Functional Analysis and Applied Optimization in Banach Spaces

Fabio Botelho
Functional Analysis
and Applied
Optimization in
Banach Spaces
Applications to Non-Convex Variational
Models
Functional Analysis and Applied Optimization
in Banach Spaces
Fabio Botelho
Functional Analysis
and Applied Optimization
in Banach Spaces
Applications to Non-Convex Variational
Models
With Contributions by Anderson Ferreira

and Alexandre Molter
123
Fabio Botelho
Department of Mathematics and Statistics
Federal University of Pelotas
Pelotas, RS-Brazil
ISBN 978-3-319-06073-6 ISBN 978-3-319-06074-3 (eBook)

DOI 10.1007/978-3-319-06074-3
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014938037
Mathematics Subject Classification: 46N10, 46E15, 46N50, 49J40, 49K20
© Springer International Publishing Switzerland 2014

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of pub-
lication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
The first objective of this work is to present, to some extent, a deep introduction
to the basic concepts on real and functional analysis.
In principle, the text is written for applied mathematicians and postgraduate stu-
dents in applied mathematics, with interest in applications of functional analysis,
calculus of variations, and optimization to problems in physics and engineering.
However, engineers, physicists, and other professionals in related areas may
find the text very interesting by the possibility of background development towards
graduate-level mathematics applicable in their respective work fields.
We have proven almost all results presented. The proofs are rigorous, but we
believe are almost all very clear and relatively easy to read, even at the most complex
text parts.
The material presented in Parts I and II concerns standard real and functional
analysis. Hence in these two parts the results in general are not new, with the excep-
tion of some sections on domains of class Ĉ1 and relating Sobolev spaces and some
sections about Lagrange multiplier results and the basic theorem about relaxation
for the scalar case, where we show a different proof concerning the original one
in the book Convex Analysis and Variational Problems (indeed such a book is the
theoretical base of the present work) by Ekeland and Témam’s.
About the basic part, specifically Chaps. 1–3 correspond to standard functional
analysis. In Chaps. 4–6 we present basic and advanced concepts in measure and in-
tegration which will be relevant in subsequent results (in fact perhaps a little more
than the minimum necessary). Moreover, Chaps. 7 and 8 correspond to a basic expo-
sition on Sobolev spaces and again, the fundamental results presented are relevant
for subsequent developments. In Chaps. 9–11 we introduce some basic and more
advanced concepts on calculus of variations, convex analysis, and optimization.
Finally, the applications presented in Chaps. 12–23 correspond to the work of
the present author along the last years, and almost all results including the applica-
tions of duality for micro-magnetism, composites in elasticity, and conductivity and
phase transitions are extensions and natural developments of prior ones presented
in the author’s Ph.D. thesis at Virginia Tech, USA, and the previous book Topics
on Functional Analysis, Calculus of Variations and Duality published by Academic
v
vi Preface
Publications. The present book overlaps to some extent with the previous one just
on a part concerning standard mathematics. The applications in the present one are
almost all new developments.
Anyway, a key feature of the present work is that while all problems studied
here are nonlinear 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 du-
ality 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 and vectorial examples in phase
transitions. In Chap. 15 we present the generalized method of lines, a numerical pro-
cedure in which the solution of the partial differential equation in question is written
on lines as functions of boundary conditions and boundary shape. In Chap. 22 we
develop some examples concerning the Navier–Stokes system.
Summary of Main Results
The main results of this work are summarized as follows.
Duality Applied to Elasticity
Chapter 12 develops duality for a model in finite elasticity. The dual formulations
obtained allow the matrix of stresses to be nonpositive or nonnegative 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). The results are based on standard tools of
convex analysis and the concept of the Legendre transform.
Duality Applied to a Plate Model
Chapter 13 develops dual variational formulations for the two-dimensional equa-

tions of the nonlinear elastic Kirchhoff–Love plate model. We obtain a convex dual
variational formulation which allows nonpositive definite membrane forces. In the
third section, similar to the triality criterion introduced in [36], we obtain sufficient
conditions of optimality for the present case. Again the results are based on the fun-
damental tools of convex analysis and the Legendre transform, which can easily be
analytically expressed for the model in question.
Preface vii
Duality Applied to Ginzburg–Landau-Type Equations
Chapters 14–16 are concerned with existence theory and the development of dual
variational formulations for Ginzburg–Landau-type equations. Since the primal for-
mulations 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 formula-
tion for the simpler real case. For such a formulation optimality conditions are also
established.
Duality Applied to Multi-well Variational Problems
The main focus of Chaps. 17 and 18 is the development of dual variational formu-
lations for multi-well optimization problems in phase transitions, conductivity, and
elasticity. 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 for a Model in Quantum Mechanics
In Chap. 19 we develop a duality principle and computation for a class of nonlin-

ear eigenvalue problems found in quantum mechanics models. We present numeri-
cal results for one- and two-dimensional problems. We highlight that this chapter is
coauthored by myself and my colleague Professor Anderson Ferreira.
Duality Applied to the Optimal Design in Elasticity
The first part of Chap. 20 develops a dual variational formulation for the optimal
design of a plate of variable thickness. The design variable, namely the plate thick-
ness, 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. Finally, we highlight that this
chapter is coauthored by myself and my colleague Professor Alexandre Molter.
viii Preface
Duality Applied to Micro-magnetism
The main focus of Chap. 21 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 cases. It is important to emphasize
that the 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 Chap. 22 we develop approximate solutions for the incompressible Navier–

Stokes system through the generalized method of lines. We also obtain a linear
system whose solution solves the steady-state incompressible Euler equations.
Duality Applied to the Optimal Control and Optimal Design

of a Beam Model
Chapter 23 develops duality for the optimal control and design of a beam model.
We emphasize the dual formulation is useful to obtain numerical results. Finally,
numerical examples of optimal design are provided, concerning the maximization
of buckling load and fundamental frequency, respectively.
Acknowledgments
This monograph is based on 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 an advanced level. I am also grateful to all the professors
who 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 ITA (Instituto Tecnológico de
Aeronáutica, SP-Brasil), my undergraduate and masters school, and Virginia Tech,
Preface ix
USA, two wonderful institutions in the American continent, forever in my heart

and memories. Specifically about ITA, among many others, I would like to express
my gratitude to Professors Leo H. Amaral and Tânia Rabello and my master thesis
advisor Antônio Marmo de Oliveira also for their valuable work.
Pelotas, Brazil Fabio Botelho

Contents
Part I Basic Functional Analysis
1 Topological Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Some Properties of Topological Vector Spaces . . . . . . . . . . . . . . . . . . 7
1.4 Compactness in Topological Vector Spaces . . . . . . . . . . . . . . . . . . . . . 9
1.5 Normed and Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Compactness in Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 The Arzela–Ascoli Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.8 Linear Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.9 Linearity and Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.10 Continuity of Operators on Banach Spaces . . . . . . . . . . . . . . . . . . . . . 23
1.11 Some Classical Results on Banach Spaces . . . . . . . . . . . . . . . . . . . . . . 24
1.12 Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.13 Orthonormal Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.13.1 The Gram–Schmidt Orthonormalization . . . . . . . . . . . . . . . . . 39
2 The Hahn–Banach Theorems and Weak Topologies . . . . . . . . . . . . . . . . 41

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2 The Hahn–Banach Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3 Weak Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.4 The Weak-Star Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.5 Weak-Star Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.6 Separable Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7 Uniformly Convex Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3 Topics on Linear Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.1 Topologies for Bounded Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2 Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 Compact Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
xi
xii Contents
3.4 The Square Root of a Positive Operator . . . . . . . . . . . . . . . . . . . . . . . . 66

