B
(−K
σ
−K
ϑ
)Wϕ
2
dudv
= (2 −µ)
__
B
(−K)Wϕ
2
dudv −
2nN
1
χ
0
__
B
(−K)Wϕ
2
dudv.
300 13 The second variation
Thus we are led to
__
B
_
[∇ϕ[
2
−µ(−K)Wϕ
2
_
dudv ≥
_
2 −µ −
2nN
1
χ
0
_
__
B
(−K)Wϕ
2
dudv
where the right hand side is nonnegative due to our assumptions. ⊓⊔
We want to give a representation of ∆χ for graphs with nonvanishing mean
curvature vector H = H(X, Z).
Let again H
σ
= H
σ
(X, N
σ
). Then it holds
∆χ = −2(2H
2
−K)Wχ +2
n
∑
σ=1
(H
σ,X
X
u
+H
σ,Z
N
σ,u
)(n
1
σ
x
2
v
−n
2
σ
x
1
v
)
−2
n
∑
σ=1
(H
σ,X
X
v
+H
σ,Z
N
σ,v
)(n
1
σ
x
2
u
−n
2
σ
x
1
u
)
if the normal bundle of the surface is ﬂat. For an explicite calculation we refer to
Fr¨ ohlich [64].
From this form it is ﬁnally possible to infer again a stability inequality of the
form
__
B
[∇ϕ[ dudv ≥ µ
__
B
(2H
2
−K)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R) with a suitable stability factor µ > 0.
13.5 Eigenvalue problems on manifolds
The number “2” in the stability inequality
__
B
[∇ϕ[
2
dudv ≥2
__
B
(−K)Wϕ
2
dudv
for minimal surfaces X : B→R
3
actually represents the ﬁrst eigenvalue of the spher
ical Laplace operator on the half sphere. In fact, a minimal graph is stable and its
spherical image is contained in a half sphere. Thus the ﬁrst eigenvalue of this Lapla
cian on the spherical image of the graph is greater than 2.
In the following we want to develop a basic theory of such eigenvalue problems,
and we present some applications of this theory to problems concerning stability
and µstability.
13.5 Eigenvalue problems on manifolds 301
The Rayleigh quotient
Let a smooth and regular mapping X : B →R
n+2
with a line element
ds
2
= g
11
du
2
+2g
12
dudv +g
22
dv
2
be given. Its trace in space is a twodimensional set M
2
⊂R
n+2
with smooth bound
ary ∂M
2
⊂R
n+2
which we identify with the mapping X = X(u, v).
Deﬁnition 13.4. The Laplace operator ∆
M
2 on the parametrically given surface
M
2
⊂R
n+2
is deﬁned by
∆
ds
2 φ :=
1
_
g
11
g
22
−g
2
12
_
_
∂
∂u
g
22
φ
u
−g
12
φ
v
_
g
11
g
22
−g
2
12
+
∂
∂v
g
11
φ
v
−g
12
φ
v
_
g
11
g
22
−g
2
12
_
_
(13.5)
for smooth functions φ ∈C
2
(B, R).
For real λ we now consider the eigenvalue problem
∆
ds
2 ϕ +λϕ = 0 in B, ϕ = 0 in ∂B. (13.6)
The function ϕ ∈C
2
0
(B, R) is called an eigenfunction to ∆
M
2 if there is a number
λ ∈Rsuch that (13.6) is satisﬁed. In this case the number λ is called an eigenvalue.
Multiplication of the eigenvalue equation (13.6) with ϕ followed by an integra
tion by parts gives us
λ
__
B
ϕ
2
W dudv =−
__
B
ϕ∆
ds
2 ϕW dudv =
__
B
∇
ds
2 (ϕ, ϕ)W dudv
with the area element W =
_
g
11
g
22
−g
2
12
and the Beltrami operator of ﬁrst kind
∇
ds
2 (φ, ψ) =
2
∑
i, j=1
g
i j
φ
u
i ψ
u
j
w.r.t. the line element ds
2
which we already met before. For such calculus rules
regarding nonEuclidean differential operators we refer the reader to Blaschke and
Leichtweiss [15].
Deﬁnition 13.5. The Rayleigh quotient is deﬁned as
R[ϕ] :=
__
B
∇
ds
2 (ϕ, ϕ)W dudv
__
B
ϕ
2
W dudv
.
302 13 The second variation
The inﬁmum of the Rayleigh quotient corresponds exactly to the ﬁrst eigenvalue
of the Laplacian on M
2
⊂R
n+2
, i.e.
λ
1
:= inf
ϕ∈V(B,R)
R[ϕ]
on the function space
V(B, R) :=
_
φ ∈ H
1,2
(B, R) ∩C
0
(B, R) : φ ,≡0, φ = 0 on ∂B
_
.
Existence and regularity of the ﬁrst eigenfunction
First of all we recall that there exist an eigenfunction ϕ ∈V(B, R) with eigenvalue
λ
1
> 0 at all, see e.g. Bandle [5] or Sakai [138] for the following result.
Proposition 13.2. The ﬁrst eigenvalue λ
1
of (13.6) is positive, and it has multiplicity
1, i.e. up to sign there is exactly one eigenfunction ϕ
1
for λ
1
. It particularly holds
ϕ
1
> 0 or ϕ
1
< 0 in B.
The uniqueness follows essentially from the usual maximum principle. Higher
regularity ϕ ∈ C
k
0
(B, R) can be inferred from Weyl’s lemma, see e.g. Hellwig [83]
or Jost [103].
The ﬁrst eigenvalue on spherical domains
As we will see immediately the ﬁrst eigenvalue and the associated eigenfunction of
the problem (13.6) on spherical domains can be employed successfully.
As usual we denote by S
2
the unit sphere. From Barbosa and do Carmo [8] we
cite the following monotonicity results.
Proposition 13.3. Let Ω ⊂S
2
and Θ ⊂S
2
be two simply connected, spherical com
pact domains with C
1
regular boundaries.
1. If Ω ⊂Θ then it holds
λ
1
(Ω) ≥λ
1
(Θ).
2. Let
S
2
ω
:=
_
Z ∈ S
2
: Z (0, 0, 1) ≥cosω
_
be a spherical cap on S
2
, and suppose that Ω ⊂S
2
satisﬁes
Area(Ω) = Area(S
2
ω
).
Then it holds
λ
1
(S
2
ω
) ≤λ
1
(Ω)
with equality if and only if Ω = S
2
ω
.
13.5 Eigenvalue problems on manifolds 303
In Sato [139] we ﬁnd some numerical values of the ﬁrst eigenvalue λ
1
(S
2
ω
) for
different angles ω which illustrate some of our later results. We present them here
in rounded form.
ω 0.03 0.16 0.31 0.63 0.79 0.94
λ
1
(S
2
ω
) 6422.2 225.39 59.73 14.24 8.92 6.20
ω 1.10 1.26 1.41 1.57 1.61 1.64
λ
1
(S
2
ω
) 4.45 3.31 2.57 2.0 1.90 1.80
ω 1.68 1.72 1.77 1.82 1.87 1.93
λ
1
(S
2
ω
) 1.70 1.60 1.50 1.40 1.30 1.20
ω 1.99 2.07 2.15 2.23 2.33 2.55
λ
1
(S
2
ω
) 1.10 1.00 0.90 0.80 0.70 0.60
ω 2.57 2.71 2.87 3.03 3.13
λ
1
(S
2
ω
) 0.50 0.40 0.30 0.20 0.10
The value ω = 1.57 represents the half sphere S
2
π
2
⊂ S
2
on which λ
1
is exactly
2 (this completes our remarks on stable minimal graphs at the beginning of this
section). Moreover there hold
lim
ω→0
+
λ
1
(S
2
ω
) = +∞, lim
ω→π
−
λ
1
(S
2
ω
) = 0.
From the above tables it seems evident that λ
1
decreases monotonically as ω
increases. In fact, for Ω ⊂Θ ⊂S
2
it holds the following monotonicity property
λ
1
(Ω) ≥λ
1
(Θ).
Furthermore, if Area(Ω) = Area(S
2
ω
) then
λ
1
(Ω) ≥λ
1
(S
2
ω
).
In other words: Symmetrical domains minimize the ﬁrst eigenvalue.
All these results can be found in more detail in the huge literature, e.g. Bandle
[5], Barbosa and do Carmo [8] or Sakai [138]. We also want to refer to the classical
work Courant and Hilbert [39].
304 13 The second variation
The ﬁrst eigenvalue and the Gauss curvature
We need some further comparison results for the ﬁrst eigenvalue λ
1
on curved man
ifolds which we also take from Barbosa and do Carmo [8].
Let M
2
(K
0
) ⊂ R
n+2
denote a smooth and regular surface with constant Gauss
curvature K
0
>0, and let a further surface M
2
⊂R
n+2
with variable Gauss curvature
be given.
Proposition 13.4. Let D ⊂ M
2
be compact and simply connected, and let K ≤ K
0
be the Gaussian curvature of M
2
with a real constant K
0
> 0. Furthermore, let
D
∗
⊂M
2
(K
0
) be a geodesic disc such that
Area(D
∗
) = Area(D).
Then it holds
λ
1
(D) ≥λ
1
(D
∗
).
In other words, the ﬁrst eigenvalue is minimized on constantly curved manifolds.
Proposition 13.5. Let D ⊂ (M
2
, ds
2
) be compact and simply connected with M
2
⊂
R
n+2
being a smooth, simply connected and regular surface with metric ds
2
, and let
ψ be a nonnegative C
2
function on M
2
which vanishes in at most isolated points.
Denote by λ
1
> 0 the ﬁrst eigenvalue to the problem
△
ds
2 ϕ +λϕ = 0 in D, ϕ = 0 on ∂D.
Now consider on M
2
the new metric d ˆ s
2
:=ψds
2
. Suppose that the Gauss curvature
´
K of (M, d ˆ s
2
) satisﬁes
´
K ≤K
0
with a real constant K
0
> 0. Then it holds
λ
1
≥λ
1
(D
∗
)
with D
∗
⊂M
2
(K
0
) being a geodesic disc with the same area as D in (M, d ˆ s
2
).
13.6 µstability due to Ruchert, Barbosa and do Carmo
Proving the existence of a lower bound χ
0
>0 for the Jacobian of the plane mapping
(x
1
(u, v), x
2
(u, v)) in the form
0 < χ
0
≤ χ =
x
1
u
x
2
v
−x
1
v
x
2
u
W
as supposed in Theorem13.4 is a rather difﬁcult problem. In case n =1 of one codi
mension this assumption demands that the third component N
3
of the unit normal
vector of the minimal graph is greater than χ
0
> 0.
13.6 µstability due to Ruchert, Barbosa and do Carmo 305
In this section we want to prove a further stability result which comprises spectral
methods from Barbosa and do Carmo [8] und Ruchert [137].
Theorem 13.5. Let the conformally parametrized minimal surface X with a torsion
free ONF N be given. For a real number κ
0
> 0 assume that
Q :=
__
B
(κ
0
−K)W dudv <ω
0
with a real constant ω
0
∈ (0, 4π). Let S
2
ω
⊂S
2
denote the spherical cap of spherical
angle ω such that
Area(S
2
ω
) =ω
0
,
and let µ >0 be the smallest eigenvalue of the spherical Laplacian ∆
S
2 on S
2
ω
. Then
the surface X is µstable with this number µ in the following sense
__
B
[∇ϕ[
2
dudv ≥ µ
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R).
For the proof we recall some important facts from chapter 6: For the curvature
χ
´
K = K−
1
2W
△logχ, K =
2
W
2
_
W
w
W
w
W
−W
ww
_
,
of the FubiniStudy metric we already have proved
´
K ≤1, see Theorem 6.3.
Proof of the theorem. Let
´
△ denote the LaplaceBeltrami operator w.r.t. the metric
´ g
i j
:= χg
i j
. Furthermore,
´
λ
1
means the ﬁrst eigenvalue to the problem
´
△ϕ +λϕ = 0 in B, ϕ = 0 on ∂B,
and let λ
∗
1
> 0 be the ﬁrst eigenvalue of
△
∗
ϕ
∗
+λ
∗
ϕ
∗
= 0 in S
2
ω
, ϕ
∗
= 0 on ∂S
2
ω
with a spherical cap S
2
ω
⊂S
2
such that Area(S
2
ω
) =Q. Since
´
K ≤1 we infer λ
∗
1
≤
´
λ
1
,
and by our assumption it holds
µ <λ
∗
1
≤
´
λ
1
≤
__
B
[∇ϕ[
2
dudv
__
B
(κ
0
−K)Wϕ
2
dudv
for any ϕ ∈C
∞
0
(B, R).
306 13 The second variation
It follows
__
B
[∇ϕ[
2
dudv > µ
__
B
(κ
0
−K)Wϕ
2
dudv > µ
__
B
(−K)Wϕ
2
dudv,
and the statement is proved. ⊓⊔
We want to clarify the connection of our theorem with a theorem from Barbosa
and do Carmo [8]:
Theorem 13.6. For a real a ∈ (0, 2] assume that
´
K ≤ a for the Gaussian curvature
of the spherical metric from our theorem above. Then a minimal surface is stable
with µ = 2 (with no curvature restrictions on the normal bundle) if
__
B
(−K)W dudv ≤
4π
1 +a
.
In fact, Barbosa and do Carmo proved
´
K ≤2
as we already mentioned in chapter 6. Our result is thus a improvement under the
assumption of ﬂat normal bundle where we get a sharper bound on
´
K.
Further stability results for weighted minimal surfaces in R
3
using these methods
from the spectral theory of the Laplacian on manifolds are presented in chapter 15.
To conclude this section we want to refer to Spruck [150] who estabished a sta
bility criteria for minimal surfaces with sufﬁciently small curvatura integra using a
generalized Sobolev inequality from Michael and Simon [119].
13.7 Calibration forms
We brieﬂy and incompletely discuss the method of calibration forms for minimal
surfaces.
Let X = (x, y, ζ(x, y)) be a minimal graph in R
3
, i.e. ζ satisﬁes the minimal sur
face equation (see chapter 9)
div
_
∇ζ
_
1 +[∇ζ[
2
_
= 0 in Ω ⊂R
2
.
Now let
¯
ζ be a comparison function such that
¯
ζ =ζ on the boundary ∂Ω. It holds
__
Ω
∇ζ ∇(ζ −
¯
ζ)
_
1 +[∇ζ[
2
= 0.
13.7 Calibration forms 307
This follows after partial integration taking (
¯
ζ −ζ)[
∂Ω
into account. Thus we can
estimate
__
Ω
_
1 +[∇ζ[
2
dxdy =
__
Ω
1 +[∇ζ[
2
_
1 +[∇ζ[
2
dxdy
=
__
Ω
1 +∇ζ ∇
¯
ζ
_
1 +[∇ζ[
2
≤
__
Ω
_
1 +[∇
¯
ζ[
2
dxdy
because
1 +∇ζ ∇
¯
ζ ≤
_
1 +[∇ζ[
2
_
1 +[∇
¯
ζ[
2
,
see Giusti [72], section 13.7.
But what happens in case of higher codimension?
Let us consider exemplarily the special case n =2 of minimal surfaces in R
4
. On
the manifold M
2
⊂ R
4
, regarded as the image under the mapping X : B →R
4
, we
consider a 2form ω = ω(X, Y ) for tangent vector ﬁelds X and Y . Such a form
is deﬁned as an alternating and differentiable mapping on the manifold such that
1. ω(X +Y , Z ) =ω(X, Z ) +ω(Y , Z);
2. ω( f X, Y ) = f ω(X, Y ) for differentiable f ;
3. ω(X, Y ) =−ω(Y , X).
Please consult the wide literature for detailed discussions on differential forms, e.g.
Blaschke and Leichtweiss [15], Heil [78] or Cartans original work [23].
Next, for complex valued vectors x, y ∈ C
2
we deﬁne the K¨ ahler form
ω(x, y) = Im
2
∑
j=1
x
j
y
j
.
The point is that this form satisﬁes
ω(X
u
/
√
W, X
v
/
√
W) = 1
if and only if X = (Φ(w),Ψ(w)) is conformally parametrized, i.e. X
u
/
√
W and
X
v
/
√
W is an orthonormal basis of the tangent space, and Φ and Ψ are holo
morphic functions. Otherwise it holds ω ≤ 1; remember particularly our example
X(w) = (w, w
2
)!
A differential form with these properties is called a calibration form.
Now denoting by X
∗
ω the pull back of ω (deﬁned on the surface) onto the do
main B by the mapping X, we compute using Stokes’ theoremand Poincare’s lemma
308 13 The second variation
(identify X : B →R
4
with the imbedding of the image X(B) into R
4
)
A[X] =
_
B
X
∗
1 =
_
B
X
∗
ω =
_
B
d(X
∗
τ) =
_
∂B
X
∗
τ =
_
∂B
¯
X
∗
τ =
_
B
¯
X
∗
ω ≤A[
¯
X]
under the assumption that
¯
X = X on ∂B.
Theorem 13.7. Let X = (Φ,Ψ) with Φ and Ψ holomorphic functions in B. Then X
minimizes the area among all mappings
¯
X : B →R
4
with the property X(w) =
¯
X(w)
for all w ∈ ∂B.
These calculations are taken from Eschenburg and Jost [52]. See also the math
ematics resource website Wolfram MathWorld for a good introduction to K¨ ahler
forms and calibration forms.
Chapter 14
Energy estimates
14.1 Geodesic discs
14.2 The area of µstable geodesic discs
14.3 An area estimate for minimal graphs
14.4 Area estimates via the curvatura integra
14.5 The area of graphs with prescribed mean curvature
14.6 The isoperimetric inequality
14.7 The spherical energy of µstable geodesic discs
310 14 Energy estimates
In this chapter we establish various estimates for the area of surfaces with prescribed mean curva
ture and for their spherical energies. We mainly consider µstable immersions and surface graphs.
14.1 Geodesic discs
We have frequently used geodesic discs: Assume the immersion X : B →R
n+2
(or a
part of it) represents a geodesic disc B
r
(X
0
) of geodesic radius r >0 about the center
X
0
∈ R
n+2
. Using geodesic polar coordinates ρ and ϕ we get a reparametrization of
the form
X = X(ρ, ϕ): [0, r] [0, 2π] −→R
n+2
.
For the corresponding line element ds
2
p
we have
ds
2
p
= X
ρ
(ρ, ϕ)dρ
2
+2X
ρ
(ρ, ϕ) X
ϕ
(ρ, ϕ)dρdϕ +X
ϕ
(ρ, ϕ)
2
dϕ
2
= dρ
2
+P(ρ, ϕ)dϕ
2
(14.1)
with a continuously differentiable function P satisfying the properties
P(ρ, ϕ) > 0 for all (ρ, ϕ) ∈ (0, r] [0, 2π) (14.2)
as well as
lim
ρ→0
+
P(ρ, ϕ) = 0, lim
ρ→0
+
∂
∂ρ
_
P(ρ, ϕ) = 1 for all ϕ ∈ [0, 2π). (14.3)
For a comprehensive introduction of geodesic discs we want to refer the reader
to Klingenberg [106]. For our purposes the following remarks are already sufﬁcient:
1. For immersions with nonpositive Gaussian curvature
K ≤0,
the socalled exponential map, which sends tangential vectors V ∈ T
X
(w) injec
tively onto geodesic curves on the surface, is globally injective. This particularly
means, given a simplyconnected and complete surface without boundary, we
can introduce geodesic discs for arbitrary r > 0.
312 14 Energy estimates
2. I some geodesic curve on a surface with positive Gaussian curvature
K ≥K
0
> 0
has length greater than
π
√
K
0
then the exponential map is not injectiv. Thus
π
√
K
0
is the supremum of our geodesic radius r > 0.
We particularly refer Klingenberg [106], Theorems 3.4.13 and 3.4.16.
14.2 The area of µstable geodesic discs
In the following we want to establish estimates for the area of geodesic discs which
are additionally µstable in the following sense
__
B
∇
ds
2 (ψ, ψ)W dudv
≥ µ
__
B
(q −K)Wψ
2
dudv −α(n)
2
∑
i, j=1
n
∑
σ,ϑ=1
__
B
g
i j
T
ϑ
σ,i
T
ϑ
σ, j
Wψ
2
dudv
(14.4)
for all ψ ∈C
∞
0
(B, R) and with a real number α(n) ≥0, see formula (13.3).
Using geodesic polar coordinates ρ and ϕ it takes the form
r
_
0
2π
_
0
∇
ds
2
P
(ψ, ψ)
_
P(ρ, ϕ)dρdϕ
=
r
_
0
2π
_
0
P(ρ, ϕ)ψ
ρ
(ρ, ϕ)
2
+ψ
ϕ
(ρ, ϕ)
2
P(ρ, ϕ)
_
P(ρ, ϕ)dρdϕ
≥ µ
r
_
0
2π
_
0
(q −K)ψ(ρ, ϕ)
2
_
P(ρ, ϕ)dρdϕ
−α(n)
n
∑
σ,ϑ=1
r
_
0
2π
_
0
_
P(ρ, ϕ)(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
ψ(ρ, ϕ)
2
dρdϕ.
The method of the proof of our ﬁrst result follows essentially Gulliver [76] and
Sauvigny [141].
14.2 The area of µstable geodesic discs 313
Theorem 14.1. Let the immersion X represent a geodesic disc B
r
(X
0
) ⊂ R
n+2
of
geodesic radius r > 0. Assume that X is µstable in the sense of (14.4) with real
µ >
1
2
.
Then it holds the area estimate
A[X] ≤
1
2µ −1
_
2πµ +2α(n)T
X
[N]
_
r
2
(14.5)
with the total torsion T
X
[N] of some chosen ONF N, i.e. the area of the geodesic
disc growth quadratically in r.
Proof. 1. First we introduce geodesic polar coordinates (ρ, ϕ) as above. Then along
curves ρ = const the integral formula of Bonnet and Gauss can be written as
2π
_
0
κ
g
(ρ, ϕ)
_
P(ρ, ϕ)dϕ +
ρ
_
0
2π
_
0
K(τ, ϕ)
_
P(τ, ϕ) dτdϕ = 2π
with the geodesic curvature κ
g
along curves with ρ = const, see e.g. Blaschke
and Leichtweiss [15]. Furthermore, from the same source [15], §81 we take
κ
g
(ρ, ϕ)
_
P(ρ, ϕ) =
∂
∂ρ
_
P(ρ, ϕ), (ρ, ϕ) ∈ (0, r] [0, 2π),
along such a curve, and we arrive at
∂
∂ρ
2π
_
0
_
P(ρ, ϕ)dϕ =
2π
_
0
κ
g
(ρ, ϕ)
_
P(ρ, ϕ)dϕ
= 2π −
ρ
_
0
2π
_
0
K(τ, ϕ)
_
P(τ, ϕ) dτdϕ
for all ϕ ∈ (0, r].
2. We abbreviate the left hand side of the latter identity by
L(ρ) :=
2π
_
0
_
P(ρ, ϕ)dϕ, ϕ ∈ (0, r].
For its ﬁrst and second derivatives we compute
L
′
(ρ) = 2π −
2π
_
0
K(τ, ϕ)
_
P(τ, ϕ) dτdϕ
314 14 Energy estimates
as well as
L
′′
(ϕ) =−
2π
_
0
K(ρ, ϕ)
_
P(ρ, ϕ)dϕ.
3. Now consider the special test function
Φ(ρ) := 1 −
ρ
r
, ϕ ∈ (0, r],
satisfying
∇
ds
2
P
(Φ, Φ) =
P(ρ, ϕ)Φ
ρ
(ϕ)
2
+Φ
ϕ
(ρ)
P(ρ, ϕ)
=Φ
′
(ρ)
2
w.r.t. the line element ds
2
P
from (14.1). Here ∇
ds
2
P
(Φ, Φ) denotes the Beltrami
operator of ﬁrst kind w.r.t. ds
2
P
. We estimate as follows
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ =
r
_
0
2π
_
0
Φ
′
(ρ)
2
_
P(ρ, ϕ)dρdϕ
≥ µ
r
_
0
2π
_
0
(q −K)Φ(ρ)
2
_
P(ρ, ϕ)dρdϕ
−α(n)
n
∑
σ,ϑ=1
r
_
0
2π
_
0
_
P(ρ, ϕ)(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
dρdϕ
≥ µ
r
_
0
L
′′
(ρ)Φ(ρ)
2
dρ
−α(n)
n
∑
σ,ϑ=1
r
_
0
2π
_
0
_
P(ρ, ϕ)(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
dρdϕ
due the µstability condition. Partial integration together with (14.3) gives
r
_
0
L
′′
(ρ)Φ(ρ)
2
dρ = L
′
(ρ)Φ(ρ)
¸
¸
¸
ρ=r
ρ=0
+
−2
r
_
0
L
′
(ρ)Φ(ρ)Φ
′
(ρ)dρ
= −2π −2L(ρ)Φ(ρ)Φ
′
(ρ)
¸
¸
¸
ρ=r
ρ=0
+
+2
r
_
0
L(ρ)Φ
′
(ρ)
2
dρ
= −2π +2
r
_
0
L(ρ)Φ
′
(ρ)
2
dρ.
14.3 An area estimate for minimal graphs 315
Since Φ ∈ [0, 1] we get
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ ≥ −2πµ +2µ
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ
−α(n)
n
∑
σ,ϑ=1
r
_
0
2π
_
0
_
P(ρ, ϕ)(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
Φ(ρ)
2
dρdϕ
≥ −2πµ +2µ
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ
−α(n)
n
∑
σ,ϑ=1
r
_
0
2π
_
0
_
P(ρ, ϕ)(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
dρdϕ.
Rearranging thus yields
(2µ −1)
r
_
0
Φ(ρ)
2
L(ρ)dρ
≤ 2πµ +α(n)
n
∑
σ,ϑ=1
r
_
0
2π
_
0
_
P(ρ, ϕ)(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
dρdϕ,
and the statement follows taking
1
r
2
A[X] =
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ
into account. ⊓⊔
The question whether a similar estimate is true for values µ <
1
2
remains open.
We refer the reader to a discussion in FischerColbrie and Schoen [60].
14.3 An area estimate for minimal graphs
Following Bergner and Fr¨ ohlich [13] want to show how one could establish area
bounds for minimal graphs independently of stability, total torsion, or other things.
Theorem 14.2. Let B
r
(X
0
) denote a geodesic disc of the minimal graph
X(w) = (w, w
2
), w ∈ B
R
:=
_
w ∈ C
2
: [w[ ≤R
_
,
with radius r > 0 and centered at X
0
:= X(0, 0).
316 14 Energy estimates
Then it holds
A[B
r
(X
0
)] ≤192πr
2
for sufﬁciently large r > 0. (14.6)
Proof. Denote by
c(t) =
_
u(t), v(t)
_
⊂B
R
, t ∈ [0, T],
an arclength parametrized and smooth curve with the properties
(i) c(0) = (0, 0) and c(T) ∈ ∂B
R
;
(ii) [ ˙ c(t)[
2
= 1 for all t ∈ [0, T].
The set of all such curves is abbreviated with C. Now let c(t) be chosen from C.
We will establish a lower bound for the length L[X ◦c] of its image on the minimal
graph. W.l.o.g. we assume [u(T)[ ≥[v(T)[. It particularly holds
[u(T)[ ≥
R
2
.
We deﬁne
t
∗
:= sup
_
t ∈ (0, T) : [u(t)[ ≤
R
4
_
. (14.7)
Note that T −t
∗
≥
R
4
for the special arclength parametrization. We estimate now as
follows:
L[X ◦ c] =
T
_
0
_
1 +4u
2
+4v
2
dt ≥
T
_
0
_
1 +4u
2
dt ≥
T
_
t
∗
_
1 +4u
2
dt
≥ 2
T
_
t
∗
[u(t)[ dt ≥
R
2
(T −t
∗
) ≥
R
2
8
.
The case [v(T)[ ≥[u(T)[ can be treated analogously. Now let
r := min
c∈C
L[X ◦ c].
Due to the graphical property of the minimal surface, the geodesic disc B
r
(X
0
)
projects onetoone into the disc B
R
. Together with (14.7) we get
A [B
r
(X
0
)] ≤
__
B
R
(1 +4u
2
+4v
2
)dudv =
2π
_
0
R
_
0
(1 +4ρ
2
)ρ dρdϕ
= 2π
_
R
2
2
+R
4
_
≤ 3πR
4
≤ 3 8
2
πr
2
= 192πr
2
for all R ≥1 proving the statement. ⊓⊔
14.4 Area estimates via the curvatura integra 317
14.4 Area estimates via the curvatura integra
With the next result and its proof we follow again Sauvigny [141] and use the in
tegral formula of Bonnet and Gauss to infer an area estimate in terms of the total
curvature of the immersion.
Theorem 14.3. Let the immersion X represent a geodesic disc B
r
(X
0
) of radius
r > 0. Suppose that its Gaussian curvature K satisﬁes
K(ρ, ϕ) ≤K
0
for all (ρ, ϕ) ∈ (0, r] [0, 2π)
with a real constant K
0
∈ [0, +∞). Then it holds
A[X] ≤r
2
_
_
_
π +
1
2
r
_
0
2π
_
0
_
K
0
−K(ρ, ϕ)
_
_
P(ρ, ϕ)dρdϕ
_
_
_
(14.8)
for its area.
Proof. As in section 14.2 we consider curves ρ = const
∂
∂ρ
2π
_
0
_
P(ρ, ϕ)dϕ = 2π −
ρ
_
0
2π
_
0
K(τ, ϕ)
_
P(τ, ϕ) dτdϕ
≤ 2π +
ρ
_
0
2π
_
0
_
K
0
−K(τ, ϕ)
_
_
P(ρ, ϕ)dτdϕ
≤ 2π +
r
_
0
2π
_
0
_
K
0
−K(τ, ϕ)
_
_
P(τ, ϕ) dτdϕ.
Integration over the radius ρ yields
2π
_
0
_
P(ρ, ϕ)dϕ ≤ρ
_
_
_
2π +
r
_
0
2π
_
0
_
K
0
−K(τ, ϕ)
_
_
P(τ, ϕ) dτdϕ
_
_
_
.
A ﬁnal integration over ρ = 0. . . r proves the statement. ⊓⊔
In particular, for minimal surfaces with K
0
= 0 we infer the estimate
A[X] ≤r
2
_
_
_
π +
1
2
r
_
0
2π
_
0
_
−K(ρ, ϕ)
__
P(ρ, ϕ)dρdϕ
_
_
_
.
We immediately obtain an area estimate in terms of the total geodesic curvature
κ
g
of the boundary curve of the immersion X.
318 14 Energy estimates
First it holds [κ
g
[ ≤ κ for the nonnegative spatial curvature of the boundary.
From the integral formula of Bonnet and Gauss we then obtain
__
B
(−K)W dudv =
_
∂B
κ
g
(s)ds −2π ≤
_
∂B
κ(s)ds −2π.
Inserting this inequality into the estimate of the previous theorem shows
A[X] ≤πr
2
+
K
0
2
A[X] r
2
−πr
2
+
r
2
2
_
∂B
κ(s)ds.
Corollary 14.1. It holds
A[X] ≤
K
0
2
A[X]r
2
+
r
2
2
_
∂B
κ(s)ds
with K
0
∈ [0, +∞) from Theorem 14.3.
Consider again the case K
0
= 0 for a minimal surface. We infer
A[X] ≤
r
2
2
_
∂B
κ(s)ds.
In this special context we would like to mention Fenchel’s theorem (see Fenchel
[56]) which states
_
∂B
κ(s)ds ≥2π
with equality if and only if the boundary curve is a plane and convex curve.
14.5 The area of graphs with prescribed mean curvature
We continue our investigations on the Gullivertype functional
__
Ω
Γ(x, y, ζ)
_
1 + p
2
+q
2
+ p
2
q
2
−(p q)
2
dxdy
for graphs ζ =ζ(x, y) with the settings
ζ = (ζ
1
, . . . , ζ
n
), p =ζ
x
∈ R
n
, q =ζ
y
∈ R
n
.
Our aim is to establish an upper bound for the area of critical points of the associated
variational problem.
14.5 The area of graphs with prescribed mean curvature 319
From Corollary 11.1 we knwo that critical points are solutions of
div
(p
σ
, q
σ
)
W
= 2H
σ
(X, N
σ
)
_
1 + p
2
σ
+q
2
σ
W
+
1
ΓW
_
_
p
2
+q
2
+ p
2
q
2
−(p q)
2
¸
Γ
z
σ
−
n
∑
ω=1
(p
σ
p
ω
+q
σ
q
ω
)Γ
z
ω
_
−
1
Γ
div
_
p
σ
q
2
−q
σ
(p q)
W
Γ,
q
σ
p
2
−p
σ
(p q)
W
Γ
_
(14.9)
for all σ = 1, . . . , n with the mean curvature ﬁeld
H
σ
(X, N
σ
) =
Γ
X
(X) N
σ
2Γ(X)
.
The case Γ(X) ≡1 represents the minimal surface case
div
(p
σ
, q
σ
)
W
+div
_
p
σ
q
2
−q
σ
(p q)
W
,
q
σ
p
2
−p
σ
(p q)
W
_
= 0 in Ω (14.10)
for all σ = 1, . . . , n. Moreover, the second divergence on the left hand side term
vanishes identically if n = 1.
In the following we suppose that Γ depends only on (x, y) ∈ Ω, i.e.
Γ =Γ(x, y) ∈C
1
(Ω, R).
Theorem 14.4. Let ζ = (ζ
1
, . . . , ζ
n
) be a solution of (14.9) with a density function
Γ =Γ(x, y) ∈C
1
(Ω, R). Let
0 <Γ
0
≤Γ(x, y) ≤Γ
1
< +∞ in Ω,
Γ
C
1
(Ω)
=: Γ
2
< +∞,
and assume furthermore
Λ := 1 −
√
2n
2
Γ
2
Γ
0
max
σ=1,...,n
ζ
σ

