Book 2011091002

TOPICS ON FUNCTIONAL
ANALYSIS, CALCULUS OF
VARIATIONS AND DUALITY
Fabio Botelho
AP
Academic Publications
1991 Mathematics Subject Classification. 46N10, 49N15

Key words and phrases. Banach spaces, convex analysis, duality,
calculus of variations, non-convex systems, generalized method of lines
Abstract. This work is a kind of revised and enlarged edition
of the title Variational Convex Analysis, published by Lambert
Academic Publishing. First we present the basic tools of analysis necessary to develop the core theory and applications. New
results concerning duality principles for systems originally modeled
by non-linear differential equations are shown in chapters 10 to 18.
A key aspect of this work is that although the original problems
are non-linear with corresponding non-convex variational formulations, the dual formulations obtained are almost always concave
and amenable to numerical computations. When the primal problem has no solution in the classical sense, the solution of dual problem is a weak limit of minimizing sequences, and the evaluation of
such average behavior is important in many practical applications.
Among the results we highlight the dual formulations for micromagnetism, phase transition models, composites in elasticity and
conductivity and others. To summarize, in the present work we
introduce convex analysis as an interesting alternative approach for
the understanding and computation of some important problems in
the modern calculus of variations.
Finally, in chapter 13 we develop a new numerical procedure,
called the generalized method of lines. Through such a method
the domain is discretized in lines (or more generally in curves) and
the solution on them is written as functions of boundary conditions
and boundary shape. In chapter 17 we apply the method to an
approximation of the incompressible Navier-Stokes system.
Author: Fabio Botelho

Address: Department of Mathematics
Federal University of Pelotas
Pelotas, RS, BRAZIL
email: fabio.silva.botelho@gmail.com
c
Academic
Publications, Ltd., 2011
Electronic book: http://www.acadpubl.eu (397 pp.)
ISBN: 978-954-2940-08-1
Contents
Introduction
Summary of Main Results
Acknowledgments
Part 1.
ix
ix
xii
Basic Functional Analysis
Chapter 1. Topological Vector Spaces

1.1. Introduction
1.2. Vector Spaces
1.3. Some Properties of Topological Vector Spaces
1.4. Compactness in Topological Vector Spaces
1.5. Normed and Metric Spaces
1.6. Compactness in Metric Spaces
1.7. The Arzela-Ascoli Theorem
1.8. Linear Mappings
1.9. Linearity and Continuity
1.10. Continuity of Operators on Banach Spaces
1.11. Some Classical Results on Banach Spaces
1.12. Hilbert Spaces
3
3
3
8
11
12
13
20
23
24
25
26
32
Chapter 2. The Hahn-Banach Theorems and

Weak Topologies
2.1. Introduction
2.2. The Hahn-Banach Theorem
2.3. Weak Topologies
2.4. The Weak-Star Topology
2.5. Weak-Star Compactness
2.6. Separable Sets
2.7. Uniformly Convex Spaces
39
39
39
45
47
48
52
53
Chapter 3. Topics on Linear Operators

3.1. Topologies for Bounded Operators
3.2. Adjoint Operators
3.3. Compact Operators
55
55
56
59
iii
iv
CONTENTS
3.4. The Square Root of a Positive Operator

3.5. About the Spectrum of a Linear Operator
3.6. The Spectral Theorem for Bounded
Self-Adjoint Operators
3.7. The Spectral Decomposition of
Unitary Transformations
3.8. Unbounded Operators
3.9. Symmetric and Self-Adjoint Operators
61
66
70
78
81
84
Chapter 4. Measure and Integration

4.1. Basic Concepts
4.2. Simple Functions
4.3. Measures
4.4. Integration of Simple Functions
4.5. The Fubini Theorem
4.6. The Lebesgue Measure in Rn
4.7. Lebesgue Measurable Functions
91
91
93
94
95
99
105
115
Chapter 5. Distributions
5.1. Basic Definitions and Results
5.2. Differentiation of Distributions
125
125
129
Chapter 6. The Lebesgue and Sobolev Spaces

6.1. Definition and Properties of Lp Spaces
6.2. The Sobolev Spaces
6.3. The Sobolev Imbedding Theorem
6.4. The Proof of the Sobolev Imbedding Theorem
6.5. Compact Imbeddings
131
131
139
144
145
166
Part 2.
171
Variational Convex Analysis
Chapter 7. Basic Concepts on the Calculus of Variations

7.1. Introduction to the Calculus of Variations
7.2. Evaluating the Gateaux variations
7.3. The Gateaux Variation in W 1,2 ()
7.4. Elementary Convexity
7.5. The Legendre-Hadamard Condition
7.6. The Weierstrass Necessary Condition
7.7. The du Bois-Reymond Lemma
7.8. The Weierstrass-Erdmann Conditions
7.9. Natural Boundary Conditions
173
173
175
177
179
182
184
187
189
191
Chapter 8. Basic Concepts on Convex Analysis
195
CONTENTS
8.1.
8.2.
8.3.
8.4.
Convex Sets and Convex Functions

Duality in Convex Optimization
Relaxation for the Scalar Case
Duality Suitable for the Vectorial Case
195
204
208
217
Chapter 9. Constrained Variational Optimization

9.1. Basic Concepts
9.2. Duality
9.3. The Lagrange Multiplier Theorem
223
223
227
229
Part 3.
233
Applications
Chapter
10.1.
10.2.
10.3.
10.4.
10.5.
10.6.
10.7.
10.8.
10. Duality Applied to a Plate Model

Introduction
The Primal Variational Formulation
The Legendre Transform
The Classical Dual Formulation
The Second Duality Principle
The Third Duality Principle
A Convex Dual Formulation
A Final Result, other Sufficient
Conditions of Optimality
10.9. Final Remarks
235
235
240
242
244
250
255
258
261
264
Chapter
11.1.
11.2.
11.3.
11.4.
11. Duality Applied to Elasticity

Introduction and Primal Formulation
The First Duality Principle
Conclusion
267
267
269
274
276
Chapter
12.1.
12.2.
12.3.
12.4.
12.5.
12.6.
12. Duality Applied to a Membrane Shell Model

The Polar Functional Related to F : U R
The Final Format of First Duality Principle
Conclusion
277
277
279
280
280
281
285
Chapter 13. Duality Applied to Ginzburg-Landau Type

Equations
13.1. Introduction
13.2. A Concave Dual Variational Formulation
13.3. Applications to Phase Transition in Polymers
287
287
289
293
vi
CONTENTS
13.4. A Numerical Example

13.5. A New Path for Relaxation
13.6. A New Numerical Procedure,
the Generalized Method of Lines
13.7. A Simple Numerical Example
13.8. Conclusion
Chapter
14.1.
14.2.
14.3.
14.4.
298
301
306
314
315
14. Duality Applied to Conductivity in Composites 317

Introduction
317
The Primal Formulation
318
The Duality Principle
318
Conclusion
320
Chapter 15. Duality Applied to the Optimal

Design in Elasticity
15.1. Introduction
15.2. The Main Duality Principles
15.3. The First Applied Duality Principle
15.4. A Concave Dual Formulation
15.5. Duality for a Two-Phase Problem in Elasticity
15.7. Conclusion
323
323
324
329
332
335
338
339
Chapter
16.1.
16.2.
16.3.
16.4.
16.5.
16.6.
16.7.
16.8.
16.9.
341
341
342
344
345
349
354
360
364
372
16. Duality Applied to Micro-Magnetism

Introduction
The Primal formulations and the Duality Principles
A Preliminary Result
The Duality Principle for the Hard Case
An Alternative Dual Formulation for the Hard Case
The Full Semi-linear Case
The Cubic Case in Micro-magnetism
Final Results, Other Duality Principles
Conclusion
Chapter
17.1.
17.2.
17.3.
17. Duality Applied to Fluid Mechanics

Linear Systems which the Solutions
Solve the Navier-Stokes One
17.4. The Method of Lines for the Navier-Stokes System
17.5. Conclusion
Chapter 18. Duality Applied to a Beam Model
373
373
375
376
379
385
387
CONTENTS
18.1. Introduction and Statement

of the Primal Formulation
18.2. Existence and Regularity Results for Problem P
18.3. A Convex Dual Formulation for the Beam Model
18.4. A Final Result, Another Duality Principle
18.5. Conclusion
Bibliography
vii
387
388
390
392
394
395
Introduction
The main objective of this work is to present recent results of the
author about applications of duality to non-convex problems in the
calculus of variations. The text is divided into chapters described
in the next page, and chapters 1 to 9 present the basic concepts on
standard analysis necessary to develop the applications.
Of course, the material presented in the first 9 chapters is not
new, with exception of the section on relaxation for the scalar case,
where we show different proofs of some theorems presented in Ekeland and Temams book Convex Analysis and Variational Problems
(indeed such a book is the theoretical base of the present work),
and the section about relaxation for the vectorial case. The applications, presented in chapters 10 to 18, correspond to the work of
the present author along the last years, and almost all results including the applications of duality for micro-magnetism, composites
in elasticity and conductivity and phase transitions, were obtained
during the Ph.D. program at Virginia Tech.
The key feature of this work is that while all problems studied
here are non-linear with corresponding non-convex variational formulation, it has been almost always possible to develop convex (in
fact concave) dual variational formulations, which in general are
more amenable to numerical computations.
The section on relaxation for the vectorial case, as its title suggests, presents duality principles that are valid even for vectorial
problems. It is worth noting that such results were used in this
text to develop concave dual variational formulations in situations
such as for conductivity in composites, vectorial examples in phase
transitions, etc.
Summary of Main Results
The main results of this work are summarized as follows.
Duality Applied to a Plate Model. Chapter 10 develops
dual variational formulations for the two dimensional equations of
ix
INTRODUCTION
the nonlinear elastic Kirchhoff-Love plate model. The first duality

principle presented is the classical one and may be found in similar format in [40], [22]. It is worth noting that such results are
valid only for positive definite membrane forces. However, we obtain new dual variational formulations which relax or even remove
such constraints. In particular we exhibit a convex dual variational
formulation which allows non positive definite membrane forces. In
the last section, similar to the triality criterion introduced in [24],
we obtain sufficient conditions of optimality for the present case.
The results are based on fundamental tools of Convex Analysis and
the Legendre Transform, which can easily be analytically expressed
for the model in question.
Duality Applied to Finite Elasticity. Chapter 11 develops
duality for a model in finite elasticity. The dual formulations obtained allow the matrix of stresses to be non positive definite. This
is in some sense, an extension of earlier results (which establish
the complementary energy as a perfect global optimization duality
principle only if the stress tensor is positive definite at the equilibrium point). Again, the results are based on standard tools of
convex analysis and on the concept of Legendre Transform.
Duality Applied to a Shell Model. The main focus of Chapter 12 is the development of dual variational formulations for a
non-linear elastic membrane shell model. In the present literature,
the concept of complementary energy can be established only if
the external load produces a critical point with positive definite
membrane forces matrix. Our idea is to obtain dual variational
formulations for which the mentioned constraint is relaxed or even
eliminated. Again, the results are obtained through basic tools of
convex analysis and the concept of Legendre Transform, which can
be analytically established for the concerned shell model.
Duality Applied to Ginzburg-Landau Type Equations.
Chapter 13 is concerned with the development of dual variational
formulations for Ginzburg-Landau type equations. Since the primal formulations are non-convex, we use specific results for distance
between two convex functions to obtain the dual approaches. Note
that we obtain a convex dual formulation and a kind of primal-dual
formulation. For the convex formulation, optimality conditions are
established. In the last section we develop the Generalized Method
of Lines, a new numerical procedure in which the solution of the
SUMMARY OF MAIN RESULTS
xi
partial differential equation in question is written on lines as functions of boundary conditions and boundary shape.
Duality Applied to Conductivity in Composites. The
main focus of Chapter 14 is the development of a dual variational
formulation for a two-phase optimization problem in conductivity.
The primal formulation may not have minimizers in the classical
sense. In this case, the solution through the dual formulation is a
weak limit of minimizing sequences for the original problem.
Duality Applied to the Optimal Design in Elasticity.
The first part of Chapter 15 develops a dual variational formulation for the optimal design of a plate of variable thickness. The
design variable, namely the plate thickness, is supposed to minimize the plate deformation work due to a given external load. The
second part is concerned with the optimal design for a two-phase
problem in elasticity. In this case, we are looking for the mixture
of two constituents that minimizes the structural internal work.
In both applications the dual formulations were obtained through
basic tools of convex analysis.
Duality Applied to Micro-Magnetism. The main focus of
Chapter 16 is the development of dual variational formulations for
functionals related to ferromagnetism models. We develop duality
principles for the so-called hard and full (semi-linear) uniaxial and
cubic cases. It is important to emphasize that the new dual formulations here presented are convex and are useful to compute the
average behavior of minimizing sequences, specially as the primal
formulation has no minimizers in the classical sense. Once more
the results are obtained through standard tools of convex analysis.
Duality Applied to Fluid Mechanics. In Chapter 17 we
use the concept of Legendre Transform to obtain dual variational
formulations for the Navier-Stokes system. What is new and relevant is the establishment of a linear system whose the solution
also solves the original problem. In the last section we present the
solution of an approximation concerning the Navier-Stokes system
through the generalized method of lines.
Duality Applied to a Beam Model. Chapter 18 develops
existence and duality for a non-linear beam model. Our final result
is a convex variational formulation for the model in question.
xii
INTRODUCTION
Acknowledgments
This monograph is based in my Ph.D. thesis at Virginia Tech
and I am especially grateful to Professor Robert C. Rogers for his
excellent work as advisor. I would like to thank the Department
of Mathematics for its constant support and this opportunity of
studying mathematics in advanced level. I am also grateful to all
Professors that have been teaching me during the last years, for
their valuable work. Among the Professors, I particularly thank
Martin Day (Calculus of Variations), James Thomson (Real Analysis) and George Hagedorn (Functional Analysis) for the excellent
lectured courses. Finally, special thanks to all my Professors at
I.T.A. (Instituto Tecnologico de Aeronautica, SP-Brasil) my undergraduate and masters school, and to Virginia Tech-USA, two
wonderful institutions in the American continent.
Part 1
Basic Functional Analysis
CHAPTER 1
Topological Vector Spaces

1.1. Introduction
The main objective of this chapter is to present an outline of
the basic tools of analysis necessary to develop the subsequent chapters. We assume the reader has a background in linear algebra and
elementary real analysis at an undergraduate level. All proofs are
developed in details.
1.2. Vector Spaces

We denote by F a scalar field. In practice this is either R or C,
the set of real or complex numbers.
Definition 1.2.1 (Vector Spaces). A vector space over F is a
set denoted by U, whose elements are called vectors, for which are
defined two operations, namely, addition denoted by (+) : U U
U, and, scalar multiplication, denoted by () : F U U, so that
the following relations are valid
(1) u + v = v + u, u, v U,
(2) u + (v + w) = (u + v) + w, u, v, w U,
(3) there exists a vector denoted by such that u + = u,
u U,
(4) for each u U, there exists a unique vector denoted by
u such that u + (u) = ,
(5) ( u) = ( ) u, , F, u U,
(6) (u + v) = u + v, F, u, v U,
(7) ( + ) u = u + u, , F, u U,
(8) 1 u = u, u U.
Remark 1.2.2. From now on we may drop the dot () in scalar
multiplications and denote u simply as u.
3
1. TOPOLOGICAL VECTOR SPACES
Definition 1.2.3 (Vector Subspace). Let U be a vector space.

A set V U is said to be a vector subspace of U if V is also a
vector space with the same operations as those of U. If V 6= U we
say that V is a proper subspace of U.
Definition 1.2.4 (Finite dimensional Space). A vector space
is said to be of finite dimension if there exists fixed u1 , u2 , ..., un U
such that for each u U there are corresponding 1 , ...., n F for
which
n
X
u=
i ui.
(1.1)
i=1
Definition 1.2.5 (Topological Spaces). A set U is said to

be a topological space if it is possible to define a collection of
subsets of U called a topology in U, for which are valid the following
properties:
(1) U ,
(2) ,
(3) if A and B then A B , and
(4) arbitrary unions of elements in also belong to .
Any A is said to be an open set.
Remark 1.2.6. When necessary, to clarify the notation, we
shall denote the vector space U endowed with the topology by
(U, ).
Definition 1.2.7 (Closed Sets). Let U be a topological space.
A set A U is said to be closed if U A is open. We also denote
U A = Ac .
Proposition 1.2.8. For closed sets we have the following
properties:
(1) U and are closed,
(2) If A and B are closed sets then A B is closed,
(3) Arbitrary intersections of closed sets are closed.
Proof.
(1) Since is open and U = c , by Definition 1.2.7
U is closed. Similarly, since U is open and = U U = U c ,
is closed.
1.2. VECTOR SPACES
(2) A, B closed implies that Ac and B c are open, and by Definition 1.2.5, Ac B c is open, so that A B = (Ac B c )c
is closed.
(3) Consider A = L A , where L is a collection of indices
and A is closed, L. We may write A = (L Ac )c
and since Ac is open L we have, by Definition 1.2.5,
that A is closed.

Definition 1.2.9 (Closure). Given A U we define the clo as the intersection of all closed sets that
sure of A, denoted by A,
contain A.
Remark 1.2.10. From Proposition 1.2.8 Item 3 we have that
A is the smallest closed set that contains A, in the sense that, if C
is closed and A C then A C.
Definition 1.2.11 (Interior). Given A U we define its
interior, denoted by
A, as the union of all open sets contained in
A.
Remark 1.2.12. It is not difficult to prove that if A is open
then A =
A.
Definition 1.2.13 (Neighborhood). Given u0 U we say that
V is a neighborhood of u0 if such a set is open and contains u0 . We
denote such neighborhoods by Vu0 .
Proposition 1.2.14. If A U is a set such that for each
u A there exists a neighborhood Vu u such that Vu A, then
A is open.
Proof. This follows from the fact that A = uA Vu and any
arbitrary union of open sets is open.

Definition 1.2.15 (Function). Let U and V be two topological
spaces. We say that f : U V is a function if f is a collection of
pairs (u, v) U V such that for each u U there exists only one
v V such that (u, v) f .
Definition 1.2.16 (Continuity at a Point). A function f :

U V is continuous at u U if for each neighborhood Vf (u) V
of f (u) there exists a neighborhood Vu U of u such that f (Vu )
Vf (u) .
Definition 1.2.17 (Continuous Function). A function f :
U V is continuous if it is continuous at each u U.
Proposition 1.2.18. A function f : U V is continuous if
and only if f 1 (V) is open for each open V V , where
f 1 (V) = {u U | f (u) V}.
(1.2)
Proof. Suppose f 1 (V) is open whenever V V is open. Pick

u U and any open V such that f (u) V. Since u f 1 (V) and
f (f 1 (V)) V, we have that f is continuous at u U. Since u U
is arbitrary we have that f is continuous. Conversely, suppose f
is continuous and pick V V open. If f 1 (V) = we are done,
since is open,. Thus, suppose u f 1 (V), since f is continuous,
there exists Vu a neighborhood of u such that f (Vu ) V. This
means Vu f 1 (V) and therefore, from Proposition 1.2.14, f 1 (V)
is open.

Definition 1.2.19. We say that (U, ) is a Hausdorff topological space if, given u1 , u2 U, u1 6= u2 , there exists V1 , V2
such that
u1 V1 , u2 V2 and V1 V2 = .
(1.3)
Definition 1.2.20 (Base). A collection is said to be a

base for if every element of may be represented as a union of
elements of .
Definition 1.2.21 (Local Base). A collection
of neighborhoods of a point u U is said to be a local base at u if each
neighborhood of u contains a member of .
Definition 1.2.22 (Topological Vector Space). A vector space
endowed with a topology, denoted by (U, ), is said to be a topological vector space if and only if
(1) Every single point of U is a closed set,
1.2. VECTOR SPACES
(2) The vector space operations (addition and scalar multiplication) are continuous with respect to .
More specifically, addition is continuous if, given u, v U and
V such that u + v V then there exists Vu u and Vv v
such that Vu + Vv V. On the other hand, scalar multiplication
is continuous if given F, u U and V u, there exists
> 0 and Vu u such that, F satisfying | | < we have
Vu V.
Given (U, ), let us associate with each u0 U and 0 F
(0 =
6 0) the functions Tu0 : U U and M0 : U U defined by
Tu0 (u) = u0 + u
(1.4)
M0 (u) = 0 u.
(1.5)
and
The continuity of such functions is a straightforward consequence

of the continuity of vector space operations (addition and scalar
multiplication). It is clear that the respective inverse maps, namely
Tu0 and M1/0 are also continuous. So if V is open then u0 + V,
that is (Tu0 )1 (V) = Tu0 (V) = u0 + V is open. By analogy 0 V
is open. Thus is completely determined by a local base, so that
the term local base will be understood henceforth as a local base
at . So to summarize, a local base of a topological vector space is
a collection of neighborhoods of , such that each neighborhood
of contains a member of .
Now we present some simple results, namely:
Proposition 1.2.23. If A U is open, then u A there
exists a neighborhood V of such that u + V A
Proof. Just take V = A u.
Proposition 1.2.24. Given a topological vector space (U, ),

any element of may be expressed as a union of translates of members of , so that the local base generates the topology .
Proof. Let A U open and u A. V = A u is a neighborhood of and by definition of local base, there exists a set Vu V
such that Vu . Thus, we may write

A = uA (u + Vu ).
(1.6)
1.3. Some Properties of Topological Vector Spaces

In this section we study some fundamental properties of topological vector spaces. We start with the following proposition:
Proposition 1.3.1. Any topological vector space U is a Hausdorff space.
Proof. Pick u0 , u1 U such that u0 6= u1 . Thus V = U {u1
u0 } is an open neighborhood of zero. As + = , by the continuity
of addition, there exist V1 and V2 neighborhoods of such that
V1 + V2 V
(1.7)
define U = V1 V2 (V1 ) (V2 ), thus U = U (symmetric) and

U + U V and hence
u0 + U + U u0 + V U {u1 }
(1.8)
u0 + v1 + v2 6= u1 , v1 , v2 U,
(1.9)
u0 + v1 6= u1 v2 , v1 , v2 U,
(1.10)
so that
or
and since U = U
(u0 + U) (u1 + U) = .
(1.11)
Definition 1.3.2 (Bounded Sets). A set A U is said to

be bounded if to each neighborhood of zero V there corresponds a
number s > 0 such that A tV for each t > s.
Definition 1.3.3 (Convex Sets). A set A U such that
if u, v A then u + (1 )v A, [0, 1],
is said to be convex.
(1.12)
1.3. SOME PROPERTIES OF TOPOLOGICAL VECTOR SPACES
Definition 1.3.4 (Locally Convex Spaces). A topological vector space U is said to be locally convex if there is a local base
whose elements are convex.
Definition 1.3.5 (Balanced sets). A set A U is said to be
balanced if A A, F such that || 1.
Theorem 1.3.6. In a topological vector space U we have:
(1) Every neighborhood of zero contains a balanced neighborhood of zero,
(2) Every convex neighborhood of zero contains a balanced
convex neighborhood of zero.
Proof.
(1) Suppose U is a neighborhood of zero. From the
continuity of scalar multiplication, there exist V (neighborhood of zero) and > 0, such that V U whenever
|| < . Define W = ||< V, thus W U is a balanced
neighborhood of zero.
(2) Suppose U is a convex neighborhood of zero in U. Define
A = {U | C, || = 1}.
(1.13)
rA = ||=1 rU = ||=1 rU.
(1.14)
As 0 = (where U denotes the zero vector) from

the continuity of scalar multiplication there exists > 0
and there is a neighborhood of zero V such that if || <
then V U. Define W as the union of all such V.
Thus W is balanced and 1 W = W as || = 1, so that
W = W U, and hence W A, which implies that the
interior
A is a neighborhood of zero. Also
A U. Since A
is an intersection of convex sets, it is convex and so is
A.
Now will show that A is balanced and complete the proof.

For this, it suffices to prove that A is balanced. Choose r
and such that 0 r 1 and || = 1. Then
Since U is a convex set that contains zero, we obtain
rU U, so that rA A, which completes the proof.

Proposition 1.3.7. Let U be a topological vector space and
V a neighborhood of zero in U. Given u U, there exists r R+
such that u V, such that || < r.
10
Proof. Observe that u+V is a neighborhood of 1u, then by the

continuity of scalar multiplication, there exists W neighborhood of
u and r > 0 such that
W u + V, such that | 1| < r,
(1.15)
u u + V,
(1.16)
( 1)u V, where | 1| < r,
(1.17)
V, such that ||
< r,
u
(1.18)
so that
or
and thus
which completes the proof.
Corollary 1.3.8. Let V be a neighborhood of zero in U, if

{rn } is a sequence such that rn > 0, n N and lim rn = , then
n
U
n=1 rn V.
Proof. Let u U, then u V for any sufficiently small,
from the last proposition u 1 V. As rn we have that
rn > 1 for n sufficiently big, so that u rn V, which completes the
proof.

Proposition 1.3.9. Suppose {n } is sequence such that n
0, n < n1 , n N and V a bounded neighborhood of zero in U,
then {n V} is a local base for U.
Proof. Let U be a neighborhood of zero, as V is bounded, there
exists t0 R+ such that V tU for any t > t0 . As lim n = 0,
there exists n0 N such that if n n0 then n <

n V U, n such that n n0 .
n
1
,
t0
so that
Definition 1.3.10 (Convergence in topological vector spaces).

Let U be a topological vector space. We say {un } converges to
u0 U, if for each neighborhood V of u0 then, there exists N N
such that
un V, n N.
1.4. COMPACTNESS IN TOPOLOGICAL VECTOR SPACES
11
1.4. Compactness in Topological Vector Spaces

We start this section with the definition of open covering.
Definition 1.4.1 (Open Covering). Given B U we say that
{O , A} is a covering of B if B A O . If O is open
A then {O } is said to be an open covering of B.
Definition 1.4.2 (Compact Sets). A set B U is said to be
compact if each open covering of B has a finite sub-covering. More
explicitly, if B A O , where O is open A, then, there
exist 1 , ..., n A such that B O1 ... On , for some n, a
finite positive integer.
Proposition 1.4.3. A compact subset of a Hausdorff space is
closed.
Proof. Let U be a Hausdorff space and consider A U, A
compact. Given x A and y Ac , there exist open sets Ox and
Oyx such that x Ox , y Oyx and Ox Oyx = . It is clear that A
xA Ox and since A is compact, we may find {x1 , x2 , ..., xn } such
that A ni=1 Oxi . For the selected y Ac we have y ni=1 Oyxi
and (ni=1 Oyxi ) (ni=1 Oxi ) = . Since ni=1 Oyxi is open, and y is an
arbitrary point of Ac we have that Ac is open, so that A is closed,

Proposition 1.4.4. A closed subset of a compact space U is
compact.
Proof. Consider {O , L} an open cover of A. Thus {Ac , O ,
L} is a cover of U. As U is compact, there exist 1 , 2 , ..., n
such that Ac (ni=1 Oi ) U, so that {Oi , i {1, ..., n}} covers
A, so that A is compact. The proof is complete.

Definition 1.4.5 (Countably Compact Sets). A set A is said
to be countably compact if every infinite subset of A has a limit
point in A.
Proposition 1.4.6. Every compact subset of a topological
space U is countably compact.
12
Proof. Let B an infinite subset of A compact and suppose

B has no limit point. Choose {x1 , x2 , ....} B and define F =
{x1 , x2 , x3 , ...}. It is clear that F has no limit point. Thus for each
n N, there exist On open such that On F = {xn }. Also, for
each x A F , there exist Ox such that x Ox and Ox F = .
Thus {Ox , x A F ; O1 , O2 , ...} is an open cover of A without a
finite subcover, which contradicts the fact that A is compact.
1.5. Normed and Metric Spaces
The idea here is to prepare a route for the study of Banach
spaces defined below. We start with the definition of norm.
Definition 1.5.1 (Norm). A vector space U is said to be a
normed space, if it is possible to define a function kkU : U R+ =
[0, +), called a norm, which satisfies the following properties:
(1) kukU > 0, if u 6= and kukU = 0 u =
(2) ku + vkU kukU + kvkU , u, v U,
(3) kukU = ||kukU , u U, F.
Now we present the definition of metric.
Definition 1.5.2 (Metric Space). A vector space U is said to
be a metric space if it is possible to define a function d : U U
R+ , called a metric on U, such that
(1) 0 d(u, v) < , u, v U,
(2) d(u, v) = 0 u = v,
(3) d(u, v) = d(v, u), u, v U,
(4) d(u, w) d(u, v) + d(v, w), u, v, w U.
A metric can be defined through a norm, that is
d(u, v) = ku vkU .
(1.19)
d(u + w, v + w) = d(u, v), u, v, w U.
(1.20)
In this case we say that the metric is induced by the norm.

The set Br (u) = {v U | d(u, v) < r} is called the open ball
with center at u and radius r. A metric d : U U R+ is said to
be invariant if
The following are some basic definitions concerning metric and
normed spaces:
1.6. COMPACTNESS IN METRIC SPACES
13
Definition 1.5.3 (Convergent Sequences). Given a metric

space U, we say that {un } U converges to u0 U as n ,
if for each > 0, there exists n0 N, such that if n n0 then
d(un , u0 ) < . In this case we write un u0 as n +.
Definition 1.5.4 (Cauchy Sequence). {un } U is said to be
a Cauchy sequence if for each > 0 there exists n0 N such that
d(un , um ) < , m, n n0
Definition 1.5.5 (Completeness). A metric space U is said
to be complete if each Cauchy sequence related to d : U U R+
converges to an element of U.
Definition 1.5.6 (Banach Spaces). A normed vector space U
is said to be a Banach Space if each Cauchy sequence related to the
metric induced by the norm converges to an element of U.
Remark 1.5.7. We say that a topology is compatible with
a metric d if any A is represented by unions and/or finite
intersections of open balls. In this case we say that d : U U R+
induces the topology .
Definition 1.5.8 (Metrizable Spaces). A topological vector
space (U, ) is said to be metrizable if is compatible with some
metric d.
Definition 1.5.9 (Normable Spaces). A topological vector
space (U, ) is said to be normable if the induced metric (by this
norm) is compatible with .
1.6. Compactness in Metric Spaces
Definition 1.6.1 (Diameter of a Set). Let (U, d) be a metric
space and A U. We define the diameter of A, denoted by diam(A)
by
diam(A) = sup{d(u, v) | u, v A}.
Definition 1.6.2. Let (U, d) be a metric space. We say that
{Fk } U is a nested sequence of sets if
F1 F2 F3 ....
14
Theorem 1.6.3. If (U, d) is a complete metric space then

every nested sequence of non-empty closed sets {Fk } such that
lim diam(Fk ) = 0
k+
has non-empty intersection, that is
k=1 Fk 6= .
Proof. Suppose {Fk } is a nested sequence and lim diam(Fk ) =
k
0. For each n N select un Fn . Suppose given > 0. Since

lim diam(Fn ) = 0,
there exists N N such that if n N then

diam(Fn ) < .
Thus if m, n > N we have um , un FN so that

d(un , um ) < .
Hence {un } is a Cauchy sequence. Being U complete, there exists

u U such that
un u as n .
Choose m N. We have that un Fm , n > m, so that
u Fm = Fm .
Since m N is arbitrary we obtain
The proof is complete.
u
m=1 Fm .
Theorem 1.6.4. Let (U, d) be a metric space. If A U is

compact then it is closed and bounded.
Proof. We have already proved that A is closed. Suppose, to
obtain contradiction that A is not bounded. Thus for each K N
there exists u, v A such that
d(u, v) > K.
Observe that
A uA B1 (u).
Since A is compact there exists u1 , u2 , ..., un A such that
A = nk=1 B1 (uk ).
15
Define
R = max{d(ui, uj ) | i, j {1, ..., n}}.
Choose u, v A such that
d(u, v) > R + 2.
(1.21)
Observe that there exist i, j {1, ..., n} such that

Thus
u B1 (ui), v B1 (uj ).
d(u, v) d(u, ui) + d(ui, uj ) + d(uj , v)

2 + R,
which contradicts (1.21). This completes the proof.
(1.22)
Definition 1.6.5 (Relative Compactness). In a metric space

(U, d) a set A U is said to be relatively compact if A is compact.
Definition 1.6.6 ( - nets). Let (U, d) be a metric space. A
set N U is sat to be a -net with respect to a set A U if for
each u A there exists v N such that
d(u, v) < .
Definition 1.6.7. Let (U, d) be a metric space. A set A U

is said to be totally bounded if for each > 0 there exists a finite
-net with respect to A.
Proposition 1.6.8. Let (U, d) be a metric space. If A U is
totally bounded then it is bounded.
Proof. Choose u, v A. Let {u1, ..., un } be the 1 net with
respect to A. Define
R = max{d(ui, uj ) | i, j {1, ..., n}}.
Observe that there exist i, j {1, ..., n} such that

d(u, ui) < 1, d(v, uj ) < 1.
Thus
d(u, v) d(u, ui) + d(ui, uj ) + d(uj , v)
R + 2.
Since u, v A are arbitrary, A is bounded.
(1.23)
16
Theorem 1.6.9. Let (U, d) be a metric space. If from each

sequence {un } A we can select a convergent subsequence {unk }
then A is totally bounded.
Proof. Suppose, to obtain contradiction, that A is not totally
bounded. Thus there exists 0 > 0 such that there exists no 0 -net
with respect to A. Choose u1 A, hence {u1 } is not a 0 -net, that
is, there exists u2 A such that
d(u1, u2 ) > 0 .
Again {u1 , u2 } is not a 0 -net for A, so that there exists u3 A
such that
d(u1 , u3) > 0 and d(u2, u3 ) > 0 .
Proceeding in this fashion we can obtain a sequence {un } such that
d(un , um ) > 0 , if m 6= n.
(1.24)
Clearly we cannot extract a convergent subsequence of {un }, otherwise such a subsequence would be Cauchy contradicting (1.24).

Definition 1.6.10 (Sequentially compact Sets). Let (U, d) be
a metric space. A set A U is said to be sequentially compact if
for each sequence {un } A there exist a subsequence {unk } and
u A such that
unk u, as k .
Theorem 1.6.11. A subset A of a metric space (U, d) is compact if and only if it is sequentially compact.
Proof. Suppose A is compact. By Proposition 1.4.6 A is countably compact. Let {un } A be a sequence. We have two situations
to consider.
(1) {un } has infinitely many equal terms, that is in this case
we have
un1 = un2 = .... = unk = ... = u A.
Thus the result follows trivially.
(2) {un } has infinitely many distinct terms. In such a case,
being A countably compact, {un } has a limit point in A,
17
so that there exist a subsequence {unk } and u A such

that
unk u, as k .
In both cases we may find a subsequence converging to some u A.

Thus A is sequentially compact.
Conversely suppose A is sequentially compact, and suppose
{G , L} is an open cover of A. For each u A define
(u) = sup{r | Br (u) G , for some L}.
First we prove that (u) > 0, u A. Choose u A. Since A

L G , there exists 0 L such that u G0 . Being G0 open,
there exists r0 > 0 such that Br0 (u) G0 .
Thus
(u) r0 > 0.
Now define 0 by
0 = inf{(u) | u A}.
Therefore, there exists a sequence {un } A such that

(un ) 0 as n .
Since A is sequentially compact, we may obtain a subsequence

{unk } and u0 A such that
(unk ) 0 and unk u0 ,
as k . Therefore, we may find K0 N such that if k > K0

then
(u0 )
.
(1.25)
d(unk , u0) <
4
We claim that
(u0 )
(unk )
, if k > K0 .
4
To prove the claim, suppose
z B (u0 ) (unk ), k > K0 ,
4
(observe that in particular from (1.25)

u0 B (u0 ) (unk ), k > K0 ).
4
Since
(u0 )
< (u0 ),
2
18
there exists some 1 L such that
B (u0 ) (u0 ) G1 .
2
However, since
d(unk , u0 ) <
(u0 )
, if k > K0 ,
4
we obtain
B (u0 ) (u0 ) B (u0 ) (unk ), if k > K0 ,
2
so that
Therefore
(unk )
(u0)
, k > K0 .
4
lim (unk ) = 0
(u0 )
.
4
Choose > 0 such that

0 > > 0.
From the last theorem since A it is sequentially compact, it is totally
bounded. For the > 0 chosen above, consider an -net contained
in A (the fact that the -net may be chosen contained in A is also
a consequence of last theorem) and denote it by N that is,
N = {v1 , ..., vn } A.
Since 0 > , there exists

such that
considering that
1 , ..., n L
B (vi ) Gi , i {1, ..., n},
(vi ) 0 > > 0, i {1, ..., n}.
For u A, since N is an -net we have
u ni=1 B (vi ) ni=1 Gi .
Since u U is arbitrary we obtain

Thus
A ni=1 Gi .
{G1 , ..., Gn }
is a finite subcover for A of
{G , L}.
Hence A is compact.
19
Theorem 1.6.12. Let (U, d) be a metric space. Thus A U

is relatively compact if and only if for each sequence in A, we may
select a convergent subsequence.
Proof. Suppose A is relatively compact. Thus A is compact so
that from the last Theorem, A is sequentially compact.
Thus from each sequence in A we may select a subsequence
which converges to some element of A. In particular, for each sequence in A A we may select a subsequence that converges to
some element of A.
Conversely, suppose that for each sequence in A we may select a
convergent subsequence. It suffices to prove that A is sequentially
compact. Let {vn } be a sequence in A. Since A is dense in A, there
exists a sequence {un } A such that
1
d(un , vn ) < .
n
From the hypothesis we may obtain a subsequence {unk } and u0
A such that
unk u0 , as k .
Thus,
vnk u0 A, as k .
Therefore A is sequentially compact so that it is compact.
Theorem 1.6.13. Let (U, d) be a metric space.

(1) If A U is relatively compact then it is totally bounded.
(2) If (U, d) is a complete metric space and A U is totaly
bounded then A is relatively compact.
Proof.
(1) Suppose A U is relatively compact. From
the last theorem, from each sequence in A we can extract
a convergent subsequence. From Theorem 1.6.9 A is totally
bounded.
(2) Let (U, d) be a metric space and let A be a totally bounded
subset of U.
Let {un } be a sequence in A. Since A is totally bounded
for each k N we find a k -net where k = 1/k, denoted
20
by Nk where
(k)
(k)
Nk = {v1 , v2 , ..., vn(k)

}.
k
In particular for k = 1 {un } is contained in the 1-net N1 .

Thus at least one ball of radius 1 of N1 contains infinitely
(1)
many points of {un }. Let us select a subsequence {unk }kN
of this infinite set (which is contained in a ball of radius 1).
Similarly, we may select a subsequence here just partially
(2)
(1)
relabeled {unl }lN of {unk } which is contained in one of
the balls of the 12 -net. Proceeding in this fashion for each
(k)
k N we may find a subsequence denoted by {unm }mN of
the original sequence contained in a ball of radius 1/k.
(k)
Now consider the diagonal sequence denoted by {unk }kN
= {zk }. Thus
2
d(zn , zm ) < , if m, n > k,
k
that is {zk } is a Cauchy sequence, and since (U, d) is complete, there exists u U such that
zk u as k .
From Theorem 1.6.12, A is relatively compact.

1.7. The Arzela-Ascoli Theorem

In this section we present a classical result in analysis, namely
the Arzela-Ascoli theorem.
Definition 1.7.1 (Equi-continuity). Let F be a collection of
complex functions defined on a metric space (U, d). We say that F
is equicontinuous if for each > 0, there exists > 0 such that if
u, v U and d(u, v) < then
|f (u) f (v)| < , f F .
Furthermore, we say that F is point-wise bounded if for each u U

there exists M(u) R such that
|f (u)| < M(u), f F .
Theorem 1.7.2 (Arzela-Ascoli). Suppose F is a point-wise

bounded equicontinuous collection of complex functions defined on
1.7. THE ARZELA-ASCOLI THEOREM
21
a metric space (U, d). Also suppose that U has a countable dense
subset E. Thus, each sequence {fn } F has a subsequence that
converges uniformly on every compact subset of U.
Proof. Let {un } be a countable dense set in (U, d). By hypothesis, {fn (u1 )} is a bounded sequence, therefore it has a convergent
subsequence, which is denoted by {fnk (u1 )}. Let us denote
fnk (u1 ) = f1,k (u1 ), k N.
Thus there exists g1 C such that
f1,k (u1 ) g1 , as k .
Observe that {fnk (u2)} is also bounded and also it has a convergent
subsequence, which similarly as above we will denote by {f2,k (u2 )}.
Again there exists g2 C such that
f2,k (u1 ) g1 , as k .
f2,k (u2 ) g2 , as k .
Proceeding in this fashion for each m N we may obtain {fm,k }

such that
fm,k (uj ) gj , as k , j {1, ..., m},
where the set {g1 , g2, ..., gm } is obtained as above. Consider the
diagonal sequence
{fk,k },
and observe that the sequence
{fk,k (um )}k>m

is such that
fk,k (um ) gm C, as k , m N.
Therefore we may conclude that from {fn } we may extract a subsequence also denoted by
{fnk } = {fk,k }
which is convergent in
E = {un }nN .
22
Now suppose K U, being K compact. Suppose given > 0.

From the equi-continuity hypothesis there exists > 0 such that if
u, v U and d(u, v) < we have
|fnk (u) fnk (v)| < , k N.

3
Observe that
K uK B (u),
2
and being K compact we may find {

u1 , ..., uM } such that
uj ).
K M
j=1 B (
2
Since E is dense in U, there exists

vj B (
uj ) E, j {1, ..., M}.
2
Fixing j {1, ..., M}, from vj E we obtain that

lim fnk (vj )
exists as k . Hence there exists K0j N such that if k, l > K0j

then
|fnk (vj ) fnl (vj )| < .

3
Pick u K, thus
uj )
u B (
2
for some j {1, ..., M}, so that

d(u, vj ) < .
Therefore if
then
k, l > max{K01 , ..., K0M },
|fnk (u) fnl (u)| |fnk (u) fnk (vj )| + |fnk (vj ) fnl (vj )|
+|fnl (vj ) fnl (u)|

+ + = .
3 3 3
(1.26)
Since u K is arbitrary, we conclude that {fnk } is uniformly

Cauchy on K.
1.8. LINEAR MAPPINGS
23
1.8. Linear Mappings

Given U, V topological vector spaces, a function (mapping) f :
U V , A U and B V , we define:
f (A) = {f (u) | u A},
(1.27)
and the inverse image of B, denoted f 1 (B) as

f 1 (B) = {u U | f (u) B}.
(1.28)
Definition 1.8.1 (Linear Functions). A function f : U V

is said to be linear if
f (u + v) = f (u) + f (v), u, v U, , F.
(1.29)
Definition 1.8.2 (Null Space and Range). Given f : U V ,

we define the null space and the range of f, denoted by N(f ) and
R(f ) respectively, as
N(f ) = {u U | f (u) = }
(1.30)
and
R(f ) = {v V | u U such that f (u) = v}.
(1.31)
Note that if f is linear then N(f ) and R(f ) are subspaces of U.

Proposition 1.8.3. Let U, V be topological vector spaces. If
f : U V is linear and continuous at , then it is continuous
everywhere.
Proof. Since f is linear we have f () = . Since f is continuous
at , given V V a neighborhood of zero, there exists U U
neighborhood of zero, such that
f (U) V.
(1.32)
v u U f (v u) = f (v) f (u) V,
(1.33)
v u + U f (v) f (u) + V,
(1.34)
Thus
or
which means that f is continuous at u. Since u is arbitrary, f is
continuous everywhere.
24
1.9. Linearity and Continuity

Definition 1.9.1 (Bounded Functions). A function f : U V
is said to be bounded if it maps bounded sets into bounded sets.
Proposition 1.9.2. A set E is bounded if and only if the
following condition is satisfied: whenever {un } E and {n } F
are such that n 0 as n we have n un as n .
Proof. Suppose E is bounded. Let U be a balanced neighborhood of in U, then E tU for some t. For {un } E, as n 0,
there exists N such that if n > N then t < |1n | . Since t1 E U
and U is balanced, we have that n un U, n > N, and thus
n un . Conversely, if E is not bounded, there is a neighborhood V of and {rn } such that rn and E is not contained in
rn V, that is, we can choose un such that rn1 un is not in V, n N,

so that {rn1 un } does not converge to .
Proposition 1.9.3. Let f : U V be a linear function.
Consider the following statements
(1) f is continuous,
(2) f is bounded,
(3) If un then {f (un )} is bounded,
(4) If un then f (un ) .
Then,
1 implies 2,
2 implies 3,
if U is metrizable then 3 implies 4, which implies 1.
Proof.
(1) 1 implies 2: Suppose f is continuous, for W
V neighborhood of zero, there exists a neighborhood of
zero in U, denoted by V, such that
f (V) W.
(1.35)
f (E) f (tV) = tf (V) tW, t t0 ,
(1.36)
If E is bounded, there exists t0 R+ such that E tV,

t t0 , so that
and thus f is bounded.

(2) 2 implies 3: Suppose un and let W be a neighborhood
of zero. Then there exists N N such that if n N
1.10. CONTINUITY OF OPERATORS ON BANACH SPACES
25
then un V W where V is a balanced neighborhood of

zero. On the other hand, for n < N, there exists Kn such
that un Kn V. Define K = max{1, K1, ..., Kn }. Then
un KV, n N and hence {un } is bounded. Finally
from 2, we have that {f (un )} is bounded.
(3) 3 implies 4: Suppose U is metrizable and let un .
Given K N, there exists nK N such that if n > nK
then d(un , ) < K12 . Define n = 1 if n < n1 and n = K,
if nK n < nK+1 so that
d(n un , ) = d(Kun , ) Kd(un , ) < K 1 .
(1.37)
Thus since 2 implies 3 we have that {f (n un )} is bounded

so that, by Proposition 1.9.2 f (un ) = n1 f (n un ) as
n .
(4) 4 implies 1: suppose 1 fails. Thus there exists a neighborhood of zero W V such that f 1 (W) contains no neighborhood of zero in U. Particularly, we can select {un } such
that un B1/n () and f (un ) not in W so that {f (un )} does
not converge to zero. Thus 4 fails.

1.10. Continuity of Operators on Banach Spaces
Let U, V be Banach spaces. We call a function A : U V an
operator.
Proposition 1.10.1. Let U, V be Banach spaces. A linear
operator A : U V is continuous if and only if there exists K R+
such that
kA(u)kV < KkukU , u U.
Proof. Suppose A is linear and continuous. From Proposition
1.9.3,
if {un } U is such that un then A(un ) .
(1.38)
We claim that for each > 0 there exists > 0 such that if
kukU < then kA(u)kV < .
Suppose, to obtain contradiction that the claim is false.
Thus there exists 0 > 0 such that for each n N there exists
un U such that kun kU n1 and kA(un )kV 0 .
Therefore un and A(un ) does not converge to , which
contradicts (1.38).
26
Thus the claim holds.

In particular, for = 1 there exists > 0 such that if kukU <
then kA(u)kV < 1. Thus given an arbitrary u U, u 6= , for
u
w=
2kukU
we have
kA(u)kV
kA(w)kV =
< 1,
2kukU
that is
2kukU
kA(u)kV <
, u U.
Defining
2
K=
the first part of the proof is complete. Reciprocally, suppose there

exists K > 0 such that
kA(u)kV < KkukU , u U.
Hence un implies kA(un )kV , so that from Proposition

1.9.3, A is continuous.

1.11. Some Classical Results on Banach Spaces
In this section we present some important results in Banach
spaces. We start with the following theorem.
Theorem 1.11.1. Let U and V be Banach spaces and let
A : U V be a linear operator. Then A is bounded if and only if
the set C U has at least one interior point, where
C = A1 [{v V | kvkV 1}].
Proof. Suppose there exists u0 U in the interior of C. Thus,

there exists r > 0 such that
Br (u0 ) = {u U | ku u0 kU < r} C.
Fix u U such that kukU < r. Thus, we have
kA(u)kV kA(u + u0 )kV + kA(u0)kV .
Observe also that
k(u + u0 ) u0 kU < r,
1.11. SOME CLASSICAL RESULTS ON BANACH SPACES
27
so that u + u0 Br (u0) C and thus

and hence
kA(u + u0 )kV 1
kA(u)kV 1 + kA(u0 )kV ,
(1.39)
u U such that kukU < r. Fix an arbitrary u U such that

u 6= . From (1.39)
u r
w=
kukU 2
is such that
kA(u)kV r
1 + kA(u0 )kV ,
kA(w)kV =
kukU 2
so that
2
kA(u)kV (1 + kA(u0)kV )kukU .
r
Since u U is arbitrary, A is bounded.
Reciprocally, suppose A is bounded. Thus
kA(u)kV KkukU , u U,
for some K > 0. In particular

1
C.
D = u U | kukU
K
Definition 1.11.2. A set S in a metric space U is said to be

nowhere dense if S has an empty interior.
Theorem 1.11.3 (Baire Category Theorem). A complete
metric space is never the union of a countable number of nowhere
dense sets.
Proof. Suppose, to obtain contradiction, that U is a complete
metric space and
U =
n=1 An
where each An is nowhere dense. Since A1 is nowhere dense, there
exist u1 U which is not in A1 , otherwise we would have U = A1 ,
which is not possible since U is open. Furthermore, Ac1 is open, so
that we may obtain u1 Ac1 and 0 < r1 < 1 such that
B1 = Br1 (u1 )
28
satisfies
B1 A1 = .
Since A2 is nowhere dense we have B1 is not contained in A2 . Therefore we may select u2 B1 \ A2 and since B1 \ A2 is open, there
exists 0 < r2 < 1/2 such that
2 = B
r (u2) B1 \ A2 ,
B
2
that is
B2 A2 = .
Proceeding inductively in this fashion, for each n N we may
obtain un Bn1 \ An such that we may choose an open ball Bn =
Brn (un ) such that
n Bn1 ,
B
and
Bn An =
0 < rn < 21n .

Observe that {un } is a Cauchy sequence, considering that if m, n >
N then un , um BN , so that
d(un , um ) < 2(21N ).
Define
u = lim un .
n
Since
we get
un BN , n > N,
N BN 1 .
uB
Therefore u is not in AN 1 , N > 1, which means u is not in
n=1 An = U, a contradiction.

Theorem 1.11.4 (The Principle of Uniform Boundedness).
Let U be a Banach space. Let F be a family of linear bounded
operators from U into a normed linear space V . Suppose for each
u U there exists a Ku R such that
kT (u)kV < Ku , T F .
Then, there exists K R such that
kT k < K, T F .
29
Proof. Define
Bn = {u U | kT (u)kV n, T F }.
By the hypotheses, given u U, u Bn for all n sufficiently big.

Thus,
U =
n=1 Bn .
Moreover each Bn is closed. By the Baire category theorem there
exists n0 N such that Bn0 has non-empty interior. That is, there
exists u0 U and r > 0 such that
Br (u0 ) Bn0 .
Thus, fixing an arbitrary T F , we have
kT (u)kV n0 , u Br (u0 ), .
Thus if kukU < r then k(u + u0 ) u0 kU < r, so that

kT (u + u0 )kV n0 ,
that is
Thus
kT (u)kV kT (u0)kV n0 .
kT (u)kV 2n0 , if kukU < r.
(1.40)
For u U arbitrary, u 6= , define

w=
ru
,
2kukU
from (1.40) we obtain

kT (w)kV =
rkT (u)kV
2n0 ,
2kukU
kT (u)kV
4n0 kukU
, u U.
r
so that
Hence
kT k
4n0
, T F .
r
Theorem 1.11.5 (The Open Mapping Theorem). Let U and

V be Banach spaces and let A : U V be a bounded onto linear
operator. Thus if O U is open then A(O) is open in V .
30
Proof. First we will prove that given r > 0, there exists r > 0
such that
A(Br ()) BrV ().
(1.41)
Here BrV () denotes a ball in V of radius r with center in . Since

A is onto
V =
n=1 A(nB1 ()).
By the Baire Category Theorem, there exists n0 N such that the
closure of A(n0 B1 ()) has non-empty interior, so that A(B1 ()) has
non-empty interior. We will show that there exists r > 0 such that
BrV () A(B1 ()).
Observe that there exists y0 V and r1 > 0 such that

BrV1 (y0 ) A(B1 ()).
(1.42)
Define u0 B1 () which satisfies A(u0 ) = y0 . We claim that

A(Br2 ()) BrV1 (),
where r2 = 1 + ku0 kU . To prove the claim, pick

y A(B1 ())
thus there exists u U such that kukU < 1 and A(u) = y. Therefore
A(u) = A(u u0 + u0 ) = A(u u0 ) + A(u0 ).
But observe that
ku u0 kU kukU + ku0 kU
< 1 + ku0kU
= r2 ,
(1.43)
so that
This means
and hence
A(u u0 ) A(Br2 ()).

y = A(u) A(u0 ) + A(Br2 ()),
A(B1 ()) A(u0 ) + A(Br2 ()).

That is, from this and (1.42), we obtain
A(u0 ) + A(Br2 ()) A(B1 ()) BrV1 (y0 ) = A(u0 ) + BrV1 (),
and therefore
A(Br2 ()) BrV1 ().
31
Since
A(Br2 ()) = r2 A(B1 ()),
we have, for some not relabeled r1 > 0 that
A(B1 ()) BrV1 ().
Thus it suffices to show that
A(B1 ()) A(B2 ()),

to prove (1.41). Let y A(B1 ()), since A is continuous we may
select u1 B1 () such that
y A(u1 ) BrV1 /2 () A(B1/2 ()).
Now select u2 B1/2 () so that
y A(u1 ) A(u2 ) BrV1 /4 ().
By induction, we may obtain
un B21n (),
such that
y
Define
n
X
j=1
A(uj ) BrV1 /2n ().
u=
un ,
n=1
we have that u B2 (), so that
X
A(un ) = A(u) A(B2 ()).
y=
n=1
Therefore
A(B1 ()) A(B2 ()).

The proof of (1.41) is complete.
To finish the proof of this theorem, assume O U is open. Let
v0 A(O). Let u0 O be such that A(u0 ) = v0 . Thus there exists
r > 0 such that
Br (u0 ) O.
From (1.41),
A(Br ()) BrV (),
for some r > 0. Thus
A(O) A(u0 ) + A(Br ()) v0 + BrV ().
32
This means that v0 is an interior point of A(O). Since v0 A(O)

is arbitrary, we may conclude that A(O) is open.

Theorem 1.11.6 (The Inverse Mapping Theorem). A continuous linear bijection of one Banach space onto another has a
continuous inverse.
Proof. Let A : U V satisfying the theorem hypotheses.
Since A is open, A1 is continuous.

Definition 1.11.7 (Graph of a Mapping). Let A : U V be
an operator, where U and V are normed linear spaces. The graph
of A denoted by (A) is defined by
(A) = {(u, v) U V | v = A(u)}.
Theorem 1.11.8 (The Closed Graph Theorem). Let U and
V be Banach spaces and let A : U V be a linear operator. Then
A is bounded if and only if its graph is closed.
Proof. Suppose (A) is closed. Since A is linear (A) is a
subspace of U V . Also, being (A) closed, it is a Banach space
with the norm
k(u, A(u)k = kukU + kA(u)kV .
Consider the continuous mappings
1 (u, A(u)) = u
and
2 (u, A(u)) = A(u).
Observe that 1 is a bijection, so that by the inverse mapping
theorem 1
1 is continuous. As
A = 2 1
1 ,
it follows that A is continuous. The converse is trivial.
1.12. Hilbert Spaces

At this point we introduce an important class of spaces, namely
the Hilbert spaces.
1.12. HILBERT SPACES
33
Definition 1.12.1. Let H be a vector space. We say that H is

a real pre-Hilbert space if there exists a function (, )H : H H R
such that
(1) (u, v)H = (v, u)H , u, v H,
(2) (u + v, w)H = (u, w)H + (v, w)H , u, v, w H,
(3) (u, v)H = (u, v)H , u, v H, R,
(4) (u, u)H > 0, u H, and (u, u)H = 0, if and only if u =
.
Remark 1.12.2. The function (, )H : H H R is called
an inner-product.
Proposition 1.12.3 (Cauchy-Schwarz inequality). Let H be
a pre-Hilbert space. Defining
p
kukH = (u, u)H , u H,
we have
|(u, v)H | kukH kvkH , u, v H.

Equality holds if and only if u = v for some R or v = .
Proof. If v = the inequality is immediate. Assume v 6= .
Given R we have
0 (u v, u v)H
= (u, u)H + 2 (v, v)H 2(u, v)H
= kuk2H + 2 kvk2H 2(u, v)H .
(1.44)
In particular for = (u, v)H /kvk2H , we obtain

0 kuk2H
that is
(u, v)2H
,
kvk2H
|(u, v)H | kukH kvkH .

The remaining conclusions are left to the reader.
Proposition 1.12.4. On a pre-Hilbert space H, the function

k kH : H R
is a norm, where as above
kukH =
(u, u).
34
Proof. The only non-trivial property to be verified, concerning

the definition of norm, is the triangle inequality.
Observe that, given u, v H, from the Cauchy-Schwarz inequality we have,
ku + vk2H =
=
=
Therefore
(u + v, u + v)H
(u, u)H + (v, v)H + 2(u, v)H
(u, u)H + (v, v)H + 2|(u, v)H |
kuk2H + kvk2H + 2kukH kvkH
(kukH + kvkH )2 .
ku + vkH kukH + kvkH , u, v H.

(1.45)
Definition 1.12.5. A pre-Hilbert space H is to be a Hilbert

space if it is complete, that is, if any cauchy sequence in H converges
to an element of H.
Definition 1.12.6 (Orthogonal Complement). Let H be a
Hilbert space. Considering M H we define its orthogonal complement, denoted by M , by
M = {u H | (u, m)H = 0, m M}.
Theorem 1.12.7. Let H be a Hilbert space, M a closed
subspace of H and suppose u H. Under such hypotheses There
exists a unique m0 M such that
ku m0 kH = min {ku mkH }.
mM
Moreover n0 = u m0 M
so that
u = m0 + n0 ,
where m0 M and n0 M . Finally, such a representation

through M M is unique.
Proof. Define d by
d = inf {ku mkH }.
mM
Let {mi } M be a sequence such that
ku mi kH d, as i .
35
Thus, from the parallelogram law we have

kmi mj k2H
= kmi u (mj u)k2H

= 2kmi uk2H + 2kmj uk2H 2k
2u + mi + mj k2H
= 2kmi uk2H + 2kmj uk2H 4k

u + (mi + mj )/2k2H
2d2 + 2d2 4d2 = 0, as i, j +. (1.46)
Thus {mi } M is a Cauchy sequence. Since M is closed, there

exists m0 M such that
so that
Define
mi m0 , as i +,
ku mi kH ku m0 kH = d.
n0 = u m0 .
We will prove that n0 M .
Pick m M and t R, thus we have
Since
we obtain
so that
d2 ku (m0 tm)k2H
= kn0 + tmk2H
= kn0 k2H + 2(n0 , m)H t + kmk2H t2 .
(1.47)
kn0 k2H = ku m0 k2H = d2 ,

2(n0 , m)H t + kmk2H t2 0, t R
(n0 , m)H = 0.
Being m M arbitrary, we obtain
n0 M .
It remains to prove the uniqueness. Let m M, thus
since
ku mk2H = ku m0 + m0 mk2H
= ku m0 k2H + km m0 k2H ,
(u m0 , m m0 )H = (n0 , m m0 )H = 0.
(1.48)
36
From (1.48) we obtain

ku mk2H > ku m0 k2H = d2 ,
if m 6= m0 .
Therefore m0 is unique.
Now suppose
u = m1 + n1 ,
where m1 M and n1 M . As above, for m M
ku mk2H = ku m1 + m1 mk2H
= ku m1 k2H + km m1 k2H ,
ku m1 kH
(1.49)
and thus since m0 such that

d = ku m0 kH
is unique, we get
m1 = m0
and therefore
n1 = u m0 = n0 .
Theorem 1.12.8 (The Riesz Lemma). Let H be a Hilbert

space and let f : H R be a continuous linear functional. Then
there exists a unique u0 H such that
f (u) = (u, u0)H , u H.
Moreover
kf kH = ku0 kH .
Proof. Define N by
N = {u H | f (u) = 0}.
Thus, as f is a continuous and linear N is a closed subspace of H.
If N = H, then f (u) = 0 = (u, )H , u H and the proof would be
complete. Thus assume N 6= H. By the last theorem there exists
v 6= such that v N .
Define
f (v)
v.
u0 =
kvk2H
37
Thus if u N we have
f (u) = 0 = (u, u0)H = 0.
On the other hand if u = v for some R, we have

f (u) = f (v)
f (v)(v, v)H
=
kvk2H

f (v)v
=
v,
kvk2H H
= (v, u0)H .
(1.50)
Therefore f (u) equals (u, u0)H in the space spanned by N and v.

Now we show that this last space (then span of N and v) is in fact
H. Just observe that given u H we may write

f (u)v
f (u)v
+
u= u
.
(1.51)
f (v)
f (v)
Since
f (u)v
u
N
f (v)
we have finished the first part of the proof, that is, we have proven
that
f (u) = (u, u0)H , u H.
To finish the proof, assume u1 H is such that
f (u) = (u, u1)H , u H.
Thus,
ku0 u1 k2H = (u0 u1 , u0 u1 )H

= (u0 u1 , u0)H (u0 u1 , u1 )H
= f (u0 u1 ) f (u0 u1 ) = 0.
(1.52)
Hence u1 = u0 .
Let us now prove that
First observe that
kf kH =
=
kf kH = ku0 kH .
sup{f (u) | u H, kukH 1}
sup{|(u, u0)H | | u H, kukH 1}
sup{kukH ku0 kH | u H, kukH 1}
ku0 kH .
(1.53)
38
On the other hand

kf kH = sup{f (u) | u H, kukH 1}

u0
f
ku0 kH
(u0 , u0 )H
=
ku0kH
= ku0 kH .
(1.54)
From (1.53) and (1.54)

kf kH = ku0 kH .
Remark 1.12.9. Similarly as above we may define a Hilbert

space H over C, that is, a complex one. In this case the complex
inner product (, )H : H H C is defined through the following
properties:
(1) (u, v)H = (v, u)H , u, v H,
(2) (u + v, w)H = (u, w)H + (v, w)H , u, v, w H,
(3) (u, v)H = (u, v)H , u, v H, C,
(4) (u, u)H > 0, u H, and (u, u) = 0, if and only if u = .
Observe that in this case we have
(u, v)H = (u, v)H , u, v H, C,
where for = a + bi C, we have = a bi. Finally, similar

results as those proven above are valid for complex Hilbert spaces.
CHAPTER 2
The Hahn-Banach Theorems and

Weak Topologies
2.1. Introduction
The notion of weak topologies and weak convergence is fundamental in the modern variational analysis. Many important
problems are non-convex and have no minimizers in the classical
sense. However the minimizing sequences in reflexive spaces may
be weakly convergent, and it is important to evaluate the average
behavior of such sequences in many practical applications.
2.2. The Hahn-Banach Theorem
In this chapter U denotes a Banach space, unless otherwise indicated. We start this section by stating and proving the HahnBanach theorem for real vector spaces, which is sufficient for our
purposes.
Theorem 2.2.1 (The Hahn-Banach Theorem).
functional p : U R satisfying
Consider a
p(u) = p(u), u U, > 0,
(2.1)
p(u + v) p(u) + p(v), u, v U.
(2.2)
Let V U be a vector subspace and let g : V R be a linear

functional such that
g(u) p(u), u V.
(2.3)
Then there exists a linear functional f : U R such that

g(u) = f (u), u V,
(2.4)
f (u) p(u), u U.
(2.5)
and
39
40
2. THE HAHN-BANACH THEOREMS AND WEAK TOPOLOGIES
Proof. Pick z U V . Denote by V the space spanned by V

and z, that is
V = {v + z | v V and R}.
(2.6)
We may define an extension of g to V , denoted by g, as

g(z + v) =
g (z) + g(v),
(2.7)
where g(z) will be appropriately defined. Suppose given v1 , v2 V ,

> 0, > 0. Then
g(v1) + g(v2 ) = g(v1 + v2 )
v1 +
v2 )
= ( + )g(
+
+
( + )p(
(v1 z) +
(v2 + z))
+
+
p(v1 z) + p(v2 + z)
(2.8)
and therefore
1
1
[p(v1 z) + g(v1 )] [p(v2 + z) g(v2 )],
v1 , v2 V, , > 0. (2.9)
Thus, there exists a R such that
1
1
sup [ (p(v z) + g(v))] a inf [ (p(v + z) g(v))].
vV,>0
vV,>0
(2.10)
If we define g(z) = a we obtain g(u) p(u), u V . Define by

E the set of extensions e of g, which satisfy e(u) p(u) on the
subspace where e is defined. We define a partial order in E by
setting e1 e2 if e2 is defined in a larger set than e1 and e1 = e2
where both are defined. Let {e }A be a linearly ordered subset
of E. Let V be the subspace on which e is defined. Define e on
A V by setting e(u) = e on V . Clearly e e so each linearly
ordered set of E has an upper bound. By the Zorns lemma, E
has a maximal element f defined on some set U such that f (u)
p(u), u U . We can conclude that U = U, otherwise if there
was an z1 U U , as above we could have a new extension f1 to
the subspace spanned by z1 and U , contradicting the maximality
of f .
2.2. THE HAHN-BANACH THEOREM
41
Definition 2.2.2 (Topological Dual Space). For a Banach

space U, we define its Topological Dual Space, as the set of all
linear continuous functionals defined on U. We suppose that such
dual space of U, may be identified with a space denoted by U
through a bilinear form h, iU : U U R (here we are refereing
to the standard representations of dual spaces concerning Lebesgue
and Sobolev spaces). That is, given f : U R linear continuous
functional, there exists u U such that
f (u) = hu, uiU , u U.
(2.11)
The norm of f , denoted by kf kU , is defined as
kf kU = sup{|hu, uiU | | kukU 1}.
(2.12)
uU
Corollary 2.2.3. Let V U a vector subspace of U and let

g : V R a linear continuous functional of norm
kgkV = sup{|g(u)| | kukV 1}.
(2.13)
uU
Then, there exists an u in U such that

hu, uiU = g(u), u V,
(2.14)
ku kU = kgkV .
(2.15)
and
Proof. Apply Theorem 2.2.1 with p(u) = kgkV kukV .
Corollary 2.2.4. Given u0 U there exists u0 U such that

ku0 kU = ku0 kU and hu0, u0 iU = ku0 k2U .
(2.16)
Proof. Apply Corollary 2.2.3 with V = {u0 | R} and

g(tu0) = tku0 k2U so that kgkV = ku0 kU .
Corollary 2.2.5. Given u U we have
kukU = sup {|hu, uiU | | kukU 1}.

u U
Proof. Suppose u 6= . Since
|hu, uiU | kukU kukU , u U, u U
(2.17)
42
we have
sup {|hu, uiU | | kukU 1} kukU .
(2.18)
u U
However, from last corollary we have that there exists u0 U such
that ku0 kU = kukU and hu, u0iU = kuk2U . Define u1 = kuk1

U u0 .
Then ku1kU = 1 and hu, u1iU = kukU .

