See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/265154383

VARIATIONAL METHODS FOR NONLINEAR

PARTIAL DIFFERENTIAL EQUATIONS

Article

CITATIONS READS

0 33

1 author:

Carlos Tello
Universidad Nacional de Ingeniera (Peru)
3 PUBLICATIONS 48 CITATIONS

SEE PROFILE

All content following this page was uploaded by Carlos Tello on 21 October 2015.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
VARIATIONAL METHODS FOR NONLINEAR PARTIAL
DIFFERENTIAL EQUATIONS

By

CARLOS TELLO

A Thesis Submitted to the Graduate Faculty of

WAKE FOREST UNIVERSITY

in Partial Fulfillment of the Requirements

for the Degree of

MASTER OF ARTS

Mathematics

December 2010

Winston-Salem, North Carolina

Approved By:

Sarah Raynor, Ph.D., Advisor

Stephen B. Robinson, Ph.D., Chairperson

Richard D. Carmichael, Ph.D.
 Table of Contents

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Chapter 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Basic Idea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Chapter 2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.6 Lower Semicontinuous Functions . . . . . . . . . . . . . . . . . . . . . 9

Chapter 3 Functional Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.1 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Normed Linear Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.1 Bounded Linear Operators . . . . . . . . . . . . . . . . . . . . 16
3.2.2 The Hahn-Banach Theorem . . . . . . . . . . . . . . . . . . . 17

Chapter 4 Weak Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.1 Weak Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Weak Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4 Weak Compactness Theorem . . . . . . . . . . . . . . . . . . . . . . . 24

Chapter 5 Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.0.1 Real-valued Functions . . . . . . . . . . . . . . . . . . . . . . 26
5.0.2 Sequences of Real-Valued Functions . . . . . . . . . . . . . . . 27
5.1 Arzela-Ascoli Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.2 C k (U ) Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.3 Measure Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3.1 Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.4 Lp Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.5 Convolution and Smoothing . . . . . . . . . . . . . . . . . . . . . . . 38

ii
Chapter 6 Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.1 The Sobolev Function Space W k,p (U ) . . . . . . . . . . . . . . . . . . 40
6.2 Sobolev Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.3 Rellich-Kondrachov Compactness Theorem . . . . . . . . . . . . . . . 43

Chapter 7 Variational Analysis in C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

7.1 Variational Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.2 Euler-Lagrange Equation . . . . . . . . . . . . . . . . . . . . . . . . . 48

Chapter 8 Variational Analysis in W 1,q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

8.1 Coercivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
8.2 Lower Semicontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
8.3 Convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
8.4 Weak Sequential Lower Semicontinuity Theorem . . . . . . . . . . . . 55
8.5 Existence of minimizers Theorem . . . . . . . . . . . . . . . . . . . . 58
8.6 Weak Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

Chapter 9 Free Boundary Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

9.1 Existence of Minimizers for problems with Free Boundaries . . . . . . 63
9.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

Chapter 10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

iii
 Acknowledgments

There are many people who helped to make this thesis possible. In first place to my
family for all their unconditional moral support. Thanks to all the great community of
the Wake Forest University, my classmates whom I shared a great learning experience.
My gratitude to all my professors, very specially to my advisor Dr. Sarah Raynor for
all her patience answering uncountable number of analysis questions, to Dr. Edward
Allen for helping me to maintain academic motivation. Also thanks to Dr. Stephen
Robinson and Dr. Richard Carmichael for their useful lessons.

iv
 Abstract

Carlos Tello

The objective of this work is to explain some basic aspects of variational methods for
solving a class of nonlinear partial differential equations. First, relevant mathematical
background of functional analysis and variational calculus is explained. One of the
main results discussed is the existence theorem for minimizers of functionals for the
case of fixed boundary problems. This theorem assumes coercivity and lower semicon-
tinuity conditions of the functional and the energy functional is defined over subsets
of Sobolev spaces. After that, the technical concepts associated to this theorem are
implemented to study some specific free boundary problems.

v
Chapter 1: Introduction

1.1 Basic Idea

Many interesting problems in the natural sciences are described by partial differen-
tial equations (PDE). For example, we would like to make better predictions for the
behavior of physical or biological systems, or, as an important first step, to deter-
mine their stationary patterns. The case of linear problems has been investigated in
great detail by mathematicians using well-established methods. We are much more
interested in the case of nonlinear problems, which often appear in applications.

First, we consider the case of problems with fixed boundary. Suppose we are
interested in solving some nonlinear PDE. Let A[u] = 0 be a PDE where A[] is
a nonlinear partial differential operator and u is the unknown function. We know
that there does not exist a general method for solving this nonlinear PDE. However
it is possible that this PDE can be obtained by minimizing an associated energy
functional. Quite often, finding the minimum of this functional is easier than solving
the nonlinear PDE A[] = 0 directly. Furthermore an important class of nonlinear
PDEs related to physical problems can be obtained from an appropriate variational
problem.

Variational analysis is a framework for minimizing functionals whose domain is

a space of functions. Since this space of functions is usually infinite dimensional we
need functional analysis to describe it formally. There are many different classes of
function spaces and it is an art to select a convenient class of function space for a
specific problem. One of most important issues that we need to resolve before using
any analytical or numerical solving method is to determine the existence of solutions.

1
 There exists a useful method to establish existence of minimizer of functionals.
Suppose we want to minimize a functional I[w] where w belongs to a some class C of
real-valued functions compatible with some boundary condition. Basically the method
we will consider works as follows: First take a minimizing sequence {un }
n=1 C that

satisfy
lim I[un ] = inf I[u],
n uC

and then show that some subsequence of {un } converges to a minimizer u C. To

do this we need three key conditions:
(a) Sequential compactness to guarantee that a minimizing sequence contains a con-
vergent subsequence. This typically require a suitable selection of a weak topology,
(b) Closedness so that the limit of such subsequence belongs to C, and
(c) Lower semicontinuity to ensure that the limit of this minimizing sequence is indeed
a minimizer for I[w].

Another case is the so called free boundary problems, which have been investigated
recently with great interest. In free boundary problems we have not only an unknown
function but also the domain of this function is itself unknown. Similarly to the
previous case, a convenient method to solve this problem is by finding a minimizing
sequence of an appropriate functional. However, in this case the functionals are
usually discontinuous and thus the connection of the variational problem to the PDE is
much more complicated, requiring weak convergence methods and geometric measure
theory.

The plan of this dissertation is to work out the mathematical background required
to deal with these classes of problems. First we discuss formally some main results of
topology and functional analysis, and then we study the properties of Sobolev spaces.
Then we formulate the variational problems over the space of smooth functions, and
after that explain in detail the theorem of existence and uniqueness of the minimizer of

2
functionals over Sobolev spaces with Dirichlet boundary conditions on fixed domains.
Imposing additional growing conditions on the energy functional it is shown that
the minimizer represents the weak solution of the nonlinear PDEs associated to the
variational problem. Finally, we discuss briefly the minimizing problem for the case
of free boundary conditions.

3
Chapter 2: Preliminaries

2.1 Topology

Definition 2.1. A topology on a set X is a collection of subsets of X satisfying:

X, .

is closed under finite intersections.

is closed under arbitrary unions.

A set X equipped with a topology is called a topological space, and it is denoted

by (, X) or simply X if the specific topology is clear. A member of is called an
open set. The complement of a set A X is denoted by Ac . The complement of
an open set is a closed set.

Definition 2.2. Let X be a topological space. A neighborhood of a point x X

is an open set containing x.

Definition 2.3. A topological space X is called Hausdorff if for any pair of points
x, y X with x 6= y there exist neighborhoods U of x and V of y such that U V = .

Definition 2.4. A non trivial set can have many different topologies. Suppose that
we have two topologies and 0 on a set X. If 0 , the we say that 0 is weaker
than .

Definition 2.5. A subfamily B of is a base for if for every x X and every open
set U containing x there is a set V B such that x V U.

4
Definition 2.6. A collection Bx of neighborhoods of a point x X is called local
base at x if every neighborhood of x contains a member of Bx .

Definition 2.7. A topological space is first countable if there exists a countable

local base at each point. A topological space is called second countable if it has a
countable base.

Remark 2.8. If B. is a base for , then each U is a union of members of B.

Definition 2.9. A subfamily S of is a subbase of if the collection of all finite

intersections of members of S is a base for .

Definition 2.10. If Y is a subset of a topological space (X, ), then the topology

induced by on Y is defined by Y = {Y V | V }.

2.2 Closure

Definition 2.11. Let X be a topological space. The interior of A X, denoted

by A , is the union of all open subsets of A. The closure of A, denoted A is the
intersection of all closed supersets of A.

Using this definitions we can verify the following results: The interior is the largest
open set included in A; The closure is the smallest closed set including A; If A
B X, then A B and A B. A set A X is open if and only if A = A , and
B is closed if and only if B = B. For any set A X, (A) = A and (A ) = A and
(A )c = (Ac ). Also, (A B) = A B and (A B) = A B .

Definition 2.12. A point x X is limit point of a set A X if any neighborhood

V of x satisfies A (V {x}) 6= . A point x X is point of closure of a set
B X if for any neighborhood U of x we have B U 6= .

5
 Given A X, the definition of limit point implies the following result: A =
A {limit points of A}.

Remark 2.13.

Definition 2.14. Given a set A X. The boundary of A is defined by A := AAc

This definition implies that if x A, then every neighborhood V of x satisfy

V A 6= and V Ac 6= . We have the following results: If A is an open subset of
X, then A = A A and A A = ;. A set is closed if and only if it contains all its
limit points.

2.3 Compactness

Definition 2.15. Let U X. A collection O of open subsets of X is said to be an

open cover of U if the union of the elements of O contains U .

Definition 2.16. Let U X. Then U is compact if only if every open cover of U

contains a finite subcover.

Some properties of compact sets are the following: Finite sets are compact. Finite
unions of compact sets are compact. Closed subsets of compact sets are compact.
Compact subsets of Hausdorff spaces are closed.

Definition 2.17. D X is dense in X if D = X.

This definition implies that a set D is dense if and only if every nonempty open
set of X contains a point in D. This means that any point in X can be approximated
by points in D.

Definition 2.18. A topological space X is separable if it has a countable dense

subset.

Theorem 2.19. If a topological space is second countable, then it is separable.

6
2.4 Sequences

Definition 2.20. A sequence in X is a function from the natural numbers N =

{1, 2, . . .} into X.

Definition 2.21. A sequence {xn } in a topological space X converges to a point

x X it for each neighborhood V of x there is an N N such that xm V for all
m N . We say that x is the limit of the sequence.

We use the short notation for sequences {uk } representing {uk }

k=1 , and for limits

xk x as k representing limk xk = x.

Definition 2.22. Let {uk } be a sequence in X, and {kj } be a strictly increasing

sequence of natural numbers. We say that the sequence {ukj } is a subsequence of
{uk }.

Theorem 2.23. A topological space is Hausdorff if and only if every sequence con-
verges to at most one point.

If a sequence {xn } in a set A X converges to x, then x A. However, the

converse is not true. There can be points in the closure of a set but no sequences in
the set A converging to that point.

Definition 2.24. U X is sequentially compact if any sequence in U has a

convergent subsequence in U .

Definition 2.25. Let U, V X. We say that V is compactly contained in U if V

is compact and V U , which is denoted by V b U .

Definition 2.26. We say that a set U X is precompact if its closure is compact.

Definition 2.27. If u is a real-valued function defined on U , then the support of u

is defined by supp(u) = {x U : u(x) 6= 0}.

