PDE and hypersurfaces

with prescribed
mean curvature

Yunelsy N. Alvarez
To my children, hoping that in future scientific events
they will remain embraced with the motherly warmth.
 Contents

Preface ix

1 Introduction 1

2 Analysis of the mean curvature equation 5

3 The existence program 15

4 Fulfillment of the existence program’s requirements 23

5 Sharpness of the Serrin condition 43

6 Prescribed mean curvature equations in manifolds 51

Bibliography 63

Index 66

vii
 Preface

This book is based on the mini-course PDE and hypersurfaces with prescribed
mean curvature, held at Federal University of São Carlos on the Workshop on Sub-
manifold Theory and Geometric Analysis, August 05 – 09, 2019.
The aim of these lectures notes is to introduce to the geometers useful tools
from the theory of Partial Differential Equations which are used to obtain hy-
persurfaces with prescribed mean curvature. In fact, the expression of the mean
curvature of a graph leads to a quasilinear elliptic partial differential equation of
second order. Hence, compact graphs with prescribed mean curvature and fixed
boundary can be obtained by solving the Dirichlet problem for the mean curvature
equation.
In order to solve the Dirichlet problem the Leray-Schauder fixed point theorem
is used in these notes. However, the beautiful Continuity Methods is another tool
that can be used for the same purpose. In any case, the application of the method
strongly depends on the existence of a priori estimates for the solutions of some
“related problems”.
These notes are organized as follows. Chapter 1 is a brief introduction to the
subject. In Chapter 2 is obtained the non-divergence form of the mean curvature of

ix
x Preface

a graph and some analytical properties about the minimal operator are highlighted.
The existence of solutions is treated in Chapter 3. The estimates needed to fulfill
the existence program are stated in Chapter 4. The sharpness of the existence
results is proved in Chapter 5. With the intention of having an understandable
text, some classical results from the theory of Partial Differential Equations needed
were established in the placed where are used. Finally, Chapter 6 contains some
recent results about this subject in more general ambient spaces.
At the end of the text, the reader should be able to understand that the existence
of graphs with prescribed mean curvature over bounded domains of the Euclidean
space is closely related with the geometry of the domain. Nevertheless, we want
to point out that these lectures notes definitely do not represent the whole subject.
For instance, some regularity issues are left to the reader as a motivation for further
studies.
Yunelsy N. Alvarez
 Acknowledgments

I would like to thank the organizing and scientific committees of the Work-
shop on Submanifold Theory and Geometric Analysis, the opportunity to teach
the mini-course at the event.
I also thank the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)
for the financial support during my stay at the event and the effort to publish these
notes (process 19/11125-5). During the elaboration of this book the author was
supported by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior -
Brasil (CAPES) - Finance Code 001.
Special thanks to professors Marcos M. Alexandrino, Ricardo Sa Earp and Bar-
bara Nelli for the encouragement.

xi
 1

Introduction

The Dirichlet problem for the prescribed mean curvature equation consists on
find a function satisfying the prescribed mean curvature equation in a bounded do-
main of the n−dimensional Euclidean ambient space and that continuously takes
on given boundary values. Precisely, given a smooth bounded domain Ω ⊂ Rn
(n > 2) and a function ϕ defined over the boundary, we ask if for a prescribed
smooth function H there exists some u satisfying
  

 ∇u

div  q
  = nH(x) in Ω,
1 + k∇uk 2
(P )




 u = ϕ in ∂Ω.

If this is the case, then the graph of u is an hypersurface in Rn+1 of mean curvature
H(x) at each point (x, u(x)).
This problem traces its roots to the Plateau’s problem, first posed by Lagrange
[18] in 1760. Lagrange wanted to find the surface with the least area among all
the surfaces having the same boundary. This question arose as an example of the
Calculus of Variations that he was developing. In order to find a minimum of the
area functional, Lagrange derived the Euler-Lagrange equation for the solutions

1
 2 Chapter 1. Introduction

of this problem. That is, if an area minimizing surface in the three dimensional
Euclidean space is a graph of a smooth function u over a bounded domain, then u
necessarily satisfies
   
1 + u2y uxx − 2ux uy uxy 1 + u2x uyy = 0 (1.1)

It was Meusnier [19], in 1776, who gave a geometrical interpretation for equa-
tion (1.1). He realized that the connection between this partial differential equation
and Geometry is given by the concept of mean curvature, formally introduced on
his work. Indeed, the mean curvature of the graph of any solution of this equation
must vanish. All those surfaces are called minimal surfaces even though a surface
ltima having vanish mean curvature is not necessarily (globally) area-minimizing.
To find examples of functions satisfying the minimal surface equation (1.1) is
not an easy task. As a matter of fact, Lagrange could only give the constants func-
tions as examples. It was Meusnier [19] who gave another example of such a func-
tion that locally represents the helicoid, in 1776. He also proved that the catenoid,
discovered by Euler in 1741, can be seen locally as the graph of a function satisfy-
ing (1.1). In 1834, almost seventy years later of the discovering of the catenoid and
the helicoid, H. F. Scherk [22] gave new examples of minimal surfaces, the most
famous one is the Sherk’s surface.
At this point, the interest was about the geometry of the domains over which
minimal graphs can actually exist, rather than find explicit expressions of the solu-
tions of (1.1). In 1910, Bernstein [8] showed that the Dirichlet problem for equation
(1.1) has solution in disks of R2 for continuous boundary data. But Bernstein also
realized that the disks could be replaced by convex domains (see [8, p. 236]). Proofs
of this fact were given independently in 1930 by Douglas [12] and Radó [21, p. 795]
PDE and hypersurfaces with prescribed mean curvature 3

who also ensured the uniqueness of the solution. In 1965, Finn [13] made an im-
portant contribution when he proved that the Dirichlet problem for equation (1.1)
may not be solvable if the domain is non-convex. That is, this convexity conditions
is sharp in order to obtain minimal graphs for arbitrary continuous boundary val-
ues. Using the results from Douglas and Radó, and following a suggestion from
Osserman, Finn stated the following sharp theorem:

Theorem A (Douglas-Radó-Finn [13, T. 4a p. 146]). The Dirichlet problem for

equation (1.1) has solution in Ω for arbitrary continuous boundary data if, and only
if, Ω is convex.

In 1966, Jenkins and Serrin [17] generalized Theorem A to higher dimensions:

Theorem B (Jenkins-Serrin [17, T. 1 p. 171]). Let Ω be a bounded domain in Rn

whose boundary is of class C 2 . Then the Dirichlet problem for the minimal surface
equation (H = 0) in Ω is uniquely solvable for arbitrary C 2 boundary values if, and
only if, the mean curvature of ∂Ω is non-negative.

It can be observed that, although the convexity of Ω is not the appropriated

generalization of the two-dimensional case (as it was thought), it is again a ge-
ometric property. Let’s recall that a domain whose boundary has non-negative
mean curvature is called a mean convex domain. Throughout this text, H∂Ω will
denote the mean curvature of ∂Ω.
Mathematicians were also questioning about the existence of graphs whose
mean curvature not necessarily vanish. It was Serrin [23] who gave a complete
answer:
4 Chapter 1. Introduction

Theorem C (Serrin [23, p. 416]). Let Ω be a bounded domain in n-dimensional

Euclidean space whose boundary is of class C 2 . Then for every constant H the Dirich-
let problem (P ) has a unique solution for arbitrary C 2 boundary data if, and only
if,
(n − 1)H∂Ω (y) > nH ∀ y ∈ ∂Ω. (1.2)

This work of Serrin was actually focused on the study of a more general class
of Dirichlet problems on which problem (P ) is contained. In fact, Theorem C is a
direct conclusion of the following result:

Theorem D (Serrin [23, T. p. 484]). Let Ω be a bounded domain in n-dimensional

Euclidean space, whose boundary is of class C 2 . Let H ∈ C 1 (Ω) and suppose that

n
k∇H(x)k 6 (H(x))2 ∀ x ∈ Ω. (1.3)
n−1

Then problem (P ) is uniquely solvable for arbitrarily given C 2 boundary values if,
and only if,
(n − 1)H∂Ω (y) > n |H(y)| ∀ y ∈ ∂Ω. (1.4)

Our focus in these notes is to proof Theorem D under higher regularity as-
sumptions (see the combination of Theorem 4.7 with Theorem 5.2).
 2

Analysis of the mean

curvature equation

2.1. The mean curvature equation

Let Σ be an oriented hypersurface in Rn+1 . Let us fix a unit normal vector field
N along Σ. Let us recall that

N : Σ −→ Sn
p 7−−→ N (p)
is called the Gauss map of Σ. Besides,

dNp : Tp Σ −→ TN (p) Sn = Tp Σ

is a self-adjoint operator. Thus, there exists an orthonormal basis in Tp Σ such that

the matrix of dNp relative to this basis is a diagonal matrix with entries in the real
numbers. The principal curvatures of Σ at a point p ∈ Σ are the eigenvalues of
−dNp , called the Weingarten map, and the mean curvature H at p is the average of
this principal curvatures at p, that is,
X
n
nH = λi = tr (−dNp ) . (2.1)
i=1

5
6 Chapter 2. Analysis of the mean curvature equation

On the other hand, let Ñ be an extension of N to an open neighbourhood

U ⊂ Rn+1 containing Σ. Obviously, Ñ |Σ = N . As Ñ = 1, we have
D E 1 ∂ D E
dÑp (Ñ ), Ñ (p) = Ñ , Ñ = 0.
2 ∂ Ñ
This implies that dÑp (v) ∈ Tp Σ for all v ∈ Rn+1 , and so dNp = dÑp |Tp Σ . Conse-
quently,
   
div −Ñ = tr −dÑp = tr (−dNp ) = nH. (2.2)

In the case of our interest, where Σ is a graph over a domain Ω ⊂ M of a

function u ∈ C 2 (Ω), the unit normal field along Σ is given by
−∇u(x) + en+1
N= ,
W
where (and from now on)
q
W = 1 + k∇u(x)k2 .

Using (2.2) we obtain,

 
∇u
nH = div  q . (2.3)
1 + k∇uk 2

Making some computations we finally get the mean curvature equation in non-
divergence form1 , !
1 X n
∂i u∂j u
δij − ∂ij u = nH. (2.4)
W i,j=1 W2
1
Equation (2.4) can be derived using the formula

X
n−1
nH = tr (dNp ) = g ij bij ,
i,j=1

where g is the induced metric on S and bij are the coefficients of the second fundamental form.
PDE and hypersurfaces with prescribed mean curvature 7

2.2. Ellipticity of the minimal operator

Note that equation (2.4) is equivalent to

n 
X 
Mu := W 2 δij − ∂i u∂j u ∂ij u = nH(x)W 3 . (2.5)
i,j=1

We call M of minimal operator. From now on will also be considered the equation

Qu := Mu − nH(x)W 3 = 0. (2.6)

The operator M is a quasilinear operator of second order. In fact the coeffi-

cients of the second order derivatives are given by
 
aij (p) = (W (p))2 δij − pi pj , (2.7)
q
where p stands for ∇u and W (p) = 1 + kpk2 . The aim of this section is to prove
that the minimal operator M is strictly elliptic. That is, A(p) = (aij (p)) is positive
definite for each fixed p, and the infimum of the smallest eigenvalue λ(p) of A(p)
is positive2 . This is an important property which is required in many of the results
from the theory of partial differential equations used in this text.
In order to prove that A(p) is positive definite, it is just to find the eigenvalues
of the quadratic form
n 
X   
Qp (q) = hA(p)q, qi = (W (p))2 δij − pi pj qi qj = 1 + kpk2 kqk2 − hp, qi2 .
i,j=1

If q = p it follows
 
hA(p)p, pi = 1 + kpk2 kpk2 − kpk4 = kpk2 ,
2
For a precise definition of ellipticity and strict ellipticity see [14, p. 259]. Observe that in the
case of the minimal operator M, the matrix A does not depend on x.
8 Chapter 2. Analysis of the mean curvature equation

and for q⊥p

 
hA(p)q, qi = 1 + kpk2 kqk2 .