3.5 About the Spectrum of a Linear Operator . . . . . . . . . . . . . . . . . . . . . . . 73
3.6 The Spectral Theorem for Bounded Self-Adjoint Operators . . . . . . . 77
3.6.1 The Spectral Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.7 The Spectral Decomposition of Unitary Transformations . . . . . . . . . . 86
3.8 Unbounded Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.9 Symmetric and Self-Adjoint Operators . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.9.1 The Spectral Theorem Using Cayley Transform . . . . . . . . . . . 94
4 Basic Results on Measure and Integration . . . . . . . . . . . . . . . . . . . . . . . . 99

4.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.2 Simple Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.4 Integration of Simple Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.5 Signed Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.6 The Radon–Nikodym Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.7 Outer Measure and Measurability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.8 The Fubini Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.8.1 Product Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5 The Lebesgue Measure in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2 Properties of the Outer Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.3 The Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.4 Properties of Measurable Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.5 Lebesgue Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6 Other Topics in Measure and Integration . . . . . . . . . . . . . . . . . . . . . . . . . 147

6.1 Some Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.2 The Riesz Representation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6.3 The Lebesgue Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.3.1 Lebesgue Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.1 Basic Definitions and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.2 Differentiation of Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
7.3 Examples of Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.3.1 First Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.3.2 Second Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.3.3 Third Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Contents xiii
8 The Lebesgue and Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

8.1 Definition and Properties of L p Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 175
8.1.1 Spaces of Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . 180
8.2 The Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.3 The Sobolev Imbedding Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
8.3.1 The Statement of the Sobolev Imbedding Theorem . . . . . . . . 186
8.4 The Proof of the Sobolev Imbedding Theorem . . . . . . . . . . . . . . . . . . 187
8.4.1 Relatively Compact Sets in L p (Ω ) . . . . . . . . . . . . . . . . . . . . . . 191
8.4.2 Some Approximation Results . . . . . . . . . . . . . . . . . . . . . . . . . . 195
8.4.3 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
8.4.4 The Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
8.5 The Trace Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
8.6 Compact Imbeddings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Part II Variational Convex Analysis
9 Basic Concepts on the Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . 225

9.1 Introduction to the Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . 225
9.2 Evaluating the Gâteaux Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
9.3 The Gâteaux Variation: A More General Case . . . . . . . . . . . . . . . . . . . 229
9.4 Fréchet Differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
9.5 Elementary Convexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
9.6 The Legendre–Hadamard Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
9.7 The Weierstrass Condition for n = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
9.8 The Weierstrass Condition: The General Case . . . . . . . . . . . . . . . . . . . 239
9.9 The du Bois–Reymond Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
9.10 The Weierstrass–Erdmann Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 245
9.11 Natural Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
10 Basic Concepts on Convex Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

10.1 Convex Sets and Convex Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
10.1.1 Lower Semicontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
10.2 Duality in Convex Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
10.3 The Min–Max Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
10.4 Relaxation for the Scalar Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
10.5 Duality Suitable for the Vectorial Case . . . . . . . . . . . . . . . . . . . . . . . . . 280
10.5.1 The Ekeland Variational Principle . . . . . . . . . . . . . . . . . . . . . . 280
11 Constrained Variational Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287

11.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
11.2 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
11.3 The Lagrange Multiplier Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
11.4 Some Examples Concerning Inequality Constraints . . . . . . . . . . . . . . 299
11.5 The Lagrange Multiplier Theorem for Equality and Inequality
Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
11.6 Second-Order Necessary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
xiv Contents
11.7 On the Banach Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 307

11.8 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
11.8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
11.8.2 The Main Results About Gâteaux Differentiability . . . . . . . . 311
Part III Applications
12 Duality Applied to Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
12.2 The Main Duality Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
12.3 Other Duality Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
12.4 A Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
12.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
13 Duality Applied to a Plate Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
13.3 Another Duality Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
13.4 An Algorithm for Obtaining Numerical Results . . . . . . . . . . . . . . . . . 353
13.5 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
13.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
14 About Ginzburg–Landau-Type Equations: The Simpler Real

Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
14.1.1 Existence of Solution for the Ginzburg–Landau Equation . . . 364
14.2 A Concave Dual Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . 365
15 The Full Complex Ginzburg–Landau System . . . . . . . . . . . . . . . . . . . . . . 373

15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
15.2 Global Existence for the Ginzburg–Landau System . . . . . . . . . . . . . . 374
15.3 A Related Optimal Control Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 377
15.4 The Generalized Method of Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
15.4.1 About the Approximation Error . . . . . . . . . . . . . . . . . . . . . . . . 386
15.4.2 The Solution of Laplace Equation for a Special Class
of Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
15.5 A First Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
15.6 A Numerical Example Concerning the Optimal Control Problem . . . 390
15.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
16 More on Duality and Computation for the Ginzburg–Landau

System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
16.2 The Duality Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
Contents xv
16.3 On the Numerical Procedures for Ginzburg–Landau-Type

Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
16.4 Numerical Results for Related PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . 403
16.4.1 A Related PDE on a Special Class of Domains . . . . . . . . . . . . 403
16.4.2 About the Matrix Version of GMOL . . . . . . . . . . . . . . . . . . . . 404
16.5 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
16.6 A New Procedure to Obtain Approximate PDE Solutions . . . . . . . . . 412
16.7 Final Results, Newton’s Method for a First-Order System . . . . . . . . . 416
16.7.1 An Example in Nuclear Physics . . . . . . . . . . . . . . . . . . . . . . . . 418
16.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
17 On Duality Principles for Scalar and Vectorial Multi-well

Variational Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
17.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
17.4 The Scalar Multi-well Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
17.4.1 The Primal Variational Formulation . . . . . . . . . . . . . . . . . . . . . 433
17.4.2 A Scalar Multi-well Formulation . . . . . . . . . . . . . . . . . . . . . . . 435
17.5 Duality for a Vectorial Multi-well Model Applicable
to Phase Transition Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
17.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
18 More on Duality Principles for Multi-well Problems . . . . . . . . . . . . . . . . 447

18.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
18.3 Another Duality Principle for Phase Transition Models . . . . . . . . . . . 450
18.4 Duality for a Problem on Conductivity in Composites . . . . . . . . . . . . 451
18.5 Optimal Design and Control for a Plate Model . . . . . . . . . . . . . . . . . . 456
18.5.1 The Duality Principle for the Plate Model . . . . . . . . . . . . . . . . 456
18.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
19 Duality and Computation for Quantum Mechanics Models . . . . . . . . . 465

Fabio Botelho and Anderson Ferreira
19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
19.3 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
19.3.1 The Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
19.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
20 Duality Applied to the Optimal Design in Elasticity . . . . . . . . . . . . . . . . 477

Fabio Botelho and Alexandre Molter
20.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
20.2 On the Duality Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
xvi Contents
20.4 Another Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486

20.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
21 Duality Applied to Micro-Magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493

21.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
21.2 Summary of Results for the Hard Uniaxial Case . . . . . . . . . . . . . . . . . 494
21.3 The Duality Principle for the Hard Case . . . . . . . . . . . . . . . . . . . . . . . . 496
21.4 The Semi-Linear Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
21.5 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 510
21.5.1 First Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
21.5.2 Second Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
21.6 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
21.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
22 The Generalized Method of Lines Applied to Fluid Mechanics . . . . . . 517

22.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
22.2 On the Solution of Steady-State Euler Equation . . . . . . . . . . . . . . . . . 517
22.3 The Generalized Method of Lines for the Navier–Stokes System . . . 519
22.3.1 The General Case for Specific Boundary Conditions . . . . . . . 520
22.3.2 A Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
22.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 530
23 Duality Applied to the Optimal Control and Optimal Design

of a Beam Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
23.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
23.3 Some Closely Related Simpler Examples with Numerical
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
23.3.1 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
23.3.2 A Dynamical Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
23.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
Acronyms
N The set of natural numbers