7
2.5 Continuous Functions

Definition 2.28. Let X, Y be topological spaces. A function f : X Y is contin-

uous if f 1 (U ) is open in X for each open set U in Y . We say that f is continuous
at the point x if f 1 (V ) is a neighborhood of x whenever V is a neighborhood of
f (x).

Definition 2.29. If X and Y are Hausdorff spaces and f : X Y is a function,

then f is sequentially continuous at x if limn f (xn ) = f (x) for every sequence
{xn } in X that satisfies limn xn = x.

Theorem 2.30. (a) If f : X Y is continuous at x, then f is sequentially contin-

uous at x.
(b) If f : X Y is sequentially continuous at x, and if X has a countable local base
at x (in particular, if X is metrizable), then f is continuous at x.

Proof. (a) Suppose xn x in X, V is a neighborhood of f (x) in Y , and U = f 1 (V ).

Since f is continuous, U is a neighborhood of x, and therefore there exists an N such
that xn U for all n N . Then f (xn ) V for all n N . Thus f (xn ) f (x) as
n .
(b) Fix x X, let {Un } be a countable local base for the topology of X at x, and
assume that f is not continuous at x. Then there is a neighborhood V of f (x) in
Y such that f 1 (V ) is not a neighborhood of x. Without lost generality we can
consider the countable local base {Un } to be nested, so there is a sequence xn , such
that xn Un , xn x as n , and xn 6 f 1 (V ). Thus f (xn ) 6 V , so that f is
not sequentially continuous.

This theorem shows that sequential continuity is a weaker condition than conti-
nuity, but they are equivalent for metric spaces.

8
Theorem 2.31. Every continuous function between topological spaces maps compact
sets to compact sets.

Corollary 2.32. A continuous real-valued function defined on a compact set achieves

its maximum and minimum values.

For a proof of this important corollary see [18] Theorems 4.14, 4.19.

2.6 Lower Semicontinuous Functions

Definition 2.33. Let X be a topological space, and let f : X R {+}. The

function f is lower semicontinuous at u if

c R, {u X | f (u) c} is closed in X.

If f is lower semicontinuous at every point u of X, then we say that f is lower

semicontinuous on X.

Lemma 2.34. f : X R {} is lower semicontinuous if and only if, for all

c R f 1 ((c, ]) = {x X | f (x) > c} is open in X.

Definition 2.35. Let X be a topological space. A function f : X R {}

is sequentially lower semicontinuous at u X if for any sequence {un } X
converging to u
f (u) lim inf f (un ).
n

If f is sequentially lower semicontinuous at every point u of X, then we say that f is

sequentially lower semicontinuous on X.

Lemma 2.36. Let X be a first countable topological space. An extended real valued
function f : X R {} is lower semicontinuous if and only if it is sequentially
lower semicontinuous.

9
 This theorem shows in particular that for metric spaces lower semicontinuity is
equivalent to sequentially lower semicontinuity. We will often deal with weak topolo-
gies of normed spaces. These spaces are neither first countable nor metrizable. How-
ever, it will be enough for our purposes to use only the sequential characterization of
lower semicontinuity.

Remark 2.37. Every extended real-valued continuous function f : X R {} is

lower semicontinuous.

Lemma 2.38. Let X be a topological space, and let f : X R {} be lower

semicontinuous, and c > 0 . Then cf is also lower semicontinuous.

Proof. Let xk x be a convergent sequence in X. Then lim inf k f (xk ) f (x).

Multiply this inequality by c and use the fact that c lim inf k f (xk ) = lim inf k [cf (xk )]
for each k. We conclude that lim inf k [cf (xk )] cf (x).

Lemma 2.39. Let X be a topological space, and let f, g : X R {} be lower

semicontinuous. Then max{f, g}, min{f, g}, and f +g are also lower semicontinuous.

Proof. Let xk x be a convergent sequence in X. Since max{f, g} f, g and by

monotonicity of lim inf we have

lim inf [max{f, g}(xk )] lim inf f (x) and

k k

lim inf [max{f, g}(xk )] lim inf g(x).

k k

Then lim inf k [max{f, g}(xk )] lim inf k [max{f + g}(x)].

On the other hand

lim inf [(f + g)(xk )] = lim inf [f (xk ) + g(xk )]

k k

lim inf f (xk ) + lim inf g(xk ) (superadditivity of lim inf)

k k

f (x) + g(x) = (f + g)(x).

10
Proposition 2.40. Let U be a open subset of a topological space X. Then the char-
acteristic function U : X R is lower semicontinuous.

Lemma 2.41. A lower semicontinuous function on a compact set K assumes its

infimum there.

11
Chapter 3: Functional Analysis

3.1 Metric Spaces

Definition 3.1. A metric space is a set M equipped with a distance function

d : M M [0, ) which satisfies the following properties:

(M1 ) d(x, y) = 0 if and only if x = y,

(M2 ) x, y M d(x, y) = d(y, x), (3.1)
(M3 ) x, y, z M d(x, z) d(x, y) + d(y, z).

The inequality in (M3 ) is the triangle inequality.

Remark 3.2. We will use the following notation and definitions: Rn is the real n-
dimensional Euclidian space. A nonempty, bounded, connected open subset U of
Rn is called a domain. A typical point in Rn is x = (x1 , x2 , . . . , xn ). The i-th
standard coordinate vector is ei = (0, . . . , 0, 1, 0, . . . , 0).

The Euclidean space Rn is a metric space with a distance defined by

d(x, y) = ( ni=1 (xi yi )2 )1/2 where x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn ).
P

Remark 3.3. The function space C(K) = {f : K R | f is continuous on K},

where K is a compact subset of Rn , is a metric space with a distance is defined by
d(f, g) := maxxK |f (x) g(x)|.

Proposition 3.4. If x, y, z M , then |d(x, z) d(y, z)| d(x, y).

Definition 3.5. Let M be a metric space and x M . The open ball in M with
center x and radius  > 0 is defined by B(x, ) := {z X : d(z, x) < }.

Definition 3.6. The topology generated by the collection of all open balls of a metric
space X is called the metric topology of X.

12
Remark 3.7. Given a subset E of a metric space X, E has empty interior in X if and
only if its complement, X \ E, is dense in X.

Remark 3.8. If E is an open subset of a metric space X, for each point x E, there
exist an open ball centered at x whose closure is contained in E.

Definition 3.9. Let X be a metric space. A sequence {uk }

k=1 X converges to

u X if  > 0 N N such that k N d(uk , u) < . We write uk u

meaning limk uk = u which is equivalent to limk d(uk , u) = 0.

Definition 3.10. Let X be a metric space. A sequence {uk }

k=1 X is called a

Cauchy sequence if  > 0 N N such that k, l N d(uk , ul ) < .

Definition 3.11. We say that a metric space X is complete if every Cauchy sequence
in X converges to a point in x.

Proposition 3.12. Let E a subset of a complete metric space X. Then the metric
space E is complete if and only if E is a closed subspace of X.

Definition 3.13. Let E be a nonempty subset of a metric space (X, d). We define
the diameter of E by
diam E = sup{d(x, y) | x, y E}

Definition 3.14. Let X be a metric space. We say that E X is bounded if it

has a finite diameter.

Proposition 3.15. For metric spaces, compactness and sequential compactness are
equivalent.

For a proof, see [17] Theorem 16, Chapter 9.

Theorem 3.16 (Heine-Borel). U Rn is compact if only if it is closed and bounded.

13
Theorem 3.17 (Bolzano-Weirstrass). Every bounded sequence in Rn has a convergent
subsequence.

Remark 3.18. A compact metric space is separable, and every subspace of a separable
metric space is separable.

Definition 3.19. We say that a countable collection of sets {En }

n=1 is nested if

En+1 En for all n N.

Definition 3.20. A nested sequence {En }

n=1 of nonempty subsets of X is called a

contracting sequence if
lim diam(En ) = 0.
n

Theorem 3.21 (The Cantor Intersection Theorem). Let X be a metric space. Then
X is complete if and only if whenever {Cn }
n=1 is a contracting sequence of nonempty

closed subsets of X, there is a point x X for which

T
n=1 Cn = {x}.

Definition 3.22. A subset E of a metric space X is called nowhere dense if its

closure E has empty interior.

Definition 3.23. A subset E of a metric space X is said to be of the first category

if E is the union of a countable collection of nowhere dense subsets of X. A set that
is not of the first category is said to be of the second category.

Theorem 3.24 (Baire Category Theorem). Let X be a complete metric space. Let
{On }
n=1 be a countable collection of open dense subsets of X. Then the intersection
T
n=1 On is also dense.

Corollary 3.25. Let X be a complete metric space. Let {Cn } n=1 be a countable

collection of closed with empty interior subsets of X. Then the union

S
n=1 Cn also

has empty interior.

14
Corollary 3.26. Let X be a complete metric space and {Cn }n=1 be a countable col-

lection of closed subsets of X. If X =

S
n=1 Cn , then at least one of the Cn s has

nonempty interior.

The Baire Category Theorem also implies the following results:

Remark 3.27. An open subset of a complete metric space is of the second category.

Remark 3.28. In a complete metric space, the union of a countable collection of

nowhere dense sets has empty interior.

Theorem 3.29. Let F be a family of continuous real-valued functions on a complete

metric space X that is pointwise bounded in the sense that for each x X, there is a
constant Mx for which
|f (x)| Mx for all f F.

Then there is a nonempty open subset O of X on which F is uniformly bounded in

the sense that there is a constant M for which

|f | M on O for all f F. (3.2)

Proof. Define the collection of sets En = {x X | |f (x)| n for all f F}

where n = 1, 2, . . . . En is closed because each function f F is continuous, En =
1
T
f F f ((, n]) and because the intersection of any collection of closed sets is
closed. Since F is pointwise bounded, for each x X there is an index n such that
|f (x)| n for all f F. This means that x En . Thus X =
S
n=1 En . Since X is

a complete metric space we can apply the previous corollary to conclude that there
exists at least one n for which En contains an open ball B(x, r). Therefore (3.2) holds
for O = B(x, r) and M = n.

15
3.2 Normed Linear Spaces

Definition 3.30. A real linear space X is an Abelian group under addition (+),
together with a scalar multiplication from R X into X such that for all , R
and x, y X,
(x + y) = x + y, ()x = (x), ( + )x = x + x, and 1x = x.

Definition 3.31. A real linear space X is called a normed linear space if there
exists a mapping k k : X [0, ) such that:
(N1 ) kxk = 0 if and only if x = 0,
(N2 ) x X, R kxk = || kxk, (3.3)
(N3 ) x, y X kx + yk kxk + kyk.
Inequality (N3 ) is the triangle inequality.

Remark 3.32. Every normed space M is a metric space, in which the distance function
d : M M R satisfies d(x, y) = kx yk.

Definition 3.33. A Banach space is a normed linear space that is complete with
respect to the metric induced by the norm.

Remark 3.34. The Euclidian space Rn is a Banach space.

3.2.1 Bounded Linear Operators

Let X and Y be real Banach spaces.

Definition 3.35. A map A : X Y is a linear operator if u, v X, , R

we have
A[u + v] = Au + Av.

Definition 3.36. A linear operator A : X Y is bounded if

kAk := sup{kAukY | kukX 1} < .

This quantity is called the operator norm of A.

16
Remark 3.37. Every bounded linear operator is continuous and every continuous lin-
ear operator is bounded.

Proposition 3.38. Let X and Y be normed linear spaces. Then the collection of all
bounded linear operators from X to Y , L(X, Y ), is a normed linear space.

Proposition 3.39. Let X and Y be normed linear spaces. If Y is a Banach space,

then so is L(X, Y ).

3.2.2 The Hahn-Banach Theorem

Definition 3.40. A functional on a linear space X is a real-valued function T :

X R. A linear functional is a functional T such that for all x, y X, and for all
, R we have
T ( x + y) = T (x) + T (y).

