130
pages

Voir plus
Voir moins

Vous aimerez aussi

Universität Ulm | 89069 Ulm | Deutschland Fakultät für

Mathematik und

Wirtschaftswissenschaften

Institut für

Angewandte Analysis

Elliptic and Parabolic Problems with

Robin Boundary Conditions on Lipschitz

Domains

Dissertation zur Erlangung des Doktorgrades Dr. rer. nat.

der Fakultät für Mathematik und Wirtschaftswissenschaften der Universität Ulm

vorgelegt von Robin Nittka aus Ehingen im Jahr 2010

Tag der Prüfung:

29. Juni 2010

Gutachter:

Prof. Dr. Wolfgang Arendt

Prof. Dr. Werner Kratz

Prof. Dr. Reiner Schätzle

Amtierender Dekan:

Prof. Dr. Werner KratzJune 22, 2010

c 2010 Robin Nittka

This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike

3.0 License. To view a copy of this license, visit

http://creativecommons.org/licenses/by-nc-sa/3.0/de/ or send a letter to Creative

Commons, 543 Howard Street, 5th Floor, San Francisco, California, 94105, USA.

ALayout: PDF-LT X2"EContents

1 Introduction 1

2 Preliminaries and notations 11

2.1 Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 Function spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.3 Lipschitz domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.4 Functional analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.5 Monotone operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.6 Interpolation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.7 Linear semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.8 Nonlinear and subdiﬀerentials . . . . . . . . . . . . . . . . . 35

2.9 Diﬀerential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3 Elliptic equations with Neumann boundary conditions 45

3.1 Hölder regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.2 Existence of solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.2.1 Linear equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.2.2 Quasilinear equations . . . . . . . . . . . . . . . . . . . . . . . . 56

3.3 Continuous dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.3.1 Linear equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.3.2 Regular equations . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.3.3 Uniformly monotone equations . . . . . . . . . . . . . . . . . . . 71

4 Elliptic equations with Robin boundary conditions 79

4.1 Existence of solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4.2 Hölder regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

5 Parabolic equations 97

5.1 Linear with Robin boundary condition . . . . . . . . . . . . . 97

5.2 equations with Wentzell-Robin boundary conditions . . . . . . . 103

5.3 Nonlinear with Robin boundary conditions . . . . . . . . . . . 110

Bibliography 119

iii1 Introduction

As the title already suggests, in this thesis we are concerned with elliptic and parabolic

problems. Roughly speaking, our investigations of the solutions can be separated into

four categories: existence, uniqueness, regularity, and continuous dependence on the

data. The emphasis is on the regularity for elliptic and parabolic equations. But it

turns out that in order to studyy one has also to understand the other three

topics. We start with a short description of our motivation on the basis of a standard

example.

One of the model equations in the theory of partial diﬀerential equations is the Neumann

problem 8

u u =f on

;<

(1.1)@u

: =g on @

@

Nwith > 0 and

an open set inR . A basic result is that if

, f, and g are of class

1 1C , then there exists a classical solution u2 C ( ) of (1.1).

In many important situations one does not have such a high regularity. For example

for numerical applications it is very natural to consider (1.1) for a domain

that has

2corners, maybe a triangle inR . Passing to a weaker notion of solution, the celebrated

Lax-Milgram theorem tells us that there exists a unique weak solution u of (1.1) even

2if

is merely a Lipschitz domain, where we only have to require f 2 L ( ) and

2g2 L (@ ) . Here, a weak solution is a priori only once weakly diﬀerentiable, i.e.,

1;2u2W ( ) .

It is now a natural question whether this weak solution has higher regularity. If

is

smooth enough, then the unique weak solution u of

8

u u =f on

;<

(1.2)@u

: = 0 on @

@

2;p pis twice weakly diﬀerentiable, and more precisely u2W ( ) if f2L ( ) . Thus if

q=2f2 L ( ) for some q > N, then the unique solution of (1.2) is continuous up to

the boundary of

, i.e., u2 C(

), by the Sobolev embedding theorems, and in fact

even Hölder continuous up to the boundary. On the other hand, for general Lipschitz

1;pdomains the solution will typically not even be in W ( ) for large p even if f is

very smooth. We can, however, still hope that u is continuous up to the boundary if

q=2f2L ( ) for some q>N.

11 Introduction

In fact, the desire to prove that the weak solution u of (1.2) is continuous up to the

q=2boundary if

is a bounded Lipschitz domain andf2L ( ) ,q>N, was the starting

point for this thesis. The basic approach to attack this problem is to extend u by

0reﬂection along the graph of @

to a function u~ deﬁned in a neighborhood

of

.

0We can show that u~ solves again a uniformly elliptic diﬀerential equation on

. But

since the derivative of a Lipschitz function is in general not continuous, the coeﬃcients

of this new equation are merely bounded. Still, the celebrated result independently

0due to De Giorgi and Nash allows us to deduce that u~ is continuous on

. Hence u is

continuous on

.