Then, the smallest eigenvalues of A(p) is λ(p) = λ = 1 with associated eigenspace

span{p}. Consequently, M and Q are strictly elliptic operators.
Observe also that Λ(p) = 1 + kpk2 is the largest (and there is no other) eigen-
value of A(p) whose associated eigenspace is p⊥ . Since Λ(p) → ∞ as kpk → ∞,
M and Q are non-uniformly elliptic operators. This is the reason why the geom-
etry of the domain is important for the solvability of problem (P ) (see [14, §12.4 p.
309, p. 345]).

2.3. Maximum Principles

The maximum principles are important tools used in the study of second order
elliptic equations. The following theorem is restricted to the particular case of the
operator Q defined in (2.6).

Theorem 2.1 (Comparison principle [14, Th. 10.1 p. 263]). Let Ω ∈ Rn be a

bounded domain and u, v ∈ C 2 (Ω) ∩ C 0 (Ω) satisfying


Qu > Qv in Ω,

 u 6 v in ∂Ω.
Then u 6 v in Ω.

Proof. Let w = u − v. Recalling the expression for the coefficients aij given in (2.7)
it follows
X X X
Qu − Qv = aij (∇u)∂ij w + aij (∇u)∂ij v − aij (∇v)∂ij v.
ij ij ij
PDE and hypersurfaces with prescribed mean curvature 9

For each x ∈ Ω let us define

X
fx (p) = aij (p)Dij v(x) − nH(x)(W (p))3 .
ij

Applying the mean value theorem to the function

ϕx (t) = f ((1 − t)∇u(x) + t∇v(x)),

we obtain

X X
Qu − Qv = aij (∇u(x))∂ij w + bi (x)∂i w,
ij i

where

X ∂akl
bi (x) = (x, (1 − t(x))∇u(x) + t(x)∇v(x))Dij v(x)
kl ∂pi
∂ (W 3 )
− nH(x) (x, (1 − t(x))∇u(x) + t(x)∇v(x)).
∂pi

So, Lw := Qu − Qv > 0. Since also w 6 0 in ∂Ω, then w 6 0 in Ω as a direct

consequence of the following theorem for linear operators.
Theorem (Week maximum principle [14, Th. 3.1 p. 32]). Let Ω ⊂ Rn be a
bounded domain. Assume that u ∈ C 2 (Ω) ∩ C 0 (Ω) satisfies
X X
Lu = aij (x)∂ij u + bi (x)∂i u > 0 in Ω,
ij i

where L is elliptic and the coefficients aij and bi are locally bounded. Then

sup u = sup u.
Ω ∂Ω

Indeed, it can easily be verified that L satisfies the hypothesis of this theorem. 
10 Chapter 2. Analysis of the mean curvature equation

The following proposition will be an important tool in Chapter 5. This variant

of the comparison principle traces its roots back to the work of Finn [13, Lemma
p. 139] about the minimal surface equation in domains of the plane. His lemma
was extended by Jenkins-Serrin [17, Prop. III p. 182] for the minimal hypersurface
equation in Rn , and subsequently by Serrin [23, Th. 1 p. 459] for more general
quasilinear elliptic operators (see also [14, Th. 14.10 p. 347]). We just refer here to
the operator Q defined in (2.6).

Proposition 2.2 ([16, L. 3.4.3 p. 109]). Let Ω ∈ M be a bounded domain and Γ a

relative open portion of ∂Ω of class C 1 . For H ∈ C 0 (Ω), let u ∈ C 2 (Ω) ∩ C 1 (Ω ∪
Γ) ∩ C 0 (Ω) and v ∈ C 2 (Ω) ∩ C 0 (Ω) satisfying






Qu > Qv in Ω,

 u6v in ∂Ω \ Γ,




 ∂v = −∞ in Γ,
∂N

where N is the inner unit normal to Γ. Then u 6 v in Γ. Consequently, u 6 v in Ω.

Proof. Suppose by contradiction that m = max(u − v) > 0. Hence, u 6 v + m in

Γ
Γ. Then u 6 v + m in ∂Ω since u 6 v in ∂Ω \ Γ by hypotheses. Also,

Q(v + m) = Qv 6 Qu.

As a consequence of Theorem 2.1, u 6 v + m in Ω.

Let now y0 ∈ Γ be such that m = u(y0 ) − v(y0 ). Then, for t > 0 near 0 one
has

u(y0 + tNy0 ) − u(y0 ) 6 (v (y0 + tNy0 ) + m) − (v(y0 ) + m) = v(y0 + tNy0 ) − v(y0 ).

PDE and hypersurfaces with prescribed mean curvature 11

Dividing the expression by t and passing to the limit as t goes to zero, it follows
that
∂u ∂v
6 = −∞,
∂N ∂N
which is a contradiction since u ∈ C 1 (Γ). Hence, u 6 v in Γ. Using again Theorem
2.1 one has u 6 v in Ω. 

2.4. Transformation formulas

The goal in this section is to derive some transformation formulas that will be
used in the next chapters. Note first that the operator M defined in (2.5) can be
written as
D E
Mu = W 2 ∆u(x) − ∇2 u · ∇u, ∇u , (2.8)

where ∇2 denotes the Hessian of a function (the matrix of second derivatives).

Let Ω ∈ Rn be a bounded domain and ϕ ∈ C 2 (Ω). Let I be an interval and
ψ ∈ C 2 (I). Let us now define

w = ψ ◦ % + ϕ,

where % is a distance function3 . Then, ∂i w = ψ 0 ∂i % + ∂i ϕ and ∇w = ψ 0 ∇% + ∇ϕ.

For this reason

∂ij w = ψ 00 ∂i %∂j % + ψ 0 ∂ij % + ∂ij ϕ, (2.9)

and
∇2 w = ψ 00 ∇% ⊗ ∇% + ψ 0 ∇2 % + ∇2 ϕ.
3
If % : Ω → R is a distance function, then ∇% ≡ 1 on Ω.
12 Chapter 2. Analysis of the mean curvature equation

Since % is a distance function, it follows

 
∆w(x) = tr ∇2 w = ψ 00 k∇%k2 + ψ 0 ∆% + ∆ϕ = ψ 00 + ψ 0 ∆% + ∆ϕ. (2.10)

Besides4 ,
D E
∇2 w · ∇w, ∇w
D E D E
=ψ 00 h∇%, ∇wi2 + ψ 0 ∇2 % · ∇w, ∇w + ∇2 ϕ · ∇w, ∇w
D E
=ψ 00 h∇%, ψ 0 ∇% + ∇ϕi + ψ 0 ∇2 % · (ψ 0 ∇% + ∇ϕ) , ψ 0 ∇% + ∇ϕ
2

D E
+ ∇2 ϕ · (ψ 0 ∇% + ∇ϕ) , ψ 0 ∇% + ∇ϕ

=ψ 00 h∇%, ψ 0 ∇% + ∇ϕi
2

 D E D E D E
+ ψ 0 ψ 02 ∇2 % · ∇%, ∇% + 2ψ 0 ∇2 % · ∇%, ∇ϕ + ∇2 % · ∇ϕ, ∇ϕ
D E
+ ∇2 ϕ · (ψ 0 ∇% + ∇ϕ) , ψ 0 ∇% + ∇ϕ .

Recalling again that % is a distance function one has

D E X 1  
∇2 % · ∇%, ej = ∂ij %∂i % = h∂j ∇%, ∇%i = ∂j k∇%k2 = 0 ∀ 1 6 j 6 n.
i 2

As a consequence,

h∇2 w · ∇w, ∇wi = ψ 00 h∇%, ψ 0 ∇% + ∇ϕi2 + ψ 0 h∇2 % · ∇ϕ, ∇ϕi

(2.11)
+h∇2 ϕ · (ψ 0 ∇% + ∇ϕ) , ψ 0 ∇% + ∇ϕi.

From (2.8), (2.10) and (2.11) we derive

 D E
Mw =W 2 (ψ 00 + ψ 0 ∆% + ∆ϕ) − ψ 00 h∇%, ψ 0 ∇% + ∇ϕi + ψ 0 ∇2 % · ∇ϕ, ∇ϕ
2

D E
+ ∇2 ϕ · (ψ 0 ∇% + ∇ϕ) , ψ 0 ∇% + ∇ϕ ,
4
Let’s recall that (u ⊗ v) w = hv, wiu for any vectors u, v and w.
PDE and hypersurfaces with prescribed mean curvature 13

q
where W = W (∇w) = 1 + kψ 0 ∇% + ∇ϕk2 . Accordingly,
D E
Mw =ψ 0 W 2 ∆% − ψ 0 ∇2 % · ∇ϕ, ∇ϕ

ψ 00 W 2 − ψ 00 h∇%, ψ 0 ∇% + ∇ϕi (2.12)

2

D E
∆ϕW 2 − ∇2 ϕ · (ψ 0 ∇% + ∇ϕ) , ψ 0 ∇% + ∇ϕ .
√
Furthermore, if ϕ is constant, then W = 1 + ψ 02 and

Mw = ψ 0 (1 + ψ 02 )∆% + ψ 00 . (2.13)
 3

The existence program

In this chapter we carry out with the existence program for the Dirichlet prob-
lem (P ). It will be seen how the elliptic theory assures that the solvability of
problem (P ) strongly depends on C 1 a priori estimates for the family of related
problems


 Mu = τ nH(x)W 3 in Ω,
(Pτ )

 u = τ ϕ in ∂Ω,

not depending on τ or u1 . Actually, the following theorem (which is also valid for
more general elliptic operators) holds.

Theorem 3.1 ([14, T. 11.4 p. 281]). Let Ω ⊂ Rn be a bounded domain with ∂Ω

of class C 2,α , ϕ ∈ C 2,α (Ω) and H ∈ C α (Ω) for some α ∈ (0, 1). Assume there
exists M > 0 independent of u and τ such that any solution u of the related problems
(Pτ ) satisfies kukC 1 (Ω) < M . Then the Dirichlet problem (P ) has a unique solution
in C 2,α (Ω).

Proof. Let β ∈ [0, 1] to be made precise later. For each v ∈ C 1,β (Ω) we consider
1
Note that problem (P ) is equivalent to problem (Pτ =1 ).

15
16 Chapter 3. The existence program

the linear operator

X
Lv u := avij (x)∂ij u, (3.1)
ij

where
q
avij (x) = (Wv (x))2 δij − ∂i v(x)∂j v(x), Wv (x) = W (∇v(x)) = 1 + ∇v(x) .

We now want to study the following linear problem



 Lv u = nH(x)(Wv (x))3 in Ω,
(LPv )

 u = ϕ in ∂Ω.

The following theorem for linear operators guaranties the existence of a unique
solution uv ∈ C 2,αβ (Ω) ⊂ C 1,β (Ω) of problem (LPv ).

Theorem (Schauder [14, T. 6.14 p. 107]). Let Ω ⊂ Rn be a C 2,α bounded

X
domain and ϕ ∈ C 2,α (Ω) for some α ∈ (0, 1). Let Lu = aij (x)∂ij u +
ij
X
bi (x)∂i u + c(x)u be a strictly elliptic operator in Ω with coefficients in C α (Ω)
i
and c 6 0. Let also f be a function in C α (Ω). Then the problem


 Lu = f in Ω,

 u = ϕ in ∂Ω,

has a unique solution in C 2,α (Ω).

In fact, notice that ∂Ω is of class C 2,αβ and ϕ ∈ C 2,αβ (Ω). Also, nHWv3 and
avij belong to C αβ (Ω) since v ∈ C 1,β (Ω), H ∈ C α (Ω) and the coefficients aij are
regular. Besides, Lv is strictly elliptic because it was proved in section 2.2 that
the infimum of the smallest eigenvalue of the matrix (aij (∇v(x))) = (avij (x)) is 1.
PDE and hypersurfaces with prescribed mean curvature 17

And so, the operator

T : C 1,β (Ω) −→ C 1,β (Ω)

v 7−−→ uv

is well defined.
In addition, the solvability of problem (P ) is equivalent to the existence of a
fixed point of T . Indeed, since T v is the only solution of (LPv ), that is,
 n  
 X

 (Wv (x))2 δij − ∂i v∂j v ∂ij (T v) = nH(x) (Wv (x))3 in Ω,

i,j=1



 T v = ϕ in ∂Ω,

then the existence of a function u ∈ C 1,β (Ω) satisfying T u = u is exactly problem

(P ). So, is evident the need for a theorem to guarantee the existence of a fixed
point of T . The following is enough for our purpose.