Definition 3.41. A functional p : X [0, ) on a linear space is said to be posi-

tively homogeneous if

p(x) = p(x) for all x X, > 0.

and it is subadditive if

p(x + y) p(x) + p(y) for all x, y X.

Remark 3.42. Any norm on a linear space is both subadditive and positively homo-
geneous.

Theorem 3.43 (The Hahn-Banach Theorem). Let p be a positively homogeneous,

subadditive functional on a linear space X and let Y be a subspace of X on which
there is defined a linear functional for which

p on Y.

Then may be extended to a linear functional on all of X for which p on X.

17
 Let X be a normed linear space. Recall that a linear functional T on X is bounded
if there is an M 0 such that for all x X we have

|T (x)| M kxk.

The infimum of all such M is the norm of T and it is denoted by kT k .

We have the following results

x, y X, |T (x) T (y)| kT k kx yk.

and
kT k = sup{T (x) | x X, kxk 1}.

Proposition 3.44. Let X be a normed linear space. Then the collection of all bounded
linear functionals on X is a linear space on which k k is a norm. This normed space
is called the dual space of X, and this space is denoted by X .

Theorem 3.45. Let X0 be a linear subspace of a normed linear space X. Then each
bounded linear functional on X0 has an extension to a bounded linear functional
on all of X that has the same norm as . In particular, for each x X, there is a
X for which
(x) = kxk and kk = 1.

Theorem 3.46 (The Uniform Boundedness Theorem). Let X be a Banach Space and
Y a normed linear space, and let F be a subset of L(X, Y ). Suppose the family F is
pointwise bounded in the sense that for each x in X there is a constant Mx for which

kT (x)k Mx for all T F.

Then the family F is uniformly bounded in the sense that there is a constant M 0
for which kT k M for all T in F.

18
Proof. For each T F, the real-valued function fT : X R defined by fT (x) = kT xk
is a real-valued continuous function on X. Since this family of continuous functions
is pointwise bounded on X and the metric space X is complete, by Theorem (3.29),
there is an open ball B(x0 , r) in X and a constant C 0 for which

kT (x)k C for all x B(x0 , r) and T F.

Thus, for each T F,

kT (x)k = kT ((x + x0 ) x0 )k kT (x + x0 )k + kT (x0 )k C + Mx0 for all x B(0, r).

C+Mx0
Therefore, setting M = r
, and using the linearity of T we have kT k M for all
T in F.

Theorem 3.47 (BanachSteinhaus Theorem). Let X be a Banach space, Y a normed

linear space, and Tn : X Y a sequence of continuous linear operators. Suppose that,
for each x X, limn Tn (x) exists in Y .
(i) Then the sequence of operators Tn is uniformly bounded.
(ii) The operator T : X Y defined by

T (x) = lim Tn (x) for all x X.

n

is linear, continuous, and

kT k lim inf kTn k.

Proof. T is linear because the pointwise limit of linear operators is linear. By the Uni-
form Boundedness Principle we observe that the sequence {Tn } is uniformly bounded.
Thus lim inf kTn k is finite. By continuity of the norm we have limn kTn (x)k =
kT (x)k. Since kTn (x)k kTn k kxk for all n, we have kT (x)k lim inf kTn k kxk.
Therefore T is bounded and kT k lim inf kTn k.

19
Definition 3.48. Let Y be a Banach space, and let X be a nonempty subset of Y .
(a) X is closed in Y if {xk }
k=1 X and limk xk = y in Y implies that y X.

(b) X is bounded if supxX kxk < .

(c) X is precompact in Y if from every sequence {xk }
k=1 X it is possible to select

a subsequence {xkj }
j=1 which converges in Y .

(d) X is compact if X is precompact and closed.

Definition 3.49. We say that the Banach space X is embedded in the Banach
space Y if
(a) X is a vector subspace of Y , and
(b) There exists a constant C > 0 such that

kukY CkukX u X.

Definition 3.50. Let X and Y be Banach spaces, X Y . We say that X is

compactly embedded in Y , written X Y if
(a) X is embedded in Y , and
(b) each bounded sequence in X is precompact in Y .

Definition 3.51. Given u X, u X we write hu , ui := u (u). The symbol

h , i represents the dual pairing of X and X. We define ku k := sup{|hu , ui| |
kuk 1}, which is the same as the operator norm defined previously for general linear
operators.

20
Chapter 4: Weak Topology

4.1 Weak Topology

If F is a collection of real-valued functions on a set X, the F-weak topology on X is

the weakest topology on X for which each function in F is continuous. We observe
that a sequence {xn } in X converges to x X with respect to the F-weak topology
if and only if
lim f (xn ) = f (x) for all f F. (4.1)
n

Definition 4.1. Let X be a normed linear space. The weak topology induced on X
by the dual space X is called the weak topology on X.

For a normed linear space we frequently call the topology induced by the norm
the strong topology.

If X is a normed linear space and u X we define the linear functional J(u) :

X R by
J(u)[u ] = u (u) for all u X .

Definition 4.2. Let X be a normed linear space. The linear operator J : X (X )

defined by
J(u)[u ] = u (u) for all u X, u X

is called the natural embedding of X into (X ) . The space X is said to be

reflexive if J(X) = (X ) .

We can check that the evaluation functional J(u) is linear and is bounded on X
with kJ(u)k kuk. Also, the operator J : X (X ) is injective and linear and
therefore J(X) is a linear subspace of (X ) .

21
Theorem 4.3. Let X be a normed linear space. Then the natural embedding J :
X X is an isometry.

Proof. By definition of the norm on the dual space, for all u X we have

|(u)| kk kuk for all X .

Thus
|J(u)()| kuk kk for all X .

Therefore J(u) is bounded and kJ(u)k kuk. On the other hand, according to
Theorem (3.45), there is a X for which kk = 1 and J(u)() = kuk. Therefore
kuk kJ(u)k. We conclude that kJ(u)k = kuk, i.e. J is an isometry.

Definition 4.4. Let X be a normed linear space. The weak topology induced on X
induced by J(X) (X ) is called the weak-* topology on X .

For a normed linear space X we have the following inclusions:

(i) weak topology on X strong topology on X,
(ii) weak-* topology on X weak topology on X strong topology on X .

4.2 Weak Convergence

Let X be a Banach space.

Definition 4.5. We say a sequence {uk }

k=1 X converges weakly to u X, written

uk * u, if hu , uk i hu , ui for each bounded linear functional u X .

Proposition 4.6. If uk u, then uk * u.

Proof. Suppose that uk u. Given any u X and using the fact that u is
continuous we have u (uk ) u (u). This means that for any u X hu , uk i
hu , ui. Thus by definition of weak convergence uk * u.

22
Theorem 4.7. Let X be a normed linear space. Then every weakly convergent se-
quence in X is bounded in norm. And, if un * u, then kuk lim inf k kun k.

Proof. Let {un } * u in X. Then J(un ) : X R is a sequence of functionals that

converges pointwise to J(u) : X R because each of the functionals J(un ) is contin-
uous with respect to the weak topology generated by X . The Uniform Boundedness
Theorem implies that {J(un )} is a bounded sequence of linear functionals on X .
Then since the natural embedding J is an isometry, the sequence {un } is bounded.
Now we show the second result. From Theorem 3.45 there exists a functional X
for which kk = 1 and (u) = kuk. Then

|(un )| kk kun k = kun k for all n.

Moreover, because un * u, |{(un )}| converges to |(u)| = kuk. Therefore using the
fact that is continuous we have

kuk = lim |(un )| lim inf kun k.

n

4.3 Hilbert Spaces

Let H be a real linear space.

Definition 4.8. A mapping ( , ) : H H R is called an inner product if

(1) (u, u) 0 u H,
(2) (u, u) = 0 if and only if u = 0,
(3) (u, v) = (v, u) u, v H,
(4) (u, v + w) = (u, v) + (u, w) u, v, w H, , R.

A linear space H together with an inner product on H is called an inner product

space.

23
Notation 4.9. If ( , ) is an inner product, the associated norm is

kuk := (u, u)1/2 .

The Cauchy-Schwarz inequality states

|(u, v)| kuk kvk u, v H.

Definition 4.10. A Hilbert space is an inner product space which is a Banach

space with respect to the associated norm.

Definition 4.11. The elements u, v H are orthogonal if (u, v) = 0. A countable

collection {wk }
k=1 H is called orthonormal if

(wk , wl ) = 0 k 6= l, and kwk k = 1 k.

Definition 4.12. Given an orthonormal collection {wk }

k=1 H, we say that this

collection is an orthonormal basis if for any u H we can write

X
u= (u, wk )wk
k=1

with the series converging in H; that is, every u H can be represented in the inner
product norm by its orthogonal expansion with respect to {wk }
k=1 .

P
Remark 4.13. If {wk } 2
k=1 is an orthonormal basis, then u H, kuk =
2
k=1 (u, wk ) .

This is called Parsevals identity.

4.4 Weak Compactness Theorem

Theorem 4.14 (The Tychonoff Product Theorem). Let {X } be a collection of

compact spaces indexed by a set . Then the Cartesian product X , with the
product topology, also is compact.

For a proof of this theorem see p. 245 in [17].

24
Theorem 4.15 (The Banach-Alaoglu Theorem). If X is a normed linear space and
B = {f X | kf k 1} is the closed unit ball in X , then B is a compact space in
the weak* topology.

Theorem 4.16 (Weak Compactness Theorem). Let X be a reflexive Banach space

and suppose the sequence {uk }
k=1 X is bounded. Then there exists a subsequence

{ukj }
j=1 {uk }k=1 and u X such that ukj * u.

Theorem 4.17 (Mazurs Theorem). A convex, closed subset of a Banach space X is

weakly closed.

25
Chapter 5: Function Spaces

5.0.1 Real-valued Functions

In this section we are going to introduce the framework of some important function
spaces. We consider real-valued functions defined over a specific domain U of the
Euclidean space Rn . For our purposes we always assume that the target space R
has the usual 1-dimensional Euclidean metric topology, in particular this is a Banach
space. Also for the domain space Rn we always consider the usual n-dimensional
Euclidean metric topology, which makes Rn into a Banach space.

Let U Rn , consider a function u : U R, and let x0 U . From our definition

of continuity in topological spaces we find that u is continuous at x0 if  > 0 > 0
such that x U and |x x0 | < |u(x) u(x0 )| < , and u is continuous on U if
it is continuous at every point of U .

Definition 5.1. Let x U Rn , and consider a function u : U R. The partial

u u(x+tei )u(x)
derivative of u at x in the direction ei is defined by xi
= limt0 t
provided
that the limit exists.

u 2u 3u
Notation 5.2. uxi = xi
, u xi xj = xi xj
, uxi xj xk = xi xj xk
, etc.

Notation 5.3. A vector of the form = (1 , 2 , . . . , n ) where each i is a nonneg-

ative integer, is called a multiindex of order || = 1 + 2 + + n .
Given a multiindex , define
|| u(x)
D u(x) = = x11 xnn u
x1 1 xnn
If k is a nonnegative integer,

Dk u(x) := {D u(x) | || = k},

26
is the set of all partial derivatives of u at x of order k. Assigning some ordering to
k
the various partial derivatives, we can consider Dk u(x) as a point in Rn . Note that
1/2
X
|Dk u| = |D u|2 .
||=k

Special Cases:
If k = 1, we can arrange the elements of Du into a vector:
D1 u Du := (ux1 , . . . , uxn )=gradient vector.
If k = 2, we arrange the elements of D2 u in a matrix:

ux1 x1 ux1 xn
D2 u =
.. .