C
0
(Ω)
> 0.
Then it holds
Λ A[ζ] ≤ [Ω[ +n[∂Ω[ max
σ=1,...,n
ζ
σ

C
0
(∂Ω)
+2nh
0
[Ω[ max
σ=1,...,n
ζ
σ

C
0
(Ω)
+
√
2n
2
[∂Ω[ max
σ=1,...,n
ζ
σ

C
0
(∂Ω)
D
⊤
ζ
σ

C
0
(∂Ω)
(14.11)
with the circumference [∂Ω[ of the boundary ∂Ω ⊂R
2
, the area [Ω[ of Ω, and the
tangential derivative D
⊤
ζ
σ
for σ = 1, . . . , n along ∂Ω.
320 14 Energy estimates
Finally we set
h
0
:= max
(x,y)∈Ω
max
[N[=1
[H(x, y, N)[.
The second row in this area estimate does not appear in case of one codimension.
We would also set Λ := 1 if n = 1.
The special situation for minimal graphs in the contents of the
Corollary 14.2. Let ζ solve the minimal surface system (14.10). Then it holds
A[ζ] ≤ [Ω[ +n[∂Ω[ max
σ=1,...,n
ζ
σ

C
0
(∂Ω)
+
√
2n
2
[∂Ω[ max
σ=1,...,n
ζ
σ

C
0
(∂Ω)
D
⊤
ζ
σ

C
0
(∂Ω)
.
(14.12)
We want to point out the dependence of the C
1
data of the solution in contrast to
the case n = 1 where it holds
A[ζ] ≤[Ω[ +c[∂Ω[ζ
C
0
(∂Ω)
with some suitable constant c > 0. For illustration consider the minimal graph
(w, w
n
) ∈ R
4
: For all n ∈ N it holds ζ
σ

C
0
(B)
=
√
2 independently of the area
actually enclosed.
Now we come to the proof of our theorem.
Proof. 1. First we sum up the n identities
ζ
σ
div
∇ζ
σ
W
= div
ζ
σ
∇ζ
σ
W
−
[∇ζ
σ
[
2
W
, σ = 1, . . . , n,
to get
n
∑
σ=1
ζ
σ
div
∇ζ
σ
W
=
n
∑
σ=1
div
ζ
σ
∇ζ
σ
W
−
1
W
n
∑
σ=1
(p
2
σ
+q
2
σ
)
=
n
∑
σ=1
div
ζ
σ
∇ζ
σ
W
−
1
W
(p
2
+q
2
).
For the area element W =
_
1 + p
2
+q
2
+ p
2
q
2
−(p q)
2
we compute
1
W
−W =
1 −
_
1 + p
2
+q
2
+ p
2
q
2
−(p q)
2
¸
W
= −
p
2
+q
2
W
−
p
2
q
2
−(p q)
2
W
.
This yields
W =
1
W
+
n
∑
σ=1
div
ζ
σ
∇ζ
σ
W
−
n
∑
σ=1
ζ
σ
div
∇ζ
σ
W
+
p
2
q
2
−(p q)
2
W
. (14.13)
14.5 The area of graphs with prescribed mean curvature 321
Now we insert the EulerLagrange equations from (14.9). First multiply all these
equations by ζ
σ
and sum up to get (note Γ
z
σ
≡0)
n
∑
σ=1
ζ
σ
div
∇ζ
σ
W
= 2
n
∑
σ=1
H
σ
(x, y, N
σ
)ζ
σ
_
1 + p
2
σ
+q
2
σ
W
−
n
∑
σ=1
ζ
σ
div
_
p
σ
q
2
−q
σ
(p q)
W
,
q
σ
p
2
−p
σ
(p q)
W
_
−
n
∑
σ=1
_
p
σ
q
2
−q
σ
(p q)
Γ(x, y)W
ζ
σ
Γ
x
(x, y) +
q
σ
p
2
−p
σ
(p q)
Γ(x, y)W
ζ
σ
Γ
y
(x, y)
_
.
In the second row we take into account
div
_
p
σ
q
2
−q
σ
(p q)
W
ζ
σ
,
q
σ
p
2
−p
σ
(p q)
W
ζ
σ
_
= ζ
σ
div
_
p
σ
q
2
−q
σ
(p q)
W
,
q
σ
p
2
−p
σ
(p q)
W
_
+
p
2
σ
q
2
−p
σ
q
σ
(p q)
W
+
q
2
σ
p
2
−p
σ
q
σ
(p q)
W
,
and after summation
n
∑
σ=1
ζ
σ
div
_
p
σ
q
2
−q
σ
(p q)
W
,
q
σ
p
2
−p
σ
(p q)
W
_
=
n
∑
σ=1
div
_
p
σ
q
2
−q
σ
(p q)
W
ζ
σ
,
q
σ
p
2
−p
σ
(p q)
W
ζ
σ
_
−
2
W
_
p
2
q
2
−(p q)
2
_
.
2. Now (14.13) can be written in the form
W =
1
W
+
n
∑
σ=1
div
ζ
σ
∇ζ
σ
W
−
p
2
q
2
−(p q)
2
W
−2
n
∑
σ=1
H
σ
(x, y, N
σ
)ζ
σ
_
1 + p
2
σ
+q
2
σ
W
+
n
∑
σ=1
div
_
p
σ
q
2
−q
σ
(p q)
W
ζ
σ
,
q
σ
p
2
−p
σ
(p q)
W
ζ
σ
_
+
n
∑
σ=1
_
p
σ
q
2
−q
σ
(p q)
Γ(x, y)W
ζ
σ
Γ
x
(x, y) +
q
σ
p
2
−p
σ
(p q)
Γ(x, y)W
ζ
σ
Γ
y
(x, y)
_
.
(14.14)
322 14 Energy estimates
Then we can integrate as follows.
(i) We start with
__
Ω
1
W
dxdy ≤[Ω[ because
1
W
≤1. (14.15)
(ii) Next for the second term we have
n
∑
σ=1
__
Ω
div
ζ
σ
∇ζ
σ
W
dxdy ≤
n
∑
σ=1
_
∂Ω
[∇ζ
σ
ν[
W
[ζ
σ
[ ds
≤ n[∂Ω[ max
σ=1,...,n−2
ζ
σ