Theorem (Leray-Schauder [14, T. 11.3 p. 280]). Let B be a Banach space and

T : B → B a completely continuous operator2 . If there exists M > 0 such that

kxkB < M

for each τ ∈ [0, 1] and x ∈ B satisfying x = τ T x, then T has a fixed point.

In order to use the Leray-Schauder fixed point theorem, we observe first that
the family of solutions of the related problems (Pτ ) is not empty once u = 0 obvi-
ously satisfies (P0 ). Also, B = C 1,β (Ω) is a Banach space.

2
A continuous mapping between two Banach spaces is called compact or completely continuous
if the images of bounded sets are precompact.
18 Chapter 3. The existence program

Since (LPv ) holds for each v ∈ C 1,β (Ω), then, for τ ∈ [0, 1], it follows
 n  
 X

 (Wv (x))2 δij − ∂i v∂j v ∂ij (τ T v) = τ nH(x) (Wv (x))3 in Ω,

i,j=1 (3.2)



 τ T v = τ ϕ in ∂Ω.

Hence, equation τ T u = u is equivalent to problem (Pτ ), and the family of solutions

of these problems is bounded in C 1 (Ω) from the hypotheses.
The following global Hölder estimate for quasilinear operators of second order
guarantees that we actually have an a priori bound in C 1,β (Ω) for some β ∈ (0, 1).

Theorem (Ladyzhenskaya-Ural’tseva [14, T. 13.7 p. 331]). Let Ω ⊂ Rn be a

bounded domain with ∂Ω of class C 2 and ϕ ∈ C 2 (Ω). Assume that u ∈ C 2 (Ω)
satisfies

 X

 Qu = aij (x, u, ∇u)∂ij u + b(x, u, ∇u) = 0 in Ω,


ij





u = ϕ in ∂Ω,

where Q is an elliptic operator such that aij ∈ C 1 (Ω × R × Rn ) and b ∈ C 0 (Ω ×

R × Rn ). Then
[ui ]α,Ω < C ∀i = 1, . . . , n,
   
where C = C n, kukC 1 (Ω) , kϕkC 2 (Ω) , Ω, λ and α = α n, kukC 1 (Ω) , Ω, λ > 0.

Indeed, the operator Q defined in (2.6) obviously satisfies the hypothesis of the
previous theorem. Observe that this estimate does not depend on u or τ because
we already have an a priori estimate in C 1 (Ω) for the family of solutions of (Pτ ).
This is the constant β that should be fixed at the beginning.
PDE and hypersurfaces with prescribed mean curvature 19

It still remains to prove that T is continuous and compact. We recall first the
following theorem for linear elliptic operators.

Theorem (Global Schauder estimate [14, Th. 6.6 p. 98]). Let Ω ⊂ Rn be

a bounded domain with ∂Ω of class C 2,α and ϕ ∈ C 2,α (Ω) for some α ∈ (0, 1).
Assume that u ∈ C 2,α (Ω) satisfies

 X X

 Lu = aij (x)∂ij u + bi (x)∂i u + c(x)u = f (x) in Ω,


ij i





u = ϕ in ∂Ω,

where f and the coefficients of the strictly elliptic operator L belong to C α (Ω). Then
 
kukC 2,α (Ω) 6 C kukC 0 (Ω) + kf kC α (Ω) + kϕkC 2,α (Ω) (3.3)
n o
for C = C(Ω, n, α, λ, K) > 0, K > max kaij kC α (Ω) , kbi kC α (Ω) , kckC α (Ω) .

In view of the previous theorem

 

kT vkC 2,αβ (Ω) 6 C kT vkC 0 (Ω) + nHWv3 + kϕkC 2,αβ (Ω)
C αβ (Ω)

for every v ∈ C 1,β (Ω). Also,

kT vkC 0 (Ω) 6 sup |ϕ| + C̃ nHWv3
∂Ω C 0 (Ω)

as a direct consequence of the following theorem for linear elliptic operators.

20 Chapter 3. The existence program

Theorem (Height estimate [14, Th. 3.7 p. 36]). Let Ω ⊂ Rn be a bounded

domain. Assume that u ∈ C 2 (Ω) ∩ C 0 (Ω) satisfies
X X
Lu = aij (x)∂ij u + bi (x)∂i u + c(x)u = f (x)
ij i

where f and the coefficients bi of the strictly elliptic operator L are bounded in Ω
and c 6 0. Then
sup |u| 6 sup |u| + C sup |f | . (3.4)
Ω ∂Ω Ω
!
where C = C λ, diam(Ω), max sup |bi | .
i Ω

Observe that the constants C and C̃ are independent of v. Hence, T maps

bounded sets in C 1,β (Ω) into bounded sets in C 2,αβ (Ω) ,→ C 1,β (Ω) ,→ C α (Ω).
 
Let now {T vm } be a sequence in T C 1,β (Ω) . We affirm that there exists
some subsequence that converges in C 1,β (Ω). Certainly, since for every x and y
in Ω we have

|T vm (x) − T vm (y)| 6 kT vm kC α (Ω) kx − ykα ,

so {T vm } is equicontinuous. This sequence is also uniformly bounded, then there

n o
exists a subsequence T vξ(m) that converges uniformly to a continuous function
w by the Arzela-Ascoli theorem.
Besides, for every 1 6 i 6 n,

|∂i T vm (x) − ∂i T vm (y)| 6 kT vm kC 1,β (Ω) kx − ykβ ,

then {∂i T vm } is also equicontinuous. The Arzela-Ascoli theorem guarantees the

n o
existence of a subsequence ∂i T vφi (m) that converges uniformly to a continuous
PDE and hypersurfaces with prescribed mean curvature 21

function wi since {∂i T vm } is uniformly bounded. This implies that ∂i w exists, is

continuous and wi = ∂i w.
Furthermore, for every 1 6 i, j 6 n,

|∂ij T vm (x) − ∂ij T vm (y)| 6 kT vm kC 2,αβ (Ω) kx − ykαβ .

Because of the C 2,αβ estimate {∂ij T vm } is equicontinuous. Due to the fact that
n o
it is also a bounded sequence, then there exists a subsequence ∂ij T vψij (m) that
converges uniformly to ∂ij w (which necessarily exists and is continuous).
This proves that w ∈ C 2 (Ω) and that the subsequence T vψ(φ(ξ(m))) , with φ =
φ1 ◦ · · · ◦ φn and ψ = ψnn ◦ ψn,n−1 ◦ · · · ◦ ψ11 , converges to w in C 2 (Ω) and also
in C 1,β (Ω).
In order to prove the continuity of T , let {vm } be a sequence converging to
v in C 1,β (Ω). Since {T vm } is precompact in C 2 (Ω), every subsequence has a
n o
convergent subsequence. Let T vξ(m) such a convergent subsequence with limit
w ∈ C 2 (Ω). Then
 n  2     3
 X

 Wvξ(m) δij − ∂i vξ(m) ∂j vξ(m) ∂ij T vξ(m) = nH(x) Wvξ(m) in Ω,

i,j=1



 T vξ(m) = ϕ in ∂Ω.
Passing to the limits and recalling the definition of T one has w = T v. Given that
every subsequence of {T vm } has at least one subsequence which converges to T v,
then {T vm } also converges to T v.
So, it had been proved that T has a fixed point u ∈ C 2,αβ (Ω) ,→ C 1,β (Ω).
To see that u ∈ C 2,α (Ω) recall that u is a solution of the linear problem LPu
(see problem (LPv )) and use the following global regularity theorem for linear
operators.
22 Chapter 3. The existence program

Theorem (Global regularity [14, Th. 6.18 p. 111]). Let Ω ⊂ Rn be a bounded

domain with ∂Ω of class C 2,α and ϕ ∈ C 2,α (Ω) for some α ∈ (0, 1). Suppose that
u ∈ C 2 (Ω) satisfies

 X X

 Lu = aij (x)∂ij u + bi ∂i u + c(x)u = f in Ω,


ij i





u = ϕ in ∂Ω,

where f and the coefficients of the strictly elliptic operator L belong to C α (Ω).
Then u ∈ C 2,α (Ω).
Finally, notice that the solution is unique as a consequence of the comparison
principle, Theorem 2.1. 

Remark 3.2. Applying the same argument to every fixed τ ∈ [0, 1] we ensure the
existence of a solution for each problem (Pτ ).
 4

Ful illment of the existence

program’s requirements

The goal in this chapter is to obtain the a priori estimates for the family of
solutions of the related problems (Pτ ) required by the existence program.
In order to derive the a priori global gradient estimate we will use the tech-
niques introduced by Caffarelli-Nirenberg-Spruck [9, p. 51].

Theorem 4.1 (A priori global gradient estimate [16, Th. 3.2.4 p. 94]). Let
Ω ∈ M be a bounded domain with ∂Ω of class C 2 . For H ∈ C 1 (Ω), let u ∈
C 3 (Ω) ∩ C 1 (Ω) be a solution of (2.5). Then
 
!
4n sup|u|(1+kHk1 )
sup k∇u(x)k 6 3 + sup k∇uk e Ω
. (4.1)
Ω ∂Ω

Proof. Let w(x) = k∇u(x)k eAu(x) where A > 1. Suppose w attains a maximum
at x0 ∈ Ω. If x0 ∈ ∂Ω, then

w(x) 6 w(x0 ) = k∇u(x0 )k eAu(x0 ) .

So,
2A sup|u|
sup k∇u(x)k 6 sup k∇uk e Ω . (4.2)
Ω ∂Ω

23
24 Chapter 4. Ful illment of the existence program’s requirements

Suppose now that x0 ∈ Ω and that ∇u(x0 ) =

6 0. Without loss of generality it can
∇u(x0 )
be assumed that e1 = k∇u(x0 )k
. Then,

∂k u(x0 ) = hek , ∇u(x0 )i = k∇u(x0 )k δk1 . (4.3)

Differentiating (2.5) with respect to x1 we derive

  X
n X
n X
n
∂1 W 2
∆u + W 2
∂1ii u − 2 ∂i u∂1j u∂ij u − ∂i u∂j u∂1ij u
i=1 i,j=1 i,j=1 (4.4)
 
= n∂1 HW 3 + nH∂1 W 3 .

The goal now is to estimate the third derivatives in (4.4). Observe first that the
function w̃(x) = ln w(x) = Au(x) + ln k∇u(x)k also attains a maximum at x0 .
Consequently, for each 0 6 k 6 n,
 
∂k k∇uk2 (x0 )
∂k w̃(x0 ) = A∂k u(x0 ) + = 0, (4.5)
2 k∇u(x0 )k2
and
 
1  −2
 
2
 ∂kk k∇uk2 (x0 )
∂kk w̃(x0 ) = A∂kk u(x0 )+ ∂k k∇uk (x0 )∂k k∇uk (x0 )+ 6 0.
2 2 k∇u(x0 )k2

Since
   −1  −2  
∂k k∇uk−2 = ∂k k∇uk2 = − k∇uk2 ∂k k∇uk2 ,

then    2  
∂k k∇uk2 (x0 ) ∂kk k∇uk2 (x0 )
A∂kk u(x0 ) − + 6 0. (4.6)
2 k∇u(x0 )k4 2 k∇u(x0 )k2
Once !
  X
n X
n
∂k k∇uk 2
= ∂k (∂i u(x)) 2
=2 ∂i u∂ki u, (4.7)
i=1 i=1
PDE and hypersurfaces with prescribed mean curvature 25

it follows from (4.3)

 
∂k k∇uk2 (x0 ) = 2 k∇u(x0 )k ∂1k u(x0 ). (4.8)

Substituting (4.3) and (4.8) in (4.5) one has

2 k∇u(x0 )k ∂1k u(x0 )
A k∇u(x0 )k δk1 + = 0,
2 k∇u(x0 )k2
thus,
∂1k u(x0 ) = −A k∇u(x0 )k2 δk1 . (4.9)

Substituting also (4.9) in (4.8) we obtain

 
∂k k∇uk2 (x0 ) = −2A k∇u(x0 )k3 δk1 . (4.10)