.
uxn x1 uxn xn
D2 u is called the Hessian matrix.

Pn
The Laplacian of u is 4u = i=1 uxi xi = tr(D2 u).

5.0.2 Sequences of Real-Valued Functions

Recall that a sequence {xk } n n

k=1 R converges to x R if  > 0 N N such that

k N |xk x| < . We write xk x meaning limk xk = x.

Definition 5.4. A sequence of functions on U is a sequence {un } in the space of

functions from U to R.

Definition 5.5. Let U Rn , and {un } be a sequence of continuous functions from

U to R. We say that {un } is convergent pointwise on U if there exists a function
u : U R such that x U  > 0 N N such that n > N |un (x)u(x)| < .

Definition 5.6. Let U Rn , and let {un } be a sequence of continuous functions

from U to R. We say that the sequence {un } is pointwise bounded if for each
x U , the sequence of real numbers {un (x)} is bounded.

27
Definition 5.7. Let U Rn be compact, and let {un } be a sequence of continuous
functions from U to R. We say that the sequence {un } is uniformly bounded on
U if M > 0 such that, n N, maxxU |un (x)| < M.

Definition 5.8. Let U Rn , and let {un } be a sequence of continuous functions from
U to R. We say that the sequence {un } is equicontinuous on U if  > 0 > 0
such that x, y U n N, |x y| < |un (x) un (y)| < .

Definition 5.9. Let U Rn , and let {un } be a sequence of continuous functions

from U to R. We say that {un } is a uniformly convergent sequence if there exists
a function u : U R such that
 > 0 N N such that x U n > N |un (x) u(x)| < .

Definition 5.10. Let U Rn , and let {un } be a sequence of continuous functions

from U to R. We say that {un } is a uniformly Cauchy sequence if  > 0 N N
such that x U m, n > N |um (x) un (x)| < .

Remark 5.11. If U Rn is compact, and {un } is a uniformly Cauchy sequence of

continuous functions, then {un } is a uniformly convergent sequence of continuous
functions and the limit function is also continuous. See theorems 7.8, 7.12 in [18].

5.1 Arzela-Ascoli Theorem

The following theorem will be very important in the theory of Sobolev spaces, which
is discussed in the next chapter.

Theorem 5.12. Every uniformly bounded, equicontinuous sequence of real-valued

functions on a compact subset of Rn has a uniformly convergent subsequence.

Proof. Let A be a compact domain in Rn , and let (fn ) be a uniformly bounded,

equicontinuous sequence of functions from A to R.

28
Let B be a countable dense subset of A, which exists because Rn is separable, and
let (bn ) be an enumeration of B.

Since fn is uniformly bounded on A, the sequence of real numbers (fn (x)) is bounded
for any x A. So the sequence of real numbers (f1 (b1 ), f2 (b1 ), f3 (b1 ), . . .) is bounded.

By the Bolzano-Weirstrass Theorem [18] this sequence has a convergent subsequence

(f1,1 (b1 ), f1,2 (b1 ), f1,3 (b1 ), . . .)

By hypothesis the subsequence (f1,1 (x), f1,2 (x), f1,3 (x), . . .) of (fn (x)) is bounded for
any x A, so (f1,1 (b2 ), f1,2 (b2 ), f1,3 (b2 ), . . .) is bounded.

By the Bolzano-Weirstrass Theorem this last subsequence has a convergent subse-

quence
(f2,1 (b2 ), f2,2 (b2 ), f2,3 (b2 ), . . .)

By hypothesis the subsequence (f2,1 (x), f2,2 (x), f2,3 (x), . . .) of (f1,m (x)) is bounded for
any x A, so (f2,1 (b3 ), f2,2 (b3 ), f2,3 (b3 ), . . .) is bounded.

Again by the Bolzano-Weirstrass Theorem this last subsequence has a convergent

subsequence
(f3,1 (b3 ), f3,2 (b3 ), f3,3 (b3 ), . . .)

Repeating the procedure explained above we generate the following collection of sub-

29
sequences of 0 (x) := (fn (x))

1 (x) = (f1,1 (x), f1,2 (x), f1,3 (x), . . .)

2 (x) = (f2,1 (x), f2,2 (x), f2,3 (x), . . .)

3 (x) = (f3,1 (x), f3,2 (x), f3,3 (x), . . .)

.. .
. = .. (5.1)

n (x) = (fn,1 (x), fn,2 (x), fn,3 (x), . . .)

.. .
. = ...

This collection of subsequences has the following properties:

1. n is a subsequence of n1 , n N.

2. n (bn ) = (fn,m (bn )) converges for n fixed and m .

3. The order of the functions on each sequence is preserved. This means that if
j < k and both fn,j and fn,k appear in n , then these functions appear in n1
with fn,j = fn1,i , fn,k = fn1,l , and i < l.

Now we select the diagonal subsequence defined by

(x) := (f1,1 (x), f2,2 (x), f3,3 (x), . . .)

Property 3 implies that if we move upward from the n-th row to the n 1-th row in
the collection of functions in (5.1), then the functions can only move to the right and
never to the left.

Properties 1 and 3 imply that (fn,n (x), fn+1,n+1 (x), fn+2,n+2 (x), . . .) is a subsequence
of n (x) for each n N because the elements of the diagonal subsequence come from
functions in n (x) after the first n 1 functions. Specifically, by property 3, for each
k n, fk,k (x) = fn,m (x) for some m n. Using mathematical induction we can

30
conclude that mn is a subsequence of (fn ).

By Property 2 we know that 1 (x) is convergent on b1 . By properties 1 and 2 2 (x)

is convergent on the points {b1 , b2 }. By properties 1 and 2 3 (x) is convergent on
the points {b1 , b2 , b3 }. Then by mathematical induction we can show that n (x) is
convergent on {bm } for all m n.

Therefore, properties 1, 2, and 3 together imply that

is a subsequence of (fn ) convergent on B A. (5.2)

Next we will use the fact that (fn ) is equicontinuous to show that the subsequence
converges uniformly on A.

The fact that is a subsequence of the original equicontinuous sequence (fn ) im-
plies that itself is also equicontinuous. Formally this means that
 > 0 > 0 such that

(x, y A) (n N) |x y| < |fn,n (x) fn,n (y)| < /3. (5.3)

The number depends on , but it is independent of the choice of x, y A, and n.

For the selected above, for each term of the enumeration bm B, m N we

associate an open ball B (bm ) Rn with radius and center bm . The infinite collec-
tion of open balls C := {B (bm ) | m N} is an open cover of A because B is dense in A.

Since A is compact, C has a finite subcover, so there are finitely many points b1 , b2 , . . . br
B for some r N such that we can write A rj=1

B (bj ).
By the previous conclusion (5.2), (x) is a convergent sequence for each point bj with

31
1 j r . This fact implies that (x) is a Cauchy sequence at each one of these
points.
Then (kr ,j N) such that

m, n > kr ,j |fm,m (bj ) fn,n (bj )| < /3. (5.4)

To avoid the dependence on j we set K = max1jr {kr ,j }.

Thus, we have found that > 0 K N such that m, n > K , x A

|fm,m (x) fn,n (x)| |fm,m (x) fm,m (bj )| + |fm,m (bj ) fn,n (bj )| + |fn,n (bj ) fn,n (x)|,

for some j with 1 j r , by the triangle inequality.

Then, using the inequality (5.3) twice, and inequality (5.4) once we get

|fm,m (x) fn,n (x)| < /3 + /3 + /3 = 

Hence we have found that the sequence , which is a subsequence of (fn ), is a uni-
formly Cauchy sequence on A. Then using the remark (5.11), we conclude that
converges uniformly on A.

5.2 C k (U ) Spaces

In this section are introduced the basic classes of real-valued continuously differen-
tiable functions. These classes of functions are defined on open subsets of the Eu-
clidean metric space Rn . After introducing some metric structure in these classes of
functions, we can use them to make useful approximations in the class of Sobolev
spaces.

Definition 5.13. Let U be an open subset of Rn .

C(U ) C 0 (U ) := {u : U Rn | u is continuous} is the linear space of continuous
real-valued functions on U .

32
Remark 5.14. C(U ) is a linear space under pointwise addition and scalar multiplica-
tion meaning that for any f, g C(U ), U R, and x U we have (f + g)(x) =
f (x) + g(x), (f )(x) = f (x).

Definition 5.15. C(U ) := {u C(U ) | u is uniformly continuous on U }.

Remark 5.16. Recall that a real-valued function u defined on the bounded open set
U Rn is uniformly continuous if and only if it can be extended to a real-valued
continuous function u defined on U .

Definition 5.17. C k (U ) := {u : U R; | u is k-times continuously differentiable} is

the linear space of k-times continuously differentiable real valued functions on U .

Definition 5.18. C k (U ) := {u C k (U ) | D u is uniformly continuous on U, for all ||

k}.

T
Definition 5.19. C (U ) := {u : U R | u is infinitely differentiable} = k=0 C k (U )
is the linear space of infinitely continuously differentiable real-valued functions on U .
We say that a function u C is smooth. C (U ) := k
T
k=0 C (U ).

Definition 5.20. Cc (U ) := {u : U R | u C (U ) and supp(u) b U } is the

linear space of smooth real valued functions on U with compact support. Similarly
defined are the linear spaces Cc (U ), Cck (U ), Cc (U ), Cck (U ), and Cc (U ).

5.3 Measure Theory

5.3.1 Lebesgue Measure

Definition 5.21. A collection M of subsets of Rn is called a -algebra if

(a) , Rn M,
(b) A M implies Ac M,
S T
(c) if {Ak }
k=1 M, then k=1 Ak , and k=1 Ak are also elements of M.

33
Theorem 5.22 (Existence of Lebesgue measure and Lebesgue measurable sets).
There exists a maximal -algebra M and a mapping

| | : M [0, +]

with the following properties:

(a) Every open subset of Rn , and thus every closed subset of Rn , belongs to M.
(b) If B is any ball in Rn , then |B| equals the n-dimensional volume of B.
(c) If {Ak }
k=1 M and the sets {Ak }k=1 are pairwise disjoint, then

[
X
| Ak | = |Ak |.
k=1 k=1

Notation 5.23. The sets in M are called Lebesgue measurable sets and | | is
the n-dimensional Lebesgue measure.

Notation 5.24. If some property holds everywhere on Rn , except for on a measur-

able set with Lebesgue measure zero, we say the property holds almost everywhere,
abbreviated a.e..

Definition 5.25. Let f : Rn R. We say f is a measurable function if for each

open subset U R,
f 1 (U ) M.

Remark 5.26. In particular, if f is a continuous function, then f is measurable. The

sum and product of two measurable functions are measurable. In addition if {fk }
k=1

is a sequence of measurable functions, then lim sup fk and lim inf fk ar also measurable
functions.

Theorem 5.27 (Egoroffs Theorem). Let {fk }

k=1 , and f be measurable functions,

and suppose that fk f pointwise a.e. on A, where A Rn is measurable, and

|A| < . Then for each  > 0 there exists a measurable subset E A such that

34
(a) |A E| , and
(b) fk f uniformly on E.

Theorem 5.28 (Monotone Convergence Theorem). Assume the functions {fk }

k=1

are nonnegative and measurable, with f1 f2 fk fk+1 .

Then
Z Z
lim fk dx = lim fk dx.
Rn k k Rn

5.4 Lp Spaces

Let U be a measurable subset of Rn .

Definition 5.29. Let f : U R be measurable. We define the essential supremum

by ess supU f := inf{C > 0 | f (x) C for almost all x U }.

Definition 5.30. Lp (U ) := {u : U R | u is measurable and kukLp (U ) < }

R
where kukLp (U ) = ( U |u|p dx)1/p and 1 p < .