R The real set
Rn The n-dimensional real set
∞ Infinity symbol
R R ∪ {+∞}
Ω A subset of Rn
∂Ω The boundary of Ω
C The complex set
U,V,Y Banach spaces
U∗ Dual space of U
·, ·U : U × U ∗ → R Duality pairing between U and U ∗
H Hilbert space
(·, ·)H Inner product in a Hilbert space H
A :U →Y Operator whose domain is U and co-domain is Y
A∗ : Y ∗ → U ∗ Adjoint operator relating A
D(A) Domain of operator A
N(A) Null space of operator A
R(A) Range of operator A
F :U →R Functional whose domain is U
F∗ : U∗ → R Fenchel conjugate functional of F
F (u), δ F(u), ∂ F(u)
∂u Notation for the Gâteaux derivative of F, at u ∈ U
δ F(u; ϕ ) The Gâteaux variation of F at u ∈ U relating the
direction ϕ
∂ F(u) The set of sub-gradients of F at u ∈ U
|·| Absolute value for real or complex numbers
· U Norm, for vectors in a normed space U
d : U × U → [0, +∞) Metric in a metric space U
0/ The empty set
S⇒T S implies T
S⇔T S if and only if T
∀ For all
xvii
xviii Acronyms
∃ Exists symbol
Cm (Ω ) Space of real functions on Ω that are continuously differ-
entiable up to order m, 0 ≤ m ≤ ∞
Ccm (Ω ) Set of functions in Cm (Ω ) with compact support in Ω
D(Ω ),Cc∞ (Ω ) Set of functions in C∞ (Ω ) with compact support in Ω
L p (Ω ) Space of measurable functions whose p-power of their
absolute values is finite integrable
W m,p (Ω ) Sobolev space {u ∈ L p (Ω ), Dα u ∈ L p (Ω ), |α | ≤ m, 1 ≤
p ≤ ∞}, where derivatives are in the distributional sense
L p (Ω ; Rn ) n-Dimensional L p
W m,p (Ω ; Rn ) n-Dimensional W m,p
Part I
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. The main references for this chapter are the excellent books on functional
analysis: Rudin [58], Bachman and Narici [6], and Reed and Simon [52]. 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 which we will de-
note 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 θ such that u + θ = u, ∀u ∈ U,
4. for each u ∈ U, there exists a unique vector denoted by
−u such that u + (−u) = θ ,
5. α · (β · u) = (α · β ) · u, ∀α , β ∈ F, u ∈ U,
6. α · (u + v) = α · u + α · v, ∀α ∈ F, u, v ∈ U,
7. (α + β ) · u = α · u + β · u, ∀α , β ∈ F, u ∈ U,
8. 1 · u = u, ∀u ∈ U.
F. Botelho, Functional Analysis and Applied Optimization in Banach Spaces: Applications 3

to Non-Convex Variational Models, DOI 10.1007/978-3-319-06074-3 1,
4 1 Topological Vector Spaces
Remark 1.2.2. From now on we may drop the dot (·) in scalar multiplications and
denote α · u simply as α u.
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 = 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 α1 , . . . ., αn ∈ F for which
n
u = ∑ αi 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 σ of subsets of U called a topology in U, for which
the following properties are valid:
1. U ∈ σ ,
2. 0/ ∈ σ ,
3. if A ∈ σ and B ∈ σ , then A ∩ B ∈ σ ,
4. arbitrary unions of elements in σ also belong to σ .
Any A ∈ σ 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 σ by (U, σ ).
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 = {u ∈ U | u ∈ A}.
Remark 1.2.8. For any sets A, B ⊂ U we denote
A \ B = {u ∈ A | u ∈ B}.
Also, when the meaning is clear we may denote A \ B by A − B.

Proposition 1.2.9. For closed sets we have the following properties:
1. U and 0/ 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 0/ is open and U = 0/ c , by Definition 1.2.7, U is closed. Similarly, since U
is open and 0/ = U \ U = U c , 0/ is closed.
2. A, B closed implies that Ac and Bc are open, and by Definition 1.2.5, Ac ∪ Bc is
open, so that A ∩ B = (Ac ∪ Bc )c is closed.
1.2 Vector Spaces 5
3. Consider A = ∩λ ∈L Aλ , where L is a collection of indices and Aλ is closed,

∀λ ∈ L. We may write A = (∪λ ∈L Acλ )c and since Acλ is open ∀λ ∈ L we have,
by Definition 1.2.5, that A is closed.
Definition 1.2.10 (Closure). Given A ⊂ U we define the closure of A, denoted by

Ā, as the intersection of all closed sets that contain A.
Remark 1.2.11. From Proposition 1.2.9 item 3 we have that Ā is the smallest closed
set that contains A, in the sense that if C is closed and A ⊂ C, then Ā ⊂ C.
Definition 1.2.12 (Interior). Given A ⊂ U we define its interior, denoted by A◦ , as

the union of all open sets contained in A.
Remark 1.2.13. It is not difficult to prove that if A is open, then A = A◦ .
Definition 1.2.14 (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 1.2.15. If A ⊂ U is a set such that for each u ∈ A there exists a neigh-
borhood Vu u such that Vu ⊂ A, then A is open.
Proof. This follows from the fact that A = ∪u∈A Vu and any arbitrary union of open
sets is open.
Definition 1.2.16 (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 1.2.17 (Continuity at a Point). A function f : U → V is continuous at

u ∈ U if for each neighborhood V f (u) ⊂ V of f (u), there exists a neighborhood
Vu ⊂ U of u such that f (Vu ) ⊂ V f (u) .
Definition 1.2.18 (Continuous Function). A function f : U → V is continuous if it

is continuous at each u ∈ U.
Proposition 1.2.19. 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 ) = 0, / we
are done, since 0/ 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 1.2.15, f −1 (V ) is open.
Definition 1.2.20. We say that (U, σ ) is a Hausdorff topological space if, given u1 ,
u2 ∈ U, u1 = u2 , there exists V1 , V2 ∈ σ such that
u1 ∈ V1 , u2 ∈ V2 and V1 ∩ V2 = 0.
/ (1.3)
Definition 1.2.21 (Base). A collection σ ⊂ σ is said to be a base for σ if every

element of σ may be represented as a union of elements of σ .
Definition 1.2.22 (Local Base). A collection σ̂ 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 σ̂ .
Definition 1.2.23 (Topological Vector Space). A vector space endowed with a
topology, denoted by (U, σ ), is said to be a topological vector space if and only if
1. every single point of U is a closed set,
2. the vector space operations (addition and scalar multiplication) are continuous
with respect to σ .
More specifically, addition is continuous if given u, v ∈ U and V ∈ σ 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 α ∈ F, u ∈ U and V α · u, there
exists δ > 0 and Vu u such that ∀β ∈ F satisfying |β − α | < δ we have β Vu ⊂ V .
Given (U, σ ), let us associate with each u0 ∈ U and α0 ∈ F (α0 = 0) the functions
Tu0 : U → U and Mα0 : U → U defined by
Tu0 (u) = u0 + u (1.4)
and
Mα0 (u) = α0 · u. (1.5)
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 T−u0 and M1/α0 , are also continuous. So if V is
open, then u0 + V , that is, (T−u0 )−1 (V ) = Tu0 (V ) = u0 + V is open. By analogy
α0 V is open. Thus σ is completely determined by a local base, so that the term local
base will be understood henceforth as a local base at θ . So to summarize, a local
base of a topological vector space is a collection Ω of neighborhoods of θ , such that
each neighborhood of θ contains a member of Ω .
Now we present some simple results.
Proposition 1.2.24. If A ⊂ U is open, then ∀u ∈ A, there exists a neighborhood V
of θ such that u + V ⊂ A.
Proof. Just take V = A − u.
Proposition 1.2.25. Given a topological vector space (U, σ ), any element of σ may
be expressed as a union of translates of members of Ω , so that the local base Ω
generates the topology σ .
1.3 Some Properties of Topological Vector Spaces 7
Proof. Let A ⊂ U open and u ∈ A. V = A − u is a neighborhood of θ and by defi-

nition of local base, there exists a set VΩ u ⊂ V such that VΩu ∈ Ω . Thus, we may
write
A = ∪u∈A (u + VΩu ). (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 = u1 . Thus V = U \ {u1 − u0 } is an open neigh-
borhood of zero. As θ + θ = θ , by the continuity of addition, there exist V1 and V2
neighborhoods of θ 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)
so that
u0 + v1 + v2 = u1 , ∀v1 , v2 ∈ U , (1.9)
or
u0 + v1 = u1 − v2 , ∀v1 , v2 ∈ U , (1.10)
and since U = −U
(u0 + U ) ∩ (u1 + U ) = 0.
/ (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 λ u + (1 − λ )v ∈ A, ∀λ ∈ [0, 1], (1.12)
is said to be convex.
Definition 1.3.4 (Locally Convex Spaces). A topological vector space U is said to
be locally convex if there is a local base Ω whose elements are convex.
Definition 1.3.5 (Balanced Sets). A set A ⊂ U is said to be balanced if α A ⊂ A,