C
0
(∂Ω)
(14.16)
with the outer unit normal vector ν at the boundary ∂Ω, taking
_
1 + p
2
σ
+q
2
σ
W
≤1 for allσ = 1, . . . , n
into account.
(iii) The third term is nonpositive and can be estimated by 0 :
−
p
2
q
2
−(p q)
2
W
≤0.
(iv) Likewise it holds
2
n
∑
σ=1
__
Ω
H
σ
(x, y, N
σ
)ζ
σ
_
1 + p
2
σ
+q
2
σ
W
dxdy
≤ 2nh
0
[Ω[ max
σ=1,...,n
ζ
σ

C
0
(Ω)
.
(14.17)
(v) Now we come back to the second row in (14.14). Note that
p
σ
q
2
−q
σ
(p q) =
n
∑
θ=1
(p
σ
q
θ
−p
θ
q
σ
)q
θ
,
q
σ
p
2
−p
σ
(p q) =
n
∑
θ=1
(q
σ
p
θ
−q
θ
p
σ
)p
θ
.
(14.18)
After multiplication by p
σ
or q
σ
and following summation we get
p
2
q
2
−(p q)
2
=
1
2
n
∑
σ,θ=1
(p
σ
q
θ
−p
θ
q
σ
)
2
. (14.19)
14.5 The area of graphs with prescribed mean curvature 323
Thus we estimate as follows:
n
∑
σ=1
__
Ω
div
_
p
σ
q
2
−q
σ
(p q)
W
ζ
σ
,
q
σ
p
2
−p
σ
(p q)
W
ζ
σ
_
dxdy
=
n
∑
σ,ω=1
__
Ω
div
_
(p
σ
q
ω
−p
ω
q
σ
)q
ω
W
ζ
σ
, −
(p
σ
q
ω
−p
ω
q
σ
)p
ω
W
ζ
σ
_
dxdy
=
n
∑
σ,ω=1
_
∂Ω
(p
σ
q
ω
−p
ω
q
σ
)ζ
σ
W
(q
ω
, −p
ω
) ν ds
≤
√
2
n
∑
σ,ω=1
_
∂Ω
[ζ
σ
[[D
⊤
ζ
ω
[ ds
≤
√
2n
2
[∂Ω[ max
σ=1,...,n
ζ
σ

C
0
(∂Ω)
[D
⊤
ζ
σ

C
0
(∂Ω)
.
(14.20)
(vi) To estimate the last terms we come back to (14.18), (14.19) and compute
[p
σ
q
2
−q
σ
(p q)[
W
≤
n
∑
ϑ=1
[p
σ
q
ϑ
−q
σ
p
ϑ
[[q
ϑ
[
W
≤
√
2
n
∑
ϑ=1
_
p
2
q
2
−(p q)
2
W
[q
ϑ
[,
and analogously
[q
σ
p
2
−p
σ
(p q)[
W
≤
√
2
n
∑
ϑ=1
_
p
2
q
2
−(p q)
2
W
[p
ϑ
[.
Thus we arrive at
n
∑
σ=1
__
Ω
_
p
σ
q
2
−q
σ
(p q)
Γ(x, y)W
ζ
σ
Γ
x
(x, y) +
q
σ
p
2
−p
σ
(p q)
Γ(x, y)W
ζ
σ
Γ
y
(x, y)
_
dxdy
≤
n
∑
σ=1
Γ
2
__
Ω
[ζ
σ
[
_
[p
σ
q
2
−q
σ
(p q)[
Γ(x, y)W
+
[q
σ
p
2
−p
σ
(p q)[
Γ(x, y)W
_
dxdy
≤
√
2n
2
Γ
2
Γ
0
max
σ=1,...,n
ζ
σ

C
0
(B)
__
Ω
_
1 + p
2
+q
2
+ p
2
q
2
−(p q)
2
dxdy.
(14.21)
For this estimate we explicitely need the assumption Γ = Γ(x, y). Otherwise
there remain derivatives p
ϑ
and q
ϑ
quadratically and we are not able to estimates
the area element.
324 14 Energy estimates
3. Putting (14.15), (14.16), (14.17), (14.20) and (14.21) together we conclude
A[ζ] ≤ [Ω[ +n[∂Ω[ max
σ=1,...,n
ζ
σ