Besides, from (4.7) we get

  n 
X 
∂kk k∇uk2 (x) =2 ∂kki u∂i u + (∂ki u)2 ,
i=1

and from (4.3) we conclude

  X
n
∂kk k∇uk2 (x0 ) = 2 k∇u(x0 )k ∂kk1 u + 2 (∂ki u(x0 ))2 . (4.11)
i=1

Using expressions (4.10) and (4.11) in (4.6) we verify that

X
n
(∂ki u(x0 ))2
∂kk1 u(x0 )
A∂kk u(x0 ) − 2A2 k∇u(x0 )k2 δk1 + + i=1
6 0.
k∇u(x0 )k k∇u(x0 )k2
From (4.9) we have for k = 1
n 
X 2
−A k∇u(x0 )k2 δi1
∂111 u(x0 )
−A k∇u(x0 )k − 2A k∇u(x0 )k +
2 2 2
+ 2 i=1
6 0,
k∇u(x0 )k k∇u(x0 )k2
26 Chapter 4. Ful illment of the existence program’s requirements

then,
∂111 u(x0 ) 6 2A2 k∇u(x0 )k3 . (4.12)

If k > 1, then
X
n
(∂ki u(x0 ))2
∂kk1 u(x0 )
A∂kk u(x0 ) + 6 − i=1 6 0,
k∇u(x0 )k k∇u(x0 )k2
so,
∂kk1 u(x0 ) 6 −A∂kk u(x0 ) k∇u(x0 )k , 1 < k 6 n. (4.13)

On the other hand, since (4.10) holds we deduce

   
∂1 W 2 (x0 ) = ∂1 k∇uk2 (x0 ) = −2A k∇u(x0 )k3 , (4.14)

and
  3  
∂1 W 3 (x0 ) = W0 ∂1 W 2 (x0 ) = −3AW0 k∇u(x0 )k3 . (4.15)
2
Using (4.3), (4.9), (4.12), (4.13), (4.14) and (4.15) in (4.4) it follows

n∂1 H(x0 )W03 − 3nAH0 W0 k∇u(x0 )k3

X X
= − 2A k∇u(x0 )k3 ∂ii u(x0 ) + W02 ∂1ii u(x0 ) + ∂111 u(x0 )
i>1 i>1

X X
6 − 2A k∇u(x0 )k3 ∂ii u(x0 ) − A k∇u(x0 )k W02 ∂ii u(x0 ) + 2A2 k∇u(x0 )k3
i>1 i>1

 X
= − A k∇u(x0 )k 1 + 3 k∇u(x0 )k2 ∂ii u(x0 ) + 2A2 k∇u(x0 )k3 ,
i>1
q
where H0 = H(x0 ) and W0 = 1 + k∇u(x0 )k2 .
PDE and hypersurfaces with prescribed mean curvature 27

In addition, since (4.3) and (4.9) hold, the mean curvature equation (2.5) at x0
takes the form
X
nH0 W03 = W02 ∆u(x0 ) − k∇u(x0 )k2 ∂11 u(x0 ) = W02 ∂ii u(x0 ) − A k∇u(x0 )k2 ,
i>1

so,
X A k∇u(x0 )k2
∂ii u(x0 ) = nH0 W0 + . (4.16)
i>1 W02
Therefore,
!
  A k∇u(x0 )k2
0 6 − A k∇u(x0 )k 1 + 3 k∇u(x0 )k 2
nH0 W0 +
W02
+ 2A2 k∇u(x0 )k3 + 3nAH0 W0 k∇u(x0 )k3 − n∂1 HW03

A2 k∇u(x0 )k3  
= − AnH0 W0 k∇u(x0 )k − n∂1 HW03 + 1 − k∇u(x 0 )k2
.
W02
Then
A2 k∇u(x0 )k3  
k∇u(x 0 )k2
− 1 6 Anh0 W0 k∇u(x0 )k + nh1 W03 ,
W02
where h0 = sup |H| and h1 = sup k∇Hk. We can suppose that k∇u(x0 )k > 1.
Ω Ω
Dividing by A2 W03 , noticing that W02 > W0 > k∇u(x0 )k and choosing
A = 2n (kHk1 + 1), it follows
 
k∇u(x0 )k3 k∇u(x0 )k2 − 1 n 1
 5/2 < kHk1 <
1 + k∇u(x0 )k2 A 2

As a consequence,
k∇u(x0 )k < 3

and
w(x) 6 w(x0 ) = k∇u(x0 )k eAu(x0 ) 6 3eAu(x0 ) .
28 Chapter 4. Ful illment of the existence program’s requirements

Hence,
2A sup|u|
sup k∇u(x)k 6 3e Ω . (4.17)
Ω

Putting together (4.2) and (4.17) we obtain

2A sup|u| 2A sup|u|
sup k∇u(x)k 6 3e Ω + sup k∇uk e Ω ,
Ω ∂Ω

which yields the desired estimate. 

The following theorem due to Serrin [23, §9 p. 434] ensures the boundary
gradient estimate. For the proof is used the classical idea of finding an upper and
a lower barriers for u on ∂Ω to get a control of ∇u along ∂Ω.

Theorem 4.2 (A priori boundary gradient estimate [16, Th. 3.2.2 p. 89]). Let
 
Ω ∈ Rn be a bounded domain with ∂Ω of class C 2 and ϕ ∈ C 2 (Ω). Let H ∈ C 1 Ω
satisfying
(n − 1)H∂Ω (y) > n |H(y)| ∀ y ∈ ∂Ω. (4.18)

If u ∈ C 2 (Ω) ∩ C 1 (Ω) is a solution of (P ), then

3
sup k∇uk 6 kϕk1 + eC (1+kHk1 +kϕk2 )(1+kϕk1 ) (kuk0 +kϕk0 ) (4.19)
∂Ω

for some C = C(n, Ω).

Proof. We set d(x) = dist(x, ∂Ω) for x ∈ Ω. Let τ > 0 be such that d is of class
C 2 over the set of points in Ω for which d(x) 6 τ (see [14, L. 14.16 p. 355], [16, L.
3.1.8 p. 84] and [23, L. 1 p. 420]). Let ψ ∈ C 2 ([0, τ ]) be a non-negative function
satisfying
PDE and hypersurfaces with prescribed mean curvature 29

P1. ψ 0 (t) > 1, P2. ψ 00 (t) 6 0, P3. tψ 0 (t) 6 1.

For a < τ to be fixed later on, we consider the set

Ωa = {x ∈ M ; d(x) < a} .

Let w± = ±ψ ◦ d + ϕ. Firstly, we estimate ±Mw± in Ωa . Using the transfor-

mation formula (2.12) one has
D E
±Mw± =ψ 0 W±2 ∆d − ψ 0 ∇2 d · ∇ϕ, ∇ϕ

+ ψ 00 W±2 − ψ 00 h∇d, ±ψ 0 ∇d + ∇ϕi (4.20)

2

D E
± W±2 ∆ϕ ∓ ∇2 ϕ · (±ψ 0 ∇d + ∇ϕ), ±ψ 0 ∇d + ∇ϕ ,

where
q q
W± = 1 + k∇w± k2 = 1 + k±ψ 0 ∇d + ∇ϕk2 .

Once d is of class C 2 in Ωa and ψ 0 > 1 we have

D E
 
ψ 0  ∇2 d · ∇ϕ, ∇ϕ  6 ψ 02 kdk2 kϕk21 . (4.21)

Considering that ψ 00 < 0 and h∇d, ±ψ 0 ∇d + ∇ϕi2 6 k±ψ 0 ∇d + ∇ϕk2 , then

ψ 00 W±2 − ψ 00 h∇d, ±ψ 0 ∇d + ∇ϕi 6 ψ 00 .

2
(4.22)

Also ϕ is of class C 2 in Ωa by hypothesis, so

 D E

±∆ϕW±2 ∓ ∇2 ϕ · (±ψ 0 ∇d + ∇ϕ), ±ψ 0 ∇d + ∇ϕ 

6 n kϕk2 W±2 + kϕk2 k±ψ 0 ∇d + ∇ϕk

2

6 2n kϕk2 W±2 .
30 Chapter 4. Ful illment of the existence program’s requirements

Notice now that

 
k±ψ 0 ∇d + ∇ϕk = ψ 02 + 2ψ 0 h±∇d, ∇ϕi + k∇ϕk2 6 (1 + kϕk1 )2 ψ 02 ,
2

thus
W±2 6 1 + (1 + kϕk1 )2 ψ 02 6 2 (1 + kϕk1 )2 ψ 02 . (4.23)

As a consequence,
 D E
 
±∆ϕW±2 ∓ ∇2 ϕ · (±ψ 0 ∇d + ∇ϕ), ±ψ 0 ∇d + ∇ϕ 
(4.24)
6 4n kϕk2 (1 + kϕk1 )2 ψ 02 .

Substituting (4.22), (4.21), (4.24) in (4.20) it follows

± Mw± 6 ψ 0 W±2 ∆d + ψ 00 + cψ 02 , (4.25)

where
c = kdk2 kϕk21 + 4n kϕk2 (1 + kϕk1 )2 . (4.26)

From (4.25) we obtain

±Qw± 6 ψ 0 W±2 ∆d + ψ 00 + cψ 02 + n |H(x)| W±3 . (4.27)

Let now y = y(x) be the point of ∂Ω nearest to x. Then x belongs to the

segment {y + tNy ; 0 6 t 6 a}, where N is the inner unit normal to ∂Ω. Let
us denote by Γt the hypersurface parallel to ∂Ω at a distance t contained in Ωa .
We define H(t) := HΓt (y + tNy ) where HΓt is the mean curvature of Γt with
respect to the normal that coincides with ∇d(x) = x−y
kx−yk
at x. Remember that
∆d(x) = −(n − 1)HΓd(x) (x) (see [14, §14.6 p 354], [16, §3.1 p. 80] and [23, §3 p.
420]). In addition,
HΓt (y + tNy ) > H∂Ω (y) (4.28)
PDE and hypersurfaces with prescribed mean curvature 31

since H0 (t) > (H(t))2 > 0 in [0, a) (see [23, p. 485]). Using also the Serrin
condition (4.18) we get

∆d(x) 6 ∆d(y) 6 −n |H(y)| ∀ x ∈ Ωa (4.29)

Substituting (4.29) in (4.27) we obtain

±Qw± 6nψ 0 W±2 (|H(x)| − |H(y)|) + n |H(x)| W±2 W± − ψ 0 + ψ 00 + cψ 02 . (4.30)

Besides,
|H(x)| − |H(y)| 6 h1 (1 + kϕk1 )d(x),

where h1 = sup k∇Hk . Recalling also (4.23) one gets

Ω

nψ 0 W±2 (|H(x)| − |H(y)|) 6 2nh1 (1 + kϕk1 )3 d(x)(ψ 0 (d(x)))3 .

Using the assumption P3 it follows

nψ 0 W±2 (|H(x)| − |H(y)|) 6 2nh1 (1 + kϕk1 )3 ψ 02 . (4.31)

On the other hand,

W± − ψ 0 6 1 + k±ψ 0 ∇d + ∇ϕk − ψ 0 6 1 + kϕk1 . (4.32)

From (4.23) and (4.32) we obtain

n |H(x)| (W± − ψ 0 ) W±2 6 2nh0 (1 + kϕk1 )3 ψ 02 , (4.33)

where h0 = sup |H| .

Ω
Using (4.31) and (4.33) in (4.30) we get
 
±Qw± 6 c + 2n kHk1 (1 + kϕk1 )3 ψ 02 + ψ 00 .
32 Chapter 4. Ful illment of the existence program’s requirements

Recalling the expression for c given in (4.26) and making some algebraic com-
putations we infer that

c + 2n kHk1 (1 + kϕk1 )3 < C (1 + kϕk2 + kHk1 ) (1 + kϕk1 )3 ,

where
C = 4n (1 + kdk2 + 1/τ ) . (4.34)

Choosing
ν = C (1 + kHk1 + kϕk2 ) (1 + kϕk1 )3 (4.35)

we define ψ by
1
ψ(t) = log(1 + kt).
ν
Accordingly,
k
ψ 0 (t) = (4.36)
ν(1 + kt)
and
k2
ψ 00 (t) = − . (4.37)
ν(1 + kt)2
That’s why
±Qw± < νψ 02 + ψ 00 = 0, in Ωa .