∀α ∈ F such that |α | ≤ 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 multiplica-
tion, there exist V (neighborhood of zero) and δ > 0, such that α V ⊂ U when-
ever |α | < δ . Define W = ∪|α |<δ α V ; thus W ⊂ U is a balanced neighborhood
of zero.
2. Suppose U is a convex neighborhood of zero in U. Define
A = {∩α U | α ∈ C, |α | = 1}. (1.13)
As 0 · θ = θ (where θ ∈ U denotes the zero vector) from the continuity of scalar

multiplication there exists δ > 0 and there is a neighborhood of zero V such that
if |β | < δ , then β V ⊂ U . Define W as the union of all such β V . Thus W is
balanced and α −1 W = W as |α | = 1, so that W = α W ⊂ α 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 we will
show that A◦ is balanced and complete the proof. For this, it suffices to prove that
A is balanced. Choose r and β such that 0 ≤ r ≤ 1 and |β | = 1. Then
rβ A = ∩|α |=1 rβ α U = ∩|α |=1 rα U . (1.14)
Since α U is a convex set that contains zero, we obtain rα U ⊂ α U , so that

rβ A ⊂ 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 β u ∈ V , ∀β such that |β | < r.
Proof. Observe that u + V is a neighborhood of 1 · u, and then by the continuity of

scalar multiplication, there exists W neighborhood of u and r > 0 such that
β W ⊂ u + V , ∀β such that |β − 1| < r, (1.15)
so that
βu ∈ u+V , (1.16)
or
(β − 1)u ∈ V , where |β − 1| < r, (1.17)

1.4 Compactness in Topological Vector Spaces 9
and thus
β̂ u ∈ V , ∀β̂ such that |β̂ | < r, (1.18)
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 U ⊂ ∪∞
n=1 rn V .
n→∞
Proof. Let u ∈ U, then α u ∈ V for any α sufficiently small, from the last proposition
u ∈ α1 V . As rn → ∞ we have that rn > α1 for n sufficiently big, so that u ∈ rn V , which
completes the proof.
Proposition 1.3.9. Suppose {δn } is a sequence such that δn → 0, δn < δn−1 , ∀n ∈ N
and V a bounded neighborhood of zero in U, then {δn 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 δn = 0, there exists n0 ∈ N such that if n ≥ n0 ,
n→∞
then δn < t10 , so that δn V ⊂ U , ∀n such that n ≥ n0 .
Definition 1.3.10 (Convergence in Topological Vector Spaces). Let U be a topo-
logical 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
We start this section with the definition of open covering.

Definition 1.4.1 (Open Covering). Given B ⊂ U we say that {Oα , α ∈ A} is a
covering of B if B ⊂ ∪α ∈A Oα . If Oα is open ∀α ∈ A, then {Oα } 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 subcovering. More explicitly, if B ⊂ ∪α ∈A Oα , where Oα
is open ∀α ∈ A, then there exist α1 , . . . , αn ∈ A such that B ⊂ Oα1 ∪ . . . ∪ Oαn , 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 = 0. / It
is clear that A ⊂ ∪x∈A 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 ) ∩
/ Since ∩ni=1 Oyxi is open and y is an arbitrary point of Ac we have that
(∪ni=1 Oxi ) = 0.
c
A is open, so that A is closed, which completes the proof.
The next result is very useful.
Theorem 1.4.4. Let {Kα , α ∈ L} be a collection of compact subsets of a Hausdorff

topological vector space U, such that the intersection of every finite subcollection
(of {Kα , α ∈ L}) is nonempty.
Under such hypotheses
∩α ∈L Kα = 0.
/
Proof. Fix α0 ∈ L. Suppose, to obtain contradiction, that
∩α ∈L Kα = 0.
/
That is,
Kα0 ∩ [∩αα = α0
∈L Kα ] = 0.
/
Thus,
∩αα = α0
∈L Kα ⊂ Kα0 ,
c
so that
Kα0 ⊂ [∩αα = α0
∈L Kα ] ,
c
α =α
Kα0 ⊂ [∪α ∈L 0 Kαc ].
However, Kα0 is compact and Kαc is open, ∀α ∈ L.
Hence, there exist α1 , . . . , αn ∈ L such that
Kα0 ⊂ ∪ni=1 Kαc i .
From this we may infer that
Kα0 ∩ [∩ni=1 Kαi ] = 0,

/
which contradicts the hypotheses.

The proof is complete.
Proposition 1.4.5. A closed subset of a compact space U is compact.
Proof. Consider {Oα , α ∈ L} an open cover of A. Thus {Ac , Oα , α ∈ L} is a cover
of U. As U is compact, there exist α1 , α2 , . . . , αn such that Ac ∪ (∪ni=1 Oαi ) ⊃ U, so
that {Oαi , i ∈ {1, . . . , n}} covers A, so that A is compact. The proof is complete.
Definition 1.4.6 (Countably Compact Sets). A set A is said to be countably com-
pact if every infinite subset of A has a limit point in A.
Proposition 1.4.7. Every compact subset of a topological space U is countably
compact.
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 = 0. / 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 11
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 pos-
sible to define a function · U : U → R+ = [0, +∞), called a norm, which satisfies
the following properties:
1. u U > 0, if u = θ and u U = 0 ⇔ u = θ ,
2. u + v U ≤ u U + v U , ∀ u, v ∈ U,
3. α u U = |α | u U , ∀u ∈ U, α ∈ 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) = u − v U . (1.19)
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
d(u + w, v + w) = d(u, v), ∀u, v, w ∈ U. (1.20)
The following are some basic definitions concerning metric and normed spaces:
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 ε > 0, there exists n0 ∈ N,
such that if n ≥ n0 , then d(un , u0 ) < ε . 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 ε > 0 there exists n0 ∈ N such that d(un , um ) < ε , ∀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 (Limit Point). Let (U, d) be a metric space and let E ⊂ U. We say
that v ∈ U is a limit point of E if for each r > 0 there exists w ∈ Br (v) ∩ E such that
w = v.
Definition 1.5.7 (Interior Point, Topology for (U, d)). Let (U, d) be a metric space
and let E ⊂ U. We say that u ∈ E is interior point if there exists r > 0 such that
Br (u) ⊂ E. We may define a topology for a metric space (U, d) by declaring as
open all set E ⊂ U such that all its points are interior. Such a topology is said to be
induced by the metric d.
Definition 1.5.8. Let (U, d) be a metric space. The set σ of all open sets, defined
through the last definition, is indeed a topology for (U, d).
Proof.
1. Obviously 0/ and U are open sets.
2. Assume A and B are open sets and define C = A ∩ B. Let u ∈ C = A ∩ B; thus,
from u ∈ A, there exists r1 > 0 such that Br1 (u) ⊂ A. Similarly from u ∈ B there
exists r2 > 0 such that Br2 (u) ⊂ B.
Define r = min{r1 , r2 }. Thus, Br (u) ⊂ A ∩ B = C, so that u is an interior point of
C. Since u ∈ C is arbitrary, we may conclude that C is open.
3. Suppose {Aα , α ∈ L} is a collection of open sets. Define E = ∪α ∈L Aα , and we
shall show that E is open.
Choose u ∈ E = ∪α ∈L Aα . Thus there exists α0 ∈ L such that u ∈ Aα0 . Since Aα0
is open there exists r > 0 such that Br (u) ⊂ Aα0 ⊂ ∪α ∈L Aα = E. Hence u is an
interior point of E, since u ∈ E is arbitrary, we may conclude that E = ∪α ∈L Aα
is open.
Definition 1.5.9. Let (U, d) be a metric space and let E ⊂ U. We define E as the set
of all the limit points of E.
Theorem 1.5.10. Let (U, d) be a metric space and let E ⊂ U. Then E is closed if
and only if E ⊂ E.
Proof. Suppose E ⊂ E. Let u ∈ E c ; thus u ∈ E and u ∈ E . Therefore there exists
r > 0 such that Br (u) ∩ E = 0, / so that Br (u) ⊂ E c . Therefore u is an interior point
of E . Since u ∈ E is arbitrary, we may infer that E c is open, so that E = (E c )c is
c c
closed.
Conversely, suppose that E is closed, that is, E c is open.
If E = 0,
/ we are done.
Thus assume E = 0/ and choose u ∈ E . Thus, for each r > 0, there exists v ∈
Br (u) ∩ E such that v = u. Thus Br (u) E c , ∀r > 0 so that u is not a interior point
of E c . Since E c is open, we have that u ∈ E c so that u ∈ E. We have thus obtained,
u ∈ E, ∀u ∈ E , so that E ⊂ E.
Remark 1.5.11. From this last result, we may conclude that in a metric space, E ⊂ U
is closed if and only if E ⊂ E.
Definition 1.5.12 (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.
1.6 Compactness in Metric Spaces 13
Remark 1.5.13. We say that a topology σ is compatible with a metric d if any A ⊂ σ