C
0
(∂Ω)
+2nh
0
[Ω[ max
σ=1,...,n
ζ
σ

C
0
(Ω)
+
√
2(n −2)
2
[∂Ω[ max
σ=1,...,n
ζ
σ

C
0
(∂Ω)
D
⊤
ζ
σ

C
0
(∂Ω)
+
√
2n
2
Γ
2
Γ
0
max
σ=1,...,n
ζ
σ

C
0
(Ω)
A[ζ].
Rearranging proves the statement. ⊓⊔
14.6 The isoperimetric inequality
Following ideas of Hurwitz [98] we want to give a proof of the isoperimetric in
equality for immersions X : B →R
n+2
with prescribed mean curvature ﬁelds. We
also want to refer the reader to Blaschke and Leichtweiss [15], Barbosa and do
Carmo [7], or Osserman’s review article [129].
Theorem 14.5. Let the immersion X : B →R
n+2
with prescribed mean curvature
ﬁeld H(X, N) and spanning a C
1
regular boundary curve Γ ∈ R
n+2
be given. As
sume that
sup
X∈R
n+2
max
[N[=1
H(X, N) ≤h
0
∈ [0, +∞).
Suppose furthermore that the following smallness condition
1 −nh
0
X −X(∂B)
C
0
(B)
> 0
is true with the mean value
X(∂B) :=
2π
_
0
Z(ϕ)dϕ,
with a representation Z = Z(ϕ) for the boundary curve Γ ⊂R
n+2
of the surface as
described in the proof below. Then it holds the isoperimetric inequality
A[X] ≤
1
4π
_
1 −nh
0
X −X(∂B)
C
0
(B)
_ L[Γ]
2
(14.22)
with the length L[Γ] of Γ ⊂R
n+2
.
Proof. Let (u, v) ∈ B be conformal parameters. We introduce geodesic polar coor
dinates to get the mapping
Y(r, ϑ) = X(r cosϑ, r sinϑ).
14.6 The isoperimetric inequality 325
Let A be a ﬁxed vector. We compute
A[X] =
1
2
__
B
_
[X
u
[
2
+[X
v
[
2
_
dudv
=
1
2
__
B
_
_
(X −A) X
u
¸
u
+
_
(X −A) X
v
¸
v
_
dudv −
1
2
__
B
(X −A) ∆X dudv
=
1
2
_
∂B
_
(X −A) X
u
, (X −A) X
v
¸
ν ds +
1
2
__
B
(X −A) ∆X dudv
≤
1
2
2π
_
0
_
Y(1, ϑ) −A
_
Y
r
(1, ϑ)dϑ +
1
2
__
B
[X −A[[∆X[ dudv
with the outer unit normal vector ν = (cos ϑ, sinϑ). Make additionally use of
X
u
= cosϑY
r
−
1
r
sinϑY
ϑ
, X
v
= sinϑY
r
+
1
r
cosϑY
ϑ
.
From the estimate
[∆X[ =
¸
¸
¸
¸
¸
2
n
∑
σ=1
H(X, N
σ
)WN
σ
¸
¸
¸
¸
¸
≤nh
0
[∇X[
2
we infer
A[X] ≤
1
2
2π
_
0
_
Y(1, ϑ) −A
_
Y
r
(1, ϑ)dϑ +
nh
0
2
__
B
[X −A[[∇X[
2
dudv.
Now let the curve Γ ⊂R
n+2
presented in arclength via
Z(ϕ) +A: [0, 2π] −→Γ, [Z
′
(ϕ)[ =
L[Γ]
2π
for all ϕ ∈ [0, 2π)
where we choose A such that for Z = Z(ϕ) it holds a Fourier expansion in the
following form
Z(ϕ) =
∞
∑
k=1
_
A
k
coskϕ +B
k
sinkϕ
_
with vectors A
k
and B
k
, k = 1, 2, . . . We arrive at
2π
_
0
_
[Z
′
(ϕ)[
2
−[Z(ϕ)[
2
_
dϕ =π
∞
∑
k=1
(k
2
−1)(A
2
k
+B
2
k
) ≥0. (14.23)
326 14 Energy estimates
Then it follows that
A[X] ≤
1
2
2π
_
0
[Y(1, ϑ) −A[[Y
ϑ
(1, ϑ)[ dϑ +
nh
0
X −X(∂B)
C
0
(B)
2
__
B
[∇X[
2
dudv
≤
1
2
2π
_
0
[Z(ϕ)[[Z
′
(ϕ)[ dϕ +nh
0
X −X(∂B)
C
0
(B)
__
B
W dudv
We continue as follows:
L[Γ]
2
−4π A[X]
= L[Γ]
2
−2π
2π
_
0
[Z(ϕ)[[Z
′
(ϕ)[ ϕ −4πnh
0
X −X(∂B)
C
0
(B)
__
B
W dudv
= 2π
2π
_
0
_
[Z
′
(ϕ)[
2
−[Z(ϕ)[[Z
′
(ϕ)[
_
dϕ −4πnh
0
X −X(∂B)
C
0
(B)
__
B
W dudv
≥ π
2π
_
0
_
[Z
′
(ϕ)[
2
−[Z(ϕ)[
2
_
dϕ −4πnh
0
X −X(∂B)
C
0
(B)
__
B
W dudv
≥ −4πnh
0
X −X(∂B)
C
0
(B)
__
B
W dudv
using Cauchy’s inequality and (14.23). After rearranging we have
L[Γ]
2
≥4π(1 −nh
0
X −X(∂B)
C
0
(∂B)
)A[X]
proving the statement. ⊓⊔
Corollary 14.3. Let the minimal immersion X : B →R
n+2
be given, spanning the
C
1
regular boundary curve Γ ⊂R
3
. Then it holds the isoperimetric inequality
A[X] ≤
1
4
L[Γ]
2
(14.24)
with the length L[Γ] of C ⊂R
3
.
Historical remarks on the isoperimetric inequality for minimal surfaces
The ﬁrst successful approach in proving the isoperimetric inequality in the plane
goes back to the Swiss mathematician Jacob Steiner (*1796 Utzendorf/Kt. Bern,
†1863 Bern).
14.7 The spherical energy of µstable geodesic discs 327
Though Steiner’s proof is incomplete, his ingenious method of symmetrization en
joys a wide range of applications, see e.g. Bandle [5].
The Swedish mathematician Tage Gills Torsten Carleman (*1892 in Visseltofta,
†1949 in Stockholm) was the ﬁrst who proved the isoperimetric inequality
A[X] ≤
L[Γ]
2
4π
for minimal surfaces. Beckenbach and Rad´ o in [9] then established this inequality
on simply connected surfaces with nonpositive Gaussian curvature.
The isoperimetric inequality encodes regularity properties of the underlying sur
face. Namely due to Federer and Fleming [54] and Maz’ya [118] it is equivalent to
the Sobolev inequality
_
_
__
B
[∇ϕ[
2
dudv
_
_
≥4π
__
B
ϕ
2
dudv.
Finally Topping in [158] derived the isoperimetric inequality in the plane and on
curved surfaces using results from the theory of curve shortening ﬂow.
For further references and applications of the isoperimetric inequality we want to
refer again to Osserman [129] and the references therein; see also the monographs
Bandle [5] and Chavel [25].
14.7 The spherical energy of µstable geodesic discs
Let us ﬁnally come back to the stability condition
__
B
∇
ds
2 (ψ, ψ)W dudv ≥ µ
__
B
(q −K)Wψ
2
dudv −2α(n)T
X
[N] (14.25)
used already in section 14.2 above.
Following methods from Sauvigny [141] we want to show how to apply this
inequality to establish an upper bound for the spherical energy of a surface.
Theorem 14.6. Let the conformally parametriced immersion X : B→R
n+2
together
with an ONF N be given. Assume that X is stable in the sense of (14.25) with a
function q satisfying
H
2
−q ≥0 in B.
328 14 Energy estimates
Then the Dirichlet energy of N = (N
1
, . . . , N
n
) can be estimated as follows
__
B
ν
[∇N[
2
dudv ≤2
__
B
ν
(2H
2
−q)W dudv +
8
µν
2
+
_
2 +
2α(n)
µ
_
T
X
[N] (14.26)
for all ν ∈ (0, 1) with the setting
[∇N[
2
:=
n
∑
σ=1
[∇N
σ
[
2
.
Proof. For ν ∈ (0, 1) we take the special test function Φ ∈C
∞
0
(B, R) such that
Φ ≡1 on B
ν
, [∇Φ[ ≤
2
ν
on B.
From the Weingarten equations we know
N
2
σ,u
=
L
2
σ,11
+L
2
σ,12
W
+
n
∑
ϑ=1
(T
ϑ
σ,1
)
2
, N
2
σ,v
=
L
2
σ,12
+L
2
σ,22
W
+
n
∑
ϑ=1
(T
ϑ
σ,2
)
2
using conformal parameters. Then we compute
n
∑
σ=1
__
B
ν
[∇N
σ
[
2
dudv
= 2
n
∑
σ=1
__
B
ν
(2H
2
σ
−K
σ
)W dudv +
n
∑
σ,ϑ=1
__
B
ν
_
(T
ϑ
σ,1
)
2
+(T
ϑ
σ,2
)
2
_
dudv
= 2
__
B
ν
(2H
2
−q)W dudv +2
__
B
ν
(q −K)W dudv +2T
X
[N]
≤ 2
__
B
ν
(2H
2
−q)W dudv +2
__
B
(q −K)WΦ
2
dudv +2T
X
[N]
≤ 2
__
B
ν
(2H
2
−q)W dudv +
2
µ
__
B
[∇Φ[
2
dudv +2
_
1 +
2α(n)
µ
_
T
X
[N]
≤ 2
__
B
ν
(2H
2
−q)W dudv +
8
µν
2
+2
_
1 +
2α(n)
µ
_
T
X
[N]
prooving the statement. ⊓⊔
Chapter 15
Fminimal surfaces in R
3
15.1 Fminimal surface
15.2 The geometry of the spherical mapping
15.3 Stability and µstability
15.4 Eigenvalue problems for the spherical mapping
15.5 A curvature estimate
15.6 Theorems of Bernstein type
330 15 Fminimal surfaces in R
3
In this ﬁnal chapter we consider general elliptic functionals and its critical points in Euclidean
space R
3
. The Lagrangian densities of these functionals depend on the surface vector X as well as
the normal direction X
u
X
v
.
15.1 Fminimal surfaces
This ﬁnal chapter is devoted to analytical studies of critical points X : B →R
3
of
anisotropic variational problems of the form
F[X] =
__
B
F(X
u
X
v
)dudv −→extr!
Such critical points are called Fminimal surfaces. As we have seen in chapter
11, after introducing the special weight matrix
W(Z)
−1
=
1
_
det F
ZZ
(Z)
F
ZZ
(Z) +(z
i
z
j
)
i, j=1,2,3
those immersions are revealed as weighted minimal surfaces.
15.2 The geometry of the spherical mapping
On the Hopf function
From Theorem 5.4 we immediately obtain
Theorem 15.1. Let the weighted conformally parametrized Fminimal surface
X : B →R
3
be given. Then its Hopf function
H (w) = L
11
(w) −L
22
(w) −2iL
12
(w)
satisﬁes the complex differential equation
H
w
(w) = A(w)H (w) +B(w)H (w) in B. (15.1)
332 15 Fminimal surfaces in R
3
Here we set
A =
1
4
(a −d +ic +ib), B =
1
4
(a +d +ic −ib)
as well as
a =Ω
1
22
+Ω
2
21
, b =Ω
2
22
−Ω
1
21
, c =Ω
1
12
+Ω
2
11
, d =Ω
2
12
−Ω
1
11
.
Thus the spherical mapping N =N(u, v) of the immersion possesses at most isolated
interior branch points, or it holds K ≡0 at all.
Here we recall the identity
N
u
N
v
= KWN
with the Gaussian curvature K.
Analogously we would proceed in case of prescribed nonweighted mean cur
vature. Then the Hopf function is holomorphic and solves the CauchyRiemann
equation
H
w
(w) = 0 in B.
The spherical mapping N thus possesses at most isolated interior branch points, or
it holds h
2
0
−K ≡0 due to [H [
2
= 4(h
2
0
−K)W
2
. We refer e.g. to Jost [101].
An asymptotical expansion
Our next result on the distribution of the branch points of the sphercial mapping
goes back Sauvigny [141].
Theorem 15.2. Let the weighted conformally parametrized Fminimal surface
X : B →R
3
be given. Then at each point w
0
∈ B there are two linearly independent
vectors A, B ∈ R
3
and a natural number n ∈ N such that the following asymptotical
expansion holds true
N(w
0
+w) −N(w
0
) =
2
n
ρ
n
(Acosnϕ +Bsinnϕ) +o(ρ
n
) for ρ →0. (15.2)
Proof. From (11.33) we already know
[∆N[ ≤(1 +ω
2
)[∇N[
2
in B.
Thus due to a theorem of Hartman and Wintner [77] (see also Hildebrandt [84]) at
each point w
0
∈ B there is a vector C := A−iB, C ,= 0, and a number n ∈ N such
that it holds
N
w
(w
0
+w) =
1
2
_
N
u
(w
0
+w) −iN
v
(w
0
+w)
_
= Cw
n−1
+o([w[
n−1
)
for w →0.
(15.3)
15.2 The geometry of the spherical mapping 333
The linear connection between the weighted fundamental forms from section 5.1
implies that III
W
(X) und I
W
(X) are diagonalized simultaneously. We get
0 =
_
W
−
1
2
(N) ◦ N
w
_
2
=
_
_
W
−
1
2
(N) ◦C
_
2
+o(1)
_
w
2n−2
for w →0.
For the nonsingular matrix F := W
−
1
2
(N(w
0
)) this means
0 = (F◦C)
2
=[F◦ A[
2
−[F◦ B[
2
−2i(F◦ A) (F◦ B),
and we conclude
(F◦ A) (F◦ B) = 0, [F◦ A[ =[F◦ B[ > 0. (15.4)
Thus A and B are linearly independent. We introduce polar coordinates and integrate
(15.3) to get
N(w
0
+w) −N(w
0
)=
1
n
(Cw
n
+Cw
n
) +o([w[
n
)
=
2
n
Re
_
(A−iB)ρ
n
(cosnϕ +i sinnϕ)
_
+o(ρ
n
)
=
2
n
ρ
n
(Acosnϕ +Bsinnϕ) +o(ρ
n
), ρ →0.
The statement is proved. ⊓⊔
In other words: The spherical mapping is open, an open neighbourhood of a
point w ∈
˚
B is mapped onto an open neighbourhood of its image N(w) ∈ S
2
.
We say w ∈ B is a branch point of order n −1 if the expansion (15.2) holds true
with n ∈ N.
The spherical mapping under the stereographic projection
Shortly we want to prove quasiconformality of the spherical mapping of weighted
minimal surfaces. For this purpose we start with some basic facts about the stereo
graphic projection from S
2
⊂R
3
into R
2
.
Let an angle ω ∈ (0, π) be given, and suppose that
N(u, v) (0, 0, 1) ≥cosω in B, ω ∈ [0, π), (15.5)
holds true for the spherical image of the surface or equivalently
N(u, v) ∈ S
2
ω
in B
334 15 Fminimal surfaces in R
3
with the spherical cap
S
2
ω
:=
_
Z ∈ S
2
: Z (0, 0, 1) ≥cosω
_
⊂S
2
.
The Gauss map is supposed to skip a neighborhood of the south pole.
Now we parametrize S
2
ω
as follows
z
1
= sinϑ cosϕ, z
2
= sinϑ sinϕ, z
3
= cosϑ,
ϑ ∈ [0, ω], ϕ ∈ [0, 2π],
(15.6)
and by
Sp: S
2
ω
−→R
2
we denote the stereographic projection with the south pole as center, analytically
given by (see e.g. Neutsch [125])
Sp(z
1
, z
2
, z
3
) =ξ +iη ≡
z
1
1 +z
3
+i
z
2
1 +z
3
, (z
1
, z
2
, z
3
) ∈ S
2
ω
. (15.7)
Its inverse sufﬁces
z
1
=
2ξ
1 +ξ
2
+η
2
, z
2
=
2η
1 +ξ
2
+η
2
, z
3
=
1 −ξ
2
−η
2
1 +ξ
2
+η
2
.
Using the spherical coordinates from (15.6) we can write
Sp(ϑ, ϕ) =
_
sinϑ cosϕ
1 +cosϑ
,
sinϑ sinϕ
1 +cosϑ
_
, (15.8)
and together with (15.7) we calculate
1 +ξ
2
+η
2
= 1 +
sin
2
ϑ cos
2
ϕ
(1 +cosϑ)
2
+
sin
2
ϑ sin
2
ϕ
(1 +cosϑ)
2
=
2
1 +cosϑ
.
In this way we are led to the socalled modulus of distortion
σ(ϑ, ϕ) :=
1
1 +cosϑ
. (15.9)
For example, the north pole is distorted by the factor
1
2
, while along the equator all
lengths remain ﬁxed.
Quasiconformality of the spherical mapping
Our next results states quasiconformality of the spherical mapping under the condi
tions previously discussed. We follow again Sauvigny [141].
15.2 The geometry of the spherical mapping 335
Proposition 15.1. Let the weighted conformally parametrized weighted minimal
surface X : B →R
3
be given. Assume that its spherical mapping satisﬁes (15.5)
with an angle ω ∈ [0, π). Then the associated plane mapping
Φ(u, v) := Sp◦ N(u, v), (u, v) ∈ B,
is bounded and quasiconform, i.e. there hold
[Φ(u, v)[ ≤
sinω
1 +cosω
and [∇Φ(u, v)[
2
≤−2(1 +ω
0
)J
Φ
(u, v) in B
with the nonpositive Jacobian J
Φ
= J
Φ
(u, v) of the mapping Φ =Φ(u, v).
Proof. The ﬁrst bound follows directly from the representation (15.8) of the stereo
graphic projection in terms of ϑ and ϕ. Furthermore, together with (15.9) we esti
mate the distortion χ = χ(ξ, η) of the inverse mapping Sp
−1
, i.e.
1 +cosω ≤ χ(ξ, η) ≤2 if Sp
−1
(ξ, η) (0, 0, 1) ≥cosω. (15.10)
Now we deﬁne the weight matrix
H(u, v) :=
1
χ(Φ(u, v))
2
∂Sp
−1
(Φ(u, v))
T
◦ W(X, N)
−1
◦ ∂Sp
−1
(Φ(u, v))
satisfying
(1 +ω
0
)
−1
[ξ[
2
≤ξ ◦ H(u, v) ◦ ξ ≤(1 +ω
0
)[ξ[
2
for all ξ ∈ R
2
(u, v) ∈ B
as well as
det H(u, v) = 1 in B.
Using the linear connection between the weighted fundamental forms we infer
∂Φ(u, v)
T
◦ H(u, v) ◦ ∂Φ(u, v) =−WKχ
2
_
1 0
0 1
_
.
Notice the diagonal form of this product. Since K ≤0 we have
−det ∂Φ = −det
_
∂Φ
T
◦ H◦ ∂Φ
_1
2
= −WKχ
2
=
1
2
Spur
_
∂Φ
t
◦ H◦ ∂Φ
_
≥
1
2(1 +ω
0
)
[∇Φ[
2
.
This proves the statement. ⊓⊔
Quasiconformality of the spherical mapping for immersions of this surface class
along with curvature estimates were already established in Simon [147] in 1977, see
also our following considerations. We additionally want to refer the reader to Hoy
[95], [96] and [97] on this subject.
336 15 Fminimal surfaces in R
3
An oscillation estimate for the spherical mapping
Now Theorem 12.3 from Gilbarg and Trudinger [71] (see also Simon [147]) pro
vides an interior H¨ older estimate for socalled (k, k
′
)quasiconformal plane map
pings (p, q) = (p(x, y), q(x, y)) in the sense of
p
2
x
+ p
2
y
+q
2
x
+q
2
y
≤2k(p
x
q
y
−p
y
q
x
) +k
′
with constants [k[ ≥ 1 and k
′
≥ 0. Together with our modulus of distortion (15.10)
we arrive at (see Sauvigny [141])
Theorem 15.3. Let the weighted conformally parametrized Fminimal surface
X : B → R
3
be given. Suppose that its spherical mapping satisﬁed (15.5) with
an angle ω ∈ [0, π). Let ﬁnally λ > 0 and ν ∈ (0, 1) be ﬁxed. Then there exists
δ =δ(ω
0
, ω, ν; λ) ∈ (0,
1
2
ν) such that for all w
0
∈ B
1−
1
2
ν
(0, 0) the following oscil
lation estimate holds true
[N(w) −N(w
1
)[ ≤λ for all w ∈ B
δ
(w
0
) (15.11)
with arbitrary w
1
∈ B
δ
(w
0
).
Application to immersion of prescribed weighted mean curvature
Let us consider more closely the spherical mapping of immersions with prescibed
mean curvature from the point of view of quasiconformality.
Due to the linear connection
III
W
(X) −2H
W
(X, N)II
W
(X) +KI
W
(X) = 0
between the weighted fundamental forms we compute
(det ∂Φ)
2
= det
_
2H
W
(X, N)L
11
−KW 2H
W
(X, N)L
12
2H
W
(X, N)L
21
2H
W
(X, N)L
22
−KW
_
χ
2
= K
2
W
2
χ
4
using weighted conformal parameters (u, v) ∈ B. i.e. for the signed Jacobian ∂Φ we
infer the identity
det ∂Φ = KWχ
2
.
Moreover, for its trace we have
trace(∂Φ
T
◦ H◦ ∂Φ) =
_
2H
W
(X, N)L
11
−KW +2H
W
(X, N)L
22
−KW
_
χ
2
= 2
_
2H
W
(X, N)
2
−K
_
Wχ
2
.
15.2 The geometry of the spherical mapping 337
Proposition 15.2. Using weighted conformal parameters (u, v) ∈ B it holds
[∇Φ[
2
≤−2(1 +ω
0
)det ∂Φ+4(1 +ω
0
)H
W
G(X, N)
2
Wχ
2
in B,
i.e. Φ =Φ(u, v) is (k, k
′
)quasiconformal with
k =−2(1 +ω
0
) and k
′
= 4(1 +ω
0
)H
W
(X, N)
2
Wχ
2
.
We want to apply this reslut to immersions with bounded gradient. Suppose
sup
(u,v)∈B
W(u, v) ≤W
0
∈ (0, +∞). (15.12)
Then we ﬁnd
Theorem 15.4. Assume that the weighted conformally parametrized immersion
X : B → R
3
satisﬁes (15.12) and that its spherical mapping fulﬁlls (15.5) with
an angle ω ∈ [0, π). Let ﬁnally λ > 0 and ν ∈ (0, 1) be ﬁxed. Then there exist
δ = δ(ω
0
, ω, ν,W
0
; λ) ∈ (0,
1
2
ν) such that for all w
0
∈ B
1−
1
2
ν
(0, 0) the following
oscillation estimate holds true
[N(w) −N(w
1
)[ ≤λ for all w ∈ B
δ
(w
0
) (15.13)
with arbitrary w
1
∈ B
δ
(w
0
).
This estimate can immediately be applied to conformally parametrized immer