Definition 2.2.6 (Affine Hyper-Plane). Let U be a Banach
space. An affine hyper-plane H is a set of the form
H = {u U | hu, uiU = }
(2.19)
for some u U and R.

Proposition 2.2.7.
closed.
A hyper-plane H defined as above is
Proof. The result follows from the continuity of hu, uiU as a

functional defined in U.

Definition 2.2.8 (Separation). Given A, B U we say that
a hyper-plane H, defined as above separates A and B if
hu, uiU , u A, and hu, uiU , u B.
(2.20)
We say that H separates A and B strictly if there exists > 0 such

that
hu, uiU , u A, and hu, uiU + , u B,
(2.21)
Theorem 2.2.9 (Hahn-Banach theorem, geometric form). Consider A, B U two convex disjoint non-empty sets, where A is
open. Then there exists a closed hyper-plane that separates A and
B.
We need the following Lemma.
Lemma 2.2.10. Consider C U a convex open set such that
C. Given u U, define
p(u) = inf{ > 0, 1 u C}.
(2.22)
2.2. THE HAHN-BANACH THEOREM
43
Thus, p is such that there exists M R+ satisfying

0 p(u) MkukU , u U,
(2.23)
C = {u U | p(u) < 1}.
(2.24)
and
Also
p(u + v) p(u) + p(v), u, v U.
Proof. Let r > 0 be such that B(, r) C, thus
kukU
, u U
(2.25)
r
which proves (2.23). Now suppose u C. Since C is open (1+)u
1
C for sufficiently small. Therefore p(u) 1+
< 1. Conversely, if
1
p(u) < 1 there exists 0 < < 1 such that u C and therefore,
since C is convex, u = (1 u) + (1 ) C.
u
v
Also, let u, v C and > 0. Thus p(u)+
C and p(v)+
C so
p(u)
(1t)v
p(u)+
tu
that p(u)+
+ p(v)+
C, t [0, 1]. Particularly for t = p(u)+p(v)+2
u+v
C, which means p(u + v) p(u) + p(v) +
we obtain p(u)+p(v)+2
2, > 0
Lemma 2.2.11. Consider C U a convex open set and let

u0 U be a vector not in C. Then there exists u U such that
hu, uiU < hu0 , uiU , u C
Proof. By a translation, we may assume C. Consider the
functional p as in the last lemma. Define V = {u0 | R}.
Define g on V , by
g(tu0 ) = t, t R.
(2.26)
We have that g(u) p(u), u V . From the Hahn-Banach theorem, there exist a linear functional f on U which extends g such
that
f (u) p(u) MkukU .
(2.27)
Here we have used lemma 2.2.10. In Particular, f (u0 ) = 1, and

(also from the last lemma) f (u) < 1, u C. The existence of u
satisfying the theorem follows from the continuity of f indicated in
(2.27).
44
Proof of Theorem 2.2.9. Define C = A + (B) so that C is

convex and 6 C. From Lemma 2.2.11, there exists u U such
that hw, uiU < 0, w C, which means
hu, uiU < hv, uiU , u A, v B.
(2.28)
Thus, there exists R such that

suphu, uiU inf hv, uiU ,
vB
uA
(2.29)
Theorem 2.2.12 (Hahn-Banach theorem, second geometric

form). Consider A, B U two convex disjoint non-empty sets.
Suppose A is closed and B is compact. Then there exists an hyperplane which separates A and B strictly.
Proof. There exists > 0 sufficiently small such that A =
A + B(0, ) and B = B + B(0, ) are convex disjoint sets. From
Theorem 2.2.9, there exists u U such that u 6= and
hu + w1 , u iU hu + w2, u iU , u A, v B, w1 , w2 B(0, 1).
(2.30)
Thus, there exists R such that
hu, uiU + ku kU hv, uiU ku kU , u A, v B.
(2.31)

Corollary 2.2.13. Suppose V U is a vector subspace such
that V 6= U. Then there exists u U such that u 6= and
hu, uiU = 0, u V.
(2.32)
Proof. Consider u0 U such that u0 6 V . Applying Theorem

2.2.9 to A = V and B = {u0 } we obtain u U and R such
that u 6= and
hu, uiU < < hu0 , u iU , u V.
Since V is a subspace we must have hu, uiU = 0, u V .
(2.33)
2.3. WEAK TOPOLOGIES
45
2.3. Weak Topologies

Definition 2.3.1 (Weak Neighborhoods and Weak Topologies).
For the topological space U and u0 U, we define a weak neighborhood of u0 , denoted by Vw as
Vw = {u U | |hu u0 , ui iU | < , i {1, ..., m}},
(2.34)
for some m N, > 0, and ui U , i {1, ..., m}. Also, we

define the weak topology for U, denoted by (U, U ) as the set of
arbitrary unions and finite intersections of weak neighborhoods in
U.
Proposition 2.3.2. Consider Z a topological vector space and
a function of Z into U. Then is continuous as U is endowed
with the weak topology, if and only if u is continuous, for all
u U .
Proof. It is clear that if is continuous with U endowed with
the weak topology, then u is continuous for all u U . Conversely, consider U a weakly open set in U. We have to show that
1 (U) is open in Z. But observe that U = L V , where each V
is a weak neighborhood. Thus 1 (U) = L 1 (V ). The result
follows considering that u is continuous for all u U , so that
1 (V ) is open, for all L.

Proposition 2.3.3. A Banach space U is Hausdorff as endowed with the weak topology (U, U ).
Proof. Pick u1 , u2 U such that u1 6= u2 . From the HahnBanach theorem, second geometric form, there exists a hyper-plane
separating {u1 } and {u2 }. That is, there exist u U and R
such that
hu1 , u iU < < hu2 , u iU .
(2.35)
Defining
Vw1 = {u U | |hu u1 , ui| < hu1 , u iU },
(2.36)
Vw2 = {u U | |hu u2 , u iU | < hu2 , u iU },
(2.37)
and
we obtain u1 Vw1 , u2 Vw2 and Vw1 Vw2 = .
46
Remark 2.3.4. if {un } U is such that un converges to u in

(U, U ) then we write un u.
Proposition 2.3.5. Let U be a Banach space. Considering
{un } U we have
(1) un u, for (U, U ) hun , u iU hu, uiU , u U ,
(2) If un u strongly (in norm) then un u weakly,
(3) If un u weakly, then {kun kU } is bounded and kukU
lim inf kun kU ,
n
(4) If un u weakly and un u strongly in U then

hun , un iU hu, uiU .
Proof.
(1) The result follows directly from the definition
of topology (U, U ).
(2) This follows from the inequality
|hun , uiU hu, uiU | ku kU kun ukU .
(2.38)
|hun, u iU | ku kU kun kU ,
(2.39)
(3) Since for every u U the sequence {hun , u iU } is bounded,

from the uniform boundedness principle we have that there
exists M > 0 such that kun kU M, n N. Furthermore,
for u U we have
and taking the limit, we obtain
|hu, uiU | lim inf ku kU kun kU .
(2.40)
kukU = sup |hu, uiU | lim inf kun kU .
(2.41)
Thus
kukU 1
(4) Just observe that

|hun , un iU hu, uiU | |hun , un u iU |
+|hu un , u iU |
kun u kU kun kU
+|hun u, uiU |
Mkun u kU
+|hun u, uiU |.
(2.42)
2.4. THE WEAK-STAR TOPOLOGY
47
Theorem 2.3.6. Consider A U a convex set. Thus A is

weakly closed if and only if it is strongly closed.
Proof. Suppose A is strongly closed. Consider u0 6 A. By
the Hahn-Banach theorem there exists a closed hyper-plane which
separates u0 and A strictly. Therefore there exists R and
u U such that
hu0 , uiU < < hv, uiU , v A.
(2.43)
V = {u U | hu, uiU < },
(2.44)
Define
so that u0 V, V U A. Since V is open for (U, U ) we have

that U A is weakly open, hence A is weakly closed. The converse
is obvious.

2.4. The Weak-Star Topology
Definition 2.4.1 (Reflexive Spaces). Let U be a Banach
space. We say that U is reflexive if the canonical injection J :
U U defined by
hu, uiU = hu , J(u)iU , u U, u U ,
(2.45)
is onto.
The weak topology for U is denoted by (U , U ). By analogy,
we can define the topology (U , U), which is called the weak-star
topology. A standard neighborhood of u0 U for the weak-star
topology, which we denoted by Vw , is given by
Vw = {u U | |hui, u u0 iU | < , i {1, ..., m}}
(2.46)
for some > 0, m N, ui U, i {1, ..., m}. It is clear that the

weak topology for U and the weak-star topology coincide if U is
reflexive.
Proposition 2.4.2. Let U be a Banach space. U as endowed
with the weak-star topology is a Hausdorff space.
Proof. The proof is similar to that of Proposition 2.3.3.
48
2.5. Weak-Star Compactness

We start with an important theorem about weak-* compactness.
Theorem 2.5.1 (Banach Alaoglu Theorem). The set BU =
{f U | kf kU 1} is compact for the topology (U , U) (the
weak-star topology).
Proof. For
Q each u U, we will associate a real number u and
denote = uU u . We have that RU and let us consider the
projections Pu : RU R, where Pu () = u . Consider the weakest
topology for which the functions Pu (u U) are continuous. For
U , with the topology (U , U) define : U RU , by
Y
(u ) =
hu, uiU , u U .
(2.47)
uU
Since for each fixed u the mapping u hu, uiU is weakly-star

continuous, we see that, is continuous, since weak-star convergence and convergence in are equivalent in U . To prove that
1 is continuous, from Proposition 2.3.2, it suffices to show that
the function hu, 1()iU is continuous on (U ). This is true
because hu, 1()iU = u on (U ). On the other hand, it is also
clear that (BU ) = K where
K = { RU | |u | kukU ,
u+v = u + v , u = u , u, v U, R}. (2.48)
To finish the proof, it is sufficient, from the continuity of 1 , to

show that K is compact in RU , concerning the topology . Observe
that K = K1 K2 where
and
K1 = { RU | |u | kukU , u U},
(2.49)
K2 = { RU | u+v = u + v , u = u , u, v U, R}.
(2.50)
Q
The set uU [kukU , kukU ] is compact as a Cartesian product of
compact intervals. Since K1 K and K1 is closed, we have that
K1 is compact (for the topology in question) . On the other hand,
K2 is closed, because defining the closed sets Au,v and B,u as
Au,v = { RU | u+v u v = 0},
(2.51)
2.5. WEAK-STAR COMPACTNESS
49
and
B,u = { RU u u = 0}
(2.52)
K2 = (u,vU Au,v ) ((,u)RU B,u ).
(2.53)
we may write
We recall that the K2 is closed because arbitrary intersections of

closed sets are closed. Finally, we have that K1 K2 is compact,

Theorem 2.5.2 (Kakutani). Let U be a Banach space. Then
U is reflexive if and only if
BU = {u U | kukU 1}
(2.54)
is compact for the weak topology (U, U ).
Proof. Suppose U is reflexive, then J(BU ) = BU . From the

last theorem BU is compact for the topology (U , U ). Therefore it suffices to verify that J 1 : U U is continuous from U
with the topology (U , U ) to U, with the topology (U, U ).
From Proposition 2.3.2 it is sufficient to show that the function
u 7 hf, J 1 uiU is continuous for the topology (U , U ), for each
f U . Since hf, J 1 uiU = hu, f iU we have completed the first
part of the proof. For the second we need two lemmas.
Lemma 2.5.3 (Helly). Let U be a Banach space, f1 , ..., fn
U and 1 , ..., n R, then 1 and 2 are equivalent, where:
(1)
Given > 0, there exists u U such that ku kU 1 and
(2)
|hu, fi iU i | < , i {1, ..., n}.

n
n

X
X

f

i i
i i

i=1
i=1
, 1 , ..., n R.
(2.55)
PnProof. 1 2: Fix 1 , ..., n R, > 0 and define S =

i=1 |i |. From 1, we have

n
n

X
X

(2.56)
hu
,
f
i

i
i U
i i < S

i=1
i=1
50
and therefore

n
n

X
X

hu
,
f
i

i
i U < S
i i

i=1
i=1
or

n
n
X
X

i i <
i fi

i=1
i=1
so that

n
X

ku kU + S
i fi

i=1

n
n

X
X

i fi
i i

i=1
since is arbitrary.
(2.57)
i=1
+ S (2.58)
(2.59)
U
Now let us show that 2 1. Define

~ = (1 , ..., n ) Rn and
consider the function (u) = (hf1 , uiU , ..., hfn , uiU ). Item 1 implies
that
~ belongs to the closure of (BU ). Let us suppose that
~ does
not belong to the closure of (BU ) and obtain a contradiction. Thus
we can separate
~ and the closure of (BU ) strictly, that is there
~
exists = (1 , ..., n ) Rn and R such that
~ u BU
(u) ~ < <
~ ,
(2.60)
Taking the supremum in u we contradict 2.
Also we need the lemma.

Lemma 2.5.4. Let U be a Banach space. Then J(BU ) is
dense in BU for the topology (U , U ).
Proof. Let u BU and consider Vu a neighborhood of u
for the topology (U , U ). It suffices to show that J(BU ) Vu 6=
. As Vu is a weak neighborhood, there exists f1 , ..., fn U and
> 0 such that
Vu = { U | hfi , u iU | < , i {1, ..., n}}.
(2.61)
Define i = hfi , u iU and thus for any given 1 , ..., n R we have

n
n
n

X
X
X

f
,
(2.62)
f
,
u
i
h
=

i i
i i
U
i i

i=1
i=1
i=1
2.5. WEAK-STAR COMPACTNESS
51
so that from Helly lemma, there exists u U such that ku kU 1

and
or,
and hence
|hu , fi iU i | < , i {1, ..., n}
(2.63)
|hfi, J(u ) u iU | < , i {1, ..., n}
(2.64)
J(u ) Vu .
(2.65)
Now we will complete the proof of Kakutani Theorem. Suppose

BU is weakly compact (that is, compact for the topology (U, U )).
Observe that J : U U is weakly continuous, that is, it is continuous with U endowed with the topology (U, U ) and U endowed
with the topology (U , U ). Thus as BU is weakly compact, we
have that J(BU ) is compact for the topology (U , U ). From
the last lemma, J(BU ) is dense BU for the topology (U , U ).
Hence J(BU ) = BU , or J(U) = U , which completes the proof.

Proposition 2.5.5. Let U be a reflexive Banach space. Let
K U be a convex closed bounded set. Then K is weakly compact.
Proof. From Theorem 2.3.6, K is weakly closed (closed for the
topology (U, U )). Since K is bounded, there exists R+ such
that K BU . Since K is weakly closed and K = K BU , we
have that it is weakly compact.

Proposition 2.5.6. Let U be a reflexive Banach space and
M U a closed subspace. Then M with the norm induced by U
is reflexive.
Proof. We can identify two weak topologies in M, namely:
(M, M ) and the trace of (U, U ).
(2.66)
It can be easily verified that these two topologies coincide (through

restrictions and extensions of linear forms). From theorem 2.4.2,
it suffices to show that BM is compact for the topology (M, M ).
But BU is compact for (U, U ) and M U is closed (strongly) and
convex so that it is weakly closed, thus from last proposition, BM
52
is compact for the topology (U, U ), and therefore it is compact

for (M, M ).

2.6. Separable Sets
Definition 2.6.1 (Separable Spaces). A metric space U is said
to be separable if there exist a set K U such that K is countable
and dense in U.
The next Proposition is proved in [9].
Proposition 2.6.2. Let U be a separable metric space. If
V U then V is separable.
Theorem 2.6.3. Let U be a Banach space such that U is
separable. Then U is separable.
Proof. Consider {un } a countable dense set in U . Observe
that
kun kU = sup{|hun, uiU | | u U and kukU = 1}
(2.67)
so that for each n N, there exists un U such that kun kU = 1

and hun , un iU 21 kun kU .
Define U0 as the vector space on Q spanned by {un }, and U1
as the vector space on R spanned by {un }. It is clear that U0 is
dense in U1 and we will show that U1 is dense in U, so that U0 is a
dense set in U. For, suppose u is such that hu, uiU = 0, u U1 .
Since {un } is dense in U , given > 0, there exists n N such that
kun u kU < , so that
1
ku kU hun , un iU = hun , un u iU + hun , u iU
2 n
kun u kU kun kU + 0 < (2.68)
or
ku kU kun u kU + kun kU < + 2 = 3.
(2.69)
Therefore, since is arbitrary, ku kU = 0, that is u = . By

Corollary 2.2.13 this completes the proof.

Proposition 2.6.4. U is reflexive if and only if U is reflexive.
2.7. UNIFORMLY CONVEX SPACES
53
Proof. Suppose U is reflexive, as BU is compact for (U , U)

and (U , U) = (U , U ) we have that BU is compact for (U ,
U ), which means that U is reflexive.
Suppose U is reflexive, from above U is reflexive. Since J(U)
is a closed subspace of U , from Proposition 2.5.6, J(U) is reflexive.
Thus, U is reflexive, since J is a isometry.

Proposition 2.6.5. Let U be a Banach space. Then U is
reflexive and separable if and only if U is reflexive and separable.
2.7. Uniformly Convex Spaces
Definition 2.7.1 (Uniformly Convex Spaces). A Banach space
U is said to be uniformly convex if for each > 0, there exists > 0
such that:
If u, v U, kukU 1, kvkU 1, and ku vkU > then
ku+vkU
< 1 .
2
Theorem 2.7.2 (Milman Pettis).
Banach space is reflexive.
Every uniformly convex
Proof. Let U be such that kkU = 1. It suffices to show

that J(BU ). Since J(BU ) is closed in U , we have only to show
that for each > 0 there exists u U such that k J(u)kU < .
Thus, suppose given > 0. Let > 0 be the corresponding
constant relating the uniformly convex property.
Choose f U such that kf kU = 1 and
hf, iU > 1 .
2
Define
(2.70)
V = { U | |hf, iU | < .
2
Observe that V is neighborhood of in (U , U ). Since J(BU )
is dense in BU concerning the topology (U , U ), we have that
V J(BU ) 6= and thus there exists u BU such that J(u) V.
Suppose, to obtain contradiction, that
k J(u)kU > .
Therefore, defining
W = (J(u) + BU )c ,
54
we have that W , where W is also a weak neighborhood of in

(U , U ), since BU is closed in (U , U ).
Hence V W J(BU ) 6= , so that there exists some v BU
such that J(v) V W. Thus, J(u) V and J(v) V , so that
|hu, f iU hf, iU | < ,

2
and
|hv, f iU hf, iU | < .

2
Hence,
2hf, iU < hu + v, f iU +
ku + vkU + .
(2.71)
ku vkU .
(2.72)
From this and (2.70) we obtain

ku + vkU
> 1 ,
2
and thus from the definition of uniform convexity, we obtain
On the other hand, since J(v) W , we have
kJ(u) J(v)kU = ku vkU > ,
which contradicts (2.72). The proof is complete.
CHAPTER 3
Topics on Linear Operators

3.1. Topologies for Bounded Operators
First we recall that the set of all bounded linear operators, denoted by L(U, Y ), is a Banach space with the norm
kAk = sup{kAukY | kukU 1}.
The topology related to the metric induced by this norm is called

the uniform operator topology.
Let us introduce now the strong operator topology, which is
defined as the weakest topology for which the functions
are continuous where
Eu : L(U, Y ) Y
Eu (A) = Au, A L(U, Y ).
For such a topology a base at origin is given by sets of the form

{A | A L(U, Y ), kAuikY < , i {1, , ..., n}},
where u1 , ..., un U and > 0.

Observe that a sequence {An } L(U, Y ) converges to A concerning this last topology if
kAn u AukY 0, as n , u U.
In the next lines we describe the weak operator topology in

L(U, Y ). Such a topology is weakest one such that the functions
are continuous, where
Eu,v : L(U, Y ) C
Eu,v (A) = hAu, viY , A L(U, Y ), u U, v Y .
For such a topology, a base at origin is given by sets of the form

{A L(U, Y ) | |hAui , vj iY | < , i {1, ..., n}, j {1, ..., m}}.
where > 0, u1 , ..., un U, v1 , ..., vm Y .

55
56
3.
TOPICS ON LINEAR OPERATORS
A sequence {An } L(U, Y ) converges to A L(U, Y ) if

|hAn u, viY hAu, viY | 0,
as n , u U, v Y .
3.2. Adjoint Operators

We start this section recalling the definition of adjoint operator.
Definition 3.2.1. Let U, Y be Banach spaces. Given a bounded
linear operator A : U Y and v Y , we have that T (u) =
hAu, v iY is such that
|T (u)| kAukY kv k kAkkv kY kukU .
Hence T (u) is a continuous linear functional on U and considering
our fundamental representation hypothesis, there exists u U
such that
T (u) = hu, uiU , u U.
We define A by setting u = A v , so that
T (u) = hu, uiU = hu, Av iU
that is,
hu, A v iU = hAu, v iY , u U, v Y .
We call A : Y U the adjoint operator relating A : U Y.
Theorem 3.2.2. Let U, Y be Banach spaces and let A : U
Y be a bounded linear operator. Then
kAk = kA k.
3.2. ADJOINT OPERATORS
57
Proof. Observe that

kAk = sup{kAuk | kukU = 1}
uU
= sup{ sup {hAu, v iY | kv kY = 1}, kukU = 1}

uU v Y
sup
(u,v )U Y
sup
(u,v )U Y
{hu, Av iU | kv kY = 1, kukU = 1}
sup {sup{hu, Av iU | kukU = 1}, kv kY = 1}
=
=
{hAu, v iY | kv kY = 1, kukU = 1}
v Y uU

sup {kA v k, kv kY = 1}
v Y
= kA k.
(3.1)
In particular, if U = Y = H where H is Hilbert space, we have

Theorem 3.2.3. Given the bounded linear operators A, B :
H H we have
(1) (AB) = B A ,
(2) (A ) = A,
(3) If A has a bounded inverse A1 then A has a bounded
inverse and
(A )1 = (A1 ) .
(4) kAA k = kAk2 .
Proof.
(1) Observe that
(ABu, v)H = (Bu, A v)H = (u, B A v)H , u, v H.
(2) Observe that
(u, Av)H = (A u, v)H = (u, A v)H , u, v H.
(3) We have that
I = AA1 = A1 A,
so that
I = I = (AA1 ) = (A1 ) A = (A1 A) = A (A1 ) .
(4) Observe that
kA Ak kAkkA k = kAk2 ,
58
3.
and
kA Ak sup{(u, AAu)H | kukU = 1}
uU
= sup{(Au, Au)H | kukU = 1}

uU
= sup{kAuk2H | kukU = 1} = kAk2 ,
(3.2)
uU
and hence
kA Ak = kAk2 .
Definition 3.2.4.
adjoint if
Given A L(H) we say that A is selfA = A .
Theorem 3.2.5. Let U and Y be Banach spaces and let

A : U Y be a bounded linear operator. Then
[R(A)] = N(A ),
where
[R(A)] = {v Y | hAu, v iY = 0, u U}.
Proof. Let v N(A ). Choose v R(A). Thus there exists u
in U such that Au = v so that
hv, v iY = hAu, v iY = hu, Av iU = 0.
Since v R(A) is arbitrary we have obtained

N(A ) [R(A)] .
Suppose v [R(A)] . Choose u U. Thus,

so that
hAu, v iY = 0,
hu, A v iU , u U.
Therefore A v = , that is, v N(A ). Since v [R(A)] is
arbitrary, we get
[R(A)] N(A ).
This completes the proof.

The proof of the next result may be found in Luenberger [27],
page 156.
3.3. COMPACT OPERATORS
59
Theorem 3.2.6. Let U and Y be Banach spaces and let

A : U Y be a bounded linear operator. If R(A) is closed then
R(A ) = [N(A)] .
3.3. Compact Operators

We start this section defining compact operators.
Definition 3.3.1. Let U and Y be Banach spaces. An operator A L(U, Y ) (linear and bounded) is said to compact if A takes
bounded sets into pre-compact sets. Summarizing, A is compact
if for each bounded sequence {un } U, {Aun } has a convergent
subsequence in Y .
Theorem 3.3.2. A compact operator maps weakly convergent
sequences into norm convergent sequences.
Proof. Let A : U Y be a compact operator. Suppose
un u weakly in U.
By the uniform boundedness theorem, {kun k} is bounded. Thus,

given v Y we have
hv , Aun iY
= hA v , un iY
hA v , uiU
= hv , AuiY .
Being v Y arbitrary, we get that
Aun Au weakly in Y.
(3.3)
(3.4)
Suppose Aun does not converge in norm to Au. Thus there exists
> 0 and a subsequence {Aunk } such that
kAunk AukY , k N.
As {unk } is bounded and A is compact, {Aunk } has a subsequence

converging para v 6= Au. But then such a sequence converges
weakly to v 6= Au, which contradicts (3.4). The proof is complete.

Theorem 3.3.3. Let H be a separable Hilbert space. Thus
each compact operator in L(H) is the limit in norm of a sequence
of finite rank operators.
60
3.
Proof. Let A be a compact operator in H. Let {j } an orthonormal basis in H. For each n N define
n = sup{kAkH | [1 , ..., n ] and kkH = 1}.
It is clear that {n } is a non-increasing sequence that converges to

a limit 0. We will show that = 0. Choose a sequence {n }
such that
n [1 , ..., n ] ,
kn kH = 1 and kAn kH /2. Now we will show that

n , weakly in H.
Let H = H, thus there exists a sequence {aj } C such that
aj j .
j=1
Suppose given > 0. We may find n0 N such that
a2j < .
j=n0
Choose n > n0 . Hence there exists {bj }j>n such that
n =
bj j ,
j=n+1
and
b2j = 1.
j=n+1
Therefore

X

(j , j )H aj bj
|(n , )H | =

j=n+1

X

aj bj
=

j=n+1
v
v
uX
u
u 2u X
t
aj t
b2j
j=n+1
j=n+1
(3.5)
3.4.
THE SQUARE ROOT OF A POSITIVE OPERATOR
61
if n > n0 . Since > 0 is arbitrary,

(n , )H 0, as n .
Since H is arbitrary, we get
n , weakly in H.
Hence, as A is compact, we have
An in norm ,
so that = 0. Finally, we may define {An } by
!
n
n
X
X
An (u) = A
(u, j )H Aj ,
(u, j )H j =
j=1
j=1
for each u H. Thus

kA An k = n 0, as n .
3.4. The Square Root of a Positive Operator

Definition 3.4.1. Let H be a Hilbert space. A mapping
E : H H is said to be a projection on M H if for each z H
we have
Ez = x
where z = x + y, x M and y M .
Observe that
(1) E is linear,
(2) E is idempotent, that is E 2 = E,
(3) R(E) = M,
(4) N(E) = M .
Also observe that from
Ez = x
we have
so that
kEzkH = kxk2H kxk2H + kyk2H = kzk2H ,

kEk 1.
62
3.
Definition 3.4.2. Let A, B L(H). We write

A
if
(Au, u)H 0, u H,
and in this case we say that A is positive. Finally, we denote
AB
if
A B .
Theorem 3.4.3. Let {An } be a sequence of self-adjoint commuting operators in L(H). Let B L(H) be a self adjoint operator
such that
Ai B = BAi , i N.
Suppose also that
A1 A2 A3 ... An ... B.
Under such hypotheses there exists a self adjoint, bounded, linear

operator A such that
An A in norm ,
and
A B.
Proof. Consider the sequence {Cn } where
Cn = B An 0, n N.
Fix u H. First, we show that {Cn u} converges. Observe that

Also, if n > m then
so that
Ci Cj = Cj Ci , i, j N.
An Am
Cm = B Am B An = Cn .
Therefore from Cm and Cm Cn we obtain
and also
Thus,
(Cm Cn )Cm , if n > m

Cn (Cm Cn ) .
2
(Cm
u, u)H (Cn Cm u, u)H (Cn2 u, u)H ,
3.4.
63
and we may conclude that

(Cn2 u, u)H
is a monotone non-increasing sequence of real numbers, bounded
below by 0, so that there exists R such that
lim (Cn2 u, u)H = .
Since each Cn is self adjoint we obtain

k(Cn Cm )uk2H
as
((Cn Cm )u, (Cn Cm )u)H

((Cn Cm )(Cn Cm )u, u)H
2
(Cn2 u, u)H 2(Cn Cm u, u) + (Cm
u, u)H
2 + = 0,
(3.6)
=
=
=
m, n .
Therefore {Cn u} is a Cauchy sequence in norm, so that there exists
the limit
lim Cn u = lim (B An )u,
n
and hence there exists

lim An u, u H.
Now define A by
Au = lim An u.
n
Since the limit

lim An u, u H
exists we have that

sup{kAn ukH }
nN
is finite for all u H. By the principle of uniform boundedness

sup{kAn k} <
nN
so that there exists K > 0 such that

Therefore
so that
kAn k K, n N.
kAn ukH KkukH ,
kAuk = lim {kAn ukH } KkukH , u H
n
64
3.
which means that A is bounded. Fixing u, v H, we have

(Au, v)H = lim (An u, v)H = lim (u, An v)H = (u, Av)H ,
n
and thus A is self adjoint. Finally

(An u, u)H (Bu, u)H , n N,
so that
(Au, u) = lim (An u, u)H (Bu, u)H , u H.
n
Hence A B.
Definition 3.4.4. Let A L(A) be a positive operator. The

self adjoint operator B L(H) such that
B2 = A
is called the square root of A. If B we denote
B = A.
Theorem 3.4.5. Suppose A L(H) is positive. Then there
exists B such that
B 2 = A.
Furthermore B commutes with any C L(H) such that commutes
with A.
Proof. There is no loss of generality in considering
kAk 1,
which means A I, because we may replace A by
so that if
A
kAk
C2 =
then
Let
A
kAk
B = kAk1/2 C.
B0 = ,
3.4.
65
and consider the sequence of operators given by

1
Bn+1 = Bn + (A Bn2 ), n N {0}.
2
Since each Bn is polynomial in A, we have that Bn is self adjoint and
commute with any operator with commutes with A. In particular
First we show that
Bi Bj = Bj Bi , i, j N.
Bn I, n N {0}.
Since B0 = , and B1 = 12 A, the statement holds for n = 1. Suppose

Bn I. Thus
1
1
I Bn+1 = I Bn A + Bn2
2
2
1
1
=
(I Bn )2 + (I A)
(3.7)
2
2
so that
Bn+1 I.
The induction is complete, that is,
Bn I, n N.
Now we prove the monotonicity also by induction. Observe that

and supposing
we have
B0 B1 ,
Bn1 Bn ,
1
1
2
Bn+1 Bn = Bn + (A Bn2 ) Bn1 (A Bn1
)
2
2
1
2
= Bn Bn1 (Bn2 Bn1
)
2
1
= Bn Bn1 (Bn + Bn1 )(Bn Bn1 )
2
1
= (I (Bn + Bn1 ))(Bn Bn1 )
2
1
((I Bn1 ) + (I Bn ))(Bn Bn1 ) .
=
2
The induction is complete, that is
= B0 B1 B2 ... Bn ... I.
66
3.
By the last theorem there exists a self adjoint operator B such that
Bn B in norm.
Fixing u H we have
1
Bn+1 u = Bn u + (A Bn2 )u,
2
so that taking the limit in norm as n , we get
= (A B 2 )u.
Being u H arbitrary we obtain
A = B 2.
It is also clear that

B0
3.5. About the Spectrum of a Linear Operator

Definition 3.5.1. Let U be a Banach space and let A L(U).
A complex number is said to be in the resolvent set (A) of A, if
I A
is a bijection with a bounded inverse. We call
R (A) = (I A)1
the resolvent of A in .
If 6 (A), we write
(A) = C (A),
where (A) is said to be the spectrum of A.
Definition 3.5.2. Let A L(U).
(1) If u 6= and Au = u for some C then u is said to be

an eigenvector of A and the corresponding eigenvalue.
If is an eigenvalue, then (I A) is not injective and
therefore (A).
The set of eigenvalues is said to be the point spectrum
of A.
3.5.
ABOUT THE SPECTRUM OF A LINEAR OPERATOR
67
(2) If is not an eigenvalue but

R(I A)
is not dense in U and therefore I A is not a bijection,

we have that (A). In this case we say that is in the
residual spectrum of A, or briefly Res[(A)].
Theorem 3.5.3. Let U be a Banach space and suppose that
A L(U). Then (A) is an open subset of C and
F () = R (A)
is an analytic function with values in L(U) on each connected component of (A). For , (A), R (A) and R (A) commute
and
R (A) R (A) = ( )R (A)R (A).
Proof. Let 0 (A). We will show that 0 is an interior point
of (A).
Observe that symbolically we may write
1
1
=
A
0 + (0 A)
1
1

=
0 A 1 0
0 A
!
X 0 n
1
=
1+
.
(3.8)
0 A
0 A
n=1
Define,
(A) = R0 (A){I +
R
X
n=1
Observe that
( 0 )n (R0 )n }.
(3.9)
k(R0 )n k kR0 kn .
Thus, the series indicated in (3.9) will converge in norm if
| 0 | < kR0 k1 .
(3.10)
Hence, for satisfying (3.10), R(A)

is well defined and we can easily
check that
(A) = I = R
(A)(I A).
(I A)R
68
3.
Therefore
(A) = R (A), if | 0 | < kR0 k1 ,
R
so that 0 is an interior point. Since 0 (A) is arbitrary, we have

that (A) is open. Finally, observe that
R (A) R (A) = R (A)(I A)R (A) R (A)(I A)R (A)
= R (A)(I)R (A) R (A)(I)R (A)
= ( )R (A)R (A)
(3.11)
Interchanging the roles of and we may conclude that R and
R commute.

Corollary 3.5.4. Let U be a Banach space and A L(U).
Then the spectrum of A is non-empty.
Proof. Observe that if
kAk
<1
||
we have
(I A)1 = [(I A/)]1

= 1 (I A/)1
!
n
X
A
= 1 I +
.
n=1
(3.12)
Therefore we may obtain
R (A) = 1
I+
n
X
A
n=1
In particular
kR (A)k 0, as || .
(3.13)

(A) = .
In such a case R (A) would be a entire bounded analytic function.

From Liouvilles theorem, R (A) would be constant, so that from
(3.13) we would have
R (A) = , C,
which is a contradiction.
3.5.
ABOUT THE SPECTRUM OF A LINEAR OPERATOR
69
Proposition 3.5.5. Let H be a Hilbert space and A L(H).

(1) If Res[(A)] then P (A ).
(2) If P (A) then P (A ) Res[(A )].
Proof.
(1) If Res[(A)] then

R(A I) 6= H.
Therefore there exists v (R(A I)) , v 6= such that

(v, (A I)u)H = 0, u H
that is
((A I)v, u)H = 0, u H
so that
(A I)v = ,
which means that P (A ).

(2) Suppose there exists v 6= such that
(A I)v = ,
and
Thus
so that
Since
6 P (A ).
(u, (A I)v))H = 0, u H,
((A I)u, v)H , u H.
(A I)u 6= , u H, u 6= ,
we get v (R(A I)) , so that R(A I) 6= H.

Hence Res[(A )].
Theorem 3.5.6. Let A L(H) be a self-adjoint operator.

then
(1) (A) R.
(2) Eigenvectors corresponding to distinct eigenvalues of A are
orthogonal.
70
3.
Proof. Let , R. Thus, given u H we have
k(A ( + i))uk2 = k(A )uk2 + 2 kuk2 ,
so that
k(A ( + i))uk2 2 kuk2 .
Therefore if 6= 0, A ( + i) has a bounded inverse on its range,

which is closed. If R(A ( + i)) 6= H then by the last result
( i) would be in the point spectrum of A, which contradicts
the last inequality. Hence, if 6= 0 then + i (A). To complete
the proof, suppose
Au1 = 1 u1 ,
and
Au2 = 2 u2 ,
where
Thus
1 , 2 R, 1 6= 2 and u1 , u2 6= .
(1 2 )(u1 , u2 )H =
=
=
=
=
1 (u1 , u2 )H 2 (u1 , u2 )H
(1 u1 , u2 )H (u1 , 2 u2 )H
(Au1 , u2 )H (u1 , Au2 )H
(u1 , Au2 )H (u1 , Au2 )H
0.
(3.14)
Since 1 2 6= 0 we get
(u1 , u2 )H = 0.

3.6. The Spectral Theorem for Bounded
Self-Adjoint Operators
Let H be a complex Hilbert space. Consider A : H H a
linear bounded operator, that is A L(H), and suppose also that
such an operator is self-adjoint. Define
m = inf {(Au, u)H | kukH = 1},
uH
and
M = sup{(Au, u)H | kukH = 1}.
uH
3.6. THE SPECTRAL THEOREM FOR BOUNDED...
71
Remark 3.6.1. It is possible to prove that for a linear selfadjoint operator A : H H we have
kAk = sup{|(Au, u)H | | u H, kukH = 1}.
This propriety, which we do not prove in the present work, is crucial

for the subsequent results, since for example for A, B linear and self
adjoint and > 0 we have
we also would have
I A B I,
kA Bk < .
Define by P the set of all real polynomials defined in R. Define

1 : P L(H),
by
1 (p()) = p(A), p P.
Thus we have
(1) 1 (p1 + p2 ) = p1 (A) + p2 (A),
(2) 1 (p1 p2 ) = p1 (A)p2 (A),
(3) 1 (p) = p(A), R, p P
(4) if p() 0, on [m, M], then p(A) ,
We will prove (4):
Consider p P . Denote the real roots of p() less or equal to
m by 1 , 2 , ..., n and denote those that are greater or equal to
M by 1 , 2 , ..., l . Finally denote all the remaining roots, real or
complex by
v1 + i1 , ..., vk + ik .
Observe that if i = 0 then vi (m, M). The assumption that
p() 0 on [m, M] implies that any real root in (m, M) must be
of even multiplicity.
Since complex roots must occur in conjugate pairs, we have the
following representation for p() :
n
l
k
Y
Y
Y
p() = a ( i ) (i ) (( vi )2 + 2i ),
i=1
where a 0. Observe that

since,
i=1
i=1
A i I ,
(Au, u)H m(u, u)H i (u, u)H , u H,
72
3.
and by analogy
i I A .
On the other hand, since Avk I is self-adjoint, its square is positive

and hence since the sum of positive operators is positive, we obtain
(A vk I)2 + 2k I .
Therefore
p(A) .
The idea is now to extend de domain of 1 to the set of upper

semi-continuous functions, and such set we will denote by C up .
Observe that if f C up , there exists a sequence of continuous
functions {gn } such that
gn f, pointwise ,
that is
gn () f (), R.
Considering the Weierstrass Theorem, since gn C([m, M]) we

may obtain a sequence of polynomials {pn } such that

g n + 1 pn < 1 ,

2n
2n
where the norm k k refers to [m, M]. Thus
pn () f (), on [m, M].

Therefore
p1 (A) p2 (A) p3 (A) ... pn (A) ...
Since pn (A) is self-adjoint for all n N, we have
pj (A)pk (A) = pk (A)pj (A), j, k N.
Then the lim pn (A) (in norm) exists, and we denote
n
lim pn (A) = f (A).
Now recall the Dinis Theorem:

Theorem 3.6.2 (Dini). Let {gn } be a sequence of continuous
functions defined on a compact set K R. Suppose gn g
point-wise and monotonically on K. Under such assumptions the
convergence in question is also uniform.
73
Now suppose that {pn } and {qn } are sequences of polynomial

such that
pn f, and qn f,
we will show that
lim pn (A) = lim qn (A).
First observe that being {pn } and {qn } sequences of continuous

functions we have that
hnk () = max{pn (), qk ()}, [m, M]
is also continuous, n, k N. Now fix n N and define
hk () = max{pk (), qn ()}.
observe that
hk () qn (), R,
so that by Dinis theorem
hk qn , uniformly on [m, M].

It follows that for each n N there exists kn N such that if
k > kn then
1
hk () qn () , [m, M].
n
Since
pk () hk (), [m, M],
we obtain
1
, [m, M].
n
By analogy, we may show that for each n N there exists kn N
such that if k > kn then
pk () qn ()
qk () pn ()
1
.
n
From above we obtain

lim pk (A) qn (A) +
1
.
n
Since the self adjoint qn (A) + 1/n commutes with the

lim pk (A)
74
3.
we obtain

1
lim pk (A) lim qn (A) +
n
k
n
lim qn (A).
(3.15)
Similarly we may obtain

lim qk (A) lim pn (A),
n
so that
lim qn (A) = lim pn (A) = f (A).
Hence, we may extend 1 : P L(H) to 2 : C up L(H) where

C up as earlier indicated, denotes the set of upper semi-continuous
functions, where
2 (f ) = f (A).
Observe that 2 has the following properties
(1) 2 (f1 + f2 ) = 2 (f1 ) + 2 (f2 ),
(2) 2 (f1 f2 ) = f1 (A)f2 (A),
(3) 2 (f ) = 2 (A), R, 0.
(4) if f1 () f2 (), [m, M], then
f1 (A) f2 (A).
The next step is to extend 2 to 3 : Cup L(H), where

Cup = {f g | f, g C up }.
For h = f g Cup we define
3 (h) = f (A) g(A).
Now we will show that 3 is well defined. Suppose that h Cup

and
h = f1 g1 and h = f2 g2 .
Thus
f1 g1 = f2 g2 ,
that is
f1 + g2 = f2 + g1 ,
so that from the definition of 2 we obtain
f1 (A) + g2 (A) = f2 (A) + g1 (A),
that is
f1 (A) g1 (A) = f2 (A) g2 (A).
75
Therefore 3 is well defined. Finally observe that for < 0

(f g) = g ()f,
where g C up and f C up . Thus
3 (f ) = f (A) = 3 (f ), R.
3.6.1. The Spectral Theorem. Consider the upper semicontinuous function

1, if ,
h () =
(3.16)
0, if > .
Denote
E() = 3 (h ) = h (A).
Observe that
so that
Therefore
h ()h () = h (), R,
[E()]2 = E(), R.
{E() | R}
is a family of orthogonal projections. Also observe that if we
have
h ()h () = h ()h () = h (),
so that
E()E() = E()E() = E(), .
If < m, then h () = 0, on [m, M], so that
E() = 0, if < m.
Similarly, if M them h () = 1, on [m, M], so that
E() = I, if M.
Next we show that the family {E()} is strongly continuous from

the right. First we will establish a sequence of polynomials {pn }
such that
pn h ,
and
pn () h+ 1 (), on [m, M].
n
Observe that for any fixed n there exists a sequence of polynomials

{pnj } such that
pnj h+1/n , point-wise .
76
3.
Consider the monotone sequence

gn () = min{prs () | r, s {1, ..., n}}.
Thus
gn () h+ 1 (), R,
n
and we obtain
lim gn () lim h+ 1 () = h ().
On the other hand

gn () prn (), R, r {1, ..., n},
so that
lim gn () lim prn ().
Therefore
lim gn ()
lim lim prn ()
r n
= h ().
(3.17)
Thus
lim gn () = h ().
Observe that gn are not necessarily polynomials. To set a sequence of polynomials, observe that we may obtain a sequence {pn }
of polynomials such that
1
|gn () + 1/n pn ()| < n , [m, M], n N.
2
so that
Thus
pn () gn () + 1/n 1/2n gn () h+1/n ().

pn (A) E(),
and
pn (A) h+ 1 (A) = E( + 1/n) E().

n
Therefore we may write
E() = lim pn (A) lim E( + 1/n) E().

n
Thus
lim E( + 1/n) = E().
From this we may easily obtain the strong continuity from the right.
77
For we have
(h () h ()) (h () h ())
(h () h ()).
(3.18)
To verify this observe that if < or > then all terms involved
in the above inequalities are zero. On the other hand if
then
h () h () = 1,
so that in any case (3.18) holds. From the monotonicity property

we have
(E() E()) A(E() E())
(E() E()).
(3.19)
Now choose a, b R such that

a < m and b M.
Suppose given > 0. Choose a partition P0 of [a, b], that is
P0 = {a = 0 , 1 , ..., n = b},
such that
max {|k k1 |} < .
k{1,...,n}
Hence
k1(E(k ) E(k1 )) A(E(k ) E(k1 ))
k (E(k ) E(k1 )). (3.20)
Summing up on k and recalling that
n
X
k=1
E(k ) E(k1 ) = I,
we obtain
n
X
k1(E(k ) E(k1 )) A
k=1
n
X
k=1
k (E(k ) E(k1 )).

(3.21)
78
3.
Let 0k [k1 , k ]. Since (k 0k ) (k k1) from (3.20) we

obtain
n
n
X
X
0
A
k (E(k ) E(k1 ))
(E(k ) E(k1 ))
k=1
k=1
= I.
(3.22)
0k (E(k ) E(k1 )).
(3.23)
By analogy
I A
Since
A
n
X
k=1
n
X
k=1
0k (E(k ) E(k1))
is self-adjoint we obtain
kA
n
X
k=1
0k (E(k ) E(k1))k < .
Being > 0 arbitrary, we may write

Z b
A=
dE(),
a
that is
A=
dE().
3.7. The Spectral Decomposition of

Unitary Transformations
Definition 3.7.1. Let H be a Hilbert space. A transformation
U : H H is said to be unitary if
(Uu, Uv)H = (u, v)H , u, v H.
Observe that in this case
U U = UU = I,
so that
U 1 = U .
3.7. THE SPECTRAL DECOMPOSITION OF...
79
Theorem 3.7.2. Every Unitary transformation U has a spectral decomposition

Z 2
U=
ei dE(),
0
where {E()} is a spectral family on [0, 2]. Furthermore E() is

continuous at 0 and it is the limit of polynomials in U and U 1 .
We present just a sketch of the proof. For the trigonometric
polynomials
n
X
i
p(e ) =
ck eik ,
k=n
consider the transformation
p(U) =
n
X
ck U k ,
k=n
where ck C, k {n, ..., 0, ..., n}.

Observe that
n
X
i
p(e ) =
ck eik ,
k=n
so that the corresponding operator is

n
n
X
X
k
p(U) =
ck U =
ck (U )k .
k=n
k=n
Also if
p(ei ) 0
there exists a polynomial q such that
so that
p(ei ) = |q(ei )|2 = q(ei )q(ei ),

p(U) = [q(U)] q(U).
Therefore
(p(U)v, v)H = (q(U) q(U)v, v)H = (q(U)v, q(U)v)H 0, v H,
which means
p(U) 0.
Define the function h () by

1, if 2k < 2k + ,
h () =
0, if 2k + < 2(k + 1),
(3.24)
80
3.
for each k {0, 1, 2, 3, ...}. Define E() = h (U). Observe

that the family {E()} are projections and in particular
E(0) = 0,
E(2) = I
and if , since
h () h (),
we have
E() E().
Suppose given > 0. Let P0 be a partition of [0, 2] that is,

P0 = {0 = 0 , 1 , ..., n = 2}
such that
max {|j j1 |} < .
j{1,...,n}
For fixed [0, 2], let j {1, ..., n} be such that

[j1 , j ].
|ei
n
X
k=1
eik (hk () hk1 ())| = |ei eij |

| j | < . (3.25)
Thus,
i
0 |e
n
X
k=1
eik (hk () hk1 ())|2 2
so that, for the corresponding operators

0 [U
n
X
k=1
ik
(E(k )E(k1 )] [U
and hence
kU
n
X
k=1
n
X
eik (E(k )E(k1 )]
k=1
eik (E(k ) E(k1 )k < .
Being > 0 arbitrary, we may infer that

Z 2
U=
ei dE().
0
2 I
3.8. UNBOUNDED OPERATORS
81
3.8. Unbounded Operators

3.8.1. Introduction. Let H be a Hilbert space. Let A : D(A)
H be an operator, where unless indicated D(A) is a dense subset
of H. We consider in this section the special case where A is unbounded.
Definition 3.8.1. Given A : D H we define the graph of
A, denoted by (A) by,
(A) = {(u, Au) | u D}.
Definition 3.8.2. An operator A : D H is said to be closed
if (A) is closed.
Definition 3.8.3. Let A1 : D1 H and A2 : D2 H
operators. We write A2 A1 if D2 D1 and
A2 u = A1 u, u D1 .
In this case we say that A2 is an extension of A1 .

Definition 3.8.4. A linear operator A : D H is said to
be closable if it has a linear closed extension. The smallest closed
extension of A is denote by A and is called the closure of A.
Proposition 3.8.5. Let A : D H be a linear operator. If
A is closable then
(A) = (A).
Proof. Suppose B is a closed extension of A. Then
(A) (B) = (B),
so that if (, ) (A) then (, ) (B), and hence = . Define
the operator C by
D(C) = { | (, ) (A) for some },
and C() = , where is the unique point such that (, ) (A).
Hence
(C) = (A) (B),
so that
A C.
82
3.
However C B and since B is an arbitrary closed extension of A

we have
C=A
so that
(C) = (A) = (A).

Definition 3.8.6. Let A : D H be a linear operator where
D is dense in H. Define D(A ) by
D(A ) = { H | (A, )H = (, )H , D for some H}.
In this case we denote
A = .
A defined in this way is called the adjoint operator related to A.
Observe that by the Riesz lemma, D(A ) if and only if there
exists K > 0 such that
|(A, )H | KkkH , D.
Also note that if
A B then B A .
Finally, as D is dense in H then
= A ()
is uniquely defined. However the domain of A may not be dense,
and in some situations we may have D(A ) = {}.
If D(A ) is dense we define
A = (A ) .
Theorem 3.8.7. Let A : D H a linear operator, being D
dense in H. Then
(1) A is closed,
(2) A is closable if and only if D(A ) is dense and in this case
A = A .
(3) If A is closable then (A) = A .
3.8. UNBOUNDED OPERATORS
83
(1) We define the operator V : H H H H
Proof.
by
V (, ) = (, ).
Let E H H be a subspace. Thus if (1 , 1 ) V (E )
then there exists (, ) E such that
V (, ) = (, ) = (1 , 1 ).
Hence
= 1 and = 1 ,
so that for (1 , 1 ) E and (w1 , w2 ) E we have
((1 , 1 ), (w1, w2 ))HH = 0 = (1 , w1 )H + (1 , w2 )H .

Thus
(1 , w2 )H + (1 , w1)H = 0,
and therefore
((1 , 1 ), (w2 , w1 ))HH = 0,
that is
((1 , 1 ), V (w1 , w2 ))HH = 0, (w1 , w2 ) E.
This means that

so that
(1 , 1 ) (V (E)) ,
V (E ) (V (E)) .
It is easily verified that the implications from which the
last inclusion results are in fact equivalences, so that
V (E ) = (V (E)) .
Suppose (, ) H H. Thus (, ) V ((A)) if
and only if
((, ), (A, ))HH = 0, D,
which holds if and only if
(, A)H = (, )H , D,
that is, if and only if

Thus
(, ) (A ).
(A ) = V ((A)) .
Since (V ((A)) is closed, A is closed.
84
3.
(2) Observe that (A) is a linear subset of H H so that

(A) = [(A) ]
= V 2 [(A) ]
= [V [V ((A)) ]]
= [V ((A )]
(3.26)
so that from the proof of item 1, if A is densely defined

we get
(A) = [(A ) ].
Conversely, suppose D(A ) is not dense. Thus there exists
[D(A )] such that 6= . Let (, A ) (A ).
Hence
((, ), (, A))HH = (, )H = 0,
so that
(, ) [(A)] .
Therefore V [(A)] is not the graph of a linear operator.
Since (A) = V [(A)] A is not closable.
(3) Observe that if A is closable then
A = (A ) = A = (A) .

3.9. Symmetric and Self-Adjoint Operators
Definition 3.9.1. Let A : D H be a linear operator, where
D is dense in H. A is said to be symmetric if A A , that is if
D D(A ) and
A = A, D.
Equivalently, A is symmetric if and only if
(A, )H = (, A)H , , D.
Definition 3.9.2. Let A : D H be a linear operator. We
say that A is self-adjoint if A = A , that is if A is symmetric and
D = D(A ).
Definition 3.9.3. Let A : D H be a symmetric operator.
We say that A is essentially self-adjoint if its closure A is selfadjoint. If A is closed, a subset E D is said to be a core for A if
A|E = A.
3.9. SYMMETRIC AND SELF-ADJOINT OPERATORS
85
Theorem 3.9.4. Let A : D H be a symmetric operator.

Then the following statements are equivalent
(1) A is self-adjoint.
(2) A is closed and N(A iI) = {}.
(3) R(A iI) = H.
1 implies 2:
Suppose A is self-adjoint let D = D(A ) be such
that
A = i
Proof.
so that
A = i.
Observe that
i(, )H =
=
=
=
=
(i, )H
(A, )H
(, A)H
(, i)H
i(, ),
(3.27)
so that (, )H = 0, that is = . Thus

N(A iI) = {}.
Similarly we prove that N(A + iI) = {}. Finally, since
A = A = A, we get that A = A is closed.
2 implies 3:
Suppose 2 holds. Thus the equation
A = i
has no non trivial solution. We will prove that R(A iI)
is dense in H. If R(A iI) then
((A iI), )H = 0, D,
so that D(A ) and
(A iI) = (A + iI) = ,
and hence by above = . Now we will prove that R(A

iI) is closed and conclude that
R(A iI) = H.
86
3.
Given D we have
k(A iI)k2H = kAk2H + kk2H .
Thus if {n } D is such that
(A iI)n 0 ,
we conclude that
n 0 ,
for some 0 and {An } is also convergent. Since A is

closed, 0 D and
(A iI)0 = 0 .
Therefore R(A iI) is closed, so that

Similarly
R(A iI) = H.
R(A + iI) = H.
3 implies 1: Let D(A ). Since R(A iI) = H, there is
an D such that
(A iI) = (A iI),
and since D D(A ) we obtain D(A ), and

(A iI)( ) = .
Since R(A + iI) = H we have N(A iI) = {}. Therefore

= , so that D(A ) = D. The proof is complete.

3.9.1. The Spectral Theorem Using Cayley Transform.
In this section H is a complex Hilbert space. We suppose A is
defined on a dense subspace of H, being A self-adjoint but possibly
unbounded. We have shown that (A + i) and (A i) are onto H
and it is possible to prove that
U = (A i)(A + i)1 ,
exists on all H and it is unitary. Furthermore on the domain of A,

A = i(I + U)(I U)1 .
The operator U is called the Cayley transform of A. We have

already proven that
Z 2
U=
ei dF (),
0
87
where {F ()} is a monotone family of orthogonal projections, strongly

continuous from the right and we may consider it such that

0, if 0,
F () =
(3.28)
I, if 2.
Since F () = 0, for all 0 and
F (0) = F (0+ )
we obtain
F (0+ ) = 0 = F (0 ),
that is, F () is continuous at = 0. We claim that F is continuous
at = 2. Observe that F (2) = F (2 + ) so that we need only to
show that
F (2 ) = F (2).
Suppose
F (2) F (2 ) 6= .
Thus there exists some u, v H such that
(F (2) F (2( )))u = v 6= .
Therefore
so that
F ()v = F ()[(F (2) F (2 ))u],

F ()v =
Observe that
U I =
and
U I =
0, if < 2,
v, if 2.
(3.29)
(ei 1)dF (),
(ei 1)dF ().
Let {n } be a partition of [0, 2]. From the monotonicity of [0, 2]

and pairwise orthogonality of
{F (n ) F (n1 )}
we can show that (this is not proved in details here)
Z 2
(U I)(U I) =
(ei 1)(ei 1)dF (),
0
88
3.
so that, given z H we have
((U I)(U I)z, z)H =

thus, for v defined above
|ei 1|2 dkF ()zk2 ,
k(U I)vk2 = ((U I)v, (U I)v)H

= ((U I) (U I)v, v)H
Z 2
=
|ei 1|2 dkF ()vk
0
|ei 1|2 dkF ()vk
= 0
(3.30)
The last two equalities results from e2i 1 = 0 and dkF ()vk =
on [0, 2). since v 6= the last equation implies that 1 P (U),
which contradicts the existence of
(I U)1 .
Thus, F is continuous at = 2.
Now choose a sequence of real numbers {n } such that n
(0, 2), n = 0, 1, 2, 3, ... such that

n
= n.
cot
2
Now define Tn = F (n ) F (n1 ). Since U commutes with F (),
U commutes with Tn . since
A = i(I + U)(I U)1 ,
this implies that the range of Tn is invariant under U and A. Observe that
X
X
Tn =
(F (n ) F (n1 ))
n
lim F () lim F ()
= I = I.
Hence
X
n
R(Tn ) = H.
(3.31)
89
Also, for u H we have that
0, if < n1,
(F () F (n1 ))u, if n1 n ,
F ()Tn u =
F (n ) F (n1 ))u, if > n ,
(3.32)
so that
(I U)Tn u =
(1 ei )dF ()Tn u
n1
Therefore
Z
n1
n
(1 ei )dF ()u.
(1 ei )1 dF ()(I U)Tn u
Z
(1 e ) dF ()
(1 ei )1 (1 ei )dF ()u
dF ()u
i 1
n1
n
n1
n
n1
(1 ei )dF ()u
n1
Z 2
dF ()Tn u = Tn u.
(3.34)
Hence
(I U)|R(Tn )
1
n1
(1 ei )1 dF ().
also from the theory earlier developed we may prove that

Z 2
f (U) =
f (ei )dF ().
0
From this, from above and as
A = i(I + U)(I U)1
we obtain
ATn u =
Therefore defining
(3.33)
n
n1
i(1 + ei )(1 ei )1 dF ()u.

= cot
,
2
90
3.
and
E() = F (2cot1 ),
we get
i
i 1
i(1 + e )(1 e )
Hence,
ATn u =
Finally, from
u=

= .
= cot
2
dE()u.
n1
Tn u,
n=
we can obtain
Au = A(
Tn u)
n=
=
=
ATn u
n=
Z n
X
n=
dE()u.
(3.35)
n1
Being the convergence in question in norm, we may write

Z
Au =
dE()u.
Since u H is arbitrary, we may denote

Z
A=
dE().
CHAPTER 4
Measure and Integration

4.1. Basic Concepts
In this chapter U denotes a topological space.
Definition 4.1.1 (-Algebra ). A collection M of subsets of
U is said to be a -Algebra if M has the following properties:
(1) U M,
(2) if A M then U A M,
(3) if An M, n N, then
n=0 An M.
Definition 4.1.2 (Measurable Spaces). If M is a -algebra
in U we say that U is a measurable space. The elements of M are
called the measurable sets of U.
Definition 4.1.3 (Measurable Function). If U is a measurable
space and V is a topological space, we say that f : U V is a
measurable function if f 1 (V) is measurable whenever V V is an
open set.
Remark 4.1.4.
(1) Observe that = U U so that from
1 and 2 in Definition 4.1.1, we have that M.
(2) From 1 and 3 from Definition 4.1.1, it is clear that ni=1 Ai
M whenever Ai M, i {1, ..., n}.
c c
(3) Since
i=1 Ai = (i=1 Ai ) also from Definition 4.1.1, it is
clear that M is closed under countable intersections.
(4) Since A B = B c A we obtain: if A, B M then
A B M.
Theorem 4.1.5. Let F be any collection of subsets of U.
Then there exists a smallest -algebra M0 in U such that F M0 .
91
92
4. MEASURE AND INTEGRATION
Proof. Let be the family of all -Algebras that contain F .

Since the set of all subsets in U is a -algebra, is non-empty.
Let M0 = M M , it is clear that M0 F , it remains to
prove that in fact M0 is a -algebra. Observe that:
(1) U M , M , so that, U M0 ,
(2) A M0 implies A M , M , so that Ac
M , M , which means Ac M0 ,
(3) {An } M0 implies {An } M , M , so that
n=1 An M , M , which means n=1 An M0 .

From Definition 4.1.1 the proof is complete.

Definition 4.1.6 (Borel Sets). Let U be a topological space,
considering the last theorem there exists a smallest -algebra in U,
denoted by B, which contains the open sets of U. The elements of
B are called the Borel sets.
Theorem 4.1.7. Suppose M is a -algebra in U and V is a
topological space. For f : U V , we have:
(1) If = {E V | f 1 (E) M}, then is a -algebra.
(2) If V = [, ], and f 1 ((, ]) M, for each R,
then f is measurable.
Proof.
(1) (a) V since f 1 (V ) = U and U M.
(b) E f 1 (E) M U f 1 (E) M
f 1 (V E) M V E .
1
(c) {Ei } f 1 (Ei ) M, i N
(Ei )
i=1 f
1
M f (i=1 Ei ) M i=1 Ei .
Thus is a -algebra.
(2) Define = {E [, ] | f 1 (E) M} from above
is a - algebra. Given R, let {n } be a real sequence
such that n as n , n < , n N . Since
(n , ] for each n and
C
[, ) =
n=1 [, n ] = n=1 (n , ] ,
(4.1)
we obtain, [, ) . Furthermore, we have (, ) =

[, ) (, ] . Since every open set in [, ]
may be expressed as a countable union of intervals (, )
we have that contains all the open sets. Thus, f 1 (E)
M whenever E is open, so that f is measurable.
4.2. SIMPLE FUNCTIONS
93
Proposition 4.1.8. If {fn : U [, ]} is a sequence of

measurable functions and g = supn1 fn and h = lim sup fn then g
n
and h are measurable.
1
Proof. Observe that g 1((, ]) =
n=1 fn ((, ]). From
last theorem g is measurable. By analogy h = inf k1 {supik fi } is
measurable.
4.2. Simple Functions

Definition 4.2.1 (Simple Functions). A function f : U
C is said to be a simple function if its range (R(f )) has only
finitely many points. If {1 , ..., n } = R(f
P ) and we set Ai = {u
U | f (u) = i }, clearly we have: f = ni=1 i Ai , where

1, if u Ai ,
Ai (u) =
(4.2)
0, otherwise.
Theorem 4.2.2. Let f : U [0, ] be a measurable function.
Thus there exists a sequence of simple functions {sn : U [0, ]}
such that
(1) 0 s1 s2 ... f ,
(2) sn (u) f (u) as n , u U.
Proof. Define n = 2n . To each n N and each t R+ , there
corresponds a unique integer K = Kn (t) such that
Kn t (K + 1)n .
(4.3)
Defining
n (t) =
Kn (t)n , if 0 t < n,
n, if t n,
(4.4)
we have that each n is a Borel function on [0, ], such that

(1) t n < n (t) t if 0 t n,
(2) 0 1 ... t,
(3) n (t) t as n , t [0, ].
It follows that the sequence {sn = n f } corresponds to the results

indicated above.
94
4.3. Measures
Definition 4.3.1 (Measure). Let M be a -algebra on a
topological space U. A function : M [0, ] is said to be
a measure if () = 0 and is countably additive, that is, given
{Ai } U, a sequence of pairwise disjoint sets then
(
i=1 Ai )
(Ai ).
(4.5)
i=1
In this case (U, M, ) is called a measure space.