Also,
kt 1
tψ 0 (t) = 6 < 1,
ν(1 + kt) ν
which is property P3. From (4.37) we see that property P2 is also satisfied. Another
consequence of (4.37) is that ψ 0 (t) > ψ 0 (a) for all t ∈ [0, a], thus property P1 is
ensured provided that
k
ψ 0 (a) = = 1. (4.38)
ν(1 + ka)
PDE and hypersurfaces with prescribed mean curvature 33

Furthermore, choosing

1
ψ(a) = log(1 + ka) = kuk0 + kϕk0 , (4.39)
ν

we have

±w± (x) = ψ(a) ± ϕ(x) = kuk0 + kϕk0 ± ϕ(x) > ±u(x) ∀ x ∈ ∂Ωa \ ∂Ω.

By combining (4.38) and (4.39) we see that

k = νeν(kuk0 +kϕk0 ) (4.40)

and, therefore,
eν(kuk0 +kϕk0 ) − 1
a= .
νeν(kuk0 +kϕk0 )
Note also that a < 1
ν
< τ as it was required.
Finally, if x ∈ ∂Ω, then w± (x) = ±ψ(0) + ϕ(x) = u(x). By the maximum
principle we can conclude that w− 6 u 6 w+ in Ωa , consequently

−ψ ◦ d 6 u − ϕ 6 ψ ◦ d in Ωa .

Observe that
−ψ ◦ d = u − ϕ = ψ ◦ d = 0 in ∂Ω,

thus, for y ∈ ∂Ω and 0 6 t 6 a, we have

−ψ(t) + ψ(0) 6 (u − ϕ)(y + tNy ) − (u − ϕ)(y) 6 ψ(t) − ψ(0).

Dividing by t > 0 and passing to the limit as t goes to zero we infer that

−ψ 0 (0) 6 h∇u(y) − ∇ϕ(y), Ny i 6 ψ 0 (0),

34 Chapter 4. Ful illment of the existence program’s requirements

so,
|h∇u(y) − ∇ϕ(y), Ny i| 6 ψ 0 (0)

which leads to
|h∇u(y), Ny i| 6 |h∇ϕ(y), Ny i| + ψ 0 (0). (4.41)

Since u ≡ ϕ in ∂Ω

∇u(y) = (∇ϕ(y))T + h∇u(y), Ny iNy .

Using (4.41) we derive

2

k∇u(y)k2 = (∇ϕ(y))T + h∇u(y), N i2
2
6 (∇ϕ(y))T + (|h∇ϕ(y), N i| + ψ 0 (0))
2

2

= (∇ϕ(y))T + h∇ϕ(y), N i2 + 2 |h∇ϕ(y), N i| ψ 0 (0) + (ψ 0 (0))2

6 k∇ϕ(y)k2 + 2 k∇ϕ(y)k ψ 0 (0) + (ψ 0 (0))2

= (k∇ϕ(y)k + ψ 0 (0)) .
2

which yields the desired estimate. 

The combination of Theorems 4.1 and 4.2 with Theorem 3.1 implies that it just
remains to find an a priori height estimate. In fact, the following theorem holds.

Theorem 4.3 ([14, Th. 16.9 p. 407]). Let Ω ⊂ Rn be a bounded domain with ∂Ω
of class C 2,α for some α ∈ (0, 1), ϕ ∈ C 2,α (Ω) and H ∈ C 1,α (Ω). Assume that the
family of solutions of the related problems (Pτ ) is uniformly bounded. If

(n − 1)H∂Ω (y) > n |H (y)| ∀ y ∈ ∂Ω, (4.42)

then the Dirichlet problem (P ) has a unique solution in C 2,α (Ω).

PDE and hypersurfaces with prescribed mean curvature 35

Proof. Let M an a priori bound for the family of solutions of the related problems
(Pτ ). For τ ∈ [0, 1] fixed, let u a solution of (Pτ ).
Note now that u ∈ C 3 (Ω) as a consequence of applying twice the following
interior regularity theorem for linear operators.
Theorem (Interior regularity [14, Th. 6.17 p. 109]). Let Ω ⊂ Rn a domain.
Suppose that u ∈ C 2 (Ω) satisfies
X X
Lu = aij (x)∂ij u + bi (x)∂i u + c(x)u = f (x)
ij i

where f and the coefficients of the elliptic operator L belong to C k,α (Ω). Then
u ∈ C k+2,α (Ω).

In fact, taking into account that H ∈ C 1,α (Ω), u can be seen as a solution of the
linear equation Lu u = τ nH(x) where Lu is defined in (3.1). Therefore, Theorem
4.1 can be applied to obtain
! !
√
sup k∇u(x)k 6 3 + sup k∇uk exp 2 sup |u| (1 + 8n (τ kHk1 ))
Ω ∂Ω Ω
!
√
6 3 + sup k∇uk exp (2M (1 + 8n (kHk1 ))) .
∂Ω

But, on account of assumptions (4.42) one has for any y ∈ ∂Ω

(n − 1)H∂Ω (y) > n |H(y)| > τ n |H(y)| .

Then, Theorem 4.2 ensures the existence of some constant C = C(n, Ω) such that
3
sup k∇uk 6 kτ ϕk1 + eC (1+kHk1 +kτ ϕk2 )(1+kτ ϕk1 ) (kuk0 +kτ ϕk0 )
∂Ω
3
6 kϕk1 + eC (1+kHk1 +kϕk2 )(1+kϕk1 ) (M +kϕk0 ) .
36 Chapter 4. Ful illment of the existence program’s requirements

Thus, the family of solutions of the related problems (Pτ ) is bounded in C 1 (Ω)
independently of τ . Theorem 3.1 ensures the existence of a unique solution u ∈
C 2,α (Ω) for our problem (P ). 

The following theorem guarantees an a priori height estimate if the function

H satisfies a further hypothesis in addition to the Serrin condition.

Theorem 4.4 (A priori height estimate [23, p. 484]). Let Ω ∈ Rn be a bounded

domain with ∂Ω of class C 2 . Let H ∈ C 1 (Ω) satisfying
n
k∇H(x)k 6 (H(x))2 ∀ x ∈ Ω (4.43)
n−1
and
(n − 1)H∂Ω (y) > n |H(y)| ∀ y ∈ ∂Ω. (4.44)

If u ∈ C 2 (Ω) ∩ C 0 (Ω) is a solution of the mean curvature equation (2.5) in Ω, then

eµδ − 1
sup |u| 6 sup |u| + , (4.45)
Ω ∂Ω µ
where µ > n sup |H| and δ = diam(Ω).
Ω

Proof. Let d(x) = dist(x, ∂Ω) for x ∈ Ω. Let Ω0 be the biggest open subset of
Ω having the unique nearest point property, then d ∈ C 2 (Ω0 ) (see [14, p. 409,
Lemmas 14.16 and 14.17 p. 355], [23, p. 481, §3 p. 420]).
We now define w = φ ◦ d + sup |u| over Ω, where
∂Ω

eµδ  
φ(t) = 1 − e−µt .
µ
If we prove that |u| 6 w in Ω we obtain the desired estimate. By the sake of
contradiction we suppose first that the function v = u − w attains a maximum
m > 0 at x0 ∈ Ω (note that u 6 w in ∂Ω).
PDE and hypersurfaces with prescribed mean curvature 37

Let y0 ∈ ∂Ω be such that d(x0 ) = dist(x0 , y0 ) = t0 and γ the straight line

segment joining x0 to y0 . Restricting u and w to γ we see that v 0 (t0 ) = 0. Hence,
u0 (t0 ) = w0 (t0 ) = φ0 (t0 ) > 0 which implies that ∇u(x0 ) 6= 0. Therefore, Γ0 =
{x ∈ Ω; u(x) = u(x0 )} is of class C 2 near x0 . Then, there exists a small ball B (z0 )
tangent to Γ0 in x0 such that

u > u(x0 ) in B (z0 ) \ {x0 }. (4.46)

We note that

dist(z0 , y0 ) 6 dist(z0 , x0 ) + dist(x0 , y0 ) =  + d(x0 ).

Hence, for z̃ lying in the intersection of ∂B (z0 ) with the straight line segment
joining z0 to y0 , we have

d(z̃) 6 dist(z̃, y0 ) = dist(z0 , y0 ) −  6 d(x0 ) +  −  = d(x0 ).

Thus, w(z̃) 6 w(x0 ) since φ is increasing. Consequently,

u(z̃) − w(x0 ) 6 u(z̃) − w(z̃) 6 u(x0 ) − w(x0 )

and u(z̃) 6 u(x0 ). By (4.46) one has that z̃ = x0 , so z0 belongs to γ and γ is or-
thogonal to Γ0 . This ensures that x0 ∈ Ω0 because if there exists y1 6= y0 satisfying
d(x0 ) = dist(x0 , y1 ), then the straight line joining y1 and x0 is also orthogonal to
Γ0 , which is a contradiction.
However, let us show that this is also impossible. Using the transformation
formula (2.13) one has

Mw = φ0 (1 + φ02 )∆d + φ00 in Ω0 . (4.47)

38 Chapter 4. Ful illment of the existence program’s requirements

We first estimate ∆d in Ω0 . For x ∈ Ω0 , let y = y(x) in ∂Ω be the nearest point to

x, so x belongs to the segment {y + tNy ; t > 0}, where N is the inner normal to
∂Ω. Note that y is now fixed. Let us denote by {Γt } the hypersurface parallel to
some portion of ∂Ω containing y at distance t. So x belongs to Γd(x) .
Let
n
h(t) = H (y + tNy ) .
n−1
Accordingly
n
h0 (t) = h∇H(y + tNy ), Ny i.
n−1
Taking into account the additional hypothesis (4.43) we see that

n
|h0 (t)| 6 k∇H(y + tNy )k 6 (h(t))2 ,
n−1

hence
|h0 (t)| − (h(t))2 6 0.

Recalling again that H0 (t) > (H(t))2 (see [23, p. 485]) it follows

H0 (t) > (H(t))2 + |h0 (t)| − (h(t))2 . (4.48)

Then,
(H(t) − h(t))0 > (H(t) + h(t)) (H(t) − h(t)) (4.49)

and
(H(t) + h(t))0 > (H(t) − h(t)) (H(t) + h(t)) . (4.50)

Let us define v(t) = H(t) − h(t) and g(t) = H(t) + h(t). From (4.49) one has

v 0 (t) > g(t)v(t)

PDE and hypersurfaces with prescribed mean curvature 39

Rt
g(s)ds
Multiplying this inequality by e 0 , it results
!0
v(t)
Rt > 0,
g(s)ds
e 0

so
v(t)
Rt > v(0) = H(0) − h(0).
e 0 g(s)ds
From the Serrin condition (4.44) it follows that

n
|h(0)| = |H (y)| 6 H∂Ω (y) = H(0).
n−1

thus, v(t) > 0 and H(t) > h(t).

Using (4.50) we obtain in a similar way that H(t) > −h(t). Therefore,

H(t) > |h(t)| ,

that is,
n |H(y + tNy )| 6 (n − 1)HΓt (y + tNy ).

Consequently,
n |H(x)| 6 (n − 1)HΓd(x) (x). (4.51)

This proves that

∆d(x) 6 −n |H (x)| ∀ x ∈ Ω0 .

Using this estimate in (4.47) we have in Ω0

Mw 6 −n |H(x)| φ0 (1 + φ02 ) + φ00 .

In addition
φ00 (t) = −µeµ(δ−t) = −µφ0 (t) < −n |H(x)| φ0 (t)
40 Chapter 4. Ful illment of the existence program’s requirements

and φ0 > 1, so
 3/2
Mw 6 −n |H (x)| φ0 (2 + φ02 ) < −n |H (x)| 1 + φ02 . (4.52)

From (4.52) we conclude that

 3/2
Q(w + m) = Qw =Mw − nH (x) 1 + φ02 6 0 = Qu.

Moreover, u 6 w + m and u(x0 ) = w(x0 ) + m. By the maximum principle

u ≡ w + m in Ω0 which is a contradiction since u < w + m in ∂Ω. This proves
that u 6 w in Ω.
Applying the same argument to the function −u we also obtain that −u 6 w
in Ω. 

Remark 4.5. The proof shows that if there exists a function H satisfying the
hypothesis (4.43) in addition to the Serrin condition (4.44), then any hypersurface
that is parallel to some portion of ∂Ω “inherit” the Serrin condition (see (4.51)).
This geometric implication is the key to obtain the height estimate for solutions of
the mean curvature equation in terms of its boundary values.