Definition 5.31. L (U ) := {u : U R | u is measurable and kukL (U ) < }

where kukL (U ) = ess supU |u|.

Definition 5.32. Lploc (U ) = {u : U R | u Lp (V ) for all V U }.

1 1 ap bq
Lemma 5.33 (Youngs Inequality). Let a, b > 0, p, q > 1, p
+ q
= 1, ab p
+ q
.

1
Theorem 5.34 (Holders Inequality). Let p, q 1, p
+ 1q = 1, u Lp (U ), v Lq (U ).
Then uv L1 (U ) and kuvk1 kukp kvkq or explicitly
Z Z Z
1 1
|u(x)v(x)|dx ( |u(x)| dx) p ( |v(x)|q dx) q .
p
U U U

Corollary 5.35 (Minkowskis Inequality). Let 1 p , u, v Lp (U ). Then

u + v Lp (U ) and ku + vkLp (U ) kukLp (U ) + kvkLp (U ) .

35
Theorem 5.36 (Interpolation Inequality for Lp -norms). Assume 1 s r t ,
1/r = /s + (1 )/t for 0 < < 1, and u Ls (U ) Lt (U ). Then u Lr (U ), and
kukLr (U ) kukLs (U ) kuk1
Lt (U ) .

Proposition 5.37. The function space Lp (U ) with 1 p is a Banach space

provided we identify two functions that agree a.e. on U .

Proposition 5.38. The dual of Lp (U ) with 1 p < is (Lp (U )) = Lq (U ) where

1/p + 1/q = 1, 1 < q .

We have the following results: (a) (L (U )) 6= L1 (U ), (b) Lp (U ) is separable if

1 p < , and (c) Lp (U ) is reflexive if 1 < p < .

Theorem 5.39. Let U Rn be a bounded domain, and 1 p q . Then

Lq (U ) Lp (U ). Furthermore

kukLp (U ) CkukLq (U ) , for all u Lq (U ),

1 1 1
where C = |U | p q if q < and C = |U | p if q = .

Theorem 5.40. Let U Rn be a bounded domain. Then uniform convergence implies

convergence in Lp (U ) for any 1 p < .

Proof. Let {uk } Lp (U ) be a sequence of Lp functions that converges uniformly to

a function u defined on U . Then for any  > 0 there exists k 0 N such that

|uk (x) u(x)| <  for all k k 0 and for all x U.

Now

(|u|)p (|uk u| + |uk |)p (Triangle Inequality)

(2 max{|uk u|, |uk |})p

2p (|uk u|p + |uk |p )

36
Then we integrate
Z Z Z
p p p p
|u| dx 2 |uk0 u| dx + 2 |uk0 |p dx
U U U
p p
2  |U | + 2 p
kuk0 kpLp (U ) < .

Thus u Lp (U ).
Then use uniform convergence inequality raised to the p power, integrate, and finally
raise to 1
p
power with k > k 0
Z Z
1 1
( |uk (x) u(x)| dx) p < ( p dx) p ,
p
U U
1
kuk (x) u(x)kLp (U ) < |U | p .

Theorem 5.41. Let U be an open subset of Rn , and 1 p < , and q the conjugate
of p. Suppose that uk * u weakly in Lp (U ) and vk v strongly in Lq (U ). Then
Z Z
lim uk vk dx = uv dx.
k U U

Proof. For each k,

Z Z Z Z Z
uk vk dx uvdx = (vk v)uk dx + vuk dx uvdx.
U U U U U

Since every weakly convergent is strongly bounded there exists a constant C > 0 such
that kuk kLp (U ) C for all k.
Z Z Z Z Z
| uk vk dx uvdx| | (vk v)uk dx| + | vuk dx uvdx| ( Triangle Ineq.)
U U U Z U Z U

Ckvk vkLq (U ) + | vuk dx uvdx|. (Holders Ineq.).

U U
R R
Finally use the fact that limk kvk vkLq (U ) = 0 and limk U
vuk dx = U
uvdx.

Theorem 5.42. If uk u in Lp (U ) then a subsequence {unk } uk converges a.e.

to u.

37
5.5 Convolution and Smoothing

This procedure will be used to construct smooth approximations to given functions.

Notation 5.43. Let U Rn be open, and  > 0. We define

U = {x U | dist(x, U ) > }

Definition 5.44. Define the standard mollifier C (Rn ) by

C exp( |x|211 ) if |x| < 1


(x) =
0 if |x| 1,
R
such that supp() B(0, 1) and C > 0 yields Rn dx = 1.
1
For each  > 0, we define  (x) := n
( x ).
The functions  are in C and satisfy
R
Rn
 (x)dx = 1, supp( ) B(0, ).

Definition 5.45. If f L1loc (U ) we define its mollification as f  :=  f. That is,

R R
f  (x) = U  (x y)f (y)dy = B(0,) (z)f (x z)dz, x U . The last equality is
obtained by using the change of variable x y = z and the fact that the support of
 is contained in the ball B(0, ).

Theorem 5.46 (Properties of mollifiers).

(a) f  C (U ),
(b) f  f a.e. as  0,
(c) If f C(U ), then f  f uniformly on compact subsets of U.
(d) If 1 p and f Lploc (U ), then f  f in Lploc (U ).

These properties are used to show the existence of a partition of unity

Theorem 5.47. Let U Rn be a domain covered by a family {Vi }iI of open sets i.e.
U iI Vi . Then there exists a partition of unity i.e a family {j }jJ Cc (Rn )
S

such that
(a) For each j J, supp(j ) Ui for some i I,
(b) 0 j (x) 1 for all j J, for all x Rn ,

38
 P
(c) jJ j (x) = 1 for all x U ,
(d) For each compact set K U , there are only finitely many j that are not identi-
cally zero on K.

39
Chapter 6: Sobolev Spaces

6.1 The Sobolev Function Space W k,p(U )

Definition 6.1. Let u, v L1loc (U ). For a multiindex , we say that v is the -weak
derivative of u on U , written D u = v, if U uD dx = (1)|| U vdx for all test
R R

functions Cc (U ).

Definition 6.2. Suppose p R such that 1 p , and k Z such that k 0.

The Sobolev space W k,p is defined as
W k,p (U ) = {u Lp (U ) | D u exists in the weak sense and is in Lp (U ) || k},
R
with the norm kukW k,p (U ) = ( ||k U |D u|p )1/p if 1 p < and kukW k, (U ) =
P

P
||k ess supU |D u|.

We identify functions in W k,p (U ) which agree a.e.

k,p
Definition 6.3. Wloc (U ) = {u | u W k,p (V ) for all V U }.

Definition 6.4. W0k,p (U ) is the closure of Cc (U ) in W k,p (U ). This means that

u W0k,p (U ) if and only if there exists a sequence {um } Cc (U ) such that um u
in W k,p (U ).

Remark 6.5. If k = 0 we have W 0,p (U ) = Lp (U ). If p = 2 and 0 k we write

H k (U ) = W k,2 (U ). H k is a Hilbert space with a inner product given by (u, v)H k =
P R 0 2
||k U (D u(x) D v(x)dx. In particular H (U ) = L (U ).

Theorem 6.6 (Properties of weak derivatives). Let u, v W k,p (U ), k| k. Then

1. D u W k|| (U ) and D (D u) = D (D u) = D+ for all , with ||+||

k.

40
 2. For all a, b R, au + bv W k,p (U ) and D (au + bv) = aD u + bD v

3. If Cc (U ), then u W k,p (U ) and D (u) !

P
!()! D D u.

Proposition 6.7. For all k 0 and for all 1 p , the Sobolev space W k,p (U )
is a Banach space which is separable for 1 p < and reflexive for 1 < p < .

Definition 6.8. Let U Rn be bounded domain and let k N. We say that the
boundary U is C k if for each point x0 U there exist r > 0 and a C k function
f : Rn1 R such that

U B(x0 , r) = {x B(x0 , r) | xn = f (x1 , x2 , , xn1 )}.

Theorem 6.9 (Global approximation by functions smooth up to the boundary).

Assume U is bounded and U is C 1 . Suppose u W k,p (U ) for some 1 p < and
for any k 1. Then there exist functions um C (U ) such that

um u in W k,p (U ). (6.1)

Theorem 6.10 (Extension Theorem). Assume U is bounded and U is C 1 . Select a

bounded open set V such that U V . Then there exists a bounded linear operator

E : W 1,p (U ) W 1,p (Rn ), (6.2)

and a constant C depending only on p, U and V such that for each u W 1,p (U ) :
(a) Eu = u a.e. in U ,
(b) Eu has support within V ,
(c) there exist a constant C > 0 that depend only on p, U , and V such that
kEukW 1,p (Rn ) CkukW 1,p (U ) .

Definition 6.11. We call Eu an extension of u to Rn .

41
Theorem 6.12 (Trace Theorem). Let U Rn be a bounded domain, and U C 1 .
Then there exists a bounded linear operator

T : W 1,p (U ) Lp (U ), (6.3)

and a constant C depending only on p and U such that

(a) T u = u|U if u W 1,p (U ) C(U ),
(b) for each u W 1,p (U ), kT ukLp (U ) CkukW 1,p (U ) .

Definition 6.13. We call T u the trace of u on U .

Theorem 6.14 (Trace-zero functions in W 1,p ). Let U Rn be a bounded domain,

and U is C 1 , and let u W 1,p (U ). Then

u W01,p (U ) if and only if T u = 0 on U. (6.4)

6.2 Sobolev Inequalities

np
Definition 6.15. If 1 p < n, the Sobolev conjugate of p is p := np
.

Theorem 6.16 (Gagliardo-Nirenberg-Sobolev inequality). Assume 1 p < n. There

exists a constant C = C(p, n) such that for all u Cc1 (Rn )

kukLp (Rn ) CkDukLp (Rn ) . (6.5)

Theorem 6.17 (Estimates for W 1,p , 1 p < n). Let U Rn be a bounded domain, and

suppose U is C 1 . Assume 1 p < n, and u W 1,p (U ). Then u Lp (U ), and
there exists a constant C = C(p, n, U ) with estimate

kukLp (U ) CkukW 1,p (U ) . (6.6)

Theorem 6.18 (Estimates for W01,p , 1 p < n). Let U Rn be a bounded domain.
Suppose u W01,p (U ) for some 1 p < n. Then we have the estimate with a constant
C = C(p, q, n, U )
kukLq (U ) CkDukLp (U ) (6.7)

42
for each q [1, p ].
In particular, for all 1 p ,

kukLp (U ) CkDukLp (U ) . (6.8)

This estimate is called Poincares inequality.

6.3 Rellich-Kondrachov Compactness Theorem

The Gagliardo-Nirenberg-Sobolev Inequality (6.16) states the embedding of W 1,p (U )

into Lp (U ) for 1 p < n. Now we will prove another result that shows that W 1,p (U )
is compactly embedded in Lq (U ) for 1 q < p .

Theorem 6.19 (Rellich-Kondrachov Compactness Theorem). Let U be a bounded,

open subset of Rn , such that U is C 1 , and 1 p < n. Then

W 1,p (U ) Lq (U ), q such that 1 q < p .

Proof. 1. Theorem (6.17) implies that W 1,p (U ) Lq (U ) with kukLq (U ) CkukW 1,p (U )
for all 1 q < p . This means that W 1,p (U ) is embedded in Lq (U ). We there-
fore only need to show that the embedding is compact, i.e. if {um }
m=1 is a

bounded sequence in W 1,p (U ), then there exists a subsequence {umj }

j=1 which

converges in Lq (U ).