sions with prescribed nonweighted mean curvature H(X, Z) which are small in the
sense of
h
0
M < 1 with M := sup
(u,v)∈B
[X(u, v)[, h
0
:= sup
(X,Z)∈R
3
R
3
[H(X, Z)[.
This follows from Heinz’ gradient estimates from section 9.4 taking
[∆X[ ≤h
0
[∇X[
2
in B
into account. Thus there is a real constant c = c(h
0
, M, ν
1
) with the property
[∇X(u, v)[ ≤c(h
0
, M, ν
1
) for all (u, v) ∈ B
1−ν
1
(0, 0).
Then the mapping N = N(u, v) is (k, k
′
)quasiconformal in B
1−ν
1
(0, 0) with k =−2
and k
′
= 4h
2
0
c(h
0
, M, ν
1
)
2
.
Furthermore, the smallness condition can be replaced by the following modulus
of continuity:
For real λ > 0 there is a δ =δ(λ) such that
[X(w) −X(w
1
)[ ≤λ for all w ∈ B
δ
(w
0
) (15.14)
for arbitrary w
1
∈ B
δ
(w
0
) and w
0
∈ B with B
δ
(w
0
) ⊂⊂B.
338 15 Fminimal surfaces in R
3
Namely, deﬁne the mapping
´
X(w) :=
1
λ
_
X(w) −X(w
1
)
_
: B
δ
(w
0
) −→R
3
with the properties
[
´
X(w)[ ≤1 and [∆
´
X(w)[ ≤λh
0
[∇
´
X(w)[
2
in B
δ
(w
0
).
Now choose λ > 0 so that λh
0
< 1. This implies an inner gradient estimate for the
surface vector and therefore an oscillation estimate for spherical mapping.
Finally we want to remark that an oscillation estimates for the composition
N(u, v) =G◦ X(u, v), (u, v) ∈ B,
with the Gauss map G: X(B) → S
2
on surface graphs of mean curvature type is
established in Gilbarg and Trudinger [71], section 16.6. Assuming a modulus of
continuity (15.14) we could then infer an oscillation estimate N = N(u, v).
15.3 Stability and µstability
The second variation
Our aim now is to compute the second variation of the parametric functional F[X].
With a test function ϕ ∈C
∞
0
(B, R) we consider the normal variation
¯
X(u, v) := X(u, v) +εϕ(u, v)N(u, v), (u, v) ∈ B.
Then it follows
¯
X
u
∧
¯
X
v
= X
u
∧X
v
+ε¦X
u
∧N
v
+N
u
∧X
v
¦ϕ +ε¦X
u
∧Nϕ
v
+N∧X
v
ϕ
u
¦
+ε
2
ϕ
2
N
u
∧N
v
+ε
2
¦N
u
∧Nϕ
v
+N∧N
v
ϕ
u
¦ϕ.
First it holds
∂
2
∂ε
2
F(
¯
X
u
∧
¯
X
v
)
¸
¸
¸
ε=0
=
∂
∂ε
(
¯
X
u
∧
¯
X
v
)
¸
¸
¸
ε=0
◦ F
ZZ
(X
u
∧X
v
) ◦
∂
∂ε
(
¯
X
u
∧
¯
X
v
)
¸
¸
¸
ε=0
+F
Z
(X
u
∧X
v
)
∂
2
∂ε
2
(
¯
X
u
∧
¯
X
v
)
¸
¸
¸
ε=0
.
15.3 Stability and µstability 339
Let (u, v) ∈ B be weighted conformal parameters. From the properties of the weight
matrix W(Z) we infer
∂
2
∂ε
2
F(
¯
X
u
∧
¯
X
v
)
¸
¸
¸
ε=0
=
1
W
_
(X
u
∧N
v
+N
u
∧X
v
)ϕ +(X
u
∧Nϕ
v
+N∧X
v
ϕ
u
)
_
◦ F
ZZ
(N) ◦ . . .
. . . ◦
_
(X
u
∧N
v
+N
u
∧X
v
)ϕ +(X
u
∧Nϕ
v
+N∧X
v
ϕ
u
)
_
+F
Z
(N)
_
2N
u
∧N
v
ϕ
2
+N
u
∧N(ϕ
2
)
v
+N∧N
v
(ϕ
2
)
u
_
=
_
det F
ZZ
(N)
W
(X
u
∧Nϕ
v
+N∧X
v
ϕ
u
) ◦ W(N)
−1
◦ (X
u
∧Nϕ
v
+N∧X
v
ϕ
u
)
+
_
_
F
Z
(N), N
u
, N
¸
ϕ
2
_
v
+
__
F
Z
(N), N, N
v
¸
ϕ
2
_
u
−
_
F
ZZ
(N) ◦ N
v
, N
u
, N
¸
ϕ
2
−
_
F
ZZ
(N) ◦ N
u
, N, N
v
¸
ϕ
2
with the usual setting [X,Y, Z] := X (Y Z). It follows that
(X
u
∧Nϕ
v
+N∧X
v
ϕ
u
) ◦ W(N)
−1
◦ (X
u
∧Nϕ
v
+N∧X
v
ϕ
u
)
=
_
W(N)
1
2
◦ X
v
ϕ
v
+W(N)
1
2
◦ X
u
ϕ
u
_
◦
_
W(N)
1
2
◦ X
v
ϕ
v
+W(N)
1
2
◦ X
u
ϕ
u
_
= W
_
ϕ
2
u
+ϕ
2
v
_
= W[∇ϕ[
2
.
For the divergence term we have
__
B
div
_
_
F
Z
(N), N, N
v
¸
ϕ
2
,
_
F
Z
(N), N
u
, N
¸
ϕ
2
_
dudv = 0
due to ϕ = 0 on ∂B = 0. Finally we know
_
N
u
, N
v
, F
ZZ
(N) ◦ N
¸
= 0,
so that we infer
_
F
ZZ
(N) ◦ N
u
, N
v
, N
¸
+
_
N
u
, F
ZZ
(N) ◦ N
v
, N
¸
=
_
F
ZZ
(N) ◦ N
u
, N
v
, N
¸
+
_
N
u
, F
ZZ
(N) ◦ N
v
, N
¸
+
_
N
u
, N
v
, F
ZZ
(N) ◦ N
¸
=
_
N
u
, N
v
, N] traceF
ZZ
(N) = KW traceF
ZZ
(N).
340 15 Fminimal surfaces in R
3
Thus we arrive at
δ
2
F[X] =
__
B
_
_
det F
ZZ
(N) [∇ϕ[
2
+KW traceF
ZZ
(N)ϕ
2
_
dudv
for all ϕ ∈C
∞
0
(B, R).
Using general parameters (u, v) ∈ B we have therefore proved
Theorem 15.5. The second variation of the parametric functional F[X] reads
δ
2
W [X] =
__
B
_
_
det F
ZZ
(N) ∇
ds
2
W
(ϕ, ϕ) +KtraceF
ZZ
(N)ϕ
2
_
W dudv
for all ϕ ∈C
∞
0
(B, R).
Stability and µstability
Again we require
M
1
[ξ[
2
≤ξ ◦ F
ZZ
(Z) ◦ ξ ≤M
2
[ξ[
2
for all ξ = (ξ
1
, ξ
2
, ξ
3
) ∈ R
3
(15.15)
and all Z ∈ S
2
with ξ Z = 0, with positive constants 0 < M
1
≤M
2
< +∞. Then we
can set
1 +ω
0
=
M
2
M
1
resp.
1
1 +ω
0
=
M
1
M
2
.
Theorem 15.6. Let X : B →R
3
be a stable Fminimal surface, i.e. assume that
δ
2
F[X] ≥0.
Furthermore, suppose that (15.15) is satisﬁed. Let ﬁnally µ > 0 be a real number
with the propery
µ ≤
2M
1
M
2
.
Then the immersion is µstable with this number µ, i.e. it holds
__
B
∇
2
ds
W
(ϕ, ϕ)W dudv ≥ µ
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R).
Proof. Namely after rearranging it follows
M
2
__
B
∇
ds
2
W
(ϕ, ϕ)W dudv ≥2M
1
__
B
(−K)Wϕ
2
dudv
for all ϕ ∈C
∞
0
(B, R).
15.4 Eigenvalue problems for the spherical mapping 341
This implies
__
B
∇
ds
2
W
(ϕ, ϕ)W dudv ≥
2M
1
M
2
__
B
(−K)Wϕ
2
dudv,
and the statement follows. ⊓⊔
15.4 Eigenvalue problems for the spherical mapping
This section is devoted to a deeper analysis of µstable Fminimal surfaces in Eu
clidean space R
3
. Our results are partly build up on our considerations from chap
ter 13, on the other hand we will meet new methods and results applicable for F
minimal surfaces in R
3
exclusively.
A comparison principle
The spherical mapping N: B→R
3
represents a branched immersion fromthe closed
unit disc B⊂R
2
into the unit sphere S
2
⊂R
3
. As we have seen above, interior branch
points are isolated.
In the following sections we are concerned with the eigenvalue problem 13.6 on
spherical domains. To establish suitable bounds for the corresponding ﬁrst eigen
value λ
1
> 0 we need the following comparison argument.
Lemma 15.1. Let the immersion X : B →R
3
be given. Let
ds
2
=
2
∑
i, j=1
g
i j
du
i
du
j
, ds
2
W
=
2
∑
i, j=1
h
i j
du
i
du
j
its nonweighted resp. weighted line elements w.r.t. to a weight matrix W(Z). Then
there hold
(1 +ω
0
)
−1
__
B
∇
ds
2 (ϕ, ϕ)W dudv ≤
__
B
∇
ds
2
W
(ϕ, ϕ)W dudv
≤ (1 +ω
0
)
__
B
∇
ds
2 (ϕ, ϕ)W dudv
(15.16)
for all ϕ ∈C
1
0
(B, R) with the invariant Beltrami operators ∇
ds
2 and ∇
ds
2
W
.
342 15 Fminimal surfaces in R
3
Proof. On the tangential plane T
X
of the immersion we compute
__
B
∇
ds
2
W
(ϕ, ϕ)W dudv
=
__
B
(ϕ
u
, ϕ
v
) ◦
_
∂X
T
◦ W(X, N) ◦ ∂X
_
−1
◦ (ϕ
u
, ϕ
v
)W dudv
=
__
B
(ϕ
u
, ϕ
v
) ◦ (∂X)
−1
◦ W(X, N)
−1
◦ (∂X
T
)
−1
◦ (ϕ
u
, ϕ
v
)W dudv
=
__
B
_
(ϕ
u
, ϕ
v
) ◦ (∂X)
−1
_
◦ W(X, N)
−1
◦
_
(ϕ
u
, ϕ
v
) ◦ (∂X)
−1
_
W dudv
from where we infer the estimates
__
B
∇
ds
2
W
(ϕ, ϕ)W dudv
≤ (1 +ω
0
)
__
B
_
(ϕ
u
, ϕ
v
) ◦ (∂X)
−1
_
_
(ϕ
u
, ϕ
v
) ◦ (∂X)
−1
_
W dudv
= (1 +ω
0
)
__
B
∇
ds
2 (ϕ, ϕ)W dudv
as well as
__
B
∇
ds
2
W
(ϕ, ϕ)W dudv
≥
1
1 +ω
0
__
B
_
(ϕ
u
, ϕ
v
) ◦ (∂X)
−1
_
_
(ϕ
u
, ϕ
v
) ◦ (∂X)
−1
_
W dudv
=
1
1 +ω
0
__
B
∇
ds
2 (ϕ, ϕ)W dudv
proving the statement. ⊓⊔
The weighted Rayleigh quotient on S
2
Let Ω ⊂ S
2
be a spherical domain covered by the Gauss map N = N(u, v) of the
Fminimal surface X : B →R
3
.
15.4 Eigenvalue problems for the spherical mapping 343
To the Rayleigh quotient R[ϕ] introduced in section 13.5 we deﬁne its weighted
counterpart
R
W
[ϕ] :=
_
Ω
∇
S
2
,W
(ϕ, ϕ)dS
_
Ω
ϕ
2
dS
for funktions ϕ ∈V(Ω, R) on the function space
V(Ω, R) :=
_
φ ∈ H
1,2
(Ω, R) ∩C
0
(B, R) : φ ,≡0, φ = 0 on ∂Ω
_
with the weighted Beltrami operator ∇
S
2
,W
on S
2
⊂R
3
and its area element dS.
To Ω ⊂S
2
we associate the ﬁrst eigenvalue of this weighted operator on S
2
:
min
ϕ∈V(Ω,R)
R
W
[ϕ] =λ
1,W
(Ω).
From the previous comparison result we conclude
1
1 +ω
0
λ
1,W
(Ω) ≤λ
1
(Ω) ≤(1 +ω
0
)λ
1,W
(Ω). (15.17)
µstability due to Schwarz
The following result characterizes the size of the spherical image of µstable F
minimal surfaces, see Fr¨ ohlich [62].
Theorem 15.7. Let the weighted conformally parametrized and µstable Fminimal
surface X : B →R
3
with µ > 0 be given. Let Ω
′
⊂⊂ Ω ⊂ S
2
denote the domain
which is covered by the spherical mapping N[
˚
B
′
on an open disc
˚
B
′
⊂⊂
˚
B. Fur
thermore assume that N(∂B
′
) = ∂Ω
′
. Then for the ﬁrst eigenvalue λ
1
(Ω
′
) of the
spherical Laplacian on Ω
′
⊂S
2
it holds
λ
1
(Ω
′
) ≥
µ
1 +ω
0
. (15.18)
Proof. For λ
1,W
(Ω
′
) we choose an eigenfunction ϕ : Ω
′
→R such that
∆
S
2
,W
ϕ +λ
1,W
(Ω
′
)ϕ = 0 in Ω
′
,
ϕ = 0 auf ∂Ω
′
,
ϕ > 0 in Ω
′
.
The ﬁnitely many branch points of the Gauss map on the closed set B
′
⊂
˚
B are
denoted by ¦w
1
, . . . , w
N
¦ with N ∈ N. Now for sufﬁciently small ε > 0 we consider
closed discs K
ε
(w
t
) ⊂ B
′
(eventually together with their intersection with B
′
) of
344 15 Fminimal surfaces in R
3
common radius ε > 0 around these branch points. The pullback ψ(u, v) := ϕ ◦
N(u, v), (u, v) ∈ B
′
, then satisﬁes the eigenvalue problem
∆
ds
2
g
ψ −λ
1,W
(Ω
′
)Kψ = 0 in B
′
ε
:= B
′
¸
N
_
t=1
K
ε
(w
t
),
ψ = 0 on ∂B
′
,
ψ > 0 in B
′
.
Considering the vanishing boundary data of ψ we evaluate the µstability condition
as follows
__
B
′
ε
[∇ψ[
2
dudv +µ
__
B
′
ε
ψ
2
KW dudv
= −
__
B
′
ε
ψ△ψdudv +µ
__
B
′
ε
ψ
2
KW dudv +
N
∑
t=1
_
∂K
ε
(w
t
)
ψ∇ψ (N
1
, N
2
)ds
=
_
λ
1,W
(Ω
′
) −µ
_
__
B
′
ε
ψ
2
(−K)W dudv +
N
∑
t=1
_
∂K
ε
(w
t
)
ψ∇ψ (N
1
, N
2
)ds
with the outer unit normal vector (N
1
, N
2
) at the boundary curve ∂K
ε
(w
t
), t =
1, . . . , N, and its line element ds. Since ψ ∈ H
2
1
(
˚
B
′
, R) ∩C
0
(B
′
, R) we infer
lim
ε→0
__
B
′
ε
ψ
2
(−K)W dudv =
__
B
′
ψ
2
(−K)W dudv,
lim
ε→0
N
∑
t=1
_
∂K
ε
(w
t
)
ψ∇ψ (N
1
, N
2
)ds = 0.
Thus if λ
1,W
(Ω
′
) −µ < 0 then it follows necessarily
__
B
′
[∇ψ[
2
dudv +µ
__
B
′
ψ
2
KW dudv < 0
contradicting the µstability. ⊓⊔
Spherical oscillation of µstable Fminimal surfaces
This result, the ideas of which go already back to H.A. Schwarz, leads us to our next
oscillation estimate for the spherical mapping, see Fr¨ ohlich [62].
15.4 Eigenvalue problems for the spherical mapping 345
Theorem 15.8. Let the weighted conformally parametrized and µstable Fminimal
surface X : B →R
3
with µ > 0 be given. Then for each ε =ε(ω
0
, µ) > 0 satisfying
(15.20) there is δ =δ(ω
0
, µ, ν; ε(µ)) ∈
_
0, (1 −ν)
2
_
with the property
osc
B
δρ
(w
0
)
N(w) := sup
w
1
,w
2
∈B
δρ
(w
0
)
[N(w
1
) −N(w
2
)[ ≤2ε(ω
0
, µ) (15.19)
on each open disc
˚
B
ρ
(w
0
) ⊂B
1−ν
(0, 0) with ﬁxed ν ∈ (0, 1) and [w
0
[ < 1 −ν.
Proof. Consider the open disc
˚
B
ρ
(w
0
) ⊂ B
1−ν
(0, 0) with ρ ∈ (0, 1 −ν −[w
0
[) and
[w
0
[ < 1 −ν, ν ∈ (0, 1). Due to the estimate
__
[w[≤1−ν
[∇N[
2
dudv ≤
8π(1 +ω
0
)
µν
2
we can apply the CourantLebesgue lemma from Heinz [80], Lemma 14, to ﬁnd for
given ε > 0 a real δ =δ(ω
0
, µ, ν; ε) ∈
_
0, (1 −ν)
2
_
and δ
∗
∈ [δ,
√
δ] such that
_
∂B
δ
∗
ρ
(w
0
)
[dN[ ≤2
¸
8π
2
(1 +ω
0
)
(−logδ)µν
2
≤ 2ε;
see the proof of our curvature estimate from chapter 10. This means in other words
N
_
∂B
δ
∗
ρ
(w
0
)
_
⊂K
ε
_
N(w
0
+δ
∗
ρ)
_
with the setting K
ε
(Z) :=¦X ∈ R
3
: [X −Z[ ≤ε¦. Now let Ω
δ
∗
ρ
⊂S
2
be covered by
the restriction N[
˚
B
δ
∗
ρ
(w
0
)
. Then it holds with a uniquely determined angle ω =ω(ε)
N
_
∂B
δ
∗
ρ
(w
0
)
_
⊂S
2
ω(ε)
(w
0
) := S
2
∩K
ε
_
N(w
0
+δ
∗
ρ)
_
.
We choose ε =ε(ω
0
, µ) sufﬁciently small with the property
λ
1
_
S
2
¸ S
2
ω(ε)
(w
0
)
_
<
µ
1 +ω
0
. (15.20)
But then it mus be satisﬁed
Ω
δ
∗
ρ
⊂S
2
ω(ε)
(w
0
).
Otherwise there is a point w
1
∈ B
δ
∗
ρ
(w
0
) with
N(w
1
) ∈ S
2
¸ S
2
ω(ε)
(w
0
).
Note that N = N(u, v) is an open mapping as we showed above. Continuation along
paths we arrive at
S
2
¸ S
2
ω(ε)
(w
0
) ⊂Ω
δ
∗
ρ
.
346 15 Fminimal surfaces in R
3
To λ
1
_
S
2
¸S
2
ω(ε)
(w
0
)
_
we now choose an eigenfunction ϕ : S
2
¸S
2
ω(ε)
(w
0
) →R such
that there hold
∆
S
2 ϕ +λ
1
_
S
2
¸ S
2
ω(ε)
(w
0
)
_
ϕ = 0 in S
2
¸ S
2
ω(ε)
(w
0
),
ϕ = 0 on ∂
_
S
2
¸ S
2
ω(ε)
(w
0
)
_
,
ϕ > 0 in S
2
¸ S
2
ω(ε)
(w
0
),
which we pull back by means of ψ(u, v) := ϕ ◦ N(u, v), (u, v) ∈ B
δ
∗
ρ
. But now the
eigenvalue bound (15.20) contradicts our previous result of Schwarz. ⊓⊔
A theorem of Barbosa and do Carmo type
With our next result from Fr¨ ohlich [62] we present some sort of inverse of Schwarz’
theorem on µstable Fminimal surfaces. Origininally it was proved in Barbosa and
do Carmo [6] for minimal surfaces.
Theorem 15.9. Let the weighted conformally parametrized Fminimal surface
X : B →R
3
be given. Let
˚
B
′
⊂⊂
˚
B denote an open disc, and with Ω
′
⊂ S
2
we mean
the spherical domain which is convered by the restriction N[
˚
B
′
. For given ω
0
≥ 0
we choose an angle ω = ω(ω
0
) ∈ (0, π) such that for the spherical cap S
2
ω
⊂ S
2
it
holds
Area(Ω
′
) < Area(S
2
ω
), λ
1
(S
2
ω
) = µ(1 +ω
0
) (15.21)
with a real number µ > 0. Then X[
˚
B
′
is µstabil with this µ > 0.
Proof. 1. Assume that X : B →R
3
is not µstable. Then on a simply connected and
open domain
˚
B
∗
⊆
˚
B
′
there is a function ψ ∈C
2
(B
∗
, R) satisfying
∆ψ(u, v) +µ(−K)Wψ(u, v) = 0 in
˚
B
∗
,
ψ(u, v) > 0 in
˚
B
∗
,
ψ(u, v) = 0 on ∂B
∗
.
(15.22)
Let Ω
∗
⊆ Ω
′
be that domain which is covered by N restricted on
˚
B
∗
⊆
˚
B
′
. We
construct a function ϕ ∈ V (Ω
∗
, R) with the property
_
Ω
∗
∇
S,W
ϕ
2
S,W
dS ≤ µ
_
Ω
∗
ϕ
2
dS. (15.23)
Then our comparison principle yields
_
Ω
∗
∇
S
ϕ
2
S
dS ≤ µ(1 +ω
0
)
_
Ω
∗
ϕ
2
dS
implying λ
1
(Ω
∗
) ≤ µ(1 +ω
0
).
15.4 Eigenvalue problems for the spherical mapping 347
Now let S
2
ω
∗ ⊂ S
2
be a spherical cap with the same area as Ω
∗
⊂ S
2
. Together
with (15.21) as well as Proposition 13.4 and 13.3 from chapter 13 we infer
µ(1 +ω
0
) =λ
1
(S
2
ω
) <λ
1
(S
2
ω
∗ ) ≤λ
1
(Ω
∗
) ≤ µ(1 +ω
0
).
Thus we arrive at a contradiction to the µstability assumption. It remains to
construct ϕ ∈V(Ω
∗
, R).
2. For some arbitrary point q ∈ Ω
∗
we ﬁnd preimages
N
−1
(q) ∩B
∗
=¦w
1
, . . . , w
n
q
¦ ⊂B
∗
with multiplicities α(w
i
) ≥1, i = 1, . . . , n
q
. We set
ϕ(q) :=
n
q
∑
i=1
α(w
i
)ψ(w
i
). (15.24)
This function is positive on Ω
∗
and vanishes on the boundary ∂Ω
∗
(recall that
the spherical mapping is open by Theorem 15.2). Following the argumentation
from Barbosa and do Carmo [6], Lemma 3.6, which we skip here, the function is
also continuous on the closure of Ω
∗
⊂S
2
. We are left to prove (15.23).
3. Let ¦q
1
, . . . , q
N
¦ ⊂ Ω
∗
, N ∈ N, the set of images of branch points in B
∗
⊆ B
′
.
Choose ε > 0 sufﬁciently small. In Ω
∗
⊆ Ω
′
we consider closed geodesic discs
K
ε
(q
i
), i = 1, . . . , N, of common radius ε > 0 and centers at q
i
for i = 1, . . . , N.
Deﬁne
T
ε
:=Ω
∗
¸
_
∂Ω
∗
∪
N
_
t=1
K
ε
(q
t
)
_
.
Following Barbosa and do Carmo [6], Lemma 3.8, there exist simply connected,
open and disjoint sets R
jk
⊂T
ε
with the following properties:
a) For any point q ∈ R
jk
, the preimage set N
−1
(q) ∩
˚
B
∗
consists of exactly j
elements;
b) it holds
T
ε
=
_
j,k
R
jk
; (15.25)
c) there exist simply connected, open and disjoint sets
˚
B
∗
jkl
⊂
˚
B
∗
such that
N
−1
(
˚
R
jk
) =
j
_
l=1
˚
B
∗
jkl
;
thus the spherical mapping restricted on such a
˚
B
∗
jkl
⊂
˚
B
∗
is diffeomorphic
onto R
jk
⊂T
ε
.
348 15 Fminimal surfaces in R
3
4. Let N
jkl
:= N[
˚
B
∗
jkl
. Then the spherical mapping is locally invertible with the in
verse mappings
N
−1
jkl
: R
jk
−→
˚
B
∗
jkl
, l = 1, . . . , j. (15.26)
Furthermore, the function ϕ : Ω
∗
→R from (15.24) satisﬁes
ϕ[
R
jk
=
j
∑
l=1
ψ ◦ N
−1
jkl
. (15.27)
It is continuously differentiable in R
jk
since R
jk
is free of branch points.
5. We are going to prove the inequality
lim
ε→0
_
T
ε
∇
S,g
ϕ
2
S,g
dS ≤ lim
ε→0
µ
_
T
ε
ϕ
2
dS. (15.28)
The right hand side here is bounded due to the continuity of ϕ : Ω
∗
→R, and so
is ϕ ∈V(Ω
∗
, R), and in the limit we obtain
_
Ω
∗
∇
S,W
ϕ
2
S,W
dS ≤ µ
_
Ω
∗
ϕ
2
dS.
Taking account of (15.25) and (15.27) we compute
_
T
ε
∇
S,W
ϕ
2
S,W
dS =
∑
j,k
_
R
jk
∇
S,W
ϕ
2
S,W
dS
=
∑
j,k
_
R
jk
_
_
_
_
_
j
∑
l=1
∇
S,W
(ψ ◦ N
−1
jkl
)
_
_
_
_
_
2
S,W
dS
=
∑
j,k
_
R
jk
j
∑
l=1
∇
S,W
(ψ ◦ N
−1
jkl
)
2
S,W
dS +I
1
(15.29)
with the setting
I
1
:=
∑
j,k
_
R
jk
2
∑
1≤r<s≤j
¸∇
S,W
(ψ ◦ N
−1
jkr
), ∇
S,W
(ψ ◦ N
−1
jks
))
S,W
dS (15.30)
where ¸, ) denotes the inner product on S
2
w.r.t. the weighted metric (h
i j
)
i, j=1,2
.
Fom the linear connection of the weighted fundamental forms together with
(15.26) we conclude (using weighted conformal parameters)
15.4 Eigenvalue problems for the spherical mapping 349
j
∑
l=1
_