is represented by unions and/or finite intersections of open balls. In this case we say
that d : U × U → R+ induces the topology σ .
Definition 1.5.14 (Metrizable Spaces). A topological vector space (U, σ ) is said to
be metrizable if σ is compatible with some metric d.
Definition 1.5.15 (Normable Spaces). A topological vector space (U, σ ) is said to
be normable if the induced metric (by this norm) is compatible with σ .
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 ⊃ . . . .
Theorem 1.6.3. If (U, d) is a complete metric space, then every nested sequence of
nonempty closed sets {Fk } such that
lim diam(Fk ) = 0
k→+∞
has nonempty intersection, that is,
∩∞
k=1 Fk = 0.
/
Proof. Suppose {Fk } is a nested sequence and lim diam(Fk ) = 0. For each n ∈ N,
k→∞
select un ∈ Fn . Suppose given ε > 0. Since
lim diam(Fn ) = 0,
n→∞
there exists N ∈ N such that if n ≥ N, then
diam(Fn ) < ε .
Thus if m, n > N we have um , un ∈ FN so that
d(un , um ) < ε .
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 ∈ F̄m = Fm .
Since m ∈ N is arbitrary we obtain
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 ⊂ ∪u∈A B1 (u).
Since A is compact there exists u1 , u2 , . . . , un ∈ A such that
A =⊂ ∪nk=1 B1 (uk ).
Define
R = max{d(ui , u j ) | 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
u ∈ B1 (ui ), v ∈ B1 (u j ).
Thus
d(u, v) ≤ d(u, ui ) + d(ui , u j ) + d(u j , v)

≤ 2 + R, (1.22)
which contradicts (1.21). This completes the proof.

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 (ε -Nets). Let (U, d) be a metric space. A set N ⊂ U is sat to be a
ε -net with respect to a set A ⊂ U if for each u ∈ A there exists v ∈ N such that
d(u, v) < ε .
Definition 1.6.7. Let (U, d) be a metric space. A set A ⊂ U is said to be totally

bounded if for each ε > 0, there exists a finite ε -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 , u j ) | i, j ∈ {1, . . . , n}}.
Observe that there exist i, j ∈ {1, . . . , n} such that
d(u, ui ) < 1, d(v, u j ) < 1.
Thus
d(u, v) ≤ d(u, ui ) + d(ui , u j ) + d(u j , v)

≤ R + 2. (1.23)
Since u, v ∈ A are arbitrary, A is bounded.
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 ε0 > 0 such that there exists no ε0 -net with respect to A. Choose u1 ∈ A; hence
{u1 } is not a ε0 -net, that is, there exists u2 ∈ A such that
d(u1 , u2 ) > ε0 .
Again {u1 , u2 } is not a ε0 -net for A, so that there exists u3 ∈ A such that
d(u1 , u3 ) > ε0 and d(u2 , u3 ) > ε0 .
Proceeding in this fashion we can obtain a sequence {un } such that
d(un , um ) > ε0 , if m = n. (1.24)
Clearly we cannot extract a convergent subsequence of {un }; otherwise such a sub-

sequence would be Cauchy contradicting (1.24). The proof is complete.
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 1.4.7 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 com-
pact, {un } has a limit point in A, 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 {Gα , α ∈ L} is an
open cover of A. For each u ∈ A define
δ (u) = sup{r | Br (u) ⊂ Gα , for some α ∈ L}.
First we prove that δ (u) > 0, ∀u ∈ A. Choose u ∈ A. Since A ⊂ ∪α ∈L Gα , there

exists α0 ∈ L such that u ∈ Gα0 . Being Gα0 open, there exists r0 > 0 such that
Br0 (u) ⊂ Gα0 .
Thus,
δ (u) ≥ r0 > 0.
Now define δ0 by
δ0 = inf{δ (u) | u ∈ A}.
Therefore, there exists a sequence {un } ⊂ A such that
δ (un ) → δ0 as n → ∞.
Since A is sequentially compact, we may obtain a subsequence {unk } and u0 ∈ A

such that
δ (unk ) → δ0 and unk → u0 ,
as k → ∞. Therefore, we may find K0 ∈ N such that if k > K0 , then
δ (u0 )
d(unk , u0 ) < . (1.25)
4
We claim that
δ (u0 )
δ (unk ) ≥ , if k > K0 .
4
To prove the claim, suppose
z ∈ B δ (u0 ) (unk ), ∀k > K0 ,

4
(observe that in particular from (1.25)
u0 ∈ B δ (u0 ) (unk ), ∀k > K0 ).

4
Since
δ (u0 )
< δ (u0 ),
2
there exists some α1 ∈ L such that
B δ (u0 ) (u0 ) ⊂ Gα1 .

2
However, since
δ (u0 )
d(unk , u0 ) < , if k > K0 ,
4
we obtain
B δ (u0 ) (u0 ) ⊃ B δ (u0 ) (unk ), if k > K0 ,
2 4
so that
δ (u0 )
δ (unk ) ≥ , ∀k > K0 .
4
Therefore
δ (u0 )
lim δ (unk ) = δ0 ≥ .
k→∞ 4
Choose ε > 0 such that
δ0 > ε > 0.
From the last theorem since A is sequentially compact, it is totally bounded. For the
ε > 0 chosen above, consider an ε -net contained in A (the fact that the ε -net may be
chosen contained in A is also a consequence of the last theorem) and denote it by N
that is,
N = {v1 , . . . , vn } ∈ A.
Since δ0 > ε , there exists
α1 , . . . , αn ∈ L
such that
Bε (vi ) ⊂ Gαi , ∀i ∈ {1, . . . , n},
considering that
δ (vi ) ≥ δ0 > ε > 0, ∀i ∈ {1, . . . , n}.
For u ∈ A, since N is an ε -net we have
u ∈ ∪ni=1 Bε (vi ) ⊂ ∪ni=1 Gαi .
Since u ∈ U is arbitrary we obtain
A ⊂ ∪ni=1 Gαi .
Thus
{Gα1 , . . . , Gαn }
is a finite subcover for A of
{Gα , α ∈ L}.
Hence, A is compact.
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 sub-
sequence. 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 totally 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
εk -net where εk = 1/k, denoted by Nk where
(k) (k) (k)
Nk = {v1 , v2 , . . . , vnk }.
1.7 The Arzela–Ascoli Theorem 19
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 many points of {un }. Let us select a subse-
(1)
quence {unk }k∈N of this infinite set (which is contained in a ball of radius 1).
(2)
Similarly, we may select a subsequence here just partially relabeled {unl }l∈N
(1)
of {unk } which is contained in one of the balls of the 12 -net. Proceeding in this
(k)
fashion for each k ∈ N we may find a subsequence denoted by {unm }m∈N of the
original sequence contained in a ball of radius 1/k.
(k)
Now consider the diagonal sequence denoted by {unk }k∈N = {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 (Equicontinuity). Let F be a collection of complex functions de-
fined on a metric space (U, d). We say that F is equicontinuous if for each ε > 0,
there exists δ > 0 such that if u, v ∈ U and d(u, v) < δ , then
| f (u) − f (v)| < ε , ∀ 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 equicontinu-