Remark 4.6. The proof shows that condition (4.43) only needs to be valid in Ω0 .

We are able to prove the following theorem from Serrin.

Theorem 4.7 (Serrin [23, p. 484]). Let Ω ⊂ Rn be a bounded domain with ∂Ω of

class C 2,α for some α ∈ (0, 1). Let H ∈ C 1,α (Ω) satisfying

n
k∇H(x)k 6 (H(x))2 ∀ x ∈ Ω. (4.53)
n−1
PDE and hypersurfaces with prescribed mean curvature 41

and
(n − 1)H∂Ω (y) > n |H (y)| ∀ y ∈ ∂Ω. (4.54)

Then for every ϕ ∈ C 2,α (Ω) there exists a unique solution u ∈ C 2,α (Ω) of the Dirich-
let problem (P ).

Proof. By Theorem 4.3 it only remains to prove that the family of solutions of the
related problems (Pτ ) is uniformly bounded. Let u be a solution of problem (Pτ )
for arbitrary τ ∈ [0, 1] and let w = φ ◦ d + sup |ϕ| as in the proof of Theorem 4.4.
∂Ω
Similarly as in the proof of that theorem, if the function u − w attains a positive
maximum m at x0 ∈ Ω, then x0 would be in Ω0 (the biggest open subset of Ω
having the unique nearest point property). But for x ∈ Ω0 we have

Qτ (w + m) = Qτ (w) 6 Mw + τ n |H(x)| (1 + φ02 )3/2 6 0

once (4.52) holds and τ ∈ [0, 1]. Proceeding as in the proof of Theorem 4.4, we get
that u 6 w also in Ω0 . Analogously it follows that −u 6 w in Ω. Hence, u satisfies
the estimate (4.45), that is,

eµδ − 1 eµδ − 1
sup |u| 6 sup |τ ϕ| + 6 sup |ϕ| + ,
Ω ∂Ω µ ∂Ω µ

where µ > n sup |H| and δ = diam(Ω). 

Ω

Remark 4.8. Observe that Theorem 4.7 is not exactly the existence part in Theo-
rem D stated in the introduction. In order to reduce the differentiability assump-
tions on the domain and the boundary data, the original problem can be approx-
imated by new problems having the C 2,α differentiability requirements. For in-
stance, ∂Ω can be approximated by C 2,α hypersurfaces. However, it can only
42 Chapter 4. Ful illment of the existence program’s requirements

be guaranteed that the mean curvature of any approximating surface Σ satisfies

(n − 1)HΣ (x) > n |H(x)| − ε for every x ∈ Σ, where ε is a positive constant.
Hence, a boundary gradient estimate sharper than that obtained in Theorem 4.2 is
needed. Furthermore, it is also required an interior gradient estimate which yields
a compactness result also used in this argument. We refer the work of Serrin [23,
§14 p. 451] for further studies.

Remark 4.9. Although the additional hypothesis (4.53) is required in order to

obtain an a priori height estimate, the fundamental role in Theorem 4.7 is played
by the Serrin condition (4.54). For example, if the function H satisfies the integral
condition
Z   n+2  p1
2 − 2
kHkLn (Ω) < n 1 + kpk dp
Rn

instead of (4.53), the height estimate is guaranteed and the same conclusion of
Theorem 4.7 holds as a consequence of Theorem 4.3 (see [16, Ths. 3.2.1 p. 87 and
3.4.1 p. 105] and [14, Th. 16.10 p. 408]).
 5

Sharpness of the Serrin condition

The goal in this chapter is to prove that the Serrin condition,

(n − 1)H(y) > n |H(y)| ∀ y ∈ ∂Ω, (5.1)

is actually sharp for the solvability of the Dirichlet problem (P ). That is, if (5.1)
fails, then there exists boundary values for which problem (P ) has no possible
solution.
The next lemma establishes an a priori height estimate for solutions of equation
(2.5) in Ω in those points of ∂Ω on which the Serrin condition (5.1) fails.

Lemma 5.1 ([16, L. 3.4.4 p. 109]). Let Ω ⊂ M be a bounded domain whose

boundary is of class C 2 . Let H ∈ C 0 (Ω) be a non-negative function and u ∈
C 2 (Ω) ∩ C 0 (Ω) satisfying (2.5). Assume that there exists y0 ∈ ∂Ω such that

(n − 1)H∂Ω (y0 ) < nH(y0 ). (5.2)

Then for each ε > 0 there exists a > 0 depending only on ε, H∂Ω (y0 ), the geometry
of Ω and the modulus of continuity of H in y0 , such that

u(y0 ) < sup u + ε. (5.3)

∂Ω\Ba (y0 )

43
44 Chapter 5. Sharpness of the Serrin condition

Proof. The proof is done in two steps. Firstly, it will be find an estimate for u(y0 )
depending on sup u for some a that does not depend on u. Secondly, an upper
∂Ba (y0 )∩Ω
bound for sup u in terms of sup u is stated.
∂Ba (y0 )∩Ω ∂Ω\Ba (y0 )

Step 1. First of all, note that from (5.2) there exists ν > 0 such that

(n − 1)H∂Ω (y0 ) < nH(y0 ) − 4ν. (5.4)

Let R1 > 0 be such that ∂BR1 (y0 ) ∩ Ω is connected and

ν
|H(x) − H(y0 )| < , ∀ x ∈ BR1 (y0 ) ∩ Ω. (5.5)
n
Let S be a quadric hypersurface inside Ω, tangent to ∂Ω at y0 and whose mean
curvature calculated with respect to the normal field N which coincides with the
inner normal to ∂Ω at y0 satisfies

ν
HS (y0 ) < H∂Ω (y0 ) + . (5.6)
(n − 1)
Let d(x) = dist(x, S) for x ∈ Ω. It is known that d is of class C 2 over the strip

Στ = {x + tNx ; x ∈ S, t ∈ [0, τ )}

for some τ > 0 (see [16, L. 3.1.8 p. 84] and [23, L. 1 p. 420]). For each t ∈ [0, τ )
fixed, let us define
St = {x + tNx ; x ∈ S}.

Let 0 < R2 < min{τ, R1 } be such that

|∆d(x) − ∆d(y0 )| < ν ∀ x ∈ BR2 (y0 ) ∩ Στ . (5.7)

PDE and hypersurfaces with prescribed mean curvature 45

Let us fix a < R2 to be made precise later. For 0 <  < a let

Ω = {x ∈ Ba (y0 ) ∩ Στ ; d(x) > }.

Let φ ∈ C 2 (, a) satisfying

P1. φ(a) = 0, P2. φ0 6 0, P3. φ00 > 0, P4. φ0 () = −∞.

It is also required that

ν(φ0 (t))3 + φ00 (t) = 0, t ∈ (, a). (5.8)

Let us define v = sup u + φ ◦ d. So, v > u in ∂Ω \ S . If u 6 v in S ,

∂Ba (y0 )∩Ω
then an estimate for u(y0 + Ny0 ) is obtained. Notice now that if N is the normal
to S inwards Ω and x ∈ S ∩ Ba (y0 ), then

∂v
(x) = h∇v(x), N (x)i = hφ0 (d(x))∇d(x), ∇d(x)i = φ0 () = −∞.
∂N

So, Proposition 2.2 can be used if Qu > Qv in Ω . This will be proved in the sequel.
For x ∈ Ω the transformation formula (2.13) yields

Qv = φ0 (1 + φ02 )∆d + φ00 − nH(x)(1 + φ02 )3/2 .

The assumptions on φ immediately gives

(1+φ02 )3/2 = (1+φ02 )1/2 (1+φ02 ) > (φ02 )1/2 (1+φ02 ) = |φ0 | (1+φ02 ) = −φ0 (1+φ02 ).

Since H > 0, then

−nH(x)(1 + φ02 )3/2 < nH(x)φ0 (1 + φ02 ).

46 Chapter 5. Sharpness of the Serrin condition

Hence,
Qv < φ0 (1 + φ02 ) (∆d(x) + nH(x)) + φ00 . (5.9)

Furthermore,

∆d(x) + nH(x) =∆d(x) − ∆d(y0 ) − (n − 1)HS (y0 ) + nH(x)

> − ν − (n − 1)HS (y0 ) + nH(x) (a)

> − 2ν − (n − 1)H∂Ω (y0 ) + nH(x) (b)

>2ν − nH(y0 ) + nH(x) (c)

>ν, (d)

where (a) follows directly from (5.7), (b) from (5.6), (c) from (5.4) and (d) from (5.5).
Using this estimate in (5.9) it follows

Qv < φ0 (1 + φ02 )ν + φ00 < φ03 ν + φ00 .

Assumption (5.8) yields Qv < 0 in Ω . From Proposition 2.2 it is deduced that

u6v= sup u + φ() in S ∩ Ba (y0 ).

∂Ba (y0 )∩Ω

Let us now define φ explicitly by (see also [14, §14.4])

s
2  
φ(t) = (a − )1/2 − (t − )1/2 . (5.10)
ν
Observe that φ satisfies P1–P4 and that φ03 ν + φ00 = 0 in (, a). Indeed,
s !1/2
0 1 2 1 2
φ (t) = − (t − )−1/2 = −
2 ν 2 ν(t − )
and s !3/2
00 1 2 ν 2
φ (t) = (t − )−3/2 = = −νφ0 (t)3 .
4 ν 8 ν(t − )
PDE and hypersurfaces with prescribed mean curvature 47

Therefore, s
2  
u(y0 + Ny0 ) 6 sup u+ (a − )1/2 .
∂Ba (y0 )∩Ω ν
Being that this estimate holds for each 0 <  < a, we can pass to the limit as 
goes to zero, so s
2a
u(y0 ) 6 sup u+ . (5.11)
∂Ba (y0 )∩Ω ν

Step 2. Let δ = diam(Ω) and ψ ∈ C 2 (a, δ) satisfying

P5. ψ(δ) = 0, P6. ψ 0 6 0, P7. ψ 00 > 0, P8. ψ 0 (a) = −∞.

It is also needed that

(ψ 0 (t))3
(n − 1) + ψ 00 (t) 6 0, t ∈ (a, δ). (5.12)
t

Let w = sup u + ψ ◦ ρ, where ρ(x) = dist(x, y0 ). Remind that ρ ∈

∂Ω\Ba (y0 )
C 2 (Rn \{y0 }), so w ∈ C 2 (Ω \ Ba (y0 )). The idea is to use Proposition 2.2 again.
Note that w > u in ∂Ω \ Ba (y0 ). Also, if Na is the normal to ∂Ba (y0 ) ∩ Ω inwards
Ω \ Ba (y0 ) and x ∈ ∂Ba (y0 ) ∩ Ω, then

∂w
(x) = h∇w(x), Na (x)i = hψ 0 (ρ(x))∇ρ(x), ∇ρ(x)i = ψ 0 (a) = −∞.
∂Na

On the other hand, the transformation formula (2.13) gives

Qw = ψ 0 (1 + ψ 02 )∆ρ + ψ 00 − nH(x)(1 + ψ 02 )3/2 .

n−1
Since H > 0 and ∆ρ(x) = it follows
ρ(x)
n−1 0 n − 1 03
Qw 6 ψ (1 + ψ 02 ) + ψ 00 < ψ + ψ 00 .
ρ ρ
48 Chapter 5. Sharpness of the Serrin condition

Assumption (5.12) yields Qw < 0 in Ω \ Ba (y0 ).

From Proposition 2.2 we conclude that u 6 w in ∂Ba (y0 ) ∩ Ω, so

sup u6 sup u + ψ(a). (5.13)

∂Ba (y0 )∩Ω ∂Ω\Ba (y0 )

Using (5.13) in (5.11) from step 1 we get

s
2a
u(y0 ) 6 sup u + ψ(a) + .
∂Ω\Ba (y0 ) ν
Let us define ψ by (see also [14, §14.4])
 1/2 Z δ  
2 r −1/2
ψ(t) = log dr. (5.14)
n−1 t a
n−1 0 3
Such a function satisfies P5–P8, and also ψ (t) + ψ 00 (t) < 0 for each t ∈
t
(a, δ). In fact,
 1/2  
0 2 t −1/2
ψ (t) = − log
n−1 a
and
 1/2  −3/2 !
00 2 1 t a1
ψ (t) = − − log
n−1 2 a ta
   
1 2 1/2
t −3/2
= log
2t n − 1 a
 3/2  
n−1 2 t −3/2
= log
4t n−1 a
n−1 0 3
=− ψ (t)
4t

Additionally, it is easy to see that lim ψ(a) = 0. Hence, for each ε > 0, a can
a→0
be chosen small enough to satisfy
s
2a
ψ(a) + < ε. 
ν
PDE and hypersurfaces with prescribed mean curvature 49

Now we are able to prove the following non-existence result.