R
jk
∇
S,W
(ψ ◦ N
−1
jkl
)
2
S,W
dS =
j
∑
l=1
__
B
∗
jkl
∇
dσ
2
W
(ψ, ψ)W
dσ
2
W
dudv
=
j
∑
l=1
__
B
∗
jkl
∇
ds
2
W
(ψ, ψ)W dudv.
(15.31)
If we set
B
∗
ε
:= N
−1
_
N
_
t=1
K
ε
(q
t
)
_
with B
∗
¸
˚
B
∗
ε
=
_
j,k,l
B
∗
jkl
it follows together with (15.29) and (15.31)
_
T
ε
∇
S,W
ϕ
2
S,W
dS =
∑
j,k,l
__
B
∗
jkl
∇
ds
2
W
(ψ, ψ)W dudv +I
1
=
__
B
∗
¸B
∗
ε
∇
ds
2
W
(ψ, ψ)W dudv +I
1
.
(15.32)
Consider the right hand side of (15.28):
_
T
ε
ϕ
2
dS =
∑
j,k
_
R
jk
_
j
∑
l=1
ψ ◦ N
−1
jkl
_
2
dS =
∑
j,k
j
∑
l=1
_
R
jk
(ψ ◦ N
−1
jkl
)
2
dS +I
2
due to (15.27) and with the setting
I
2
:=
∑
j,k
_
R
jk
2
∑
1≤r<s≤j
(ψ ◦ N
−1
jkr
)(ψ ◦ N
−1
jks
)dS. (15.33)
Using weighted conformal parameters this means
_
T
ε
ϕ
2
dS =
∑
j,k,l
__
B
∗
jkl
ψ
2
(−K)W dudv +I
2
=
__
B
∗
¸B
∗
ε
ψ
2
(−K)W dudv +I
2
.
(15.34)
350 15 Fminimal surfaces in R
3
6. From (15.32), (15.34) and (15.22) we infer
_
T
ε
∇
S,W
ϕ
2
S,W
dS −µ
_
T
ε
ϕ
2
dS
=
__
B
∗
¸B
∗
ε
[∇ψ[
2
dudv −µ
__
B
∗
¸B
∗
ε
ψ
2
(−K)W dudv +I
1
−µI
2
=
__
B
∗
¸B
∗
ε
[∇ψ[
2
dudv +
__
B
∗
¸B
∗
ε
ψ△ψdudv +I
1
−µI
2
=
_
∂B
∗
ε
(ψ∇ψ) (N
1
, N
2
)ds +I
1
−µI
2
with the outer unit normal vector ﬁeld (N
1
, N
2
) at the boundary ∂B
∗
ε
and its line
element ds. The function ψ is continuously differentiable in B
∗
⊆B
′
and vanishes
outside of this set such that
lim
ε→0
_
∂B
∗
ε
(ψ∇ψ) (N
1
, N
2
)ds = 0.
Thus it remains to show
lim
ε→0
(I
1
−µI
2
) ≤0. (15.35)
7. Now let
χ
ℓ
:=ψ ◦ N
−1
jkl
: R
jk
−→R, ℓ = 1, . . . , j.
Due to (15.27) we have ϕ[
R
jk
= χ
1
+. . . +χ
j
. Furthermore, there hold
∆
S,W
χ
ℓ
+µχ
ℓ
= 0 in R
jk
, ℓ = 1, . . . , j,
in view of (15.22). We compute
∆
S,W
(χ
r
χ
s
) = χ
r
∆
S,W
χ
s
+χ
s
∆
S,g
χ
r
+2¸∇
S,W
χ
r
, ∇
S,W
χ
s
)
S,g
= −2µχ
r
χ
s
+2¸∇
S,W
χ
r
, ∇
S,W
χ
s
)
S,W
.
Together with (15.30) and (15.33) it therefore follows
I
1
−µI
2
=
∑
j,k
∑
1≤r<s≤j
_
R
jk
_
2¸∇
S,W
χ
r
, ∇
S,W
χ
s
)
S,W
−2µχ
r
χ
s
_
dS
=
∑
j,k
∑
1≤r<s≤j
_
R
jk
∆
S,g
(χ
r
χ
s
)dS
=
∑
j,k
∑
1≤r<s≤j
_
∂R
jk
¸∇
S,W
(χ
r
χ
s
), N)
S,W
dσ
S
15.5 A curvature estimate 351
with the unit normal vector Nat the positively oriented boundary curve ∂R
jk
and
its line element dσ
S
. Now if χ
r
χ
s
= 0 along an arc γ ⊂ ∂R
jk
, γ ,⊂ ∂K
ε
(q
t
) for
t ∈¦1, . . . , N¦, then the inner product in the integral is nonositive due to χ
r
χ
s
>0
in R
jk
. In case χ
r
χ
s
,= 0 along γ ⊂ R
jk
and γ ,⊂ ∂K
ε
(q
t
) for t ∈ ¦1, . . . , N¦, this
arc is boundary of two such domains, and we integrate in opposite directions
respectively. Now let γ
jk
⊂∂K
ε
(q
t
) for t ∈ ¦1, . . . , N¦. We must show
lim
ε→0
∑
j,k
∑
1≤r<s≤j
_
γ
jk
¸∇
S,W
(χ
r
χ
s
), N )
S,W
dσ
S
≤0.
Due to the boundness of ψ =ψ(u, v) on B
∗
⊆B
′
it follows
¸
¸
¸
_
γ
jk
¸∇
S,W
(χ
r
χ
s
), N )
S,W
dσ
S
¸
¸
¸ =
¸
¸
¸
_
γ
jk
¸χ
r
∇
S,W
χ
s
+χ
s
∇
S,W
χ
r
, N )
S,W
dσ
S
¸
¸
¸
=
_
γ
jk
[χ
r
[[¸∇
S,W
χ
s
, N )
S,W
[ dσ
S
+
_
γ
jk
[χ
s
[[¸∇
S,W
χ
r
, N )
S,W
[ dσ
S
= const
_
γ
jk
_
[¸∇
S,W
χ
s
, N )
S,W
[ +[¸∇
S,W
χ
r
, N )
S,W
[
_
dσ
S
.
An arc β := N
−1
jkr
(γ
jk
) in
˚
B ⊂R
2
degenerates to a point
˚
B for ε →0. We get
_
γ
jk
¸∇
S,W
χ
r
, N)
S,W
dσ
S
=
_
γ
jk
¸∇
S,W
(ψ ◦ N
−1
jkr
), N )
S,W
dσ
S
=
_
β
∇ψ (N
1
, N
2
)dt
where we note
1 =¸N, N)
dσ
2
g
= (−K)W[N[
2
, N= (N
1
, N
2
)
as well as
dσ
S
=
√
−KW dt.
We ﬁnally proceed as in point 6 of the proof to show (15.35). The statement
follows. ⊓⊔
15.5 A curvature estimate
Application of the previous general curvature estimate
With the considerations so far we are able to realize the assumptions we made for
our curvature estimate from section 11.9 in case of Fminimal surfaces with at least
special weights W = W(Z).
352 15 Fminimal surfaces in R
3
Theorem 15.10. Let the weighted conformally parametrized and µstable Fmini
mal surface X : B →R
3
be given. Suppose that
µ >
1 +ω
0
2
.
Let furthermore the surface represent a geodesic disc B
r
(X
0
) of radius r > 0 with
center X
0
:=X(0, 0). Then there is a constant Θ =Θ(ω
0
, ω
2
, µ) ∈[0, +∞) such that
it holds the curvature estimate
κ
1
(0, 0)
2
+κ
2
(0, 0)
2
≤
1
r
2
Θ(ω
0
, ω
2
, µ)
with the principal curvatures κ
1
and κ
2
of the surface.
Proof. We only give a sketch of the proof.
1. From chapter 11, formula (11.33) we ﬁrst recall
[∆(r
−1
X)[ ≤2(1 +ω
0
)ω
2
[∇(r
−1
X)[[∇N[,
[∆N[ ≤(1 +ω
2
)[∇N[
2
in B. Note that this system is not coupled!
2. From Theorem 15.8 we infer an oscillation estimate for the spherical mapping N
such that we can proceed as in the proof of the curvature estimate Theorem11.13
to obtain a gradient bound of the following form
[∇N(w
0
)[ ≤k
3
(ω
0
, ω
2
, µ) for all w
0
∈ B
1−ν
0
(0, 0).
Inserting this into the differential inequality for the surface vector linearized the
system.
3. The constant d
0
∈ (0, +∞) for the Dirichlet growth estimate
D[X] ≤d
0
r
2
follows immediately fromthe previous lemma. Finally, the modulus of continuity
for the surface vector X follows from a convexhull property Theorem 11.12
taking account of K ≤0 for the Gaussian curvature K.
A curvature estimate due to Hoy
Following the considerations in Hoy [97] we prove a further curvature estimate for
µstable Wminimal surfaces assuming a growth condition for the curvature integra
which thus replaces the above smallness condition for the stability constant µ > 0.
15.5 A curvature estimate 353
Theorem 15.11. Let the µstable Fminimal surface X : B → R
3
with µ > 0 be
given. Assume that it represents a geodesi disc B
r
(X
0
) of radius r > 0 with cen
ter X
0
= X(0, 0). Then it holds the integral curvature estimate
−
σ
_
0
2π
_
0
K(ρ, ϕ)
_
P(ρ, ϕ)dρdϕ ≤
1 +ω
0
µ ln
r
σ
_
_
_
2π −
r
_
0
2π
_
0
K(ρ, ϕ)
_
P(ρ, ϕ)dρdϕ
_
_
_
for all 0 <σ ≤r.
Proof. For positive σ ∈ (0, r] we consider the nonnegative test function
Φ(ρ) :=
_
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
_
r
_
σ
dt
L(t)
≡c
0
∈ R for all 0 ≤ρ ≤σ
r
_
ρ
dt
L(t)
for σ <ρ ≤r
with the function
L(ρ) =
2π
_
0
_
P(ρ, ϕ)dϕ, 0 <ρ ≤r
satisfying
L
′
(ρ) = 2π −
ρ
_
0
2π
_
0
K(τ, ϕ)
_
P(τ, ϕ)dτdϕ,
L
′′
(ρ) = −
2π
_
0
K(ρ, ϕ)
_
P(ρ, ϕ)dϕ,
see section 14.2. Note that due to K ≤ 0 the derivatives L
′
(ρ) and L
′′
(ρ) are also
nonnegative for all ρ ∈ [0, r]. Together with the estimate
r
_
0
L
′′
(ρ)Φ(ρ)
2
dρ ≤
1 +ω
0
µ
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ
(compare with the derivation of a similar inequality in section 14.2) we compute
c
2
0
σ
_
0
L
′′
(ρ)dρ =
σ
_
0
L
′′
(ρ)Φ(ρ)
2
dρ ≤
r
_
0
L
′′
(ρ)Φ(ρ)
2
dρ
≤
1 +ω
0
µ
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ ≤ . . .
354 15 Fminimal surfaces in R
3
. . . =
1 +ω
0
µ
σ
_
0
Φ
′
(ρ)
2
L(ρ)dρ +
1 +ω
0
µ
r
_
σ
Φ
′
(ρ)
2
L(ρ)dρ
=
1 +ω
0
µ
r
_
σ
1
L(ρ)
2
L(ρ)dρ =
1 +ω
0
µ
r
_
σ
1
L(ρ)
dρ
=
(1 +ω
0
)c
0
µ
.
(15.36)
Comparing the graphical shape of the function L(ρ) with a usual theorem of inter
secting lines in plane Euclidean geometry we see
ρ
L(ρ)
≥
r
L(r)
for all ρ ∈ [σ, r]. Thus we obtain
c
0
=
r
_
σ
dρ
L(ρ)
≥
r
_
σ
r
L(r)ρ
dρ =
r
L(r)
ln
r
σ
,
and the previous estimate yields
σ
_
0
L
′′
(ρ)dρ ≤
1 +ω
0
µc
0
≤
(1 +ω
0
)L(r)
µr ln
r
σ
.
On the other hand, integrating L
′
(ρ) over ρ = 0. . . r gives (notice that L(0) = 0 and
K ≤0)
L(r)
r
≤2π −
r
_
0
2π
_
0
K(τ, ϕ)
_
P(τ, ϕ)dτdϕ,
such that we infer
−
σ
_
0
L
′′
(ρ)dρ ≤
1 +ω
0
µ ln
r
σ
_
_
_
2π −
r
_
0
2π
_
0
K(τ, ϕ)
_
P(τ, ϕ) dτdϕ
_
_
_
.
This proves the statement. ⊓⊔
Remarks
We want to conclude this section with the following remarks.
1. Under the Osserman condition
N(u, v) (0, 0, 1) ≥cosω for all (u, v) ∈ B
15.6 Theorems of Bernstein type 355
with an angle ω ∈ (0, π) as well as the smallness condition
−
__
B
KW dudv ≤e
1
< +∞
Sauvigny in [141], Corollary 1, proves a curvature estimate for Fminimal sur
faces X : B →R
3
.
2. Hoy in [97] proves a curvature estimate for immersions with quasiconformal
spherical image, socalled Quasiminimal surfaces, under an Osserman condition
of the above form. The important integral inequality
r
_
0
L
′′
(ρ)Φ(ρ)
2
dρ ≤const
r
_
0
Φ
′
(ρ)
2
L(ρ)dρ
(and therefore (15.36)) is essentially obtained from the quasiconformality.
15.6 Theorems of Bernstein type
We conclude our considerations with presenting various theorems of Bernstein type
for surfaces of minimal surface type with prescibed special weights W=W(Z). For
the following results we refer the reader to Fr¨ ohlich [64].
Complete and µstable Fminimal surfaces
First we clarify the terminology µstability for complete surfaces.
Deﬁnition 15.1. The complete immersion X : R
2
→R
3
is called µstable with a real
number µ > 0 if its restriction to an arbitrary compact domain Ω ⊂ R
2
is µstable
with this number.
Now from Theorem 15.10 we immediately infer
Theorem 15.12. Let X : R
2
→R
3
be a complete and µstable Fminimal surface
with µ >
1+ω
0
2
. Then this surface represents a plane.
Proof. Due to K ≤ 0 we can introduce geodesic discs B
r
(X
0
) for arbitrary r > 0.
Using conformal parameters we infer the curvature estimate
κ
1
(ξ, η)
2
+κ
2
(ξ, η)
2
≤
1
r
2
Θ(ω
0
, ω
2
, µ).
The statement follows from r →∞. ⊓⊔
Let us focus on some special cases.
356 15 Fminimal surfaces in R
3
Stable Fminimal surfaces
Let X : B →R
3
be a stable and critical point of the parametric functional F[X]. As
we have seen in section 15.3, the surface is also µstable with µ ≤ 2M
1
M
−1
2
. The
condition µ >
1+ω
0
2
thus leads to
2M
1
> M
2
. (15.37)
Corollary 15.1. Let X : B →R
3
be a conformally parametrized and µstable F
minimal surface representing a geodesic disc B
r
(X
0
) of radius r > 0 and with
center X
0
∈ X(0, 0). Suppose (15.37) holds true. Then there is a constant Θ =
Θ(ω
0
, ω
2
, µ) ∈ [0, +∞) such that
κ
1
(0, 0)
2
+κ
2
(0, 0)
2
≤
1
r
2
Θ(ω
0
, ω
2
, µ).
If additionally X : R
2
→R
3
is complete then it is afﬁnelinear.
Surfaces with prescribed spherical image
Let Ω
′
⊆Ω ⊂S
2
be covered by the restriction N[
B
′
. For ω
0
≥0 let a spherical angle
ω =ω(ω
0
) ∈ (0, +π) be chosen such that for the spherical cap S
2
ω
⊂S
2
there hold
Area(Ω) < Area(S
2
ω
), λ
1
(S
2
ω
) = µ(1 +ω
0
) (15.38)
with a real number µ >0. By the generalized theoremof Barbosa and do Carmo the
Fminimal surface X[
B
′
is then µstable with this number, i.e. we have
__
B
′
[∇ϕ[
2
dudv ≥ µ
__
B
′
ϕ
2
(−K)W dudv for all ϕ ∈C
∞
0
(B
′
, R).
Now this inequality is true for all sets B
′
⊂⊂B, and in each such set there are at
most ﬁnitely man branch points of the spherical mapping. On the other hand, note
that each admissable test function has compact support. Then we conclude
__
B
[∇ϕ[
2
dudv ≥ µ
__
B
ϕ
2
(−K)W dudv for all ϕ ∈C
∞
0
(B, R).
Corollary 15.2. Let X : B→R
3
be a weighted conformally parametrized Fminimal
surface representing a geodesic disc B
r
(X
0
). Let furthermore S
2
ω
⊂ S
2
be a spher
ical cap satisfying (15.38). Then with a real constant Θ =Θ(ω
0
, ω
2
, µ) ∈ [0, +∞),
where µ := (1 +ω
0
)
−1
λ
1
(S
2
ω
), it holds the curvature estimate
κ
1
(0, 0)
2
+κ
2
(0, 0)
2
≤
1
r
2
Θ(ω
0
, ω
2
, µ).
15.6 Theorems of Bernstein type 357
If the immersion is additionally deﬁned on the whole plane R
2
and complete then it
represents a plane.
Graphs of minimal surface type
Consider immersions of minimal surface type which can be represented as graphs
over the [x, y]plane. Introduce weighted conformal parameters (u, v) ∈ B. We thus
get a mapping with the property
Area(Ω) < 2π
for its spherical image. Recalling λ
1
(S
2
π/2
) = 2 for the ﬁrst eigenvalue of the spher
ical Laplacian on the half sphere(see section 13.5)) we set
µ :=
2
1 +ω
0
.
Moreover, µ >
1+ω
0
2
must be fulﬁlled.
Corollary 15.3. Let X : B →R
3
be a weighted minimal surface which can be repre
sented as a graph over the [x, y]plane and which represents a geodesic disc B
r
(X
0
).
Suppose that 0 ≤ω
0
<1. Then there is a real constant Θ =Θ(ω
0
, ω
2
, µ) ∈[0, +∞),
µ := 2(1 +ω
0
)
−1
, such that
κ
1
(0, 0)
2
+κ
2
(0, 0)
2
≤
1
r
2
Θ(ω
0
, ω
2
, µ).
If the graph is complete then it represents a plane.
Graphs with prescribed growth for the curvatura integra
Let us return to Hoy’s curvature estimate.
Corollary 15.4. Let the complete and µstable Fminimal surface X : R
2
→R
3
be
given. Assume that
lim
r→∞
1
ln
r
σ
r
_
0
2π
_
0
K(ρ, ϕ)
_
P(ρ, ϕ)dρdϕ = 0
for all σ > 0. Then the surface represents a plane.
358 15 Fminimal surfaces in R
3
Final remarks
1. Our version of this Bernsteintype theorem is somewhat weaker than Hoy’s ver
sion from [97]. There the reader ﬁnds the assumption
lim
r→∞
1
ln
α
(1 +r)
r
_
0
2π
_
0
K(ρ, ϕ)
_
P(ρ, ϕ)dρdϕ = 0
for arbitrary α ∈ (0, +∞).
2. Following [97], a growth condition of the form
r
_
0
L(ρ)dρ ≤d
0
r
2
, d
0
∈ (0, +∞),
with the above function L = L(ρ) already implies the condition
lim
r→∞
1
ln
α
(1 +r)
r
_
0
2π
_
0
K(ρ, ϕ)
_
P(ρ, ϕ)dρdϕ = 0.
Since due to the monotonicity of L = L(ρ) we know
L(r)r ≤
2r
_
r
L(ρ)dρ ≤
2r
_
0
L(ρ)dρ ≤4d
0
r
2
,
and therefore L(r) ≤4d
0
r. Analogously we infer
L
′
(r)r ≤
2r
_
r
L
′
(ρ)dρ ≤
2r
_
0
L
′
(ρ)dρ = L(2r) −L(0) = L(2r) ≤8d
0
r
which leads us to L
′
(r) ≤8d
0
. Summarizing we get the statement with
r
_
0
L
′′
(ρ)dρ
ln
α
(1 +r)
=
L
′
(r) −L
′
(0)
ln
α
(1 +r)
≤
8d
0
−2π
ln
α
(1 +r)
.
Within our framework such a d
0
is realizable by a a lower bound µ >
1+ω
0
2
.
References
1. Abresch, U.
Constant mean curvature tori in terms of elliptic functions
Journal f¨ ur die Reine und Angewandte Mathematik 374, 169–192, 1987
2. Almgren, F.
Plateau’s problem
W.A. Benjamin, Inc., 1966
3. Alt, W.
Lineare Funktionalanalysis
Springer, 2006
4. B¨ ar, C.
Elementare Differentialgeometrie
Walter de Gruyter, 2001
5. Bandle, C.
Isoperimetric inequalities and applications
Pitman Advanced Publishing Program, 1980
6. Barbosa, J.L.; do Carmo, M.
On the size of stable minimal surfaces in R
3
The American Journal of Mathematics 98, 515–528, 1974
7. Barbosa, J.L.; do Carmo, M.
A proof of a general isoperimetric inequality for surfaces
Mathematische Zeitschrift 162, 245–261, 1978
8. Barbosa, J.L.; do Carmo, M.
Stability of minimal surfaces and eigenvalues of the Laplacian
Mathematische Zeitschrift 173, 13–28, 1980.
9. Beckenbach, E.F.; Rad´ o, T.
Subharmonic functions and surfaces of negative curvature
Transactions of the American Mathematical Society 35, 662–674, 1933
10. Begehr, H.G.W.; Wen, G.C.
Nonlinear boundary value problems and their applications
Addison Wesley Longman Ltd., 1990
11. Bergner, M.
Konstruktion von Fl ¨ achen vorgeschriebener mittlerer Kr¨ ummung mit der Kontinui
t ¨ atsmethode
Dissertation, Darmstadt, 2006
359
360 References
12. Bergner, M.; Dittrich, J.
On surfaces of prescribed weighted mean curvature
Calculus of Variations and Partial Differential Equations 33, 169–185, 2008
13. Bergner, M.; Fr¨ ohlich, S.
On twodimensional immersions of prescribed mean curvature in R
n
Zeitschrift f¨ ur Analysis und ihre Anwendungen 27, No. 1, 31–52, 2008
14. Bers, L.
Univalent solutions of Linear elliptic systems
Communications on Pure and Applied Mathematics VI, 513–526, 1953
15. Blaschke, W.; Leichtweiss, K.
Elementare Differentialgeometrie
Springer, 1973
16. B¨ ohme, R.; Hildebrandt, S.; Tausch, E.
The twodimensional analogue of the catenary
Paciﬁc Journal of Mathematics 88, No. 2, 1980
17. Bourgain, J.; Brezis, H; Mironescu, P.
Lifting in Sobolev spaces
Journal d’Analyse Mathematique 80, 37–86, 2000
18. Brauner, H.
Differentialgeometrie
Friedr. Vieweg & Sohn Verlagsgesellschaft mbH, 1981
19. Brezis, H.; Coron, J.M.
Multiple solutions of Hsystems and Rellich’s conjecture
Communications on Pure and Applied Mathematics 37, 149–187, 1984
20. Burago, Y.D.; Zalgaller, V.A.
Geometric inequalities
Springer, 1988.
21. do Carmo, M.P.
Differential geometry of curves and surfaces
PrenticeHall, Inc., 1976
22. do Carmo, M.P.
Riemannian geometry
Birkh¨ auser, 1992
23. Cartan, H.
Differentialformen.
BI Wissenschaftsverlag, 1974.
24. Cesari, L.
An existence theorem of the calculus of variations for integrals on parametric surfaces