ous collection of complex functions defined on a metric space (U, d). Also suppose
that U has a countable dense subset E. Thus, each sequence { fn } ⊂ F has a sub-
sequence 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 ) = f˜1,k (u1 ), ∀k ∈ N.
Thus there exists g1 ∈ C such that
f˜1,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 { f˜2,k (u2 )}. Again there exists g2 ∈ C
such that
f˜2,k (u1 ) → g1 , as k → ∞.
f˜2,k (u2 ) → g2 , as k → ∞.
Proceeding in this fashion for each m ∈ N we may obtain { f˜m,k } such that
f˜m,k (u j ) → g j , as k → ∞, ∀ j ∈ {1, . . . , m},
where the set {g1 , g2 , . . . , gm } is obtained as above. Consider the diagonal sequence
{ f˜k,k },
and observe that the sequence
{ f˜k,k (um )}k>m
is such that
f˜k,k (um ) → gm ∈ C, as k → ∞, ∀m ∈ N.
Therefore we may conclude that from { fn } we may extract a subsequence also de-
noted by
{ fnk } = { f˜k,k }
which is convergent in
E = {un }n∈N .
Now suppose K ⊂ U, being K compact. Suppose given ε > 0. From the equiconti-
nuity hypothesis there exists δ > 0 such that if u, v ∈ U and d(u, v) < δ we have
ε
| fnk (u) − fnk (v)| < , ∀k ∈ N.
3
Observe that
K ⊂ ∪u∈K B δ (u),
2
and being K compact we may find {ũ1 , . . . , ũM } such that
K ⊂ ∪M
j=1 B δ (ũ j ).
2
Since E is dense in U, there exists
v j ∈ B δ (ũ j ) ∩ E, ∀ j ∈ {1, . . . , M}.

2
1.8 Linear Mappings 21
Fixing j ∈ {1, . . . , M}, from v j ∈ E we obtain that
lim fnk (v j )
k→∞
exists as k → ∞. Hence there exists K0 j ∈ N such that if k, l > K0 j , then
ε
| fnk (v j ) − fnl (v j )| < .
3
Pick u ∈ K; thus
u ∈ B δ (ũ jˆ)
2
for some jˆ ∈ {1, . . . , M}, so that
d(u, v jˆ) < δ .
Therefore if
k, l > max{K01 , . . . , K0M },
then
| fnk (u) − fnl (u)| ≤ | fnk (u) − fnk (v jˆ)| + | fnk (v jˆ) − fnl (v jˆ)|
+| fnl (v jˆ) − fnl (u)|
ε ε ε
≤ + + = ε. (1.26)
3 3 3
Since u ∈ K is arbitrary, we conclude that { fnk } is uniformly Cauchy on K.
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 (α u + β v) = α f (u) + β f (v), ∀u, v ∈ U, α , β ∈ 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) = θ } (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 and V , respectively.

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

where.
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 sat-
isfied: whenever {un } ⊂ E and {αn } ⊂ F are such that αn → 0 as n → ∞ we have
αn un → θ as n → ∞.
Proof. Suppose E is bounded. Let U be a balanced neighborhood of θ in U and
then E ⊂ tU for some t. For {un } ⊂ E, as αn → 0, there exists N such that if n > N,
then t < |α1n | . Since t −1 E ⊂ U and U is balanced, we have that αn un ∈ U , ∀n > N,
and thus αn un → θ . Conversely, if E is not bounded, there is a neighborhood V of
θ and {rn } such that rn → ∞ and E is not contained in rn V , that is, we can choose
un such that rn−1 un is not in V , ∀n ∈ N, so that {rn−1 un } does not converge to θ .
Proposition 1.9.3. Let f : U → V be a linear function. Consider the following
statements:
1.10 Continuity of Operators on Banach Spaces 23
1. f is continuous,
2. f is bounded,
3. if un → θ , then { f (un )} is bounded,
4. if un → θ , then f (un ) → θ .
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)
If E is bounded, there exists t0 ∈ R+ such that E ⊂ tV , ∀t ≥ t0 , so that
f (E) ⊂ f (tV ) = t f (V ) ⊂ tW , ∀t ≥ t0 , (1.36)
and thus f is bounded.
2. 2 implies 3: Suppose un → θ and let W be a neighborhood of zero. Then, there
exists N ∈ N such that if n ≥ N, 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 → θ . Given K ∈ N, there exists
nK ∈ N such that if n > nK , then d(un , θ ) < K12 . Define γn = 1 if n < n1 and
γn = K, if nK ≤ n < nK+1 so that
d(γn un , θ ) = d(Kun , θ ) ≤ Kd(un , θ ) < K −1 . (1.37)
Thus since 2 implies 3 we have that { f (γn un )} is bounded so that, by
Proposition 1.9.2, f (un ) = γn−1 f (γn un ) → θ 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 (θ ) 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
A(u) V < K u U , ∀u ∈ U.
Proof. Suppose A is linear and continuous. From Proposition 1.9.3,
if {un } ⊂ U is such that un → θ then A(un ) → θ . (1.38)
We claim that for each ε > 0 there exists δ > 0 such that if u U < δ , then
A(u) V < ε .
Suppose, to obtain contradiction, that the claim is false.
Thus there exists ε0 > 0 such that for each n ∈ N there exists un ∈ U such that
un U ≤ 1n and A(un ) V ≥ ε0 .
Therefore un → θ and A(un ) does not converge to θ , which contradicts (1.38).
Thus the claim holds.
In particular, for ε = 1, there exists δ > 0 such that if u U < δ , then
A(u) V < 1. Thus given an arbitrary not relabeled u ∈ U, u = θ , for
δu
w=
2 u U
we have
δ A(u) V
A(w) V = < 1,
2 u U
that is
2 u U
A(u) V < , ∀u ∈ U.
δ
Defining
2
K=
δ
the first part of the proof is complete. Reciprocally, suppose there exists K > 0 such
that
A(u) V < K u U , ∀u ∈ U.
Hence un → θ implies A(un ) V → θ , so that from Proposition 1.9.3, A is continu-
ous.
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 = A−1 [{v ∈ V | v V ≤ 1}].
1.11 Some Classical Results on Banach Spaces 25
Proof. Suppose there exists u0 ∈ U in the interior of C. Thus, there exists r > 0 such
that
Br (u0 ) = {u ∈ U | u − u0 U < r} ⊂ C.
Fix u ∈ U such that u U < r. Thus, we have
A(u) V ≤ A(u + u0) V + A(u0) V .
Observe also that

(u + u0) − u0 U < r,
so that u + u0 ∈ Br (u0 ) ⊂ C and thus
A(u + u0) V ≤ 1
and hence
A(u) V ≤ 1 + A(u0) V , (1.39)
∀u ∈ U such that u U < r. Fix an arbitrary not relabeled u ∈ U such that u = θ .