Theorem 5.2 ([16, Th. 3.4.5 p. 112]). Let Ω ⊂ M be a bounded domain whose
boundary is of class C 2 . Let H ∈ C 0 (Ω) be a function either non-positive or non-
negative. Assume that there exists y0 ∈ ∂Ω such that

(n − 1)H∂Ω (y0 ) < n |H(y0 )| .

Then, for any ε > 0, there exists ϕ ∈ C ∞ (Ω) with |ϕ| < ε on ∂Ω, such that there
exists no u ∈ C 2 (Ω) ∩ C 0 (Ω) satisfying problem (P ).

Proof. Obviously it can be supposed that H > 0. For any ε > 0 take a as in the
previous lemma. Let ϕ ∈ C ∞ (Ω) such that ϕ = 0 in ∂Ω \ Ba (y0 ), 0 6 ϕ 6 ε on
∂Ω ∩ Ba (y0 ) and ϕ(y0 ) = ε. Hence, no solution of equation (2.5) in Ω could have
ϕ as boundary values because such a function does not satisfy (5.3). 
 6

Prescribed mean curvature

equations in manifolds

las secciones no combinan con los resultados de ricardo e nietshe

Dirichlet problems for equations whose solutions describe hypersurfaces with

prescribed mean curvature have been also studied outside of the Euclidean space.

6.1. Vertical graphs with prescribed mean curvature in M × R

There are also some results of this type in the Riemannian product M × R,
where M is a complete Riemannian manifold of dimension n > 2. Analogously to
the Euclidean setting, the solutions of the equation
 
∇M u
divM  q  = nH (6.1)
1 + k∇M ukM
2

are vertical graphs in M × R with mean curvature H at each point of the graph
(see equation (2.3)). To see the difference with the Euclidean case we notice that,
in a coordinate system (x1 , . . . , xn ) in M , the non-divergence form of equation

51
52 Chapter 6. Prescribed mean curvature equations in manifolds

(6.1) is equivalent to
!
1 X n
ui uj
σ ij − ∇2ij u = nH, (6.2)
W i,j=1 W2
X
n
where (σ ) is the inverse of the metric (σij ) of M , u =
ij i
σ ij ∂j u are the coordi-
  j=1
nates of ∇u and ∇2ij u(x) = ∇2 u(x) ∂
, ∂
∂xi ∂xj
(compare with equation (2.4)).

6.1.1. The minimal case: H = 0

In this context, Aiolfi-Ripoll-Soret proved (see [1, Th. 1 p. 72]) that there always
exists a vertical minimal graph (H = 0) in M × R over a mean convex, smooth
and bounded domain Ω in M for arbitrary continuous boundary data. This result
generalizes the existence part in Theorem B stated in the introduction.
On the other hand, the mean convexity condition is sharp when M is a Hadamard
manifold whose sectional curvature is bounded above by −1 due to a work M.
Telichevesky (see [25, Th. 6 p. 246]). The combination of these two results gener-
alizes Theorem B to these types of Hadamard manifolds.

6.1.2. The constant case: H ∈ R

Actually, the pioneer in the study of the Dirichlet problem for equation (6.1)
was Spruck. He proved (see [24, Th. 1.4 p. 787]) that for a constant H satisfying
n2
RiccM > − H 2, (6.3)
n−1
there exists a vertical graphs with constant curvature H over a bounded domain
Ω ⊂ M for arbitrary continuous boundary values if

(n − 1)H∂Ω (y) > n |H| ∀ y ∈ ∂Ω.

PDE and hypersurfaces with prescribed mean curvature 53

Notice that, unlike the minimal case, this result of Spruck does not generalizes
the existence part in Theorem C because of the additional hypothesis (6.3). It will
get clear ………………………………………………………

6.1.3. The non‑constant case

The author of these notes has generalized the aforementioned non-existence

result in the M × R context on her PhD thesis [4], supervised by professor R. Sa
Earp.
First of all, it was proved the following generalizations of Theorem 5.2 for a
function H depending also on the height:

Theorem 6.1 ([6, Cor. 5 p. 4]). Let M be a Cartan-Hadamard manifold and Ω ⊂

M a bounded domain whose boundary is of class C 2 . Let H ∈ C 0 (Ω × R) be a
function either non-negative or non-positive and non-decreasing in the variable z.
Suppose there exists y0 ∈ ∂Ω such that

(n − 1)H∂Ω (y0 ) < n sup |H(y0 , z)| .

z∈R

Then there exists ϕ ∈ C ∞ (Ω) such that there is no u ∈ C 0 (Ω) ∩ C 2 (Ω) satisfying
equation (6.1) with u = ϕ in ∂Ω.

Theorem 6.2 ([6, Cor. 6 p. 4]). Let M be a compact and simply connected manifold
which is strictly 1/4−pinched1 . Let Ω ⊂ M be a domain with diam(Ω) < √π
2 K0
and
whose boundary is of class C 2 . Let H ∈ C 0 (Ω×R) be a function either non-negative
1
A Riemannian manifold is said to be strictly 1/4−pinched if the sectional curvature K of M
satisfies 14 K0 < K 6 K0 for a positive constant K0 .
54 Chapter 6. Prescribed mean curvature equations in manifolds

or non-positive and non-decreasing in the variable z. Suppose there exists y0 ∈ ∂Ω

such that
(n − 1)H∂Ω (y0 ) < n sup |H(y0 , z)| .
z∈R

Then there exists ϕ ∈ C ∞ (Ω) such that there is no u ∈ C 0 (Ω) ∩ C 2 (Ω) satisfying
equation (6.1) with u = ϕ in ∂Ω.

The statements of these theorems ensure that the strong Serrin condition

(n − 1)H∂Ω (y) > n sup |H(y, z)| ∀ y ∈ ∂Ω (6.4)

z∈R

is a necessary condition for the solvability of the Dirichlet problem for equation
(6.1) in these types of manifolds.
Some direct consequences derived from these non-existence results are the fol-
lowing. Firstly, the combination of Theorem 6.1 with the existence theorem from
Aiolfi-Ripoll-Soret [1, Th. 1 p. 72] for the minimal case mentioned above shows
that the sharp solvability criterion of Jenkins-Serrin, Theorem B stated in the in-
troduction, actually holds in every Cartan-Hadamard manifolds:

Theorem 6.3. Let M be a Cartan-Hadamard manifold and Ω ⊂ M a bounded

domain whose boundary is of class C 2 . Then the Dirichlet problem for the minimal
vertical equation in Ω has a unique solution for arbitrary continuous boundary data
if, and only if, Ω is mean convex.

Secondly, combining Theorem 6.2 with an existence result of Spruck [24, Th.
1.4 p. 787] we infer the following:
checar regularidade do teorema a seguir
PDE and hypersurfaces with prescribed mean curvature 55

Theorem 6.4. Let M be a compact and simply connected manifold which is strictly
1/4−pinched. Let Ω ⊂ M be a domain with diam(Ω) < √π
2 K0
and whose boundary
is of class C 2,α for some α ∈ (0, 1). Then for every constant H the Dirichlet problem
for equation (6.1) has a unique solution for arbitrary continuous boundary data if,
and only if, (n − 1)H∂Ω > n |H|.

On the other hand, we also proved a generalization of Theorem 4.7 for a func-
tion H depending also on the height:

Theorem 6.5 ([5, Th. 3 p. 2]). Let Ω ⊂ M be a bounded domain with ∂Ω of class
C 2,α for some α ∈ (0, 1). Let H ∈ C 1,α (Ω × R) satisfying ∂z H > 0 and

n2
Riccx > n sup k∇x H(x, z)k − inf (H(x, z))2 ∀ x ∈ Ω. (6.5)
z∈R n − 1 z∈R

If
(n − 1)H∂Ω (y) > n sup |H (y, z)| ∀ y ∈ ∂Ω, (6.6)
z∈R

then for every ϕ ∈ C 2,α (Ω) there exists a unique solution u ∈ C 2,α (Ω) of the Dirich-
let problem for equation (6.1).

We hightlight that Theorem 6.5 also generalizes the result for constant H in
the M × R setting proved by Spruck (see [24, T 1.4 p. 787]).
As in the Euclidean case, for the proof we used the technique based on the
Leray-Schauder fixed point theorem explained in Chapter 3. Thus, we established
an a priori estimate in the C 1 class for the solutions of a family of Dirichlet prob-
lems related to equation (6.1). However, outside the Euclidean ambient space we
face some geometric properties that make the study more complex.
56 Chapter 6. Prescribed mean curvature equations in manifolds

For instance, we recall that to obtain the a priori gradient estimate in Theorem
4.2, was used the fact that in every mean convex domain of the Euclidean space the
mean curvature of hypersurfaces which are parallel to the boundary is increasing
along the inner normal lines (see expression (4.28)). This geometric property is
not always true in mean convex domains of Hadamard manifolds2 . Surprisingly,
the combination of (6.5) and (6.6) implies that the mean curvature of the parallel
hypersurfaces to the boundary is increasing along the inner normal geodesics.
Also, as we have seen in the proof of Theorem 4.7, the combination of (4.53)
and (4.54) implies that any hypersurface inside Ω which is parallel to a portion
of ∂Ω “inherits” condition (4.54) (see equation (4.51)). In [5] we proved that this
geometric property is also true in domains of a complete Riemannian manifold on
which there exists a function H satisfying (6.5) and (6.6). Again, this is one of the
keys to establish the a priori height estimate.
The global gradient estimate obtained in [5] depends explicitly on the height
and the boundary gradient estimates as well as in the Euclidean case, and also in
a lower bound for the Ricci curvature of M in Ω.
Theorem 6.5 in combination with Theorem 6.1 leads to the following general-
2
Let us consider a mean convex domain Ω in the hyperbolic space Hn . For y ∈ ∂Ω let λi (t) be
the ith principal curvature of Γt at γy (t), then (see [2, p. 17])

− tanh t + λi (0)
λi (t) =
1 − λi (0) tanh t

and so
 
2
sech2 (t) (λi (0)) − 1
λ0i (t) = 2 .
(1 − λi (0) tanh t)

Thus, HΓt (γy (t)) decrease if |λi (0)| < 1 for all 1 6 i 6 n.
PDE and hypersurfaces with prescribed mean curvature 57

ization in the C 2,α class of Theorem D of Serrin stated in the introduction:

Theorem 6.6. Let M be a Cartan-Hadamard manifold and Ω ⊂ M a bounded

domain whose boundary is of class C 2,α for some α ∈ (0, 1). Suppose that H ∈
C 1,α (Ω × R) is either non-negative or non-positive in Ω × R, ∂z H > 0 and
n2
Riccx > n sup k∇x H(x, z)k − inf (H(x, z))2 , ∀ x ∈ Ω.
z∈R n − 1 z∈R
Then the Dirichlet problem for equation (6.1) has a unique solution u ∈ C 2,α (Ω) for
every ϕ ∈ C 2,α (Ω) if, and only if,

(n − 1)H∂Ω (y) > n sup |H(y, z)| ∀ y ∈ ∂Ω.

z∈R

Using Theorem 6.2 we derived:

Theorem 6.7. Let M be a compact and simply connected manifold which is strictly
1/4−pinched. Let Ω ⊂ M be a domain with diam(Ω) < √π
2 K0
and whose boundary
is of class C 2,α for some α ∈ (0, 1). Suppose that H ∈ C 1,α (Ω × R) is either non-
negative or non-positive in Ω × R, ∂z H > 0 and
n2
Riccx > n sup k∇x H(x, z)k − inf (H(x, z))2 , ∀ x ∈ Ω.
z∈R n − 1 z∈R
Then the Dirichlet problem for equation (6.1) has a unique solution u ∈ C 2,α (Ω) for
every ϕ ∈ C 2,α (Ω) if, and only if,