Proposition 4.3.2. Let : M [0, ], where M is a algebra of U. Then we have the following.
(1) (A1 ... An ) = (A1 ) + ... + (An ) for any given {Ai }
of pairwise disjoint measurable sets of M.
(2) If A, B M and A B then (A) (B).
(3) If {An } M, A =
n=1 An and
A1 A2 A3 ...
(4.6)
then, lim (An ) = (A).

n
(4) If {An } M, A =
n=1 An , A1 A2 A3 .... and
(A1 ) is finite then,
lim (An ) = (A).
(4.7)
Proof.
(1) Take An+1 = An+2 = .... = in Definition 4.1.1
item 1,
(2) Observe that B = A (B A) and A (B A) = so
that by above (A (B A)) = (A) + (B A) (A),
(3) Let B1 = A1 and let Bn = An An1 then Bn M,
Bi Bj = if i 6= j, An = B1 ... Bn and A =
i=1 Bi .
Thus
(A) =
(
i=1 Bi )
X
n=1
(Bi ) = lim
n
X
(Bi )
i=1
= lim (An ). (4.8)

n
(4) Let Cn = A1 An . Then C1 C2 ..., (Cn ) = (A1 )

(An ), A1 A =
n=1 Cn .
4.4. INTEGRATION OF SIMPLE FUNCTIONS
95
Thus by 3 we have
(A1 ) (A) = (A1 A) = lim (Cn )
n
= (A1 ) lim (An ). (4.9)

n
4.4. Integration of Simple Functions

Definition 4.4.1 (Integral for Simple Functions). For s : U
[0, ], a measurable simple function, that is,
s=
n
X
i Ai ,
(4.10)
i=1
where
Ai (u) =
1, if u Ai ,
0, otherwise,
we define the integral of s over E M, denoted by

Z
n
X
s d =
i (Ai E).
E
(4.11)
R
s d as
(4.12)
i=1
The convention 0. = 0 is used here.
Definition 4.4.2 (Integral for Non-negative Measurable Functions). If f : U [0, ] is measurable,

for E M, we define the
R
integral of f on E, denoted by E f d, as
Z
Z
f d = sup{ sd},
(4.13)
sA
where
A = {s simple and measurable | 0 s f }.
(4.14)
Definition 4.4.3 (Integrals for Measurable Functions). For

a measurable f : U [, ] and E M, we define f + =
max{f,
R 0}, f = max{f, 0} and the integral of f on E, denoted
by E f d, as
Z
Z
Z
+
f d =
f d
f d.
E
96
Theorem 4.4.4 (Lebesgues Monotone Convergence Theorem).

Let {fn } be a sequence of real measurable functions on U and suppose that
(1) 0 f1 (u) f2 (u) ... , u U,
(2) fn (u) f (u) as n , u U.
Then,
(a) fR is measurable,
R
(b) U fn d U f d as n .
Proof. Since
[0, ] such that
fn d
fn+1 d, n N, there exists
fn d , as n ,
(4.15)
By Proposition 4.1.8, f is measurable, and since fn f we have

Z
Z
fn d
f d.
(4.16)
U
From (4.15) and (4.16), we obtain

Z
f d.
(4.17)
Let s be any simple function such that 0 s f , and let c R

such that 0 < c < 1. For each n N we define
En = {u U | fn (u) cs(u)}.
(4.18)
Clearly En is measurable and E1 E2 ... and U = nN En .

Observe that
Z
Z
Z
fn d
fn d c
sd.
(4.19)
U
En
En
Letting n and applying Proposition 4.3.2, we obtain

Z
Z
= lim
fn d c sd,
n
(4.20)
so that
sd, s simple and measurable such that 0 s f.

(4.21)
4.4. INTEGRATION OF SIMPLE FUNCTIONS
97
This implies
Z
f d.
(4.22)
From (4.17) and (4.22) the proof is complete.
The next result we do not prove it (it is a direct consequence of

last theorem). For a proof see [36].
Corollary 4.4.5. Let {fn } be a sequence of non-negative
measurable functions
P defined on U (fn : U [0, ], n N).
Defining f (u) =
n=1 fn (u), u U, we have
Z
Z
X
f d =
fn d.
U
n=1
Theorem 4.4.6 (Fatous Lemma). If {fn : U [0, ]} is a

sequence of measurable functions, then
Z
Z
lim inf fn d lim inf
fn d.
(4.23)
n
Proof. For each k N define gk : U [0, ] by

gk (u) = inf {fi (u)}.
(4.24)
gk fk
(4.25)
ik
Then
so that
Z
gk d
fk d, k N.
Also 0 g1 g2 ..., each gk is measurable, and
lim gk (u) = lim inf fn (u), u U.

n
From the Lebesgue monotone convergence theorem

Z
Z
Z
gk d =
lim inf fn d.
gk d = lim
lim inf
k
(4.26)
(4.27)
(4.28)
From (4.26) we have
lim inf
k
gk d lim inf {
k
fk d}.
(4.29)
98
Thus, from (4.28) and (4.29) we obtain

Z
Z
lim inf fn d lim inf
fn d.
U
(4.30)
Theorem 4.4.7 (Lebesgues Dominated Convergence Theorem). Suppose {fn } is sequence of complex measurable functions
on U such that
lim fn (u) = f (u), u U.
(4.31)
If there exists a measurable function g : U R+ such that

and |fn (u)| g(u), u U, n N, then
R
(1) U |fR| d < ,
(2) lim U |fn f | d = 0.
g d <
Proof.
(1) This inequality holds since f is measurable and
|f | g.
(2) Since 2g |fn f | 0 we may apply the Fatous Lemma
and obtain:
Z
Z
2gd lim inf (2g |fn f |)d,
(4.32)
n
so that
lim sup
n
Hence
lim
|fn f | d 0.
|fn f | d = 0.
(4.33)
(4.34)

We finish this section with an important remark:
Remark 4.4.8. In a measurable space U we say that a property holds almost everywhere (a.e.) if it holds on U except for a
set of measure zero. Finally, since integrals are not changed by the
redefinition of the functions in question on sets of zero measure,
the proprieties of items (1) and (2) of the Lebesgues monotone
4.5. THE FUBINI THEOREM
99
convergence may be considered a.e. in U, instead of in all U. Similar remarks are valid for the Fatous lemma and the Lebesgues
dominated convergence theorem.
4.5. The Fubini Theorem
We start this section with the definition of complete measure
space.
Definition 4.5.1. We say that a measure space (U, M, ) is
complete if M contains all subsets of sets of zero measure. That is
if A M, (A) = 0 and B A then B M.
Definition 4.5.2. We say that C U is a semi-algebra in U
if the two conditions below are valid.
(1) if A, B C then A B C,
(2) For each A C, Ac is a finite disjoint union of elements in
C.
4.5.1. Product Measures. Let (U, M1 , 1 ) and (V, M2 , 2 )
be two complete measure spaces. We recall that the cartesian product between U and V , denoted by U V is defined by
U V = {(u, v) | u U and v V }.
If A U and B V we call A B a rectangle. If A M1
and B M2 we say that A B is a measurable rectangle. The
collection R of measurable rectangles is a semi-algebra since
(A B) (C D) = (A C) (B D),
and
(A B)c = (Ac B) (A B c ) (Ac B c ).
We define : M1 M2 R+ by
(A B) = 1 (A)2 (B).
Lemma 4.5.3. Let {Ai Bi }iN be a countable disjoint collection of measurable rectangles whose the union is the rectangle
A B. Then
X
(A B) =
1 (Ai )2 (Bi ).
i=1
100
Proof. Let u A. Thus each v B is such that (u, v) is

exactly in one Ai Bi . Therefore
X
AB (u, v) =
Ai (u)Bi (v).
i=1
Hence for the fixed u in question, from the corollary of Lebesgue

monotone convergence theorem we may write
Z
Z X
AB (u, v)d2(v) =
Ai (u)Bi (v)d2(v)
V
i=1
Ai (u)2 (Bi )
(4.35)
i=1
so that also from the mentioned corollary

Z
Z
X
1 (Ai )2 (Bi ).
d1 (u) AB (u, v)d2(v) =
U
i=1
Observe that
Z
Z
Z
Z
d1(u) AB (u, v)d2(v) =
d1 (u) A (u)B (v)d2 (v)
U
= 1 (A)2 (B).
From the last two equations we may write
X
(A B) = 1 (A)2 (B) =
1 (Ai )2 (Bi ).
i=1
Definition 4.5.4. Let E U V . We define Eu and Ev by

and
Eu = {v | (u, v) E},
Ev = {u | (u, v) E}.
Observe that
Eu (v) = E (u, v),
(E c )u = (Eu )c ,
and
for any collection {E }.
(E )u = (E )u ,
101
We denote by R as the collection of sets which are countable

unions of measurable rectangles. Also, R will denote the collection of sets which are countable intersections of elements of R .
Lemma 4.5.5. Let u U and E R . Then Eu is a measurable subset of V .
Proof. If E R the result is trivial. Let E R . Then E may
be expressed as a disjoint union
E =
i=1 Ei ,
where Ei R, i N. Thus,
Eu (v) = E (u, v)
= sup Ei (u, v)
iN
= sup (Ei )u (v).
(4.36)
iN
Since each (Ei )u is measurable we have that

(Ei )u (v)
is a measurable function of v, so that
Eu (v)
is measurable, which implies that Eu is measurable. Suppose now
E =
i=1 Ei ,
where Ei+1 Ei , i N. Then
Eu (v) = E (u, v)
= inf Ei (u, v)
iN
= inf (Ei )u (v).

iN
(4.37)
Thus as from above (Ei )u (v) is measurable for each i N, we have

that Eu is also measurable so that Eu is measurable.
Lemma 4.5.6. Let E be a set in R with (1 2 )(E) < .
Then the function g defined by
g(u) = 2 (Eu )
102
is a measurable function and

Z
g d1(u) = (1 2 )(E).
U
Proof. The lemma is true if E is a measurable rectangle. Let

{Ei } be a disjoint sequence of measurable rectangles and E =
i=1 Ei . Set
gi (u) = 2 ((Ei )u ).
Then each gi is a non-negative measurable function and
X
g=
gi .
i=1
Thus g is measurable, and by the corollary of the Lebesgue monotone convergence theorem, we have
Z
Z
X
g(u)d1(u) =
gi (u)d1(u)
U
i=1
X
i=1
(1 2 )(Ei )
= (1 2 )(E).
(4.38)
Let E be a set of finite measure in R . Then there is a sequence

in R such that
Ei+1 Ei
and
E =
i=1 Ei .
Let gi (u) = 2 ((Ei )u ), since
Z
g1 (u) = (1 2 )(E1 ) < ,
U
we have that
g1 (u) < a.e. in E1 .

For an u E1 such that g1 (u) < we have that {(Ei )u } is a
sequence of measurable sets of finite measure whose intersection is
Eu . Thus
g(u) = 2 (Eu ) = lim 2 ((Ei )u ) = lim gi (u),
i
that is,
gi g, a.e. in E.
(4.39)
103
We may conclude that g is also measurable. Since

0 gi g, i N
the Lebesgue dominated convergence theorem implies that

Z
Z
g(u) d1(u) = lim gi d1(u) = lim (1 2 )(Ei )
E
= (1 2 )(E).
Lemma 4.5.7. Let E be a set such that (1 2 )(E) = 0.

then for almost all u U we have
2 (Eu ) = 0.
Proof. Observe that there is a set in R such that E F and

From the last lemma
(1 2 )(F ) = 0.
2 (Fu ) = 0
fora almost all u. From Eu Fu we obtain
2 (Eu ) = 0
for almost all u, since 2 is complete.
Proposition 4.5.8. Let E be a measurable subset of U V

such that (1 2 )(E) is finite. The for almost all u the set Eu is
a mesurable subset of V . The function g defined by
g(u) = 2 (Eu )
is measurable and
Z
g d1 (u) = (1 2 )(E).
Proof. First observe that there is a set F R such that

E F and
(1 2 )(F ) = (1 2 )(E).
Let G = F E. Since F and G are measurable, G is measurable,
and
(1 2 )(G) = 0.
By the last lemma we obtain
2 (Gu ) = 0,
104
for almost all u so that

g(u) = 2 (Eu ) = 2 (Fu ), a.e. in U.
By Lemma 4.5.6 we may conclude that g is measurable and
Z
g d1 (u) = (1 2 )(F ) = (1 2 )(E).
Theorem 4.5.9 (Fubini). Let (U, M1 , 1 ) and (V, M2 , 2 )
be two complete measure spaces and f an integrable function on
U V . Then
(1) fu (v) = f (u, v) is measurable and integrable for almost all

u.
(2) fv (u) = f (u, v) is measurable and integrable for almost all
v.
R
(3) h1 (u) = R V f (u, v) d2 (v) is integrable on U.
(4) h2 (v) = U f (u, v) d1(u) is integrable on V .
(5)

Z Z
Z Z
f d2(v) d1 (u) =
f d1 (u) d2 (v)
U
V
V
U
Z
=
f d(1 2 ).
(4.40)
U V
Proof. It suffices to consider the case where f is non-negative

(we can then apply the result to f + = max(f, 0) and f = max(f, 0)).
The last proposition asserts that the theorem is true if f is a simple
function which vanishes outside a set of finite measure. Similarly
as in in Theorem 4.2.2, we may obtain a sequence of non-negative
simple functions {n } such that
n f.
Observe that, given u U, fu is such that
(n )u fu , a.e. .
By the Lebesgue monotone convergence theorem we get
Z
Z
f (u, v) d2 (v) = lim
n (u, v) d2 (v),
V
4.6. THE LEBESGUE MEASURE IN Rn
105
so that this last resulting function is integrable in U. Again by the

Lebesgue monotone convergence theorem, we obtain

Z Z
Z Z
n d2 (v) d1 (u)
f d2 (v) d1 (u) = lim
n U
V
U
V
Z
= lim
n d(1 2 )
n U V
Z
=
f d(1 2 ).
(4.41)
U V
4.6. The Lebesgue Measure in Rn

In this section we will define the Lebesgue measure and the concept of Lebesgue measurable set. We show that the set of Lebesgue
measurable sets is a algebra so that the earlier results, proven
for more general measure spaces, remain valid in the present context (such as the Lebesgue monotone and dominated convergence
theorems).
We start with the following theorems without proofs.
Theorem 4.6.1. Every open set A R may be expressed as
a countable union of disjoint open intervals.
Remark 4.6.2. In this text Qj denotes
a closed cube in Rn
Qn
and |Qj | its volume, that is, |Qj | = i=1 (bi ai ), where Qj =
Q
n
i=1 [ai , bi ]. Also we assume that if two Q1 and Q2 closed or not,
1 |. We recall that two
have the same interior, then |Q1 | = |Q2 | = |Q
n
cubes Q1 , Q2 R are said to be quasi-disjoint if their interiors are
disjoint.
Theorem 4.6.3. Every open set A Rn , where n 1 may
be expressed as a countable union of quasi-disjoint closed cubes.
Definition 4.6.4 (Outer Measure). Let E Rn . The outer
measure of E, denoted by m (E), is defined by
)
(
X
,
m (E) = inf
|Qj | : E
j=1 Qj
j=1
where Qj is a closed cube, j N.
106
4.6.1. Properties of the Outer Measure. First observe that

given > 0, there exists a sequence {Qj } such that
E
j=1 Qj
and
X
j=1
|Qj | m (E) + .
(1) Monotonicity: If E1 E2 then m (E1 ) m (E2 ). This

follows from the fact that if E2
j=1 Qj then E1
j=1 Qj .
(2) P
Countable sub-additivity : If E
j=1 Ej , then m (E)
j=1 m (Ej ).
Proof. First assume that m (Ej ) < , j N, otherwise the result is obvious. Thus, given > 0 for each
j N there exists a sequence {Qk,j }kN such that
Ej
k=1 Qk,j
and
X
k=1
Hence
|Qk,j | < m (Ej ) +

E
j,k=1 Qk,j
and therefore
m (E)
.
2j
|Qk,j | =
j,k=1

X
X
X
j=1
m (Ej ) +
j=1
k=1
|Qk,j |
2j
m (Ej ) + .
(4.42)
j=1
Being > 0 arbitrary, we obtain
m (E)
X
j=1
m (Ej ).
107
(3) If
E Rn ,
and
then
= inf{m (A) | A is open and E A},

m (E) = .
Proof. From the monotonicity, we have

Thus
m (E) m (A), A E, A open .
m (E) .
Suppose given > 0. Choose a sequence {Qj } of closed
cubes such that
and
E
j=1
X
j=1
|Qj | m (E) + .
j } be a sequence of open cubes such that Q

j Qj
Let {Q
j | |Qj | + , j N.
|Q
2j
Define
A =
j=1 Qj ,
hence A is open, A E and
X
j |
m (A)
|Q
j=1

X
j=1
X
j=1
therefore

|Qj | + j
2
|Qj | +
m (E) + 2.
m (E) + 2.
Being > 0 arbitrary, we have
m (E).
(4.43)
108
(4) If E = E1 E2 e d(E1 , E2 ) > 0, then
m (E) = m (E1 ) + m (E2 ).
Proof. First observe that being E = E1 E2 we have

m (E) m (E1 ) + m (E2 ).
Let > 0. Choose {Qj } a sequence of closed cubes such

that
E
j=1 Qj ,
e
X
|Qj | m (E) + .
j=1
Let > 0 such that
d(E1 , E2 ) > > 0.

Dividing the cubes Qj if necessary, we may assume that
the diameter of each cube Qj is smaller than . Thus each
Qj intersects just one of the sets E1 and E2 . Denote by J1
and J2 the sets of indices j such that Qj intersects E1 and
E2 respectively. thus,
hence,
E1 jJ1 Qj e E2 jJ2 Qj .
m (E1 ) + m (E2 )
Being > 0 arbitrary,
jJ1
X
j=1
|Qj | +
jJ2
|Qj | m (E) + .
m (E1 ) + m (E2 ) m (E).
|Qj |
(4.44)
(5) If a set E is a countable union of cubes quasi disjoints, that

is
E =
j=1 Qj
then
m (E) =
|Qj |.
j=1
109
Proof. Let > 0.

j } be open cubes such that Q
j Qj (that is,
Let {Q
j is contained in the interior of Qj ) and
the closure of Q
.
2j
1 , ..., Q
N are disjoint and
Thus, for each N N the cubes Q
each pair have a finite distance. Hence,
j| +
|Qj | |Q
(N
j=1 Qj )
N
X
j=1
Being
j |
|Q
N
X

|Qj | j .
2
j=1
N
j=1 Qj E
we obtain
m (E)
Therefore
X
j=1
N
X
j=1
j |
|Q
N
X
j=1
|Qj | .
|Qj | m (E) + .
Being > 0 arbitrary, we may conclude that
X
j=1
|Qj | m (E).
4.6.2. The Lebesgue Measure.

Definition 4.6.5. A set E Rn is said to be Lebesgue measurable if for each > 0 there exists A Rn open such that
EA
and
m (A E) .
If E is measurable, we define its Lebesgue measure, denoted by

m(E), as
m(E) = m (E).
110
4.6.3. Properties of Measurable Sets.

(1) Each open set is measurable.
(2) If m (E) = 0 then E is measurable. In particular if E A
and m (A) = 0 then E is measurable.
Proof. Let E Rn be such that m (E) = 0. Suppose
given > 0, thus there exists A Rn open such that
E A and m (A) < . Therefore
m (A E) < .

(3) A countable union of measurable sets is measurable.
Proof. Suppose
E =
j=1 Ej
where each Ej is measurable. Suppose given > 0. For

each j N, there exists Aj Rn open such that
e
Ej Aj
.
2j
Define A =
j=1 Aj . thus E A and
m (Aj Ej )
(A E)
j=1 (Aj Ej ).
From the monotonicity and countable sub-additivity of the

outer measure we have,
m (A E)
m (Aj Ej ) < .

j=1
(4) Closed sets are measurable.

Proof. Observe that
F =
k=1 F Bk ,
where Bk denotes a closed ball of radius k with center at

origin. Thus F may be expressed as a countable union of
compact sets. Hence, we have only to show that if F is
compact then it is measurable. Let F be a compact set.
Observe that
m (F ) < .
111
Let > 0, thus there exists an open A Rn such that

F Ae
m (A) m (F ) + .
Being F closed, A F is open, and therefore, A F may
be expressed as a countable union of quasi disjoint closed
cubes. Hence
For each N N
A F =
j=1 Qj .
K = N
j=1 Qj
is compact, therefore
d(K, F ) > 0.
Being K F A, we have
m (A) m (F K) = m (F ) + m (K) = m (F ) +
Therefore
N
X
j=1
N
X
j=1
|Qj |.
|Qj | m (A) m (F ) .
Finally,
m (A F )
X
j=1
|Qj | < .
(5) If E Rn is measurable, then E c is measurable.
Proof. A point x Rn is denoted by x = (x1 , x2 , ..., xn )

where xi R for each i {1, ..., n}. Let E be a measurable
set. For each k N there exists an open Ak E such that
1
m (Ak E) < .
k
c
Observe that Ak is closed and therefore measurable, k
N. Thus
c
S =
k=1 Ak
is also measurable. On the other hand
S Ec
112
and if x (E c S) then x E c and x 6 S, so that x 6 E

and x
6 Ack , k N. Hence x 6 E and x Ak , k N
and finally x (Ak E), k N, that is,
Therefore
E c S Ak E, k N.
1
, k N.
k
m (E c S)
Thus
m (E c S) = 0.
This means that E c S is measurable, so that,
E c = S (E c S)
is measurable. The proof is complete.
(6) A countable intersection of measurable sets is measurable.

Proof. This follows from items 3 and 5 just observing
that
c c
j=1 Ej = (j=1 Ej ) .

Theorem 4.6.6. If {Ei } is sequence of measurable pairwise
disjoint sets and E =
j=1 Ei then
m(E) =
m(Ej ).
j=1
Proof. First assume that Ej is bounded. Being Ejc measurable,

given > 0 there exists an open Hj Ejc such that
m (Hj Ejc ) < j , j N.

2
c
Denoting Fj = Hj we have that Fj Ej is closed and
m (Ej Fj ) < j , j N.
2
For each N N the sets F1 , ..., FN are compact and disjoint, so
that
N
X
N
m(j=1 Fj ) =
m(Fj ).
j=1
113
As
N
j=1 Fj E
we have
m(E)
Hence
N
X
j=1
m(Fj )
m(E)
X
j=1
N
X
j=1
m(Ej ) .
m(Ej ) .
being > 0 arbitrary, we obtain

m(E)
m(Ej ).
j=1
As the reverse inequality is always valid, we have
X
m(Ej ).
m(E) =
j=1
For the general case, select a sequence of cubes {Qk } such that
Rn =
k=1 Qk
and Qk Qk+1 k N. Define S1 = Q1 e Sk = Qk Qk1 , k 2.

Also define
Ej,k = Ej Sk , j, k N.
Thus
E =
j=1 (k=1 Ej,k ) = j,k=1 Ej,k ,
where such an union is disjoint and each Ej,k is bounded. Through
the last result, we get
X
m(E) =
m(Ej,k )
=
=
j,k=1
X
X
j=1 k=1
m(Ej,k )
m(Ej ).
(4.45)
j=1
Theorem 4.6.7. Suppose E Rn is a measurable set. Then

for each > 0:
114
(1) There exists an open set A Rn such that E A and

m(A E) < .
(2) There exists a closed set F Rn such that F E and

m(E F ) < .
(3) If m(E) is finite, there exists a compact set K E such
that
m(E K) < .
(4) If m(E) is finite there exist a finite union of closed cubes

F = N
j=1 Qj
such that
where
m(E F ) ,
E F = (E F ) (F E).
Proof.
(1) This item follows from the definition of measurable set.
(2) Being E c measurable, there exist an open B Rn such
that E c B and
m (B E c ) < .
Defining F = B c , we have that F is closed, F E and

E F = B E c . Therefore
m(E F ) < .
(3) Choose a closed set such that F E e
m(E F ) < .
2
Let Bn be a closed ball with center at origin and radius
n. Define Kn = F Bn and observe that Kn is compact,
n N. Thus
E Kn E F.
Being m(E) < we have
m(E Kn ) < ,
for all n sufficiently big.
4.7. LEBESGUE MEASURABLE FUNCTIONS
115
(4) Choose a sequence of closed cubes {Qj } such that

E
j=1 Qj ,
and
|Qj | m(E) + .
2
j=1
Being m(E) < the series converges and there exists

N0 N such that
N0
|Qj | < .
2
+1
0
Defining F = N
j=1 Qj , we have
m(E F ) = m(E F ) + m(F E)

m
j=N0 +1 Qj + m j=1 Qj E
X
X
|Qj | +
|Qj | m(E)
j=N0 +1
j=1
+ = .
2 2
(4.46)
4.7. Lebesgue Measurable Functions

Definition 4.7.1. Let E Rn be a measurable set. A function f : E [, +] is said to be Lebesgue measurable if for
each a R, the set
f 1 ([, a)) = {x E | f (x) < a}
is measurable.
Observe that:
(1) If
f 1 ([, a))
is measurable for each a R then
1
f 1 ([, a]) =
([, a + 1/k))
k=1 f
is measurable for each a R.
116
(2) If
f 1 ([, a])
is measurable for each a R then
1
f 1 ([, a)) =
([, a 1/k])
k=1 f
is also measurable for each a R.

(3) Given a R, observe that
f 1 ([, a)) is measurable E f 1 ([, a)) is measurable

f 1 (R) f 1 ([, a)) f 1 (R [, a)) is measurable
f 1 ([a, +]) is measurable .
(4.47)
(4) From above, we can prove that

f 1 ([, a))
is measurable a R if, and only if
f 1 ((a, b))
is measurable for each a, b R such that a < b. Therefore f is measurable if and only if f 1 (O) is measurable
whenever O R is open.
(5) Thus f is measurable if f 1 (F ) is measurable whenever
F R is closed.
Proposition 4.7.2. If f is continuous in Rn , then f is measurable. If f is measurable and real and is continuous, then f
is measurable.
Proof. The first implication is obvious. For the second, being
continuous
1 ([, a))
is open, and therefore
( f )1 (([, a)) = f 1 (1 ([, a)))
is measurable, a R.
Proposition 4.7.3. Suppose {fk } is a sequence of measurable

functions. Then
sup fk (x), inf fk (x),
kN
kN
117
and
lim sup fk (x), lim inf fk (x)
k
are measurable.
Proof. We will prove only that supnN fn (x) is measurable. The
remaining proofs are analogous. Let
f (x) = sup fn (x).
nN
Thus
1
f 1 ((a, +]) =
n=1 fn ((a, +]).
Being each fn measurable, such a set is measurable, a R. By

analogy
inf fk (x)
kN
is measurable,
lim sup fk (x) = inf sup fj (x),
k1 jk
and
lim inf fk (x) = sup inf fj (x)
k
k1 jk
are measurable.
Proposition 4.7.4.
functions such that
Let {fk } be a sequence of measurable

lim fk (x) = f (x).
Then f is measurable.
Proof. Just observe that
f (x) = lim fk (x) = lim sup fk (x).
k

The next result we do not prove it. For a proof see [38].
Proposition 4.7.5. If f and g are measurable functions, then
(1) f 2 is measurable.
(2) f + g and f g are measurable if both assume finite values.
118
Proposition 4.7.6. Let E Rn a measurable set. Suppose

f : E R is measurable. Thus if g : E R is such that
g(x) = f (x), a.e. in E
then g is measurable.
Proof. Define
A = {x E | f (x) 6= g(x)},
and
B = {x E | f (x) = g(x)}.
A is measurable since m (A) = m(A) = 0 and therefore B = E A

is also measurable. Let a R. Hence

g 1 ((a, +]) = g 1 ((a, +]) A g 1 ((a, +]) B .
Observe that
x g 1 ((a, +]) B x B and g(x) (a, +]

x B and f (x) (a, +]
x B f 1 ((a, +]).
(4.48)
Thus g 1 ((a, +]) B is measurable. As g 1((a, +]) A A we

have m (g 1((a, +]) A) = 0, that is, such a set is measurable.
Hence being g 1 ((a, +]) the union of two measurable sets is also
measurable. Being a R arbitrary, g is measurable.

Theorem 4.7.7. Suppose f is a non-negative measurable
function on Rn . Then there exists a increasing sequence of nonnegative simple functions {k } such that
lim k (x) = f (x), x Rn .
Proof. Let N N. Let QN be the cube with center at origin

and side of measure N. Define
f (x), if x QN and f (x) N,

N,
if x QN and f (x) > N,
FN (x) =
0,
otherwise.
Thus FN (x) f (x) as N , x Rn . Fixing M, N N define

l
l+1
El,M = x QN :
,
FN (x)
M
M
for 0 l N M. Defining
FN,M
119
NM
X
l
E ,
=
M l,M
l=0
we have that FN,M is a simple function and

0 FN (x) FN,M (x)
1
.
M
If K (x) = FK,K (x) we obtain

0 |FK (x) K (x)|
1
.
K
Hence
|f (x) K (x)| |f (x) FK (x)| + |FK (x) K (x)|.
Therefore
lim |f (x) K (x)| = 0, x Rn .
Theorem 4.7.8. Suppose that f is a measurable function

defined on Rn . Then there exists a sequence of simple functions
{k } such that
and
|k (x)| |k+1(x)|, x Rn , k N,
lim k (x) = f (x), x Rn .
Proof. Write
where
f (x) = f + (x) f (x),

f + (x) = max{f (x), 0}
and
f (x) = max{f (x), 0}.
Thus f + and f are non-negative measurable functions so that from
the last theorem there exist increasing sequences of non-negative
simple functions such that
(1)
k (x) f + (x), x Rn ,
(2)
k (x) f (x), x Rn ,
120
as k . Defining
we obtain
(1)
(2)
k (x) = k (x) k (x),

k (x) f (x), x Rn
as k and
(1)
(2)
|k (x)| = k (x) + k (x) |f (x)|, x Rn ,
as k .
Theorem 4.7.9. Suppose f is a measurable function in Rn .

Then there exists a sequence of step functions {k } which converges
to f a.e. in Rn .
Proof. From the last theorem, it suffices to prove that if E is
measurable and m(E) < then E may be approximated almost
everywhere in E by step functions. Suppose given > 0. Observe
that from Proposition 4.6.7, there exist cubes Q1 , ..., QN such that
m(E N
j=1 Qj ) < .
j such that M R
We may obtain almost disjoints rectangles R

j=1 j =
N
j such that
j=1 Qj and disjoints rectangles Rj R
m(E M
j=1 Rj ) < 2.
Thus
f (x) =
M
X
Rj ,
j=1
possibly except in a set of measure < 2. Hence, for each k > 0,

there exists a step function k such that m(Ek ) < 2k where
Defining
Ek = {x Rn | f (x) 6= k (x)}.
and
Fk =
j=k+1 Ej
F =
k=1 Fk
we have m(F ) = 0 considering that
Finally, observe that
m(F ) 2k , k N.
k (x) f (x), x F c .
121
Theorem 4.7.10 (Egorov). Suppose that {fk } is a sequence

of measurable functions defined in a measurable set E such that
m(E) < . Assume that fk f, a.e in E. Thus given > 0 we
may find a closed set A E such that fk f uniformly in A
and m(E A ) < .
Proof. Without loosing generality we may assume that
For each N, k N define
fk f, x E.
EkN = {x E | |fj (x) f (x)| < 1/N, j k}.
Fixing N N, we may observe that
N
EkN Ek+1
,
N
and that
k=1 Ek = E. Thus we may obtain kN such that
1
m(E EkNN ) < N .
2
Observe that
1
|fj (x) f (x)| < , j kN , x EkNN .
N
Choose M N such that
2k .
2
k=M
Define
Thus
N
A =
N M EkN .
m(E A )
m(E EkNN ) < .

2
N =M
Suppose given > 0. Let N N be such that N > M and 1/N < .
Thus if x A then x EkNN so that
|fj (x) f (x)| < , j > kN .
Hence fk f uniformly in A . Observe that A is measurable and
thus there exists a closed set A A such that
m(A A ) < .
2
122
That is:

m(E A ) m(E A ) + m(A A ) < + = ,
2 2
and
fk f
uniformly in A . The proof is complete.
Definition 4.7.11. We say that a set A L1 (Rn ) is dense in

L (Rn ), if for each f L1 (Rn ) and each > 0 there exists g A
such that
Z
kf gkL1(Rn ) =
|f g| dx < .
1
Rn
Theorem 4.7.12. About dense sets in L1 (Rn ) we have:

(1) The set of simple functions is dense in L1 (Rn ).
(2) The set of step functions is dense in L1 (Rn ).
(3) the set of continuous functions with compact support is
dense in L1 (Rn ).
Proof.
(1) From the last theorems given f L1 (Rn ) there
exists a sequence of simple functions such that
k (x) f (x) a.e. in Rn .
Since {k } may be also such that
|k | f, k N
from the Lebesgue dominated converge theorem, we have
kk f kL1 (Rn ) 0,
as k .
(2) From the last item , it suffices to show that simple functions
may approximated by step functions. As a simple function
is a linear combination of characteristic functions of sets
of finite measure, it suffices to prove that given > 0 and
a set of finite measure, there exists a step function such
that
kE kL1 (Rn ) < .
This may be made similar as in Theorem 4.7.9.
123
(3) From the last item, it suffices to establish the result as

f is a characteristic function of a rectangle in Rn . First
consider the case of a interval [a, b]. we may approximate
f = [a,b] by g(x), where g is continuous, and linear on
(a , a) e (b, b + ) and

1, if a x b,
g(x) =
0, if x a or x b + .
Thus
kf gkL1 (Rn ) < 2.

for the general case of a rectangle in Rn , we just recall that
in this case f is the product of the characteristic functions
of n intervals. Therefore we may approximate f by the
product of n functions similar to g defined above.
CHAPTER 5
Distributions
5.1. Basic Definitions and Results
Definition 5.1.1 (Test Functions, the Space D()). Let
Rn be a nonempty open set. For each K compact, consider
the space DK , the set of all C () functions with support in K.
We define the space of test functions, denoted by D() as
D() = K DK , K compact.
(5.1)
Thus D() if and only if C () and the support of is a

compact subset of .
Definition 5.1.2 (Topology for D()). Let Rn be an
open set.
(1) For every K compact, K denotes the topology which
a local base is defined by {VN,k }, where N, k N,
and
VN,k = { DK | kkN < 1/k}
(5.2)
kkN = max{|D (x)| | x , || N}.
(5.3)
(2)
denotes the collection of all convex balanced sets W
D() such that W DK K for every compact K .
(3) We define in D() as the collection of all unions of sets
of the form + W, for D() and W
.
Theorem 5.1.3. Concerning the last definition we have the
following.
(1) is a topology in D().
(2) Through , D() is made into a locally convex topological
vector space.
Proof.
(1) From item 3 of Definition 5.1.2, it is clear that
arbitrary unions of elements of are elements of . Let
125
126
5. DISTRIBUTIONS
us now show that finite intersections of elements of also

belongs to . Suppose V1 and V2 , if V1 V2 =
we are done. Thus, suppose V1 V2 . By the definition
of there exist two sets of indices L1 and L2 , such that
Vi = Li (i + Wi ), for i = 1, 2,
(5.4)
i + Wi , for i = 1, 2.
(5.5)
i (1 i )Wi , for i {1, 2}.
(5.6)
i + i Wi (1 i )Wi + i Wi = Wi
(5.7)
+ i Wi i + Wi Vi , for i {1, 2}.
(5.8)
+ W Vi ,
(5.9)
V1 V2 = V1 V2 ( + W ) .
(5.10)
V = { D() | kk0 < k1 2 k0 }.
(5.11)
and as V1 V2 there exist i D() and Wi

such
that
Thus there exists K such that i DK for i {1, 2}.
Since DK Wi K , DK Wi is open in DK so that from
(5.5) there exists 0 < i < 1 such that
From (5.6) and from the convexity of Wi we have
so that
Define W = (1 W1 ) (2 W2 ) so that
and therefore we may write

(2) It suffices to show that single points are closed sets in D()
and the vector space operations are continuous.
(a) Pick 1 , 2 D() such that 1 6= 2 and define
Thus V
and 1 6 2 + V. As 2 + V is open
is contained D() {1 } and 2 6= 1 is arbitrary,
it follows that D() {1 } is open, so that {1 } is
closed.
(b) The proof that addition is -continuous follows from
the convexity of any element of
. Thus given 1 , 2
D() and V
we have
1
1
1 + V + 2 + V = 1 + 2 + V.
(5.12)
2
2
5.1. BASIC DEFINITIONS AND RESULTS
127
(c) To prove the continuity of scalar multiplication, first

consider 0 D() and 0 R. Then:
0 0 = ( 0 ) + ( 0 )0 .
(5.13)
For V
there exists > 0 such that 0 21 V. Let
us define c = 12 (|0 | + ). Thus if | 0 | < then
( 0 )0 21 V. Let D() such that
0 cV =
so that
1
V,
2(|0 | + )
(5.14)
1
(5.15)
(|0 | + )( 0 ) V.
2
This means
1
1
(5.16)
( 0 ) + ( 0 )0 V + V = V.
2
2
Therefore 0 0 V whenever | 0 | < and
0 cV.

For the next result the proof may be found in Rudin [35].
Proposition 5.1.4. A convex balanced set V D() is open
if and only if V .
Proposition 5.1.5. The topology K of DK D() coincides
with the topology that DK inherits from D().
Proof. From Proposition 5.1.4 we have
V implies DK V K .
(5.17)
{ DK | k kN < } V.
(5.18)
U = { D() | kkN < }.
(5.19)
DK ( + U ) = + (DK U ) V.
(5.20)
Now suppose V K , we must show that there exists A such

that V = ADK . The definition of K implies that for every V,
there exist N N and > 0 such that
Define
Then U
and
128
5. DISTRIBUTIONS
Defining A = V ( + U ), we have completed the proof.
The proof for the next two results may also be found in Rudin
[35].
Proposition 5.1.6. If A is a bounded set of D() then A
DK for some K , and there are MN < such that kkN
MN , A, N N.
Proposition 5.1.7. If {n } is a Cauchy sequence in D(),
then {n } DK for some K compact, and
lim ki j kN = 0, N N.
i,j
(5.21)
Proposition 5.1.8. If n 0 in D(), then there exists a

compact K which contains the support of n , n N and
D n 0 uniformly, for each multi-index .
The proof follows directly from last proposition.
Theorem 5.1.9. Suppose T : D() V is linear, where
V is a locally convex space. Then the following statements are
equivalent.
(1) T is continuous.
(2) T is bounded.
(3) If n in D() then T (n ) as n .
(4) The restrictions of T to each DK are continuous.
Proof.
1 2. This follows from Proposition 1.9.3 .
2 3. Suppose T is bounded and n 0 in D(), by last
proposition n 0 in some DK so that {n } is bounded
and {T (n )} is also bounded. Hence by Proposition 1.9.3,
T (n ) 0 in V .
3 4. Assume 3 holds and consider {n } DK . If
n then, by Proposition 5.1.5, n in D(), so
that, by above T (n ) in V . Since DK is metrizable,
also by proposition 1.9.3 we have that 4 follows.
4 1. Assume 4 holds and let V be a convex balanced
neighborhood of zero in V . Define U = T 1 (V). Thus U
is balanced and convex. By Proposition 5.1.5, U is open
in D() if and only if DK U is open in DK for each
5.2. DIFFERENTIATION OF DISTRIBUTIONS
129
compact K , thus if the restrictions of T to each DK

are continuous at , then T is continuous at , hence 4
implies 1.

Definition 5.1.10 (Distribution). A linear functional in D()
which is continuous with respect to is said to be a Distribution.
Proposition 5.1.11. Every differential operator is a continuous mapping from D() into D().
Proof. Since kD kN kk||+N , N N, D is continuous
on each DK , so that by Theorem 5.1.9, D is continuous on D().

Theorem 5.1.12. Denoting by D () the dual space of D()
we have that T : D() R D () if and only if for each compact
set K there exists an N N and c R+ such that
|T ()| ckkN , DK .
(5.22)
Proof. The proof follows from the equivalence of 1 and 4 in

Theorem 5.1.9.

5.2. Differentiation of Distributions
Definition 5.2.1 (Derivatives for Distributions). Given T
D () and a multi-index , we define the D derivative of T as
D T () = (1)|| T (D ), D().
(5.23)
Remark 5.2.2. Observe that if |T ()| ckkN , D()

for some c R+ , then
|D T ()| ckD kN ckkN +|| , D(),
(5.24)
thus D T D (). Therefore, derivatives of distributions are also

distributions.
Theorem 5.2.3. Suppose {Tn } D (). Let T : D() R
be defined by
T () = lim Tn (), D().
n
(5.25)
130
5. DISTRIBUTIONS
Then T D (), and
D Tn D T in D ().
(5.26)
Proof. Let K be an arbitrary compact subset of . Since (5.25)

holds for every DK , the principle of uniform boundedness implies that the restriction of T to DK is continuous. It follows from
Theorem 5.1.9 that T is continuous in D(), that is, T D ().
On the other hand
(D T )() = (1)|| T (D ) = (1)|| lim Tn (D )
n
= lim (D Tn ()), D(). (5.27)

n
CHAPTER 6
The Lebesgue and Sobolev Spaces

We start with the definition of Lebesgue spaces, denoted by
L (), where 1 p and Rn is an open set.
p
6.1. Definition and Properties of Lp Spaces

Definition 6.1.1 (Lp Spaces). For 1 p < , we say that
u Lp () if u : R is measurable and
Z
|u|pdx < .
(6.1)
R
We also denote kukp = [ |u|p dx]1/p and will show that k kp is a
norm.
Definition 6.1.2 (L Spaces). We say that u L () if
u is measurable and there exists M R+ , such that |u(x)|
M, a.e. in . We define
kuk = inf{M > 0 | |u(x)| M, a.e. in }.
We will show that k k is
q by the relations
+,
p
,
q=
p1
1,
(6.2)
a norm. For 1 p , we define

if p = 1,
if 1 < p < +,
if p = +,
so that symbolically we have
1 1
+ = 1.
p q
The next result is fundamental in the proof of the Sobolev Imbedding Theorem.
131
132
6. THE LEBESGUE AND SOBOLEV SPACES
Theorem 6.1.3 (Holder Inequality). Consider u Lp () and

v Lq (), with 1 p . Then uv L1 () and
Z
|uv|dx kukp kvkq .
(6.3)
Proof. The result is clear if p = 1 or p = . You may assume

kukp , kvkq > 0, otherwise the result is also obvious. Thus suppose
1 < p < . From the concavity of log function on (0, ) we obtain

1 p 1 q
1
1
log
a + b log ap + log bq = log(ab).
(6.4)
p
q
p
q
Thus,
1
1
ab (ap ) + (bq ), a 0, b 0.
p
q
(6.5)
1
1
|u(x)||v(x)| |u(x)|p + |v(x)|q , a.e. in .
p
q
(6.6)
Therefore
Hence |uv| L1 () and

Z
1
1
|uv|dx kukpp + kvkqq .
p
q
Replacing u by u in (6.7) > 0, we obtain

Z
p1
1
|uv|dx
kukpp + kvkqq .
p
q
q/p
For = kuk1
p kvkq
we obtain the Holder inequality.
(6.7)
(6.8)
The next step is to prove that k kp is a norm.

Theorem 6.1.4. Lp () is a vector space and k kp is norm p
such that 1 p .
Proof. The only non trivial property to be proved concerning
the norm definition, is the triangle inequality. If p = 1 or p =
the result is clear. Thus, suppose 1 < p < . For u, v Lp () we
have
|u(x) + v(x)|p (|u(x)| + |v(x)|)p 2p (|u(x)|p + |v(x)|p ), (6.9)
6.1. DEFINITION AND PROPERTIES OF Lp SPACES
133
so that u + v Lp (). On the other hand

Z
p
ku + vkp =
|u + v|p1 |u + v|dx
Z
Z
p1
|u + v| |u|dx + |u + v|p1|v|dx, (6.10)
and hence, from Holders inequality

that is,
p1
ku + vkpp ku + vkp1
p kukp + ku + vkp kvkp ,
(6.11)
ku + vkp kukp + kvkp , u, v Lp ().
(6.12)
Theorem 6.1.5. Lp () is a Banach space for any p such that

1 p .
Proof. Suppose p = . Suppose {un } is Cauchy sequence
in L (). Thus, given k N there exists Nk N such that, if
m, n Nk then
1
(6.13)
kum un k < .
k
Therefore, for each k, there exist a set Ek such that m(Ek ) = 0,
and
1
|um (x) un (x)| < , x Ek , m, n Nk .
(6.14)
k
Observe that E =
k=1 Ek is such that m(E) = 0. Thus {un (x)} is
a real Cauchy sequence at each x E. Define u(x) = lim un (x)
n
on E. Letting m in (6.14) we obtain

1
(6.15)
|u(x) un (x)| < , x E, n Nk .
k
Thus u L () and kun uk 0 as n .
Now suppose 1 p < . Let {un } a Cauchy sequence in Lp ().
We can extract a subsequence {unk } such that
1
kunk+1 unk kp k , k N.
(6.16)
2
To simplify the notation we write uk in place of unk , so that
1
kuk+1 uk kp k , k N.
(6.17)
2
134
Defining
gn (x) =
n
X
k=1
|uk+1(x) uk (x)|,
(6.18)
we obtain
kgn kp 1, n N.
(6.19)
From the monotone convergence theorem and (6.19), gn (x) converges to a limit g(x) with g Lp (). On the the other hand, for
m n 2 we have
|um (x) un (x)| |um (x) um1 (x)| + ... + |un+1(x) un (x)|
g(x) gn1 (x), a.e. in . (6.20)
Hence {un (x)} is Cauchy a.e. in and converges to a limit u(x) so

that
|u(x) un (x)| g(x), a.e. in , for n 2,
(6.21)
which means u Lp (). Finally from |un (x)u(x)| 0, a.e. in ,

|un (x) u(x)|p |g(x)|p and the Lebesgue dominated convergence
theorem we get
kun ukp 0 as n .
(6.22)
Theorem 6.1.6. Let {un } Lp () and u Lp () such that

kun ukp 0. Then there exists a subsequence {unk } such that
(1) unk (x) u(x), a.e. in ,
(2) |unk (x)| h(x), a.e. in , k N, for some h Lp ().
Proof. The result is clear for p = . Suppose 1 p < .

From the last theorem we can easily obtain that |unk (x)u(x)| 0
as k , a.e. in . To complete the proof, just take h = u + g,
where g is defined in the proof of last theorem.

Theorem 6.1.7. Lp () is reflexive for all p such that 1 < p <
.
Proof. We divide the proof into 3 parts.
135
(1) For 2 p < we have that

u v p
u + v p
1
p
p

2 p + 2 p 2 (kukLp () + kvkLp () ),
L ()
L ()
u, v Lp (). (6.23)
Proof. Observe that
p + p (2 + 2 )p/2 , , 0.
(6.24)
ab
a+b
Now taking = 2 and = 2 in (6.24), we obtain
(using the convexity of tp/2 ),
a+b p
ab p
a+b 2
a b 2 p/2
a2 b2
| +|
| (|
| +|
| ) = ( + )p/2
2
2
2
2
2
2
1 p 1 p
|a| + |b| . (6.25)
2
2
The inequality (6.23) follows immediately.
(2) Lp () is uniformly convex, and therefore reflexive for 2
p < .
Proof. Suppose given > 0 and suppose that
kukp 1, kvkp 1 and ku vkp > .
From part 1, we obtain

p
u + v p
<1 ,

2
2
(6.26)
(6.27)
and therefore

u + v

2 < 1 ,
p
(6.28)
for = 1 (1 (/2)p )1/p > 0. Thus Lp () is uniformly

convex and from Theorem 2.7.2 it is reflexive.
(3) Lp () is reflexive for 1 < p 2. Let 1 < p 2, from 2
we can conclude that Lq is reflexive. We will define T :
Lp () (Lq ) by
Z
hT u, f iLq () =
uf dx, u Lp (), f Lq ().
(6.29)
From the Holder inequality, we obtain

|hT u, f iLq () | kukpkf kq ,
(6.30)
136
so that
kT uk(Lq ()) kukp .
(6.31)
Pick u Lp () and define f0 (x) = |u(x)|p2u(x) (f0 (x) =

0 if u(x) = 0). Thus, we have that f0 Lq (), kf0 kq =
kukp1
and hT u, f0iLq () = kukpp . Therefore,
p
kT uk(Lq ())
hT u, f0iLq ()
= kukp
kf0 kq
(6.32)
Hence from (6.31) and (6.32) we have
kT uk(Lq ()) = kukp , u Lp ().
(6.33)
Thus T is an isometry from Lp () to a closed subspace of

(Lq ()) . Since from the first part Lq () is reflexive, we
have that (Lq ()) is reflexive. Hence T (Lp ()) and Lp ()
are reflexive.

Theorem 6.1.8 (Riesz Representation Theorem). Let 1 <
p < and let f be a continuous linear functional on Lp (). Then
there exists a unique u0 Lq such that
Z
f (v) =
vu0 dx, v Lp ().
(6.34)
Furthermore
kf k(Lp ) = ku0 kq .
(6.35)
Proof. First we define the operator T : Lq () (Lp ()) by

Z
hT u, viLp () =
uv dx, v Lp ().
(6.36)
Similarly to last theorem, we obtain
kT uk(Lp ()) = kukq .
(6.37)
q
We have to show that T is onto. Define E = T (L ()). As E is

a closed subspace, it suffices to show that E is dense in (Lp ()) .
Suppose h (Lp ) = Lp is such that
hT u, hiLp () = 0, u Lq ().
(6.38)
Choosing u = |h|p2h we may conclude that h = 0 which, by

Corollary 2.2.13 completes the first part of the proof. The proof of
uniqueness is left to the reader.
137
Definition 6.1.9. Let 1 p . We say that u Lploc () if

uK Lp () for all compact K .
6.1.1. Spaces of Continuous Functions. We introduce some
definitions and properties concerning spaces of continuous functions. First, we recall that by a domain we mean an open set
in Rn . Thus for a domain Rn and for any nonnegative integer
m we define by C m () the set of all functions u which the partial
derivatives D u are continuous on for any such that || m,
where if D = D11 D22 ...Dnn we have || = 1 + ... + n . We
m
0
define C () =
m=0 C () and denote C () = C(). Given a
function : R, its support, denoted by spt() is given by
spt() = {x | (x) 6= 0}.
Cc () denotes the set of functions in C () with compact support
contained in .
The sets C0 () and C0 () consist of the closure of Cc () (which
is the set of functions in C() with compact support in ) and
Cc () respectively, relating the uniform convergence norm. On
the other hand, CBm () denotes the set of functions u C m () for
which D u is bounded on for 0 || m. Observe that CBm ()
is a Banach space with the norm denoted by k kB,m given by
kukB,m = max sup {|D u(x)|} .
0||m x
as the set of functions u C m () for which

Also, we define C m ()
D u is bounded and uniformly continuous on for 0 || m.
is a closed subspace of C m () and is also a
Observe that C m ()
B
Banach space with the norm inherited from CBm (). Finally we
define the spaces of Holder continuous functions.
Definition 6.1.10 (Spaces of Holder Continuous Functions).
If 0 < < 1, for a nonnegative integer m we define the space of
as the subspace
Holder continuous functions denoted by C m, (),
m
of C () consisting of those functions u for which, for 0 || m,
there exists a constant K such that
|D u(x) D u(y)| K|x y|, x, y .
138
is a Banach space with the norm denoted by k km, given

C m, ()
by

|D u(x) D u(y)|
, x 6= y .
kukm, = kukB,m + max sup
0||m x,y
|x y|
Theorem 6.1.11.
1 p < .
The space C0 () is dense in Lp (), for
Proof. For the proof we need the following lemma:

Lemma 6.1.12. Let f L1loc () such that
Z
f u dx = 0, u C0 ().
(6.39)
Then f = 0 a.e. in .
Suppose f L1 () and m() < . Given > 0, since C0 ()

is dense in L1 (), there exists f1 C0 () such that kf f1 k1 <
and thus, from (6.39) we obtain
Z
(6.40)
| f1 u dx| kuk , u C0 ().
Defining
K1 = {x | f1 (x) },
(6.41)
K2 = {x | f1 (x) }.
(6.42)
and
As K1 and K2 are disjoint compact sets, by the Urysohn Theorem

there exists u0 C0 () such that

+1, if x K1 ,
(6.43)
u0 (x) =
1, if x K2
and
|u0(x)| 1, x .
(6.44)
Also defining K = K1 K2 , we may write

Z
Z
Z
f1 u0 dx =
f1 u0 dx +
f1 u0 dx.
(6.45)
6.2. THE SOBOLEV SPACES
Observe that, from (6.40)

Z
Z
|f1 | dx
|f1 u0 | dx
K
so that
Z
|f1 | dx =
|f1 | dx +
139
(6.46)
|f1 | dx + m().
(6.47)
Hence
kf k1 kf f1 k1 + kf1 k1 2 + m().
(6.48)
Since > 0 is arbitrary, we have that f = 0 a.e. in . Finally, if

m() = , define
n = {x | dist(x, c ) > 1/n and |x| < n}.
(6.49)
It is clear that =
n=1 n and from above f = 0 a.e. on n , n
N, so that f = 0 a.e. in .
Finally, to finish the proof of Theorem 6.1.11, suppose h
Lq () is such that
Z
hu dx = 0, u C0 ().
(6.50)
R
Observe that h L1loc () since K |h| dx khkq m(K)1/p < .
From last lemma h = 0 a.e. in , which by Corollary 2.2.13 completes the proof.

Theorem 6.1.13. Lp () is separable for any 1 p < .
Proof. The result follows from last theorem and from the fact
that C0 (K) is separable for each K compact (from the Weierstrass theorem, polynomials with rational coefficients are dense
C0 (K)). Observe that =
n=1 n , n defined as in (6.49), where
n is compact, n N.

6.2. The Sobolev Spaces
Now we define the Sobolev spaces, denoted by W m,p ().
Definition 6.2.1 (Sobolev Spaces). We say that u W m,p ()
if u Lp () and D u Lp (), for all such that 0 || m,
where the derivatives are understood in the distributional sense.
140
Definition 6.2.2. We define the norm k km,p for W m,p (),

where m N and 1 p , as
1/p
X
, if 1 p < ,
(6.51)
kukm,p =
kD ukpp
0||m
and
kukm, = max kD uk .
0||m
(6.52)
Theorem 6.2.3. W m,p () is a Banach space.

Proof. Consider {un } a Cauchy sequence in W m,p (). Then
{D un } is a Cauchy sequence for each 0 || m. Since Lp () is
complete there exist functions u and u , for 0 || m, in Lp ()
such that un u and D un u in Lp () as n . From above
Lp () L1loc () and so un determines a distribution Tun D ().
For any D() we have, by Holders inequality
Z
|un (x) u(x)||(x)|dx kkq kun ukp .
|Tun () Tu ()|
(6.53)
Hence Tun () Tu () for every D() as n . Similarly

TD un () Tu () for every D(). We have that
Tu () = lim TD un () = lim (1)|| Tun (D )
n
= (1)|| Tu (D ) = TD u (), (6.54)

for every D(). Thus u = D u in the sense of distributions,
for 0 || m, and u W m,p (). As lim ku un km,p = 0,
W m,p () is complete.
Remark 6.2.4. Observe that distributional and classical derivatives coincide when the latter exist and are continuous. We define
S W m,p () by
S = { C m () | kkm,p < }
(6.55)
Thus, the completion of S concerning the norm k km,p is denoted

by H m,p ().
Corollary 6.2.5. H m,p () W m,p ()
141
Proof. Since W m,p () is complete we have that H m,p ()

W ().

m,p
Theorem 6.2.6. W m,p () is separable if 1 p < , and is

reflexive and uniformly convex if 1 < p < . Particularly, W m,2 ()
is a separable Hilbert space with the inner product
X
(u, v)m =
hD u, D viL2 () .
(6.56)
0||m
Proof.
We can see W m,p () as a subspace of Lp (, RN ), where
P
N = 0||m 1. From the relevant properties for Lp (), we have
that Lp (; RN ) is a reflexive and uniformly convex for 1 < p <
and separable for 1 p < . Given u W m,p (), we may
associate the vector P u Lp (; RN ) defined by
P u = {D u}0||m .
(6.57)
Since kP ukpN = kukm,p , we have that W m,p is closed subspace of

Lp (; RN ). Thus from Theorem 1.21 in Adams [1], we have that
W m,p () is separable if 1 p < and, reflexive and uniformly
convex, if 1 < p < .

Lemma 6.2.7. Let 1 p < and define U = Lp (; RN ).
For every continuous linear functional f on U, there exists a unique
v Lq (; RN ) = U such that
f (u) =
N
X
i=1
Moreover,
hui , vi i, u U.
kf kU = kvkqN ,
(6.58)
(6.59)
where k kqN = k kLq (,RN ) .

Proof. For u = (u1 , ..., un ) Lp (; RN ) we may write
f (u) = f ((u1 , 0, ..., 0)) + ... + f ((0, ..., 0, uj , 0, ..., 0))
+ ... + f ((0, ..., 0, un)), (6.60)
and since f ((0, ..., 0, uj , 0, ..., 0)) is continuous linear functional on
uj Lp (), there exists a unique vj Lq () such that f (0, ..., 0, uj ,
142
0, ..., 0) = huj , vj iL2 () , uj Lp (), 1 j N, so that

f (u) =
N
X
i=1
hui , vi i, u U.
(6.61)
From Holders inequality we obtain

|f (u)|
N
X
j=1
kuj kp kvj kq kukpN kvkqN ,
(6.62)
and hence kf kU kvkqN . The equality in (6.62) is achieved for

u Lp (, RN ), 1 0, there is a measurable set A such that m(A) > 0 and |vk (x)|
kvk k , x A. Defining u(x) as

vk /vk , if i = k, x A and vk (x) 6= 0
ui (x) =
(6.64)
0,
otherwise,
we have
f (uk ) = hu, vk iL2 () =
|vk |dx (k(vk k )kuk k1

= (kvkN )kuk1N . (6.65)
Since is arbitrary, the proof is complete.
Theorem 6.2.8. Let 1 p < . Given a continuous linear

functional f on W m,p (), there exists v Lq (, RN ) such that
X
f (u) =
hD u, v iL2 () .
(6.66)
0||m
Proof. Consider f a continuous linear operator on U = W m,p ().

By the Hahn Banach Theorem, we can extend f to f, on Lp (; RN ),
qN = kf kU and by the last theorem, there exists {v }
so that kfk
Lq (; RN ) such that
X
f(
u) =
h
u, v iL2 () , v Lp (; RN ).
(6.67)
0||m
143
In particular for u W m,p (), defining u = {D u} Lp (; RN )

we obtain
X
f (u) = f(
u) =
hD u, v iL2 () .
(6.68)
1||m
Finally, observe that, also from the Hahn-Banach theorem kf kU =

kfkqN = kvkqN .

Definition 6.2.9. Let Rn be a domain. For m a positive
integer and 1 p < we define W0m,p () as the closure in kkm,p of
Cc (), where we recall that Cc () denotes the the set of C ()
functions with compact support contained in . Finally, we also
recall that the support of : R, denoted by spt(), is given
by
spt() = {x |(x) 6= 0}.
Now we enunciate two theorems relating the concept of trace.
We do not prove such results. For proofs see Evans, [18].
Theorem 6.2.10 (The trace theorem). Let Rn be an
open bounded set such that is C 1 . Thus there exists a bounded
linear operator
T : W 1,p () Lp (),
such that
(1) T u = u| if u W 1,p () C() and,
(2) kT ukp, Ckuk1,p,, u W 1,p (), where the constant
C depends only on p and .
T u is called the trace of u on .
Theorem 6.2.11. Let Rn be an open bounded set such
that is C 1 . Then u W01,p () if and only if u W 1,p () and
T u = 0 on .
Remark 6.2.12. Similar results are valid for W0m,p , however
in this case the traces relative to derivatives of order up to m 1
are involved.
144
6.3. The Sobolev Imbedding Theorem

6.3.1. The Statement of Sobolev Imbedding Theorem.
Now we present the Sobolev Imbedding Theorem. We recall that
for normed spaces X, Y the notation
X Y
means that X Y and there exists a constant K > 0 such that
kukY KkukX , u X.
If in addition the imbedding is compact then for any bounded sequence {un } X there exists a convergent subsequence {unk },
which converges to some u in the norm k kY .
Theorem 6.3.1 (The Sobolev Imbedding Theorem). Let
be an open bounded set in Rn such that is C 1 . Let j 0 and
m 1 be integers and let 1 p < .
(1) Part I
(a) Case A If either mp > n or m = n and p = 1 then
W j+m,p () CBj ().
(6.69)
Moreover,
W j+m,p () W j,q (), for p q ,
(6.70)
and, in particular
W m,p () Lq (), for p q .
(6.71)
(b) Case B If mp = n, then

W j+m,p () W j,q (), for p q < ,
(6.72)
and, in particular
W m,p () Lq (), for p q < .
(c) Case C If mp < n and p = 1 , then
np
W j+m,p() W j,q (), for p q p =
,
n mp
(6.73)
(6.74)
and, in particular
W m,p () Lq (), for p q p =
np
.
n mp
(6.75)
6.4. THE PROOF OF THE SOBOLEV IMBEDDING THEOREM
145
(2) Part II
If mp > n > (m 1)p, then
W j+m,p C j,(), for 0 < m (n/p),
(6.76)
and if n = (m 1)p, then

W j+m,p C j, (), for 0 < < 1.
(6.77)
Also, if n = m 1 and p = 1, then (6.77) holds for = 1

as well.
(3) Part III All imbeddings in Parts A and B are valid for arbitrary domains if the W space undergoing the imbedding is replaced with the corresponding W0 space.
6.4. The Proof of the Sobolev Imbedding Theorem
Now we present a collection of results which imply the proof of
the Sobolev imbedding theorem. We start with the approximation
by smooth functions.
Definition 6.4.1. Let Rn be an open bounded set. For
each > 0 define
= {x | dist(x, ) > }.
Definition 6.4.2. Define Cc (Rn ) by
(

C exp |x|211 , if |x|2 < 1,
2
(x) =
0,
if |x|2 1,
where | |2 refers to the Euclidean norm in Rn , that is for x =
(x1 , ..., xn ) Rn , we have
q
|x|2 = x21 + ... + x2n .
Moreover, C > 0 is chosen so that

Z
dx = 1.
Rn
For each > 0, set

(x) =
1 x
.
146
The function is said to be the fundamental mollifier. The functions Cc (Rn ) and satisfy
Z
dx = 1,
Rn
and spt( ) B(0, ).
Definition 6.4.3. If f : Rn is locally integrable, we

define its mollification, denoted by f : R as:
f = f,
that is,
f (x) =
=
(x y)f (y) dy
B(0,)
(y)f (x y) dy.
(6.78)
Theorem 6.4.4 (Properties of mollifiers). The mollifiers have

the following properties:
(1) f C ( ),
(2) f f a.e. as 0,
(3) If f C() then f f uniformly on compact subsets of
.
Proof.
(1) fix x , i {1, ..., n} and a h small enough
such that
x + hei .
Thus

Z
x + hei y
xy
f (x + hei ) f (x) 1
1
= n
h
h
f (y) dy

Z
1
x + hei y
xy
1
= n
V h
f (y) dy,
(6.79)
for an appropriate compact V . As

1
x + hei y
xy
1 x y
,
h
xi
147
as h 0, uniformly on V , we obtain
Z
(x y)
f (x)
=
f (y) dy.
xi
xi
By analogy, we may show that

Z
D f (x) =
D (x y)f (y) dy, x .
(2) From the Lebesgue differentiation theorem we have

Z
1
|f (y) f (x)| dy = 0,
(6.80)
lim
r0 |B(x, r)| B(x,r)
for almost all x . Fix x such that (6.80) holds.
Hence,
Z
|f (x) f (x)| =
(x y)[f (x) f (y)] dy
B(x,)

Z
1
xy
n
[f (x) f (y)] dy
B(x,)
Z
C
|f (y) f (x)| dy
(6.81)
|B(x, )| B(x,)
for an appropriate constant C > 0. From (6.80), we obtain
f f as 0.
(3) Assume f C(). Given V choose W such that
V W ,
and note that f is uniformly continuous on W Thus the
limit indicated in (6.80) holds uniformly on V , and therefore f f uniformly on V .