It turns out that this strategy still works for the inhomogeneous equation (1.1). More-

over, the proof goes through for a much larger class of diﬀerential equations, including

degenerate and singular quasilinear elliptic equations. In addition, it is possible to

obtain the analogous result for Robin boundary conditions via reduction to the Neu-

mann case. Finally, we can apply these elliptic results to show that the corresponding

parabolic problems are well-posed in the space C(

), i.e., we have continuous solutions

for continuous initial data, or, in other words, the elliptic operator generates a strongly

continuous semigroup on C( ) .

We now discuss the general setting. In this thesis we consider quasilinear, elliptic,

second order partial diﬀerential equations in divergence form with inhomogeneous

NRobin boundary conditions on a bounded Lipschitz domain

R , i.e., equations

that essentially look like

(

divA(x;u;ru) =f on

;

(1.3)

A(x;u;ru) +h(u) =g on @

;

where denotes the outer unit normal of

. The model problems we have in mind

arise for

2A(x;u;ru) = (1 + arctanjruj )ru (1.4)

and

m 2A(x;u;ru) =jruj ru (1.5)

with m2 (1;1). For the latter, problem (1.3) is called the m-Laplace equation.

Equations of this type appear in many mathematical models of physical processes, e.g.

nonlinear diﬀusion and ﬁltration [Phi61], deformation plasticity [AC84], and viscoelastic

materials [LT90]. Also the special case of linear equations is included, and frequently

we pay special attention to it. Our main results for these problems are summarized in

Examples 4.2.7, 4.2.9, 4.2.10, and 4.2.16.

Several authors have shown under varying assumptions on the coeﬃcients that every

solution of (1.3) is locally Hölder continuous in the interior of

, see for example [LU68,

Ser64, Tru67]. Without exaggeration, it can be said that it is well understood how

interior regularity can be obtained for equations like (1.3).

Regarding regularity at the boundary, classically the domains are assumed to be

smooth. More recently, however, the class of Lipschitz has attracted much

2interest. As examples we refer to the articles by Kenig and Rule [KR09], Mitrea

and Monniaux [MM09], Mitrea and Taylor [MT09], Agranovich [Agr08], Shen [She08],

Haller-Dintelmann and Rehberg [HDR09], and Wood [Woo07], to name only a few of

the newest contributions.

Among all possible boundary conditions, Dirichlet boundary conditions are the most

popular, meaning that one prescribes values for u on the boundary of

. Regularity

of solutions of Dirichlet problems lies at the very heart of potential theory, and sharp

conditions are known under which the solution is continuous up to the boundary of

.

For the linear case, this is the celebrated Wiener criterion. Quasilinear generalizations

have been found and studied by Maz’ya [Maz70], Gariepy and Ziemer [GZ77], and

Kilpeläinen and Malý [KM94]. In fact, much is known about the regularity of the

solution and its derivatives even if the right hand side is very rough, see for example

recent articles by Mingione [Min07, Min10] and Duzaar and Mingione [DM09].

For Robin boundary conditions, on the other hand, and even for the special case of

Neumann b i.e., if h(u) = 0 in (1.3), the situation is not as well

understood. There are, however, results due to Lieberman [Lie83, Lie92] if the domain

is smooth except for a small set. One of the main goals of this thesis is to establish

regularity up to the boundary also for these boundary conditions if

is a Lipschitz

domain. More precisely, we want to show that every solution is Hölder continuous up

to the boundary of

. This means that u allows for a continuous extension to

which

is Hölder continuous for the same exponent and the same Hölder constant.

For Neumann boundary conditions, we use the elegant method that we had already

described for the Laplace operator to deduce regularity at the boundary from interior

regularity results. In fact, if u is a solution of (1.3), then we extend u to a function u~

0on a larger domain

containing

by reﬂection along the boundary of

. If g = 0,

0then u~ solves an equation of the same type as (1.3) on

. Thus we can deduce that u~

0is locally Hölder continuous in the interior of

and hence in particular on

, where it

coincides with u.

Let us compare this trick with the standard approach to regularity at the boundary:

for smooth boundary, a typical technique is to locally transform the problem into

a diﬀerential equation on a half-space by a smooth, nonlinear transformation in the

spatial variables. Regularity on a half-space can often be shown by direct calculations.

Alternatively, one may extend the function from the half-space to the whole space by

reﬂection and apply the same arguments as above. In this sense, our idea is somewhat

similar to the usual one, but we merged the two steps into a single one. Moreover,

due to the lack of smoothness of the boundary, we cannot apply the usual spatial

transformation, which would correspond to reﬂection along the outer normal of

, but

have to reﬂect along the graph.

There are still some details that require additional care. Firstly, we run into trouble

because the structure of the equation is not maintained under reﬂection if g = 0.