2. By the extension theorem (6.10), and without lost of generality we can assume
that all the functions {um }
m=1 have compact support in some bounded open

set V Rn . We may also assume that

sup kum kW 1,p (V ) < . (6.9)

m

3. Let > 0 and consider the smoothed functions

um := um m N

43
 Here, is the standard mollifier. We may also assume that for sufficiently
small the functions {um }
m=1 all have support in V because supp(um ) supp(um )+

supp( ).

4. First, we make the following claim:

um um in Lq (U ) as 0, uniformly in m. (6.10)

To prove this claim, note that if um is smooth, then

Z

um (x) um (x) = (y)(um (x y) um (x))dy,
B(0,1)
(using the first fundamental theorem of calculus)
Z Z 1
d
= (y) (um (x ty)dtdy,
B(0,1) 0 dt
(using the chain rule)
Z Z 1
= (y) Dum (x ty) y dtdy.
B(0,1) 0

Now using the property of Lebesgue integrals of bounded measurable functions

R R
f on bounded domains, | V f | V |f |.
Z Z Z 1Z

|um (x) um (x)|dx (y) |Dum (x ty)|dxdtdy,
V B(0,1) 0 V
(setting z = x ty and integrating over y, z
Z
|Dum (z)|dz.
V

We have found that

kum um kL1 (V ) kDum kL1 (V )

CkDum kLp (V ) (because V is bounded).

R
For um W 1,p (V ) the integral V
|Dum (z)|dz is finite for all m.
By (6.9) we have

um um in L1 (V ), uniformly in m. (6.11)

44
 Then, since 1 q < p , we see using the interpolation inequality for Lp -norms
(5.36) that

kum um kLq (V ) kum um kL1 (V ) kum um k1

Lp (V )
,

1 1
where is chosen so that q
= 1
+ p
, 0 < < 1.
Now inequality (6.9) and the Gagliardo-Nirenberg-Sobolev Inequality (6.16) im-
ply that
kum um kLq (V ) Ckum um kL1 (V ) ,

hence result (6.11) implies (6.10).

5. Next we make the following second claim:

For each fixed > 0, the sequence {um }

m=1 is uniformly bounded and equicontinuous.

(6.12)
In fact, if x Rn , then
Z

|um (x)| (x y)|um (y)|dy,
B(x,)
k kL (Rn ) kum kL1 (V )

(because kum kL1 (V ) is bounded, whenever V is bounded)

C
< ,
n

for all m N. This result means that the sequence {un } is uniformly bounded
In the same way
Z
|Dum (x)| |D (x y)||um (y)|dy,
B(x,)
C
kD kL (Rn ) kum kL1 (V ) < ,
n+1
for all m N. This result means that the sequence {un } is equicontinuous.
These two estimates imply (6.12).

45
6. Now fix > 0. We can show that there exists a subsequence {umj }
j=1

{um }
m=1 such that

lim sup kumj umk kLq (v) . (6.13)

j,k

7. To show this, we use the claim (6.10) to select such that

m N, kum um kLq (V ) /2. (6.14)

8. Since the functions {um }

m=1 and {um }m=1 have support in a fixed bounded set

V Rn , we can use the claim (6.12) and the Arzela-Ascoli theorem, (5.12), to
obtain a subsequence {umj }
j=1 {um }m=1 which converges uniformly on V . In

particular this subsequence is a Cauchy sequence therefore

lim sup kumj umk kLq (V ) = 0. (6.15)

j,k

Now results (6.14), (6.15), together the triangle inequality imply that

lim sup kumj umk kLq (V ) .

j,k

so we have proved (6.13).

9. We use (6.13) with = 1, 12 , 31 , . . .. Then use a standard diagonal argument to

extract a subsequence {umj }
j=1 {um }m=1 satisfying

lim sup kumj umk kLq (v) = 0.

j,k

46
Chapter 7: Variational Analysis in C

7.1 Variational Problem

Suppose that a real-valued function f : [a, b] R R is given. By definition, f has

a local minimum at x0 [a, b] if and only if there exists a neighborhood U (x0 ) of x0
such that f (x) f (x0 ) for all x U (x0 ), where x 6= x0 .

If f has a local minimum at x0 , assuming that x0 is an interior point of [a, b], and
the derivative f 0 (x0 ) exists, then f 0 (x0 ) = 0.

We will generalize this extremal problem from real calculus to variational calculus.
Given a functional F : D X R where X is a Banach space. If F has a local

minimum at u0 , and u0 D, then we have an operator equation F 0 (u0 ) = 0. The
generalized derivative F 0 (x0 ) is a Frechet derivative.

A variational problem is a generalization of extremal problems in real calculus

where the domain D, also called the admissible set, is a space of functions. Since
spaces of functions are infinite dimensional we need the framework of functional anal-
ysis to characterize the problem.

The functional F is given by an integral involving the unknown function and

its first derivative, and it is not necessarily linear. The generalized derivative is
determine by calculating the first variation of the functional, and if the function is
sufficiently smooth yields a second-order PDE called the Euler-Lagrange Equation of
the variational problem.

In this chapter, we discuss the simple case where the domain of the functional
is the space of smooth functions that satisfying a Dirichlet boundary condition on a
fixed boundary. The technique to calculate the first variation is explained explicitly.

47
7.2 Euler-Lagrange Equation

Definition 7.1. Let U Rn be a nonempty, connected, open set. We call U a

domain.

Definition 7.2. Let U Rn be a bounded domain with smooth boundary U , and

let L : Rn R U R be a smooth function. We call L a Lagrangian.

Definition 7.3. Given a fixed smooth function g : U R, we define the set of

admissible functions

A = {w C (U ) | w|U = g|U }

This set of admissible functions corresponds to a Dirichlet boundary condition.

Notation 7.4. L = L(p, z, x) = L(p1 , . . . , pn , z, x1 , . . . , xn ) for p Rn , z R, and

x U . Here, if w is any function in A, p is the variable that represents the gradient
vector Dw = (wx1 , . . . , wxn ), z is the variable that represents w(x). We also we use
the following short notations for the derivatives of L Dp L = (Lp1 , . . . , Lpn ), Dz L =
Lz , Dx L = (Lx1 , . . . , Lxn ).

Definition 7.5. The functional I[] is defined by

Z
I[w] = L(Dw(x), w(x), x)dx (7.1)
U

for w A.

Assume that some particular smooth function u is a minimizer of I[] and satisfies
the boundary condition u|U = g|U . We will show that u is a solution of a specific
nonlinear PDE. Take any smooth function v Cc (U ) and define the function i :
R R by
i( ) := I[u + v]. (7.2)

48
Since v has compact support in U we have v|U = 0. Then (u+ v)|U = u|U + v|U =
u|U = g, so that u + v is admissible for any . Therefore, since u is a minimizer
of I[], i() has a minimum at = 0. Therefore = 0 is a critical point of this
one-variable function, i.e. the necessary condition

i0 (0) = 0, (7.3)

must be satisfied. Writing out i( ) explicitly we have

Z
i( ) = L(Du + Dv, u + v, x)dx. (7.4)
U

Calculating the derivative i0 ( ), using the fact that is a parameter inside the integral
and applying the chain rule, we obtain
Z Xn
0
i ( ) = [ Lpk (Du + Dv, u + v, x)vxk + Lz (Du + Dv, u + v, x)v]dx. (7.5)
U k=1

Setting = 0 and substituting into (7.3), we get

Z Xn
0
i (0) = [ Lpk (Du, u, x)vxk + Lz (Du, u, x)v]dx = 0. (7.6)
U k=1

Now, integrating by parts on the first term and using the fact that v has compact
support within U , we get
Z n
X
[ (Lpk (Du, u, x))xk + Lz (Du, u, x)]vdx = 0. (7.7)
U k=1

Since this equality holds for arbitrary test functions v, we conclude that u satisfies
the following second-order, possibly nonlinear PDE
n
X
(Lpk (Du, u, x))xk + Lz (Du, u, x) = 0. (7.8)
k=1

This is the Euler-Lagrange equation associated with the functional I[] defined in (8.1).
Any smooth minimizer of I[] is a solution of the Euler-Lagrange equation (7.8), and
thus we can try to find a solution of equation (7.8) by searching for minimizers of

49
(8.1). But there may be other, non-minimizing solutions to the PDE and also the
minimizer may not be smooth. It is important to point out that the Euler-Lagrange
Equation is a possibly quasilinear second order PDE in divergence form.

One important example related with problems in physics is the Poisson equation.
Consider L(p, z, x) = 12 |p|2 zf (x). Then Lpi = pi with i = 1, . . . , n, Lz = f (x).
The functional is
Z
1
I[w] := [ |Dw|2 wf ]dx,
U 2

and the associated Euler-Lagrange equation is M u = f in U.

50
Chapter 8: Variational Analysis in W 1,q

Let U Rn be a bounded domain such that U is C 1 , let the Lagrangian function

L be smooth, and let the admissible set of functions A be a subset of some appropriate
Sobolev space. Take also the same kind of functional discussed in previous chapter
Z
I[w] = L(Dw(x), w(x), x)dx (8.1)
U

for w A.

As we anticipated in the introduction in order to have a existence proof for min-

imizers we will need three key requirements: compactness, closedness, and lower
semicontinuity. When these requirements are satisfied we have available powerful
weak convergence methods to guarantee existence of minimizers.

We have some reasons why we want to define the functional over a Sobolev space.
First, we know that Sobolev spaces are more general than the space of smooth func-
tions because it includes discontinuous functions. Second, Sobolev spaces are reflexive
Banach spaces so that we can use the corresponding weak sequential compactness the-
orem which is a very important property to prove existence of minimizers.

We will need additional conditions, other than smoothness, on the Lagrangian L.

We will identify some specific conditions on the Lagrangian L that ensures that the
functional I[] defined over a Sobolev spaces indeed has a minimizer.

8.1 Coercivity

We know that in general arbitrary smooth functionals do not necessarily attain a

global minimum or even a local one. This problem becomes clear if we think of

51
 2
smooth functions such as f (x) = ex . The problem happens in finite dimensions
only when the smooth function has horizontal asymptotes. For this reason we need
to control from below the growth of the functional I[w] for large functions w. One
way to do this is to require that I[w] as |w| . We will assume that for any
q R with 1 < q < there exist constants > 0, 0 such that

L(p, z, x) |p|q , (8.2)

for all p Rn , z R, x U. This is called a coercivity condition on L. Then

integrating over U and using monotonicity property of integral we get

I[w] kDwkqLq (U ) (8.3)

for := |U |. Thus I[w] as kDwkLq .

We say that (8.3) is a coercivity condition on I[w].

The crucial fact that U is bounded, implies that is finite i.e 0 < . Then
inequality (8.3) implies that I[w] is bounded below i.e < I[w]. On the
other hand we know, by Poincares Theorem, that the Lq -norm of Dw controls the
Lq -norm of w. These two norms contributes to the norm of W 1,q (U ).

These results suggests that we can define the functional I[w] for functions w
W 1,q (U ) and not only for smooth functions as we did in the previous chapter. Fur-
thermore we can also require a Dirichlet boundary condition w|U = g|U that must
be interpreted now in the trace sense.

The class of admissible functions is defined by

A := {w W 1,q (U ) : w|U = g|U in the trace sense}.

8.2 Lower Semicontinuity

By the coercivity condition, I(w) is bounded below, so we can set

m := inf I[w]. (8.4)

wA

52
Using the definition of infimum we can select a sequence of functions uk A so that

I[uk ] m as k . (8.5)

We call {uk }
k=1 a minimizing sequence. We want to show that some subsequence of

{uk }
k=1 converges to an actual minimizer u A. To do this it is necessary to use a

kind of sequential compactness of the admissible set A. This condition is not possible
using only norm convergence because the space W 1,q (U ) is infinite dimensional. Recall
that in infinite dimensional spaces the unit closed ball is not necessarily compact.