Theorem 6.4.5. Let u Lp (), where 1 p < . Then
u Lp (),
and
k ukp kukp , > 0

lim k u ukp = 0.
0+
148
Proof. Suppose u Lp () and 1 < p < . Defining q =

p/(p 1), from Holders inequality we have

Z

(x y)u(y) dy
| u(x)| =
n
ZR

(11/p)
1/p

=
[ (x y)]
[ (x y)] u(y) dy
Rn
Z
Rn
Z
Rn
(x y) dy
1/q Z
(x y)|u(y)|p dy
(x y)|u(y)| dy
Rn
1/p
1/p
(6.82)
From this and Fubini theorem, we obtain

Z
Z Z
p
| u(x)| dx
(x y)|u(y)|p dy dx
n
n

Z
ZR R
p
(x y) dx dy
=
|u(y)|
Rn
Rn
kukpp .
(6.83)
Suppose given > 0. As C0 () is dense in Lp (), there exists

C0 () such that
ku kp < /3.
From the fact that
as 0, uniformly in we have that there exists > 0 such that

k kp < /3
if < . Thus for any < (), we get
k u ukp = k u + + ukp
k u kp + k kp + k ukp
/3 + /3 + /3 = .
(6.84)
Since > 0 is arbitrary, the proof is complete.
For the next theorem we denote

u(x), if x ,
u(x) =
0,
if x Rn .
149
6.4.1. Relatively Compact Sets in Lp ().

Theorem 6.4.6. Consider 1 p < . A bounded set K
L () is relatively compact if and only if for each > 0, there exists
> 0 and G (here G denotes that G ) such that
for each u K and h Rn such that |h| < we have
(1)
Z
|
u(x + h) u(x)|p dx 0. As K is compact we may find a /6-net for K. Denote such
a /6-net by N where
N = {v1 , ..., vm } Lp ().
Since Cc () is dense in Lp (), for each k {1, ..., m} there exists

k Cc () such that
kk vk kp < .
6
Thus defining
S = {1 , ..., m },
given u K, we may select vk N such that
ku vk kp < ,
6
so that
kk ukp kk vk kp + kvk ukp

+ = .
6 6
3
Define
where
We have that
G = m
k=1 spt(k ),
spt(k ) = {x Rn | k (x) 6= 0}.
G ,
(6.87)
150
where as above mentioned this means G . Observe that

Z
p
p
> ku k kp
|u(x)|p dx.
G
Since u K is arbitrary, (6.86) is proven. Since k is continuous

and spt(k ) is compact we have that k is uniformly continuous,
that is, for the given above, there exists > 0 such that if |h| <
1} then
min{,
, x G,
|k (x + h) k (x)| <
3(|G| + 1)
Thus,
Z
p
.
|k (x + h) k (x)|p dx <
3
Also observe that since
ku k kp < ,
3
we have that
kTh u Th k kp < ,
3
1}, we obtain
where Th u = u(x + h). Thus if |h| < = min{,
kTh u ukp kTh u Th k kp + kTh k k kp

+kk ukp

+ + = .
(6.88)
<
3 3 3
For the converse, it suffices to consider the special case = Rn ,
= {
because for the general we can define K
u | u K}. Suppose
n
given > 0 and choose G R such that for all u K we have
Z
|u(x)|p dx < .
3
Rn G
For each > 0 the function u C (Rn ), and in particular

u C(G). Suppose C0 (Rn ). Fix > 0. By Holders
inequality we have
p
Z

p
(y)((x y) (x)) dy
| (x) (x)| =
n
p
ZR

11/p
1/p

( (y))
( (y)) (Ty (x) (x)) dy
=
n
ZR
( (y))|Ty (x) (x)|p dy.

(6.89)
B ()
151
Hence, from the Fubini theorem we may write

Z
| (x) (x)|p dx
n
R
Z
Z
( (y))
|Ty (x) (x)|p dx dy, (6.90)
Rn
B ()
so that we may write

k kp sup {kTh kp }.
(6.91)
hB ()
Fix u Lp (Rn ). We may obtain a sequence {k } Cc (Rn ) such

that
k u, in Lp (Rn ).
Observe that
as k . Also
k u, in Lp (Rn ),
Th k Th u, in Lp (Rn ),
as k . Thus
kTh k k kp kTh u ukp ,

in particular
lim { sup {kTh k k k sup {kTh u ukp }.
k hB ()
hB ()
Therefore as
k k k kp k u ukp ,
as k , from (6.91) we get
k u ukp sup {kTh u ukp }.

hB()

k u ukp 0, uniformly in K as 0.
Fix 0 > 0 such that
Z
|0 u u|p dx <
G
, u K.
3 2p1
152
Observe that
Z

|0 u(x)| =
0 (x y)u(y) dy
n

ZR

(11/p)
1/p
[0 (x y)]
[0 (x y)] u(y) dy
=
Rn
Z
1/q Z
1/p
p
0 (x y) dy
0 (x y)|u(y)| dy
Rn
Z
Rn
Rn
0 (x y)|u(y)|p dy
1/p
(6.92)
From this, we may write,

1/p

|0 u(x)| sup 0 (y)
kukp K1 , x Rn , u K
yRn
where K1 = K2 K3 ,
K2 =
sup 0 (y)
yRn
1/p
and K3 is any constant such that

kukp < K3 , u K.
Similarly
|0 u(x + h) 0 u(x)|
sup 0 (y)
yRn
1/p
kTh u ukp ,
and thus from (6.85) we obtain

0 u(x + h) 0 u(x), as h 0
uniformly in Rn and for u K.

By the Arzela-Ascoli Theorem
{0 u | u K}
is relatively compact in C(G), and it is totally bounded so that
there exists a 0 -net N = {v1 , ..., vm } where
.
0 =
p1
3 2 |G|
Thus for some k {1, ..., m} we have
kvk 0 uk < 0 .
Hence,
Z
Z
p
|u(x) vk (x)| dx =
|u(x) | dx +
|u(x) vk (x)|p dx
G
Z
+ 2p1 (|u(x) (0 u)(x)|p
3
G
+|0 u(x) vk (x)|p ) dx

|G|
p1
+2
+
3
3 2p1 3 2p1 |G|
= .
(6.93)
Rn
153
Rn G
Thus K is totally bounded and therefore it is relatively compact.


6.4.2. Some Approximation Results.
Theorem 6.4.7. Let Rn be an open set. Assume u
W () for some 1 p < , and set
m,p
u = u in .
Then,
(1) u C ( ), > 0,
m,p
(2) u u in Wloc
(), as 0,
Proof. Assertion 1 has been already proved. Let us prove 2.
We will show that if || m, then
D u = D u, in .
For, let x . Thus,
Z
(x y)u(y) dy
D u (x) = D
Z
=
Dx (x y)u(y) dy
Z
||
Dy ( (x y))u(y) dy.
= (1)
Observe that for fixed x the function

(y) = (x y) Cc ().
(6.94)
154
Therefore,
Z
Z
||
(x y)Dyu(y) dy,
Dy ( (x y)) u(y) dy = (1)
and hence,
||+||
D u (x) = (1)
(x y)D u(y) dy
= ( D u)(x).
(6.95)
Now choose any open bounded set such that V . We have

that
D u D u, in Lp () as 0,
for each || m.
Thus,
ku ukpm,p,V =
||m
kD u D ukp,V 0,
as 0.
Theorem 6.4.8. Let Rn be a bounded open set and

suppose u W m,p () for some 1 p < . Then there exists a
sequence {uk } C () such that
uk u in W m,p ().
Proof. Observe that

=
i=1 i ,
where
Define
i = {x | dist(x, ) > 1/i}.

i+1 ,
Vi = i+3
and choose any open set V0 such that V0 , so that

=
i=0 Vi .
Let {i}
i=0 be a smooth partition of unit subordinate to the open
sets {Vi }
i=0 . That is,

0P i 1, i Cc (Vi )
on ,
i=0 i = 1,
155
Now suppose u W m,p (). Thus i u W m,p () and spt(i u)

Vi . Choose > 0. For each i N choose i > 0 small enough
so that
ui = i (i u)
satisfies
kui i ukm,p,
2i+1
i Vi . Define
and spt(ui ) Wi where Wi = i+4
v=
ui .
i=0
Thus such a function belongs to C (), since for each open V

there are at most finitely many non-zero terms in the sum. Since
u=
i u,
i=0
we have that for a fixed V ,

kv ukm,p,V
kui i ukm,p,V
i=0
i=0
1
2i+1
= .
(6.96)
Taking the supremum over sets V we obtain

kv ukm,p, < .
Since > 0 is arbitrary, the proof is complete.
The next result is also relevant. For a proof see Evans, [18] page
232.
Theorem 6.4.9. Let Rn be a bounded set such that
is C 1 . Suppose u W m,p () where 1 p < . Thus there exists
a sequence {un } C () such that
un u in W m,p (), as n .
156
6.4.3. Extensions. In this section we study extensions of Sobolev

spaces from a domain Rn to Rn .
Theorem 6.4.10. Assume Rn is an open bounded set,
and that is C 1 . Let V be a bounded open set such that V .
Then there exists a bounded linear operator
E : W 1,p () W 1,p (Rn ),
such that for each u W 1,p () we have:

(1) Eu = u, a.e. in ,
(2) Eu has support in V , and
(3) kEuk1,p,Rn Ckuk1,p,, where the constant depend only
on p, , and V.
Proof. Fix x0 and suppose first that is flat near x0 ,
lying in the plane {xn } = 0. Thus there exists an open ball B with
center in x0 and radius r > 0 such that
+
B = B {x : xn 0}
B = B {x : xn 0} Rn ,
and define u by
First assume u C ()

u(x),
u =
3u(x1 , ..., xn1 , xn ) + 4u(x1 , ..., xn1 , xn /2),
if x B +
if x B .
We will show that u C 1 (B). Indeed,
u (x)
u(x1 , ..., xn1 , xn )
u(x1 , ..., xn1 , xn /2)
=3
2
,
xn
xn
xn
so that
We can also verify that
u |xn =0 = u+ |xn =0 .
u+
xi = uxi , i {1, ..., n 1}.
Observing the expression of the partial derivatives of u in B + and

B we have
k
uk1,p,B Ckuk1,p,B+ ,
where C does not depend on u.

Let us now consider the situation such that is not necessarily
flat near x0 . Let be a C 1 mapping which straightens out near
157
x0 . The inverse of is denoted by . Write y = (x) so that

x = (y), and define u1 by
u1 = u((y)).
Therefore, there exists r > 0 such that for B(x0 , r) = B1 we have
similar result as above, that is u1 is C 1 and
k
u1 k1,p,B1 C1 k
u1 k1,p,B1+ .
Define W = (B1 ), then we may write
k
uk1,p,W C2 kuk1,p,W + .
Hence u is an extension of u from W + = (B1+ ) to W = (B1 ).
Since is compact, there exists finitely many points x0i
such that the union of corresponding Wi covers . Thus
N
i=1 Wi = W for some N N. Take W0 such that
N
i=0 Wi , and let {i}N
i=0 be an associated partition of unity. Denoting by ui the extension of u from Wi+ to Wi and u0 = u defining
u =
N
X
i u i ,
i=0
we obtain u = u, in and
k
uk1,p,Rn C3 kuk1,p,.
Observe that the partition of unity may be chosen such that the
support of u be in V . Henceforth we denote
Eu = u.
Recall that we have considered u C (). Suppose now u
W 1,p () and choose {uk } C () converging to u in W 1,p ().
Thus, from above
kEuk Eul k1,p,Rn C3 kuk ul k1,p, ,
and thus {Euk } is a Cauchy sequence converging to some u. Defining u = Eu, we obtain Eu = u, a.e. in and
k
uk1,p,Rn C3 kuk1,p,,
where the constant does not depend on u. The proof is complete.
158
6.4.4. The Main Results.

Definition 6.4.11. For 1 p < n we define r =
np
.
np
The next theorem we do not prove it. For a proof see Evans,
[18] page 263.
Theorem 6.4.12 (Gagliardo-Nirenberg-Sobolev inequality). Let
1 p < n. Thus there exists a constant K > 0 depending only p
and n such that
kukr,Rn KkDukp,Rn , u Cc1 (Rn ).
Theorem 6.4.13. Let Rn be a bounded open set. Suppose is C 1 , 1 p < n and u W 1,p ().
Then u Lr () and
kukr, Kkuk1,p,,
where the constant depends only on p, n and .
Proof. Since is C 1 , from Theorem 6.4.10, there exists an
extension Eu = u W 1,p (Rn ) such that u = u in the support of
u is compact and
k
uk1,p,Rn Ckuk1,p, ,
where C does not depend on u. As u has compact support, from

Theorem 6.4.8, there exists a sequence {uk } Cc (Rn ) such that
uk u in W 1,p (Rn ).
from the last theorem
kuk ul kr,Rn KkDuk Dul kp,Rn .
Hence,
uk u in Lr (Rn ).
also from the last theorem
kuk kr,Rn KkDuk kp,Rn , k N,

so that
ukp,Rn .
k
ukr,Rn KkD
159
Therefore, we may get

kukr,
k
ukr,Rn
KkD
ukp,Rn
K1 k
uk1,p,Rn
K2 kuk1,p, .
(6.97)
Theorem 6.4.14. Let Rn be a bounded open set such

that C 1 . If mp < n, then W m,p () Lq () for p q
(np)/(n mp).
Proof. Define q0 = np/(n mp). We first prove by induction
on m that
W m,p Lq0 ().
The last result is exactly the case for m = 1. Assume
W m1,p Lr1 (),
(6.98)
where
r1 = np/(n (m 1)p) = np/(n np + p),
whenever n > (m 1)p. If u W m,p () where n > mp, then u and

Dj u are in W m1,p (), so that from (6.98) we have u W 1,r1 ()
and
kuk1,r1, Kkukm,p, .
(6.99)
Since n > mp we have that r1 = np/((n mp) + p) < n, from

q0 = nr1 /(n r1 ) = np/(n mp) by the last theorem we have
kukq0, K2 kuk1,r1, ,
where the constant K2 does not depend on u, and therefore from
this and (6.99) we obtain
kukq0, K2 kuk1,r1 , K3 kukm,p,.
(6.100)
The induction is complete. Now suppose p q q0 . Define

s = (q0 q)p/(q0 p) and t = p/s = (q0 p)/(q0 q).
160
Through Holders inequality, we get

Z
q
kukq, =
|u(x)|s |u(x)|qs dx
Z
1/t Z
1/t
(qs)t
|u(x)| dx
|u(x)|
dx
st
p/t
q /t
kukp,kukq00 ,
p/t
q /t
0
kukp,(K3 )q0 /t kukm,p,
p/t
q /t
0
(K3 )q0 /t kukm,p,kukm,p,
= (K3 )q0 /t kukqm,p,,
since
(6.101)
p/t + q0 /t = q.
Corollary 6.4.15.
q < .
If mp = n, then W m,p () Lq for p
Proof. If q p = p/(p 1) then q = ns/(n ms) where

s = pq/(p + q) is such that 1 s p. observe that
W m,p () W m,s ()
with the imbedding constant depending only on ||. Since ms < n,

by the last theorem we obtain
W m,p () W m,s () Lq ().
Now if p q p , from above we have W m,p () Lp () and the

obvious imbedding W m,p () Lp (). Define s = (p q)p/(p p)
and the result follows from a reasoning analogous to the final chain
of inequalities of last theorem, indicated in (??).

About the next theorem, note that its hypotheses are satisfied
if is C 1 (here we do not give the details).
Theorem 6.4.16. Let Rn be an open bounded set, such
that for each x there exists a convex set Cx whose shape
depends on x, but such that |Cx | > , for some > 0 that does
not depend on x. Thus if mp > n, then
W m,p () CB0 ().
161
Proof. Suppose first m = 1 so that p > n. Fix x and pick

y Cx . For C (), from the fundamental theorem of calculus,
we have
Z 1
d((x + t(y x))
dt.
(y) (x) =
dt
0
Thus,

Z 1
d((x + t(y x))
dt,

|(x)| |(y)| +

dt
0
and hence
Z
Z
|(x)| dy

Z 1
d((x + t(y x))

dt dy,
|(y)| dy +

dt
Cx 0
Cx
Cx
so that, from Holders inequality and Fubini theorem we get,

|(x)| |(x)| |Cx |
kkp,|Cx |
1/p
Therefore
|(x)| kkp,||
1/p

d((x + t(y x))
dy dt.

dt
Cx
Z 1Z
0
|(z)|tn dz dt,
where V t, |V | = t |Cx | and denotes the diameter of . From

Holders inequality again, we obtain
1/p
Z 1 Z
1/p
p
|(x)| kkp, || +
|(z)| dy
tn (tn |Cx |)1/p dt,
0
and thus
|(x)| kkp,||
1/p
+ |Cx |
Since p > n we obtain

Z 1
Z
n(11/p )
t
dt =
0
1/p
kkp,
tn/p dt =
tn(11/p ) dt.
1
.
1 n/p
From this, the last inequality and from the fact that |Cx | ||, we
have that there exists K > 0 such that
|(x)| Kkk1,p, , x , C ().
(6.102)
Here the constant K depends only on p, n and . Consider now

u W 1,p ().
162
Thus there exists a sequence {k } C () such that

k u, in W 1,p ().
Up to a not relabeled subsequence, we have

k u, a.e. in .
(6.103)
Fix x such that the limit indicate in (6.103) holds. Suppose

given > 0. Therefore, there exists k0 N such that
|k0 (x) u(x)| /2
and
Thus,
kk0 uk1,p, < /(2K).

|u(x)| |k0 (x)| + /2
Kkk0 k1,p, + /2
Kkuk1,p, + .
(6.104)
Since > 0 is arbitrary, the proof for m = 1 is complete, because

for {k } C () such that k u in W 1,p (), from (6.102) we
have that {k } is a uniformly Cauchy sequence, so that it converges
to a continuous u , where u = u, a.e. in .
For m > 1 but p > n we still have
|u(x)| Kkuk1,p, K1 kukm,p, , a.e. in , u W m,p ().
If p n mp, there exists j satisfying 1 j m 1 such that
jp n (j + 1)p. If jp < n set
r = np/(n jp),
so that by above and the last theorem:
kuk K1 kuk1,r, K1 kukmj,r, K1 kukm,r, K2 kukm,p, .
If jp = n choosing r = max(n, p) also by the last theorem we obtain
the same last chain of inequalities. This completes the proof.

Theorem 6.4.17.
p q .
If mp > n, then W m,p () Lq () for
Proof. From the proof of the last theorem, we may obtain

kuk, Kkukm,p,, u W m,p ().
163
If p q < , we have
kukqq,
|u(x)|p |u(x)|qp dx
|u(x)|p (Kkukm,p,)qp dx
qp
K qp kukpp,kukm,p,
qp
K qp kukpm,p,kukm,p,
= K qp kukqm,p,.
(6.105)
Theorem 6.4.18. Let S Rn be a n-dimensional sphere of

radius bigger than 3. If n < p, then there exists a constant C,
depending only on p and n, such that
kukC 0, (S) kuk1,p,S , u C 1 (S),
where 0 < 1 n/p.
Proof. First consider = 1 n/p and u C 1 (S). Let x, y S
such that |x y| < 1 and define = |x y|. Consider a fixed cube
. For z R ,
denoted by R S such that |R | = n and x, y R
we may write:
u(x) u(z) =
du(x + t(z x))

dt,
dt
that is,
u(x) =
u(z) dz
u(x + t(z x)) (z x) dt dz.
164
Thus, denoting in the next lines V by an appropriate set such that

|V | = tn |R |, we obtain
Z Z 1
Z
1n
n
|u(x + t(z x))| dt dz
u(z) dz/ | n
|u(x)
n 1n
n 1n
R
1
|u(z)| dz dt
tn kukp,S |V |1/p dt
0
Z 1
1n n/p
tn tn/p dt
n kukp,S
0
Z 1
tn/p dt
n 1n/p kukp,S
0
1n/p
kuk1,p,S K,
(6.106)
where
K=
tn/p dt =
n/(1 n/p).
A similar inequality holds with y in place of x, so that

|u(x) u(y)| 2K|x y|1n/p kuk1,p,S , x, y R .
Now consider 0 < < 1 n/p. Observe that, as |x y|
|x y|1n/p if |x y| < 1, we have,

|u(x) u(y)|
| x 6= y, |x y| < 1
sup
|x y|
x,yS

|u(x) u(y)|
sup
| x 6= y, |x y| < 1 Kkuk1,p,S . (6.107)
|x y|1n/p
x,yS
Also,
sup
x,yS
|u(x) u(y)|
| |x y| 1
|x y|
2kuk,S 2K1 kuk1,p,S
so that

|u(x) u(y)|
sup
| x 6= y (K + 2K1 )kuk1,p,S , u W 1,p (S).
|x y|
x,yS
165
Theorem 6.4.19. Let Rn be an open bounded set such

that is C 1 Assume n < p and u W 1,p (). Then
W 1,p () C 0, (),
for all 0 < 1 n/p.
Proof. Fix 0 < 1 n/p and let u W 1,p (). Since is
C , from Theorem 6.4.10, there exists an extension Eu = u such
that u = u, a.e. in , and
1
k
uk1,p,Rn Kkuk1,p,,
where the constant K does not depend on u. From the proof of this
same theorem, we may assume that spt(
u) is on a n-dimensional
sphere S with sufficiently big radius and such sphere does not
depend on u. Thus, in fact, we have
k
uk1,p,S Kkuk1,p,.
Since C (S) is dense in W 1,p (S), there exists a sequence {k }
C (S) such that
uk u, in W 1,p (S).
(6.108)
Up to a not relabeled subsequence, we have

uk u, a.e. in .
From last theorem we have
kuk ul kC 0, (S) Ckuk ul k1,p,S ,
so that {uk } is a cauchy sequence in C 0, (S), and thus uk u for
some u C 0, (S). Hence, from this and (6.108), we have
u = u, a.e. in S.
Finally, from above and last theorem we may write:
uk1,p,S K2 kuk1,p,.
ku kC 0, () ku kC 0, (S) K1 k
166
6.5. Compact Imbeddings

Theorem 6.5.1. Let m be a non-negative integer and let
0 < < 1. Then the following imbeddings exist:
C m+1 () C m (),
(6.109)
C m, () C m (),
(6.110)
C m, () C m, ().
(6.111)
If is bounded then imbeddings (6.110) and (6.111) are compact.

Proof. The Imbeddings (6.109) and (6.110) follows from the
inequalites
kkC m () kkC m+1 () ,
kkC m () kkC m, () .
To establish (6.111) note that for || m

|D (x) D (y)|
sup
| x 6= y, |x y| < 1
|x y|
x,y

|D (x) D (y)|
sup
| x 6= y, |x y| < 1 (6.112)
,
|x y|
x,y
and also,

D (x) D (y)|
| |x y| 1 2 sup{|D |}.(6.113)
sup |
|x y|
x
x,y
Therefore, we may conclude that
m,
kkC m, ()
().
3kkC m, () , C
Now suppose is bounded. If A is a bounded set in C 0, () then

there exists M > 0 such that
kkC 0, () M, A.
But then
|(x) (y)| M|x y| , x, y , A,
so that by the Ascoli-Arzela theorem, A is pre-compact in C().
This proves the compactness of (6.110) for m = 0.
6.5. COMPACT IMBEDDINGS
167
If m 1 and A is bounded in C m, (), then A is bounded

in C 0, (). Thus, by above there is a sequence {k } A and
C 0, () such that
k in C().
However, {Di k } is also bounded in C 0, (), so that there exist a

not relabeled subsequence, also denoted by {k } and i such that
Di k i , in C().
being the uniform one, we have i =
The convergence in C()
Di . We can proceed extracting (not relabeled) subsequences until
obtaining
D k D , in C(), 0 || m.
This completes the proof of compactness of (6.110). For (6.111),
let S be a bounded set in C m, (). Observe that

/
|D (x) D (y)|
|D (x) D (y)|
=
|x y|
|x y|
|D (x) D (y)|1/
K|D (x) D (y)|1/,(6.114)
for all S. From (6.110), S has a converging subsequence in

C m (). From (6.114) such a subsequence is also converging in
C m, (). The proof is complete.
Theorem 6.5.2 (Rellich-Kondrachov). Let Rn be an

open bounded set such that is C 1 . Let j, m be integers, j
0, m 1, and let 1 p < .
(1) Part I- If mp n, then the following imbeddings are
compact:
W j+m,p () W j,q (),
if 0 < n mp < n and 1 np/(n mp), (6.115)
W j+m,p() W j,q (), if n = mp, 1 q < .
(6.116)
(2) Part II- If mp > n, then the following imbeddings are

compact:
W j+m,p CBj (),
(6.117)
168
W j+m,p () W j,q (), if 1 q .
(6.118)
W j+m,p () C j (), if mp > n,
(6.119)
(3) Part III-The following imbeddings are compact:
W j+m,p () C j, (),
if mp > n (m 1)p and 0 < < m n/p. (6.120)
(4) Part IV- All the above imbeddings are compact if we

replace W j+m,p () by W0j+m,p().
Remark 6.5.3. Given X, Y, Z spaces, for which we have the

imbeddings X Y and Y Z and if one of these imbeddings
is compact then the composite imbedding X Z is compact.
Since the extension operator u u where u(x) = u(x) if x
and u(x) = 0 if x Rn , defines an imbedding W0j+m,p ()
W j+m,p (Rn ) we have that Part-IV of above theorem follows from
the application of Parts I-III to Rn (despite the fact we are assuming
bounded, the general results may be found in Adams [1]).
Remark 6.5.4. To prove the compactness of any of above
imbeddings it is sufficient to consider the case j = 0. Suppose,
for example, that the first imbedding has been proved for j = 0.
For j 1 and {ui } bounded sequence in W j+m,p() we have that
{D ui } is bounded in W m,p () for each such that || j. From
the case j = 0 it is possible to extract a subsequence (similarly to
a diagonal process) {uik } for which {D uik } converges in Lq () for
each such that || j, so that {uik } converges in W j,q ().
Remark 6.5.5.
1 q . In fact
Since is bounded, CB0 () Lq () for
kuk0,q, kukCB0 [vol()]1/q .
(6.121)
Thus the compactness of (6.118) (for j = 0) follows from that of

(6.117).
Proof of Parts II and III. If mp > n > (m1)p and 0 < <
(m n)/p, then there exists such that < < m (n/p). Since
is bounded, the imbedding C 0, () C 0, () is compact by
Theorem 6.5.1. Since by the Sobolev Imbedding Theorem we have
W m,p () C 0, (), we have that imbedding (6.120) is compact.
6.5. COMPACT IMBEDDINGS
169
If mp > n, let j be the non-negative integer satisfying (m

j )p > n (m j 1)p. Thus we have the chain of imbeddings
W m,p () W mj ,p () C 0, () C(),
(6.122)
where 0 < < m j (n/p). The last imbedding in (6.122) is

compact by Theorem 6.5.1, so that (6.119) is compact for j = 0.
By analogy (6.117) is compact for j = 0. Therefore from the above
remarks, (6.118) is also compact. For the proof of Part I, we need
the following lemma:
Lemma 6.5.6. Let be an bounded domain in Rn . Let
1 q1 q0 and suppose
W m,p () Lq0 (),
(6.123)
W m,p () Lq1 .
(6.124)
Suppose also that (6.124) is compact. If q1 q < q0 , then the

imbedding
W m,p Lq ()
(6.125)
is compact.
Proof. Define = q1 (q0 q)/(q(q0 q1 )) and = q0 (q
q1 )/(q(q0 q1 )). We have that > 0 and 0. From Holders
inequality and (6.123) there exists K R+ such that,
kuk0,q, kuk0,q1, kuk0,q0, Kkuk0,q1, kukm,p,,
u W m,p (). (6.126)
Thus considering a sequence {ui } bounded in W m,p (), since (6.124)

is compact there exists a subsequence {unk } that converges, and is
therefore a Cauchy sequence in Lq1 (). From (6.126), {unk } is also
a Cauchy sequence in Lq (), so that (6.125) is compact.

Proof of Part I. Consider j = 0. Define q0 = np/(n mp).
To prove the imbedding
W m,p () Lq (), 1 q < q0 ,
(6.127)
is compact, by last lemma it suffices to do so only for q = 1. For

k N, define
k = {x | dist(x, ) > 2/k}.
(6.128)
170
Suppose A is a bounded set of functions in W m,p (), that is suppose

there exists K1 > 0 such that
kukW m,p () < K1 , u A.
Also, suppose given > 0, and define, for u W m,p (), u(x) =
u(x) if x , u(x) = 0, if x Rn . Fix u A. From Holders
inequality and considering that W m,p () Lq0 (), we have
Z
1/q0 Z
11/q0
Z
q0
|u(x)|dx
|u(x)| dx
1dx
k
k
11/q0
K1 kukm,p, [vol( k )]
(6.129)
Thus, since A is bounded in W m,p , there exists K0 N such that

if k K0 then
Z
|u(x)|dx < , u A
(6.130)
k
and, now fixing a not relabeled k > K0 , we get,

Z
|
u(x + h) u(x)|dx < 2, u A, h Rn .
(6.131)
Observe that if |h| < 1/k, then x + th 2k provided x k and

0 t 1. If u C () we have that
Z
Z
Z 1
du(x + th)
|u(x + h) u(x)|
dx
|
|dt
dt
k
k
0
Z 1 Z
|h|
dt
|u(y)|dy |h|kuk1,1,
0
2k
K2 |h|kukm,p,.
(6.132)
Since C () is dense in W m,p (), from above for |h| sufficiently

small
Z
|
u(x + h) u(x)|dx < 3, u A.
(6.133)
From Theorem 6.4.6, A is relatively compact in L1 () and therefore the imbedding indicated (6.127) is compact for q = 1. This
completes the proof.
Part 2
Variational Convex Analysis
CHAPTER 7
Basic Concepts on the Calculus of Variations

7.1. Introduction to the Calculus of Variations
We recall that a functional is a function whose the co-domain is
the real set. We denote such functionals by F : U R, where U is
a Banach space. In our work format, we consider the special cases
R
(1) F (u) = f (x, u, u) dx, where Rn is an open,
bounded,
R connected set.
(2) F (u) = f (x, u, u, D 2u) dx, here

ui
Du = u =
xj
and
2
D u = {D ui } =
2 ui
xk xl
for i {1, ..., N} and k, l {1, ..., N}.

Also, f : RN RN n is denoted by f (x, s, ) and we assume
(1)
f (x, s, )
s
and
(2)
f (x, s, )
are continuous (x, s, ) RN RN n .

Remark 7.1.1.
may be used
We also recall that the notation u = Du
Now we define our general problem, namely problem P where

Problem P : minimize F (u) on U,
173
174
7. BASIC CONCEPTS ON THE CALCULUS OF VARIATIONS
that is, to find u0 U such that

F (u0 ) = min{F (u)}.
uU
At this point, we introduce some essential definitions.

Definition 7.1.2 (Space of admissible variations). Given F :
U R we define the space of admissible variations for F , denoted
by V as
V = { | u + U, u U}.
For example, for F : U R given by
Z
1
F (u) =
u u dx hu, f iU ,
2
where R3 and
U = {u W 1,2 () | u = u on }
we have
V = W01,2 ().
Observe that in this example U is a subset of a Banach space.

Definition 7.1.3 (Local minimum). Given F : U R, we
say that u0 U is a local minimum for F , if there exists > 0 such
that
F (u) F (u0), u U, such that ku u0 kU < ,
or equivalently
F (u0 + ) F (u0), V, such that kkU < .

Definition 7.1.4 (Gateaux variation). Given F : U R we
define the Gateaux variation of F on the direction V, denoted
by F (u, ) as
F (u + ) F (u)
,
0
F (u, ) = lim
if such a limit is well defined. Furthermore, if there exists u U

such that
F (u, ) = h, uiU , U,

7.2. EVALUATING THE GATEAUX
VARIATIONS
175
we say that F is Gateaux differentiable at u U, and u U is

said to be the Gateaux derivative of F at u. Finally we denote
F (u)
u = F (u) or u =
.
u
7.2. Evaluating the G
ateaux variations
Consider F : U R such that
Z
F (u) =
f (x, u, u) dx
where the hypothesis indicated in the last section are assumed.

RN ) and C 1 (;
RN ) and let us evaluate
Consider u C 1 (;
c
F (u, ):
From Definition 7.1.4,
F (u + ) F (u)
.
F (u, ) = lim
0
Observe that
f (x, u + , u + ) f (x, u, u)
0
f (x, u, u)
f (x, u, u)
=
+
.
s
Define
f (x, u + , u + ) f (x, u, u)
,
G(x, u, , ) =
and
u, ) = f (x, u, u) + f (x, u, u) .
G(x,
s
Thus we have
u, ).
lim G(x, u, , ) = G(x,
lim
Now will show that

Z
Z
u, ) dx.
lim G(x, u, , ) dx =
G(x,
0
It suffices to show that

Z
Z
u, ) dx.
lim
G(x, u, , 1/n) dx =
G(x,
n
such that
For, for each n N, define xn
n , u(xn ), (xn ))| = cn ,
|G(xn , u(xn ), (xn ), 1/n) G(x
176
where
u(x), (x))|}.
cn = max{|G(x, u(x), (x), 1/n) G(x,
Since the function in question is continuous on the compact set ,

is compact, there
{xn } is well defined. Also from the fact that
exists a subsequence xnj and x0 such that

lim xnj = x0 .
j+
Thus
lim cnj = c0
j+
0 , u(x0 ), (x0 ))|} = 0.

= lim {|G(x0 , u(x0 ), (x0 ), 1/(nj )) G(x
j+
Therefore, given > 0, there exists j0 N such that if j > j0 then

cnj < /||.
Thus, if j > j0 , we have:
Z

Z

G(x, u, , 1/(nj )) dx G(x,
u, ) dx

Z
u, )| dx cn || < . (7.1)
|G(x, u, , 1/(nj )) G(x,

j
Hence, we may write

Z
Z
u, ) dx,
lim
G(x, u, , 1/(nj )) dx =
G(x,
j+
that is,
F (u, ) =
Z
f (x, u, u)
f (x, u, u)
+

s
dx.
Theorem 7.2.1 (Fundamental lemma of calculus of variations).

Consider an open set Rn and u L1loc () such that
Z
u dx = 0, Cc ().
Then u = 0, a.e. in .
Remark 7.2.2. Of course a similar result is valid for the

vectorial case. A proof of such a result was given in Chapter 6.

7.3. THE GATEAUX
VARIATION IN W 1,2 ()
177
Theorem 7.2.3 (Necessary conditions for a local minimum).

Suppose u U is a local minimum for a Gateaux differentiable
F : U R. Then
F (u, ) = 0, V.
Proof. Fix V. Define () = F (u + ). Since by hypothesis is differentiable and attains a minimum at = 0, from
the standard necessary condition (0) = 0, we obtain (0) =
F (u, ) = 0.

Theorem 7.2.4. Consider the hypotheses stated at section
7.1 on F : U R. Suppose F attains a local minimum at
RN ) and additionally assume that f C 2 (, RN , RN n ).
u C 2 (;
Then the necessary conditions for a local minimum for F are given
by the Euler-Lagrange equations:

f (x, u, u)
f (x, u, u)
= , in .
div
s
Proof. From Theorem 7.2.3, the necessary condition stands for

F (u, ) = 0, V. From above this implies, after integration by
parts

Z
f (x, u, u)
f (x, u, u)
dx = 0,
div
s
Cc (, RN ).
The result then follows from the fundamental lemma of calculus of

variations.

7.3. The G
ateaux Variation in W 1,2 ()
Theorem 7.3.1. Consider the functional F : U R, where
Suppose
U = {u W 1,2 (, RN ) | u = u0 in }.
F (u) =
f (x, u, u) dx,
where f : RN RN n is such that

|f (x, s1 , 1 ) f (x, s2 , 2 )| < K(|s1 s2 | + |1 2 |),
s1 , s2 RN , 1 , 2 RN n .
178
Also assume the hypothesis of section 7.1. Under such assumptions,

for each u U and V = W01,2 (; RN ), we have

Z
f (x, u, u)
f (x, u, u)
+
dx.
F (u, ) =
s
Proof. From Definition 7.1.4,

F (u + ) F (u)
.
0
F (u, ) = lim
Observe that
f (x, u + , u + ) f (x, u, u)
0
f (x, u, u)
f (x, u, u)
+
, a.e in .
=
s
Define
f (x, u + , u + ) f (x, u, u)
G(x, u, , ) =
,
and
u, ) = f (x, u, u) + f (x, u, u) .
G(x,
s
Thus we have
u, ), a.e in .
lim G(x, u, , ) = G(x,
lim
Now will show that

Z
Z
u, ) dx.
lim G(x, u, , ) dx =
G(x,
0
It suffices to show that

Z
Z
u, ) dx.
G(x, u, , 1/n) dx =
G(x,
lim
n
Observe that
Z
Z
|G(x, u, , 1/n)| dx
K(|| + ||) dx
K(kk2N + kk2Nn )||1/2 .
By the Lebesgue dominated convergence theorem, we obtain

Z
Z
u, ) dx,
lim
G(x, u, , 1/(n)) dx =
G(x,
n+
7.4. ELEMENTARY CONVEXITY
179
that is,
F (u, ) =
Z
f (x, u, u)
f (x, u, u)
+

s
dx.
7.4. Elementary Convexity

if
Definition 7.4.1. A function f : Rn R is said to be convex

f (x + (1 )y) f (x) + (1 )f (y), x, y Rn , [0, 1].
Proposition 7.4.2. If f : Rn R is convex and differentiable,

then
f (y) f (x) hf (x), y xiRn , x, y Rn .
Proof. Pick x, y Rn . By hypothesis
Thus
f ((1 )x + y) (1 )f (x) + f (y), [0, 1].
f (x + (y x)) f (x)
f (y) f (x), (0, 1].
Letting 0+ we obtain
f (y) f (x) hf (x), y xiRn .
Since x, y Rn are arbitrary, the proof is complete.
Proposition 7.4.3. Let f : Rn R be a differentiable function. If

f (y) f (x) hf (x), y xiRn , x, y Rn ,
then f is convex.
Proof. Define f (x ) by
f (x ) = sup {hx, x iRn f (x)}.
xRn
Such a function f is called the Fenchel conjugate of f . Observe

that, by hypothesis,
f (f (x)) = sup {hy, f (x)iRn f (y)} = hx, f (x)iRn f (x).
yRn
(7.2)
180
On the other hand

that is
f (x ) hx, x iRn f (x), x, x Rn ,
f (x) hx, x iRn f (x ), x, x Rn .

Observe that, from (7.2)
and thus
f (x) = hx, f (x)iRn f (f (x))

f (x) = sup {hx, x iRn f (x )}, x Rn .
x Rn
Pick x, y Rn and [0, 1]. Thus, we may write
f (x + (1 )y) = sup {hx + (1 )y, x iRn f (x )}

x Rn
= sup {hx, x iRn + (1 )hy, x iRn f (x )

x Rn
(1 )f (x )}
{ sup {hx, x iRn f (x )}}

x Rn
+ (1 ){ sup {hy, xiRn f (x )}}

x Rn
=f (x) + (1 )f (y).
(7.3)
Since x, y Rn and [0, 1] are arbitrary, we have that f is

convex.

Corollary 7.4.4. Let f : Rn R be twice differentiable and

2
f (x)
,
xi xj
positive definite, for all x Rn . Then f is convex.
Proof. Pick x, y Rn . Using Taylors expansion we obtain
n X
n
X
2 f (
x)
(yi xi )(yj xj ),
f (y) = f (x) + hf (x), y xiRn +
x
x
i
j
i=1 j=1
for x = x + (1 )y (for some [0, 1]). From the hypothesis we

obtain
f (y) f (x) hf (x), y xiRn 0.
Since x, y Rn are arbitrary, the proof is complete.

Similarly we may obtain the following result.
7.4. ELEMENTARY CONVEXITY
181
Corollary 7.4.5. Let U be a Banach space. Consider F :

U R Gateaux differentiable. Then F is convex if and only if
F (v) F (u) hF (u), v uiU , u, v U.
Corollary 7.4.6. let U be a Banach space. Suppose F : U
R is twice Gateaux differentiable and that
2 F (u, ) 0, u U, V.
Then, F is convex.
Proof. Pick u, v U. Define () = F (u + (v u)). By
hypothesis, is twice differentiable, so that
(1) = (0) + (0) + (
),
where |
| 1. Thus
F (v) = F (u) + F (u, v u) + 2 F (u + (v u), v u).
Therefore, by hypothesis
F (v) F (u) + F (u, v u).
Since F is Gateaux differentiable, we obtain
F (v) F (u) + hF (u), v uiU .
Being u, v U arbitrary, the proof is complete.
Corollary 7.4.7. Let U be a Banach space. Let F : U R

be a convex Gateaux differentiable functional. If F (u) = , then
F (v) F (u), v U,
that is, u U is a global minimizer for F .
Proof. Just observe that
F (v) F (u) + hF (u), v uiU , u, v U.
Therefore, from F (u) = , we obtain
F (v) F (u), v U.
182
7.5. The Legendre-Hadamard Condition

RN ) is such that
Theorem 7.5.1. If u C 1 (;
then
2 F (u, ) 0, Cc (, RN ),
fi k (x, u(x), u(x))i k 0, x , RN , Rn .

Such a condition is known as the Legendre-Hadamard condition.
Proof. Suppose
2 F (u, ) 0, Cc (; RN ).
We denote 2 F (u, ) by
Z
2
F (u, ) = a(x)D(x) D(x) dx
Z
Z
+
b(x)(x) D(x) dx + c(x)(x) (x) dx,
(7.4)
where
a(x) = f (x, u(x), Du(x)),
b(x) = fs (x, u(x), Du(x)),
and
c(x) = fss (x, u(x), Du(x)).
Now consider v Cc (B1 (0), RN ). Thus given
x0 for sufxx0
is an admissible dificiently small we have that (x) = v
rection. Now we introduce the new coordinates y = (y 1 , ..., y n ) by
setting y = 1 (x x0 ) and multiply (7.4) by n to obtain
Z
{a(x0 + y)Dv(y) Dv(y) + 2b(x0 + y)v(y) Dv(y)
B1 (0)
+ 2 c(x0 + y)v(y) v(y)} dy > 0,
where a = {a
ij }, b = {bjk } and c = {cjk }. Since a, b and c are
continuous, we have
a(x0 + y)Dv(y) Dv(y) a(x0 )Dv(y) Dv(y),

and
b(x0 + y)v(y) Dv(y) 0,

2 c(x0 + y)v(y) v(y) 0,
7.5. THE LEGENDRE-HADAMARD CONDITION
as 0. Thus this limit give us

uniformly on
Z
fjk
D v j D v k dx 0, v Cc (B1 (0); RN ),
183
(7.5)
B1 (0)
where
fjk
= a
i k (x0 , u(x0 ), u(x0 )).
jk (x0 ) = f
Now define v = (v 1 , ..., v N ) where

v j = j cos(( y)t)(y)
= (1 , ..., N ) RN
and
= (1 , ..., n ) Rn
and Cc (B1 (0)). From (7.5) we obtain
Z
j k
( t(sin(( y)t) + cos(( y)t)D )

0 fjk
B1 (0)
( t(sin(( y)t) + cos(( y)t)D ) dy}
(7.6)
By analogy for
v j = j sin(( y)t)(y)
we obtain
j k
0 fjk
Z
B1 (0)
( t(cos(( y)t) + sin(( y)t)D )
( t(cos(( y)t) + sin(( y)t)D ) dy}
(7.7)
Summing up these last two equations, dividing the result by t2 and

letting t + we obtain
Z
j k
0 fjk
2 dy,
B1 (0)
for all Cc (B1 (0)), which implies

j k
0 fjk
.
184
7.6. The Weierstrass Necessary Condition

In this section we state without proving the result concerning
the Weierstrass Condition for the case n, N 1.
RN ) has the strong minTheorem 7.6.1. Suppose u C 1 (;
imum property. Then for each x and for each rank-one matrix
{i = i } the Weierstrass condition is satisfied, that is,
EF (x, u(x), Du(x) + ) 0,
where
EF (x, u(x), Du(x) + ) = f (x, u(x), Du(x) + )
f (x, u(x), Du(x)) i fi (x, u(x), Du(x)).
The function EF is known as the Weierstrass excess function.

7.6.1. The Weierstrass Condition for n = 1. Here we
present the Weierstrass condition for the special case N 1 and
n = 1. We start with a definition.
Definition 7.6.2. We say that u C([a,

b]; RN ) if u : [a, b]
R is continuous in [a, b], and Du is continuous except on a finite
set of points in [a, b].
N
Theorem 7.6.3 (Weierstrass). Let = (a, b) and f :

RN RN R be such that fs (x, s, ) and f (x, s, ) are continuous
RN RN .
on
Define F : U R by
Z b
F (u) =
f (x, u(x), u(x)) dx,
a
where
U = {u C 1 ([a, b]; RN ) | u(a) = , u(b) = }.

Suppose u U minimizes locally F on U, that is, suppose that
there exists 0 > 0 such that
F (u) F (v), v U, such that ku vk < 0 .
Under such hypotheses, we have
E(x, u(x), u (x+), w) 0, x [a, b], w RN ,
7.6. THE WEIERSTRASS NECESSARY CONDITION
185
and
where
E(x, u(x), u(x), w) 0, x [a, b], w RN ,

u (x+) = lim+ u (x + h),
h0
u (x+) = lim u (x + h),
h0
and,
E(x, s, , w) = f (x, s, w) f (x, s, ) f (x, s, )(w ).
Remark 7.6.4. The function E is known as the Weierstrass
Excess Function.
Proof. Fix x0 (a, b) and w RN . Choose 0 < < 1 and
h > 0 such that u + v U and
kvk < 0
where v(x) is given by
(x x0 )w,
(h x + x0 )w,
v(x) =
0,
where
if 0 x x0 h,
if h x x0 h,
otherwise,
.
1
From
F (u + v) F (u) 0
we obtain
Z x0 +h
f (x, u(x) + v(x), u(x) + v (x)) dx
x0
x0 +h
x0
Define
x =
f (x, u(x), u(x)) dx 0. (7.8)
x x0
,
h
so that
d
x=
dx
.
h
186
From (7.8) we obtain

Z 1
h
f (x0 + xh, u(x0 + xh)+v(x0 + xh), u (x0 + xh)+v (x0 + xh) d
x
0
Z 1
h
f (x0 + xh, u(x0 + xh), u(x0 + xh)) d
x 0. (7.9)
0
where the derivatives are related to x.

Therefore
Z
f (x0 + xh, u(x0 + xh) + v(x0 + xh), u (x0 + xh) + w) d
x
0
Z
f (x0 + xh, u(x0 + xh), u (x0 + xh)) d

x
0
Z 1
+
f (x0 + xh, u(x0 + xh) + v(x0 + xh), u (x0 + xh) w) d
x
Z 1
f (x0 + xh, u(x0 + xh), u(x0 + xh)) d

x
0.
(7.10)
Letting h 0 we obtain
(f (x0 , u(x0 ), u (x0 +) + w) f (x0 , u(x0 ), u (x0 +))

+(1 )(f (x0 , u(x0 ), u (x0 +) w) f (x0 , u(x0 ), u(x0 +))) 0.
Hence, by the mean value theorem we get

(f (x0 , u(x0 ), u(x0 +) + w) f (x0 , u(x0 ), u(x0 +))
(1 )
(f (x0 , u(x0 ), u(x0 +) + (
)w)) w 0. (7.11)
Dividing by and letting 0, so that 0 and (
) 0 we
finally obtain
f (x0 , u(x0 ), u(x0 +) + w) f (x0 , u(x0 ), u (x0 +))
f (x0 , u( x0 ), u(x0 +)) w 0.
Similarly we may get

f (x0 , u(x0 ), u(x0 ) + w) f (x0 , u(x0 ), u (x0 ))
f (x0 , u( x0 ), u(x0 )) w 0.
Since x0 [a, b] and w RN are arbitrary, the proof is complete.

7.7. THE DU BOIS-REYMOND LEMMA
187
7.7. The du Bois-Reymond Lemma

We present now a simpler version of the fundamental lemma of
calculus of variations. The result is specific for n = 1 and is known
as du Bois-Reymond Lemma.
Lemma 7.7.1 (du Bois-Reymond). If u C and
Z b
u dx = 0, V,
a
where
V = { C 1 [a, b] | (a) = (b) = 0},
then there exists c R such that
u(x) = c, x [a, b].

Proof. Define
1
c=
ba
and
(x) =
u(t) dt,
a
x
a
(u(t) c) dt.
Thus we have, (a) = 0, and

Z b
(b) =
u(t) dt c(b a) = 0.
a
Moreover C ([a, b]) so that
V.
Therefore
0
=
(u(x) c)2 dx
a
b
(u(x) c) (x) dx
u(x) (x) dx c[(x)]ba = 0.
Thus
(u(x) c)2 dx = 0,
(7.12)
188
and being u(x) c continuous, we finally obtain

u(x) c = 0, x [a, b].
Proposition 7.7.2. If u, v C([a, b]) and

Z b
(u(x)(x) + v(x) (x)) dx = 0,
a
V, where
then
V = { C 1 [a, b] | (a) = (b) = 0},

v C 1 ([a, b])
and
v (x) = u(x), x [a, b].

Proof. Define
u1 (x) =
Thus u1 C 1 ([a, b]) and
u(t) dt, x [a, b].
u1 (x) = u(x), x [a, b].
Hence, for V, we have

Z b
0 =
(u(x)(x) + v(x) (x) dx
a
Z b
=
(u1 (x) (x) + v (x)) dx + [u1 (x)(x)]ba
a
Z b
=
(v(x) u1 (x)) (x) dx.
(7.13)
a
That is,
(v(x) u1 (x)) (x) dx, V.
By the du Bois- Reymond lemma, there exists c R such that

Hence
v(x) u1(x) = c, x [a, b].

v = u1 + c C 1 ([a, b]),
7.8.
THE WEIERSTRASS-ERDMANN CONDITIONS
189
so that
v (x) = u1 (x) = u(x), x [a, b].
7.8. The Weierstrass-Erdmann Conditions

We start with a definition.
Definition 7.8.1. Define I = [a, b]. A function u C([a,

b]; RN )
is said to be a weak Lipschitz extremal of
Z b
F (u) =
f (x, u(x), u(x)) dx,
a
if
(fs (x, u(x), u (x)) + f (x, u(x), u(x)) (x)) dx = 0,
Cc ([a, b]; RN ).
Proposition 7.8.2. For any Lipschitz extremal of
Z b
F (u) =
f (x, u(x), u(x)) dx
a
there exists a constant c RN such that

Z x
f (x, u(x), u (x)) = c +

fs (t, u(t), u(t)) dt, x [a, b]. (7.14)
a
Proof. Fix Cc ([a, b]; RN ). Integration by parts of the extremal condition

F (u, ) = 0,
implies that
Z b
f (x, u(x), u(x)) (x) dx
a
Z bZ x
fs (t, u(t), u(t)) dt (x) dx = 0.

a
Since is arbitrary, from the du Bois-Reymond lemma, there exists

c RN such that
Z x
f (x, u(x), u (x))

fs (t, u(t), u(t)) dt = c, x [a, b].
a
190
Theorem 7.8.3 (Weierstrass-Erdmann Corner Conditions). Let

I = [a, b]. Suppose u C 1 ([a, b]; RN ) is such that
F (u) F (v), v Cr ,
for some r > 0. where
Cr = {v C 1 ([a, b]; RN ) | v(a) = u(a), v(b) = u(b),
and ku vk < r}.
Let x0 (a, b) be a corner point of u. Denoting u0 = u(x0 ),

= u (x0 + 0) and 0 = u (x0 0), then the following relations
are valid:
(1) f (x0 , u0 , 0 ) = f (x0 , u0, 0+ ),
(2)
0+
f (x0 , u0 , 0 ) 0 f (x0 , u0 , 0 )
= f (x0 , u0 , 0+ ) 0+ f (x0 , u0 , 0+ ).
Remark 7.8.4. The conditions above are known as the Weierstrass-Erdmann corner conditions.
Proof. Condition (1) is just a consequence of equation (7.14).
For (2), define
(x) = x + (x),
where Cc (I). Observe that (a) = a and (b) = b, > 0.
Also 0 (x) = x. Choose 0 > 0 sufficiently small such that for each
satisfying || < 0 , we have (x) > 0 and
u (x) = (u 1 )(x) Cr .
Define
() = F (x, u , u (x)).
Thus has a local minimum at 0, so that (0) = 0, that is
d(F (x, u
, u (x)))
|=0 = 0.
d
Observe that
d 1 (x)
d
u
= u (1 (x))
,
dx
dx
7.9. NATURAL BOUNDARY CONDITIONS
and
191
d1 (x)
1
=
.
dx
1 + (1 (x))
Thus,
F (
u ) =
Defining
b
a

1
1
f x, u( (x)), u ( (x))
1
1 + (1 (x))

dx.
x = 1 (x),
we obtain
d
x=
1
dx,
1 + (
x)
that is
dx = (1 + (
x)) d
x.
Dropping the bar for the new variable, we may write

Z b
u (x)
F (
u ) =
f x + (x), u(x),
(1 + (x)) dx.
1 + (x)
a
From
dF (
u )
|=0 ,
d
we obtain
Z b
(fx (x, u(x), u (x)) + (x)(f (x, u(x), u (x))
a
u (x)f (x, u(x), u(x)))) dx = 0. (7.15)
Since is arbitrary, from proposition 7.7.2, we obtain

Z x
f (x, u(x), u (x))u (x)f (x, u(x), u (x))

fx (t, u(t), u(t)) dt = c1
a
for some cR1 R.

x
Being a fx (t, u(t), u(t)) dt + c1 a continuous function (in fact
absolutely continuous), the proof is complete.

7.9. Natural Boundary Conditions
Consider the functional f : U R, where
Z
F (u) f (x, u(x), u(x)) dx,
RN , RN n ),
f (x, s, ) C 1 (,
and Rn is an open bounded connected set.
192
Proposition 7.9.1. Assume

U = {u W 1,2 (; RN ); u = u0 on 0 },
where 0 is closed and = = 0 1 being 1 open in

RN , RN n ) and
and 0 1 = . Thus if C 1 , f C 2 (,
2
N
u C (; R ), and also
RN ), such that = 0 on 0 ,
F (u, ) = 0, C 1 (;
then u is a extremal of F which satisfies the following natural

boundary conditions,
n fxi i (x, u(x)u(x)) = 0, a.e. on 1 , i {1, ..., N}.
Proof. Observe that F (u, ) = 0, Cc (; RN ), thus u is
a extremal of F and through integration by parts and the fundamental lemma of calculus of variations, we obtain
Lf (u) = 0, in ,
where
Lf (u) = fs (x, u(x), u(x)) div(f (x, u(x), u(x)).
Defining
V = { C 1 (; RN ) | = 0 on 0 },
for an arbitrary V, we obtain

Z
F (u, ) =
Lf (u) dx
Z
n fxi i (x, u(x), u(x))i (x) d
+
Z 1
n fxi i (x, u(x), u(x))i(x) d
=
1
= 0, V.

n fxi i (x0 , u(x0 ), u(x0 )) = > 0,
for some x0 1 and some i {1, ..., N}. Defining

G(x) = n fxi i (x, u(x), u(x)),
by the continuity of G, there exists r > 0 such that

G(x) > /2, in Br (x0 ),
(7.16)
7.9. NATURAL BOUNDARY CONDITIONS
193
and in particular
G(x) > /2, in Br (x0 ) 1 .
Choose 0 < r1 < r such that Br1 (x0 ) 0 = . This is possible since
0 is closed and x0 1 .
Choose i Cc (Br1 (x0 )) such that i 0 in Br1 (x0 ) and i >
0 in Br1 /2 (x0 ). Therefore
Z
Z
i
i dx > 0,
G(x) (x) dx >
2
1
1
and this contradicts (7.16). Thus
and by analogy
so that
G(x) 0, x 1 ,
G(x) 0, x 1 ,
G(x) = 0, x 1 .
CHAPTER 8
Basic Concepts on Convex Analysis

For this chapter the most relevant reference is Ekeland and
Temam, [17].
8.1. Convex Sets and Convex Functions
Let S be a subset of a vector space U. We recall that S is convex
if given u, v S then
u + (1 )v S, [0, 1].
(8.1)
Definition 8.1.1 (Convex hull). Let S be a subset of a vector

space U, we define the convex hull of S, denoted by Co(S) as
( n
X
Co(S) =
i ui | n N,
i=1
n
X
i=1
i = 1, i 0, ui S, i {1, ..., n} . (8.2)
Definition 8.1.2 (Convex Functional). Let S be convex sub = R{+, }

set of the vector space U. A functional F : S R
is said to be convex if
F (u + (1 )v) F (u) + (1 )F (v), u, v S, [0, 1].
(8.3)
Definition 8.1.3 (Lower Semi-continuity). Let U be Banach
is lower semi-continuous (l.s.c.) at
space. We say that F : U R
u U, if
lim inf F (un ) F (u),
(8.4)
un u strongly (in norm).
(8.5)
n+
whenever
195
196
8. BASIC CONCEPTS ON CONVEX ANALYSIS
Definition 8.1.4 (Weak Lower Semi-Continuity). Let U be

is weakly lower semiBanach space. We say that F : U R
continuous (w.l.s.c.) at u U, if
lim inf F (un ) F (u),
(8.6)
un u, weakly.
(8.7)
n+
whenever
Remark 8.1.5. We say that F is a (weak) lower semi-continuous

is (weak) lower semi-continuous u U.
function, if F : U R
we define its
Definition 8.1.6 (Epigraph). Given F : U R
Epigraph, denoted by Epi(F ) as
Epi(F ) = {(u, a) U R | a F (u)}.
Now we present a very important result but which we do not
prove. For a proof see Ekeland and Temam, [17].
is l.s.c. (lower
Proposition 8.1.7. A function F : U R
semi-continuous) if and only if its epigraph is closed.
is
Corollary 8.1.8. Every convex l.s.c. function F : U R
also w.l.s.c. (weakly lower semi-continuous).
Proof. The result follows from the fact that the epigraph of F
is convex and closed convex sets are weakly closed.

Definition 8.1.9 (Affine Continuous Function). Let U be
a Banach space. A functional F : U R is said to be affine
continuous if there exist u U and R such that
F (u) = hu, uiU + , u U.
(8.8)
Definition 8.1.10 ((U)). Let U be a Banach space, we say

belongs to (U) and write F (U) if F can
that F : U R
be represented as the point-wise supremum of a family of affine
continuous functions. If F (U), and F (u) R for some u U
then we write F 0 (U).
8.1. CONVEX SETS AND CONVEX FUNCTIONS
197
Proposition 8.1.11. Let U be a Banach space, then F (U)

if and only if F is convex and l.s.c., and if F takes the value
then F .
Definition 8.1.12 (Convex Envelope). Let U be a Banach
we define its convex envelope, denoted
space. Given F : U R,
by CF : U R by
CF (u) =
sup {hu, ui + },
(8.9)
(u ,)A
where
A = {(u , ) U R | hv, uiU + F (v), v U}
(8.10)
we
Definition 8.1.13 (Polar Functionals). Given F : U R,
define the related polar functional, denoted by F : U R, by

F (u ) = sup{hu, uiU F (u)}, u U .
(8.11)
uU
Definition 8.1.14 (Bipolar Functional). Given F : U R,
we define the related bipolar functional, denoted by F : U R,

as
F (u) = sup {hu, uiU F (u )}, u U.
(8.12)
u U
then F (u) = CF (u)

Proposition 8.1.15. Given F : U R,
and in particular if F (U) then F (u) = F (u).

Proof. By definition, the convex envelope of F is the supremum
of all affine continuous minorants of F . We can consider only the
maximal minorants, that functions of the form
u 7 hu, uiU F (u ).
(8.13)
Thus,
CF (u) = sup {hu, uiU F (u )} = F (u).
(8.14)
u U

we have F = F .
Corollary 8.1.16. Given F : U R,
198
Proof. Since F F we obtain
F F .
(8.15)
F (u) hu, uiU F (u ),
(8.16)
On the other hand, we have

so that
F (u ) = sup{hu, uiU F (u)} F (u ).
(8.17)
From (8.15) and (8.17) we obtain F (u) = F (u ).
uU
Definition 8.1.17 (Gateaux Differentiability). A functional

is said to be Gateaux differentiable at u U if there
F :U R
exists u U such that:

F (u + h) F (u)
= hh, uiU , h U.
(8.18)
lim
0
The vector u is said to be the Gateaux derivative of F : U R

at u and may be denoted as follows:
F (u)
u =
or u = F (u)
(8.19)
u
we
Definition 8.1.18 (Sub-gradients). Given F : U R,
define the set of sub-gradients of F at u, denoted by F (u) as:
F (u) = {u U , such that hv u, uiU + F (u) F (v),
v U}. (8.20)
Definition 8.1.19 (Adjoint Operator). Let U and Y be Banach spaces and : U Y a continuous linear operator. The
Adjoint Operator related to , denoted by : Y U is defined
through the equation:
hu, v iU = hu, v iY , u U, v Y .
(8.21)
Lemma 8.1.20 (Continuity of Convex Functions). If in a

neighborhood of a point u U, a convex function F is bounded
above by a finite constant, then F is continuous at u.
Proof. By translation, we may reduce the problem to the case
where u = and F (u) = 0. Let V be a neighborhood of origin such
that F (v) a < +, v V. Define W = V (V) (which is
199
a symmetric neighborhood of origin). Pick (0, 1). If v W,

since F is convex and
v
V
(8.22)
we may infer that

Also
F (v) (1 )F (0) + F (v/) a.
(8.23)
v
V.
(8.24)
F (v) (1 + )F (0) F (v/) a.
(8.25)
|F (v)| a, v W,
(8.26)
Thus
Therefore
that is, F is continuous at u = .
be a convex function
Proposition 8.1.21. Let F : U R
finite and continuous at u U. Then F (u) 6= .
Proof. Since F is convex, Epi(F ) is convex, as F is continuous
at u, we have that Epi(F ) is non-empty. Observe that (u, F (u))
belongs to the boundary of Epi(F ), so that denoting A = Epi(F ),
we may separate (u, F (u)) from
A by a closed hyper-plane H, which
may be written as
H = {(v, a) U R | hv, uiU + a = },
(8.27)
hv, uiU + a , (v, a) Epi(F ),
(8.28)
hu, uiU + F (u) = ,
(8.29)
hv u, uiU 0, v U,
(8.30)
hv, u/iU F (v),
(8.31)
for some fixed , R and u U , so that

and
where (, , u) 6= (0, 0, ). Suppose = 0, thus we have
and thus we obtain u = , and = 0. Therefore we may assume

> 0 (considering (8.28)) so that v U we have
200
and
hu, u/iU = F (u),
(8.32)
hv u, u /iU + F (u) F (v), v U,
(8.33)
u / F (u).
(8.34)
or
so that

Definition 8.1.22 (Caratheodory Mapping ). Let S Rn be
an open set, we say that that g : S Rl R is a Caratheodory
mapping if:
Rl , x 7 g(x, ) is a measurable function,
and
for almost all x S, 7 g(x, ) is a continuous function.
The proof of next results may be found in Ekeland and Temam
[17].
Proposition 8.1.23. Let E and F be two Banach spaces, S
a Borel subset of Rn , and g : S E F a Caratheodory mapping. For each measurable function u : S E, let G1 (u) be the
measurable function x 7 g(x, u(x)) F .
If G1 maps Lp (S, E) into Lr (S, F ) for 1 p, r < , then G1 is
continuous in the norm topology.
R
For the functional G : U R, defined by G(u) = S g(x, u(x))dS
, where U = U = [L2 (S)]l (this is a especial case of the more general hypothesis presented in [17]) we have the following result.
Proposition 8.1.24. Considering the last proposition we can
as :
express G : U R
Z

G (u ) =
g (x, u (x))dS,
(8.35)
S
where g (x, y) = sup (y g(x, )), almost everywhere in S.