However, this is a minor issue. In fact, the extension u~ still solves an elliptic problem,

but with a distributional right hand side, and interior regularity results are known also

in this case.

3

61 Introduction

Secondly, a major diﬃculty arises if we want to allow non-trivial functions h. The

strategy the author found to overcome this issue is as follows: ﬁx a solution u of (1.3)

:and consider g~ =g h(u) as ﬁxed. Then (1.3) takes the form

(

divA(x;u;ru) =f on

;

A(x;u;ru) =g~ on @

;

i.e., (1.3) is an inhomogeneous Neumann problem. We can hope that the result for the

Neumann case applies to this situation. In typical situations, however, it is not obvious

that g~ is regular enough to apply the result for Neumann equations. This is where the

technicalities start.

In order to show that h(u) is well behaved, we use nonlinear interpolation and boot-

strapping. More precisely, we assume that the Neumann problem has a unique solution

for every right hand side, its resolvent is Hölder continuous, and that for suﬃciently

regular right hand side the solution is Hölder continuous. These assumptions allow

us to apply a nonlinear interpolation theorem in order to deduce that the resolvent

pis regularizing on a scale of L -spaces. Now a bootstrap procedure gives the desired

regularity for h(u) and hence Hölder continuity of u. We also exhibit large classes

of equations that satisfy these assumptions. In particular, the aforementioned model

cases are covered.

Finally, we apply the elliptic theory to parabolic problems. In fact, having very good

elliptic results at hand, it is not diﬃcult to show that a parabolic problem like the heat

equation with Neumann boundary conditions

8

> u_(t) = u(t) on

for t> 0;><

@u(t)

= 0 on @

for t> 0;

> @>:

u(0) =u on

0

is well-posed on C(

), provided one knows the appropriate theorems of semigroup

theory. Here, by well-posedness we mean that for every continuous function u there0

exists a unique solution in a weak sense and that this solution, considered as a function

on the parabolic cylinder [0;1)

, is continuous. We prove this regularity for general

linear equations with Robin or Wentzell-Robin boundary conditions and also for the

two quasilinear parabolic equations with Robin b arising from our

model examples (1.4) and (1.5).

We conclude the introduction with a more precise outline of the results, including

citations to related research articles and books.

Chapter 2 contains the deﬁnitions and fundamental results that we apply in the main

body of this manuscript. More precisely, Section 2.1 consists of several estimates for

Nreal numbers, or rather for vectors inR . These are well known to people working

with the m-Laplace equation. In Section 2.2 we introduce the Sobolev spaces, in

4which weak solutions are to be found, and spaces of Hölder continuous functions. In

this context, we recall several rules for the calculus of Lipschitz continuous functions

such as the change of variables formula and the chain rule. Section 2.3 addresses

the geometric notion of a Lipschitz domain, which comprises the class of polygons

that is particularly important in applications. We quote some analytic facts for later

use, most prominently the divergence theorem, the Sobolev embedding theorems and

the trace theorems. Section 2.4 concludes the elementary part of the preliminaries

with a very short summary of selected theorems from functional analysis that are not

entirely renowned, which is included for easier reference. For an introduction to the

language of functional analysis, which is used freely, the reader is referred to the books

by Werner [Wer00] and Lax [Lax02].

Starting with Section 2.5, which contains existence theorems for nonlinear equations,

the material becomes more sophisticated. There are only few general ideas to solve

quasilinear equations. A particularly successful one is the approach via monotonic-

ity conditions, which we use here. The theory of monotone operators goes back to

Golomb [Gol35], Kačurovski [Kač60], and Zarantonello [Zar60], and it has been devel-

oped much further by many authors, among them Brézis, Minty, Browder, Rockafellar,

Leray, and Lions. It generalizes to the larger class of pseudo-monotone operators

introduced by Brézis [Bré68], which is the type of operator that we consider here

because this class is more stable under perturbations. We provide the basic deﬁnitions

of the theory and the results that are important for our applications, the main source

being Showalter’s book [Sho97]. Detailed information about monotone operators can

also be found in Brézis’ book [Bré73].

Section 2.6 is divided into two parts, both concerning interpolation theory. For general

information about interpolation theory we refer to Triebel’s book [Tri95]. The ﬁrst part

of Section 2.6 is loosely related to interpolation theory, but rather a direct estimate

pbounding a Hölder continuous function in terms of its Hölder norm and an L -norm.

This result due to Lê [Lê07] is interesting from the point of view of interpolation theory

1 q 0; because it tells us thatL ( ) is of classJ(;L ( ) ; C ( )) , see [Tri95, §1.10.1]. The

purpose for which we use it is the following: if u and u~ are solutions of a quasilinear

~equation of the form (1.3) for right hand sides (f;g) and (f;g~), and if we know that

0; ~we can bound u and u~ in C in terms of the norms of f, f, g, and g~, and if we

m