(n − 1)H∂Ω (y) > n sup |H(y, z)| ∀ y ∈ ∂Ω.

z∈R

6.1.4. Hn × R

As we observed at the Introduction, the classical Theorem C is a direct conse-

quence of Theorem D since (1.3) is satisfied by every H ∈ R.
58 Chapter 6. Prescribed mean curvature equations in manifolds

However, when we consider the case of constant mean curvature in the more
general case M ×R, we can not conclude from Theorem 6.5 that the Serrin condition
is sufficient by itself for the solvability of the Dirichlet problem for equation (6.1)
since the Ricci curvature of M is involved in (6.5) (see also Spruck’s existence result
[24, T 1.4 p. 787] for constant H in M × R).
By way of illustrating better this fact notice that if M = Hn and H ∈ R,
relation (6.5) reduces to |H| > n−1
n
. Thus, in the opposite case |H| < n−1
n
, we
can not ensure that the parallel hypersurfaces to the boundary “inherit” the Serrin
condition and that it mean curvature is increasing along the normal geodesics.
Nevertheless, in this case Spruck was able to stated an existence theorem assuming
that the strict inequality
(n − 1)H∂Ω > n |H| (6.7)

holds. In this case we observe that if (6.7) holds, then (n − 1)HΣ > n |H| for every
hypersurface Σ contained in a small tubular neighbourhood of the boundary and
parallel to it. Besides, in Hn × R there exists an entire vertical graph of constant
mean curvature n−1
n
(see [7, Th. 2.1 p. 22] for explicit formulas) whose vertical
translations and reflections are barriers for vertical graphs with mean curvature
|H| < n−1
n
. (see [24, Th. 5.4 p. 797]).
Also in [5] we extended this result of Spruck in the hyperbolic space including
the inequality in the Serrin condition even for not necessarily constant H:

Theorem 6.8. Let Ω ⊂ Hn be a bounded domain with ∂Ω of class C 2,α for some
α ∈ (0, 1). Let H ∈ C 1,α (Ω × R) satisfying ∂z H > 0 and sup |H| 6 n−1
n
. If
Ω×R

(n − 1)H∂Ω (y) > n sup |H (y, z)| ∀ y ∈ ∂Ω,

z∈R
PDE and hypersurfaces with prescribed mean curvature 59

then for every ϕ ∈ C 2,α (Ω) there exists a unique solution u ∈ C 2,α (Ω) of the Dirich-
let problem for equation (6.1).

In this particular case we used the explicit expressions of the principal curva-
tures of the parallel hypersurfaces to the boundary in order to avoid the problem of
knowing the monotonicity of it mean curvature along the inner normal geodesics.
Since Theorem 6.1 guaranties that Theorem 6.8 is sharp, we derive the following
solvability criterion (see also [6, Th. 7 p. 5]):

Theorem 6.9. Let Ω ⊂ Hn be a bounded domain with ∂Ω of class C 2,α for some
α ∈ (0, 1). Let H ∈ C 1,α (Ω × R) be a function satisfying ∂z H > 0 and 0 6 H 6
n−1
n
in Ω × R. Then the Dirichlet problem for equation (6.1) has a unique solution
u ∈ C 2,α (Ω) for every ϕ ∈ C 2,α (Ω) if, and only if,

(n − 1)H∂Ω (y) > n sup |H (y, z)| ∀ y ∈ ∂Ω,

z∈R

One of the implications of having been able to include equality in (6.7) is the
fact that this proves that the Serrin condition is sufficient for the solvability of the
Dirichlet problem for any constant H also in hyperbolic space. Thus, the Serrin
sharp solvability criterion for constant H stated in Theorem C also holds in the
hyperbolic space in the C 2,α class (see also [?, Th. 8 p. 5]):

Theorem 6.10 (Serrin type solvability criterion in Hn × R). Let Ω ⊂ Hn be a

bounded domain with ∂Ω of class C 2,α for some α ∈ (0, 1). Let H ∈ C 1,α (Ω × R)
be a function satisfying ∂z H > 0 and 0 6 H 6 n−1
n
in Ω × R. Then the Dirichlet
problem for equation (6.1) has a unique solution u ∈ C 2,α (Ω) for every ϕ ∈ C 2,α (Ω)
if, and only if, the strong Serrin condition (6.4) holds.
60 Chapter 6. Prescribed mean curvature equations in manifolds

6.2. Killing Graphs

Mucho antes de los trabajos en el contexto M × R, las personas estudiaron

Dirichlet problems en ambientes conhecidos tales como el espacio hiper-
bólico. Poner las datas⁇⁇

For instance, in the half-space setting of the hyperbolic space Hn+1 , P.-A.
Nitsche [20] studied graphs-like hypersurfaces over the totally geodesic hypersur-
face S = {x ∈ Rn+1
+ ; (x0 ) + · · · + (xn ) = 1}. Nitsche proved that there exists a
2 2

radial graph over Ω with prescribed mean curvature a function H(x) taking values
on [0, 1] for arbitrary continuous boundary values if |H(y)| < HC (y) everywhere
on ∂Ω; while if |H(y)| > HC (y) for some y ∈ ∂Ω, it can be constructed a smooth
curve that cannot be the boundary of any graph over Ω with curvature H.
Also in this setting, Guio-Sa Earp [15] considered hypersurfaces which in-
tersect at must in one point the horizontal horocycles orthogonal to a bounded
domain Ω contained in a vertical hyperplane. They studied the Dirichlet prob-
lem for the prescribed mean curvature equation of these types of graphs and ob-
tained a sharp result when HC (y) > |H(y)| for each y ∈ ∂Ω provided again that
|H(x)| 6 1 in Ω.
We observe that all the graphs … mentioned in this chapter are examples of
Killin graphs, a concept introduced by Dajczer-Hinojosa-Lira [, ].
Precisely, let M be a complete Riemannian manifold of dimension n+1, n > 2.
Let us suppose that M is endowed with a non-singular Killing vector field Y whose
orthogonal distribution is integrable. We fix some integral leaf M and we denote
by Φ : M × R → M the flow generated by Y . The Killing graph of a function u
PDE and hypersurfaces with prescribed mean curvature 61

defined over a bounded domain Ω ⊂ M , firstly defined by Dajczer-Hinojosa-Lira

in [10], is the hypersurface

Gr u = {Φ(x, u(x)); x ∈ Ω}. (6.8)

We observe that all the cases mentioned before are examples of Killing Graphs.
If H(Φ(x, u(x))) is the mean curvature of Σ at each point Φ(x, u(x)), then u
necessarily satisfies the equation
Colocar Killing al final de todos los labels de esta parte

  
W 2 ∆u − ∇2 u(∇u, ∇u) − ρ−2 ρ−2 + W 2 ∇Y Y, ∇u = nH(Φ(x, u))W 3 ,
(6.9)
where
ρ = ρ(x) = kY kM , (6.10)
q
W = ρ−2 + k∇uk2 . (6.11)

Other works have considered a Serrin type condition that provides some exis-
tence theorems in more general context (see [3], [10], [11] as examples). However,
to the best of our knowledge, no other Serrin-type solvability criterion has been
proved in settings different from the Euclidean one.
 Bibliography

[1] A. Aiolfi, J. Ripoll, and M. SoRet, The Dirichlet problem for the mini-
mal hypersurface equation on arbitrary domains of a Riemannian manifold,
Manuscripta Math., 149 (2016), pp. 71–81.

[2] M. AlexandRino, Hipersuperfícies de Nível de uma função Transnormal, Mas-

ter’s thesis, PUC-Rio, 09 1997.

[3] L. J. AlÍas and M. DajczeR, Constant mean curvature graphs in a class of

warped product spaces, Geom. Dedicata, 131 (2008), pp. 173–179.

[4] Y. N. AlvaRez, Critérios de solubilidade de tipo Serrin para problemas de Dirich-

let para equações de curvatura média pré-determinada em variedades, PhD the-
sis, PUC-Rio, december, 2018.

[5] Y. N. AlvaRez and R. Sa EaRp, Existence Serrin type results for the Dirichlet
problem for the prescribed mean curvature equation in Riemannian manifolds,
arXiv e-prints, (2019), p. arXiv:1902.10774.

63
64 Bibliography

[6] Y. N. AlvaRez and R. Sa EaRp, Sharp solvability criteria for Dirichlet problems
of mean curvature type in Riemannian manifolds: non-existence results, Calc.
Var. Partial Differential Equations, 58 (2019), p. 197.

[7] P. BÉRaRd and R. Sa EaRp, Examples of H-hypersurfaces in Hn × R and geo-

metric applications, Mat. Contemp., 34 (2008), pp. 19–51.

[8] S. BeRnstein, Sur les surfaces définies au moyen de leur courbure moyenne ou
totale, Ann. Sci. Éc. Norm. Supér. (4), 27 (1910), pp. 233–256.

[9] L. CaffaRelli, L. NiRenbeRg, and J. SpRucK, Nonlinear Second-Order Elliptic

Equations V: The Dirichlet Problem for Weingarten Hypersurfaces, Comm. Pure
Appl. Math., 41 (1988), pp. 47–70.

[10] M. DajczeR, P. A. Hinojosa, and J. H. de LiRa, Killing graphs with prescribed

mean curvature, Calc. Var. Partial Differential Equations, 33 (2008), pp. 231–
248.

[11] M. DajczeR and J. Ripoll, An extension of a theorem of Serrin to graphs in

warped products, J Geom. Anal., 15 (2005), pp. 193–205.

[12] J. Douglas, Solution of the problem of Plateau, Trans. Amer. Math. Soc., 33
(1931), pp. 263–321.

[13] R. Finn, Remarks relevant to minimal surfaces, and to surfaces of prescribed

mean curvature, J. Anal. Math., 14 (1965), pp. 139–160.

[14] D. GilbaRg and N. S. TRudingeR, Elliptic Partial Differential Equations of

Second Order, Classics in Mathematics, Springer-Verlag, 2001.
PDE and hypersurfaces with prescribed mean curvature 65

[15] E. M. Guio and R. SÁ EaRp, Existence and non-existence for a mean curvature
equation in hyperbolic space, Comm. Pure Appl. Math., 4 (2005), pp. 549–568.

[16] Q. Han, Nonlinear Elliptic Equations of the Second Order, Graduate Studies in
Mathematics, American Mathematical Society, 2016.

[17] H. JenKins and J. SeRRin, The Dirichlet problem for the minimal surface equa-
tion in higher dimensions., J. Reine Angew. Math., 229 (1968), pp. 170–187.

[18] J. L. LagRange, Essai dúne nouvelle méthode pour détermminer les maxima e les
minima des formules intégrales indéfinies, Miscellanea Taurinensia, 2 (1760).

[19] J. B. M. C. MeusnieR, Mémoire sur la courbure des surfaces, Acad. Sci. Paris,
10 (1777/1785), pp. 477–510.

[20] P.-A. Nitsche, Existence of prescribed mean curvature graphs in hyperbolic

space, manuscripta mathematica, 108 (2002), pp. 349–367.

[21] T. RadÓ, The problem of the least area and the problem of Plateau, Math. Z., 32
(1930), pp. 763–796.

[22] H. F. ScheRK, Bemerkung über der kleinste Fläche innerhalb gegebener Grenzen,
J. Reine Angew. Math, 13 (1834), pp. 185–208.

[23] J. SeRRin, The Problem of Dirichlet for Quasilinear Elliptic Differential Equa-
tions with Many Independent Variables, Philos. Trans. Roy. Soc. A, 264 (1969),
pp. 413–496.

[24] J. SpRucK, Interior gradient estimates and existence theorems for constant mean
curvature graphs in M n × R, Pure Appl. Math. Q, 3 (2007), pp. 785–800.
66 Bibliography

[25] M. TelichevesKy, A note on minimal graphs over certain unbounded domains

of Hadamard manifolds, Pacific J. Math., 281 (2016), pp. 243–255.
 Index

Gauss map, 5

mean convex domain, 3

mean curvature, 5

principal curvatures, 5

Weingarten map, 5

67
 Yunelsy N. Alvarez
Departamento de Matemá tica
Instituto de Matemá tica e Estadı́stica
Universidade de Sã o Paulo
Sã o Paulo, Brazil, CEP 05508‑090
[email protected]; [email protected]