Rl
201
For non-convex functionals it may be sometimes difficult to express analytically conditions for a global extremum. This fact motivates the definition of Legendre Transform, which is established
through a local extremum.
Definition 8.1.25 (Legendres Transform and Associated Functional). Consider a differentiable function g : Rn R. Its Legendre Transform, denoted by gL : RLn R is expressed as:
gL (y ) = x0i yi g(x0 ),
(8.36)
where x0 is the solution of the system:
g(x0 )
,
(8.37)
xi
and RLn = {y Rn such that equation (8.37) has a unique solution}.
Furthermore,
considering the functional G : Y R defined as
R
G(v) = S g(v)dS, we define the Associated Legendre Transform
Functional, denoted by GL : YL R as:
Z
GL (v ) =
gL (v )dS,
(8.38)
yi =
where
YL
= {v Y | v (x)
S
n
RL ,
a.e. in S}.
About the Legendre transform we still have the following results:

Proposition 8.1.26. Considering the last definitions, suppose
that for each y RLn at least in a neighborhood (of y ) it is possible
to define a differentiable function by the expression
g
x0 (y ) = [ ]1 (y ).
(8.39)
x
Then, i {1, ..., n}we may write:
yi =
g (y )
g(x0 )
x0i = L
xi
yi
(8.40)
Proof. Suppose firstly that:

yi =
g(x0 )
, i {1, ..., n},
xi
(8.41)
thus:
gL (y ) = yi x0i g(x0 )
(8.42)
202
and taking derivatives for this expression we have:

g(x0 ) x0j
x0j
gL (y )
= yj + x0i
,
yi
yi
xj yi
(8.43)
gL (y )
g(x0 ) x0j
= (yj
) + x0i
yi
xj
yi
(8.44)
or
which from (8.41) implies that:

gL (y )
= x0i , i {1, ..., n}.
yi
(8.45)
This completes the first half of the proof. Conversely, suppose now
that:
x0i =
gL (y )
, i {1, ..., n}.
yi
(8.46)
As y RLn there exists x0 Rn such that:

yi =
g(
x0 )
i {1, ..., n},
xi
(8.47)
and,
gL (y ) = yi x0i g(
x0 )
(8.48)
and therefore taking derivatives for this expression we can obtain:

gL (y )
g(
x0 ) x0j
x0j
= yj + x0i
,
yi
yi
xj yi
(8.49)
i {1, ..., n}, so that:
g(
x0 ) x0j
gL (y )
= (yj
) + x0i
yi
xj
yi
(8.50)
i {1, ..., n}, which from (8.46) and (8.47), implies that:
x0i =
gL (y )
= x0i , i {1, ..., n},
yi
(8.51)
from this and (8.47) we have:

yi =
g(
x0 )
g(x0 )
=
i {1, ..., n}.
xi
xi
(8.52)
203
deTheorem 8.1.27. Consider the functional J : U R

fined as J(u) = (G )(u) hu, f iU where (= {i }) : U Y
(i {1, ..., n}) is a continuous linear operator Rand, G : Y R is
a functional that can be expressed as G(v) = S g(v)dS, v Y
(here g : Rn R is a differentiable function that admits Legendre
Transform denoted by gL : RLn R. That is, the hypothesis mentioned at Proposition 8.1.26 are satisfied).
Under these assumptions we have:
J(u0 ) = (GL (v0 ) + hu0 , v0 f iU ) = ,
(8.53)
0 ))
where v0 = G((u
is supposed to be such that v0 (x) RLn , a.e.
v
in S and in this case:
J(u0 ) = GL (v0 ).
(8.54)
Proof. Suppose first that J(u0 ) = , that is:
which, as v0 =
and
G(u0 )
v
G(u0 )
f =
v
(8.55)
implies that:
v0 f = ,
g(u0 )
.
xi
Thus from the last proposition we can write:
g (v )
i (u0 ) = L 0 , for i {1, .., n}
yi
which means:
GL (v0 )
u0 =
.
v
Therefore from (8.56) and (8.59) we have:
v0i =
(8.56)
(8.57)
(8.58)
(8.59)
(GL (v0 ) + hu0 , v0 f iU ) = .
(8.60)
(GL (v0 ) + hu0 , v0 f iU ) = ,
(8.61)
v0 f =
(8.62)
This completes the first part of the proof.

Conversely, suppose now that:
that is:
204
and
GL (v0 )
.
(8.63)
v
Clearly, from (8.63), the last proposition and (8.62) we can write:
u0 =
v0 =
G((u0 ))
v
(8.64)
and
G(u0 )
f = ,
v
(8.65)
which implies:
J(u0 ) = .
(8.66)
J(u0 ) = G(u0 ) hu0, f iU
(8.67)
Finally, we have:
From this, (8.62) and (8.64) we have
J(u0 ) = G(u0 ) hu0 , v0 iU = G(u0 ) hu0 , v0 iY
= GL (v0 ). (8.68)
8.2. Duality in Convex Optimization

(F 0 (U)) we
Let U be a Banach space. Given F : U R
define the problem P as
P : minimize F (u) on U.
(8.69)
(u, 0) = F (u),
(8.70)
We say that u0 U is a solution of problem P if F (u0) = inf uU F (u).

such that
Consider a function (u, p) ( : U Y R)
we define the problem P , as
Observe that
(0, p ) =
P : maximize (0, p ) on Y .
sup
(u,p)U Y
(8.71)
{h0, uiU + hp, p iY (u, p)} (u, 0),

(8.72)
or
inf {(u, 0)} sup { (0, p)}.
uU
p Y
(8.73)
8.2. DUALITY IN CONVEX OPTIMIZATION
205
Proposition 8.2.1. Consider 0 (U Y ). If we define

h(p) = inf {(u, p)},
(8.74)
uU
then h is convex.
Proof. We have to show that given p, q Y and (0, 1), we
have
h(p + (1 )q) h(p) + (1 )h(q).
(8.75)
If h(p) = + or h(q) = + we are done. Thus let us assume

h(p) < + and h(q) < +. For each a > h(p) there exists u U
such that
h(p) (u, p) a,
(8.76)
and, if b > h(q), there exists v U such that

h(q) (v, q) b.
(8.77)
Thus
h(p + (1 )q) inf {(w, p + (1 )q)}
wU
(u + (1 )v, p + (1 )q) (u, p) + (1 )(v, q)
a + (1 )b. (8.78)
Letting a h(p) and b h(q) we obtain

h(p + (1 )q) h(p) + (1 )h(q).
(8.79)
Proposition 8.2.2. For h as above, we have h (p ) = (0, p ),

p Y , so that
h (0) = sup { (0, p)}.
(8.80)
p Y
Proof. Observe that

h (p ) = sup{hp, p iY h(p)} = sup{hp, p iY inf {(u, p)}},
pY
pY
uU
(8.81)
so that
h (p ) =
sup
(u,p)U Y
{hp, p iY (u, p)} = (0, p ).
(8.82)
206
Proposition 8.2.3. The set of solutions of the problem P

(the dual problem) is identical to h (0).
Proof. Consider p0 Y a solution of Problem P , that is,
(0, p0 ) (0, p ), p Y ,
(8.83)
which is equivalent to
h (p0 ) h (p ), p Y ,
(8.84)
h(p0 ) = sup {h0, piY h (p )} h (p0 ) = h (0)
p Y
p0 h (0). (8.85)
convex. Assume
Theorem 8.2.4. Consider : U Y R
inf uU {(u, 0)} R and there exists u0 U such that p 7 (u0 , p)
is finite and continuous at 0 Y , then
inf {(u, 0)} = sup { (0, p )},
uU
(8.86)
p Y
and the dual problem has at least one solution.

Proof. By hypothesis h(0) R and as was shown above, h is
convex. As the function p 7 (u0, p) is convex and continuous at
0 Y , there exists a neighborhood V of zero in Y such that
(u0 , p) M < +, p V,
(8.87)
for some M R. Thus, we may write

h(p) = inf {(u, p)} (u0 , p) M, p V.
uU
(8.88)
Hence, from Lemma 8.1.20, h is continuous at 0. Thus by Proposition 8.1.21, h is sub-differentiable at 0, which means h(0) = h (0).
Therefore by Proposition 8.2.3, the dual problem has solutions and
h(0) = inf {(u, 0)} = sup { (0, p )} = h (0).
uU
(8.89)
p Y
Now we apply the last results to (u, p) = G(u + p) + F (u),

where : U Y is a continuous linear operator whose adjoint
operator is denoted by : Y U . We may enunciate the
following theorem.
8.2. DUALITY IN CONVEX OPTIMIZATION
207
Theorem 8.2.5. Suppose U is a reflexive Banach space and

define J : U R by
J(u) = G(u) + F (u) = (u, 0),
(8.90)
where lim J(u) = + as kukU and F 0 (U), G 0 (Y ).

Also suppose there exists u U such that J(
u) < + with the
function p 7 G(p) continuous at
u. Under such hypothesis, there
exist u0 U and p0 Y such that
{G (p ) F ( p )}
J(u0 ) = min{J(u)} = max
uU
p Y
= G (p0 ) F ( p0 ). (8.91)
Proof. The existence of solutions for the primal problem follows
from the direct method of calculus of variations. That is, considering a minimizing sequence, from above (coercivity hypothesis),
such a sequence is bounded and has a weakly convergent subsequence to some u0 U. Finally, from the lower semi-continuity of
primal formulation, we may conclude that u0 is a minimizer. The
other conclusions follow from Theorem 8.2.4 just observing that
(0, p ) =
sup {hp, piY G(u + p) F (u)}
uU,pY
sup {hq, p i G(q) hu, pi F (u)}, (8.92)
uU,qY
so that
(0, p ) = G (p ) + sup{hu, p iU F (u)}
uU
= G (p ) + F ( p ). (8.93)
Thus,
inf {(u, 0)} = sup { (0, p )}
uU
(8.94)
p Y
and solutions u0 and p0 for the primal and dual problems, respectively, imply that
{G (p ) F ( p )}
J(u0 ) = min{J(u)} = max
uU
p Y
= G (p0 ) F ( p0 ). (8.95)
208
8.3. Relaxation for the Scalar Case

In this section, RN denotes a bounded open set with a
locally Lipschitz boundary. That is, for each point x there
exists a neighborhood Ux whose the intersection with is the
graph of a Lipschitz continuous function.
We start with the following definition.
Definition 8.3.1. A function u : R is said to be affine
if u is constant on . Furthermore, we say that u : R is
piecewise affine if it is continuous and there exists a partition of
into a set of zero measure and finite number of open sets on which
u is affine.
The proof of next result is found in [17].
Theorem 8.3.2. Let r N and let uk , 1 k r be piecewise
affine functionsPfrom into R and {k } such that k > 0, k
{1, ..., r} and rk=1 k = 1. Given > 0, there exists a locally
Lipschitz function u : R and r disjoint open sets k , 1 k r,
such that
|m(k ) k m()| < k , k {1, ..., r},
(8.96)
u(x) = uk (x), a.e. on k ,
(8.97)
|u(x)| max {|uk (x)|}, a.e. on ,
(8.98)
1kr

r

X

k uk < , x ,
u(x)

(8.99)
k=1
u(x) =
r
X
k=1
k uk (x), x .
(8.100)
The next result is also found in [17].

Proposition 8.3.3. Let r N and let uk , 1 k r be
piecewise affine functions from into R. Consider a Caratheodory
function f : RN R and a positive function c L1 () which
8.3. RELAXATION FOR THE SCALAR CASE
209
satisfy
c(x) sup{|f (x, )| | || max {kuk k }}.
1kr
(8.101)
Given > 0, there exists a locally Lipschitz function u : R

such that

Z
Z
r

X

(8.102)
f
(x,
u)dx
f
(x,
u
)dx
< ,

k
k

k=1
|u(x)| max {|uk (x)|}, a.e. in ,

1kr
|u(x)
r
X
k=1
u(x) =
k uk (x)| < , x
r
X
k=1
k uk (x), x .
(8.103)
(8.104)
(8.105)
Proof. It is sufficient to establish the result for functions uk

affine over , since can be divided into pieces on which uk are
affine, and such pieces can be put together through (8.105). Let
> 0 be given. We know that simple functions are dense in L1 (),
concerning the L1 norm. Thus there exists a partition of into a
finite number of open sets Oi , 1 i N1 and a negligible set, and
there exists fk constant functions over each Oi such that
Z
|f (x, uk (x)) fk (x)|dx < , 1 k r.
(8.106)
Now choose > 0 such that
N1 (1 + max1kr {kfk k })
and if B is a measurable set
Z
m(B) <
c(x)dx /N1.
(8.107)
(8.108)
Now we apply Theorem 8.3.2, to each of the open sets Oi , therefore

there exists a locally Lipschitz function u : Oi R and there exist
r open disjoints spaces ik , 1 k r, such that
|m(ik ) k m(Oi )| k , for 1 k r,
u = uk , a.e. in ik ,
(8.109)
(8.110)
210
|u(x)| max {|uk (x)|}, a.e. Oi ,

1kr

r

X

k uk (x) , x Oi
u(x)

(8.111)
(8.112)
k=1
u(x) =
r
X
k=1
k uk (x), x Oi .
(8.113)
P
1
We can define u = rk=1 k uk on N
i=1 Oi . Therefore u is continuous and locally Lipschitz. Now observe that
Z
Oi
f (x, u(x))dx
r Z
X
k=1
ik
f (x, uk (x))dx
Oi rk=1 ik
f (x, u(x))dx. (8.114)
From |f (x, u(x))| c(x), m(Oi rk=1 ik ) and (8.108) we

obtain

Z
r Z

X

f (x, uk (x)dx
f (x, u(x))dx

Oi
i
k

Zk=1

f (x, u(x))dx /N1. (8.115)
=

Oi rk=1 ik
Considering that fk is constant in Oi , from (8.107), (8.108) and

(8.109) we obtain
Z
r
X
|
k=1
ik
fk (x)dx k
fk (x)dx| < /N1.

Oi
(8.116)
211
i
1
We recall that k = N
i=1 k so that

Z
Z
r

X

k
f (x, uk (x))dx
f (x, u(x))dx

k=1

Z
r Z

X

f (x, uk (x))dx
f (x, u(x))dx

k=1 k
r Z
X
|f (x, uk (x) fk (x)|dx
+
+
+
k=1 k
r Z
X
k=1
r
X
k=1
fk (x)dx k

fk (x)dx
|fk (x) f (x, uk (x))|dx.
(8.117)
From (8.115), (8.106),(8.116) and (8.106) again, we obtain

Z
Z
r

X

f
(x,
u(x))dx
f
(x,
u
)dx
< 4. (8.118)

k
k

k=1
The next result we do not prove it. It is a well known result

from the finite element theory.
Proposition 8.3.4. If u W01,p () there exists a sequence

{un } of piecewise affine functions over , null on , such that
and
un u, in Lp ()
(8.119)
un u, in Lp (; RN ).
(8.120)
Proposition 8.3.5. For p such that 1 0 such that
a2 (x) + c2 ||p f (x, ) a1 (x) + c1 ||p , x , RN . (8.121)
Then, given u W 1,p () piecewise affine, > 0 and a neighborhood

V of zero in the topology (Lp (, RN ), Lq (, RN )) there exists a
function v W 1,p () such that
v u V,
u = v on ,
(8.122)
212
kv uk < ,
and

Z
Z

< .
f (x, v(x))dx
f
(x,
u(x))dx

(8.123)
(8.124)
Proof. Suppose given > 0, u W 1,p () piecewise affine continuous, and a neighborhood V of zero, which may be expressed
as

Z

p
N
V = {w L (, R ) | hm wdx < ,
m {1, ..., M}}, (8.125)
where M N, hm Lq (, RN ), R+ . By hypothesis, there

exists a partition of into a negligible set 0 and open subspaces
i , 1 i r, over which u(x) is constant. From standard
results of convex analysis in RN , for each i {1, ..., r} we can
P +1
obtain {k 0}1kN +1, and k such that N
k=1 k = 1 and
N
+1
X
k k = u, x i ,
(8.126)
k f (x, k ) = f (x, u(x)).
(8.127)
k=1
and
N
+1
X
k=1
Define i = maxk{1,...,N +1}{|k | on i }, and 1 = maxi{1,...,r}{i },

and = max{1 , kuk }. Now, observe that we can obtain func m C (; RN ) such that
tions h
0
m hm kLq (,RN ) <

max kh
.
(8.128)
m{1,...,M }
4m()
m )kLq () and we can also define
Define C = maxm{1,...,M } kdiv(h
1 = min{/4, 1/(m()1/p ), /(2Cm()1/p ), 1/m()}
(8.129)
We recall that does not depend on . Furthermore, for each

i {1, ..., r} there exists a compact subset Ki i such that
Z
1
[a1 (x) + c1 (x) max{||p}]dx < .
(8.130)
||
r
i Ki
Also, observe that the restrictions of f and f to Ki B are
continuous, so that from this and from the compactness of B,
213
for all x Ki , we can find an open ball x with center in x and

contained in , such that
1
, y x Ki , (8.131)
|f (y, u(x)) f (x, u(x))| <
m()
and
|f (y, ) f (x, )| <
1
, y x Ki , B.
m()
(8.132)
Therefore, from this and (8.127) we may write

N
+1

X
21

, y x Ki . (8.133)
k f (y, k ) <
f (y, u(x))
m()

k=1
We can cover the compact set Ki with a finite number of those

open balls x , denoted by j , 1 j l. Consider the open sets
j1
j = j i=1
i , we have that lj=1

j = lj=1
j . Defining functions
P +1
uk , for 1 k N + 1 such that uk = k and u = N
k=1 k uk we
may apply Proposition 8.3.3 to each of the open sets j , so that we

obtain functions vi W 1,p () such that

Z
Z
N
+1

X

1
f (x, k )dx < ,
k
f (x, vi (x)dx
(8.134)

j
rl
j
k=1
and
Finally we set
|vi | < , x j ,
(8.135)
|vi (x) u(x)| < 1 , x j ,
(8.136)
vi (x) = u(x), x j .
(8.137)
vi = u on i lj=1 j .
(8.138)
v(x) = vi (x), if x i ,
(8.139)
v(x) = u(x), if x 0 .
(8.140)
We may define a continuous mapping v : R by
We have that v(x) = u(x), x and kvk < . Also, from

(8.130)
Z
1
(8.141)
|f (x, u(x)|dx <
r
i Ki
214
and
Z
i Ki
|f (x, v(x)|dx <
1
.
r
(8.142)
On the other hand, from (8.133) and (8.134)

Z

Z

f (x, v(x))dx
f (x, u(x))dx

Ki j

Ki j
1 1 m(j Ki )
+
(8.143)
rl
m()
so that
Z
Z
|
f (x, v(x))dx
Ki
Ki
f (x, u(x))dx|
1 1 m(Ki )
+
. (8.144)
r
m()
Now summing up in i and considering (8.141) and (8.142) we obtain

(8.124), that is
Z
Z
| f (x, v(x))dx
f (x, u(x))dx| < 41 . (8.145)
Also, observe that from above, we have

kv uk < 1 ,
(8.146)
and thus

Z

Z

m )(v(x) u(x))dx
m (v(x) u(x))dx = div(h
h

m )kLq () kv ukLp (S)

kdiv(h
C1 m()1/p
<
.
2
Also we have that

Z

m hm ) (v u)dx
(h

(8.147)
m hm kLq (,RN ) kv ukLp (,RN ) . (8.148)

kh
2
215
Thus

Z

hm (v u)dx < , m {1, ..., M}.

(8.149)
Theorem 8.3.6. Assuming the hypothesis of last theorem,

given a function u W01,p (), given > 0 and a neighborhood
of zero V in (Lp (, RN ), Lq (, RN )), we have that there exists a
function v W01,p () such that
v u V,
(8.150)
and

Z
Z

< .
f (x, v(x))dx
f
(x,
u(x))dx

(8.151)
Proof. We can approximate u by a function w which is piecewise affine and null on the boundary. Thus, there exists > 0 such
that we can obtain w W01,p () piecewise affine such that
ku wk1,p <
(8.152)
1
w u V,
2
(8.153)
so that
and

Z
Z

< .
f (x, w(x))dx
f
(x,
u(x))dx
2

(8.154)
From Proposition 8.3.5 we may obtain v W01,p () such that

1
v w V,
2
(8.155)
and

Z
Z

< .
f (x, w(x))dx
f
(x,
v(x))dx
2

From (8.154) and (8.156)

Z

Z

f (x, u(x))dx
f (x, v(x))dx < .

(8.156)
(8.157)
216
Finally, from (8.153), (8.155) and from the fact the weak neighborhoods are convex, we have
v u V.
(8.158)
To finish this chapter, we present two theorems which summarize the last results.
Theorem 8.3.7. Let f be a Carath
eodory function from
RN into R which satisfies
a2 (x) + c2 ||p f (x, ) a1 (x) + c1 ||p
(8.159)
where a1 , a2 L1 (), 1 0. Under

such assumptions, defining U = W01,p (), we have
Z

Z

f (x, u)dx
(8.160)
f (x, u)dx = min
inf
uU
uU
The solutions of relaxed problem are weak cluster points in W01,p ()

of the minimizing sequences of primal problem.
Proof. The existence of solutions for the convex relaxed formulation is a consequence of the reflexivity of U and coercivity
hypothesis, which allows an application of the direct method of
calculus of variations. That is, considering a minimizing sequence,
from above (coercivity hypothesis), such a sequence is bounded and
has a weakly convergent subsequence to some u W 1,p (). Finally,
from the lower semi-continuity of relaxed formulation, we may conclude that u is a minimizer. The relation (8.160) follows from last
theorem.

Theorem 8.3.8. Let f be a Carath
eodory function from
RN into R which satisfies
a2 (x) + c2 ||p f (x, ) a1 (x) + c1 ||p
(8.161)
where a1 , a2 L1 (), 1 0. Let

= {u | u u0
u0 W 1,p (). Under such assumptions, defining U
1,p
W0 ()}, we have
Z

Z

f (x, u)dx
(8.162)
f (x, u)dx = min
inf
uU
uU
8.4. DUALITY SUITABLE FOR THE VECTORIAL CASE
217
The solutions of relaxed problem are weak cluster points in

W 1,p () of the minimizing sequences of primal problem.
Proof. Just apply the last theorem to the integrand g(x, ) =
f (x, + u0 ). For details see [17].

8.4. Duality Suitable for the Vectorial Case
Definition 8.4.1 (A Cone and its Partial Order Relation). Let
U be a Banach space and m > 0. We define C(m) as
C(m) = {(u, a) U R | a + mkukU 0}.
(8.163)
Also, we define an order relation for the cone C(m), namely

(u, a) (v, b) (v u, b a) C(m).
(8.164)
Proposition 8.4.2. Let S U R be a closed set such that

inf{a | (u, a) S} > .
(8.165)
Then S has a maximal element under the order relation of last

definition.
For a proof see [17], page 28.
The next result is particularly relevant for non-convex functionals.
be lower semi-continuous
Theorem 8.4.3. Let F : U R
functional such that < inf uU {F (u)} < +. Given > 0,
suppose u U is such that
F (u) inf {F (u)} + ,
uU
(8.166)
then, for each > 0, there exists u U such that

ku u kU and F (u) F (u).
(8.167)
Proof. We will apply the last proposition to S = Epi(F ),

which is a closed set. For the order relation associated with C(/),
there exists a maximal element, which we denote by (u, a ). Thus
(u , a ) (u, F (u)). Since (u , a , ) is maximal, we have a =
F (u). Also observe that
(u, F (u)) (u , F (u)),
(8.168)
218
so that
ku u kU F (u) F (u ).
(8.169)
inf {F (u)} F (u ) F (u) kuu kU F (u) inf {F (u)}+,

uU
uU
so that
0 F (u) F (u) ,
(8.170)
ku ukU .
(8.171)
and therefore

Remark 8.4.4. Observe that
F (u) tkvkU F (u + tv), t [0, 1], v U,
so that, if F is Gateaux differentiable, we obtain
kvkU hF (u ), viU .
Thus
kF (u)kU /.
Now, for =
(8.172)
(8.173)
(8.174)
we obtain the following result.
Theorem 8.4.5. Let F : U R be a Gateaux differentiable

functional. Given > 0 suppose that u U is such that
F (u) inf {F (u)} + .
uU
(8.175)
Then there exists v U such that
F (v) F (u),
ku vkU
and
kF (v)kU
(8.176)
(8.177)
.
(8.178)
The next theorem easily follows from above results.
Theorem 8.4.6. Let J : U R, be defined by

where
219
J(u) = G(u) hf, uiL2 (S;RN ) ,
(8.179)
U = W01,2 (S; RN ),
(8.180)
We suppose G is Gateaux-differentiable and J bounded from below.

Then, given > 0, there exists u U such that
J(u ) inf {J(u)} < ,
uU
and
kJ(u )kU <
.
(8.181)
(8.182)
We finish this Chapter with the most important result we have

obtained for vectorial problems in the Calculus of Variations, namely:
Theorem 8.4.7. Let U be a reflexive Banach space. Consider
(G ) : U R and (F 1 ) : U R l.s.c. functionals such that
J : U R defined as
J(u) = (G )(u) (F 1 )(u) hu, f iU
is below bounded. (Here : U Y and 1 : U Y1 are continuous linear operators whose adjoint operators are denoted by
: Y U and 1 : Y U , respectively). Also we suppose
the existence of L : Y1 Y continuous and linear operator such
that L is onto and
(u) = L(1 (u)), u U.
Under such assumptions, we have
inf {J(u)} sup { inf {F (L z ) G (v + z )}},
uU
v A z Y1
where
A = {v Y | v = f }.
In addition we assume (F 1 ) : U R is convex and Gateaux
differentiable, and suppose there exists a solution (v0 , z0 ) of the
dual formulation, so that,

F (L z0 )
L
G (v0 + z0 ),
v
v0 f = 0.
220
Suppose u0 U is such that

F (L z0 )
= 1 u 0 ,
v
so that
u0 G (v0 + z0 ).
Also we assume that there exists a sequence {un } U such that
un u0 weakly in U and
G(un ) G (u0 ) as n .
Under these additional assumptions we have
{ inf {F (L z ) G (v + z )}}
inf {J(u)} = max
uU
v A
z Y1
= F (L z0 ) G (v0 + z0 ).
Proof. Observe that
G (v + z ) hu, v iY + hu, z iY G(u), u U,
that is,
F (L z ) + G (v + z ) hu, f iU F (L z ) + h1 u, L z iY1
G(u), u U, v A
so that
sup {F (L z ) + G (v + z )}
z Y1
sup {hu, f iU F (L z ) + h1 u, L z iY1 G(u)},

z Y1
v A , u U and therefore
G(u) F (1 u) hu, f iU inf {F (L z ) G (v + z )},
z Y1
v A , u U
which means
inf {J(u)} sup { inf {F (L z ) G (v + z )}},
uU
v A z Y1
where
A = {v Y | v = f }.
Now suppose
L
F (L z0 )
v
G (v0 + z0 ),
and u0 U is such that
221
F (L z0 )
= 1 u 0 .
v
Observe that
implies that
u0 = L(1 u0 ) G(v0 + z0 )
G (v0 + z0 ) = hu0, v0 iY + hu0, z0 iY G (u0 ).
From the hypothesis
un u0 weakly in U
and
G(un ) G (u0) as n .
Thus, given > 0, there exists n0 N such that if n n0 then
G (v0 + z0 ) hun , v0 iY hun , z0 iY + G(un ) < /2.
On the other hand, since F (1 u) is convex and l.s.c. we have

lim sup{F (1 un )} F (1 u0).
n
Hence, there exists n1 N such that if n n1 then
hun , z0 iY F (1 un ) hu0, z0 iY F (1 u0 ) + = F (L z0 ) + ,
2
2
so that for all n max{n0 , n1 } we obtain
G (v0 + z0 ) F (L z0 ) hun , f iU F (1 un ) + G(un ) < .
Since is arbitrary, the proof is complete.
CHAPTER 9
Constrained Variational Optimization

9.1. Basic Concepts
For this chapter the most relevant reference is the excellent book
of Luenberger, [27], where more details may be found. We start
with the definition of cone:
Definition 9.1.1 (Cone). Given U a Banach space, we say
that C U is a cone with vertex at origin, if given u C, we have
that u C, 0. By analogy we define a cone with vertex at
p U as P = p + C, where C is any cone with vertex at origin.
Definition 9.1.2. Let P be a convex cone in U. For u, v U
we write u v (with respect to P ) if u v P . In particular u
if and only if u C. Also
P + = {u U | hu, uiU 0, u P }.
(9.1)
If u P + we write u .
Proposition 9.1.3. Let U be a Banach space and P be a
closed cone in U. If u U satisfies hu, uiU 0, u , then
u .
Proof. We prove the contrapositive. Assume u 6 P . Then
by the separating hyperplane theorem there is an u U such
that hu, uiU < hp, uiU , p P . Since P is cone we must have
hp, uiU 0, otherwise we would have hu, ui > hp, uiU for some
> 0. Thus u P + . Finally, since inf pP {hp, uiU } = 0, we
obtain hu, uiU < 0 which completes the proof.

Definition 9.1.4 (Convex Mapping).
Let U, Z be vector
spaces. Let P Z be a cone. A mapping G : U Z is said
223
224
9. CONSTRAINED VARIATIONAL OPTIMIZATION
to be convex if the domain of G is convex and

G(u1 + (1 )u2 ) G(u1) + (1 )G(u2),
u1 , u2 U, [0, 1]. (9.2)
Consider the problem P, defined as

P roblem P : Minimize F : U R subject to u , and G(u)
Define
(z) = inf{F (u) | u and G(u) z}.
(9.3)
For such a functional we have the following result.

Proposition 9.1.5. If F is a real convex functional and G is
convex, then is convex.
Proof. Observe that
(z1 + (1 )z2 ) = inf{F (u) | u
and G(u) z1 + (1 )z2 }
(9.4)
inf{F (u) | u = u1 + (1 )u2 u1 , u2

and G(u1 ) z1 , G(u2 ) z2 }
(9.5)
inf{F (u1 ) | u1 , G(u1 ) z1 }
+ (1 ) inf{F (u2) | u2 , G(u2 ) z2 }
(9.6)
(z1 ) + (1 )(z2 ).
(9.7)
Now we establish the Lagrange multiplier theorem for convex

global optimization.
Theorem 9.1.6. Let U be a vector space, Z a Banach space,
a convex subset of U, P a positive cone of Z. Assume that P
contains an interior point. Let F be a real convex functional on
and G a convex mapping from into Z. Assume the existence of
9.1. BASIC CONCEPTS
225
u1 such that G(u1 ) < . Defining

0 = inf {F (u) | G(u) },
(9.8)
then there exists z0 , z0 Z such that
0 = inf {F (u) + hG(u), z0 iZ }.
(9.9)
Furthermore, if the infimum in (9.8) is attained by u0 U such

that G(u0 ) , it is also attained in (9.9) by the same u0 and also
hG(u0 ), z0 iZ = 0. We refer to z0 as the Lagrangian Multiplier.
Proof. Consider the space W = R Z and the sets A, B where
A = {(r, z) (R, Z) | r F (u), z G(u) f or some u },
(9.10)
and
B = {(r, z) (R, Z) | r 0 , z },
(9.11)
where 0 = inf u {F (u) | G(u) }. Since F and G are convex,

A and B are convex sets. It is clear that A contains no interior
point of B, and since N = P contains an interior point , the set
B contains an interior point. Thus, from the separating hyperplane
theorem, there is a non-zero element w0 = (r0 , z0 ) W such that
r0 r1 + hz1 , z0 iZ r0 r2 + hz2 , z0 iZ , (r1 , z1 ) A, (r2 , z2 ) B.

(9.12)
From the nature of B it is clear that w0 . That is, r0 0 and

z0 . We will show that r0 > 0. The point (0 , ) B, hence
r0 r + hz, z0 iZ r0 0 , (r, z) A.
(9.13)
If r0 = 0 then hG(u1 ), z0 iZ 0 and z0 6= . Since G(u1 ) < and

z0 we have a contradiction. Therefore r0 > 0 and, without
loss of generality we may assume r0 = 1. Since the point (0 , ) is
arbitrarily close to A and B, we have
0 = inf {r + hz, z0 iZ } inf {F (u) + hG(u), z0 iZ }
(r,z)A
inf{F (u) | u , G(u) } = 0 . (9.14)

Also, if there exists u0 such that G(u0 ) , 0 = F (u0), then
0 F (u0 ) + hG(u0), z0 iZ F (u0 ) = 0 .
(9.15)
226
Hence
hG(u0), z0 iZ = 0.
(9.16)
Corollary 9.1.7. Let the hypothesis of the last theorem hold.

Suppose
F (u0) = inf {F (u) | G(u) }.
u
(9.17)
Then there exists z0 such that the Lagrangian L : U Z R

defined by
L(u, z ) = F (u) + hG(u), z iZ
(9.18)
has a saddle point at (u0 , z0 ). That is
L(u0 , z ) L(u0 , z0 ) L(u, z0 ), u , z .
(9.19)
Proof. For z0 obtained in the last theorem, we have

L(u0 , z0 ) L(u, z0 ), u .
(9.20)
As hG(u0 ), z0 iZ = 0, we have
L(u0 , z ) L(u0 , z0 ) = hG(u0 ), z iZ hG(u0), z0 iZ
= hG(u0), z iZ 0. (9.21)
We now prove two theorems relevant to develop the subsequent

section.
Theorem 9.1.8. Let F : U R and G : Z. Let

P Z be a cone. Suppose there exists (u0 , z0 ) U Z where
z0 and u0 are such that
F (u0 ) + hG(u0 ), z0 iZ F (u) + hG(u), z0 iZ , u .
(9.22)
Then
F (u0 ) + hG(u0), z0 iZ
= inf{F (u) | u and G(u) G(u0)}. (9.23)
Proof. Suppose there is a u1 such that F (u1 ) < F (u0 ) and

G(u1 ) G(u0). Thus
so that
hG(u1 ), z0 iZ hG(u0), z0 iZ
F (u1 ) + hG(u1), z0 iZ < F (u0) + hG(u0 ), z0 iZ ,
(9.24)
(9.25)
9.2. DUALITY
which contradicts the hypothesis of the theorem.
227
Theorem 9.1.9. Let F be a convex real functional and G :

Z convex and let u0 and u1 be solutions to the problems P0
and P1 respectively, where
and
P0 : minimize F (u) subject to u and G(u) z0 ,
(9.26)
P1 : minimize F (u) subject to u and G(u) z1 .
(9.27)
hz1 z0 , z1 iZ F (u0) F (u1 ) hz1 z0 , z0 iZ .
(9.28)
Suppose z0 and z1 are the Lagrange multipliers related to these

problems. Then
Proof. For u0 , z0 we have

F (u0 ) + hG(u0 ) z0 , z0 iZ F (u) + hG(u) z0 , z0 iZ , u ,
(9.29)
and, particularly for u = u1 and considering that hG(u0 )z0 , z0 iZ =
0, we obtain
F (u0 ) F (u1 ) hG(u1) z0 , z0 iZ hz1 z0 , z0 iZ .
(9.30)
A similar argument applied to u1 , z1 provides us the other inequality.

9.2. Duality
Consider the basic convex programming problem:
Minimize F (u) subject to G(u) , u ,
(9.31)
where F : U R is a convex functional, G : U Z is convex

mapping, and is a convex set. We define : Z R by
(z ) = inf {F (u) + hG(u), z iZ }.
u
(9.32)
Proposition 9.2.1. is concave and

(z ) = inf {(z) + hz, z iZ },
(9.33)
(z) = inf {F (u) | G(u) z},
(9.34)
where
u
228
and
= {z Z | G(u) z f or some u }.
Proof. Observe that
(z ) = inf {F (u) + hG(u), z iZ }
u
inf {F (u) + hz, z iZ | G(u) z}

u
= (z) + hz, z iZ , z , z .
(9.35)
On the other hand, for any u1 , defining z1 = G(u1 ), we obtain

F (u1 ) + hG(u1), z iZ inf {F (u) + hz1 , z iZ | G(u) z1 }
u
= (z1 ) + hz1 , z iZ , (9.36)

so that
(z ) inf {(z) + hz, z iZ }.
z
(9.37)
Theorem 9.2.2 (Lagrange Duality). Consider F : U

R a convex functional, a convex set, and G : U Z a convex
mapping. Suppose there exists a u1 such that G(u1 ) < and that
inf u {F (u) | G(u) } < . Under such assumptions, we have
inf {F (u) | G(u) } = max
{(z )}.
(9.38)
If the infimum on the left side in (13.20) is achieved at some u0 U

and the max on the right side at z0 Z , then
hG(u0), z0 iZ = 0
(9.39)
and u0 minimizes F (u) + hG(u), z0 iZ on .

Proof. For z we have
inf {F (u) + hG(u), z iZ }
inf
u,G(u)
{F (u) + hG(u), z iZ }
inf
u,G(u)
F (u) 0 . (9.40)
or
(z ) 0 .
(9.41)
The result follows from Theorem 9.1.6.
9.3. THE LAGRANGE MULTIPLIER THEOREM
229
9.3. The Lagrange Multiplier Theorem

Definition 9.3.1. Let T : D V be a continuously Frechet
differentiable operator, where D U is open, being U and V Banach spaces. We say that u0 D is a regular point of T if T (u0 )
maps U onto V .
Theorem 9.3.2 (Generalized Inverse Function Theorem). Let
u0 be a regular point of T : U V , where U and V are Banach
spaces. Defining v0 = T (u0), there exists a neighborhood BrV (v0 )
and a constant K > 0 such that the equation
v = T (u)
has a solution for each v BrV (v0 ) and such a solution is such that
ku u0 kU Kkv v0 kV .
Proof. Define L0 = N(T (u0 )). Observe that U/L0 is a Banach

space, for which we may define
A : U/L0 V
by
A(
u) = T (u0 )u.
Since T (u0 ) is onto, so is A, so that by the inverse mapping theorem, A has a continuous inverse A1 . Choose > 0 such that
1
<
4kA1 k
Choose r > 0 such that if ku u0kU < r then kT (u) T (u0 )k < .
Let v Br (v0 ) where r = 4kAr1 k . Define g0 = L0 and also
define
L1 = A1 [v T (u0 g0 )].
Choose g1 L1 such that
kg1 g0 kU 2kL1 L0 k.
This is possible since
kL1 L0 k = inf {kg g0 kU }.

gL1
Therefore
so that
L1 = A1 [v v0 ],
kL1 k kA1 kkv v0 kV ,
230
and hence
kg1 kU 2kL1 k
2kA1 kkv v0 kV
2kA1 kr
r
.
2
Now reasoning by induction, for n 2, assume that
define Ln by
(9.42)
kgn1 kU < r and kgn2kU < r,
Ln Ln1 = A1 (v T (u0 + gn1 )),
(9.43)
and choose gn Ln such that
kgn gn1 kU 2kLn Ln1 k.
This is possible since
kLn Ln1 k = inf {kg gn1 kU }.

gLn
Observe that we may write

Ln1 = gn1 = A1 [A[
gn1 ]] = A1 [T (u0 )gn1 ],
and thus
and
so that
Ln = A1 [v T (u0 + gn1) + T (u0)gn1 ],

Ln1 = A1 [v T (u0 + gn2 ) + T (u0 )gn2],
Ln Ln1 = A1 [T (u0 +gn1 )T (u0 +gn2 )T (u0 )(gn1 gn2 )].

Define
gt = tgn1 + (1 t)gn2 .
By the generalized mean value inequality, we obtain
kLn Ln1 k kA1 kkgn1 gn2 kU sup {kT (u0 + gt ) T (u0 )k}
t(0,1)
being
Hence
kgt kU < r.
kLn Ln1 k kA1 kkgn1 gn2 kU ,
9.3. THE LAGRANGE MULTIPLIER THEOREM
231
so that
kgn gn1 kU 2kLn Ln1 k
2kA1 kkgn1 gn2 kU
1
kgn1 gn2 kU .
2
(9.44)
Thus
kgn kU = kgn gn1 + gn1 gn2 + gn2 gn3 + ... g0 kU
1
1
kg1 kU (1 + + ... + n )
2
2
< 2kg1 kU r.
(9.45)
Therefore
kgn kU < r, n N
and
1
kgn gn1 kU kgn1 gn2 kU , n N.
2
Finally, since {gn } is a Cauchy sequence and U is a Banach space,
there exists g U such that gn g as n , and thus
Ln L = g, as n .
From this and (9.43) we get
= L L = A1 [v T (u0 + g)],
and since A1 is a bijection, we obtain
v = T (u0 + g).
Furthermore
kgkU 2kg1kU 4kA1 kkv v0 kV ,
so that we may choose
K = 4kA1 k.
Before the final result, we need the lemma:

Lemma 9.3.3. Suppose the functional F : U R achieves
a local extremum under the constraint H(u) = at the point u0 .
Also assume that F and H are continuously Frechet differentiable
in an open set containing u0 and that u0 is a regular point of H.
Then hh, F (u0 )iU = 0 for all h satisfying H (u0 )h = .
232
Proof. Without loss of generality, suppose the local extremum

is a minimum. Consider the transformation T : U R Z
defined by T (u) = (F (u), H(u)). Suppose there exists a h such
that H (u0 )h = , F (u0 )h 6= 0, then T (u0 ) = (F (u0 ), H (u0 )) :
U R Z is onto since H (u0 ) is onto Z (and obviously since
F (u0 )h 6= 0, F (u0 ) is onto R). By the inverse function theorem,
given > 0 there exists u U and > 0 with ku u0 kU < such
that T (u) = (F (u0) , ), which contradicts the assumption that
u0 is a local minimum.

Theorem 9.3.4 (Lagrange Multiplier). Suppose F is a continuously Frechet differentiable functional which has a local extremum
under the constraint H(u) = at the regular point u0, then there
exists a Lagrangian multiplier z0 Z such that the Lagrangian
functional
L(u) = F (u) + hH(u), z0 iZ
(9.46)
is stationary at u0 , that is, F (u0 ) + H (u0 ) z0 = .

Proof. From last lemma we have that F (u0 ) is orthogonal to
the null space of H (u0 ). Since the the range of H (u0 ) is closed, it
follows that
therefore there exists
F (u0 ) R[(H (u0 )) ],
(9.47)
F (u0 ) = H (u0 ) z0 .
(9.48)
z0
Z such that
Part 3
Applications
CHAPTER 10
Duality Applied to a Plate Model

10.1. Introduction
The main objective of the present chapter is to develop systematic approaches for obtaining dual variational formulations for
systems originally modeled by non-linear differential equations.
Duality for linear systems is well established and is the main
subject of classical convex analysis, since in case of linearity, both
primal and dual formulations are generally convex. In case of nonlinear differential equations, some complications occur and the standard models of duality for convex analysis must be modified and
extended.
In particular in the case of Kirchhoff-Love plate model, there
is a non-linearity concerning the strain tensor (that is, a geometric
non-linearity). To apply the classical results of convex analysis and
obtain the complementary formulation is possible only for a special
class of external loads. This leads to non-compressed plates, please
see Telega [40], Gao [22] and other references therein.
We now describe the primal formulation and related duality
principles. Consider a plate whose middle surface is represented by
an open bounded set S R2 , whose boundary is denoted by ,
subjected to a load to be specified. We denote by u : S R ( =
1, 2) the horizontal displacements and by w : S R, the vertical
displacement field. The boundary value form of the Kirchhoff-Love
model can be expressed by the equations:
(
N, = 0,
(10.1)
Q, + M, + P = 0, a.e. in S
and
N .n P = 0,
(M t n )
(Q + M, )n +
P = 0,
s
M n n Mn = 0, on t ,
235
(10.2)
236
10. DUALITY APPLIED TO A PLATE MODEL
where,
N = H ,
M = h
and,
1
(u) = (u, + u, + w, w, ),
2
(u) = w, ,
with the boundary conditions
w
= 0, on u .
u = w =
n
Here, {N } denote the membrane forces, {M } denote the moments and {Q } = {N w, } stand for functions related to the
rotation work of membrane forces, P L2 (S) is a field of vertical distributed forces applied on S, (P , P ) (L2 (t ))3 denote
forces applied to t concerning the horizontal directions defined by
= 1, 2 and vertical direction respectively. Mn are distributed moments applied also to t , where is such that u , = u t
and u t = . Finally, the matrices {H } and {h } are
related to the coefficients of Hookes Law.
The corresponding primal variational formulation to this boundary value model is represented by the functional J : U R, where
Z
Z
Z
1
1
J(u) =
H dS +
h dS P wdS
2 S
2 S
S
Z
w
)d
(P w + P u Mn
n
t
and
U = {(u , w) W 1,2 (S) W 1,2 (S) W 2,2 (S),

w
u = w =
= 0 on u }.
n
The first duality principle presented is the classical one (again we
mention the earlier similar results in Telega [40], Gao [22]) , and is
obtained by applying a little change of Rockafellars approach for
convex analysis. We have developed a different proof from the one
found in [40], by using the definition of Legendre Transform and
related properties. Such a result may be summarized as
inf {J(u)} =
uU
sup
v A C
{GL (v )}
(10.3)
10.1. INTRODUCTION
237
is exThe dual functional, denoted by GL : A C R

pressed as
GL (v )
Z
Z
1
1
M M dS
=
H N N dS +
h
2 S
2 S

Z
1
N Q Q dS ,
+
2 S
where C is defined by equations (10.1) and (10.2) and

A = {v Y | N11 > 0, N22 > 0, and
2
N11 N22 N12
> 0, a.e. in S}, (10.4)
here v = {N , M , Q } Y = L2 (S; R10 ) L2 (S)

Therefore, since the functional GL (v ) is convex in A , the duality is perfect if the optimal solution for the primal formulation
satisfies the constraints indicated in (10.4), however it is important
to emphasize that such constraints imply no compression along the
plate.
For the second and third principles, we emphasize that our dual
formulations remove or relax the constraints on the external load,
and are valid even for compressed plates.
Still for these two principles, we use a theorem (Toland, [42])
which does not require convexity of primal functionals. Such a
result can be summarized as:
inf {G(u) F (u)} = inf {F (u ) G (u )}
uU
u U
Here G : U R and F : U R and, F : U R and G : U

R denote the primal and dual functionals respectively.
In particular for the second principle, we modify the above result
by applying it to a not one to one relation between primal and dual
variables, obtaining the final duality principle expressed as follows
inf
(u,p)U Y
{JK (u, p)}
inf
(
u,v )U Y
{JK
(
u, v )}
where
JK (u, p) = G(u + p) F (u) +
K
hp, piL2 (S)
2
238
and

2

g
(v
)
L

= F ( v ) GL (v ) + K
u

y L2 (S)
1
+
hv , v iL2 (S) .
2K
Here K R is a positive constant and we are particularly concerned
with the fact that
JK
(
u, v )
JK (uK , pK ) J(u0 ), as K +
and
where
JK
(
uK , vK
) J(u0 ), as K +
JK (uK , pK ) =
JK
(
uK , vK
)=
inf
(u,p)U Y
inf
{JK (u, p)},
(
u,v )U Y
{JK
(
u, v )}
and
J(u0 ) = inf {J(u) = G(u) F (u)}.
uU
Even though we do not prove it in the present article, postponing

a more rigorous analysis concerning the behavior of uK indicated
above as K +, for a future work.
For the third duality principle, the dual variables must satisfy
the following constraints :
N11 + K > 0, N22 + K > 0 and
2
(N11 + K)(N22 + K) N12
> 0, a.e. in S. (10.5)
Such a principle may be summarized by the following result,

inf {G(u)F (1u)hu, piU } inf { sup {F (z )GL (v )}},
uU
z Y
v B (z )
where
B (z ) = {v Y | v 1 z p = 0}
Therefore the constant K > 0 must be chosen so that the optimal point of the primal formulation satisfies the constraints indicated in (10.5). This is because these relations also define an
enlarged region in which the analytical expression of the functional
GL : Y R is convex, so that, in this case, negative membrane
forces are allowed.
10.1. INTRODUCTION
239
In Section 10.7, we present a convex dual variational formulation

which may be expressed through the following duality principle:
inf {J(u)} =
uU
sup
(v ,z )E B
{G (v ) + hz , z iL2 (S) /(2K)}
where,
G (v ) =
GL (v )
1
=
2
N N dS+ 1
H
2
S
1
+
2
if v E , where
ZS
M M dS
h
N
Q, Q, dS
v = {{N }, {M }, {Q }} E v L2 (S, R10 )
and
N11 + K > 0
N22 + K > 0
and
where
2
(N11 + K)(N22 + K) N12
> 0, a.e. in S,
K }
{N
N11 + K
N12
N12
N22 + K
1
(10.6)
and
(v , z ) B
N, + P = 0,
Q, + M, z,
+ P = 0,
1212 M12 + z /K = 0,
h
1,2
z1,2 = z2,1
, a.e. in S, and, z n = 0 on .
We are assuming the existence of u0 U such that J(u0 ) =

, and so that there exists K > 0 for which N11 (u0 ) + K > 0,
N22 (u0 ) + K > 0 , (N11 (u0 ) + K)(N22 (u0 ) + K) N12 (u0 )2 > 0 (a.e
in S) and h1212 /(2K0 ) > K where K0 is the constant related to a
Poincare type Inequality and,
N (u0 ) = H (u0 ).
Finally, in the last section, we prove a result similar to those
obtained through the triality criterion introduced in Gao [24] and
establish sufficient conditions for the existence of a minimizer for
240
the primal formulation. Such conditions may be summarized by

J(u0 ) = and
Z
Z
1
1
N (u0 )w,w, dS +
h w, w, dS 0,
2 S
2 S
w W02,2 (S). 2
For this last result, our proof is new. The statement of results
themselves follows those of Gao [24].
We are now ready to state the result of Toland [42], through
which will be constructed three duality principles.
Theorem 10.1.1. Let J : U R be a functional defined as
J(u) = G(u) F (u), u U, where there exists u0 U such that
J(u0 ) = inf {J(u)} and F (u0 ) 6= , then
uU
inf {G(u) F (u)} = inf {F (u ) G (u )}
uU
u U
and for u0 F (u0 ) we have,

F (u0 ) G (u0 ) = inf {F (u ) G (u )}.
u U
Furthermore u0 G(u0 ).
10.2. The Primal Variational Formulation
Let S R2 be an open bounded set (with a boundary denoted
by ) which represents the middle surface of a plate of thickness
h. The vectorial basis related to the Cartesian system {x1 , x2 , x3 }
is denoted by (a , a3 ), where = 1, 2 (in general Greek indices
stand for 1 or 2), a3 denotes the vector normal to S, t is the vector
tangent to and n is the outer normal to S. The displacements
will be denoted by:
= {
u
u , u3 } = u a + u3 a3 ,
The Kirchhoff-Love relations are
u (x1 , x2 , x3 ) = u (x1 , x2 ) x3 w(x1 , x2 ),
and
u3 (x1 , x2 , x3 ) = w(x1 , x2 ),
10.2. THE PRIMAL VARIATIONAL FORMULATION
241
where h/2 x3 h/2 so that we have u = (u , w) U where

U = (u , w) W 1,2 (S) W 1,2 (S) W 2,2 (S),

w
u = w =
= 0 on u .
n
We divide the boundary into two parts, so that u , = u t

and u t = . The strain tensors are denoted by
1
(u) = [1 (u) + 2 (u)2 (u)]
(10.7)
2
and
(u) = 3 (u)
(10.8)
where: = {{1 }, {2 }, {3 }} : U Y = Y = L2 (S; R10 )

L2 (S) is defined by:
1 (u) = u, + u,,
(10.9)
2 (u) = w,
(10.10)
3 (u) = w, .
(10.11)
N = H ,
(10.12)
M = h
(10.13)
and
The constitutive relations are expressed as
where: {H } and {h = h12 H }, are positive definite matrices and such that H = H = H = H . Furthermore {N } denote the membrane forces and {M } the moments.
The plate stored energy, denoted by (G ) : U R is expressed
as
Z
Z
1
1
N dS +
M dS
(10.14)
(G )(u) =
2 S
2 S
and the external work, denoted as F : U R, is given by
Z
Z
w
(P w + P u Mn
F (u) =
P wdS +
)d,
(10.15)
n
S
t
where P denotes a vertical distributed load applied in S and P , P
are forces applied on t related to directions defined by a3 and
242
a respectively, and, Mn denote moments also applied on t . The

potential energy, denoted by J : U R is expressed as:
J(u) = (G )(u) F (u)
It is important to emphasize that conditions for the existence

of a minimizer (here denoted by u0 ) related to G(u) F (u) were
presented in Ciarlet [14]. Such u0 U satisfies the equation:
(G(u0) F (u0 )) =
and we should expect at least one minimizer if kP kL2 (t ) is small

enough and m(u ) > 0 (where m stands for the Lebesgue measure)
and with no restrictions concerning the magnitude of kP kL2 (S) if
m() = m(u ), so that in the latter case, we consider a field of
distributed forces {P } applied on S.
Some inequalities of Sobolev type are necessary to prove the
above result, and in this work we assume some regularity hypothesis
concerning S and its boundary, namely: in addition to S being
open and bounded, also we assume it is connected with a Lipschitz
continuous boundary , so that S is locally on one side of .
The formal proof of existence of a minimizer for J(u) = G(u)
F (u) is obtained through the Direct Method of Calculus of variations. We do not repeat this procedure here, we just refer to Ciarlet
[14] for details.
10.3. The Legendre Transform
In this section we determine the Legendre Transform related to
the function g : R10 R where:
1
g(y) = H [(y1 + y1 + y2 y2 )/2][(y1 + y1 + y2 y2 )/2]
2
1
+ h y3 y3 (10.16)
2
and we recall that
Z
G(u) =
g(u)dS.
S
From Definition 8.1.25 we may write
gL (y ) = hy0 , y iR10 g(y0)
where y0 is the unique solution of the system,

g(y0 )
yi =
yi
10.3. THE LEGENDRE TRANSFORM
243
which for the above function g, implies:

y1 = H (y1 + y2 y2 /2)
y2 = H (y1 , +y2 y2 /2)y2 = y1 y2 ,
and
y3 = h y3 .
Inverting this system we obtain
y021 = (y122 .y21 y112 .y22 )/,
y022 = (y112 .y21 + y111 .y22 )/,
and
where
y y02 .y02 /2
y01 = H
1
} = {H }1 ,
{H
= y111 y122 (y112 )2 (we recall that y112 = y121 , as a result of the
symmetries of {H }).
By analogy,
v
y03 = h
3
where:
} = {h }1 .
{h
Thus we can define the set RLn , concerning Definition 8.1.25 as

RLn = {y R10 | 6= 0}.
(10.17)
After some simple algebraic manipulations we obtain the expression

for gL : RLn R, that is,
1
1
1
gL (y ) = H
y1 y1 + h y3 y3 + y1 y02 y02 .
2
2
2
(10.18)
Also from Definition 8.1.25, we have

YL = {v Y = L2 (S; R10 ) L2 (S) | v (x) RLn a.e. in S}
so that GL : YL R may be expressed as
Z
GL (v ) =
gL (v )dS.
S
244
Or, from (10.18),

Z
Z
1
1
v v dS
v v dS +
GL (v ) =
H
h
1
1
3
3
2 S
2 S
Z
1
v v02 v02 dS. 2
+
2 S 1
Changing the notation, as indicated below,
v1 = N , v2 = Q = v1 v02 = N v02 , v3 = M
we could express GL : YL R as
Z
Z
1
1
h M M dS
N N dS +
GL (v ) =
H
2 S
2 S
Z
1
Q Q dS,
+
N
2 S
where
= {N }1 .
N
Remark 10.3.1. Also we can use the transformation
Q = N w,
and obtain
1
=
2
N N dS + 1
H
2
S
h M M dS
Z
1
N w, w, dS.
+
2 S
The term denoted by Gp : Y U R and expressed as
Z
1
Gp (v , w) =
N w, w, dS
2 S
is known as the gap function.
GL (v )
10.4. The Classical Dual Formulation

In this section we establish the dual variational formulation in
the classical sense.
We recall that J : U R is expressed by
J(u) = (G )(u) F (u),
where (G ) : U R and F : U R were defined by equations

(10.14) and (10.15) respectively. It is known and easy to see that
10.4. THE CLASSICAL DUAL FORMULATION
inf {G(u) + F (u)} sup {G (v ) F ( v )}.
uU
245
(10.19)
v Y
Now we prove a result concerning the representation of the polar

functional, namely:
Proposition 10.4.1. Considering the earlier definitions and
assumptions
on G : Y R (see section 10.2), expressed by G(v) =
R
g(v)dS, where g : R10 R is indicated in (10.16), we have
S
v A G (v ) = GL (v )
where
GL (v )
1
=
2
N N dS + 1
H
2
S
and
h M M dS
S
Z
1
Q Q dS
+
N
2 S
A = {v = {N , M , Q } Y | N11 > 0, N22 > 0, and
2
N11 N22 N12
> 0, a.e. in S}. (10.20)
Proof. First, consider the quadratic inequality in x as indicated

below,
a
x2 + bx + c 0, x R,
(
a < 0 and b2 4
ac 0) or (
a = 0, b = 0 and c 0).
(10.21)
Consider now the inequality

a1 x2 + b1 xy + c1 y 2 + d1 x + e1 y + f1 0, x, y R2
(10.22)
and the quadratic equation related to the variable x, for

a
= a1 , b = b1 y + d1 and c = c1 y 2 + e1 y + f1 ,
and for a1 < 0, from (10.21) the inequality (10.22) is equivalent to
(b21 4a1 c1 )y 2 + (2b1 d1 4a1 e1 )y + d21 4a1 f1 0, y R.
Finally, for
a
= b21 4a1 c1 < 0, b = 2b1 d1 4a1 e1 and c = d21 4a1 f1 ,
246
also from (10.21), the last inequality is equivalent to

c1 d21 a1 e21 + b1 d1 e1 (b21 4a1 c1 )f1 0.
(10.23)
In order to represent the polar functional related to the plate stored

energy, we first consider the polar functional related to g1 (y), where
1
1
1
g1 (y) = H (y1 + y2 y2 )(y1 + y2 y2 ),
2
2
2
g(y) = g1 (y) + g2 (y)
and
1
g2 (y) = h y3 y3 .
2
In fact we determine a set in which the polar functional is repre
sented by the Legendre Transform g1L
(y ), where, from (10.18),
1
g1L
(y ) = H
y1 y1
2
y 11 (y22 )2 2.y112 y21 y22 + y122 (y21 )2
. (10.24)
+ 1
2[y111 y122 (y112 )2 ]
Thus, since
g1 (y ) = sup {y1 y1 + y2 y2 g1 (y)}
yR6
we can write
g1L
(y ) = g1 (y ) g1L
(y ) y1 y1 + y2 y2 g1 (y), y R6 .
Or
1
1
1
y1 y1 + y2 y2 H (y1 + y2 y2 )(y1 + y2 y2 )
2
2
2
g1L
(y ) 0, y R6 . (10.25)
However, considering the transformation
1
y1 = y1 + y2 y2 ,
2
1
y1 = y1 y2 y2 ,
2
and substituting such relations into (10.25), we obtain
g1L
(y ) = g1 (y )
(10.26)
247
1
1
y1 (
y1 y2 y2 ) + y2 y2 H y1 y1 g1L
(y ) 0,
2
2
{
y1 , y2 } R6 . (10.27)
On the other hand, since {H } is a positive definite matrix we
have
1
1
sup {y1 y1 H y1 y1 } = H
y1 y1 . (10.28)
2
2
{
y1 }R4
Thus considering (10.28) and the expression of gL (y ) indicated in
(10.24) , inequality (10.27) is satisfied if
y 11 .(y22 )2 2.y112 .y21 y22 + y122 .(y21 )2
1
y1 y2 y2 + y2 y2 1
2
2[y111 y122 (y112 )2 ]
So, for
0, {y2 } R2 . (10.29)
1
1
a1 = y111 , b1 = y112 , c1 = y122 , , d1 = y21 , e1 = y22
2
2
and
y 11 .(y22 )2 2.y112 .y21 y22 + y122 .(y21 )2
f1 = 1
2[y111 y122 (y112 )2 ]
we obtain
c1 d21 a1 e21 + b1 d1 e1 (b21 4a1 c1 )f1 = 0
Therefore from (10.23), the inequality (10.25) is satisfied if a1 < 0

(y111 > 0) and b21 4a1 c1 < 0 (y111 y122 (y112 )2 > 0 which implies
y122 > 0).
Thus we have shown that
y A g1(y ) = g1L
(y ),
(10.30)
where
A = {y R6 | y111 > 0, y122 > 0, y111 y122 (y112 )2 > 0}.
On the other hand, by analogy to above results, it can easily be

proved that
where
g2 (y ) = g2L
(y ), {y3 } R3
(10.31)
g2L
(y ) = h
y3 y3
2
(10.32)
248
and
1
g2 (y ) = sup {y3 y3 h y3 y3 }.
2
yR3
From (10.30) and (10.31), we can write
if y A then g1 (y )+g2(y ) = g1L

(y )+g2L
(y ) (g1 +g2 ) (y ).
As (g1 + g2 ) (y ) g1 (y ) + g2 (y ) we have
if y A then gL (y ) = g1L
(y ) + g2L
(y )
= (g1 + g2 ) (y ) = g (y ). (10.33)
However, from Proposition 8.1.24

Z

G (v ) =
g (v )dS
(10.34)
so that from (10.33) and (10.34) we obtain

Z

gL (v )dS = GL (v )
v A G (v ) =
S
where,
A = {v Y | v (x) A , a.e. in S}.
Alternatively,
A = {v Y | v111 > 0, v122 > 0, and v111 v122 (v112 )2 > 0,
a.e. in S}. (10.35)
Thus, through the notation

v1 = N , v2 = Q = v1 v02 = N v02 , v3 = M
we have
A = {v = {N , M , Q } Y | N11 > 0, N22 > 0, and
2
N11 N22 N12
> 0, a.e. in S}. (10.36)
We
10.4.1. The Polar Functional Related to F : U R.
are concerned with the evaluation of the extremum,
F ( v ) = sup{hu, v iU F (u)},
uU
or
F ( v ) = sup{hu, v iY F (u)}.
uU
249
Considering
F (u) =
Z
we have

w
(P w + P u Mn
P wdS +
)d
n
S
t
= hu, f iU (10.37)
Z
0, if v C ,
+, otherwise,
(10.38)
where v C v Y and

v1, = 0,
v2, + v3, + P = 0, a.e. in S,
(10.39)
F ( v ) =
and

v1 .n P = 0,
(v t n )
3
P = 0,
(v
+
v
).n
+
2
3,
s

v3 n n Mn = 0, on t .
(10.40)
Remark 10.4.2. We can also denote
C = {v Y | v = f },
(10.41)
where the relation v = f is defined by (10.39) and (10.40).