The inequality (8.3) of the coercivity condition will only allows us to conclude
that the minimizing sequence is contained in a bounded subset of W 1,q (U ). This does
not imply that there exists any subsequence of {uk }
k=1 which actually converges in

the W 1,q (U ) norm topology.

However, we can recover enough sequential compactness if we consider the weak

topology. We know that when 1 < q < , the space Lq (U ) is a reflexive Banach
space. Recall that the weak compactness theorem (4.16) states that if a sequence in
a reflexive Banach space is bounded, then there exists a subsequence that converges
weakly to some limiting element in the given Banach space. Using this theorem for
our sequence {uk }
k=1 and our Banach space W
1,q
(U ) which is reflexive, we conclude
that there exist {ukj }
j=1 {uk }k=1 and a function u W
1,q
(U ) such that

ukj * u weakly in Lq (U )

Dukj * Du weakly in Lq (U ; Rn ). (8.6)

We abbreviate this result by saying

ukj * u weakly in W 1,q (U ). (8.7)

Now we observe that we have another potential problem because in almost interesting
applications the functional I[] is not continuous with respect to weak convergence.

53
This means that from (8.7) and (8.5) we can not conclude that

I[u] = lim I[ukj ], (8.8)

j

meaning that u is a really a minimizer. The reason is that in general Dukj * Du

does not imply Dukj Du in norm. However, is not necessary to require equality in
(8.8). It is enough to know only that

I[u] lim inf I[ukj ]. (8.9)

j

From this inequality and (8.5) we can deduce I[u] m. On the other hand (8.4)
implies I[u] m. Thus I[w] = m. This means that, if the lower semicontinuity
condition (8.9) holds u will indeed be a minimizer.

Definition 8.1. We say that a functional I[] is sequentially weakly lower semicon-
tinuous on W 1,q (U ), if

uk * u weakly in W 1,q (U ) implies that I[u] lim inf I[uk ]. (8.10)

j

The next objective is to determine reasonable conditions on the Lagrangian L that

ensure that I[] is weakly sequentially lower semicontinuous.

8.3 Convexity

To determine an absolute minimum of I[w] we need a global condition analogous

to the positivity of the Hessian used in extremal problems of finite dimensional real
analysis. This argument suggest that it is reasonable to require that L be convex in
its first argument p.

Definition 8.2. A function f : Rn R is called convex if f (x + y) f (x) +

f (y) for all x, y Rn , , > 0 such that + = 1.

54
Theorem 8.3 (Supporting hyperplanes). Suppose f : Rn R is convex. Then for
each x Rn there exists r Rn such that

f (y) f (x) + r (y x) y Rn . (8.11)

Remark 8.4. (a) The mapping y 7 f (x) + r (y x) determines the supporting

hyperplane to f at x. The inequality above says that the graph of f lies above each
supporting hyperplane. If f is differentiable at x, r is unique and r = Df (x).
(b) If f is C 2 , then f is convex if and only if D2 f 0.

The following theorem proves that a convexity condition on L implies weak lower
semicontinuity which is a crucial result that will be used in the proof of the existence
of a minimizer in the next section.

8.4 Weak Sequential Lower Semicontinuity Theo-

rem

Theorem 8.5 (Weak Sequential Lower Semicontinuity Theorem). Assume that L is

smooth and bounded below, and in addition the mapping p 7 L(p, z, x) is convex,
z R, x U . Then I[] is weakly lower semicontinuous on W 1,q (U ).

Proof. 1. Take any sequence {uk }

k=1 such that uk * u weakly in W
1,q
(U ), and let
l := lim inf k I[uk ]. We want to show that I[u] l. Note that the sequence
{uk } is not necessarily convergent strongly in W 1,q (U ) and it is not necessarily
a minimizing sequence.

2. From the functional analysis result (4.7) we know that any weakly convergent
sequence is bounded in norm. Thus, we have

sup kuk kW 1,q (U ) < . (8.12)

k

55
 From the definition of lim inf we know that l is the smallest cluster point of the
real sequence {I[uk ]}. We also know that if a sequence converges, then its limit
and its liminf coincide. Therefore, using a subsequence of uk if necessary, we
can suppose
l = lim I[uk ]. (8.13)
j

3. We know from the Rellich-Kondrachov compactness theorem (6.19) that from

any bounded sequence in W 1,q (U ) we can extract a subsequence {ukj } that
converges strongly on Lq (U ). For convenience we will use the notation {uk } for
this subsequence. So we have

uk u, strongly in Lq (U ).

4. Using the Riesz-Fischer Theorem (see [17] page 148) we can, if necessary, extract
another subsequence such that

uk u a.e. in U. (8.14)

5. Fix  > 0. Then using the previous result and Egoroffs Theorem (5.27) we
obtain that there exists a smaller measurable set E U with

|U E | . (8.15)

so that
uk u uniformly on E . (8.16)

To restrict temporarily the limit function and its derivative to be in L (U ) it

is convenient to define another subset of U

1
F := {x U : |u(x)| + |Du(x)| }. (8.17)


Observe that
|U F | 0 as  0, (8.18)

56
 because as  0 the restrictions on the on u and Du are removed. This is a
basic property of measurable functions, see for example [17].
To combine the restriction of uniform convergence and to be inside L1 at the
same time we set
G := E F . (8.19)

From relations (8.15), (8.18), we observe that |U G | 0 as  0.

6. Let  > 0 be fixed. By the coercivity condition (8.2), we know that L is bounded
below by . Without lost of generality we can consider L 0. Otherwise set
= L + . So
L
Z Z
I[uk ] = L(Duk , uk , x)dx L(Duk , uk , x)dx
U G
Z Z
L(Du, uk , x)dx + Dp L(Du, uk , x) (Duk Du)dx.
G G
(8.20)

The first inequality follows from the monotonicity of Lebesgue integrals.

The second inequality follows from the convexity of L in its first argument using
the smoothness of L and the supporting hyperplane theorem (8.3).

7. From results (8.16), (8.17), and (8.19), and using the fact that for uniformly
convergent sequences the limit of the sequence of integrals is equal to the integral
of its limit, we see that
Z Z
lim L(Du, uk , x)dx = L(Du, u, x)dx. (8.21)
k G G

The result (8.17) is required to maintain the integrand on the left-hand side
bounded, thus integrable, and to make L uniformly continuous in its arguments.

8. Duk * Du weakly in Lq (U ; Rn ) because uk * u weakly in W 1,q (U ). The

expression Dp L(Du, uk , x) Dp L(Du, u, x) converges uniformly on G because

57
 L is smooth and Dp L is uniformly continuous on bounded functions. Then by
result (5.4) the expression converges strongly in Lp (G ) for any 1 p < ; in
particular it converges in the dual of Lq ((G )). Thus from (5.41)
Z
lim Dp L(Du, uk , x) (Duk Du)dx = 0. (8.22)
k G

Substituting results (8.21), and (8.22) into (8.20) we obtain

Z
l = lim I[uk ] L(Du, u, x)dx. (8.23)
k G

9. Finally letting  0 and using the monotone convergence theorem (5.28) we

conclude that
Z
l L(Du, u, x)dx = I[u].
U

Observe that this result is independent of the specific admissible set A used i.e.
the boundary condition of the problem.

8.5 Existence of minimizers Theorem

This following theorem assumes the same hypothesis on the domain U as well as on
the lagrangian L, smoothness, and coercivity conditions of L, and the same definition
of functional I[] over the same admissible set A.

Theorem 8.6 (Existence of Minimizer Theorem). Assume that L satisfies the coer-
civity inequality (8.2) and is convex in the variable p. Suppose also that the admissible
set A is nonempty. Then there exists at least one function u A solving

I[u] = min I[w].

wA

58
Proof. 1. We know that the functional I[] is bounded below by the coercivity
condition. Since A is not empty by hypothesis, the set {I[w] | w A} is not
empty and bounded below. Thus the infimum of this set exists, and we can set
m := inf wA I[w], with < m +. If m = + we are done, we do not
need to prove nothing. So we assume that m is finite. Meaning < m < .

2. Select a minimizing sequence {uk }

k=1 W
1,q
(U ), which by definition satisfies

I[uk ] m as k . (8.24)

Observe that we only say that the real sequence {I[uk ]} converges to a limit m
in R, and the minimizing sequence {uk } does not yet necessarily converge either
strongly or weakly in W 1,q (U ).

3. Without lost of generality, we can take the constant in L be zero, since

:= L + used before. Thus using
otherwise we can make the same shifting L
the coercivity condition, we have L |p|q with > 0, so
Z
I[w] |Dw|q dx. (8.25)
U

Since m is finite we conclude from (8.24) and (8.25) that

sup kDuk kLq (U ) < . (8.26)

k

because the sequence of real numbers {I[uk ]} is convergent, thus bounded, and
I[uk ] controls the norm kDuk kLq (U ) for each k.

4. Fix any function w A. Since uk and w are both equal to g on U in the

trace sense, we have uk w W01,q (U ). Therefore using the triangle inequality,
(8.25), the fact that w A W 1,q (U ), the fact that U is a bounded set in Rn ,
and the Poincares inequality (6.8) we have

kuk kLq (U ) kuk wkLq (U ) + kwkLq (U )

C1 kDuk DwkLq (U ) + C2 C3 ,

59
 where C1 , C2 , C3 are real positive constants. Hence supk kuk kLq (U ) < . This es-
timate together with the finiteness (8.25) imply {uk }
k=1 is bounded in W
1,q
(U ).

5. Now we recall the important fact that W 1,q (U ) is a reflexive Banach space.
Applying the weak compactness theorem for W 1,q (U ), there exist a subsequence
{ukj }
j=1 {uk }k=1 and a function u W
1,q
(U ) such that

ukj * u weakly in W 1,q (U ). (8.27)

6. Now the next important step to do is to check that u A. This is the closedness
condition in our proof. To see this, note that for w A as above, uk w
W01,q (U ).
Now W01,q (U ) is a linear subspace of W 1,q (U ). Every linear subspace is convex.
Since W01,q (U ) is defined as a closure it is closed. Then, by Mazurs theorem
(4.17), W01,q (U ) is weakly closed. Hence u w W01,q (U ). Consequently the
trace of u on U is g. The Dirichlet boundary condition is satisfied in the trace
sense. so u A.

7. In the previous weak lower semicontinuity theorem (8.5), we proved that under
the given conditions,
I[u] lim inf I[ukj ] = m.
j

But since the infimum m is realized, we conclude that

I[u] = m = min I[w]. (8.28)

wA

60
8.6 Weak Solutions

Consider the boundary-value problem for the Euler-Lagrange PDE associated the
functional L
n
X
(Lpi (Du, u, x))xi + Lz (Du, u, x) = 0 in U, (8.29)
i=1
u = g on U. (8.30)

If we multiply (8.29) by a test function v Cc (U ), and integrate by parts, we get

Z X n
[ Lpi (Du, u, x)vxi + Lz (Du, u, x)v]dx = 0. (8.31)
U i=1

Now we assume that the following growth conditions on L and its derivatives hold.

|L(p, z, x)| C(|p|q + |z|q + 1), (8.32)

|Dp L(p, z, x)| C(|p|q1 + |z|q1 + 1), (8.33)

|Dz L(p, z, x)| C(|p|q1 + |z|q1 + 1) (8.34)

for some constant C and all p R, x U .

Under these growth conditions we can verify that the equation (8.31) is valid for
a minimizer u and any v W01,q (U ).