From (1.39)
u r
w=
u U 2
is such that
A(u) V r
A(w) V = ≤ 1 + A(u0) V ,
u U 2
so that
2
A(u) V ≤ (1 + A(u0) V ) u U .
r
Since u ∈ U is arbitrary, A is bounded.
Reciprocally, suppose A is bounded. Thus
A(u) V ≤ K u U , ∀u ∈ U,
for some K > 0. In particular

1
D = u ∈ U | u U ≤ ⊂ C.
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 Ā1 ; otherwise we would have U = Ā1 , which is not possible since
U is open. Furthermore, Āc1 is open, so that we may obtain u1 ∈ Ac1 and 0 < r1 < 1
such that
B1 = Br1 (u1 )
satisfies
B1 ∩ A1 = 0.
/
Since A2 is nowhere dense we have B1 is not contained in Ā2 . Therefore we may
select u2 ∈ B1 \ Ā2 and since B1 \ Ā2 is open, there exists 0 < r2 < 1/2 such that
B̄2 = B̄r2 (u2 ) ⊂ B1 \ Ā2 ,
that is,
B2 ∩ A2 = 0.
/
Proceeding inductively in this fashion, for each n ∈ N, we may obtain un ∈ Bn−1 \ Ān
such that we may choose an open ball Bn = Brn (un ) such that
B̄n ⊂ Bn−1 ,
Bn ∩ An = 0,
/
and
0 < rn < 21−n.
Observe that {un } is a Cauchy sequence, considering that if m, n > N, then un , um ∈
BN , so that
d(un , um ) < 2(21−N ).
Define
u = lim un .
n→∞
Since
un ∈ BN , ∀n > N,
we get
u ∈ B̄N ⊂ BN−1 .
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
T (u) V < Ku , ∀T ∈ F .
Then, there exists K ∈ R such that
T < K, ∀T ∈ F .
Proof. Define
Bn = {u ∈ U | T (u) V ≤ n, ∀T ∈ F }.
By the hypotheses, given u ∈ U, u ∈ Bn for all n is 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 nonempty interior. That is, there exists u0 ∈ U and r > 0 such that
Br (u0 ) ⊂ Bn0 .
Thus, fixing an arbitrary T ∈ F , we have
T (u) V ≤ n0 , ∀u ∈ Br (u0 ).
Thus if u U < r then (u + u0) − u0 U < r, so that
T (u + u0) V ≤ n0 ,
that is,
T (u) V − T (u0 ) V ≤ n0 .
Thus,
T (u) V ≤ 2n0 , if u U < r. (1.40)
For u ∈ U arbitrary, u = θ , define
ru
w= ,
2 u U
from (1.40) we obtain
r T (u) V
T (w) V = ≤ 2n0 ,
2 u U
so that
4n0 u U
T (u) V ≤ , ∀u ∈ U.
r
Hence
4n0
T ≤ , ∀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 .
Proof. First we will prove that given r > 0, there exists r > 0 such that
A(Br (θ )) ⊃ BVr (θ ). (1.41)
Here BVr (θ ) denotes a ball in V of radius r with center in θ . Since A is onto
V = ∪∞
n=1 A(nB1 (θ )).
By the Baire category theorem, there exists n0 ∈ N such that the closure of
A(n0 B1 (θ )) has nonempty interior, so that A(B1 (θ )) has nonempty interior. We
will show that there exists r > 0 such that
BVr (θ ) ⊂ A(B1 (θ )).
Observe that there exists y0 ∈ V and r1 > 0 such that
BVr1 (y0 ) ⊂ A(B1 (θ )). (1.42)
Define u0 ∈ B1 (θ ) which satisfies A(u0 ) = y0 . We claim that
A(Br2 (θ )) ⊃ BVr1 (θ ),
where r2 = 1 + u0 U . To prove the claim, pick
y ∈ A(B1 (θ ))
thus there exists u ∈ U such that u U < 1 and A(u) = y. Therefore
A(u) = A(u − u0 + u0 ) = A(u − u0) + A(u0).
But observe that
u − u0 U ≤ u U + u0 U
< 1 + u0 U
= r2 , (1.43)
so that
A(u − u0) ∈ A(Br2 (θ )).
This means
y = A(u) ∈ A(u0 ) + A(Br2 (θ )),
and hence
A(B1 (θ )) ⊂ A(u0 ) + A(Br2 (θ )).
That is, from this and (1.42), we obtain
A(u0 ) + A(Br2 (θ )) ⊃ A(B1 (θ )) ⊃ BVr1 (y0 ) = A(u0 ) + BVr1 (θ ),
and therefore
A(Br2 (θ )) ⊃ BVr1 (θ ).
Since
A(Br2 (θ )) = r2 A(B1 (θ )),
we have, for some not relabeled r1 > 0, that
A(B1 (θ )) ⊃ BVr1 (θ ).
Thus it suffices to show that
A(B1 (θ )) ⊂ A(B2 (θ )),
to prove (1.41). Let y ∈ A(B1 (θ )); since A is continuous, we may select u1 ∈ B1 (θ )

such that
y − A(u1) ∈ BVr1 /2 (θ ) ⊂ A(B1/2 (θ )).
Now select u2 ∈ B1/2 (θ ) so that
y − A(u1) − A(u2) ∈ BVr1 /4 (θ ).
By induction, we may obtain

un ∈ B21−n (θ ),
such that
n
y − ∑ A(u j ) ∈ BVr1 /2n (θ ).
j=1
Define
∞
u= ∑ un ,
n=1
we have that u ∈ B2 (θ ), so that

∞
y= ∑ A(un) = A(u) ∈ A(B2(θ )).
n=1
Therefore
A(B1 (θ )) ⊂ A(B2 (θ )).
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 (θ )) ⊃ BVr (θ ),
for some r > 0. Thus
A(O) ⊃ A(u0 ) + A(Br (θ )) ⊃ v0 + BVr (θ ).
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, A−1 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 Γ (A) is defined by
Γ (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 Γ (A) is closed. Since A is linear, Γ (A) is a subspace of U ⊕V . Also,

being Γ (A) closed, it is a Banach space with the norm
(u, A(u) = u U + A(u) V .
Consider the continuous mappings
Π1 (u, A(u)) = u
and
Π2 (u, A(u)) = A(u).
Observe that Π1 is a bijection, so that by the inverse mapping theorem, Π1−1 is
continuous. As
A = Π2 ◦ Π1−1 ,
it follows that A is continuous. The converse is trivial.
1.12 Hilbert Spaces 31
1.12 Hilbert Spaces
At this point we introduce an important class of spaces, namely the Hilbert

spaces.
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. (α u, v)H = α (u, v)H , ∀u, v ∈ H, α ∈ R,
4. (u, u)H ≥ 0, ∀u ∈ H, and (u, u)H = 0, if and only if u = θ .
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
u H = (u, u)H , ∀u ∈ H,
we have
|(u, v)H | ≤ u H v H , ∀u, v ∈ H.
Equality holds if and only if u = α v for some α ∈ R or v = θ .
Proof. If v = θ , the inequality is immediate. Assume v = θ . Given α ∈ R we have
0 ≤ (u − α v, u − α v)H
= (u, u)H + α 2 (v, v)H − 2α (u, v)H
= u 2H + α 2 v 2H − 2α (u, v)H . (1.44)
In particular, for α = (u, v)H / v 2H , we obtain
(u, v)2H
0 ≤ u 2H − ,
v 2H
that is,
|(u, v)H | ≤ u H v H .
The remaining conclusions are left to the reader.
Proposition 1.12.4. On a pre-Hilbert space H, the function
· H : H → R
is a norm, where as above

u H = (u, u).
Proof. The only nontrivial 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
u + v 2H = (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 |
≤ u 2H + v 2H + 2 u H v H
= ( u H + v H )2 . (1.45)
Therefore
u + v H ≤ u H + v H , ∀u, v ∈ H.
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. Consider-