The American Journal of Mathematics 74, 265–295, 1952
25. Chavel, I.
Isoperimetric inequalities
Cambridge University Press, 2001
26. Chen, B.Y.
Surfaces with parallel mean curvature vector
Bulletin of the American Mathematical Society 78, No. 5, 709–710, 1972
27. Chen, B.Y.
Geometry of submanifolds
Marcel Dekker, Inc., 1973
28. Chern, S.S.
An elementary proof of the existence of isothermal parameters on a surface
Proceedings of the American Mathematical Society 6, No. 5, 771–782, 1955
References 361
29. Chern, S.S.; Hartman, P.; Wintner, A.
On isothermic coordinates
Commentarii Mathematici Helvetici 28, 301–309, 1954
30. Clarenz, U.
Enclosure theorems for extremals of elliptic parametric functionals
Calculus of Variations and Partial Differential Equations 15, 313–324, 2002
31. Clarenz, U.
A stability criterion for extremals of elliptic parametric functionals
Paciﬁc Journal of Mathematics 208, No. 2, 231–242, 2003
32. Clarenz, U.; von der Mosel, H.
On surfaces of prescribed Fmean curvature
Paciﬁc Journal of Mathematics 123, 15–36, 2004
33. Cluttberbuck, J.
Interior gradient estimates for anisotropic mean curvature ﬂow
Paciﬁc Journal of Mathematics 229, No. 1, 119–136, 2007
34. Coifman, R.; Lions, P.L.; Meyer, Y.; Semmes, S.
Compensated compactness and Hardy spaces
Journal de Mathematique Pures et Appliquees 9, No. 3, 247–286, 1993
35. Colding, T.H.; Minicozzi, W.P.
Minimal surfaces
Courant Institute of Mathematical Sciences, 1999
36. Colding, T.H.; Minicozzi, W.P.
Estimates for parametric elliptic integrands
International Mathematics Research Notices 4, 291–297, 2002
37. da Costa, R.C.T.
Constraints in quantum mechanics
Physical Review A 25, No. 6, 2893–2900, 1982
38. Courant, R.
Dirichlet’s principle, conformal mappings, and minimal surfaces
Interscience publishing, Inc., New York, 1950
39. Courant, R.; Hilbert, D.
Methoden der mathematischen Physik
Springer, 1993
40. Courant, R.; Hilbert, D.
Methods of mathematical physics 2
John Wiley & Sons, Inc., 1962
41. Danskin, J.M.
On the existence of minimizing surfaces in parametric double integral problems in the
calculus of variations
Rivista di Matematica della Universita di Parma 3, 43–63, 1952.
42. Dierkes, U.
Maximum principles and nonexistence results for minimal surfaces
manuscripta mathematica 69, 203–218, 1990
43. Dierkes, U.
Maximum principles for submanifolds of arbitrary codimension and bounded mean cur
vature
Calculus of variations and partial differential equations 22, 173–184, 2005
44. Dierkes, U.; Hildebrandt, S.; K¨ uster, A.; Wohlrab, O.
Minimal surfaces
Springer, 1992
362 References
45. Dobriner, H.
Die Fl ¨ achen constanter Kr¨ ummung mit einem System sph¨ arischer Kr¨ ummungslinien dar
gestellt mithilfe von Thetafunctionen
Acta Mathematics 9, 73–104, 1886
46. Dobrowolski, M.
Angewandte Funktionalanalysis
Springer, 2005
47. Douglas, J.
Solution of the problem of Plateau
Transactions of the American Mathematical Society 33, No. 1, 263–321, 1931
48. Douglas, J.
The higher topological form of the problem of Plateau
Annali della Scuola Normale Superiore di Pisa (2) 8, 195–218, 1939
49. Douglas, J.
The most general form of the problem of Plateau
The American Journal of Mathematics 61, 59–608, 1939
50. Ecker, K.
Regularity theory for mean curvature ﬂow
Birk¨ auser, 2004
51. Enmoto, K.
Umbilical points on surfaces in R
N
Nagoya Mathematics Journal 100, 135–143, 1985
52. Eschenburg, J.H.; Jost, J.
Differentialgeometrie und Minimalﬂ¨ achen
Springer, 2007
53. Evans, L.C.
Partial Differential equations
Oxford University Press, 1998
54. Federer, H.; Fleming, W.H.
Normal and integral currents
Annals of Mathematics 72, 458–520, 1960
55. Fefferman, C.; Stein, E.M.
H
p
spaces of several variables
Acta Mathematica 129, 137–193, 1972
56. Fenchel, W.
¨
Uber Kr¨ ummung und Windung geschlossener Raumkurven
Mathematische Annalen 101, 238–252, 1929
57. Ferapontov, E.V.
Surfaces with ﬂat normal bundle: An explicit construction
Differential geometry and its Applications 14, No. 1, 15–37, 2001
58. Finn, R.
On equations of minimal surface type
Annals of Mathematics 60, No.3, 1954
59. Finsler, P.
¨
Uber Kurven und Fl ¨ achen in allgemeinen R¨ aumen
Birkh¨ auser, 1951
60. FischerColbrie, D.; Schoen, R.
The structure of complete stable minimal surfaces in 3manifolds of nonnegative scalar
curvature
Communications on Pure and Applied Mathematics 33, 199–211, 1980
References 363
61. Fomin, S.V.; Gelfand, I.M.
Calculus of variations
PrenticeHall, Inc., 1963
62. Fr¨ ohlich, S.
Curvature estimates for µstable Gminimal surfaces and theorems of Bernstein type
Analysis 22, 109–130, 2002
63. Fr¨ ohlich, S.
On 2surfaces in R
4
and R
n
Proceedings of the 5th Conference of Balkan Society of Geometers, Mangalia 2005
64. Fr¨ ohlich, S.
¨
Uber zweidimensionale elliptische Systeme der geometrischen Analysis
Habilitation thesis, Technical University Darmstadt, 2008
65. Fr¨ ohlich, S.; M¨ uller, F.
On critical normal sections for twodimensional immersions in R
4
and a RiemannHilbert
problem
Differential Geometry and its Applications 26, No. 5, 508–513, 2008
66. Fr¨ ohlich, S.; M¨ uller, F.
On critical normal sections for twodimensional immersions in R
n+2
Calculus of Variations and Partial Differential Equations 35, No. 4, 497–515, 2009
67. Fr¨ ohlich, S.; Winklmann, S.
Curvature estimates for graphs with prescribed mean curvature and ﬂat normal bundle
manuscripta mathematica 122, 149–162, 2007
68. Funk, P.
Variationsrechnung und ihre Anwendungen in Physik und Technik
Springer, 1962
69. Garnier, R.
Le probl´ em de Plateau
Annales Scientiﬁques de l’Ecole Normale Superieure (3) 45, 53–144, 1928
70. Gauss, C.F.
General investigations of curved surfaces
Dover Publications, Inc., 2005
71. Gilbarg, D.; Trudinger, N.S.
Elliptic Partial Differential Equations of Second Order
Springer, 1983
72. Giusti, E.
Minimal surfaces and functions of bounded variation
Birkh¨ auser, 1984
73. Glaeser, G.; Polthier, K.
Bilder der Mathematik
Spektrum Akademischer Verlag, 2009
74. Grassmann, H.
Die neuere Algebra und die Ausdehnungslehre
Mathematische Annalen 7, Nr. 4, 1874
75. Gulliver, R.
Existence of surfaces with prescribed mean curvature vector
Mathematische Zeitschrift 131, 117–140, 1973
76. Gulliver, R.
Minimal surfaces of ﬁnite index in manifolds of positive scalar curvature
In: Calculus of Variations and partial differential equations, S. Hildebrandt, D. Kinder
lehrer, M. Miranda, Ed., Springer, 1988
364 References
77. Hartman, P.; Wintner, A.
On the local behavior of solutions of nonparabolic partial differential equations
The American Journal of Mathematics 75, 449–476, 953
78. Heil, E.
Differentialformen und Anwendungen auf Vektoranalysis, Differentialgleichungen, Geo
metrie
BI Wissenschaftsverlag, 1974
79. Heinz, E.
¨
Uber Fl ¨ achen mit eineindeutiger Projektion auf eine Ebene, deren Kr¨ ummungen durch
Ungleichungen eingeschr¨ ankt sind
Mathematische Annalen 129, 451–454, 1955
80. Heinz, E.
On certain nonlinear elliptic differential equations and univalent mappings
J. d’Analyse Mathematique 5, 197–272, 1957
81. Heinz, E.
¨
Uber das Randverhalten quasilinearer elliptischer Systeme mit isothermen Parametern
Mathematische Zeitschrift 113, 99–105, 1970
82. H´ elein, F.
Harmonic maps, conservation laws and moving frames
Cambridge University Press, 2002
83. Hellwig, G.
Partial differential equations
B.G. Teubner, 1977
84. Hildebrandt, S.
Einige Bemerkungen ¨ uber Fl ¨ achen beschr¨ ankter mittlerer Kr¨ ummung
Mathematische Zeitschrift 159, 169–178, 1970
85. Hildebrandt, S.
¨
Uber einen neuen Existenzsatz f ¨ ur Fl ¨ achen vorgeschriebener mittlerer Kr¨ ummung
Mathematische Zeitschrift 119, 267–272, 1971
86. Hildebrandt, S.; von der Mosel, H.
On twodimensional parametric variational problems
Calculus of variations and partial differential equations 9, 1999, 249–267
87. Hildebrandt, S.; von der Mosel, H.
Plateau’s problem for parametric double integrals: Part I. Existence and regularity in the
interior
Communications on Pure and Applied Mathematics 56, 926–955, 2003
88. Hildebrandt, S.; von der Mosel, H.
Plateaus’s problem for parametric double integrals: Part II. Regularity at the boundary
Journal f¨ ur die reine und angewandte Mathematik 565, 207–233, 2003
89. Hildebrandt, S.; von der Mosel, H.
Dominance functions for parametric Lagrangians
In: Geometric analysis and nonlinear partial differential equations, Springer, 297–326,
2003
90. Hildebrandt, S.; von der Mosel, H.
On Lichtenstein’s theorem about globally conformal mappings
Calculus of variations and partial differential equations 23, 415–424, 2005
91. Hildebrandt, S.; von der Mosel, H.
Conformal representation of surfaces, and Plateau’s problem for Cartan functionals
Rivista di Matematica della Universita di Parma 7, No. 4, 1–43, 2005
References 365
92. Hoffman, D.A.
Surfaces in constant curvature manifolds with parallel mean curvature vector ﬁeld
Bulletin of the American Mathematical Society 78, No. 2, 247–250, 1972
93. Hoffman, D.A.; Osserman, R.
Geometry of the generalized Gauss map
Memoirs of the American Mathematical Society 28, No. 236, 1980.
94. Hopf, H.
¨
Uber Fl ¨ achen mit einer Relation zwischen den Hauptkr¨ ummungen
Mathematische Nachrichten 4, 232–249, 1950/51
95. Hoy, E.
¨
Uber Fl ¨ achen mit quasikonformer Gaußscher Abbildung
Wissenschaftliche Zeitschrift der MartinLutherUniversit¨ at HalleWittenberg, Mathe
matischNaturwissenschaftliche Reihe 38, No. 4, 123–129, 1989
96. Hoy, E.
Zum Bernsteinschen Satz f ¨ ur Quasiminimalﬂ¨ achen
Zeitschrift f¨ ur Analysis und ihre Anwendungen 10, No. 4, 569–572, 1991
97. Hoy, E.
On Bernstein’s theorem for quasiminial surfaces II
Zeitschrift f¨ ur Analysis und ihre Anwendungen 16, No. 4, 1027–1032, 1997
98. Hurwitz, A.
Quelques applications geometriques des series de Fourier
Annales de l’
´
Ecole Normale 19, No. 3, 357–408, 1902
99. Jenkins, H.B.
On twodimensional variational problems in parametric form
Archive for Rational Mechanics and Analysis 8, 181–206, 1961
100. Jorge, L.P.; Tomi, F.
The barrier principle for minimal submanifolds of arbitrary codimension
Annals of Global Analysis and Geometry 24, 261–267, 2003
101. Jost, J.
Twodimensional variations problems
Wiley Interscience, 1991
102. Jost, J.
Riemannian geometry and geometric analysis
Springer, 1995
103. Jost, J.
Partielle Differentialgleichungen
Springer, 1998
104. Jost, J.; Xin, Y.L.
Bernstein type theorems for higher codimension
Calculus of variations and Partial differential equations 9, No. 4, 277–296, 1999
105. Kawai, S.
A theorem of Bernstein type for minimal surfaces in R
4
Tohoku Mathematical Journal 36, 377–384, 1984
106. Klingenberg, W.P.A.
Klassische Differentialgeometrie
Edition am Gutenbergplatz, 2004
107. Koiso, M.; Palmer, B.
Geometry and stability of surfaces with constant anisotropic mean curvature
Indiana Univiversity Mathematics Journal 54, 1817–1852, 2005
366 References
108. Koiso, M.; Palmer, B.
Stability of anisotropic capillary surfaces between two parallel planes
Calculus of variations and partial differential equations 25, 275–298, 2006
109. K¨ uhnel, W.
Differentialgeometrie
Friedr. Vieweg & Sohn Verlagsgesellschaft mbH, 1999
110. Kurzke, M.; von der Mosel, H.
The Douglas problem for parametric double integrals
manuscripta mathematica 110, 93–114, 2003
111. Laugwitz, D.
Differentialgeometrie
B.G. Teubner, 1977
112. Lawson, H.B.
Lectures on minimal submanifolds
Publish or Perish, Inc., 1980
113. Liebmann, H.
Eine neue Eigenschaft der Kugel.
Nachrichten der K¨ oniglichen Gesellschaft Wissenschaften zu G¨ ottingen, Mathematisch
Physikalische Klasse, 4455, 1899
114. Liebmann, H.
Ueber die Verbiegung der geschlossenen Fl ¨ achen positiver Kr¨ ummung
Mathematische Annalen 53, 91–112, 1900
115. Lin, F.H.
Estimates for surfaces which are stationary for an elliptic parametric integral
Journal of Partial Differential Equations 3, 78–92, 1990
116. Liu, X.; Terng, C.L.
The mean curvature ﬂow for isoparametric submanifolds
Duke Mathematical Journal 147, 157–179, 2009
117. Lonseth, A.T.
The problem of Plateau in hyperbolic spaces
The American Journal of Mathematics 64, No. 1, 229–259, 1942
118. Maz’ya, V. G.
Classes of domains and imbedding theorems for function spaces
Soviet Mathematics Doklady 1, 882–885, 1960
119. Michael, J.H.; Simon, L.M.
Sobolev and meanvalue inequalities on generalized submanifolds of R
n
Communications on Pure and Applied Mathematics XXVI, 361–379, 1973
120. Morgan, F.
Geometric measure theory: A beginner’s guide
Academic Press Inc., 2000
121. Morrey, C.B.
The problem of Plateau on a Riemannian manifold
Annals of Mathematics 49, 807–851, 1948
122. Morrey, C.B.
The parametric variational problem for double integrals
Communications on Pure and Applied Mathematics 14, 569–575, 1961
123. Morrey, C.B.
Multiple integrals in the calculus of variations
Springer, 1966
References 367
124. M¨ uller, F.; Schikorra, A.
Boundary regularity via UhlenbeckRiviere decomposition
Analysis 29, 199–220, 2009
125. Neutsch, W.
Koordinaten. Theorie und Anwendungen
Spektrum Akademischer Verlag, 1995
126. Nitsche, J.C.C.
Vorlesungen ¨ uber Minimalﬂ¨ achen
Springer, 1975
127. Osserman, R.
Global properties of minimal surfaces in E
3
and E
n
Annals of Mathematics 80, No. 2, 340–364, 1964
128. Osserman, R.
A proof of the regularity everywhere of the classical solution to Plateau’s problem
Annals of Mathematics 91, 550–569, 1970.
129. Osserman, R.
The isoperimetric inequality
Bulleting of the American Mathematical Society 84, No. 6, 1978
130. Osserman, R.
A survey of minimal surfaces
Dover Publications, Inc., 1986
131. Osserman, R.; Ru, M.
An estimate for the Gauss curvature of minimal surfaces in R m whose Gauss map omits
a set of hyperplanes
Journal of Differential Geometry 46, 578–593, 1997
132. Pozzi, P.
Anisotropic mean curvature ﬂow in higher codimension
PAMM 8, No. 1, 10849–10850, 2009.
133. Rad´ o, T.
On Plateau’s problem
Annals of Mathematics 31, No. 2, 457–469, 1930
134. R¨ awer, K.
Stabile Extremalen parametrischer Doppelintegrale im R
3
Dissertation, Bonn, 1993
135. Raschewski, P.K.
Riemannsche Geometrie und Tensoranalysis
Verlag Harri Deutsch, 1995
136. Rivi` ere, T.
Conservation laws for conformally invariant variational problems
Inventiones Mathematicae 168, 1–22, 2007.
137. Ruchert, H.
Ein Eindeutigkeitssatz f ¨ ur Fl ¨ achen konstanter mittlerer Kr¨ ummung
Archiv der Mathematik 33, 91–104, 1979
138. Sakai, T.
Riemannian geometry
Translation of mathematical monographs 149, AMS, 1996
139. Sato, S.
Barta’s inequality and the ﬁrst eigenvalue of a cap domain of a 2sphere
Mathematische Zeitschrift 181, No. 3, 313–318, 1982
368 References
140. Sauvigny, F.
AprioriAbsch¨ atzungen der Hauptkr¨ ummungen f ¨ ur Immersionen vom Mittleren
Kr¨ ummungsType mittels Uniformisierung und S¨ atze vom BernsteinTyp
Habilitationsschrift, G¨ ottingen, 1989
141. Sauvigny, F.
Curvature estimates for immersions of minimal surface type via uniformization and theo
rems of Bernstein type
manuscripta mathematica 67, 567–582, 1990
142. Sauvigny, F.
Introduction of isothermal parameters into a Riemannian metric by the continuity method
Analysis 19, No. 3, 235–243, 1999
143. Sauvigny, F.
Partielle Differentialgleichungen der Geometrie und der Physik
Springer, 2004/2005
144. Schauder, J.
¨
Uber das Dirichletsche Problem im Großen f ¨ ur nichtlineare elliptische Differentialglei
chungen
Mathematische Zeitschrift 37, 623–634, 1933
145. Schoen, R.; Simon, L.; Yau, S.T.
Curvature estimates for minimal hypersurfaces
Acta Mathematica 134, 275–288, 1975
146. Schulz, F.
Regularity theory for quasilinear elliptic systems and MongeAmper´ e equations in two
dimensions
Springer, 1990
147. Simon, L.