Definition 8.7. We say that u A is a weak solution of the boundary-value

problem (8.29) for the Euler Lagrange equation if
Z X n
[ Lpi (Du, u, x)vxi + Lz (Du, u, x)v]dx = 0. (8.35)
U i=1

for all v W01,q (U ).

Theorem 8.8 (Solution of the Euler-Lagrange equation). Assume L verifies the

growth conditions (8.32,8.33,8.34), and u satisfies

I[u] = min I[w]. (8.36)

wA

Then u is a weak solution of (8.29).

61
Proof. Take v W01,q and set

i( ) := I[u + v] for all R. (8.37)

Using continuity of i( ), the growing conditions (8.32,8.33,8.34), the Youngs inequal-

ity, and the Dominated Convergence Theorem we can verify that
n
i( ) i(0)
Z X
0
i (0) = lim = [ Lpi (Du, u, x)vxi + Lz (Du, u, x)v]dx = 0. (8.38)
i0 U i=1

thus u is a weak solution.

Remark 8.9. In general the Euler-Lagrange equation (8.31) is a only a necessary

condition for minimizers. However, if the joint map (p, z) 7 L(p, z, x) is convex for
each x, then each weak solution is in fact a minimizer.

62
Chapter 9: Free Boundary Problems

9.1 Existence of Minimizers for problems with Free

Boundaries

Let U Rn be a domain. Consider the the following functional with a variable

domain of integration
Z
J[u] = f (x, u(x), u(x))dx.
U {u>0}

We want to minimize J(u) min, with the conditions: u = 0 on := U {u > 0},

u = 0 on U \ {u > 0}. The set is called the free boundary.

If we assume that the Lagrangian function f = f (x, z, p) is smooth and, the free
boundary is smooth, we can calculate the first variation using a Taylors expansion
around the absolute minimum u of J. This is done for example in the book Calculus
of Variations by Gelfand and Fomin. The Euler-Lagrange equation is

fp fz = 0 in N := U {u > 0}, (9.1)

and the free boundary conditions

u = 0, fp u f = 0 in . (9.2)

Two global conditions for an absolute minimum are

f convex in p, f (x, z, 0) 0

Consider the Lagrangian f (x, z, p) := |p|2 + Q2 , where Q = Q(x) is a function.

Using (9.1), and (9.2) we obtain

4u = 0 in N, u = 0, |u| = Q in . (9.3)

63
 If we do not impose any restriction on the class of sets {u > 0} there is an existence
proof done in reference [3]. Here we describe the basic steps of a bounded version of
this proof.

Let U Rn be a bounded domain, U is C 1 . S U is open and has n 1

dimensional Hausdorff measure |S| > 0. The Dirichlet boundary condition on S is
determined by a function w H 1 (U ) such that w 0. Consider the functional
Z
J[u] := (|u|2 + {u>0} Q2 )dx
U

in the admissible set K := {u H 1 (U ) | u = w on S} so K is a closed convex

subspace of H 1 (U ). Observe that the functional J[u] is not differentiable as a function
of u. For this reason we need some technicality to determine, in appropriate sense,
its first variation and boundary conditions on the fixed boundary U and on the free
boundary .

Theorem 9.1 (Existence Theorem for a Free Boundary Problem). There exists an
absolute minimum u K of the functional J.

Proof. We assume that 0 < Q < Qmax < , and the domain U is a bounded. Since
0 J[u] for all u H 1 (U ) we can set the number m := inf{J[u] | u K}. If
m = + we are done. So we assume that m is finite and nonnegative 0 m < .
Then there exists a minimizing sequence {un } K that satisfies J[uk ] m.

64
 R
We can rewrite the functional as J[u] = J1 [u] + J2 [u] where J1 [u] := U |u|2 dx,
R
J2 [u] := U {u>0} Q2 dx. The first term J1 [] satisfies coercivity because its integrand
|p|2 is coercive. The second term J2 [] is bounded because 0 {u>0} Q2 < implies
R
0 U {u>0} Q2 dx < . Adding a bounded functional to a coercive functional the
coercivity is preserved. Thus J[] satisfies coercivity.

Using the same argument in previous existence theorem (8.6), we can have a
sequence {u} such that uk * u weakly in K.

We need to show that J[] is weakly lower semicontinuous. According with the
weak lower semicontinuity theorem (8.5), the first term J1 [] is is weak lower semi-
continuous because its integrand |p|2 is smooth, convex in p and bounded below.

We will show that the second term J2 [] is also weakly lower semicontinuous. Using
the similar steps at the beginning of the proof the weak lower semicontinuity theorem
(8.5) we have uk u pointwise a.e. in U . Now, for any x U , {u>0} (x) > 0 implies
u(x) > 0. So uk (x) > 0 for large enough k. Then fixing such large enough k we have
for almost all x U
{u>0} (x) {uk >0} (x)

because {uk >0} (x) = 1 for k sufficiently large for almost all x U such that
{uk >0} (x) = 1. Hence using monotonicity of lim inf

{u>0} (x) lim inf {uk >0} (x)

k

Multiplying both sides of the last inequality by Q2 (x) and integrating over U , we have
Z Z
2
Q (x){u>0} (x)dx lim inf [Q2 {uk >0} (x)]dx (monotonicity)
U U k Z
lim inf {uk >0} (x)dx (Fatous lemma).
k U

Finally we use the fact that the sum of two weakly lower semicontinuous functionals
is also a weakly lower semicontinuous functional by (2.39). Thus there exists at least
one u K such that J[u] = min{J[v] | v K.

65
 Consider the following example of solution of free boundary problem in two di-
mensions
U = {(x, y) R2 | |x| < 2 and |y| < 1}

and S = {(x, y) U | |x| = 2}, w = 1, Q = 1. The absolute minimum solution is

u(x) = max{0, |x| 1}. In this case we have U = U1 U2 U3 where U1 = {(x, y)
U | 2 < x < 1}, U2 = {(x, y) U | |x| < 1}, U3 = {(x, y) U | 1 < x < 2.
u(x) = 0 in R2 , and u(x) is harmonic in R1 R3 . The Dirichlet data is u(x)|S = 1.
The free-boundary is = 1 2 where 1 = {(x, y) U | x = 1} and 2 =
{(x, y) R2 | x = 1}. The free-boundary conditions are u(x)|S = 0 and u = 1.

9.2 Applications

In this section we are going describe some applications problems that are related with
free-boundary problems. The free-boundary conditions we have considered appears,
for example, in fluid problems such as jets and cavities. To simplify the description
we consider the case of incompressible, irrotational, and inviscid fluids. In this case
the velocity of the fluid is given by a where is the velocity potential. In two
dimensions the harmonic conjugate is called the stream function. and are
related by the Cauchy-Riemann equations x = y , y = x . The stream lines are
determined by = const.

The velocity potential satisfy the Bernoullis law ||2 + 2p = const where p is
the pressure. The free-boundary in this case is the interface between the fluid and the
air. On the free boundary, the pressure is constant and we have that || is constant.
The stream function satisfies

4 = 0 in the fluid, (9.4)

= 0, =c on the free boundary, (9.5)

66
where c is a constant. Using conformal mappings the solution of the two-dimensional
jet problem has been established in the literature.

In three dimensions we can not use conformal methods to solve free boundary
problems. However in the reference [4] they considered axially symmetric jet problem
and used the variational approach to establish existence of solutions. This work was
based on the general existence theorem by Alt and Caffarelli [3].

67
Chapter 10: Conclusion

We have considered a class of boundary value problems that can be reformulated

as a variational problem. This kind of problems includes many important problems
in science and technology. We only used the Lagrangian formulation of variational
problems. This formulation is suitable for elliptic problems or steady state of evolution
problems. The specific method to prove existence of minimizers is the direct method.
One advantage of this method is that we can deal with nonlinear problems. The class
of nonlinear problems are very interesting because it can capture new features that
are not captured by linear models.

The source of nonlinearity we have been motivated by in the present work comes
from the fact that in many cases the domain of the problem is a priori unknown, so
that we have free-boundaries. It turns out that the direct method for existence theo-
rems can also be used successfully for some specific free boundary problems. Two key
concepts are weak sequential compactness and lower semicontinuity in Sobolev spaces.
The goal of the present work is to provide a basic exploration of the mathematical
tools required to begin to study applications problems related to free-boundary prob-
lems. As possible next step it would be interesting to study more advanced aspects of
stationary problems, after that evolution problems for example free-moving problems.

68
 Bibliography

[1] Adams, R. and Fournier, J. Sobolev Spaces, Second Edition. Elsevier Science
Ltd, Oxford, United Kindom, 2003.

[2] Aliprantis, C. and Border, K. Infinite Dimensional Analysis, Third Edition.

Springer, New York, 2006.

[3] Alt, H. and Caffarelli, L. Existence and Regularity for a Minimum Problem with
Free Boundary. J. Reine Angew. Math. 325 (1981), 105-144.

[4] Alt, H. and Caffarelli, L. and Friedman, A. Axially Symmetric Jet Flows. Arch.
Rational Mech. Anal. 81 (1983), 97-149.

[5] Dacorogna, B. Introduction to Calculus of Variations. Imperial College Press,

London, United Kindom, 2004.

[6] Evans, L. Partial Differential Equations, Second Edition. American Mathemati-

cal Society, Providence, Rhode Island, 2010.

[7] Evans, L. and Gariepy, R. Measure Theory and Fine Properties of Functions.
CRC Press, Boca Raton, Florida, 1992.

[8] Fleming, W. Functions of Several Variables. Spinger-Verlag, New York, 1977.

[9] Folland, G. Real Analysis, Second Edition. John Wiley & Sons Inc., New York,
1999.

[10] Friedman, A. Free Boundary Problems in Science and Technology. Notices

Amer. Math. Soc. (2000), 854-861.

69
[11] Friedman, A. Variational Principles and Free Boundary Problems. John Wiley
& Sons Inc., New York, New York, 1982.

[12] Jost, J. and Li-Jost, X Calculus of Variations. Cambridge University Press, Cam-
bridge, United Kindom, 2008.

[13] Leoni, G. A First Course in Sobolev Spaces. American Mathematical Society,

Providence, Rhode Island 2009.

[14] Meyer, R. Introduction to Mathematical Fluid Dynamics. Dover Publications

Inc., Mineola, New York, 2007.

[15] Munkres, J. Topology, Second Edition. Prentice Hall Inc., Upper Saddle River,
New Jersey, 2000.

[16] Raynor, S. Neuman Fixed Boundary Regularity for an Elliptic Free Boundary
Problem. Communications in Partial Differential Equations, 33, (2008), 1975-
1995.

[17] Royden, H. and Fitzpatrick, P. Real Analysis, Fourth Edition. Prentice Hall, Up-
per Saddle River, New Jersey, 2010.

[18] Rudin, W. Principles of Mathematical Analysis, Third Edition. McGraw-Hill

Inc., New York, 1976.

[19] Rudin, W. Functional Analysis, Second Edition. McGraw-Hill Inc., New York,
1991.

[20] Zeidler, E. Nonlinear Functional Analysis and its Aplications III. Springer, New
York, 1985.

70
 Vita

Carlos Tello was born in Lima, Peru. He did his undergraduate study in Peru,
obtaining a BS degree in Physics. In 2001 he completed a PhD degree in Physics in
Brazil. Next he moved to USA and began working as an adjunct faculty teaching
Physics and Mathematics in New Jersey and North Carolina for 5 years. Recently he
finished his graduate study at Wake Forest University for a MA degree in Mathemat-
ics. He also has worked at the Institutional Research Office of this institution. He has
interest in continue his career in applied research and development sector requiring
good mathematical physics background.

71

View publication stats