10.4.2. The First Duality Principle. Considering inequality (10.19), the expression of G (v ), and the set C above defined,
we can write
inf {(G )(u) F (u)}
uU
sup
v A C
{GL (v )}
(10.42)
so that the final form of the concerned duality principle results from
the following theorem.
Theorem 10.4.3. Let (G ) : U R and F : U R be
defined by (10.14) and (10.15) respectively (and here we express F
as F (u) = hu, f iU ). If GL : YL R attains a local extremum at
v0 A under the constraint v f = 0, then
inf {(G )(u) + F (u)} =
uU
sup
v A C
{GL (v )}
250
and u0 U and v0 Y such that:
{GL (v0 ) + hu0 , v0 f iU } =
are also such that
J(u0 ) = GL (v0 ) and J(u0 ) = .

The proof of above theorem is consequence of the standard necessary conditions for a local extremum for GL : YL R under the
constraint v f = , the inequality (10.42) plus an application
of Theorem 8.1.27.
Therefore, in a more explicit format we would have
Z
Z
1
1
H dS +
h dS
inf
uU
2 S
2 S
Z

Z
Z
Z
w
Mn
P u d
P wdS +
P wdS +
d
n
t
t
S
t

Z
Z
1
1
M M dS
=
sup
H N N dS
h
2
2
v A C
S
S

Z
1
Q Q dS
N
(10.43)
2 S
where v C v Y and,

N, = 0,
Q, + M, + P = 0, a.e. in S
and
N .n P = 0,
(M t n )
(Q + M, )n +
P = 0,
s
with the set A defined by (10.20) and

} = {N }1 .
{N
10.5. The Second Duality Principle
The next result is a extension of Theorem 10.1.1 and, instead of

calculating the polar functional related to the main part of primal
formulation, it is determined its Legendre Transform and associated
functional.
10.5. THE SECOND DUALITY PRINCIPLE
251
Theorem 10.5.1. Consider G

ateaux differentiable function and F 1 : U R
where only the second
als G : U R
one is necessarily convex, through which is defined the functional
expressed as
JK : U Y R
Khp, piL2 (S)

hu, u0 iU .
2
Suppose there exists (u0 , p0 ) U Y such that
JK (u, p) = G(u+p)+Khp, piL2 (S) F (1 u)
JK (u0 , p0 ) =
inf
(u,p)U Y
{JK (u, p)}
and JK (u0 , p0 ) = . Here = {i } : U Y and 1 : U Y are

continuous linear operators whose adjoint operators are denoted by
: Y U and 1 : Y U respectively.
Furthermore assume there exists a differentiable function de may be expressed as
noted byR g : Rn R so that G : Y R
G(v) = g(v)dS, v Y where g admits differentiable Legendre
transform denoted by gL : RLn R.
Under these assumptions we have
inf
(u,p)U Y
{JK (u, p)}
inf
(z ,v ,
u)E
{JK
(z , v , u)},
where
JK
(z , v , u) = F (z ) + (1/2K)hv , v iL2 (S) GL (v )

n
2
X

g
(v
)
L
i u

+K

y 2
i=1
L (S)
and
E = {(z , v , u
) Y YL U | 1 z + v u0 = }.
Also, the functions z0 , v0 , and u0 , defined by

z0 =
v0 =
F (1 u0 )
,
v
G(u0 + p0 )
,
v
and
u0 = u0
are such that
1 z0 + v0 u0 = ,
252
and thus
JK (u0 , p0 )
inf
(z ,v ,
u)E
{JK
(z , v , u)} JK (u0 , p0 )
+ 2Khp0 , p0 iL2 (S) (10.44)
where we are assuming that v0 YL .
Proof. Defining = inf (u,p)U Y {JK (u, p)}, G1 (u, p) = G(u+

p)+Khp, piL2 (S) and G2 (u, p) = F (1 u)+(K/2)hp, piL2 (S) +hu, u0iU
we have:
G1 (u, p) G2 (u, p) + , (u, p) U Y.
Thus, v YL , we have
sup
(u,p)U Y
{hv , u + piL2 (S) G2 (u, p)} hv , u + piL2 (S)

G1 (u, p) + , (u, p) U Y. (10.45)
From Theorem 8.2.5:

sup
(u,p)U Y
{hv , u + piL2 (S) G2 (u, p)}

=
inf
z C (v )
{F (z ) + (1/2K)hv , v iL2 () } (10.46)
where
C (v ) = {z Y | 1 z + v u0 = }.
Furthermore
hv , u + piL2 (S) G1 (u, p)
= hv , u + piL2 (S) G(u + p) Khp, piL2 (S) .
Choosing u = u and p satisfying the equations
G(
u + p)
,
vi
from a well known Legendre Transform property, we obtain
vi =
pi =
GL (v )
i u
vi
so that
hv , u+piL2 (S) G1 (u, p) =
GL (v )K
2
X

g
(v
)
L
i u

2 .
y
i
L (S)
i=1
253
From last results and inequality (10.45) we have

inf
z C (v )
{F (z ) + (1/2K)hv , v iL2 (S) } GL (v )

2
n
X

gL (v )

+ K
i u y 2
i
L (S)
i=1

=
inf
(u,p)U Y
{JK (u, p)}
(10.47)
that is,

n
2
X

g
(v
)
1
L

i u
hv , v iL2 (S) GL (v ) + K
F (z ) +
2

2K
y
i
L (S)
i=1

=
inf
(u,p)U Y
{JK (u, p)}, if z C (v ).
(10.48)
Hence
n
F (z ) + (1/2K)hv , v iL2 (S) GL (v )
,
u)E

n
o
2
X

g
(v
)
L

+ K
iu y 2
i
L (S)
i=1
inf
(z ,v

=
inf
(u,p)U Y
{JK (u, p)}
(10.49)
so that:
inf
(z ,v ,
u)E
{JK
(z , v , u)}
inf
(u,p)U Y
{JK (u, p)}
where E = C (v ) YL U, and the remaining conclusions follow
from the expressions of JK (u0 , p0 ) and JK

(z0 , v0, u0 ).

Remark 10.5.2. We conjecture that the duality gap between
the primal and dual formulations, namely 2Khp0 , p0 iL2 (S) , goes to
zero as K +, since p0 Y satisfies the extremal condition:
1 G(u0 + p0 )
+ p0 = 0,
K
v
and JK (u, p) is bounded from below. We do not prove it in the
present work.
254
In the application of last theorem to the Kirchhoff-Love plate

model, we would have F (1 u) = , and therefore the variable z is
not present in the dual formulation. Also,

Z
Z
w
d (10.50)
P u + P w Mn
hu, u0iU =
P wdS +
n
S
t
and thus the relevant duality principle could be expressed as

K
inf G(u + p) + Khp, piL2 (S) hu, u0iU hp, piL2 (S)
uU
2

Z
Z
1
1
M M dS
inf
H N N dS
h
(v ,
u)E
2 S
2 S
Z
1
Q Q dS+
N
2 S
Z
Z
1
1
N N dS +
M M dS
2K S
2K S
2
2
X

1
1
N + v02 v02
(u, + u,) H
+K
2
2
2
L (S)
,=1
+K
2
X
=1
kw, v02 k2L2 (S) + K
2
X
,=1
M k2 2
k w, h
L (S)
where (v , u) E = C U (v , u) YL U and,

N, = 0,
Q, + M, + P = 0, a.e. in S
and
N .n P = 0,
(M t n )
(Q + M, )n +
P = 0,
s
where {v02 } is defined through the equations

Q = N v02
and,
} = {N }1 .
{N
(10.51)
10.6. THE THIRD DUALITY PRINCIPLE
255
Finally, we recall that

YL = {v Y | = N11 N22 (N12 )2 6= 0, a.e in S}.
10.6. The Third Duality Principle
Now we establish the third result, which may be summarized
by the following theorem:
Theorem 10.6.1. Let U be a reflexive Banach space, (G ) :
a convex G
U R
ateaux differentiable functional and (F 1 ) :
convex, coercive and lower semi-continuous (l.s.c.) such
U R
that the functional
J(u) = (G )(u) F (1 u) hu, piU
is bounded from below , where : U Y and 1 : U Y are

continuous linear operators.
Then we may write:
inf
sup {F (z ) G (v )} inf {J(u)}

uU
z Y v B (z )
where B (z ) = {v Y such that v 1 z p = 0}

Proof. By hypothesis there exists R ( = inf uU {J(u)})
so that J(u) , u U.
That is,
(G )(u) F (1u) + hu, piU + , u U.
The above inequality clearly implies that
sup{hu, uiU F (1 u) hu, piU } sup{hu, uiU (G )(u)} +

uU
uU
u U . Since F is convex, coercive and l.s.c., by Theorem 8.2.5

we may write
sup{hu, uiU F (1 u) hu, piU } =
uU
inf
z A (u )
{F (z )},
where,
A (u ) = {z Y | 1 z + p = u }.
Since G also satisfies the hypothesis of Theorem 8.2.5, we have
sup{hu, uiU (G )(u)} =
uU
inf
{G (v )},
v D (u )
where
D (u ) = {v Y | v = u }.
256
Therefore we may summarize the last results as

F (z ) +
sup
v D (u )
{G (v )} , z A (u ).
This inequality implies

F (z ) +
sup {G (v )} ,
v B (z )
so that we can write

inf

uU
z Y v B (z )
where B (z ) = {v Y | v 1 z p = 0}.
We will apply the last theorem to a changed functional concerning the primal formulation related to the Kirchhoff-Love plate
and (F 1 ) : U R
as
model. We redefine (G ) : U R
Z
1
(G )(u) =
H (u) (u)dS
2 S
Z
Z
1
1
h (u) (u)dS + K w, w, dS
+
2 S
2
S
if N11 (u) + K > 0, N22 (u) + K > 0 and (N11 (u) + K)(N22 (u) +
K) N12 (u)2 > 0 and, + otherwise.
is convex and
Remark 10.6.2. Notice that (G ) : U R
G
ateaux differentiable on its effective domain, which is sufficient
for our purposes, since the concerned Fenchel conjugate may be
easily expressed through the region of interest.
Also, we define
Z
1
F (1 u) = K w, w, dS
2
S
Z
Z
hu, piU =
P wdS + P u dS
S
where
u = (u , w) U = W01,2 (S) W01,2 (S) W02,2 (S).
These boundary conditions refer to a clamped plate. Furthermore,

and
1 (u) = {w,1 , w,2 }

= {1 , 2 , 3 }
10.6. THE THIRD DUALITY PRINCIPLE
257
as indicated in (10.9), (10.10) and (10.11).

and F : Y R
we would obtain
Calculating G : Y R
Z
Z
1
M M dS+
N N dS + 1 h
H
G (v ) =
2 S
2 S
1
+
2
1
N w, w, dS + K
2
S
if v E . Here
w, w, dS
(10.52)
v = {N , M , w,},
E = {v Y | N11 + K > 0, N22 + K > 0 and
2
(N11 + K)(N22 + K) N12
> 0, a.e. in S}
and,
1
F (z ) =
2K
(z1 )2 dS
1
+
2K
(z2 )2 dS.
Furthermore, v B (z ) v Y and,

N, + P = 0,
(z ), + (N w, ), + M, + Kw, + P = 0, a.e. in S.
Finally, we can express the application of last theorem as:
Z
Z
1
1
2
{
(z1 ) dS +
(z )2 dS
sup
inf
z Y v B (z ) T E 2K S
2K S 2
Z
Z
1
1
M M dS+
H N N dS
h
2 S
2 S
Z
Z
1
1
N w, w, dS K w, w, dS}
2 S
2
S
inf {J(u)}.
uU
(10.53)
The above inequality can in fact represents an equality if the

positive real constant K is chosen so that the point of local ex0)
tremum v0 = G(u
E (which means N11 (u0 )+K > 0, N22 (u0 )+
v
K > 0, and (N11 (u0 ) + K)(N22 (u0 ) + K) N12 (u0 )2 > 0). The mentioned equality is a result of a little change concerning Theorem
8.1.27.
258
Remark 10.6.3. For the determination of G (v ) in (10.52)

we have used the transformation
Q = N w, + Kw, ,
similarly as indicated in Remark 10.3.1.
10.7. A Convex Dual Formulation
Remark 10.7.1. In this section we assume
40 0
H = h{
+ 20 ( + )},
0 + 20
and
h2 H
,
h =
12
where denotes the Kronecker delta and 0 , 0 are appropriate
constants.
The next result may be summarized by the following Theorem:
Theorem 10.7.2. Consider the functionals (G ) : U R,

and hu, piU defined as
(F 1 ) : U R
Z
1
(G )(u) =
H (u) (u)dS
2 S
Z
Z
1
1
h (u) (u)dS + K w, w, dS,
+
2 S
2
S
Z
1
F (1 u) = K w, w, dS,
2
S
and
Z
Z
hu, piU =
P wdS + P u dS,
S
where
u = (u , w) U = W01,2 (S) W01,2 (S) W02,2 (S).
The operators { } and { } are defined in (10.7) and (10.8),

respectively. Furthermore, we define J(u) = (G )(u) F (1 u)
hu, piU , and
1 (u) = {w,1 , w,2 }.
Suppose there exists u0 U such that J(u0 ) = 0, and that
there exists K > 0 for which N11 (u0 ) + K > 0, N22 (u0 ) + K >
0 , (N11 (u0 ) + K)(N22 (u0 ) + K) N12 (u0 )2 > 0 (a.e in S) and
10.7. A CONVEX DUAL FORMULATION
259
h1212 /(2K0 ) > K where K0 is the constant related to Poincare Inequality and,
N (u0 ) = H (u0 ).
Then,
J(u0 ) = min{J(u)} =
uU
max
{G (v ) + hz , z iL2 (S) /(2K)}
(v ,z )E B
(10.54)
G (v0 ) + hz0 , z0 iL2 (S) /(2K)
where,
G(u0 )
and z0 = Kw0, ,
v
Z
1

N N dS
H
G (v ) = GL (v ) =
2 S
Z
Z
1
1
K Q, Q, dS
N
h M M dS +
+
2 S
2 S
v0 =
if v E , where v = {{N }, {M }, {Q }} E v
L2 (S, R10 ) and
2
N11 +K > 0 N22 +K > 0 and (N11 +K)(N22 +K)N12
> 0, a.e. in S
and,
(v , z ) B
N, + P = 0,
+ P = 0,
Q, + M, z,
1212 M12 + z /K = 0,
1,2
z = z , a.e. in S, and, z n = 0 on ,
1,2
2,1
K } as indicated in (10.6).
being {N
Proof. Similarly to Proposition 10.4.1, we may obtain the following result. If v E then
GL (v ) = G (v ) hv , uiY G(u), u U,
so that
GL (v )
1
1
hz , z iL2 (S) hv , uiY
hz , z iL2 (S) G(u),
2K
2K
u U,
260
and thus, as v 1 z p = 0 (see the definition of B ) we obtain
Q, + M, z,
+ P = 0 a.e. in S.
Through this equation we may symbolically write
M12 = 1
12 {(Q, + z, M, P )/2},
3
(10.55)
, denotes M11,11 + M22,22 , in S, so that substituting

where M
such a relation in the last inequality we have
Z
Z
Z
1
1
2
1122 M11 M22 dS
H N N dS +
h1111 M11 dS + h
2 S
2 S
S
Z
1
2222 M 2 dS
+
h
22
2 S
Z
2
1212 (1
+ 2 h
312 (v , z )) dS
ZS
1
K Q, Q, dS
N
+
2 S
1
hz , z iL2 (S)
2K
1
hz , z iL2 (S)
2K
G(u) + hu, piU , u U,
(10.56)
h1 u, z iL2 (S;R2 )
where M12 is made explicit through equation (10.55). This equation

makes z an independent variable, so that evaluating the supremum
concerning z , particularly for the left side of above inequality, the
global extremum is achieved through the equation :

1

([1
a.e. in S.
12 ] [h1212 12 (v , z )]), z /K = 0,
3
This means

1212 1
h
312 (v , z ) z, /K = 0, a.e. in S and z n = 0 on
or
1212 M12 + z /K = 0, a.e. in S and z n = 0 on

h
,
for (, ) = (1, 2) and (2, 1). Therefore, after evaluating the suprema
in both sides of (10.56), we may write
GL (v )
1
hz , z iL2 (S) F (1u) G(u) + hu, piU ,
2K
u U, and (v , z ) B E .
10.8. A FINAL RESULT, OTHER SUFFICIENT...
261
and it seems to be clear that the condition h1212 /(2K0 ) > K guarantees coercivity for the expression of left side in the last inequality
(see the next remark), so that the unique local extremum concerning z is also a global extremum. The equality and remaining conclusions results from the Gateaux differentiability of primal and
dual formulations and an application (with little changes) of Theorem 8.1.27.

Remark 10.7.3.
expressed as
Observe that the dual functional could be
1
hz , z iL2 (S)
GL (v )
2K
Z
Z
Z
1
1
2
1122 M11 M22 dS
=
H N N dS +
h1111 M11 dS + h
2 S
2 S
S
Z
Z
Z
1
1
K
2
2222 M dS +
Q, Q, dS + h1212 (z )2 /K 2 dS
N
h
+
1,2
22
2 S
2 S
S
Z
1
2
+ h1212 (z2,1
) /K 2 dS
hz , z iL2 (S) .
(10.57)
2K
S
Thus, through the relation h1212 /(2K0 ) > K (where K0 is the constant related to a Poincare type inequality, now specified through
the last format of the dual formulation), it is now clear that the
dual formulation is convex on E B .
10.8. A Final Result, other Sufficient
Conditions of Optimality
This final result is developed similarly to the triality criterion
introduced in Gao [24], which describes, in some situations, sufficient conditions for optimality.
We prove the following result.
Theorem 10.8.1. Consider J : U R where J(u) = G(u)+
F (u),
Z
Z
1
1
G(u) =
H (u) (u)dS +
h w, w, dS.
2 S
2 S
Here the operators are defined as in (10.7),
Z
F (u) = P wdS hu, f iU ,
S
262
and,
U = W01,2 (S) W01,2 (S) W02,2 (S).
Then, if u0 U is such that J(u0 ) = and
1
2
1
N (u0 )w,w, dS +
2
S
h w, w, dS 0,
w W02,2 (S), (10.58)
we have
J(u0 ) = min{J(u)}.
uU
Proof. It is clear that

G(u) + F (u) (G ) (u ) F (u ), u U, u U ,
so that
G(u) + F (u) (G ) ( v ) F ( v ),
u U, v Y . (10.59)
Consider u0 for which J(u0 ) = and such that (10.58) is satisfied.

Defining
G(u0 )
,
v0 =
v
from Theorem 8.1.27 we have that
(GL (v0 ) + hu0 , v0 f iU ) = ,
J(u0 ) = GL (v0 ),
and
v0 = f.
This means
F ( v0 ) = 0.
On the other hand
(G ) ( v0 ) = sup{hu, v0iY G(u)},
uU
10.8. A FINAL RESULT, OTHER SUFFICIENT...
263
or
(G )
(v0 )
= sup
uU
(
u, + u,
, N (u0 )
2
L2 (S)
+hw, , M (u0 )iL2 (S)

Z
1
+hw,Q (u0 )iL2 (S)
H (u)(u)dS
2 S
Z
1
h w, w, dS }.
(10.60)
2 S
Since
u, + u, 1
+ w, w, ,
2
2
from the last equality, we may write
n
w, w,

, N (u0 )iL2 (S)
(G) ( v0 ) = sup h (u), N (u0 )iL2 (S) h
2
uU
+ hw, , M (u0 )iL2 (S) + hw,, Q (u0 )iL2 (S)

Z
Z
1
1
H (u) (u)dS
h w, w, dS . (10.61)
2 S
2 S
(u) =
As (Q (u0 )), + (M (u0)), + P = 0, we obtain
(G ) ( v0 )
D
Z
E
w, w,
1
sup
h w, w, dS
, N (u0 )
2
2 S
L2 (S)
uU

Z
Z
1
N (u0 )N (u0)dS. (10.62)
H
+ P wdS +
2
S
S
Therefore, from hypothesis (10.58) the extremum indicated in (10.62)

is attained for functions satisfying
(N (u0 )w
), (h w ), + P = 0.
(10.63)
From J(u0 ) = and boundary conditions we obtain

w = w0 , a.e. in S,
so that
(G)
( v0 )
Dw
+
0, w0,
1
2
2
Z
1
, N (u0 )
+
2
L (S) 2
E
h w0, w0, dS
S
N (u0 )N (u0 )dS.

H
(10.64)
264
However, since
Q (u0 ) = N (u0 )w0, ,
and
M (u0 ) = h w0,
from (10.64) we obtain
Z
1

(u0 )Q (u0 )Q (u0 )dS
N
(G ) ( v0 )
2
Z
1
M (u0 )M (u0 )dS
h
+
2 S
Z
1
N (u0 )N (u0)dS. (10.65)
H
+
2 S
Hence
(G ) ( v0 ) GL (v0 ) = J(u0 ),
and thus as F ( v0 ) = 0, we have that
J(u0 ) (G ) ( v0 ) F ( v0 ),
which, from (10.59) completes the proof.
10.9. Final Remarks

In this chapter we presented four different dual variational formulations for the Kirchhoff-Love plate model. Earlier results (see
references [40],[22]) present a constraint concerning the gap functional to establish the complementary energy (dual formulation).
In the present work the dual formulations are established on the
hypothesis of existence of a global extremum for the primal functional and the results are applicable even for compressed plates. In
particular the second duality principle is obtained through an extension of a theorem met in [42], and in this case we are concerned
with the solution behavior as K +, even though a rigorous
and complete analysis of such behavior has been postponed for a
future work. However, what seems to be interesting is that the dual
formulation as indicated in (10.51) is represented by a natural extension of the results found in [42] (particularly Theorem 10.1.1),
plus a kind of penalization concerning the inversion of constitutive
equations.
It is worth noting that the third dual formulation was based on
the same theorem, despite the fact such a result had not been directly used, we followed a similar idea to prove the duality principle.
For this last result, the membrane forces are allowed to be negative
10.9. FINAL REMARKS
265
since it is observed the restriction N11 + K > 0, N22 + K > 0 and

2
(N11 + K)(N22 + K) N12
> 0, a.e. in S, where K R is a
positive suitable constant.
In section 10.7, we obtained a convex dual variational formulation for the plate model, which allows non positive definite membrane force matrices. In this formulation, the Poincare inequality
plays a fundamental role.
Finally, in the last section, we developed a result similar to
Gaos triality criterion presented in [24]. In the plate application
this gives sufficient conditions for optimality. We present a new
proof of sufficient conditions of existence of a global extremum for
the primal problem. As earlier mentioned, such conditions may be
summarized by J(u0 ) = and
Z
Z
1
1
N (u0 )w,w, dS+
h w, w, dS 0, w W02,2 (S).
2 S
2 S
CHAPTER 11
Duality Applied to Elasticity

11.1. Introduction and Primal Formulation
Our first objective in the present chapter is to establish a dual
variational formulation for a finite elasticity model. Even though
existence of solutions for this model has been proven in Ciarlet
[13], the concept of complementary energy, as a global optimization
approach, is possible to be defined only if the stress tensor is positive
definite at a critical point. Thus we have the goal of relaxing such
constraints and start by describing the primal formulation.
Consider S R3 an open, bounded, connected set, which represents the reference volume occupied by an elastic solid under
the load f L2 (S; R3 ). We denote by the boundary of S.
The field of displacements under the action of f is denoted by
u (u1 , u2 , u3 ) U, where u1 , u2 , and u3 denotes the displacements
related to directions x, y, and z respectively, on the cartesian basis
(x, y, z). Here U is defined as
U = {u = (u1 , u2 , u3 ) W 1,4 (S; R3 ) |
u = (0, 0, 0) on 0 } (11.1)
and = 0 1 , 0 1 = . Denoting the stress tensor by {ij },

where

1
ij = Hijkl
(uk,l + ul,k + um,k um,l )
(11.2)
2
and {Hijkl} is a positive definite matrix related to the coefficients of
Hookes Law, the boundary value form of the finite elasticity model
is given by
ij,j + (im um,j ),j + fi = 0, in S,

(11.3)
u = , on 0 .
267
268
11. DUALITY APPLIED TO ELASTICITY
The corresponding primal variational formulation is represented by

J : U R, where
1
J(u) =
2

1
Hijkl
(ui,j + uj,i + um,i um,j )
2
S

1
(uk,l + ul,k + um,k um,l ) dx hu, f iL2(S;R3 ) (11.4)
2
Remark 11.1.1. Derivatives must be always understood in

distributional sense, whereas boundary conditions are in the sense
of traces. Finally, we denote k k2 as the standard norm for L2 (S).
The next result has been proven in Chapter 10.
Theorem 11.1.2. Consider Gateaux differentiable functionals
and F 1 : U R
where only the second
G : U R
one is necessarily convex, through which is defined the functional
expressed as:
JK : U Y R
JK (u, p) = G(u + p) + Khp, piL2 (S,Rn ) F (1 u)
K
hp, piL2 (S,Rn ) hu, u0iU (11.5)
2
so that it is supposed the existence of (u0 , p0 ) U Y such that
JK (u0 , p0 ) =
inf
(u,p)U Y
{JK (u, p)}
(11.6)
and JK (u0 , p0 ) = (here = {i } : U Y and 1 : U Y are

continuous linear operators whose adjoint operators are denoted by
: Y U and 1 : Y U respectively).
Furthermore it is assumed the existence of a differentiable func may be expressed
tion denotedR by g : Rn R so that G : Y R
as: G(v) = g(v)dS, v Y where g admits differentiable Legendre Transform denoted by gL : RLn R.
Under these assumptions we have:
inf
(u,p)U Y
{JK (u, p)}
inf
(z ,v ,
u)E
(z , v , u)}
{JK
(11.7)
11.2. THE FIRST DUALITY PRINCIPLE
269
where
JK
(z , v , u) = F (z ) + 1/(2K)hv , v iL2 (S,Rn ) GL (v )
2
n
X

gL (v )

+K
i u y
i=1
and
E =
{(z , v , u
) Y YL U| 1 z + v u0 = }. (11.8)
Also, the functions z0 , v0 , and u0 , defined by
F (1 u0 )
,
v
(11.9)
G(u0 + p0 )
v
(11.10)
u0 = u0
(11.11)
1 z0 + v0 u0 =
(11.12)
z0 =
v0 =
and
are such that
and thus
JK (u0 , p0 )
inf
(z ,v ,
u)E
{JK
(z , v , u)}
JK (u0 , p0 ) + 2Khp0 , p0 iL2 (S,Rn ) (11.13)

where we are assuming that v0 YL .
11.2. The First Duality Principle
In this section we establish the first duality principle. We start
with the following theorem.
Theorem 11.2.1. Define J : U R as
J(u) = G (u) F1 (u)
(11.14)
270
where

1
Hijkl
(ui,j + ui,j + um,i um,j )
2
S

1
(uk,l + uk,l + um,k um,l ) dx
2
K
+ hum,i , um,iiL2 (S) ,
2
1
G(u) =
2

F1 (u) = F (u) hu, f iL2(S;R3 ) ,
(11.15)
(11.16)
and
F (u) =
K
hum,i, um,i iL2 (S) .
2
(11.17)
Here : U Y = Y1 Y2 L2 (S; R9 ) L2 (S, R9 ) is given by

1
u = {1 u, 2u} {{ (ui,j + ui,j )}, {um,i}}.
2
Thus we can write
inf {J(u)} inf sup {F (z ) G (v )},
uU
z Y2 v E
(11.18)
(11.19)
where
v {, Q},
E = E1 E2 ,
(11.20)
E1 = {(v , z ) Y | ij,j + Qij,j zij,j

+ fi = 0, a.e. in S},
(11.21)
E2 = {(v , z ) Y | ij nj + Qij nj = zij nj , on 1 },

and Y = Y Y1 . Also,
F (z ) =
1
hz , z iL2 (S)
2K im im
(11.22)
(11.23)
and
G (v ) =
GL (v )
1
=
2
ijkl ij kl dx
H
S
1
+
2
ijK Qmi Qmj dx (11.24)
if K is positive definite in
11 + K
21
K =
31
271
S, where
12
13
22 + K 23
32
33 + K
(11.25)
ijkl } = {Hijkl}1 .
and also {
ijK } = K 1 . Finally, {H
Remark 11.2.2. The proof of a similar result was given in

F.Botelho [6]. However in the present case some details are slightly
different so that we provide a complete proof.
Proof. Defining inf uU {J(u)} we have
or
so that
G (u) F1 (u) , u U,
(11.26)
F1 (u) G (u) + , u U,
(11.27)
sup{h1 u, w iY1 F1 (u)} sup{h1u, w iY1 G(u)} + .

uU
uU
(11.28)
However, from Theorem 8.2.5
F1 (u ) = sup{h1u, w iY1 F1 (u)} =
uU
inf
z C1 C2
{F (z )} (11.29)
where
C1 = {z Y2 | zij,j
wij,j
fi = 0, a.e. in S},
(11.30)
C2 = {z Y2 | zij nj = wij nj , on 1 }.
(11.31)
and
On the other hand, also from Theorem 8.2.5

(G ) (u ) = sup{h1u, w iY1 G (u)} =
uU
inf
v D1 D2
{G (v )}
(11.32)
where
D1 = {v Y | ij,j + Qij,j wij,j

= 0, a.e. in S}, (11.33)
and
D2 = {v Y | ij nj + Qij nj = wij nj , on 1 }.
(11.34)
272
We can summarize the last results by

inf {J(u)} = F1 (u ) (G ) (u ) F (z )
uU
sup
v D1 D2
{G (v )}, (11.35)
z C1 C2 .
Hence we can write
inf {J(u)} inf sup {F (z ) G (v )}
uU
z Y1 v E
(11.36)
where E = E1 E2 ,
E1 = {(v , z ) Y | ij,j + Qij,j zij,j

+ fi = 0, a.e. in S},
(11.37)
E2 = {(v , z ) Y | ij nj + Qij nj = zij nj , on 1 },
(11.38)
and Y = Y Y1 .
Finally, we have to prove that GL (v ) = G (v ) if K is positive
definite in S. We start by formally calculating gL (y ), the Legendre
transform of g(y), where

1
1
y1kl + y2mk y2ml
g(y) = Hijkl y1ij + y2mi y2mj
2
2
K
+ y2mi y2mi . (11.39)
2
We recall that
gL (y ) = hy, y iR18 g(y)
where y R18 is solution of equation

g(y)
y =
.
y
Thus

1
y1ij = ij = Hijkl y1kl + y2mk y2ml

2
and

1
y2mi = Qmi = Hijkl y1kl + y2ok y2ol y2mj + Ky2mi

2
so that
Qmi = ij y2mj + Ky2mi .
(11.40)
(11.41)
(11.42)
(11.43)
(11.44)
273
Inverting these last equations, we have

y2mi =
ijK Qmj
(11.45)
where
1
{
ijK } = K
and also
13
11 + K 12
21
22 + K 23
=

32
33 + K
31
ijkl kl 1 y2mi y2mj .
y1ij = H
2
(11.46)
(11.47)
Finally
1 K
1
Qmi Qmj .
(11.48)
gL (, Q) = H
ijkl ij kl +
2
2 ij
Now we will prove that gL (v ) = g (v ) if K is positive definite.
First observe that
g (v ) =
sup {hy1 , iR9 + hy2 , QiR9
yR18

1
1
1
Hijkl y1ij + y2mi y2mj
y1kl + y2mk y2ml
2
2
2

K
y2mi y2mi
2

1
h
y1ij y2mi y2mj , ij iR + hy2 , QiR9
=
sup
2
(
y1 ,y2 )R9 R9

K
1
y1kl ] y2mi y2mi .
y1ij ][
Hijkl [
2
2
The result follows just observing that

1
1
y1ij , ij iR Hijkl[
sup h
y1ij ][
y1kl ] = H
ijkl ij kl
2
2
y1 R9
(11.49)
and

1
K
sup h y2mi y2mj , ij iR + hy2, QiR9 y2mi y2mi
2
2
y2 R9
1 K
Qmi Qmj (11.50)
=
2 ij
if K is positive definite.
274

The second duality principle is summarized by the following
theorem, which is similar to Theorem 11.1.2. However some details
are different, so that again we provide a complete proof. Note that
this result is particularly applicable as the optimal matrix {ij } is
negative definite.
Theorem 11.3.1. Define J : U Y R as
K
hp, piY
2
hu, f iL2(S;R3 ) . (11.51)
JK (u, p) = G(u + p) + Khp, piY F (1u)
Suppose K is big enough so that JK : U Y R is bounded below,

where

Z
1+
1
G(u) =
Hijkl
2
2
S

1
(uk,l + uk,l + um,k um,l ) dx (11.52)
2
and

Z
1
F (1 u) =
Hijkl
2 S
2

1
(uk,l + uk,l + um,k um,l ) dx. (11.53)
2
Here : U Y is expressed as:

1
umi umj
u =
(ui,j + ui,j ) +
2
2
and 1 : U Y = Y1 Y2 L2 (S; R9 ) L2 (S, R9 ) is given by

1
(ui,j + ui,j ) , {um,i} .
(11.54)
1 u = {11 u, 12 u}
2
Thus we can write
n
inf {JK (u)}
inf
F (z , Q, ) GL (z )
uU
(z ,(,Q),u)Y A U

)
3
X
ijkl 2
ui,j + uj,i um,i um,j
1
H

+
hz , z iY +
K
+
zkl

,
2K
2
2
1
+
2
i,j=1
(11.55)
275
where,
A = {(, Q) Y | ij,j Qij,j + fi = 0,
in S, (ij Qij )nj = 0, on 1 }, (11.56)
1
F (z , , Q) =
2
ijkl (z ij )(z kl ) dx
H
ij
kl
S
Z
1
ij Qmi Qmj dx (11.57)
2 S
if {ij } is negative definite in S, and also {

ij }={ij }1 .
Finally,
Z
1
ijkl zij zkl

GL (z ) =
H
dx.
2(1 + ) S
(11.58)
Proof. Defining = inf (u,p)U Y {JK (u, p)},
G2 (u, p) = G(u + p) + Khp, piY
and
G1 (u, p) = F (1u) +
K
hp, piY + hu, f iL2(S;R3 )
2
we have,
That is
J(u, p) = G2 (u, p) G1 (u, p) , u U, p Y.

G1 (u, p) G2 (u, p) + , u U, p Y,
so that
sup
(u,p)U Y
{hu + p, z iY G1 (u, p)} hu + p, z iY G2 (u, p) + .

(11.59)
However, from the standard results of variational convex analysis

sup
(u,p)U Y
{hu + p, z iY G1 (u, p)}

=
inf
(,Q))A
{F (z , , Q)} +
1
hz , z iY .
2K
On the other hand, particularly for p satisfying the relation,

z =
G(u + p)
,
z
276
from the well known proprieties of Legendre transform, we have:

GL (z )
u.
p=
z
Therefore
hu + p, z iY G2 (u, p) = GL (z )

3
X
ijkl 2
H

K
+
zkl .
2
2
1
+
2
i,j=1
Replacing the last results into (11.59) we obtain,

n
inf {JK (u)} inf F (z , Q, ) GL (z )
uU
(z ,(,Q))Y A
1
hz , z iY
(11.60)
2K
)

3
X
ijkl 2
H

+
zkl
+
K
,

2
2
1
+
2
i,j=1
Remark 11.3.2. From the expression of the dual formulation

we conjecture that the duality gap between the primal and dual
problems goes to zero as K , if we have a critical point u0 U
for which the corresponding matrix {ij (u0 )} is negative definite
a.e. in S and
Z
Z
1
1
ijkl vij vkl dx 0,
ij (u0 )vmi vmj dx

H
2 S
2 S
v L2 (S; R9 ).
A proof and/or full justification of such a conjecture is planned for

a future work.
11.4. Conclusion
In this chapter we have presented three different dual variational
formulations for a non-linear model in Elasticity. The second principle is particularly useful when there is a critical point for which
the matrix of stresses is negative definite.
CHAPTER 12
Duality Applied to a Membrane Shell Model

In this chapter we establish dual variational formulations for the
elastic membrane shell model presented in [15] (P.Ciarlet). Ciarlet
proves existence of solutions, however, the complementary energy
as developed in [40]( J.J.Telega), is possible only for a special class
of external loads, that generate a critical point with positive definite
membrane tensor. In this chapter we have the goal of relaxing such
constraints, and start by describing the primal formulation.
Consider a domain S R2 and a injective mapping ~ : S
denotes the middle surface of a
[, ] R3 such that S0 = ~(S)
shell of thickness 2.
The mapping ~ may be expressed as:
~(x) = ~1 (y) + x3 a 3 (y)
where a 3 = (a 1 a 2 )/ka 1 a 2 k and a = ~, y = (y1 , y2 ) denote
the curvilinear coordinates, x = (y, x3 ) and x3 .
The contravariant basis of the tangent plane to S in y, denoted
by {a } is defined through the relations
a a = , and a 3 = a 3
The covariant components of the metric tensor, denoted by

{a }, are defined as
a = a a
and the concerned contravariant components are denoted by {a }

and expressed as
a = a a
It is not difficult to show that
a = a a , {a } = {a }1
p
and a = a a . We also define a(y) = ka1 (y) a 2 (y)k.
The curvature tensor denoted by {b } is expressed as
b = a 3 a .
277
278
12. DUALITY APPLIED TO A MEMBRANE SHELL MODEL
Concerning the displacements due to external loads action, we

denote them by = i a i , and the admissible displacements field is
denoted by U, where
U = { W 1,4 (S; R3 ) | = (0, 0, 0) on 0 }
and here = 0 1 (0 1 = ) denotes the boundary of S.

We now state the Theorem 9-1-1 of reference [15] (Mathematical
Elasticity , Vol. III -Theory of shells), by P.Ciarlet.
R3 ) be
Theorem 12.1.1. Let S R2 a domain and ~ C 2 (S;
an injective mapping such that the two vectors a = ~ are linearly
let a3 = (a1 a2 )/ka1 a2 k, and
independent in all points of S,
i
let the vectors a be defined by ai .aj = ji . Given a displacement
~ S),
let the covariant components of
field i ai of the surface S0 = (
the change of metric tensor associated with this displacement field
be defined by
G () = (a () a ),
where a and a () denote the covariant components of the metric
and (~ + i ai )(S)
respectively. Then
tensors of the surfaces ~(S)
G () = (k + k + amn mk nk )/2
where k = b 3 and 3k = 3 + b .
We define the constitutive relations as
G ()
N = a
(12.1)
where {N } denotes the membrane forces, and
4
a a + 2(a a + a a ))
a
= (
4 + 2
(12.2)
The potential energy (stored energy plus external work) is expressed by the functional J : U R where,
Z
Z
J() =
a G ()G () a dy pi i a dy
2 S
S
R
i
i
i
2
where p = f dx3 , here {f } L (S [, ]; R3 ) denotes the
external load density.
We will define (G ) : U R and F : U R as
Z
1
a
G ()G () a dy
(12.3)
(G )() =
2 S
279
and,
F () =
pi i a dy,
(12.4)
where {} = {1 , 2m }, 1 () = (k +k )/2 and 2m () =

mk . Thus, the primal variational formulation is given by J : U
R, where
J() = G() F ().
(12.5)

We will be concerned with the Legendre Transform related to
the function g : R10 R expressed as
1
1
1
(y1 + amn y2m y2n )(y1 + akl y2k y2l ) (12.6)
g(y) = a
2
2
2
Its Legendre Transform, denoted by gL : RL10 R is given by
(12.7)
where y R10 is solution of the system

y =
g(y)
.
y
(12.8)
That is,
1
y1 = a
(y1 + akl y2k y2l )
(12.9)
1
(y1 + akl y2k y2l )amn y2n
y2m = a
(12.10)
and
or
y2m = y1 amn y2n .
(12.11)
Therefore, after simple algebraic manipulations we would obtain

1
1
gL (y ) = a
y1 y1 + Rmn y2m y2n ,
2
2
(12.12)
where
Rmn = {y1 amn }1 .
(12.13)
280
Thus, denoting v = {N , Qm } we have

Z
Z
a
N N a dy
GL (v ) =
gL (v ) a dy =
S
S
Z
1
+
Rmn Qm Qn a dy. (12.14)
2 w
We now obtain the polar functional related to the external load.
12.3. The Polar Functional Related to F : U R
The polar functional related to F : U R, for v = {N , Qm }

is expressed by
F ( v ) = sup{h, v iU F (u)}.
(12.15)
That is,
F ( v ) = sup{h1 , N aiL2 (S) + h2m , Qm aiL2 (S)

U
Z
pi i a dy}. (12.16)
S
Thus
F ( v ) =
where
v C
0, if v C ,
+, otherwise,
(N + Q )| + b Q3 = p in S,
b (N + Q ) Q3 | = p3 in S,
(12.17)
(12.18)
(N + Q ) = 0 on 1 ,
Q3 | = 0 on .

1
12.4. The Final Format of First Duality Principle

The duality principle presented in Theorem 10.5.1 is applied to
the present case. Thus we obtain
inf {J()} inf {
v C
1
hv , v iL2 (S,R10 ) GL (v )
2K

2

g
(v
)
u L
+K

y 2
L (S)
281
More explicitly

Z
Z
i
a
G ()G () a dy p i a dy
inf
U
2 S
S

Z
1

m m
inf
a dy + Q Q
a dy
N N
v C
2K
S
S
Z
Z
1
1
a dy
a
N N
Rmn Qm Qn a dy
2 S
2 S

2
X
,=1
Z

K k a
N akl v02k v02l L2 (S)
+
3 X
2
X
m=1 =1
Kkmk v02m kL2 (S)

Our objective now is to establish a different dual formulation.
First we will write the primal formulation as the difference of two
convex functionals and then will obtain the second duality principle,
as indicated in the next theorem.
Theorem 12.5.1. Define J : U R as
J() = G()
F (2)
(12.19)
where
G()
= G () h, pi,
1
G() =
2
(12.20)
a
G ()G () a dy
S
Z
K
mk mk a dy, (12.21)
+
2 S
Z
K
mk mk a dy,
F (2) =
(12.22)
2 S
Z
h, pi =
pi i a dy,
(12.23)
282
and : U Y1 Y2 L2 (S; R4 ) L2 (S; R6) is defined as

1
(k + k ) , {mk } .
(12.24)
= {1, 2 }
2
Thus, we may write
inf {J()} inf
sup {F (u ) G (v )}
u U v C (u )
(12.25)
where
v {N, Q},
Z
1

F (u ) =
umi umi a dy,
2K S
G (v ) =
GL (v )
(12.26)
(12.27)
1
a
N N ady
ady =
2 S
Z
1
+
Rmn Qm Qn a dy (12.28)
2 S
gL (v )
if NK = {N amn +K mn } is positive definite in S. Furthermore

and
Finally,
{Rmn } = NK1
(12.29)
{
a } = {a
}1 .
(12.30)
v {N, Q} C (u )
(N + Q u )| + b (Q3 u3 ) = p in S,
b (N + Q u ) (Q3 u3 )| = p3 in S,
(12.31)
(N + (Q u )) = 0 on 1 ,
(Q3 u )| = 0 on .
1
3
Proof. Defining inf U {J()} we have
or
G()
F (2) , U,
(12.32)
F (2 ) G()
+ , U,
(12.33)
283
so that
sup{h2 , ui F ()} sup{h2, u i G()}

+ .
U
(12.34)
However,
F (u ) sup{h2 , ui F (2 )}.
(12.35)
On the other hand, from Theorem 8.2.5

) (2 u ) = sup{h2 , ui G()}
(G
=
U
inf
{G (v )}
v C (u )
(12.36)
where
v {N, Q} C (u )
(N + Q u )| + b (Q3 u3 ) = p in S,
b (N + Q u ) (Q3 u3 )| = p3 in S,
(12.37)
(N + (Q u )) = 0 on 1 ,
(Q3 u )| = 0 on .
1
3
We can summarize the last results by
) ( u ) = F (u )
inf {J()} = F (u ) (G
2
sup {G (v )}, (12.38)
v C (u )
u U .
Finally, we have to prove that GL (v ) = G (v ) if NK is positive
definite in S. We start by formally calculating gL (y ), the Legendre
transform of g(y), where g : R10 R is expressed as:
1
1
1
g(y) = a
(y1 + amn y2m y2n )(y1 + akl y2k y2l )
2
2
2
K
+ y2m y2m . (12.39)
2
Observe that gL : RL10 R is given by
(12.40)
284
where y R10 is solution of the system

y =
g(y)
.
y
(12.41)
That is,
1
y1 = a
(y1 + akl y2k y2l )
(12.42)
and,
1
(y1 + akl y2k y2l )amn y2n + Ky2m
y2m = a
(12.43)
or
y2m = y1 amn y2n + Ky2m .
(12.44)
Therefore, after simple algebraic manipulations we would obtain

1
1
gL (y ) = a
y1 y1 + Rmn y2m y2n
(12.45)
2
2
where
Rmn = {y1 amn + K mn }1 .
(12.46)
Thus, through the notation v = {N , Qm } we would obtain

Z
Z
a
N N a dy
GL (v ) =
gL (v ) a dy =
2 S
w
Z
1
Rmn Qm Qn a dy (12.47)
+
2 S
where
Rmn = {N amn + K mn }1 .
(12.48)
Now we will prove that gL (v ) = g (v ) if NK is positive definite.

First observe that
g (v ) = sup {hy1 , NiR4 + hy2, QiR6
yR10
1
1
K
1
a
[y1 + amn y2m y2n ][y1 + akl y2k y2l ] y2m y2m }
2
2
2
2
1 mn
{h
y1 a y2m y2m , N iR + hy2 , QiR6
=
sup
2
(
y1 ,y2 )R4 R6
1
K
a
[
y1 ][
y1 ] y2m y2m }. (12.49)
2
2
12.6. CONCLUSION
285
The result follows just observing that

1
1
[
y1 ][
y1 ] = a
N N (12.50)
y1 , N iR a
sup {h
2
2
4
y1 R
and
1
K
sup {h amn y2m y2m , N iR + hy2 , QiR6 y2m y2m }
2
2
y2 R6
1
= Rmn Qm Qn (12.51)
2
if NK is positive definite.

12.6. Conclusion
We obtained three different dual variational formulations for the
non-linear elastic shell model studied in P.Ciarlet [15] (a membrane
shell model).The first duality principle presented is an extension of
a theorem found in Toland [42]. The solution behavior as K +
is of particular interest for a future work.
The second duality principle relaxes the condition of positive
definite membrane tensor, and thus the constant K must be chosen
so that the matrix NK is positive definite at the equilibrium point,
where NK = {N amn + K mn }.
CHAPTER 13
Duality Applied to Ginzburg-Landau Type

Equations
13.1. Introduction
In this article, our first objectives are to show existence and
develop dual formulations concerning the real semi-linear GinzburgLandau equation. We start by describing the primal formulation.
By S R3 we denote an open connected bounded set with
a sufficiently regular boundary = S (regular enough so that
the Sobolev Imbedding Theorem holds). The Ginzburg-Landau
equation is given by:

2
2 u + ( u2 )u f = 0 in S,
(13.1)
u = 0 on ,
where u : S R denotes the primal field and f L2 (S). Moreover,
, are real positive constants.
Remark 13.1.1. The complex Ginzburg-Landau equation

plays a fundamental role in the theory of superconductivity (see
[3], for details). In the present work we deal with the simpler real
form however, the results obtained may be easily extended to the
complex case.
The corresponding variational formulation is given by the functional J : U R where:
Z
Z
Z
1
u2
2
2
J(u) =
|u| dx +
( ) dx f u dx
(13.2)
2 S
2 S 2
S
where U = {u W 1,2 (S) | u = 0 on } = W01,2 (S).

We are particularly concerned with the fact that equations indicated in (13.1) are necessary conditions for the solution of Problem P,
where:
Problem P : to find u0 U such that J(u0 ) = min{J(u)}.
uU
287
288 13. DUALITY APPLIED TO GINZBURG-LANDAU TYPE EQUATIONS
13.1.1. Existence of Solution for the Ginzburg-Landau

Equation. We start with a remark.
Remark 13.1.2. From the Sobolev Imbedding Theorem for
mp < n, n mp < n, p q p = np/(n mp),
we have:
W j+m,p() W j,q ().

Therefore, considering n = 3, m = 1, j = 0, p = 2, and q = 4, we
obtain:
W 1,2 () L4 () L2 ()
and thus
kukL4 () + kukW 1,2 () +.
Furthermore, from above and Poincare Inequality it is clear that
for J given by (13.2), we have:
J(u) + as kukW 1,2(S) +,
that is, J is coercive.

Now we establish the existence of a minimizer for J : U R.
It is a well known procedure (the direct method of Calculus of
Variations). We present it here for the sake of completeness.
Theorem 13.1.3. For , R+ , f L2 (S) there exists at
least one u0 U such that:
J(u0 ) = min{J(u)}
uU
where
1
J(u) =
2
|u| dx +
2
S
u2
( )2 dx
S 2
and U = {u W 1,2 (S) | u = 0 on } = W01,2 (S).
f u dx
S
Proof. From Remark 13.1.2 we have:

J(u) + as kukU +.
Also from Poincare inequality, there exists 1 R such that

1 = inf uU {J(u)}, so that, for {un } minimizing sequence, in the
sense that:
J(un ) 1 as n +
(13.3)
13.2. A CONCAVE DUAL VARIATIONAL FORMULATION
289
we have that kun kU is bounded, and thus, as W01,2 (S) is reflexive,

there exists u0 W01,2 (S) and a subsequence {unj } {un } such
that:
unj u0 , weakly in W01,2 (S).
(13.4)
From (13.4), by the Rellich-Kondrachov theorem, up to a subsequence, which is also denoted by {unj }, we have:
unj u0 , strongly in L2 (S).
(13.5)
lim inf {J1 (unj )} J1 (u0 ).
(13.6)
Furthermore, defining J1 : U R as:

Z
Z
Z
1
2
4
|u| dx +
u dx f u dx
J1 (u) =
2 S
8 S
S
we have that J1 : U R is convex and strongly continuous, therefore weakly lower semi-continuous, so that
j+
On the other hand, from (13.5):

Z
Z
2
(unj ) dx
u20 dx, as j +
S
(13.7)
and thus, from (13.6) and (13.7) we may write:
1 = inf {J(u)} = lim inf {J(unj )} J(u0 ).

uU
j+

13.2. A Concave Dual Variational Formulation
We start this section by enunciating the following theorem which
has been proven in chapter 10.
a convex G
U R
U R convex, coercive and lower semi-continuous (l.s.c.) such

that the functional
J(u) = (G )(u) F (1 u) hu, u0iU
is bounded from below , where : U Y and 1 : U Y1 are

Then we may write:
inf
z Y1 v B (z )
uU
where B (z ) = {v Y such that v 1 z u0 = 0}.

Our next result refers to a convex dual variational formulation,
through which we obtain sufficient conditions for optimality.
Theorem 13.2.2. Consider J : U R, where
Z
Z
Z
1
u2
2
2
J(u) =
|u| dx +
( ) dx f u dx,
S 2
S 2 2
S
and U = W01,2 (S). For K = 1/K0 , where K0 stands for the constant related to Poincare inequality, we have the following duality
principle
inf {J(u)}
sup
{GL (v , z )}
uU
(z ,v1 ,v0 )B
where
GL (v , z )
and
Z
Z
Z
1
(v1 )2
1
1
2
2
=
(z
)
dx
+
|z
|
dx
dx
2K 2 S
2K S
2 S v0 + K
Z
Z
1
2
+
(v ) dx + v0 dx,
(13.8)
2 S 0
S
B = {(z , v1 , v0) L2 (S; R3 ) |

1
2 z + v1 z = f, v0 + K > 0, a.e. in S, z = 0 on }.
K
If in addition there exists u0 U such that J(u0 ) = and v0 +K =
(/2)u20 + K > 0, a.e. in S, then
J(u0 ) = min{J(u)} =
uU
where
max
(z ,v1 ,v0 )B
{GL (v , z )} = GL (
v , z ),
2
u ,
2 0
v1 = (
v0 + K)u0
v0 =
and
z = Ku0 .
Proof. Observe that we may write
J(u) = G(u) F (1u)
f u dx,
13.2. A CONCAVE DUAL VARIATIONAL FORMULATION
291
where
G(u) =
where
u2
K
( + 0)2 dx +
2
S 2 2
Z
K
F (1 u) =
u2 dx,
2 S
1
|u|2dx +
2
u2 dx,
u = {0 u, 1 u, 2u},
and
0 u = 0, 1 u = u, 2 u = u.
From Theorem 13.2.1 (here this is an auxiliary theorem through
which we obtain A , below indicated), we have
inf {J(u)} = inf sup {F (z ) G (v )},
uU
where
z Y1 v A
1
F (z ) =
2K
and
(z )2 dx,
S
Z
Z
Z
Z
1
1
(v1 )2
1
2
2
|v2 | dx+
(v0 ) dx+ v0 dx,
dx+
G (v ) =
2 S
2 S v0 + K
2 S
S
if v0 + K > 0, a.e. in S, and
A = {v Y | v 1 z f = 0},
or
A = {(z , v ) L2 (S) L2 (S; R5 ) |

Observe that
and thus
div(v2 ) + v1 z f = 0, a.e. in S}.
G (v ) hu, v iY G(u), u U, v A ,
F (z ) + G (v ) F (z ) + h1 u, z iL2 (S) + hu, f iU G(u),

(13.9)
and hence, making z an independent variable through A , from
(13.9) we may write
sup {F (z ) + G (v2 (v1 , z ), v1, v0 )}
z L2 (S)
+h1 u, z iL2 (S) +
sup {F (z )
z L2 (S)
f u dx G(u) , (13.10)
so that
sup
z L2 (S)
2K
Z
Z
(v1 )2
1
1

2
(z ) dx +
(v2 (z , v1 )) dx +
dx
2 S
2 S v0 + K
S

Z
Z
1
2
+
(v ) dx + v0 dx
2 S 0
ZS
F (1 u) + f udx G(u). (13.11)
Therefore if K 1/K0 (here K0 denotes the constant concerning

Poincare Inequality), the supremum in the left side of (13.11) is
attained through the relations
v2 =
z
and z = 0 on ,
K
so that the final format of our duality principle is given by

Z
Z
1
1
2
(z )2 dx
inf {J(u)}
sup
|z | dx +
2
uU
2K S
2K S
(z ,v1 ,v0 )B

Z
Z
Z
1
(v1 )2
1
2
(v ) dx v0 dx , (13.12)
dx
2 S v0 + K
2 S 0
S
where
B = {(z , v1 , v0) L2 (S; R3 ) |
1
2 z + v1 z = f, v0 + K > 0, a.e. in S, z = 0 on }.
K
The remaining conclusions follow from an application (with little
changes) of Theorem 8.1.27.

Remark 13.2.3. The relations
v2 =
z
and z = 0 on ,
K
are sufficient for attainability of supremum indicated in (13.11) but

just partially necessary, however we assume them because the expression of dual problem is simplified without violating inequality
(13.12) (in fact the difference between the primal and dual functionals even increases under such relations).
13.3. APPLICATIONS TO PHASE TRANSITION IN POLYMERS
293
13.3. Applications to Phase Transition in Polymers

We consider now a variational problem closely related to the
Ginzburg-Landau formulation. See [12] and other references therein
for more information how this applies to phase transition in polymers). For an open bounded S R3 with a sufficient regular
boundary denoted by , let us define J : U V R, as
Z
Z
Z
1
2
2
2
|u| dx +
(u 1) dx +
|v|2 dx,
J(u, v) =
2 S
4 S
2 S
under the constraints,
1
|S|
udx = m,
(13.13)
2 v = u m,
(13.14)
and
Z
vdx = 0.
(13.15)
Here U = V = W 1,2 (S), is a small constant and 1 < m < 1. We

may rewrite the primal functional, now denoting it by J : U V
= R {+} as
R
J(u, v) = G1 ((u, v)) F (u).
Here : U V Y L2 (S) L4 (S) L2 (S; R3 ) L2 (S; R3 ) is

defined as
(u, v) = {0 u = 0, 1u = u, 2 u = u, 3 v = v},
also
G1 ((u, v)) = G((u, v)) + Ind1 (u, v) + Ind2 (u, v) + Ind3 (u, v).
Hence
G((u, v)) =
2
1
|u| dx +
4
S
Z
K
(u 1 + 0) dx +
u2 dx
2
S
S
Z
+
|v|2dx,
2 S
Z
0, if 1
udx = m,
Ind1 (u, v) =
|S| S
+, otherwise,
Z
Ind2 (u, v) =
0, if 2 v + u m = 0, a.e. in S,
+, otherwise,
Z
0, if
vdx = 0,
Ind3 (u, v) =
S
+, otherwise,
and,
Z
K
u2 dx.
F (u) =
2 S
Similarly to Theorem 13.2.1, through appropriate Lagrange multipliers for the constraints, we may write
{J(u, v)}

Z

F (z ) G (u , v ) + 1 m + 2 mdx ,
sup
inf
inf
(u,v)U V
z Y1 (u ,v ,)A
where
1
G (u , v ) =
2
|u1 |2 dx
1
+
2
if 2u0 + K > 0, a.e. in S, and

1
F (z ) =
2K
Z
(u2 )2
dx + (u0 )2 dx
2u0 + K
S
Z
Z
1
|v |2 dx,
+ u0 dx +
2
S
S
(z )2 dx.
Also, defining Y = L2 (S) Y R L2 (S) R, we have

A = A1 A2 A3 ,A1 = {(z , u, ) Y |
1
2 = 0 a.e. in S and u1 n = 0 on },
div(u1 ) u2 + z
|S|
A2 = {(z , u, ) Y | 2 2 + 3 div(v ) = 0 a.e. in S,
2
v n +
= 2 = 0, on },
n
and
A3 = {(z , u, ) Y | 2u0 + K > 0, a.e. in S}.
295
Similarly to the case of Ginzburg-Landau formulation we can obtain

inf
(u,v)U V
{J(u, v)}
Z
Z
Z
(u2)2
1
1
2
2
(z ) dx
|z | dx+
dx
{ 2
sup
2K S
2K S
2 S 2u0 + K
(z ,u ,v ,)C
Z
Z
Z
Z
1
2
2
|v | dx 1 m 2 mdx},
(u0 ) dx + u0 dx
2 S
S
S
S
where
C = C1 A2 A3 ,
2 z
1
C1 = {(z , u, ) Y |
u2 +z
2 = 0 a.e. in S
K
|S|
z
and
= 0 on }.
n
Remark 13.3.1. It is important to emphasize that by analogy
to Section 13.2, we may obtain sufficient conditions for optimality.
That is, if there exists a critical point for the dual formulation for
which 2u0 +K > 0 a.e. in S and K = /K0 (where K0 stands for the
constant related to P oincar
e inequality), then the corresponding
primal point through Legendre transform relations is also a global
minimizer and the last inequality is in fact an equality, as has been
pointed out in Theorem 13.2.2.
13.3.1. Another Two Phase Model in Polymers. The following problem has applications in two phase models in Polymers.
To minimize the functional J : U V R (here consider S R3
as above), where
Z
J(u, v) = |Du|(S) +
|v|2 dx,
2 S
under the constraints
udx = m,
S
2 v = u m,
(13.16)
U = BV (S, {1, 1}),
(13.17)
where
and V = W 1,2 (S) (here BV denotes the space of functions with

bounded variation in S and |Du|(S) denotes the total variation of
u in S).
Redefining U as U = W 1,1 (S) L2 (S), we rewrite the primal
formulation, through suitable Lagrange multipliers, now denoting
v, 1, 2 , 3 ) (J : U V L2 (S) R L2 (S) R), as
it by J(u,
Z
Z
Z
1 2
2
J(u, v, ) =
|u|dx +
|v| dx +
(u 1)dx
2 S
S
S 2
Z
Z
+ 2
udx m + 3 (2 v + u m)dx, (13.18)
S
where = (1 , 2 , 3 ). We may write
J(u, v, ) = G((u, v)) + F (u, v, ),

where
|v|2 dx,
G((u, v)) =
|u|dx +
2
S
S
here : U V Y = L2 (S; R3 ) L2 (S; R3 ) is defined as
Z
(u, v) = {1 u = u, 2v = v}
and
F (u, v, ) = 2
Z
Z
udx m + 3 (2 v + u m)dx
S
S
Z
1 2
(u 1)dx.
+
S 2
Defining

Z
U = (u, v) U V |
udx = m, 2 v + u m = 0,
S

u2 = 1 a.e. in S ,
from the Lagrange Multiplier version of Theorem 8.2.5 we have that

inf {G((u, v)) + F (u, v, )}
(u,v)U
sup
(v ,)Y B
{G (v ) F ( v , )},
where
1
G (v ) = sup{hv, v iY G(v)} =
2
vY
|v2 |2 dx + Ind0 (v1 ),
and
Ind0 (v1 )
297
0, if |v1 |2 1, a.e. in S,
+, otherwise.
We may define
Ind1 (v1 )
0, if v1 n = 0, on ,
+, otherwise,
and
0, if div(v2 ) 2 3 = 0, a.e. in S,
3
= 3 = 0 on ,
v2 n +
n
Ind2 (v2 ) =
+, otherwise,
so that
F ( v , ) =
|div(v1 ) 2 3 |dx + Ind2 (v2 ) + Ind1 (v1 )

S
Z
+ 2 m + 3 mdx.
S
Therefore, we can summarize the last results by the following duality principle,

Z
1
inf {G((u, v)) + F (u, v, )} =
sup
|v2 |2 dx
2
(u,v)U
(v ,)A B
S

Z
Z
|div(v1 ) 2 3 |dx 2 m 3 mdx ,

S
where
A = {v Y = L2 (S; R3 ) L2 (S; R3) | |v1 (x)|2 1,

and
a.e. in S and v1 n = 0, on }.
B = {(v2 , 3 ) L2 (S; R3 )L2 (S)|div(v2)2 3 = 0, a.e. in S,

3
v2 n +
= 3 = 0 on }. (13.19)
n
Remark 13.3.2. It is worth noting that the last dual formulation represents a standard convex non-smooth optimization problem. The non-smoothness is the responsible by a possible microstructure formation. Furthermore such a formulation seems to be
amenable to numerical computation (in a simpler way as compared

to the primal approach).
In this section we present numerical results for a one-dimensional
example originally due to Bolza (see [29] for details about the primal formulation).
Consider J : U R expressed as
Z
Z
1 1
1 1
2
2
J(u) =
((u,x ) 1) dx +
(u f )2 dx
2 0
2 0
or, defining S = [0, 1],
1
G(u) =
2
and
1
F (u) =
2
we may write
((u,x )2 1)2 dx
1
0
(u f )2 dx
J(u) = G(u) + F (u)

where, for convenience we define, : U Y L4 (S) L2 (S) as
Furthermore, we have
u = {u,x , 0}.
U = {u W 1,4 (S) | u(0) = 0 and u(1) = 0.5}
For Y = Y = L4 (S) L2 (S), defining

Z
1
((u,x + p1 )2 1.0 + p0 )2 dx
G(u + p) =
2 S
for v0 > 0 we obtain
G(u)+F (u) inf {hp0 , v0 iL2 (S) hp1 , v1 iL2 (S) +G(u+p)+F (u)}
pY
or
G(u) + F (u) inf {hq0 , v0 iL2 (S) hq1 , v1 iL2 (S) + G(q)
pY
Here q = u + p so that
+ h0, v0iL2 (S) + hu , v1iL2 (S) + F (u)}.
G(u) + F (u) G (v ) + h0, v0iL2 (S) + hu,x , v1 iL2 (S) + F (u).
13.4. A NUMERICAL EXAMPLE
299
That is
G(u) + F (u) G (v ) + inf {h0, v0iL2 (S) + hu,x, v1 iL2 (S) + F (u)},
uU
or
inf {G(u) + F (u)} sup {G (v ) F ( v )}
uU
v A
where
Z
Z
1
1
(v1 )2
dx +
G (v ) =
(v )2 dx,
2 S v0
2 S 0
if v0 > 0, a.e. in S. Also
Z
1

[(v1 ),x ]2 dx + hf, (v1 ),x iL2 (S) v1 (1)u(1)
F ( v ) =
2 S
and
A = {v Y | v0 > 0, a.e. in S}.
Remark 13.4.1. Through the extremal condition v0 = ((u,x)2

1) and Weierstrass condition (u,x )2 1.0 0 we can see that the
dual formulation is convex for v0 > 0, however it is possible that
the primal formulation has no minimizers, and we could expect a
microstructure formation through v0 = 0 (that is, u,x = 1, depending on f(x)). To allow v0 = 0 we will redefine the primal
functional as below indicated.
Define G1 : U R and F1 : U R by
Z
K
G1 (u) = G(u) + F (u) +
(u,x )2 dx
2 S
and
Z
K
(u,x)2 dx.
F1 (u) =
2 S
R
Also defining G(u)

= G(u) + K2 S (u,x )2 dx, from Theorem 13.2.1
we can write
(v , v ) F (v )}
sup {F (z ) G
inf {J(u)} inf
uU
z Y v B (z )
(13.20)
where
Z
1
(z )2 dx,
=
2K S
Z
Z
2
(v
)
1
1
2
(v , v ) =
(v0 )2 dx,
dx +
G
0 2
2 S v0 + K
2 S
F1 (z )
0.6
0.5
0.4
0.3
0.2
0.1
0.1
0.2
0.4
0.6
0.8
Figure 1. Vertical axis: u0 (x)-weak limit of minimizing sequences for f(x)=0
(v1 )
1
=
2
(v1 )2 dx + hf, v1 iL2 (S) v2 (1)u(1)
and
B (z ) = {v Y | (v2 ),x +v1 z = 0 and v0 0 a.e. in S}.
We developed an algorithm based on the dual formulation indicated in (13.20). It is relevant to emphasize that such a dual
formulation is convex if the supremum indicated is evaluated under
the constraint v0 0 a.e. in S (this results follows from the traditional Weierstrass condition, so that there is no duality gap between
the primal and dual formulations and the inequality indicated in
(13.20) is in fact an equality).
We present numerical results for f (x) = 0 (see figure 1), f (x) =
0.3 Sin( x) (figure 2) and f (x) = 0.3 Cos( x) (figure 3).
The solutions indicated as optima through the dual formulations
(denoted by u0 ), are in fact weak cluster points of minimizing sequences for the primal formulations.
13.5.
A NEW PATH FOR RELAXATION
301
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.2
0.4
0.6
0.8
Figure 2. Vertical axis: u0 (x)-weak limit of minimizing sequences for f (x) = 0.3 Sin( x)
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.2
0.4
0.6
0.8
Figure 3. Vertical axis: u0 (x)-weak limit of minimizing sequences for f (x) = 0.3 Cos( x)
13.5. A New Path for Relaxation
Similarly as in Theorem 13.2.1, we may prove:
inf {G1 (u) + G2 (u) F (u) hu, f iU }
uU
inf { sup {F (z ) G2 (z + div(v ) + f ) G1 (v )}},
z Y v Y
where,
G1 (u) =
|u|2 dx,
2 S
Z
|u|4 dx,
G2 (u) =
4 S
and
Z
|u|2 dx.
F (u) =
2 S
Also
Z
1

F (z ) =
|z |2 dx,
2 S
Z
3

G2 (z + div(v ) + f ) =
(z + div(v ) + f )4/3 dx.
41/3 S
and
Z
1

|v |2 dx.
G1 (v ) =
2 S
Let us denote
and
J (v , z ) = F (z ) G2 (z + div(v ) + f ) G1 (v ),
J1 (v , z ) = F (z ) G2 (z + div(v ) + f ) hu, v iY
Z
+
|u|2 dx.
2 S
The Euler-Lagrange equations for this last functional, are given by
F (z ) G2 (z + div(v ) + f )
= 0, in S
(13.21)
z
z
and

G2 (z + div(v ) + f )
u
= 0, in S
(13.22)
z
that is
1/3

z
z + div(v ) + f
=
, in S
(13.23)
and
1/3

z + div(v ) + f
u
= 0, in S.
(13.24)
From (13.23) we have
(z )3 = ()3 (z + div(v ) + f ), in S.
(13.25)
13.5.
303
Let us denote by {zk }, where k {1, 2, 3}, the three roots of (13.25),
which at first are assumed to be all real, and define
Jk (v , zk ) = F (zk ) G2 (zk + div(v ) + f )
Z
1
hzk , v iY /() +
|zk |2 dx. (13.26)
2()2 S
Denoting
uk =
zk
F (zk )
=
z
we obtain
and from (13.23),
F (zk ) = huk , zk iY F (uk ),

uk =
G2 (zk + div(v ) + f )
,
z
that is
G2 (zk + div(v ) + f ) = huk , zk + div(v ) + f iU G2 (uk ).
Therefore
J1 (v , zk )
Henceforth we denote
=
4
|uk | dx
|uk |2 dx
2 S
S
Z
+
|uk |2 dx inf {J(u)}.
uU
2 S
Z
F (u) =
G2 (u) =
f(u) dx,
where
Therefore
g2 (u) dx,
S
2
u dx,
f(u) =
2
g2 (u) = |u|4 .
4
Z

F (z ) =
f (z ) dx,
S
Z
G2 (v ) =
g2 (v ) dx,
S
(13.27)
being
1
(z )2 ,
f (z ) =
2
and
g2 (v ) =
3
|v |4/3 .
1/3
4
Let 1 , 2 , 3 be given such that k (x) {0, 1}, a.e. in S, and

P3
k=1 k (x) = 1, a.e. in S. Again denote by {zk } for k {1, 2, 3}

the three roots of equation (13.25) (at first assumed to be real).
Define
3
X
u=
k zk /(),
k=1
and
z =
3
X
k zk .
k=1
Observe that from the values assumed by 1 , 2 , 3 we may write:
F (z ) =
Z X
3
S k=1
= h
= h
3
X
k f (zk ) dx
k u k ,
k=1
k=1
3
X
3
X
k u k ,
3
X
k=1
k=1
k zk iY
k zk iY
Z X
3
k ) dx
k f(u
S k=1
F(
3
X
k uk ),
(13.28)
k=1
and by analogy
G2 (z
+ div(v ) + f ) =
Z X
3
S k=1
= h
3
X
k g2 (zk + div(v ) + f ) dx
k u k ,
k=1
k=1
G2 (
3
X
3
X
k=1
k zk + div(v ) + f iY
k uk ).
(13.29)
13.5.
305
Thus,
J1 (v , z ) = F (z ) G2 (z + div(v ) + f )
Z
|u|2 dx
hu, v iY +
2 S
Z X
Z X
3
3
=
k f (zk ) dx
S k=1
S k=1
3
X
zk ), v iY /() +
h(
2()2
k=1
!2
Z
3

X

k zk dx

S
k=1
= hu, z iY F (u) hu, z + div(v ) + f iY + G2 (u)

hu, v iY + G1 (u)
= J(u) inf {J(u)}.
uU
(13.30)
We may summarize the last results by

(Z 3
Z X
3
X

inf
k f (zk ) dx
(v ,)Y B
3
X
k=1
S k=1
S k=1
k zk
, v
!2
Z
3

/() +
k zk dx

2()2 S
k=1
inf {J(u)}, (13.31)

uU
where
B = { = (1 , 2 , 3 )
measurable |
3
X
k = 1, a.e. in S}.
k {0, 1},
k=1
Remark 13.5.1. This primal-dual formulation is particularly

useful to compute the global optimal point as is small. In such
a case we expect the solution moves between the wells along the
domain.
Remark 13.5.2. In case just one root zk is real, a little adaptation is necessary. For numerical results, we may for example,
replace the complex roots by a big fixed integer N, in order to not
change the global optimal point.
13.6. A New Numerical Procedure,

the Generalized Method of Lines
Consider first the equation
2 u = 0, in S R2 ,
(13.32)
u = 0 on 0 and u = uf , on 1 .
Here 0 denotes the internal boundary of S and 1 the external
one. Assume 0 is smooth and consider the simpler case where
1 = 20 .
Finally, also suppose there exists r(), a smooth function such that
0 = {(, r()) | 0 2},
being r(0) = r(2).