A H¨ older estimate for quasiconformal maps between surfaces in Euclidean space
Acta Mathematica 139, 19–51, 1977
148. Smoczyk, K.; Wang, G.; Xin, Y.L.
Mean curvature ﬂow with ﬂat normal bundle
arXiv:math/0411010v1, 2004
149. Smoczyk, K.; Wang, G.; Xin, Y.L.
Bernstein type theorems with ﬂat normal bundle
Calculus of variations and partial differential equations 26, No. 1, 2006
150. Spruck, J.
Remarks on the stability of minimal submanifolds in R
n
Mathematische Zeitschrift 144, 169–174, 1975
151. Stein, E.M.
Harmonic Analysis: Realvariable methods, orthogonality, and oscillatory integrals
Princeton University Press, 1993
152. Strubecker, K.
Differentialgeometrie
Walter de Gruyter & Co., 1964/1969
153. Taylor, J.E.
Geometric models of crystal growth
Acta Metallurgica et Materialia 40, No. 7, 1443–1474, 1992
154. Taylor, J.E.
Mean curvature and weighted mean curvature
Acta Metallurgica et Materialia 40, No. 7, 1475–1485, 1992
References 369
155. Taylor, M.E.
Partial differential equations III: Nonlinear equations
Springer, 2004
156. Terng, C.L.
Isoparametric submanifolds and their Coxeter group
Journal Differential of Geometry 21, 79–107, 1985
157. Tomi, F.
Ein einfacher Beweis eines Regularit ¨ atssatzes f ¨ ur schwache L¨ osungen gewisser ellipti
scher Systeme
Mathematische Zeitschrift 112, 214–218, 1969
158. Topping, P.
Mean curvature ﬂow and geometric inequalities
Journal f¨ ur die reine und angewandte Mathematik 503, 47–61, 1988
159. Topping, P.
The optimal constant in Wente’s L
∞
estimate
Commentarii Mathematici Helvetici 72, 316–328, 1997
160. Vekua, I.N.
Systeme von Differentialgleichungen vom elliptischen Typus und Randwertaufgaben
Deutscher Verlag der Wissenschaften, 1956.
161. Vekua, I. N.
Verallgemeinerte analytische Funktionen
AkademieVerlag, 1963
162. Wang, M.T.
Stability and curvature estimates for minimal graphs with ﬂat normal bundles
arXiv:math/0411169v2, 2004
163. Weisstein, E.
Wolfram MathWorld
Created, developed, and nurtured by Eric Weisstein with contributions from the world’s
mathematical community
164. Wendland, W.L.
Elliptic systems in the plane
Pitman Publishing Ltd., 1979
165. Wente, H. C.
The differential equation ∆x = 2Hx
u
∧x
v
with vanishing boundary values
Proceedings of the American Mathematical Society 50, 131137, 1975
166. Wente, H.C.
Large solutions to the volume constrained Plateau problem
Archive for Rational Mechanics and Analysis 75, 59–77, 1980
167. Wente, H.C.
Counterexample to a conjecture of H. Hopf
Paciﬁc Journal of Mathematics 121, No. 1, 193–243, 1986
168. Weyl, H.
Zur Inﬁnitesimalgeometrie: pdimensionale Fl ¨ achen im ndimensionalen Raum
Mathematische Zeitschrift 12, 154–160, 1922
169. Weyl, H.
SpaceTimeMatter
Dover Publications, Inc., 1952
170. White, B.
Curvature estimates and compactness theorems in 3manifolds for surfaces that are sta
tionary for parametric elliptic functionals
Inventiones Mathematicae 88, 243–256, 1987
370 References
171. White, B.
Existence of smooth embedded surfaces of prescribed genus that minimize parametric
even elliptic functionals on 3manifolds
Journal of Differential Geometry 33, 413–443, 1991
172. Winklmann, S.
Integral curvature estimates for Fstable hypersurfaces
Calculus of variations and partial differential equations 23, 391–414, 2005
173. Winklmann, S.
Pointwise curvature estimates for Fstable hypersurfaces
Annales de l’Institut Henry Poincar´ e (C) Analyse Non Lin´ eaire 22, 543–555, 2005
174. Wulff, G.
Zur Frage der Geschwindigkeit des Wachstums und der Auﬂ¨ osung der Krystallﬂ¨ achen
Zeitschrift f¨ ur Krystallographie und Mineralogie 34, 449–530, 1901
175. Xin, Y.L.; Yang, L.
Curvature estimates for minimal submanifolds of higher codimension
arXiv:0709.3686, 2007
176. Yau, S.T.:
Submanifolds with constant mean curvature I.
The American journal of mathematics 96, No. 2, 346–366, 1974.
List of Names
Abresch, U., 36
Almgren, F., 268
Ammann, B., 36
B¨ ar, C., 17
B¨ ohme, R., 189, 195
Bandle, C., 302, 303, 327
Barbosa, J.L., 302–306, 324, 346, 347
Bauner, H., 56
Beckenbach, E.F., 327
Begehr, H.G.W., 114
Bergner, M., 61, 65, 184, 195, 315
Bers, L., 76, 83, 233, 234
Blaschke, L, 42
Blaschke, W., 14, 22, 52, 282, 301,
307, 313, 324
Bourgain, J., 154
Brauner, H., 30, 63
Brezis, H., 154, 157
Burago, Y.D., 168
Carleman, T.G.T., 327
Cartan, H., 87, 307
Cesari, L., 268
Chavel, I., 327
Chen, B.Y., 30, 35, 112
Chern, S.S., 13, 14
Clarenz, U., 61, 62, 65, 255, 271
Clutterbuck, J., 65, 271
Codazzi, D., 42
Coifman, R., 155
Colding, T.H., 26, 288
Coron, J.M., 157
Courant, R., 26, 83, 114, 303
da Costa, R.C.T., 51
Danskin, J.M., 268
Dierkes, U., 26, 76, 168, 184, 195,
255
Dittrich, J., 61, 65
do Carmo, M.P., 17, 30, 50, 129, 302–
306, 324, 346, 347
Dobriner, H., 36
Dobrowolski, M., 148
Douglas, J., 267
Ecker, K., 208, 271
Enmoto, K., 36
Eschenburg, J., 26, 308
Euler, L.P., 211
Evans, L.C., 164
Federer, H., 327
Fefferman, C., 155
Fenchel, W., 318
Ferapontov, E.V., 50
Finn, R., 238
Finsler, P., 60
FischerColbrie, D., 288, 315
Fleming, W.H., 327
Fomin, S.V., 164
Fr¨ ohlich, S., 61, 62, 196, 271, 288,
300, 315, 343, 344, 346
Funk, P., 60, 164
Garnier, R., 267
Gauss, C.F., 12, 21, 33
Gelfand, I.M., 164
Gerhardt, C., 36
Gilbarg, D., 143, 148, 170, 171, 185,
207, 336, 338
Giusti, E., 26, 307
Glaeser, G., 36
Grassmann, H., 87
GrosseBrauckmann, K., 36
Gulliver, R., 189, 312
Hartman, P., 14, 332
Heil, E., 87, 307
Heinz, E., 139, 168, 199–201, 208,
235, 345
371
372 LIST OF NAMES
Helein, F., 36, 110, 149–151
Hellwig, G., 302
Henneberg, E.L., 212
Hilbert, D., 83, 114, 303
Hildebrandt, S., 167, 189, 190, 195,
268–270, 332
Hoffman, D., 36, 94, 98, 101
Hopf, H., 26, 36, 76, 79
Hoppe, R., 12
Hoy, E., 335, 352, 355, 357, 358
Hurwitz, A., 324
Isenberg, J., 36
Jenkins, H.B., 268
Jorge, L.P., 184
Jost, J., 26, 76, 79, 208, 302, 308, 332
K¨ uhnel, W., 17
Kapouleas, N., 36
Karcher, H., 36
Kawai, S., 173
Kenmotsu, K., 36
Kilian, M., 36
Klingenberg, W., 14, 171, 229, 311,
312
Koiso, M., 271
Kurzke, M., 268
Kusner, R., 36
Laugwitz, D., 14, 229
Lawson, H.B., 26
Leichtweiss, K., 14, 22, 42, 52, 282,
301, 307, 313, 324
Liebmann, H., 35
Lin, F.H., 271
Lions, P.L., 155
Liu, X., 126
Lonseth, A.T:, 268
Lopez, R., 36
M¨ uller, F., 157
Mainardi, G., 42
Martin, F., 36
Maz’ya, V.G., 327
Meeks, W.H., 36
Meusnier de la Place, J.B.M.C., 212
Meyer, Y., 155
Michael, J.H., 306
Minicozzi, W.P., 26, 288
Mironescu, P., 154
Morgan, F., 214
Morrey, C.B., 268
Neil, W., 7
Neutsch, W., 334
Nitsche, J.C.C., 26, 168, 213, 238,
289
Osserman, R., 7, 26, 94–98, 101, 168,
173, 174, 176, 196, 324, 327
Palmer, B., 271
Pedit, F., 36
Polthier, K., 36
Pozzi, P., 271
R¨ awer, K., 271
Rado, T., 267, 327
Raschewski, P.K., 17
Riemann, B., 43
Riviere, T., 157
Ruchert, H., 101, 305
Sakai, T., 302, 303
Sato, S., 303
Sauvigny, F., 13, 59, 61, 62, 64, 83,
113–115, 144, 145, 168, 170,
171, 196, 204, 208, 214, 220,
231, 235, 237, 238, 254, 257,
271, 312, 317, 332, 334, 336,
355
Schauder, J., 184, 185
Scherk, H., 212
Schikorra, A., 157
Schmidt, N., 36
Schoen, R., 208, 271, 288, 315
Schwarz, H.A., 289, 344
Semmes, S., 155
Simon, L., 208, 271, 306, 335
Smoczyk, K., 126, 208
Spruck, J., 306
LIST OF NAMES 373
Stein, E.M., 150
Steiner, J., 326
Strubecker, K., 12, 14, 212
Sullivan, J., 36
Tausch, E., 189, 195
Taylor, J.E., 214, 267
Taylor, M.E., 182
Terng, C.L., 126
Tomi, F., 158, 184
Topping, P., 140, 327
Trudinger, N.S., 143, 148, 170, 171,
185, 207, 336, 338
Vekua, I.N., 76, 78, 83, 114, 116
von der Mosel, H., 61, 65, 268–271
Wang, G., 126, 208
Wang, M.T., 208
Weber, M., 36
Weingarten, J., 31
Weisstein, E., 7
Wen, G.C., 114
Wendland, W.L., 82, 83, 114
Wente, H., 36, 134, 136, 140, 149
Weyl, H., 21, 30, 56
White, B., 268
Winklmann, S., 61, 62, 271
Wintner, A., 14, 332
Wulff, G., 214, 267
Xin, Y.L., 126, 208
Yau, S.T., 208, 271
Zalgaller, V.A., 168
Index
L
p
estimate, 148
µstable, 288, 313
angle between curves, 12
area element, 6, 165
area estimate, 313, 316, 317, 319
area functional, 161, 189, 211, 224,
267
ﬁrst variation, 161
nonparametric, 165, 231
second variation, 279
associated Lagrangian, 269
Beltrami operator
ﬁrst, 220, 222, 282
Bour theorem, 212
branch point, 7, 332
calibration, 7
calibration form, 307
Catalan surface, 212
catenary curve, 211
catenoid, 14, 211
catenoidhelicoid evolution, 212
Cauchy integral operator, 115
CauchyRiemann equation, 9, 114, 332
inhomogeneous, 79
Christoffel symbols, 21
conformal form, 26
corrective terms, 63
Clifford surface, 50
CodazziMainardi equations, 39–41
weighted form, 67
complete immersion, 355
conformal parameters, 9, 13
weighted, 64
conformality relations, 13
constant mean curvature surface, 225
convex hull property, 255
Coulomb gauge, 110
CourantLebesgue lemma, 202, 345
crystall growth, 214
crystalline variational problem, 252,
267
curvatura integra, 75, 306, 317
curvature matrix of a normal bundle,
92
curvature of the normal bundle, 49
evolution, 121
curvature tensor, 43
curvature tensor of the normal bun
dle, 46
curvature vector of the normal bun
dle, 49
curve
binormal vector, 107
curvature, 108
tangential vector, 107
torsion, 108
unit normal vector, 107
curve shortening ﬂow, 327
dilatation, 238
Dirichlet boundary value problem, 112,
147
divergence operator, 190
dominance function, 269
positive deﬁnite, 269
quadratic, 269
DuBoisReymond lemma, 154
eigenfunction, 301, 302
eigenvalue, 301, 302
monotonicity of the ﬁrst, 302
eigenvalue problem, 301, 341
enclosure theorem, 255
Enneper surface, 212
Euclidean gradient, 9
Euclidean metric, 11
Euler homogeneity condition, 215, 272
374
INDEX 375
Euler rotation, 10
Euler unit normal vector, 8
exponential map, 14, 171, 311
exterior product, 87, 272
Fminimal surface, 59, 214, 226, 251,
331
µstable, 340, 343
Bernstein type theorem, 355–357
curvature estimate, 352, 353
spherical oscillation, 344
Fenchel theorem, 318
Finsler space, 60
ﬂat normal bundle, 35, 47, 112, 289,
306
Fourier expansion, 325
Frenet equations, 108
Friedrich approximation, 156
FubiniStudy metric, 96, 305
fundamental form
ﬁrst, 11, 18, 59, 162, 165
second, 12, 229
third, 12
weighted ﬁrst, 59, 62
weighted second, 62
weighted third, 62
fundamental theorem of surface the
ory, 51
Gauss curvature, 26, 33, 44, 92, 96,
172
conformal form, 44
Gauss equations, 21, 275
weighted form, 63
Gauss integrability equations, 42
Gauss map, 72
energy, 74
GaussOsserman map, 95
generalized analytic function, 114
geodesic curve, 171
geodesic disc, 14, 172, 311
area estimate, 313, 317
geodesic ﬂow, 129
geodesic polar coordinates, 14, 311
geodesic radius, 311
geometric maximum principle, 195,
254
gradient estimate, 260, 337
global, 169
interior, 169
graph of minimal surface type, 238
Grassmann curvature vector, 92
Grassmann manifold, 94
Grassmann normal space, 89
GrassmannWeingarten equations, 90,
275
Green function, 113, 140, 143
Gulliver functional, 189, 197, 283, 318
H¨ older inequality, 143
Hadamard theorem, 172, 207
Hardy space, 149, 155
RieszFourier transformdeﬁnition,
149
tempered distribution deﬁnition, 149
Harnack inequality, 201, 260
HartmanWintner theorem, 332
helicoid, 14, 212
Henneberg surface, 7, 212
Hilbert selection theorem, 153
Hildebrandt functional, 190, 193, 213,
225
holomorphic function, 9
bounded and complete, 172
Hopf conjecture, 36
Hopf function, 78, 331
Hopf vector, 81
immersion, 6
complete, 171
integrability conditions, 39
isoperimetric inequality, 324, 326
isotropic vector, 95
Jacobian
estimate, 169
Jacobian matrix, 6
K¨ ahler form, 307
Kronecker symbol, 8
376 INDEX
Lagrange identity, 136
Laplace operator, 301
Green function, 113, 140, 143
on the sphere, 300
weighted, 65
length of curves, 12
Liouville theorem, 172
Lorentz space, 150
maximum principle, 195
mean curvature, 24, 190
weighted, 220, 226
mean curvature ﬂow, 120
mean curvature graph
curvature estimate, 196
mean curvature system, 26, 64, 225
inhomogeneous, 134
weighted, 65, 235, 237
mean curvature vector, 24, 92
parallel, 125
mean curvatures system
weighted, 220, 227
Minding formula, 76
minimal graph, 6, 173
µstable, 298, 299
area estimate, 316, 320
Bernstein type theorem, 172, 181,
207
boundary gradient estimate, 182
curvature estimate, 170
global gradient estimate, 184
minimal surface, 7, 24, 26, 72, 161,
211, 224, 251, 306
µstable, 288
algebraic, 213
convex hull property, 167
curvature estimate, 173
embedding in subspace, 166
isoperimetric inequality, 326
stable, 286, 289
weighted, 65, 72, 79, 82
minimal surface system, 240
nonparametric, 166, 231, 233
weighted, 240
moving frame, 9
Neil parabola, 7
Neumann boundary value problem,
110, 130
nonlinear elliptic system, 135
nonparametric divergence equation, 234
nonparametric functional, 231
normal Coulomb frame, 110, 124, 130,
137, 138, 144
classical, 157
evolution, 125
weak, 151
normal curvature, 228
normal Gauss curvature matrix, 29
normal mean curvature matrix, 29, 48
normal space, 6, 46
normal vector, 7
orthogonal mapping, 10
rotation, 10
orthonormal normal frame, 8
Osserman curvature estimate, 173, 197,
208
parallel in the normal bundle, 34
parallel mean curvature vector, 125
parallel ONF, 108
parallel surface, 50
parameter transformation, 15
parametric functional, 211
ﬁrst variation, 217
second variation, 338
Pascali system, 81
perfect dominance function, 269
plane mapping, 169
plane normal section, 228
Plateau problem, 267
Poincare inequality, 143
Poincare lemma, 112, 132, 154, 307
Poisson equation, 133
Poisson representation formula, 113,
140, 142
principal curvature, 32, 227–229
principal curvature direction, 228
pseudoholomorphic function, 83, 331
quasiminimal surface, 355
INDEX 377
quasilinear elliptic equation, 235
quasilinear elliptic system, 168
Rado theorem, 185
Rayleigh quotient, 301
weighted, 343
Rellich embedding theorem, 153
Ricci curvature tensor, 43
Ricci equations, 40, 45, 46
Ricci scalar curvature, 43
Riemann mapping theorem, 5
Riemann tensor, 43
RiemannChristoffel tensor, 43
RiemannHilbert problem, 114
Riemannian metric, 13
Rodrigues equations, 229
rotation in R
4
, 10
rotation in R
5
, 10
rotation in R
n
, 129
saddle surface, 185
Schauder estimate, 148, 171, 204, 207
Scherk surface, 212
Schwarz lemma, 176
Schwarz theorem, 344
Sobolev embedding theorem, 148, 158
Sobolev inequality, 143, 306, 327
spherical map, 72, 333
asymptical expansion, 332
energy, 74
oscillation estimate, 336, 337
quasiconformality, 333
spherical mapping system, 241, 244
spherical surface, 49
stable, 286
Steiner symmetrization, 327
stereographic projection, 195, 333
Stokes theorem, 307
surface graph, 6
holomorphic, 119
surface with constant mean curvature,
79
surface with prescribed mean curva
ture, 251
tangent Coulomb frame, 151
tangential space, 6
tangential vector, 6, 18
tensor
contravariant, 16
covariant, 16
theorem of Bonnet and Gauss, 76,
313, 318
theorema egregium, 39, 44
threepoint condition, 185
Tomi regularity result, 158
torsion
complexvalued, 114
torsion coefﬁcients, 30
evolution, 123
torsion matrix, 92
torsion of an ONF, 73, 107
total torsion of an ONF, 75, 107, 129,
313
umbilical point, 79, 229
unit disc, 5
vector ﬁeld, 18
vector product, 88
weight matrix, 59, 61, 220, 235, 331
weighted conformal parameters, 64
weighted fundamental forms
linear dependence, 71
weighted mean curvature, 220, 226
weighted mean curvature system, 220,
227, 235, 237
curvature estimate, 255
weighted metric, 60, 222
weighted minimal surface, 65, 72, 79,
82
weighted minimal surface system, 240
weigthed mean curvature system, 65
Weingarten equations, 30, 155, 158,
275, 280, 299
weighted conformal form, 83
weighted form, 64, 242
Weingarten form, 32
Wente L
∞
estimate, 148
Wente result, 136, 157
Weyl lemma, 302
378 INDEX
Wirtinger symbols, 78
Wulff surface, 214
This action might not be possible to undo. Are you sure you want to continue?