In polar coordinates the above equation may be written as
2 u 1 u
1 2u
+
+
= 0, in S,
r 2
r r r 2 2
(13.33)
and
u = 0 on 0 and u = uf , on 1 .
Define the variable t by
r
t=
.
r()
Also defining u by
u(r, ) = u(t, ),
dropping the bar in u, equation (13.32) is equivalent to
1
u
2u
+ ( f2 ())
2
t
t
t
2
1
u
f4 () 2 u
+
f3 ()
+ 2
= 0,
t
t
t 2
in S. Here f2 (), f3 () and f4 () are known functions.
More specifically, denoting
f1 () =
we have:
f2 () = 1 +
r ()
,
r()
f1 ()
,
1 + f1 ()2
(13.34)
13.6. A NEW NUMERICAL PROCEDURE...
f3 () =
307
2f1 ()
,
1 + f1 ()2
and
1
,
1 + f1 ()2
Observe that t [1, 2] in S. Discretizing in t (N equal pieces
which will generate N lines ) we obtain the equation
un+1 2un + un1 un un1 1
+
( f2 ())
d2
d
t
(un un1 ) 1
2 un f4 ()
+
f3 () +
= 0,
(13.35)
td
2 t2
n {1, ..., N}. Here, un () corresponds to the solution on the line
n. Thus we may write
f4 () =
un = T (un1, un , un+1 ),
where

un+1 + un1 d2 un un1 1
T (un1, un , un+1) =
( f2 ())
2
2
d
t

2
un f4 ()
(un un1 ) 1
f3 () +
. (13.36)
+
td
2 t2
Now we recall a classical definition.
Definition 13.6.1. Let C be a subset of a Banach space U and

let T : C C be an operator. Thus T is said to be a contraction
mapping if there exists 0 < 1 such that
kT (x1 ) T (x2 )kU kx1 x2 kU , x1 , x2 C.
Remark 13.6.2. Observe that if kT (x)kU < 1, on a

convex set C then T is a contraction mapping, since by the mean
value inequality,
kT (x1 ) T (x2 )kU sup{kT (x)k}kx1 x2 kU , x1 , x2 C.
xC
The next result is the base of our generalized method of lines.

For a proof see [27].
Theorem 13.6.3 (Contraction Mapping Theorem). Let C
be a closed subset of a Banach space U. Assume T is contraction
mapping on C, then there exists a unique x C such that x = T (

x).
Moreover, for an arbitrary x0 C defining the sequence
we have
x1 = T (x0 ) and xk+1 = T (xk ), k N

xk x, in norm, as k +.
From equation (13.36), if d = 1/N is small enough, since un1

un it is clear that for fixed un+1 , G(un ) = T (un1, un , un+1 ) is a
contraction mapping, considering that kG (un )k < 1, for some
0 < < 1 in a set that contains the solution of equation in question.
In particular for n = 1 we have
u1 = T (0, u1, u2).
we may use the Contraction Mapping Theorem to calculate u1 as
a function of u2 . The procedure would be
(1) Set x0 = u2 .
(2) Obtain x1 = T (0, x0 , u2 ),
(3) obtain recursively
xk+1 = T (0, xk , u2),
(4) and finally get
u1 = lim xk = g1 (u2 ).
k
We have obtained thus

u1 = g1 (u2).
We can repeat the process for n = 2, that is we can solve the
equation
u2 = T (u1 , u2 , u3),
which from above stands for:
u2 = T (g1 (u2 ), u2 , u3).
The procedure would be:
(1) Set x0 = u3 ,
(2) calculate
xk+1 = T (g1 (xk ), xk , u3).
(3) obtain
u2 = lim xk = g2 (u3 ).
k
309
We proceed in this fashion until obtaining

uN 1 = gN 1 (uN ) = gN 1 (uf ).
Being uf known we have obtained uN 1 . We may then calculate
uN 2 = gN 2 (uN 1),
uN 3 = gN 3 (uN 2),
and so on, up to finding
u1 = g1 (u2).
The problem is then solved.
Remark 13.6.4. Here we consider some points concerning the
convergence of the method.
In the next lines the norm indicated as in kxk k refers to W 2,2 ([0, 2]).
In particular for n = 1 from above we have:
u1 = T (0, u1, u2).
We will construct the sequence xk (in a little different way as above)
by
x1 = u2 /2,
and
xk+1 = T (0, xk , u2 ) = u2 /2 + dT (xk ),
where the operator T is properly defined from the expression of T .
Observe that
kxk+2 xk+1 k dkT kkxk+1 xk k,
and if
0 = dkTk < 1,
we have that {xk } is (Cauchy) convergent. Through a standard
procedure for this kind of sequence, we may obtain
1
kxk+1 x1 k
kx2 x1 k,
1
so that denoting u1 = lim xk , we get
k
1
dkT kku2/2k,
1
Having such an estimate, we may similarly obtain
ku1 u2 /2k
u2 u3 + O(d),
and generically
un un+1 + O(d), n {1, ..., N 1}.
This last calculation is just to clarify that the procedure of obtaining the relation between consecutive lines through the contraction
mapping theorem is well defined. Anyway, we postpone a more
complete analysis of the approximation error for a future work.
311
Denoting x = , particularly for N = 10, truncating the series

up the term in d2 , we obtain the following expression for the lines
Line 1 :
u1 (x) = 0.1uf (x) + 0.034f2 (x)uf (x)
+0.034f3 (x)uf (x) + 0.008f4 (x)uf (x)
Line 2 :
u2 (x) = 0.2uf (x) + 0.058f2 (x)uf (x)
+0.058f3 (x)uf (x) + 0.015f4 (x)uf (x)
Line 3 :
u3 (x) = 0.3uf (x) + 0.075f2 (x)uf (x)
+0.075f3 (x)uf (x) + 0.020f4 (x)uf (x)
Line 4 :
u4 (x) = 0.4uf (x) + 0.083f2 (x)uf (x)
+0.083f3 (x)uf (x) + 0.024f4 (x)uf (x)
Line 5 :
u5 (x) = 0.5uf (x) + 0.084f2 (x)uf (x)
+0.084f3 (x)uf (x) + 0.026f4 (x)uf (x)
Line 6 :
u6 (x) = 0.6uf (x) + 0.079f2 (x)uf (x)
+0.079f3 (x)uf (x) + 0.026f4 (x)uf (x)
Line 7 :
u7 (x) = 0.7uf (x) + 0.068f2 (x)uf (x)
+0.068f3 (x)uf (x) + 0.022f4 (x)uf (x)
Line 8 :
u8 (x) = 0.8uf (x) + 0.051f2 (x)uf (x)
+0.051f3 (x)uf (x) + 0.018f4 (x)uf (x)
Line 9 :
u9 (x) = 0.9uf (x) + 0.028f2 (x)uf (x)
+0.028f3 (x)uf (x) + 0.010f4 (x)uf (x)
Remark 13.6.5. As mentioned above we have truncated the

series up to the terms in d2 . Of course higher order approximations
are possible, but it seems that to obtain better results, the best
procedure is just to discretize more (that is, to increase N). We can
use higher values of N as 100, 1, 000, or even 100, 000 still keeping
the method viable.
Remark 13.6.6. Finally, it is worth noting that the Generalized Method of Lines may be used to solve a large class of non-linear
problems. As an example we solve the Ginzburg-Landau type equation:
2 u + Au3 ABu = 0, in S R2 ,
(13.37)
with the boundary conditions:

u = 0 on 0 and u = uf , on 1 .
Here A, B > 0, 0 denotes the internal boundary of S and 1 the
external one. Again we assume 0 is a smooth function and consider
the simpler case where
1 = 20 .
Finally, also suppose there exists r(), a smooth function such that
0 = {(, r()) | 0 2},
being r(0) = r(2). Particularly, for N = 10, truncating the series
up the term in d2 , we obtain the following expression for the lines
(here x stands for , being f1 (x), f2 (x), f3 (x) and f4 (x) as above
and f5 (x) = r 2 (x)f4 (x)):
Line 1 :
u1(x) = 0.1uf (x) + 0.034f2(x)uf (x)
+0.034f3(x)uf (x) + 0.008f4 (x)uf (x)
+0.017ABf5 (x)uf (x) 0.005Af5 (x)u3f (x)
Line 2 :
u2(x) = 0.2uf (x) + 0.058f2(x)f5 (x)uf (x)
+0.058f3(x)uf (x) + 0.015f4 (x)uf (x)
+0.032ABf5 (x)uf (x) 0.009Af5 (x)u3f (x)
Line 3 :
u3(x) = 0.3uf (x) + 0.075f2(x)uf (x)
+0.075f3(x)uf (x) + 0.020f4 (x)uf (x)
+0.045ABf5 (x)uf (x) 0.015Af5 (x)u3f (x)
Line 4 :
u4(x) = 0.4uf (x) + 0.083f2(x)uf (x)
+0.083f3(x)uf (x) + 0.024f4 (x)uf (x)
+0.056ABf5 (x)uf (x) 0.019Af5 (x)u3f (x)
Line 5 :
u5(x) = 0.5uf (x) + 0.084f2(x)uf (x)
+0.084f3(x)uf (x) + 0.026f4 (x)uf (x)
+0.062ABf5 (x)uf (x) 0.023Af5 (x)u3f (x)
Line 6 :
u6(x) = 0.6uf (x) + 0.079f2(x)uf (x)
+0.079f3(x)uf (x) + 0.026f4 (x)uf (x)
+0.063ABf5 (x)uf (x) 0.026Af5 (x)u3f (x)
Line 7 :
u7(x) = 0.7uf (x) + 0.068f2(x)uf (x)
+0.068f3(x)uf (x) + 0.022f4 (x)uf (x)
+0.059ABf5 (x)uf (x) 0.026Af5 (x)u3f (x)
Line 8 :
u8(x) = 0.8uf (x) + 0.051f2(x)uf (x)
+0.051f3(x)uf (x) + 0.018f4 (x)uf (x)
+0.048ABf5 (x)uf (x) 0.023Af5 (x)u3f (x)
313
Line 9 :
u9 (x) = 0.9uf (x) + 0.045f2 (x)uf (x)
+0.045f3 (x)uf (x) + 0.010f4 (x)uf (x)
+0.028ABf5 (x)uf (x) 0.015Af5 (x)u3f (x).
Remark 13.6.7. Observe that 1 could be any multiple of
0 , in fact any real number greater than 1 multiplied by 0 . Of
particular interest is the case where 1 = k0 and k +. In
this case we simulate + through a big real number. Finally,
the Generalized Method of Lines may be adapted very easily for
cartesian coordinates.
13.7. A Simple Numerical Example
Just to illustrate the possibilities of the generalized method of
lines, consider the equation
2 u = 2 u, in S,
where
S = {(r, ) | 1 r 2, 0 2},
u = u on 0 and 1 ,
where 0 and 1 are boundaries of the circles with centers at origin

and radius 1 and 2 respectively. Finally, in polar coordinates (here
x stands for ),
u = r 2 cos(x).
See below the values of the 10 lines obtained by the generalized
method of lines (u[n]) and the exact values ( u[n] for the same
lines):
Line 1
u[1] = 1.21683 cos(x),
u[1] = 1.21 cos(x)
u[2] = 1.44713 cos(x),
u[2] = 1.44 cos(x)
u[3] = 1.69354 cos(x),
u[3] = 1.69 cos(x)
Line 2
Line 3
13.8. CONCLUSION
315
Line 4
u[4] = 1.95811 cos(x) u[4] = 1.96 cos(x)
Line 5
u[5] = 2.24248 cos(x),
u[5] = 2.25 cos(x)
u[6] = 2.54796 cos(x),
u[6] = 2.56 cos(x)
u[7] = 2.87563 cos(x),
u[7] = 2.89 cos(x)
u[8] = 3.22638 cos(x),
u[8] = 3.24 cos(x)
u[9] = 3.60096 cos(x),
u[9] = 3.61 cos(x)
Line 6
Line 7
Line 8
Line 9
Remark 13.7.1. We expect the error decreases as we increase the number of lines. Specifically for this example high order
approximations are not necessary.
13.8. Conclusion
In this article we have developed a dual variational formulation
for a Ginzburg-Landau type equation. The dual approach obtained
is represented by a concave functional. Also, sufficient conditions
of optimality are provided. Finally, in the last section we introduce
the new Generalized Method of Lines and it is important to emphasize that the results developed may be applied to a great variety
of situations. An application for the compressible Navier-Stokes
system is planned for a future work.
CHAPTER 14
Duality Applied to Conductivity in Composites

14.1. Introduction
For the primal formulation we repeat the statements found in
reference [19] (U.Fidalgo, P.Pedregal). Consider a material confined into a bounded domain RN , N > 1. The medium is obtained by mixing two constituents with different electric permitivity
and conductivity. Let Q0 and Q1 denote the two N N symmetric
matrices of electric permitivity corresponding to each phase. For
each phase, we also denote by Lj , j = 0, 1, the anisotropic N N
symmetric matrix of conductivity. Let 0 t1 1 be the proportion of the constituent 1 into the mixture. Constituent 1 occupies a
space in the physical domain which we denote by E . Regarding the set E as our design variable, we introduce the characteristic
function : {0, 1}:

1, if x E,
(14.1)
(x) =
0, otherwise,
Thus,
Z
dx =
(x)dx = t1
dx = t1 ||.
(14.2)
The matrix of conductivity corresponding to the material as a

whole is L = L1 + (1 )L0 .
Finally, the electrostatic potential, denoted by u : R is
supposed to satisfy the equation
div[L1 u + (1 )L0 u] = P (x), in ,
(14.3)

u = u0 , on
(14.4)
where P : R is a given source or sink of current (we assume

P L2 ()).
317
318
14. DUALITY APPLIED TO CONDUCTIVITY IN COMPOSITES
14.2. The Primal Formulation

Now and on we assume N = 3. Consider the problem of minimizing the cost functional,

Z
(1 )
T
T
(u) Q1 u +
(u) Q0 u dx (14.5)
I(, u) =
2
2
subject to
div[L1 u + (1 )L0 u] = P (x)
(14.6)
where u U, here U = {u W 1,2 () | u = u0 on }.

We will rewrite this problem as the minimization of J : U Y
R, where Y = L2 (S; R3),
J(u, f ) =

Z
(1 t)
t
T
T
((u) Q1 u) +
((u) Q0 u) + Ind1 (u, f ) dx
inf
tB
2
2
+ Ind2 (u, f ),
Ind1 (u, f ) =
0, if (tL1 + (1 t)L0 )u f = 0,
+, otherwise,
and
Ind2 (u, f ) =
0, if div(f ) = P a.e. in ,
+, otherwise.
Here
B = {t measurable | t(x) {0, 1}, a.e. in ,
t(x)dx = t1 ||}.
14.3. The Duality Principle

Our main duality principle is summarized by the following theorem.
be given by:
Theorem 14.3.1. Let J : U Y R
J(u) = G(u, f ) + G1 (f ) F (u),
14.3. THE DUALITY PRINCIPLE
319
where
G(u, f ) =

Z
(1 t)
t
T
T
((u) Q1 u) +
((u) Q0 u) + Ind1 (u, f ) dx
inf
tB
2
2
Z
K
+
|u|2 dx, (14.7)
2
Z
K
F (u) =
|u|2 dx
2
and
G1 (f ) = Ind2 (f ).
Thus, we may write
inf
(u,f )U Y
{J(u, f )}
sup
(v ,)C U
{ inf {F (z ) G (v + z , )
z Y
+h, P iL2() hu0, v .niL2 () },
(14.8)
where
G (v + z , f ) =
sup
(u,f )U Y
{hv, v + z iY + hf, f iY G(v, f )},
that is,
G (v + z , )
Z

1
= sup
((v + z , ) )Q(t)((v + z , )) dx ,
2
tB
and
(v + z , ) = (v + z + (t(L1 ) + (1 t)L0 )T ),
1
F (z ) =
2K
Finally,
|z |2 dx.
= (tQ1 + (1 t)Q0 + KI)1 ,

Q(t)
where I denotes the identity matrix,
C = {(v , ) Y U | div(v ) = 0, a.e. in , = 0 on },

and
B = {t measurable | t(x) {0, 1}, a.e. in ,
t(x)dx = t1 ||}.
320
14. DUALITY APPLIED TO CONDUCTIVITY IN COMPOSITES
Proof. Observe that

G (v + z , ) h, P iL2() + hu0 , v .niL2 ()
hf, iY +hu, v +z iY h, P iL2() +hu0 , v .niL2 () G(u, f ),
u U, v , z Y , f Y , so that
G (v + z , ) h, P iL2() + hu0 , v .niL2 ()
hu, z iY Ind2 (f ) G(u, f ), (14.9)
u U, v C , z Y , f Y , and thus
F (z ) + G (v + z , ) h, P iL2() + hu0, v .niL2 ()
F (z ) + hu, z iY Ind2 (f ) G(u, f ), (14.10)
u U, v C , z Y , f Y .
Taking the supremum in z at both sides of last inequality, we
obtain
sup {F (z ) + G (v + z , ) h, P iL2() + hu0, v .niL2 () }
z Y
F (u) Ind2 (f ) G(u, f ), (14.11)
u U, v C , f Y . and hence

inf {F (z ) G (v + z , )
inf {J(u, f )}
sup
(u,f )U Y
(v ,)C
z Y
+h, P iL2() hu0 , v .niL2 () }}.
(14.12)
Remark 14.3.2. We conjecture that if K > 0 is big enough

then the duality gap is zero. In fact we are supposing that
G(un , fn ) = G (un , fn )
for any n N sufficiently big, where {(un , fn )} is a minimizing
sequence for the primal problem.
14.4. Conclusion
In this chapter we developed duality for a two-phase non-convex
variational problem in conductivity. As we may not have minimizers for this kind of problem, the solution of the dual variational
formulation reflects the average behavior of minimizing sequences,
14.4. CONCLUSION
321
as a weak cluster point of such (minimizing) sequences. Finally, it

seems that the solution of dual problem is not difficult to compute.
CHAPTER 15
Duality Applied to the Optimal

Design in Elasticity
15.1. Introduction
The first objective of the present article is the establishment of
a dual variational formulation for the optimal design, concerning
the minimization of internal work, of a plate of variable thickness
(in [29] we have reference to similar problems). Such a thickness
is denoted by h(x) and allowed to assume the values between a
minimum h0 and maximum h1 . The total plate volume, assumed
to be fixed, is a design constraint denoted by V . Now we pass to
describe the primal formulation.
Consider a plate which the middle surface is denoted by S R2 ,
where S is an open bounded connected set with a sufficiently regular
boundary denoted by . As mentioned above the design variable
h(x) is such that h0 h(x) h1 , where x = (x1 , x2 ) S R2
. The field of normal displacements to S, due to a external load
P L2 (S), is denoted by w : S R.
The optimization problem is the minimization of J : U R,
where

Z
H (t)
J(w) = inf
w, w, dx,
(15.1)
tC S
2
subject to
(H (t)w, ), = P, in S
(15.2)
and
Z
(th1 + (1 t)h0 ) dx = t1 h1 |S| = V ,
where
C = {t measurable | t(x) [0, 1], a.e. in S},
323
(15.3)
324
15. DUALITY APPLIED TO THE OPTIMAL...
0 < t1 < 1 and |S| denotes the Lebesgue measure of S. Moreover,

w
2,2
2,2
= 0 on .
(15.4)
U = W0 (S) = w W (S) | w =
n
Finally,
H (t) = (th1 + (1 t)h0 )3 A
(15.5)
where h(x) = t(x)h1 + (1 t(x))h0 and {A } is a positive definite matrix related to Hookes Law. Observe that 0 t(x)
1, a.e. in S.
15.2. The Main Duality Principles
The next theorem is also relevant for the subsequent results. A
similar result may be found in [6]. We develop a proof again for
the sake of completeness.
Theorem 15.2.1. Let U be a Banach space. Consider the
(F 1 ) : U R,
such that
l.s.c. functionals (G ) : U R,
is bounded below, where
J : U R
J(u)
= G (u) F (1 u) hu, f iU .
Here : U Y , and 1 : U Y1 are continuous linear operators

such that there exists L : Y1 Y a continuous linear operator
satisfying
u = L(1 u), u U.
Under these hypotheses, we have

inf {J(u)} inf sup {F (L z ) G (v + z )} , (15.6)

uU
z Y
v A
where,
A = {v Y | v = f }.
Proof. Defining = inf uU {J(u)},

we have
Hence
F (1 u) G (u) + hu, f iU + , u U.
h1u, L z iY1 F (1 u) hu, z iY G (u) + hu, f iU + ,
u U, z Y . Therefore
sup {hv1 , L z iY1 F (v1 )} sup{hu, z iY G (u)+hu, f iU }+,
v1 Y1
uU
15.2. THE MAIN DUALITY PRINCIPLES
325
so that, from this and Theorem 8.2.5 we may write,

F (L z ) inf {G (v + z )} + .
v A
That is,
inf
z Y
sup {F (L z ) G (v + z )}
v A
Our main theoretical result is given by the following theorem.

Theorem 15.2.2. Let U be a reflexive Banach space. Consider (G ) : U R and (F 1 ) : U R l.s.c. functionals such
that J : U R defined as
J(u) = (G )(u) (F 1 )(u) hu, f iU
is bounded below. Here : U Y and 1 : U Y1 are continuous
linear operators whose adjoint operators are denoted by : Y
U and 1 : Y U , respectively. Also we suppose the existence
of L : Y1 Y a continuous linear injective operator such that
L : Y Y1 is onto and
(u) = L(1 (u)), u U.
Under such assumptions, we have

inf {F (L z ) G (v + z )} ,
inf {J(u)} sup
uU
v A
z Y
where
A = {v Y | v = f }.
Now in addition assume that (F 1 ) : U R is convex and

Gateaux differentiable, G : Y R is Gateaux differentiable and
strongly continuous and, (G ) : U R, and (F L ) : Y R
are Gateaux differentiable functionals. Also assume that for each
v Y , the functional J : Y Y R, where
J (v , z ) = F (L z ) G (v + z ),
is such that
J (v , z ) +, as kz kY or kL z kY1 .
326
Furthermore, suppose that if {zn } Y is such that kL zn kY1 <

K, n N for some K > 0 then there exists z Y such that, for
a not relabeled subsequence, we have
L zn L z , weakly in Y1 ,
and
zn z , strongly in Y .
Under such assumptions, for each v Y we may find z Y
here denoted by z (v ) such that
(v ) inf {J (v , z )} = J (v , z (v )).
z Y
Observe that z (v ) is obtained through the extremal equation

F (L z ) G (v + z )
= .
z
z
Moreover, assume the second variation related to z Y of J (v , z )
is positive definite at z (v ), in the sense that
z2 (J (v , z (v )), ) > 0, C0 , 6= .
Finally, suppose that there exists (u0 , v0 ) U Y such that
{hu0 , v0 f iU + F (L z (v0 )) G (v0 + z0 (v0 ))} = . (15.7)
Under such additional assumptions, we have

inf {G (u) F (1 u) hu, f iU } =
uu
G (u0 ) F (1u0 ) hu0 , f iU = F (L z0 ) G (v0 + z0 ) =

(15.8)
inf {F (L z ) G (v + z )} .
sup
v A
z Y
Furthermore, the following chain of equalities is satisfied,

inf sup {F (L z ) G (v + z )}
z Y
v A
= inf {G (u) F (1 u) hu, f iU }

uU

inf {F (L z ) G (v + z )} .
= sup
v A
z Y
Proof. Observe that

G (v + z ) hu, v iY + hu, z iY G(u),
u U, z Y , v Y
15.2. THE MAIN DUALITY PRINCIPLES
327
that is,
F (L z )+G (v +z ) hu, f iU F (L z )+h1u, L z iY1 G(u),
u U, z Y , v A , so that
sup {F (L z ) + G (v + z )}
z Y
sup {hu, f iU F (L z ) + h1 u, L z iY1 G(u)}, (15.9)

z Y
v A , u U and therefore, since L : Y Y1 is onto,

sup {F (L z ) + h1 u, L z iY1 } = F (1 u).
z Y
From this and (15.9), we get

G(u) F (1 u) hu, f iU inf {F (L z ) G (v + z )},
z Y
v A , u U
which means
inf {J(u)} sup
uU
v A
where
inf {F (L z ) G (v + z )} ,
z Y
A = {v Y | v = f }.
Similarly we may obtain
inf {G (u) F (1u) hu, f iU }

inf {F (L z ) G (v + z )} . (15.10)
sup
uU
v A
z Y
Fix v Y . Recall that
(v ) inf {J (v , z )}.
z Y
From the coercivity hypothesis, a minimizing sequence {zn } Y

is such that
kL zn kY1 < K1 , n N,
for some K1 > 0. Also from the hypotheses, there exists z Y
such that, up to a not relabeled subsequence
L zn L z , weakly in Y1 ,
and
zn z , strongly in Y .
328
Thus
lim inf {F (L zn )} F (L z ),
n
and, since G : Y R is strongly continuous,

Therefore,
G (v + zn ) G (v + z ), as n .
(v ) = lim inf {F (L zn ) G (v + zn )} F (L z ) G (v + z ).
n
In particular, denoting z0 = z (v0 ) = z0 , we recall that z0 is obtained through the extremal equation
F (L z0 ) G (v0 + z0 )
= .
(15.11)
z
z
Taking the variation of (15.11) in v , we obtain
2 F (L z0 ) z0 2 G (v0 + z0 ) z0 2 G (v0 + z0 )
= ,
(z )2 v
(z )2
v
z v
and considering the hypothesis about the second variation in z of
J (v0 , z ), we may infer that
z0
v
is well defined. From the extremal equation (15.7)

F (L z0 ) G (v0 + z0 ) z0 G (v0 + z0 )
u0 +
= ,
z
z
v
v
and (15.11), we obtain
u0
G (v0 + z0 )
= ,
v
(15.12)
so that
G (v0 + z0 ) = hu0 , v0 + z0 iY G (u0).
(15.13)
From (15.11) and (15.12), we get

F (L z0 )
L
u0 = ,
w
and thus, since L : Y1 Y is injective,
F (L z0 )
1 u0 = ,
w
where w = L z . Hence
F (L z0 ) = h1 u0 , L z0 iY1 F (1 u0 ).
(15.14)
15.3. THE FIRST APPLIED DUALITY PRINCIPLE
329
From the Gateaux variation related to u U in (15.7), we obtain

v0 = f.
(15.15)
From (15.13), (15.14) and (15.15), we get

F (L z0 ) G (v0 + z0 ) = G (u0 ) F (1u0 ) hu0 , f iU .
From this and (15.10) we have that (15.8) holds, that is,
inf {G (u) F (1 u) hu, f iU } =
uu
G (u0 ) F (1u0 ) hu0 , f iU = F (L z0 ) G (v0 + z0 ) =

inf {F (L z ) G (v + z )} .
sup
v A
z Y
Finally, observe that (15.12) and (15.15) means

G (v0 + z0 ) = sup {G (v + z0 )},
v A
and thus
F (L z0 ) + sup {G (v + z0 )} = F (L z0 ) G (v0 + z0 ).
v A
From this, (15.6) and (15.10), we get

inf sup {F (L z ) G (v + z )}
z Y
v A
= inf {G (u) F (1 u) hu, f iU }

uU

inf {F (L z ) G (v + z )} .
= sup
v A
z Y
15.3. The First Applied Duality Principle

Now we rewrite the primal formulation, so that we express J :
as
U Y R,

Z
H (t)
J(w, f ) = inf
w, w, + Ind1 (w, f ) dx
tB
2
S
+ Ind2 (w, f ), (15.16)
where,
Ind1 (w, f ) =
0, if f = H (t)w, ,
+, otherwise,
(15.17)
330
Ind2 (w, f ) =
0, if f, = P, a.e in S,
+, otherwise,
(15.18)
: U Y is given by
w = {w, },
and
B = {t measurable | t(x) [0, 1] a.e. in S,

Z
(th1 + (1 t)h0 ) dx = t1 h1 |S| = V , (15.19)

S
and also Y = L2 (S; R4).

Observe that
inf
(w,f )U Y
{J(w, f )} = inf
tB
inf {G(w, f, t) + F (w, f )} , (15.20)
wU
where
G(w, f, t) =
Z
S
H (t)
w, w, + Ind1 (w, f )
2
dx,
and
F (w, f ) = Ind2 (w, f ).
From Theorem 8.2.5, we may write
inf
(w,f )U Y
{G(w, f, t) + F (w, f )}
sup
(v ,f )Y Y
{G (v , f , t) F ( v , f )}, (15.21)
where
G (v , f , t) =
sup
(v,f )Y Y
{hv , v
iL2 (S) +hf , f
iL2 (S) G(v, f, t)},
or
G (v , f , t) =
sup
(v,f )Y Y
hv , v
iL2 (S) + hf , f
iL2 (S)
Z
S

H (t)
v v + Ind1 (v, f ) dx ,
2
15.3. THE FIRST APPLIED DUALITY PRINCIPLE
331
so that
G (v , f , t) =
sup
(v,f )Y Y
hv , v
iL2 (S) + hH (t)v , f
iL2 (S)

H (t)
v v dx . (15.22)
2
Thus we may write

Z
1
H (t)f
f
dx + hv
, f
iL2 (S)
G (v , f , t) =
2 S
Z
1
(t)v v dx,
H
+

2 S
where
(t)} = {H (t)}1 .
{H
On the other hand
F ( v , f )
=
sup
(w,f )U Y
iL2 (S) hf , f
iL2 (S) F (w, f )},
{hw, , v
or
F ( v , f )
=
sup
(w,f )U Y
{hw, , v
iL2 (S) hf , f
iL2 (S) Ind2 (w, f )},
that is,
F ( v , f )
sup
(w,f )U Y
{hw, , v
iL2 (S) hf , f
iL2 (S) +hw,
f, P iL2 (S) },
where w is an appropriate Lagrange multiplier, so that

hw,
P iL2(S) , if (v , f ) B ,
F ( v , f ) =
+, otherwise,
(15.23)
where
B = {(v , f ) Y Y | f
= w, , v,
= 0, in S}.
332
Hence, the duality principle indicated in (15.21), may be expressed

as
{G(w, f, t) + F (w, f )}

Z
1
H (t)w, w, dx hv
, w, iL2 (S)
sup
2
(v ,w)B
S

Z
1
H (t)v v dx + hw,
P iL2 (S) .
2 S
inf
(w,f )U Y
We may evaluate the last supremum and obtain v = , and therefore,

inf
(w,f )U Y
{G(w, f, t) + F (w, f )}

Z
1
= sup
H (t)w, w, dx + hw,
P iL2 (S) ,
2 S
wU
However, also from Theorem 8.2.5, we may conclude that

Z
1
sup
H (t)w, w, dx + hw,
P iL2(S)
2 S
wU

Z
1
H (t)M M dS ,
=
inf
{M }D
2 S
where
D = {{M } Y | M, + P = 0, in S}.
And thus, the final format of the concerned duality principle would
be
Z

1
(t)M M dS .
inf {J(w, f )} =
inf
H
(w,f )U Y
(t,{M })BD
2 S
15.4. A Concave Dual Formulation
Our next result is a concave dual variational formulation.
as
Theorem 15.4.1. Define J : U Y R,
J(w, f ) = G(w) + G1 (w, f ) F (w),
15.4. A CONCAVE DUAL FORMULATION
333
where
G(w, f ) = inf
tB
Z
S

H (t)
w, w, + Ind1 (w, f ) dx
2
Z
K
+
w, w, dx, (15.24)
2 S
G1 (w, f ) = Ind2 (w, f ),

Ind1 (w, f ) =
Ind2 (w, f ) =
0, if f = H (t)w, ,
+, otherwise,
(15.25)
0, if f, = P, a.e in S,
+, otherwise,
(15.26)
: U Y is given by
w = {w, },
K
F (w) =
2
and
w, w, dx
B = { t measurable | t(x) [0, 1] a.e. in S,

Z
(th1 + (1 t)h0 ) dx = t1 h1 |S| .
S
Under such assumptions we have,

{J(w, f )}

sup
+ hw,
P iL2 (S) } ,
inf {F (z ) G (z + v , w)
inf
(w,f )U Y
(wv
)U B
z Y
where
1
F (z ) =
2K
and
G (z , f ) =
sup
(v,f )Y Y
z
z
dx,
{hv, z iY + hf, f iY G(v, f )},
334
or more explicitly,
Z
1
K (t)z z dx
H
G (z , f ) = sup

2 S
tB
Z
1
K (t)H (t)f z dx
+
H
(15.27)

2 S

Z
1
K

H (t)H (t)H (t)f f dx .

+
2 S
Here
(t)} = {H (t)}1 and {H
K (t)} = {H (t) + KI}1 ,
{H
and, I denotes the identity matrix. Finally,
B = {v Y | v,
= 0 in S}.
Proof. Observe that

G1 ( v , f ) =
sup
(w,f )U Y
{hw, v iY +hf, f iY Ind2 (w, f )},
or
G1 ( v , f )
=
sup
(w,f )U Y
{hw, v iY + hf, f iY + hw,

f, P iY }.
Thus
G1 ( v , f )
where
hw,
P iL2(S) , if (v , f ) B1 ,
+, otherwise,
B1 = {(v , f ) Y Y | f
= w, , v,
= 0, in S}.
On the other hand
G (z + v , w)
hw,P
iL2 (S) = hw, z iY + hw, v iY
+ hf, wi
Y G(w, f ) hw,
P iL2 (S)
hw, z iY G(w, f ) G1 (w, f ),
(z , v , w)
Y B U, w U, f Y. From last inequality we
obtain
F (z ) + G (z + v , w)
hw,
P iL2(S)
F (z ) + hw, z iY G(w, f ) G1 (w, f ),
15.5. DUALITY FOR A TWO-PHASE PROBLEM IN ELASTICITY
335
(z , v , w)
Y B U, w U, f Y. Taking the supremum
in z in both sides of last inequality we obtain:

sup {F (z ) + G (z + v , w)}
hw,
P iL2(S)
z Y
F (w) G(w, f ) G1 (w, f ),

w U, w U, f Y, v B , and finally, we may write
inf
(w,f )U Y
{J(w, f )}
sup
(w,v
)U B
inf {F (z )
z Y
G (z + v , w)
+ hw,
P iL2 (S)
. (15.28)
Remark 15.4.2. The last results are similar to those of Theorem 15.2.2, however the latter problem has a format slightly different from the one of such a theorem, so that we preferred to develop
in details the dual formulation. We conjecture that if K > 0 is
big enough, so that a minimizing sequence for the primal problem
is inside the region of convexity of G(w, f ) + G1 (w, f ), then the
inequality indicated in (15.28) is in fact an equality. Finally, it is
worth noting that the dual problem seems to be simple to compute.
15.5. Duality for a Two-Phase Problem in Elasticity
In this section we develop duality for a two phase problem in
elasticity. Consider V R3 , and open connected bounded set with
a sufficiently regular boundary denoted by V = S0 S1 , where
S0 S1 = . Here V stands for the volume of a elastic solid under
the action of a load P L2 (V ; R3 ). The field of displacements is
denoted by u = (u1 , u2, u3 ) U where
U = {u W 1,2 (V ; R3 ) | u = (0, 0, 0) on S0 }.
(15.29)
The strain tensor, denoted by e = {eij }, is defined as

1
eij (u) = (ui,j + uj,i).
2
(15.30)
The solid V is supposed to be composed by mixing two constituents,

namely 1 and 0, with elasticity matrices related to Hookes Law
1
0
denoted by Hijkl
and Hijkl
, respectively. The part occupied by
constituent 1 is denoted by E and represented by the characteristic
336
function : V {0, 1} where

1, if x E,
(x) =
0, otherwise,
(15.31)
Now we define the optimization problem of minimizing J(u, )

where
Z

1
1
0
Hijkl
eij (u)ekl (u) + (1 )Hijkl
eij (u)ekl (u) dx,
J(u, ) =
2 V
(15.32)
subject to
1
0
Hijkl
ekl (u) + (1 )Hijkl
ekl (u)
u U and
,j
+ Pi = 0, in V,
dx t1 |V |,
(15.33)
(15.34)
here 0 < t1 < 1 and |V | denotes the Lebesgue measure of V .

We rewrite the primal formulation, now denoting it by J : U
Y R as

Z
Hijkl (t)
eij (u)ekl (u) + Ind1 ({eij (u)}, f ) dx
J(u, f ) = inf
tB V
2
+ Ind2 (u, f ), (15.35)
where
1
0
Hijkl(t) = tHijkl
+ (1 t)Hijkl
,

1
0
0, if fij = tHijkl
ekl (u) + (1 t)Hijkl
ekl (u),
Ind1 ({eij (u)}, f ) =
+, otherwise,

0, if fij,j + Pi = 0, a.e in S,
Ind2 (u, f ) =
+, otherwise,

Z
B = t measurable | t(x) {0, 1}, a.e. in V,
t(x) dx t1 |V | ,
V
and also Y = L2 (V ; R9 ).
By analogy to last section, we may obtain
Z

1
inf {J(u, f )} =
inf
Hijkl (t)ij kl dx ,
(u,f )U Y
(t,)BB
2 V
where
ijkl } = {Hijkl(t)}1 ,
{H
B = { Y | ij,j + Pi = 0, in V, ij nj = 0 on S1 },
and
15.5. DUALITY FOR A TWO-PHASE PROBLEM IN ELASTICITY
337
Remark 15.5.1. Similarly to earlier sections, we may also

obtain the duality principle:

inf {J(u, f )}
sup
u))
inf {F (z ) G (z + v , e(
(u,f )U Y
z Y
(
u,v )U C
where
1
F (z ) =
2K
and

+h
u, P iL2(S;R3 ) } ,
zij zij dx,
Z
1
K (t)z z dx
G (z , f ) = sup
H
ijkl
ij kl
2
tB
S
Z
1

K (t)Hklop (t)fop
+
H
zij dx
(15.36)
2 S ijkl

Z
1
K
(t)Hopmn (t)f f dx .
Hijkl (t)H
+
klop
mn ij
2 S
Here
ijkl (t)} = t{H 1 }1 + (1 t){H 0 }1

{H
ijkl
ijkl
and
K (t)} = t{H 1 + KI}1 + (1 t){H 0 + KI}1 .

{H
ijkl
ijkl
ijkl
Finally
C = {v Y | vij,j
= 0 in V, vij nj = 0, on S1 }.
Also observe that the last dual functional, now denoted by J (v , u),
may be written as

Z
1
J (v , u) = inf inf
zij zij dx
tB z Y
2K S
Z
1
K (t)(z + v )(z + v ) dx
H
ij
ij
kl
kl
2 S ijkl
Z
1
K
ijkl
H
(t)Hklop (t)eop (
u)(zij + vij ) dx
2 S

Z
1
K
(t)Hopmn (t)emn (
Hijkl(t)H
u)eij (
u) dx
klop
2 S
+h
u, P iL2 (S;R3 ) .
(15.37)
After the evaluation of the infimum in z , we obtain a concave

functional in (v , u
), as the infimum (in t) of a collection of concave
338
functionals in (v , u), even though we have not proved it in the

present work.
Consider again the optimization problem which is given by the
minimization of J : U R, where

Z
H (t)
J(w) = inf
w, w, dx,
(15.38)
tC S
2
subject to
(H (t)w, ), = P, in S
(15.39)
and
Z
(th1 + (1 t)h0 ) dx = t1 h1 |S| = V ,
(15.40)
where
C = {t measurable | t(x) [0, 1], a.e. in S},
t1 = 0.46 (in this example) and |S| denotes the Lebesgue measure
of S. Moreover,

U = W02,2 (S) = w W 2,2 (S) | w = 0 on .
(15.41)
We develop numerical results for the particular case where,
H (t) = H(t) = h(t)3 E,
(15.42)
where h(t) = th1 + (1 t)h0 and h1 = 0.1 and h0 = 104 and

E = 106 , with the units related to the International System (here we
have denoted x = (x1 , x2 )). Observe that 0 t(x) 1, a.e. in S.
The concerned duality principle would be
Z

1
inf {J(w, f )} =
inf
H(t)M
M dS .
(w,f )U Y
(t,{M })BD
2 S
where
D = {{M } Y | M, + P = 0, in S, M n n = 0 on }.
We have computed the dual problem for S = [0, 1] [0, 1] and a
vertical load acting on the plate given by P (x) = 10000, obtaining
the results indicated in the respective figures, for t(x) and w0 (x).
15.7. CONCLUSION
339
1
0.8
0.6
0.4
0.2
0
1
1
0.8
0.5
0.6
0.4
0
0.2
0
Figure 1. Vertical axis: t(x) where t(x)h1 is the plate thickness
0
0.002
0.004
0.006
0.008
0.01
0.012
1
1
0.8
0.5
0.6
0.4
0
0.2
0
Figure 2. Vertical axis: w0 (x)-field of displacements

for P (x) = 10000
15.7. Conclusion
In this article we have developed dual variational formulations
first for the optimal design of the variable thickness of a plate and,
340
in a second step, for a two-phase problem in elasticity. The infima in t indicated in the dual formulations represent the structure
search for stiffness in the optimization process, which implies the
minimization of the internal work. In some cases the primal problem may not have solutions, so that the solution of dual problem is
a weak cluster point of minimizing sequences for the primal formulation. We expect the results obtained can be used as engineering
project tools.
CHAPTER 16
Duality Applied to Micro-Magnetism

16.1. Introduction
In this article we develop dual variational formulations for models in micro-magnetism. For the primal formulation we refer to
references [26, 30] for details.
Let R3 be an open bounded set with a finite Lebesgue
measure and a regular boundary denoted by , being the corresponding outer normal denoted by n. Consider the model of
micro-magnetism in which the magnetization m : R3 , is given
by the minimization of the functional
Z
Z
Z
2
|m|2 dx + (m(x)) dx H(x) m dx
J(m, f ) =
2
Z
1
+
|f (z)|22 dz, (16.1)
2 R3
where
m = (m1 , m2 , m3 ) W 1,2 (; R3 ) Y1 , |m(x)|2 = 1, a.e. in
(16.2)
and f L2 (R3 ; R3 ) Y2 is the unique field determined by the
simplified Maxwells equations
Curl(f ) = 0, div(f + m ) = 0, a.e. in R3 .
(16.3)
Here H L2 (; R3 ) is a known external field and is a function

defined by

1, if x ,
(16.4)
(x) =
0, otherwise.
The term
|m|22 dx
341
342
16. DUALITY APPLIED TO MICRO-MAGNETISM
is called the exchange energy, where

v
u 3
uX
m2
|m| = t
2
k=1
and
|m|22 =
3
X
k=1
|mk |22 .
Finally, (m) represents the anisotropic contribution and is given

by a multi-well functional whose minima establish the preferred
directions of magnetization.
16.2. The Primal formulations and the Duality Principles
16.2.1. Summary of Results for the Hard Uniaxial Case.
We examine first the case of uniaxial material with no exchange
energy. That is, = 0 and (x) = (1 |m e|).
Observe that
(m) = min{(1 + m e), (1 m e)}
where > 0 and e R3 is a unit vector. Thus we can express the

= R {+}, as
the functional J : Y1 Y2 R
J(m, f ) = G(m, f ) + F (m)
where
Z
1
G(m, f ) =
min{g1 (m), g2 (m)} dx +
|f (z)|22 dz
2
3
R
+ Ind0 (m) + Ind1 (f ) + Ind2 (m, f ),
Z
and
F (m) =
Here,
H(x) m dx.
g1 (m) = (1 + m e),
g2 (m) = (1 m e),

0,
if |m(x)|2 = 1 a.e. in ,
Ind0 (m) =
+, otherwise,

0,
if div(f + m ) = 0 a.e. in R3 ,
Ind1 (m, f ) =
+, otherwise,
16.2. THE PRIMAL FORMULATIONS AND THE DUALITY PRINCIPLES343
and
0,
if Curl(f ) = , a.e. in R3 ,
+, otherwise.
The dual functional for such a variational formulation can be
expressed by the following duality principle:
inf (m,f )Y1 Y2 {J(m, f )}
( ( Z 3
X 2
+ Hi + (1 2t)ei )2 )1/2 dx
sup
inf ( (
tB
k=1 xi
(1 ,2 )Y
Z
Z
1
|Curl 1 2 |2 dx
+ dx (16.5)
2 R3
Ind2 (f ) =
where
B = {t measurable | t(x) [0, 1], a.e. in }.
and
Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.

16.2.2. The Results for the Full Semi-linear Case. Now
we present the duality principle for the full semi-linear case, that
and F : Y1 Y2 R
is, for > 0. First we define G : Y1 Y2 R

as
Z
Z
2
|m|2 dx + min{g1 (m), g2 (m)} dx
G(m, f ) =
2
Z
1
+
|f (z)|22 dz + Ind0 (m) + Ind1 (m) + Ind2 (m), (16.6)
2 R3
and
Z
F (m, f ) =
H m dx.
Also,
g1 (m) = (1 + m e),
and
g2 (m) = (1 m e),

0,
if |m(x)|2 = 1 a.e. in ,
Ind0 (m) =
+, otherwise,

0,
if div(f + m ) = 0 a.e. in R3 ,
Ind1 (m, f ) =
+, otherwise,
Ind2 (f ) =
0,
+, otherwise.
344
For J(m, f ) = G(m, f ) + F (m, f ), the dual variational formulation

is given by the following duality principle

Z
1
|y |22 dx
inf
{J(m, f )}
sup
inf
tB
(m,f )Y1 Y2
2
(1 ,2 ,y )Y Y
0
)
Z X
3
2 2 1/2
( (div(yi) + Hi + (1 2t)ei +
) ) dx
x
i
i=1
Z
Z
1
2
|2 Curl 1 |2 dz +
dx, (16.7)
2 R3
where
B = {t measurable | t(x) [0, 1], a.e. in },
Y0 = {y W 1,2 (; R33 ) | yi n = 0 on , i {1, 2, 3}},
and
Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.

Remark 16.2.1. It is important to emphasize that in both
cases the dual formulations are concave. Thus the dual problems
always have solutions, even when for the hard uniaxial case the
minimizer in the primal problem is not attained.
16.3. A Preliminary Result
Now we present a simple but very useful result, through which
we establish our first duality principles.
Theorem 16.3.1. Consider (G ) : V R (not necessarily
convex) such that J : V R defined by
J(m) = G(m) hm, f iV , m V,
is bounded from below (here as usual : U Y is a continuous
linear operator). Under such assumptions, we have
inf {J(m)} = sup {(G ) ( y )}
mV
y A
where
A = {y Y | y f = 0}.
16.4. THE DUALITY PRINCIPLE FOR THE HARD CASE
345
Proof. The proof is simple, just observe that

(G ) ( y ) = (G ) (f ) = sup {hm, f iV G(m)},
mV
y A .
Remark 16.3.2. What seems to be relevant is that, when
computing (G ) ( y ), we obtain a duality which is perfect
concerning the convex envelope of the primal formulation.
16.4. The Duality Principle for the Hard Case
We recall the primal formulation for the hard uniaxial case,
expressed by J(m, f ) where
J(m, f ) = G(m, f ) + F (m),
Z
Z
1
|f (z)|22 dz
G(m, f ) =
min{g1 (m), g2 (m)}dx +
2
3
R
+ Ind0 (m) + Ind1 (f ) + Ind2 (m, f ),
and
F (m, f ) =
Also,
H(x) m dx.
g1 (m) = (1 + m e),
g (m) = (1 m e),
2
0,
if |m(x)|2 = 1 a.e. in ,
Ind0 (m) =
+, otherwise,

0,
if div(f + m ) = 0 a.e. in R3 ,
Ind1 (m, f ) =
+, otherwise,
and

0,
Ind2 (f ) =
+, otherwise.
From Theorem 16.3.1, we may write
inf
(m,f )Y1 Y2
{J(m, f )}
sup
(m ,f )Y1 Y2
{G (m , f ) F (m , f ))}. (16.8)
We now calculate the dual functionals. First we have that

G (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) +hf, f iL2 (R3 ;R3 ) G(m, f )},
346
or
G (m , f ) =
sup
(m,f )Y1 Y2
hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
min{g1 (m), g2 (m)} dx
R3
|f (z)|22 dz
Ind0 (m) Ind1 (f ) Ind2 (m, f )} .
(16.9)
That is,
G (m , f ) =
sup
(m,f )Y1 Y2
inf
tB
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
(tg1 (m) + (1 t)tg2 (m)) dx
Ind0 (m) Ind1 (f ) Ind2 (m, f )},
R3
|f (z)|22 dz
(16.10)
where
Thus,
G (m , f ) =
sup
(m,f,t)Y1 Y2 B
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
(tg1 (m) + (1 t)g2 (m) )dx
R3
Ind0 (m) Ind1 (f ) Ind2 (m, f )},
|f (z)|22 dz
(16.11)
or,
G (m , f ) = sup{
sup
tB (m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
(tg1 (m) + (1 t)g2 (m)) dx
R3
Ind0 (m) Ind1 (f ) Ind2 (m, f )}}.
|f (z)|22 dz
(16.12)
16.4. THE DUALITY PRINCIPLE FOR THE HARD CASE
347
Hence
G (m , f ) = sup{
sup
tB (m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
(tg1 (m) + (1 t)g2 (m))dx
R3
|f (z)|22 dz
3
X 2
(
m 1)dx hCurl(f ), 1iL2 (R3 ,R3 )
2 i=1 i
hdiv(f + m ), 2 iL2 (R3 ) }},
where , 1 and 2 are appropriate Lagrange Multipliers concerning

the respective constraints.
Therefore,
G (m , f ) = sup{ sup {hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
tB (m,f )Y1 Y2
Z
(t(1 + m.e) + (1 t)(1 m.e))dx
Z
|f (z)|22 dz
R3
3
X 2
(
m 1)dx hCurl(f ), 1iL2 (R3 ;R3 )
2 i=1 i
hdiv(f + m ), 2 iL2 (R3 ) }}.
The last indicated supremum is attained for functions satisfying

the equations
mi + (1 2t)ei mi +
2
=0
xi
or
mi =
mi + (1 2t)ei +
2
xi
=0
and thus from the constraint

3
X
i=1
we obtain
m2i 1 = 0
3
X
2 2 1/2
= ( (mi + (1 2t)ei +
) ) .
x
i
i=1
348
Also, the supremum in f is achieved for functions satisfying

f f + Curl1 2 = 0.
Observe that we need the condition 2 = 0 on to have a finite

supremum, so that
(Z 3
X
2 2 1/2
( (mi + (1 2t)ei +
G (m , f ) = sup inf
) ) dx
xi
tB (1 ,2 )Y
i=1
Z
Z
1
2
|f + Curl 1 2 |2 dx
dx.
+
2 R3
Furthermore
F (m , f )
=
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 ) F (m, f )},
F (m , f )
=
Z
{hm, m iL2 (;R3 ) +hf, f iL2 (R3 ;R3 ) H(x)m dx},
sup
(m,f )Y1 Y2
so that
F (m , f ) =
where
0,
+,
if (m , f ) B ,
otherwise,
(16.13)
B {(m , f ) Y1 Y2 | m = H a.e. in , f = , a.e. in R3 }.
Therefore we may summarize the duality principle indicated in

(16.8) as
inf
(m,f )Y1 Y2
{J(m, f )} =
( Z
3
X
2 2 1/2
) ) dx
tB
xi
i=1
(1 ,2 )Y

Z
Z
1
2
|Curl 1 2 |2 dx + dx , (16.14)
2 R3
where
and
Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.
inf
sup
(Hi + (1 2t)ei +
16.5. AN ALTERNATIVE DUAL FORMULATION FOR THE HARD CASE

349
Thus, interchanging the infimum and supremum in the dual formulation we finally obtain,
{G(m, f ) + F (m, f )}
( ( Z 3
X 2
+ Hi + (1 2t)ei )2 )1/2 dx
sup
inf ( (
tB
x
i
i=1
(1 ,2 )Y
Z
Z
1
2
|Curl 1 2 |2 dx
+ dx
(16.15)
2 R3
where
B = {t measurable | t(x) [0, 1] a.e. in }.
and
Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.
inf
(m,f )Y1 Y2
16.5. An Alternative Dual Formulation for the Hard Case

16.5.1. The Primal Formulation. We examine again the
case where = 0 and (x) = (1 |m.e|), the hard uniaxial
case.
Observe that
(m) = min{(1 + m.e), (1 m.e)}
where > 0 and e R3 is a unitary vector. Thus we can express

= R {+}, (here V Y1 Y2 ), as
the functional J : V R
J(m, f ) = G(m, f ) + F (m)
where
1
G(m, f ) =
min{g1 (m), g2 (m)}dx +
2
R3
|f (z)|2 dz
K
K
hm, miY1 ||,
2
2
and
F (m, f ) = Ind0 (m) + Ind1 (f ) + Ind2 (m, f )
Here we have defined,
H(x) mdx.
g1 (m) = (1 + m.e),
g (m) = (1 m.e),
2
0,
if |m(x)|2 = 1 a.e. in ,
Ind0 (m) =
+, otherwise,
350
Ind1 (m, f ) =
0,
if div(f + m ) = 0 a.e. in R3 ,
+, otherwise,
and
Ind2 (f ) =
0,
+, otherwise.
16.5.2. The Duality Principle. Similarly as in Theorem 8.2.5,

we may write
inf
(m,f )Y1 Y2
{G (m, f ) + F (m, f ))}

sup
(m ,f )Y1 Y2
{G (m , f ) F (m , f ))}. (16.16)
Having in mind the application of such a duality principle, we

pass to the calculation of dual functionals. First we have that
G (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) +hf, f iL2 (R3 ;R3 ) G(m, f )},
or
G (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
min{g1 (m), g2 (m)}dx
R3
|f (z)|2 dz
K
K
hm, miY1 + ||}, (16.17)
2
2
that is,
G (m , f )
2

Z
Z
1
K
2
2
|m | dx +
|m e| dx +
||
=
|e| +
2K
K
2
2
Z
1
|f |2 dx.
+
2 R3
On the other hand
F (m , f )
=
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 ) F (m, f )},

351
that is,
F (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iY2
Ind0 (m) Ind1 (f ) Ind2 (m, f ) +
H(x) mdx.}. (16.18)
Hence, as the evaluation of this last supremum is a quadratic optimization problem, so that we may write:
F (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
3
X
(
m2i 1)dx hCurl(f ), 1iL2 (R3 ,R3 )
2
i=1
Z
hdiv(f + m ), 2 iL2 (R3 ) + H(x) m dx},
(16.19)
where , 1 and 2 are appropriate Lagrange Multipliers concerning

the respective constraints.
The last supremum in m is attained for functions satisfying the
equations
2
=0
mi + Hi mi +
xi
or
2
mi + Hi +
xi
mi =
=0
and thus from the constraint

3
X
m2i 1 = 0
i=1
we obtain
3
X
2 2
)
(mi + Hi +
x
i
i=1
!1/2
The supremum in f equals +, unless the following relation is

satisfied:
f Curl 1 2 = , in R3 .
Observe that we need the condition 2 = 0 on to have a finite
supremum, so that
!1/2
Z
3
X
2 2
F (m , f ) = inf {
(mi + Hi +
dx}.
)
x
(1 ,2 )Y
i
i=1
352
Therefore we may summarize the duality principle indicated in

(16.16) as:
{G (m, f ) + F (m, f )}
Z
Z
1
2
sup {
|m |2 dx
|m e| dx
2K
K
m
Y

2
K
||
|e|22 +
2
2
!1/2
Z
3
X
2 2
(mi + Hi +
dx
)
x
i
Z i=1
1
|Curl 1 2 |22 dx},
2 R3
inf
(m,f )Y1 Y2
where
(16.20)
m
= (m , 1 , 2 ),
Y = Y Y ,
1
and
Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.
Remark 16.5.1. The last inequality is in fact an equality, considering that the dual formulation is concave and coercive, so that
through an application of the direct method of calculus of variations, we may infer that supremum is attained with a corresponding
point that solves the primal formulation. Thus the equality follows
from the standard relations between primal and dual functionals.
Denoting
Z
K
K
G1 (m, f ) =
min{g1 (m), g2 (m)}dx + hm, miY1 ||,
2
2
1 (m ) = 1
G
2K
|m |22
dx +
K
and
(m , 2 ) =
G
2
3
X
|m e| dx

2
K
2
||,
|e| +
+
2 2
2
2 2
(mi + Hi +
)
x
i
i=1
!1/2
dx,

353
the optimality conditions for the dual problem are given by:
and
1 (m ) + G
2 (m , 2 ),
G

div(Curl 1 2 ) div G2 (m , 2 ) ,
Curl(Curl1 2 ) = , in R3 .
Hence, denoting f0 = Curl 1 2 , there exists m0 Y1 such

that
(m ),
m0 G
1
and therefore,
(m ) = hm0i , m iL2 () G
(m0 ).
G
1
i
1
(16.21)
Also
that is:
(m , 2 ),
m0 G
2
(m , 2 ) = hm0 , m + 2 iL2 (,R3 ) G

(m0 )
G
2
2
= hm0 , m + H + 2 iL2 (,R3 ) Ind0 (m0 )
Ind1 (m0 , f0 ) Ind2 (f0 ),
(16.22)
and,
div(m0 f0 ) = 0, in R3 .
(16.23)
Finally,
Z
1
|Curl1 2 |22 dx = hf0 , Curl 2 iL2 (R3 ,;R3 )
2 R3
Z
1
|f0 |22 dx. (16.24)

2 R3
Thus, from (16.21), (16.22), (16.23) and (16.24), we have
Z
1
(m0 ) +
|f0 |22 dx hm0 , HiL2 (,R3 )
G
1
2 R3
Ind0 (m0 ) + Ind1 (m0 , f0 ) + Ind2 (f0 )
Z
1
(m ) G
(m , 2 )
= G
|Curl 1 2 |22 dx.
1
2
2 R3
354
Remark 16.5.2. We conjecture that if K > 0 is big enough,

so that for a minimizing sequence of the original problem we have
(mn ) = G
1 (mn ) for all n N sufficiently big, then we can
G
1
(m) by G
1 (m) in the last duality principle, and hence
replace G
1
the duality gap between the primal and dual problems is zero.
16.6. The Full Semi-linear Case
Now we present a study concerning duality for the full semilinear case, that is, for > 0. First we recall the definition of
and F : Y1 Y2 R,
that is,
G : Y1 Y2 R
Z
Z
2
|m|2 dx + min{g1 (m), g2 (m)}dx
G(m, f ) =
2
Z
1
+
|f (z)|22 dz + Ind0 (m) + Ind1 (m) + Ind2 (m),
2 R3
and
Z
F (m, f ) =
H m dx.
Also,
g1 (m) = (1 + m e),
and
g2 (m) = (1 m e),

0,
if |m(x)|2 = 1 a.e. in ,
Ind0 (m) =
+, otherwise,

0,
if div(f + m ) = 0 a.e. in R3 ,
Ind1 (m, f ) =
+, otherwise,
0,
if Curl(f ) = 0, a.e. in R3 ,
+, otherwise.
From Theorem 16.3.1, we have
Ind2 (f ) =
inf
(m,f )Y1 Y2
{G(m, f ) + F (m, f )}
sup
(m ,f )Y1 Y2
{G (m , f ) F (m , f )}, (16.25)
where
G (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 ) G(m, f )},
16.6. THE FULL SEMI-LINEAR CASE
355
and
F (m , f )
=
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 ) F (m, f )}.
Thus, from above definitions we may write

hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
G (m , f ) =
sup
(m,f )Y1 Y2
Z
Z
|m|2 dx
min{g1 (m), g2 (m)}dx
2

Z
1
2
|f (z)|2 dz Ind0 (m) Ind1 (m) Ind2 (m) .
2 R3
Or
G (m , f ) =
sup {hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )

(m,f )Y1 Y2
Z
|m|22 dx
2
Z
inf (tg1 (m) + (1 t)g2 (m))dx
tB
Z
1
|f (z)|22 dz
2 R3
Ind0 (m) Ind1 (m) Ind2 (m)},
(16.26)
where
Hence
G (m , f ) =
sup
(m,f,t)Y1 Y2 B
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
|m|22 dx
2
Z
Z
1
|f (z)|22 dz
(tg1 (m) + (1 t)g2 (m))dx
2 R3
Ind0 (m) Ind1 (m) Ind2 (m)},
(16.27)
356
or
G (m , f ) = sup
sup
tB (m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
|m|22 dx
2
Z
Z
1
|f (z)|22 dz
(tg1 (m) + (1 t)g2 (m))dx
2
3
R
Ind0 (m) Ind1 (m) Ind2 (m)}.
(16.28)
Thus, as the second supremum is a quadratic optimization problem,

there exist Lagrange multipliers (, 1 , 2 ) L2 (S) L2 (S; R3 )
L2 (S), such that
G (m , f ) = sup
sup
tB (m,f )Y1 Y2
{hm, m iL2 (;R3 ) + hf, f iL2 (R3 ;R3 )
|m|22 dx
2
Z
Z
1
(tg1 (m) + (1 t)g2 (m))dx
|f (z)|22 dz
2 R3
Z
3
X 2
(
mi 1)dx hCurl(f ), 1iL2 (R3 ;R3 )
2 i=1
hdiv(f + m ), 2 iL2 (R3 ) }.
Hence we may write,
G (m , f ) = sup{ sup {G(m)

F (m, t)} + F1 (f )},
tB mY1
where
F1 (f )
and
1
=
2
|f + Curl 1 2 |22 dx
Z
G(m) =
|m|22 dx,
2
m = m,
R3
F (m, t) = hm, m iL2 (;R3 ) +
tg1 (m) + (1 t)g2 (m)dx
3
X 2
(
m 1)dx + hdiv(m ), 2 iL2 (R3 ) . (16.29)
2 i=1 i
357
Therefore, similarly as in Theorem 8.2.5,

sup
(m,f )Y1 Y2
(y ) + F ( y , t)},
{G(m)
F (m, t)} inf {G
y Y0
where
G (y ) =
sup
{hy, y iL2 (;R33 )
2
yL2 (,R33 )
|y|22dx}
1
=
2
|y |22 dx
and
F ( y , t) = sup {hmi , yiiL2 (;R3 ) F (m, t)}.
mY1
Thus we may write

F ( y , t) = sup {hmi , yiiL2 (,R3 ) + hm, m iL2 (;R3 )
mY1
Z
(t(1 + m e) + (1 t)(1 m e))dx
3
X 2
m 1)dx hdiv(m ), 2 iL2 (R3 ) }.
(
2 i=1 i
(16.30)
The last supremum is attained for functions satisfying yi n+2 ni =

0 on , for all i {1, 2, 3}, where n denotes the outer normal to
(such a condition is necessary to guarantee a finite supremum).
Furthermore
div(yi) + mi + (1 2t)ei mi +
2
= 0,
xi
or
mi =
div(yi ) + mi + (1 2t)ei +
From the constraint

=
P3
i=1
2
xi
m2i 1 = 0, we obtain
2
3
X
2
div(yi ) + mi + (1 2t)ei +
xi
i=1
!1/2
358
so that
F ( y , t) =
2 !1/2
Z
3
X
2
dx
div(yi) + mi + (1 2t)ei +
x
i
i=1
Z
dx.
Therefore, summarizing the last results, we may write

Z
1
G (m , f ) sup inf {
|y |22 dx
tB y Y0 2
Z X
3
2 2 1/2
+ ( (div(yi) + mi + (1 2t)ei +
) ) dx}
xi
i=1
Z
Z
1
2
| f + Curl 1 2 |2 dz dx,
+
2 R3
where, Y0 = {y W 1,2(; R33 ) | yi n + 2 ni = 0 on , i

{1, 2, 3}} . On the other hand
F (m , f ) =
{hm, m iL2 (;R3 ) + hf, f iL2 (;R3 )

(m,f )Y1 Y2
Z
+ H mdx}
sup
so that
F (m , f ) =
0,
if (m , f ) B ,
+, otherwise,
where
B = {(m , f ) Y1 Y2 | m = H, a.e. in , f = a.e. in R3 }.
359
Therefore, we could summarize the duality principle indicated in

(16.25) as
(

Z
1
|y |22 dx
sup
sup
inf
{J(m, f )} inf
tB
(m,f )Y1 Y2
2
(1 ,2 )Y y Y0
)
Z X
3
2
( (div(yi) + Hi + (1 2t)ei +
)2 )1/2 dx
x
i
i=1
Z
Z
1
2
|2 Curl 1 |2 dz + dx,
(16.31)
2 R3
where
Y0 = {y W 1,2 (; R33 ) | yi n + 2ni = 0 on , i {1, 2, 3}},
and
Y = W 1,2 (R3 ; R3 ) W 1,2 (R3 ).
Hence, we may obtain the following duality principle, for which the
dual for formulation is concave.
{G(m, f ) + F (m, f )}

Z
1
sup
inf
|y |22 dx
tB
2
(1 ,2 ,y )Y Y
inf
(m,f )Y1 Y2
)
Z X
3
2
( (div(yi) + Hi + (1 2t)ei +
)2 )1/2 dx
x
i
i=1
Z
Z
1
|2 Curl 1 |22 dz + dx,
(16.32)
2 R3
where
Y0 = {y W 1,2 (; R33 ) | yi n = 0 on , i {1, 2, 3}},
and
Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.
360
16.7. The Cubic Case in Micro-magnetism

In this section we present results for the cubic case. For the
primal formulation we refer to reference [26] for details.
Let R3 be an open bounded set with a finite Lebesgue
measure and a regular boundary denoted by and, consider the
model of micro-magnetism in which the magnetization m : R3 ,
is given by the minimization of the functional J(m, f ), where:
Z
Z
Z
2
|m|2 dx + (m(x)) dx H m dx
J(m, f ) =
2
Z
1
+
|f (z)|22 dz,
2 R3
m W 1,2 (; R3 ) Y1 , |m(x)|2 = 1, a.e. in
and f L2 (R3 ; R3 ) Y2 is the unique field determined by the

simplified Maxwells equations
Curl(f ) = 0, div(f + m ) = 0, in R3 .
Here the function (m), for the cubic anisotropy is given by:
X
(m) = K0 + K1
m2i m2j + K2 m21 m22 m23 ,
i6=j
where K1 , K2 > 0. Also H L2 (; R3 ) is a known external field

and is a function defined as

1, if x ,
(x) =
0, otherwise.
The term
|m|22 dx
stands for the exchange energy. Finally, (m) represents the anisotropic contribution and is given by a multi-well functional whose minima establish the preferred directions of magnetization.
Remark 16.7.1. It is worth noting that (m) has six points
of minimum, namely r1 = (1, 0, 0), r2 = (0, 1, 0) and r3 = (0, 0, 1),
r4 = (1, 0, 0), r5 = (0, 1, 0) and r6 = (0, 0, 1) which define the
preferred directions of magnetization.
16.7. THE CUBIC CASE IN MICRO-MAGNETISM
361
16.7.1. The Primal Formulation. For = 0, we define the

= R {+}, where
primal formulation as J : Y1 Y2 R
J(m, f ) = G(m, f ) + F (m, f ).
= R {+} is defined as
Here G : Y1 Y2 R
K
G(m, f ) =
(m) dx +
2
Z X
3
m2k 1) dx
(
(16.33)
k=1
and we shall rewrite as the approximation (and it must be emphasized that this is an approximation for the original problem)
(m) = min{g1 (m), g2 (m), g3 (m), g4 (m), g5 (m), g6 (m)},
where
gk (m) = (rk ) +
3
X
(rk )
mi
i=1
(mi rki )
1 X X 2 (rk )
(mi rki )(mj rkj )
+
2 i=1 j=1 mi mj
and, as above mentioned, r1 = (1, 0, 0), r2 = (0, 1, 0), r3 = (0, 0, 1),

r4 = (1, 0, 0), r5 = (0, 1, 0) and r6 = (0, 0, 1) are the points
through which the possible microstructure is formed for the cubic
as
case, what justify such expansions. Also, we define F : V R,
Z
Z
1
2
|f (z)|2 dz
H m dx
F (m, f ) =
2 R3
+ Ind1 (m) + Ind2 (m, f ) + Ind3 (f ), (16.34)

where
Ind1 (m) =
Ind2 (m, f ) =
0,
if |m(x)|2 = 1 a.e. in ,
+, otherwise,
0,
if div(f + m ) = 0 a.e. in R3 ,
+, otherwise,
and
Ind3 (f ) =
0,
+, otherwise.
362
16.7.2. The Duality Principles. Similarly as in Theorem

8.2.5, we have
inf
(m,f )Y1 Y2
{G (m, f ) + F (m, f )}
sup
(m ,f )Y1 Y2
{G (m , f ) F (m , f )}, (16.35)
where
G (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) +hf, f iL2 (;R3 ) G(m, f )},
so that, defining gk (m) = gk (m) +
K
(
2
P3
k=1
G (m, f ) =
R
maxk{1,...,6}{
gk(m)} dx +
+,
m2k ), we have
K
||,
2
if (m , f ) B ,
otherwise,
where
B = {(m , f ) Y1 Y2 | f = (0, 0, 0) , a.e. in R3 }.
Also,
F (m , f )
=
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) hf, f iL2 (;R3 ) F (m, f )},
or
F (m , f ) =
Z
1
sup {hm, m iL2 (;R3 ) hf, f iL2 (;R3 )
|f (z)|22 dz
2 R3
(m,f )Y1 Y2
Z
+ H m dx Ind1 (m) Ind2 (m, f ) Ind3 (f )},
that is
F (m , f ) =
sup
(m,f )Y1 Y2
{hm, m iL2 (;R3 ) hf, f iL2 (;R3 )
Z
Z
1
2
|f (z)|2 dz + H m dx
2 R3
Z
3
X 2
(
mk 1)dx hCurl(f ), 1iL2 (R3 ;R3 )
2 k=1
h(div(f + m ), 2 iL2 (R3 ) },
16.7. THE CUBIC CASE IN MICRO-MAGNETISM
363
so that
Z X
3
2
F (m , f ) = ( (
mi + Hi )2 )1/2 dx
x
i
i=1
Z
1
|f + Curl1 2 |22 dx}.
2 R3
Hence the duality principle indicated in (16.35) may be expressed

as
inf
(m,f )Y1 Y2
{G (m, f ) + F (m, f )}
Z
sup
max {
gk (m )} dx
k{1,...,6}
(m ,1 ,2 )Y
Z X
3
2
mi + Hi )2 )1/2 dx
( (
i=1 xi

Z
1
K
|Curl 1 2 |2 dz ||, (16.36)

2 R3
2
where
Y = {(m , 1 , 2 ) Y1 W 1,2 (R3 ; R3 ) W 1,2 (R3 )) |
2 = 0 on }. 2
By analogy, we may obtain the results for the semi-linear cubic case
(for > 0), namely
{G(m,
f ) + F (m, f )}
Z
Z
1
sup
max {
gk (m )} dx
|y |22 dx
k{1,...,6}
2
(m ,1 ,2 ,y )Y Y0
Z X
3
2
mi + Hi )2 )1/2 dx
( (div(yi) +
xi
i=1

Z
1
K
|Curl 1 2 |2 dx ||,
2 R3
2
inf
(m,f )Y1 Y2
where
G(m,
f) =
2
Y0
= {y W
1,2
|m|22 dx + G(m, f ),
(; R33 ) | yi n = 0 on },
364
and
Y = {(m , 1 , 2 ) Y1 W 1,2 (R3 ; R3 ) W 1,2 (R3 ) |
2 = 0 on }. 2
16.8. Final Results, Other Duality Principles

In this section we establish new duality principles for the problems addressed above and finish it with a conjecture concerning
optimality conditions. We start with the following result.
be defined by
Theorem 16.8.1. Let J : V Y2 R
Z
Z
1
|f (x)|22 dx
J(m, f ) = hmi , mi iY + (1 |m e|) dx +
2
2 R3
hmi , Hi iY + Ind0 (m) + Ind1 (m, f ) + Ind2 (m, f ),

where
Ind0 (m) =
Ind1 (m, f ) =
and
0,
if |m(x)|2 = 1 a.e. in ,
+, otherwise,
0,
if div(f + m ) = 0 a.e. in R3 ,
+, otherwise,
0,
+, otherwise.
Also V = W 1,2 (; R3 ), Y = L2 (), Y = L2 (; R3 ), and Y2 =
L2 (R3 ; R3 ). Observe that we may write
Ind2 (f ) =
J(m, f ) = G1 (m) + G2 (m) + G3 (m) + G4 (f ) + Ind1 (m, f )

+Ind2 (f ) hmi , HiiY F (m),
(16.37)
where
G1 (m) = hmi , mi iY ,
2
Z
K
G2 (m) = (1 |m e|) dx + hmi , mi iY ,
2
K1
G3 (m) =
hmi , mi iY + Ind0 (m, f ),
2
Z
1
G4 (f ) =
|f (x)|22 dx,
2 R3
16.8. FINAL RESULTS, OTHER DUALITY PRINCIPLES
365
and
(K + K1 )
hmi , mi iY .
2
Through such a notation, the following duality principle holds
F (m) =
inf
(m,f )V Y2
{J(m, f )}
sup
(v ,)A Y
inf {F (z ) G (v1 ) G2 (v2 )
z Y
G3 (v3 + z 2 ) G4 (2 Curl 1 )}} , (16.38)

where
F (z ) =
1
hz , z i ,
2(K + K1 ) i i Y
1
hv i , v i i ,
2 1 1 Y
2

Z
Z
1
2
2

|v | dx +
|v e| dx +
|e| ||,
G2 (v2 ) =
2K 2 2
K 2
2 2
R
|v + z 2 |2 dx, if 2 = 0 on ,

3
G3 (v3 2 ) =
+,
otherwise,
G1 (v1 ) =
and
1
G4 (2 Curl 1 ) = k2 Curl 1 k22,R3 .
2
Finally, A = A1 A2 , where
A1 = {v W 1,2 (; R33 )Y Y | div(v1i )v2i v3i +Hi = 0, in },
and
A2 = {v W 1,2 (; R33 ) Y Y | v1i n = 0 on },

Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.
Proof. Concerning the dual functionals, we will make a explicit

calculation only of G3 and G4 (the remaining calculations are elementary). We have that
G3 (v3 + z 2 ) = sup {hm, v3 + z iY + hdiv(m), 2 iY
mV
G3 (m) Ind1 (m)},
(16.39)
366
so that through a suitable Lagrange multiplier 0 , we may write,

G3 (v3 +z 2 ) = sup {hm, v3 +z iY hm, 2 iY +h2 , mniL2 ()
mV
!
#
" 3
Z
X
1
K1
m2k 1 dx}. (16.40)
0
hmi , mi iY
2
2
k=1
The supremum is attained for functions satisfying the following

equation
v3 + z 2 K1 (m) 0 m = ,
(16.41)
that is,
v3 + z 2
m=
.
K1 + 0
From the condition
P3
k=1
m2k = 1 in , 0 must be such that
m=
v3 + z 2
,
|v3 + z 2 |2
where | |2 denotes the euclidean norm in R3 . Thus

R
|v + z 2 |2 dx, if 2 = 0 on ,

3
G3 (v3 + z 2 ) =
+,
otherwise.
For the expression of G4 (2 Curl 1 ), we have that
G4 (2 Curl 1 ) =

Z
1
2
|f (x)|2 dx .
sup h2 , div(f )iY + h1 , Curl(f )iY
2 R3
f Y
The supremum is attained for functions satisfying
f = 2 Curl 1 ,
(16.42)
so that
1
G4 (2 Curl 1 ) = k2 Curl 1 k22,R3 .
2
Now, observe that
(16.43)
G1 (v1 ) + G2 (v2 ) + G3 (v3 + z 2 ) + G4 (2 Curl1 )

hv1i , mi iY + hv2i , mi iY + hm, v3 + z 2 iY
+ hf, 2 Curl 1 iY G1 (m) G2 (m) G3 (m) G4 (f ),

(16.44)
367
(v , ) A , m V, f Y. Thus,
F (z ) + G1 (v1 ) + G2 (v2 ) + G3 (v3 + z 2 )

+G4 (2 Curl 1 )
F (z ) + hm, z iY + hmi , HiiY Ind1 (m, f ) Ind2 (f )
G1 (m) G2 (m) G3 (m) G4 (f ),
(16.45)
(v , ) A , m V, f Y, z Y . Taking the supremum in z

in both sides of last inequality, we obtain
sup {F (z ) + G1 (v1 ) + G2 (v2 ) + G3 (v3 + z 2 )
z Y
+G4 (2
Curl 1 )}
F (m) + hmi , Hi iY Ind1 (m, f ) Ind2 (f )
G1 (m) G2 (m) G3 (m) G4 (f ),
(16.46)
(v , ) A , m V, f Y, and therefore
inf
(m,f )V Y2
{J(m, f )}
sup
(v ,)A Y
inf {F (z ) G (v1 ) G2 (v2 )
z Y
G3 (v3 + z 2 ) G4 (2 Curl 1 )}} . (16.47)

In a similar fashion, we may prove the following result.
be defined by
Theorem 16.8.2. Let J : V Y2 R
Z
Z
1
J(m, f ) = hmi , mi iY + (1 |m e|) dx +
|f (x)|22 dx
2
2 R3
hmi , Hi iY + Ind0 (m) + Ind1 (m, f ) + Ind2 (m, f ).

(16.48)
Observe that we may write
J(m, f ) = G1 (m) + G2 (m) + G3 (m) + G4 (f ) + Ind1 (m, f )
+Ind2 (f ) hmi , HiiY F (m),
(16.49)
where
G1 (m) = hmi , mi iY ,
2
Z
K
G2 (m) = (1 |m e|) dx + hmi , mi iY ,
2
K1
hmi , mi iY + Ind0 (m),
G3 (m) =
2
368
1
G4 (f ) =
2
and
R3
|f (x)|22 dx,
(K + K1 )
hmi , mi iY .
2
Through such a notation, the following duality principle holds
F (m) =
inf
{J(m, f )}

sup
inf {F (div(z )) G (v1 + z ) G2 (v2 )
(m,f )V Y
(v ,)A Y
z Y
G3 (v3 2 ) G4 (2 Curl 1 )}} , (16.50)
where
F (div(z )) =
1
hdiv(zi ), div(zi )iY ,
2(K + K1 )
1
hv1i + zi , v1i + zi iY ,
2
2

Z
Z
1
2
2

|v | dx +
|v e| dx +
|e| ||,
G2 (v2 ) =
2K 2 2
K 2
2 2
R
|v 2 |2 dx, if 2 = 0 on ,

3
G3 (v3 2 ) =
+,
otherwise,
G1 (v1 + z ) =
and
1
G4 (2 Curl 1 ) = k2 Curl 1 k22,R3 .
2
Here, A = A1 A2 , where
A1 = {v W 1,2 (; R33 )Y Y | div(v1i )v2i v3i +Hi = 0, in },

and defining Y1 = W 1,2 (; R33 ) Y Y and Y2 = Y Y Y we
denote:
A2 = {(v , z ) Y1 Y2 | v1i n = 0 and zi n = 0 on },
and
Y = {(1 , 2 ) W 1,2 (R3 ; R3 ) W 1,2 (R3 ) | 2 = 0 on }.
Finally, if K > 0 and K1 > 0 are such that
J1 (z , v ) = F (div(z )) G1 (v1 + z )
369
is coercive in z , for each v1 W 1,2(; R3 ) and there exists (m0 , v , z )

V Y1 Y2 such that
{hm0 , div(v1 ) + v2 + v3 HiY + F (div(z )) G1 (v1 + z )
G (v2 ) G3 (v3 2 ) G4 (2 Curl 1 )} = , (16.51)
then, for
f0 = 2 Curl1 ,
we have
inf
(m,f )V Y2
{G1 (m) + G
2 (m) + G3 (m)
+ G4 (f ) + Ind1 (m, f ) + Ind2 (f ) hmi , HiiY F (m)}
= G1 (m0 )+G
2 (m0 )+G3 (m0 )+G4 (f0 )+Ind1 (m0 , f0 )+Ind2 (f0 )
hm0i , Hi iY F (m0 )
= F (div(z ))G1 (v1 +z )G2 (v2 )G3 (v3 2 )G4 (2 Curl 1 )

=
sup
inf {F (div(z )) G (v1 + z ) G2 (v2 )
(v ,)A Y
z Y
G3 (v3 2 ) G4 (2 Curl 1 )}} .
Proof. Similarly to last theorem, we may prove

inf
{J(m, f )}

sup
inf {F (div(z )) G (v1 + z ) G2 (v2 )
(m,f )V Y
(v ,)A Y
z Y
G3 (v3 2 ) G4 (2 Curl 1 )}} , (16.52)
and
inf
(m,f )V Y2
{G1 (m) + G
2 (m) + G3 (m) + G4 (f )
+Ind1 (m, f ) + Ind2 (f ) hmi , HiiY F (m)}

sup
inf {F (div(z )) G (v1 + z ) G2 (v2 )
z Y
(v ,)A Y
G3 (v3 2 )
G4 (2 Curl 1 )}} .
(16.53)
370
By hypothesis, (m0 , v , z ) V Y1 Y2 is a solution of equation,

{hm0 , div(v1 ) + v2 + v3 HiY + F (div(z )) G1 (v1 + z )
G (v2 ) G3 (v3 2 ) G4 (2 Curl 1 )} = . (16.54)
Thus, from the variation in z we obtain,
(div(zi ))/(K + K1 ) = (v1i + zi )/, in ,
(16.55)
and from the variation in v1 ,
v1 + z = (m0 ).
(16.56)
From these two equations together with the boundary conditions

for v1 and z we obtain,
div(z ) = (K + K1 )m0 ,
(16.57)
so that
F (div(z )) = hm0 , div(z )iY F (m0 ),
(16.58)
G1 (v1 + z ) = hm0 , v1 + z iY G1 (m0 ).
(16.59)
and
The variation relating v2 give us
v2 =
G
2 (m0 )
,
m
(16.60)
that is
G (v2 ) = hm0 , v2 iY G
2 (m0 ).
(16.61)
v3 2
= m0 , in ,
|v3 2 |2
(16.62)
v3 = |v3 2 |2 m0 + 2 , in .
(16.63)
0 = |v3 2 |2 K1 , in ,
(16.64)
v3 = 2 + (0 + K1 )m0 , in ,
(16.65)
From the variation in v3 we have
that is,
Now denoting
we obtain
371
so that
K1
hm0 , m0 iY
G3 (v3 2 ) = hm0 , v3 2 iY
2
!
#
" 3
Z
X
1
2
m0k 1 dx, (16.66)

0
2
k=1
and thus
G3 (v3 2 ) = hm0 , v3 2 iY G3 (m0 ).
(16.67)
The variation in 2 and 1 , give us

v3 2
+ div(2 Curl 1 ) = 0, in R3 ,
div
|v3 2 |2
(16.68)
and
Curl(2 Curl 1 ) = , in R3 .
(16.69)
Hence from (16.68) and (16.62), defining f0 = 2 Curl 1 we

have,
div(m0 f0 ) = 0 in R3 ,
(16.70)
Curl(f0 ) = , in R3 .
(16.71)
and
Also,
G4 (2 Curl1 ) = hf0 , 2 Curl 1 iY G4 (f0 ).
(16.72)
Therefore, from (16.58), (16.59), (16.61), (16.67), (16.70) and (16.71)

we obtain
F (div(z ))G1 (v1 +z )G2 (v2 )G3 (v3 2 )G4 (2 Curl 1 )
= G1 (m0 ) + G
2 (m0 ) + G3 (m0 ) + G4 (f0 )
+ Ind1 (m0 , f0 ) + Ind2 (f0 ) hm0i , Hi iY F (m0 ). (16.73)

Remark 16.8.3. The variation relating m in (16.54) give us

div(v1 ) + v2 + v3 H = 0, in ,
(16.74)
372
which, from (16.56), (16.57), (16.60) and (16.65) implies that,

G
2 (m0 )
div(m0 ) (K + K1 )m0 +
m
+(0 + K1 )m0 H + 2 = , in .
(16.75)
Equations (16.75), (16.70), (16.71), are the Euler-Lagrange equations for the relaxed problem. However the derivatives
G
G2 (v2 )
2 (m0 )
and
v
=
m0 =
2
v2
m
must be understood after a regularization, considering that formally
G2 is not Gateaux differentiable. Finally, we conjecture that if K >
0 and K1 > 0 are big enough (but still satisfying the hypotheses
of last theorem), so that for a minimizing sequence {(mn , fn )} we
have 0 + K1 > 0 and G
2 (mn ) = G2 (mn ) for all n N sufficiently
big, then duality gap between the original primal problem and the
dual one is zero. A formal proof of such a conjecture is planned for
a future work.
16.9. Conclusion
In this chapter we develop duality principles for models in ferromagnetism met in reference [26], for example . Almost all dual
variational formulations here presented are convex (in fact concave)
either for the uniaxial or cubic cases. The results are obtained
through standard tools of convex analysis. It is important to emphasize that in some situations (specially the hard cases), the minima may not be attained through the primal approaches, so that
the minimizers of the dual formulations reflect the average behavior
of minimizing sequences for the primal problems, as weak cluster
points of such sequences.
CHAPTER 17
Duality Applied to Fluid Mechanics

In this article our first result is a dual variational formulation
for the incompressible two-dimensional Navier-Stokes system. We
establish as a primal formulation, the sum of L2 norm of each of
equations, and thus obtain the dual formulation through the Legendre Transform concept. Now we pass to present the primal
formulation.
Consider S R2 an open, bounded and connected set, whose
the internal boundary is denoted by 0 and, the external one is
denoted by 1 . Denoting by u : S R the field of velocity in
direction x of the Cartesian system (x, y), by v : S R, the
velocity field in the direction y, by p : S R, the pressure field, so
that P = p/, where is the constant fluid density, is the viscosity
coefficient and, g is the gravity constant, the Navier-Stokes PDE
system is expressed by:
2 u ux u vy u x P + gx = 0, in S,
2 v ux v vy v y P + gy = 0, in S,
(17.1)
x u + y v = 0,
in S,
(
u = v = 0,
on 0 ,
(17.2)
u = u , v = 0, P = P , on 1
The primal variational formulation, denoted by J : U R, is
expressed as:

1
2
2
2
J(u) =
kL1 (u)kL2 (S) + kL2 (u)kL2 (S) + kL3 (u)kL2 (S)
(17.3)
2
where u = (u, v, P ) U, and
U = {u H 2 (S) H 2 (S) H 1(S) |
u = v = 0, 0 , and, u = u , v = 0, P = P on 1 }. (17.4)
373
374
17. DUALITY APPLIED TO FLUID MECHANICS
Also
L1 (u) = 2 u ux u vy u x P + gx ,
(17.5)
L2 (u) = 2 v ux v vy v y P + gy
(17.6)
L3 (u) = x u + y v.
(17.7)
and
Clearly we can write
J(u) =
g(u)dS
(17.8)
where u = {i u}, for i {1, ..., 14} or more explicitly

1 u = 2 u, 2 u = u, 3 u = x u,
4 u = v, 5 u = y u, 6 u = x P,
7 u = 2 v, 8 u = u, 9 u = x v,
10 u = v, 11 u = y v, 12 u = y P,
13 u = x u, 14 u = y v.
(17.9)
g(y) = g1 (y) + g2 (y) + g3 (y)
(17.10)
Here
where
1
g1 (y) = (y1 + y2 y3 + y4 y5 + y6 + gx )2 ,
2
(17.11)
1
g2 (y) = (y7 + y8 y9 + y10 y11 + y12 + gy )2 ,
2
(17.12)
1
g3 (y) = (y13 + y14 )2 .
2
(17.13)
Remark 17.1.1. Through the space U derivatives must be

always understood in distributional sense, whereas boundary conditions are related to the sense of traces. Finally, it is worth emphasizing that the results of this article are valid for the special
cases in which the external forces are represented by a gradient,
with exception of the generalized method of lines, which may be
adapted for more general situations.
375

We will apply Definition 8.1.25 to g(y) = g1 (y) + g2 (y) + g3 (y).
First for g1 (y) we have
g1L
(y )
6
X
i=1
where {yi} is such that
yi yi g1 (y),
(17.14)
g1 (y)
i {1, ..., 6},
yi
yi =
that is
y1 = y6 = (y1 + y2 y3 + y4 y5 + y6 + gx ),
y2 = (y1 + y2 y3 + y4 y5 + y6 + gx )y3 = y1y3 ,
y3 = (y1 + y2 y3 + y4 y5 + y6 + gx )y2 = y1y2 ,
y4 = (y1 + y2 y3 + y4 y5 + y6 + gx )y5 = y1y5 ,
y5 = (y1 + y2 y3 + y4 y5 + y6 + gx )y4 = y1y4 .
Inverting such relations we obtain
y2 = y3/y1 ,
y3 = y2/y1 ,
y4 = y5/y1 ,
y5 = y4/y1 ,
y1 + y6 = y1 y2y3 /(y1)2 y4 y5/(y1 )2 gx .
Replacing such expressions in (17.14) we obtain
g1L
(y ) =
y2 y3 y4 y5 (y1 )2
+ +
gx y1,
y1
y1
2
(17.15)
Similarly, we may obtain
g2L
(y ) =

y11 (y7 )2
y8y9 y10
+
+
gy y7 ,
y7
y7
2
(17.16)
and
1 2
) .
g3L
(y ) = (y13
2
so that gL (y ) = g1L
(y ) + g2L (y ) + g3L (y ).
(17.17)
376
Remark 17.2.1. Observe that to solve the system

(GL (v ) + hu, v iU ) =
(17.18)
implies the solution of the Navier-Stokes system.

Here, v = denotes:
and
2 v1 + v2 + v8 + x v3 + y v5 x v13
= 0, in S,
(17.19)
2 v7 + v4 + v10
+ x v9 + y v11
y v13
= 0, in S,
(17.20)
x v1 + y v7 = 0, in S.
(17.21)
Also,
GL (v )
g1L
(v )dS
g2L
(v )dS
g3L
(v )dS
(17.22)
or, more explicitly:

Z
Z
Z
1
v4 v5
v2 v3
dS +
dS +
(v1 )2 dS
GL (v ) =
2 S
S v1
S v1
Z
Z
Z
v
v
v10 v11
8 9
gx v1 dS +
dS +
dS
v7
S
S v7
S
Z
Z
1
2
+
(v ) dS gy v7 dS.
(17.23)
2 S 7
S
17.3. Linear Systems which the Solutions
Solve the Navier-Stokes One
Through the next result we obtain a linear system whose the solution also solves the Navier-Stokes one for a special class of boundary conditions. Specifically for such a result we consider a domain
S with a boundary denoted by
Theorem 17.3.1. A solution of Navier-Stokes system earlier

indicated, that is
(17.24)
u + v = 0,
in S,
x
y
17.3. LINEAR SYSTEMS WHICH THE SOLUTIONS...
~u n = 0,
on ,
where ~u = (u, v), is given by
u = x w0 ,
377
(17.25)
(17.26)
v= w ,
y 0
where w0 is a solution of the equation
in S,
2 w 0 = 0
w0 n = 0,
(17.27)
on .
Proof. We will suppose that the solution may be expressed by

v
v
u = 2 , v = 3 .
v1
v1
It is easy to verify that v2 = x v1 and v3 = y v1 are such that

1 (u) = x L
2 (u)
y L
where
and
1 (u) = 2 u ux u vy u
L
2 (u) = 2 v ux v vy v.
L
Therefore, the variable v1 may be used to solve the continuity
equation, so that
x u + y v = 0
is equivalent to

x v1
y v1
x
+ y
.
v1
v1
That is
2 (ln(v1 )) = 0.
Defining
w0 = ln(v1 ),
the equation of continuity stands for
2 w0 = 0, in S
which with the concerned boundary conditions,

w0 n = 0, on .
gives a solution of the system in question.
(17.28)
378
Remark 17.3.2. To prove such a result we could simply have

replaced (17.26) into (17.24) to obtain
and
where
x F x P + gx = 0, in S,
(17.29)
y F y P + gy = 0, in S,
(17.30)
F = 2 w0 (x w0 )2 /2 (y w0 )2 /2.
Being w0 known as a solution of
2 w0 = 0 in S
with corresponding boundary conditions, it is clear that equations

(17.29) and (17.30) have a solution in P with the proper boundary
conditions above described.
In the final result in this section we work with the full system
of boundary conditions.
Theorem 17.3.3. A solution of the Navier-Stokes system
(17.31)
u + v = 0,
in S,
x
y
on 0 ,
u = v = 0,
(17.32)
u = u , v = 0, P = P , on 1 ,
is given by
u = x w0 + x w1 ,
v = y w0 y w1
if (w0 , w1) is a solution of the system:
2 w0 + xx w1 yy w1 = 0,
xy w1 = 0,
(17.33)
in S,
in S,
(17.34)
17.4. THE METHOD OF LINES FOR THE NAVIER-STOKES SYSTEM 379
u = x w0 + x w1 = 0,
v = y w0 y w1 = 0,
u = x w0 + x w1 = u ,
v = y w0 y w1 = 0,
on 0 ,
on 0 ,
(17.35)
on 1 ,
on 1 .
Proof. It is easy to verify that u = x w0 + x w1 and v =

y w0 y w1 such that xy w1 = 0 are also such that
1 (u) = x L
2 (u)
y L
where
and
1 (u) = 2 u ux u vy u
L
2 (u) = 2 v ux v vy v.
L
On the other hand, for u and v defined above, the equation of
continuity is given by:
2 w0 + xx w1 yy w1 = 0, in S.
Therefore the solution of this last equation simultaneously with

xy w1 = 0, in S, and the boundary conditions
and
u = x w0 + x w1 = 0, v = y w0 y w1 = 0, on 0 ,
u = x w0 + x w1 = u , v = y w0 y w1 = 0, on 1 ,
gives us a solution of the Navier-Stokes system.
17.4. The Method of Lines for the Navier-Stokes System

Similarly as in chapter 13 (in such a chapter details may be
found), we develop the solution for the Navier-Stokes system through
the generalized method of lines. We write the solution for an approximation of the system in question.
In fact we add the term
tt P
to the equation of continuity after the transformation of coordinates. The results are related to the special case 1 = 20 , however
380
it is possible to obtain them for more general situations, specially

for 1 = K0 , where K > 1 is big number, so that 1 simulates
conditions at infinity.
The boundary conditions are
u = v = 0 and P = P0 (x) on 0 ,
and
u = uf (x), v = vf (x) and P = Pf (x), on 1 .
The optimal P0 (x) is the one which minimizes
Z
(x u + y v)2 dS,
S
in order to provide the best approximation for the continuity equation. Finally, we consider g = 0, = 0.08, = 0.082 and = 1.
Units refers to the international system. Here the functions f1 (x)
up to f17 (x) are relating the boundary shape and are determined
through the transformations of coordinates, similarly as obtained
in chapter 13.
We start presenting the lines (for N=10) for the field of velocity
u (here x stands for ):
Line 1
u[1] =0.1uf (x) + 0.562f16 (x)P0 (x) 0.563f16 (x)Pf (x) + 0.033f11 (x)uf (x)
0.206f14 (x)uf (x)2 0.206f16 (x)uf (x)vf (x) 0.281f17 (x)P0 (x)
0.140f17 (x)Pf (x) + 0.045f12 (x)uf (x) 0.0656f15 (x)uf (x)uf (x)
0.103f17 (x)vf (x)uf (x) + 0.008f13 (x)uf (x),

Line 2
0.205f17 (x)vf (x)uf (x) + 0.015f13 (x)uf (x)

Line 3
0.301f17 (x)vf (x)uf (x) + 0.020f13 (x)uf (x)
Line 4
457f17 (x)Pf (x) + 0.120f12 (x)uf (x) 0.242f15 (x)uf (x)uf (x)
0.387f17 (x)vf (x)uf (x) + 0.024f13 (x)uf (x)
Line 5
0.452f17 (x)vf (x)uf (x) + 0.026f13 (x)uf (x)
Line 6
u[6] =0.6uf (x) + 1.499f16 (x)P0 (x) 1.501f16 (x)Pf [x] + 0.080f11 (x)uf (x)
0.487f17 (x)vf (x)uf (x) + 0.026f13 (x)uf (x)
Line 7
0.476f17 (x)vf (x)uf (x) + 0.023f13 (x)uf (x)
Line 8
0.404f17 (x)vf (x)uf (x) + 0.018f13 (x)uf (x)
382
Line 9
0.211f 17(x)Pf (x) + 0.045f12 (x)uf (x) 0.145f15 (x)uf (x)uf (x)
0.252f17 (x)vf (x)uf (x) + 0.010f13 (x)uf (x)
The Line s for the field of velocities v are presented below:

Line 1
v[1] =0.1vf (x) + 0.562f16 (x)P0 (x) 0.563f16 (x)Pf (x) + 0.033f11 (x)vf (x)
0.206f14 (x)vf (x)2 0.206f16 (x)uf (x)vf (x) 0.281f17 (x)P0 (x)
0.140f17 (x)Pf (x) + 0.045f12 (x)uf (x) 0.0656f15 (x)uf (x)vf (x)
0.103f17 (x)vf (x)vf (x) + 0.008f13 (x)vf (x)

Line 2
0.270f17 (x)Pf (x) + 0.080f12 (x)vf (x) 0.130f15 (x)uf (x)vf (x)
0.205f17 (x)vf (x)vf (x) + 0.015f13 (x)vf (x)

Line 3
0.301f17 (x)vf (x)vf (x) + 0.020f13 (x)vf (x)

Line 4
0.387f17 (x)vf (x)vf (x) + 0.024f13 (x)vf (x)
Line 5
0.501f17 (x)Pf (x) + 0.125f12 (x)uf (x) 0.279f15 (x)uf (x)vf (x)
0.452f17 (x)vf (x)vf (x) + 0.026f13 (x)vf (x)

Line 6
v[6] =0.6vf (x) + 1.499f16 (x)P0 (x) 1.501f16 (x)Pf [x] + 0.080f11 (x)vf (x)
0.487f17 (x)vf (x)vf (x) + 0.026f13 (x)vf (x)

Line 7
0.476f17 (x)vf (x)vf (x) + 0.023f13 (x)vf (x)

Line 8
0.404f17 (x)vf (x)vf (x) + 0.018f13 (x)vf (x)

Line 9
0.211f 17(x)Pf (x) + 0.045f12 (x)vf (x) 0.145f15 (x)uf (x)vf (x)
0.252f17 (x)vf (x)vf (x) + 0.010f13 (x)vf (x)
Finally the lines for the field of pressure P is presented below:

Line 1
P [1] = 0.1Pf (x) + 0.9P0 (x) + 7.03f6 (x)uf (x) + 7.03f4 (x)vf (x)
+1.75f7 (x)uf (x) + 1.75f5 (x)vf (x)
384
Line 2
P [2] = 0.2Pf (x) + 0.8P0 (x) + 12.50f6 (x)uf (x) + 12.50f4(x)vf (x)
+3.37f7 (x)uf (x) + 3.37f5 (x)vf (x)
Line 3
+4.72f7 (x)uf (x) + 4.72f5 (x)vf (x)
Line 4
+5.72f7 (x)uf (x) + 5.72f5 (x)vf (x)
Line 5
+6.26f7 (x)uf (x) + 6.26f5 (x)vf (x)
Line 6
+6.28f7 (x)uf (x) + 6.28f5 (x)vf (x)
Line 7
+5.72f7 (x)uf (x) + 5.72f5 (x)vf (x)
Line 8
+5.72f7 (x)uf [x] + 5.72f5 (x)vf (x)
Line 9
P [9] = 0.9Pf (x) + 0.1P0 (x) + 7.04f6 (x)uf (x) + 7.04f4 (x)vf (x)
+2.63f7 (x)uf (x) + 2.63f5 (x)vf (x).
17.5. CONCLUSION
385
17.5. Conclusion
In this chapter we develop a study about Legendre Transform
applied to the two-dimensional incompressible Navier-Stokes system. The final section presents the solution by the generalized
method of lines. The extension of results to R3 , compressible and
time dependent cases is planned for a future work.
CHAPTER 18
Duality Applied to a Beam Model

18.1. Introduction and Statement
of the Primal Formulation
In this article we present an existence result and duality theory
concerning the non-linear beam model proposed in Gao, [23].
The boundary value form of Gaos beam model is represented
by the equation
EIw,xxxx a(w,x )2 w,xx + w,xx = f, in [0, l]
(18.1)
subject to the conditions:

w(0) = w(l) = w,x (0) = w,x (l) = 0,
(18.2)
where w : [0, l] R denotes the field of vertical displacements.

The corresponding primal variational formulation for such a
model, is expressed by the functional J : U R, where:
Z l
Z l
1
a
2
4
2
J(w) =
(EI(w,xx ) + (w,x ) (w,x ) )dx
f wdx
6
0 2
0
(18.3)
Here E denotes the Young Modulus related to a specific mate3
rial, I = bh
for a beam with rectangular cross section (rectangle
12
basis b and height h), a is a constant related to the cross section
area. Furthermore, l denotes the beam length (in fact the beam is
represented by the set [0, l] = {x R | 0 x l}, denotes an
axial compressive load applied to x = l and finally, f (x) denotes
the distributed vertical load.
Also, we define:
U = {w W 2,2 ([0, l]) W 1,4 ([0, l]) |
w(0) = w(l) = 0 = w,x (0) = w,x (l)}. (18.4)
Remark 18.1.1. The boundary conditions refer to a clamped

beam at x=0 and x=l.
387
388
18. DUALITY APPLIED TO A BEAM MODEL
We consider the problem,

Problem P :
To find w0 U, such that J(w0 ) = inf {J(w)} (18.5)

wU
Equation (18.1) stands for the necessary conditions for Problem P.

Remark 18.1.2. From the Sobolev Imbedding Theorem we
have the following result (case A for mp > n):
W j+m,p() W j,q (),
(18.6)
for p q < . For the present case we have m = n = 1, p = 2,

j = 1 and q = 4, which means:
W 2,2 ([0, l]) W 1,4 ([0, l])
(18.7)
W 2,2 ([0, l]) W 1,4 ([0, l]) = W 2,2 ([0, l]).
(18.8)
so that
18.2. Existence and Regularity Results for Problem P

In this section we show the existence of a minimizer for the unconstrained problem P. More specifically, we establish the following
result:
Theorem 18.2.1. Given b, h, a, l, E, R+ and f L2 ([0, l])
there exists at least one w0 U such that
J(w0 ) = inf {J(w)},
wU
(18.9)
where
U = {w W 2,2 ([0, l]) | w(0) = w(l) = 0 = w,x (0) = w,x (l)}
(18.10)
and
J(w) =
Z l
1
a
2
4
2
(EI(w,xx) + (w,x ) (w,x ) )dx
f wdx,
2
6
0
w U. (18.11)
Proof. From Poincare Inequality it is clear that J is coercive,

that is:
18.2. EXISTENCE AND REGULARITY RESULTS FOR Problem P
lim
kwkU +
J(w) = +
389
(18.12)
where
kwkU = kwkW 2,2([0,l]) , w U.
(18.13)
Therefore since J is strongly continuous, there exists R

such that
= inf {J(w)}.
(18.14)
wU
Thus, if {wn }nN is a minimizing sequence (in the sense that

lim J(wn ) = ),
n+
then {kwn kW 2,2 ([0,l]) } and {kwn kW 1,4 ([0,l]) } are bounded sequences in
0
reflexive Banach spaces (see Remark 18.1.2).
Hence, there exists w0 W02,2 ([0, L]) and a subsequence {wnj }
{wn } such that
wnj w0 as j +, weakly in W02,2 ([0, l]).
(18.15)
From the Rellich Kondrachov theorem, up to a subsequence, which

we also denote by {wnj } we have
wnj,x w0,x as j +, strongly in L2 ([0, l]),
(18.16)
Furthermore we have
J(w) = J1 (w)
2
where
J1 (w) =
(w,x )2 dx
(18.17)
a
1
(EI(w,xx )2 + (w,x )4 ) dx
2
6
f w dx.
(18.18)
As J1 is convex and continuous, it is also weakly lower semi-continuous,

so that
lim inf J1 (wnj ) J1 (w0 )
j+
(18.19)
From this and equation (18.16), as {wnj } is also a minimizing sequence, we can conclude that:
= inf {J(w)} = lim inf J(wnj ) J(w0 )
wU
j+
(18.20)
390
which implies
J(w0 ) = = inf {J(w)}.
wU
(18.21)
Remark 18.2.2. We recall that from the Rellich-Kondrachov

Theorem, Part III, for mp > n we have the following compact
imbedding
0 ).
W j+m,p () C j (
(18.22)
In our case consider n = m = 1, p = 2 and j = 1, that is, as
w0 W 2,2 ((0, l)) (here = 0 = (0, 1)) we can conclude that
w0 C 1 ([0, l]), which means that w0 has continuous derivative
in [0, l] (no corners). In fact such a regularity result refers to the
space W 2,2 ([0, l]) as a whole, not only to this solution w0 . To obtain
deeper results concerning regularity, we would need to evaluate the
effect of necessary conditions on the solution w0 .
Now we present a duality principle proven in [6].
a convex G
U R
convex, coercive and lower semi-continuous (l.s.c.) such
U R
that the functional
J(u) = (G )(u) F (1 u) hu, piU
is bounded from below , where : U Y and 1 : U Y are

Then we may write:
inf
z Y
v B (z )
uU
where B (z ) = {v Y such that v 1 z p = 0}

18.3. A Convex Dual Formulation for the Beam Model
Now, we rewrite the primal variational formulation in a standard double-well format, as J : U R, where
Z l
Z l
Z l
2
EI
w,x
2
2
J(w) =
(w,xx ) dx +
(
) dx
f wdx,
2
2
0
0 2
0
(18.23)
18.3. A CONVEX DUAL FORMULATION FOR THE BEAM MODEL 391
where U = W02,2 ([0, l]) and , are appropriate positive real constants. We may also write
J(w) = G(w) F (1w) hw, f iL2([0,l]) ,
(18.24)
where
G(w) =
EI
(w,xx )2 dx +
2
2
K
w,x
(
)2 dx +
2 2
2
K
F (1 w) =
2
l
0
2
w,x
dx,
(18.25)
2
w,x
dx,
(18.26)
where : U Y = L2 ([0, l]; R3 ), 1 : U Y1 = Y1 = L2 ([0, l])

and 2 : U Y2 = L2 ([0, l]), are given by
w = {1 w, 2w, w},
(18.27)
1 w = w,x , 2 w = w,xx .
(18.28)
and

inf {J(w)} sup { inf {F (L z ) G (v , z )}},
wU
v A z Y1
where
1
F (L z ) =
2K
and
G (v , z ) =
GL (v , z )
l
2
(z,x
) dx,
(18.29)
Z
1 l (v1 )2
dx
+ z ) dx +
2 0 v0 + K
0
Z l
Z l
1
2
(v ) dx +
v0 dx, (18.30)
+
2 0 0
0
1
=
2EI
(v2
if v0 + K > 0, a.e. in [0, l], and
A = {v Y | v f = 0},
(18.31)
that is,
A = {v Y | v2,xx v1,x = f, a.e. in [0, l]}.
(18.32)
392
The final format of the concerned duality principle is given by,

Z l
Z l
1
1
2
inf {J(w)} sup { inf {
(z,x ) dx
(v +z )2 dx
wU
2EI 0 2
v A B z Y0 2K 0
Z
Z l
Z l
1
1 l (v1 )2
2
(v ) dx +
dx
v0 dx} (18.33)
2 0 v0 + K
2 0 0
0
where
B = {v Y | v0 + K > 0 a.e. in [0, l]}
and
Y0 = {z Y1 | z (0) = z (l) = 0}.
Remark 18.3.1. It is important to emphasize that the inequality indicated in (18.33) is an equality if there exists a critical
point for the dual formulation such that v0 + K > 0 a.e. in [0, l]
and K < EI/K0 , where, as above mentioned, K0 is the constant
concerning the Poincare inequality. In such a case, the dual formulation is convex.
18.4. A Final Result, Another Duality Principle
inf {J(w)} = inf sup {F (z ) G (v )},
wU
z Y
(18.34)
v A
where
1
F (z ) =
2K
1
G (v ) =
2EI
(v2 )2 dx
(z )2 dx,
(18.35)
and
(v1 )2
dx
0 v0 + K
Z l
Z l
1
2
+
(v0 ) dx +
v0 dx, (18.36)
2 0
0
1
+
2
if v0 + K > 0, a.e. in [0, l], and

or
A = {v Y | v 1 z f = 0},
A = {(v , z ) L2 ([0, l], R4 ) |
(18.37)
= f, in [0, l]}. (18.38)

v2,xx v1,x + z,x
18.4.
A FINAL RESULT, ANOTHER DUALITY PRINCIPLE
393
Observe that
G (v ) hw, v iY G(w), w U, v Y .
(18.39)
Thus
F (z ) + G (v )
F (z ) + h1 w, z i + hw, f iU G(w). (18.40)
We can make z an independent variable though A , that is, for

v2 (z, v1 ) given by
Z x
Z x

v2 (z , v1 ) = (v2 ) (0)x + v2 (0) +

v1 (t) dt
z (t) dt
0
0
Z x Z t1
+
f dt dt1 . (18.41)
0
From (18.40), we may write

sup
{F (z ) + G (v2 (v1 , z ), v1 , v0 )}
z L2 ([0,l])
sup
{F (z ) + h1w, z i + hw, f iU G(w)}, (18.42)
z L2 ([0,1])
so that we may infer that

Z 1
Z 1
1
1
2
(z ) dx
(v2 (z , v1 ))2 dx
inf
sup
L2 ([0,1])
z
2K
2EI
2
(v1 ,v0 )L ([0,l])C

0
0

Z 1
Z 1
Z 1
2
1
1
(v1 )
2
(v ) dx
v0 dx
dx
2 0 v0 + K
2 0 0
0
inf {J(w)}, (18.43)

wU
where
C = {v0 L2 ([0, l]) | v0 + K > 0 a.e. in [0, l]}.
Observe that the infimum for the dual formulation indicated in
(18.43) is attained, for K < EI/K0 (here K0 denotes the constant
concerning Poincare Inequality), through the relation
v2 =
EIz,x
, z (0) = z (l) = 0
K
(18.44)
394
so that the final format of our duality principle is given by

Z 1
Z 1
1
EI
2
(z )2 dx
(z,x ) dx +
2
inf {J(w)}
sup
uU
2K 0
2K 0
(z ,v1 ,v0 )B C

Z 1
Z
Z
1
1
1
1
(v1 )2
2
dx
(v ) dx
v0 dx . (18.45)
2 0 v0 + K
2 0 0
0
Defining Y0 = W01,2 [0, l] L2 ([0, l], R2 ) we have
B = {(z , v1 , v0) Y0 |
EI
z,xxx v1,x + z,x

= f, a.e. in [0, l]}. (18.46)
K
Remark 18.4.1. It is important to emphasize that equality

holds in (18.45) only if there exists a critical point for the dual formulation such that v0 +K > 0 a.e. in [0, l] and K < EI/K0 , where,
as above mentioned, K0 is the constant concerning the Poincare inequality. In such a case, the dual formulation is convex.
18.5. Conclusion
In this chapter we present an existence result and dual variational formulations for the non-convex (in variational format) Gaos
beam model. In both duality principles, the Poincare inequality
plays a fundamental role for the establishment of optimality conditions.
Bibliography
[1] R.A. Adams, Sobolev Spaces, Academic Press, New York (1975).
[2] R.A. Adams and J.F. Fournier, Sobolev Spaces, second edition , Elsevier
(2003).
[3] J.F. Annet, Superconductivity, Superfluids and Condensates, Oxford Master Series in Condensed Matter Physics, Oxford University Press, Reprint
(2010)
[4] G. Bachman and L. Narici, Functional Analysis, Dover Publications,
Reprint (2000).
[5] J.M. Ball and R.D. James, Fine mixtures as minimizers of energy , Archive
for Rational Mechanics and Analysis, 100, 15-52 (1987).
[6] F. Botelho, Dual Variational Formulations for a Non-linear Model of
Plates, Journal of Convex Analysis, 17 , No. 1, 131-158 (2010).
[7] F. Botelho, Duality for Ginzburg-Landau Type Equations and the Generalized Method of Lines, submitted.
[8] F. Botelho, Dual Variational Formulations for Models in Micro-magnetism
- The Uniaxial and Cubic Cases, submitted.
[9] H.Brezis, Analyse Fonctionnelle, Masson (1987).
[10] I.V. Chenchiah and K. Bhattacharya, The Relaxation of Two-well Energies
with Possibly Unequal Moduli, Arch. Rational Mech. Anal. 187 (2008), 409479.
[11] M. Chipot, Approximation and oscillations, Microstructure and Phase
Transition, the IMA volumes in mathematics and applications, 54, 2738 (1993).
[12] R. Choksi, M.A. Peletier, and J.F. Williams, On the Phase Diagram for
Microphase Separation of Diblock Copolymers: an Approach via a Nonlocal
Cahn-Hilliard Functional, to appear in SIAM J. Appl. Math. (2009).
[13] P.Ciarlet, Mathematical Elasticity, Vol. I Three Dimensional Elasticity,
North Holland Elsivier (1988).
[14] P.Ciarlet, Mathematical Elasticity, Vol. II Theory of Plates, North Holland Elsevier (1997).
[15] P.Ciarlet, Mathematical Elasticity, Vol. III Theory of Shells, North Holland Elsevier (2000).
[16] B. Dacorogna, Direct methods in the Calculus of Variations, SpringerVerlag (1989).
[17] I.Ekeland and R.Temam, Convex Analysis and Variational Problems.
North Holland (1976).
[18] L. C. Evans, Partial Differential Equations, Graduate Studies in Mathematics, 19, AMS (1998).
395
396
BIBLIOGRAPHY
[19] U.Fidalgo and P.Pedregal, A General Lower Bound for the Relaxation of
an Optimal Design Problem in Conductivity with a Quadratic Cost Functional, Pre-print (2007).
[20] N.B. Firoozye and R.V. Khon, Geometric Parameters and the Relaxation
for Multiwell Energies, Microstructure and Phase Transition, the IMA
volumes in mathematics and applications, 54, 85-110 (1993).
[21] D.Y.Gao and G.Strang, Geometric Nonlinearity: Potential Energy, Complementary Energy and the Gap Function, Quartely Journal of Applied
Mathematics, 47, 487-504 (1989a).
[22] D.Y.Gao, On the extreme variational principles for non-linear elastic
plates. Quarterly of Applied Mathematics, XLVIII, No. 2 (June 1990),
361-370.
[23] D.Y.Gao, Finite Deformation Beam Models and Triality Theory in Dynamical Post-Buckling Analysis. International Journal of Non-linear Mechanics 35 (2000), 103-131.
[24] D.Y.Gao, Duality Principles in Nonconvex Systems, Theory, Methods and
Applications, Kluwer, Dordrecht,(2000).
[25] R.D.James and D. Kinderlehrer Frustration in ferromagnetic materials,
Continuum Mech. Thermodyn. 2 (1990) 215-239.
[26] H. Kronmuller and M. Fahnle, Micromagnetism and the Microstructure of
Ferromagnetic Solids, Cambridge University Press (2003).
[27] D.G. Luenberger, Optimization by Vector Space Methods, John Wiley and
Sons, Inc. (1969).
[28] G.W. Milton, Theory of composites, Cambridge Monographs on Applied
and Computational Mathematics. Cambridge University Press, Cambridge
(2002).
[29] P.Pedregal, Parametrized measures and variational principles, Progress in
Nonlinear Differential Equations and Their Applications, 30, Birkhauser
(1997).
[30] P.Pedregal and B.Yan, On Two Dimensional Ferromagnetism, pre-print
(2007).
[31] M. Reed and B. Simon, Methods of Modern Mathematical Physics, Volume
I, Functional Analysis, Reprint Elsevier (Singapore, 2003).
[32] R. T. Rockafellar, Convex Analysis, Princeton Univ. Press, (1970).
[33] R.C. Rogers, A nonlocal model of the exchange energy in ferromagnet materials, Journal of Integral Equations and Applications, 3, No. 1 (Winter
1991).
[34] R.C. Rogers, Nonlocal variational problems in nonlinear electromagnetoelastostatics, SIAM J. Math, Anal. 19, No. 6 (November 1988).
[35] W. Rudin, Functional Analysis, second edition, McGraw-Hill (1991).
[36] W. Rudin, Real and Complex Analysis, third edition, McGraw-Hill (1987).
[37] D.R.S.Talbot and J.R.Willis, Bounds for the effective contitutive relation
of a nonlinear composite, Proc. R. Soc. Lond. (2004), 460, 2705-2723.
[38] E. M. Stein and R. Shakarchi, Real Analysis, Princeton Lectures in Analysis III, Princeton University Press (2005).
[39] R. Temam, Navier-Stokes Equations, AMS Chelsea, reprint (2001).
BIBLIOGRAPHY
397
[40] J.J. Telega, On the complementary energy principle in non-linear elasticity. Part I: Von Karman plates and three dimensional solids, C.R. Acad.
Sci. Paris, Serie II, 308, 1193-1198; Part II: Linear elastic solid and nonconvex boundary condition. Minimax approach, ibid, pp. 1313-1317 (1989)
[41] A.Galka and J.J.Telega Duality and the complementary energy principle
for a class of geometrically non-linear structures. Part I. Five parameter
shell model; Part II. Anomalous dual variational priciples for compressed
elastic beams, Arch. Mech. 47 (1995) 677-698, 699-724.
[42] J.F. Toland, A duality principle for non-convex optimisation and the calculus of variations, Arch. Rath. Mech. Anal., 71, No. 1 (1979), 41-61.
[43] J. L. Troutman, Variational Calculus and Optimal Control, second edition,
Springer (1996).

Book 2011091002

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Book 2011091002

Enviado por

Direitos autorais:

Formatos disponíveis

TOPICS ON FUNCTIONAL

1991 Mathematics Subject Classification. 46N10, 49N15

Author: Fabio Botelho

Basic Functional Analysis

Chapter 1. Topological Vector Spaces

Chapter 2. The Hahn-Banach Theorems and

Chapter 3. Topics on Linear Operators

3.4. The Square Root of a Positive Operator

Chapter 4. Measure and Integration

Chapter 6. The Lebesgue and Sobolev Spaces

Variational Convex Analysis

Chapter 7. Basic Concepts on the Calculus of Variations

Chapter 8. Basic Concepts on Convex Analysis

Convex Sets and Convex Functions

Chapter 9. Constrained Variational Optimization

10. Duality Applied to a Plate Model

11. Duality Applied to Elasticity

12. Duality Applied to a Membrane Shell Model

Chapter 13. Duality Applied to Ginzburg-Landau Type

13.4. A Numerical Example

14. Duality Applied to Conductivity in Composites 317

Chapter 15. Duality Applied to the Optimal

16. Duality Applied to Micro-Magnetism

17. Duality Applied to Fluid Mechanics

Chapter 18. Duality Applied to a Beam Model

18.1. Introduction and Statement

the nonlinear elastic Kirchhoff-Love plate model. The first duality

SUMMARY OF MAIN RESULTS

Basic Functional Analysis

Topological Vector Spaces

1.2. Vector Spaces

1. TOPOLOGICAL VECTOR SPACES

Definition 1.2.3 (Vector Subspace). Let U be a vector space.

Definition 1.2.5 (Topological Spaces). A set U is said to

1.2. VECTOR SPACES

1. TOPOLOGICAL VECTOR SPACES

Definition 1.2.16 (Continuity at a Point). A function f :

Proof. Suppose f 1 (V) is open whenever V V is open. Pick

Definition 1.2.20 (Base). A collection is said to be a

1.2. VECTOR SPACES

The continuity of such functions is a straightforward consequence

Proposition 1.2.24. Given a topological vector space (U, ),

1. TOPOLOGICAL VECTOR SPACES

such that Vu . Thus, we may write

1.3. Some Properties of Topological Vector Spaces

define U = V1 V2 (V1 ) (V2 ), thus U = U (symmetric) and

Definition 1.3.2 (Bounded Sets). A set A U is said to

1.3. SOME PROPERTIES OF TOPOLOGICAL VECTOR SPACES

rA = ||=1 rU = ||=1 rU.

As 0 = (where U denotes the zero vector) from

Now will show that A is balanced and complete the proof.

1. TOPOLOGICAL VECTOR SPACES

Proof. Observe that u+V is a neighborhood of 1u, then by the

( 1)u V, where | 1| < r,

which completes the proof.

Corollary 1.3.8. Let V be a neighborhood of zero in U, if

there exists n0 N such that if n n0 then n <

Definition 1.3.10 (Convergence in topological vector spaces).

1.4. COMPACTNESS IN TOPOLOGICAL VECTOR SPACES

1.4. Compactness in Topological Vector Spaces

1. TOPOLOGICAL VECTOR SPACES

Proof. Let B an infinite subset of A compact and suppose

d(u + w, v + w) = d(u, v), u, v, w U.

In this case we say that the metric is induced by the norm.