ing 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 and M a closed subspace of H and sup-
pose u ∈ H. Under such hypotheses there exists a unique m0 ∈ M such that
u − m0 H = min { u − m H }.
m∈M
Moreover n0 = u − m0 ∈ M ⊥ so that
u = m0 + n 0 ,
where m0 ∈ M and n0 ∈ M ⊥ . Finally, such a representation through M ⊕ M ⊥ is

unique.
Proof. Define d by
d = inf { u − m H }.
m∈M
Let {mi } ⊂ M be a sequence such that
u − mi H → d, as i → ∞.
Thus, from the parallelogram law, we have
mi − m j 2H = mi − u − (m j − u) 2H
= 2 mi − u 2H + 2 m j − u 2H
−2 − 2u + mi + m j 2H
= 2 mi − u 2H + 2 m j − u 2H
−4 − u + (mi + m j )/2 2H
→ 2d 2 + 2d 2 − 4d 2 = 0, as i, j → +∞. (1.46)
Thus {mi } ⊂ M is a Cauchy sequence. Since M is closed, there exists m0 ∈ M such

that
mi → m0 , as i → +∞,
so that
u − mi H → u − m0 H = d.
Define
n 0 = u − m0 .
⊥
We will prove that n0 ∈ M .
Pick m ∈ M and t ∈ R, and thus we have
d 2 ≤ u − (m0 − tm) 2H
= n0 + tm 2H
= n0 2H + 2(n0, m)H t + m 2H t 2 . (1.47)
Since
n0 2H = u − m0 2H = d 2 ,
we obtain
2(n0 , m)H t + m 2H t 2 ≥ 0, ∀t ∈ R
so that
(n0 , m)H = 0.
Being m ∈ M arbitrary, we obtain
n0 ∈ M ⊥ .
It remains to prove the uniqueness. Let m ∈ M, and thus
u − m 2H = u − m0 + m0 − m 2H
= u − m0 2H + m − m0 2H , (1.48)
since
(u − m0, m − m0 )H = (n0 , m − m0)H = 0.
From (1.48) we obtain
u − m 2H > u − m0 2H = d 2 ,
if m = m0 .
Therefore m0 is unique.
Now suppose
u = m1 + n 1 ,
where m1 ∈ M and n1 ∈ M ⊥ . As above, for m ∈ M
u − m 2H = u − m1 + m1 − m 2H
= u − m1 2H + m − m1 2H ,
≥ u − m1 H (1.49)
and thus since m0 such that

d = u − m0 H
is unique, we get
m1 = m0
and therefore
n 1 = u − m0 = n 0 .
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
f H ∗ = u0 H .
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, θ )H , ∀u ∈ H and the proof would be complete. Thus, assume N = H.
By the last theorem there exists v = θ such that v ∈ N ⊥ .
Define
f (v)
u0 = v.
v 2H
Thus,if u ∈ N we have
f (u) = 0 = (u, u0 )H = 0.
On the other hand, if u = α v for some α ∈ R, we have
f (u) = α f (v)
f (v)(α v, v)H
=
v 2H

f (v)v
= α v,
v 2H H
= (α 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,
u0 − u1 2H = (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
f H ∗ = u0 H .
First observe that
f H ∗ = sup{ f (u) | u ∈ H, u H ≤ 1}
= sup{|(u, u0 )H | | u ∈ H, u H ≤ 1}
≤ sup{ u H u0 H | u ∈ H, u H ≤ 1}
≤ u0 H . (1.53)
On the other hand
f H ∗ = sup{ f (u) | u ∈ H, u H ≤ 1}

u0
≥ f
u0 H
(u0 , u0 )H
=
u0 H
= u0 H . (1.54)
From (1.53) and (1.54)

f H ∗ = u0 H .
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. (α u, v)H = α (u, v)H , ∀u, v ∈ H, α ∈ C,
4. (u, u)H ≥ 0, ∀u ∈ H, and (u, u) = 0, if and only if u = θ .
Observe that in this case we have
(u, α v)H = α (u, v)H , ∀u, v ∈ H, α ∈ C,
where for α = a + bi ∈ C, we have α = a − bi. Finally, similar results as those

proven above are valid for complex Hilbert spaces.
1.13 Orthonormal Basis
In this section we study separable Hilbert spaces and the related orthonormal
bases.
Definition 1.13.1. Let H be a Hilbert space. A set S ⊂ H is said to be orthonormal if
u H = 1,
and
(u, v)H = 0, ∀u, v ∈ S, such that u = v.
If S is not properly contained in any other orthonormal set, it is said to be an or-
thonormal basis for H.
Theorem 1.13.2. Let H be a Hilbert space and let {un }Nn=1 be an orthonormal set.
Then, for all u ∈ H, we have
2
N N

u 2H = ∑ |(u, un )H | + u − ∑ (u, un )H un .
2
n=1
n=1

H
Proof. Observe that

N N
u= ∑ (u, un )H un + u − ∑ (u, un )H un .
n=1 n=1
Furthermore, we may easily obtain that

N N
∑ (u, un)H un and u − ∑ (u, un )H un
n=1 n=1
1.13 Orthonormal Basis 37
are orthogonal vectors so that
u 2H = (u, u)H
2 2
N N

= ∑ |(u, un )H un + u − ∑ (u, un )H un
n=1 n=1

H H
2
N N

= ∑ |(u, un )H |2 + u − ∑ (u, un )H un . (1.55)
n=1
n=1

H
Corollary 1.13.3 (Bessel Inequality). Let H be a Hilbert space and let {un }Nn=1 be
an orthonormal set. Then, for all u ∈ H, we have
N
u 2H ≥ ∑ |(u, un)H |2 .
n=1
Theorem 1.13.4. Each Hilbert space has an orthonormal basis.
Proof. Define by C the collection of all orthonormal sets in H. Define an order in C

by stating S1 ≺ S2 if S1 ⊂ S2 . Then, C is partially ordered and obviously nonempty,
since
v/ v H ∈ C, ∀v ∈ H, v = θ .
Now let {Sα }α ∈L be a linearly ordered subset of C. Clearly, ∪α ∈L Sα is an orthonor-
mal set which is an upper bound for {Sα }α ∈L .
Therefore, every linearly ordered subset has an upper bound, so that by Zorn’s
lemma C has a maximal element, that is, an orthonormal set not properly contained
in any other orthonormal set.
This completes the proof.
Theorem 1.13.5. Let H be a Hilbert space and let S = {uα }α ∈L be an orthonormal

basis. Then for each v ∈ H we have
v= ∑ (uα , v)H uα ,
α ∈L
and
v 2H = ∑ |(uα , v)H |2 .
α ∈L
Proof. Let L ⊂ L be a finite subset of L. From Bessel’s inequality we have
∑ |(uα , v)H | ≤ v 2H .
α ∈L
From this, we may infer that the set An = {α ∈ L | |(uα , v)H | > 1/n} is finite, so that
A = {α ∈ L | |(uα , v)H | > 0} = ∪∞

n=1 An
is at most countable.
Thus (uα , v)H = 0 for at most countably many α s ∈ L, which we order by
{αn }n∈N . Since the sequence
N
sN = ∑ |(uαi , v)H |2 ,
i=1
is monotone and bounded, it is converging to some real limit as N → ∞. Define

n
vn = ∑ (uαi , v)H uαi ,
i=1
so that for n > m we have

2
n

vn − vm 2H = ∑ (uαi , v)H uαi
i=m+1
H
n
= ∑ |(uαi , v)H |2
i=m+1
= |sn − sm |. (1.56)
Hence, {vn } is a Cauchy sequence which converges to some v ∈ H.

Observe that
N
(v − v, uαl )H = lim (v − ∑ (uαi , v)H uαi , uαl )H
N→∞
i=1
= (v, uαl )H − (v, uαl )H
= 0. (1.57)
Also, if α = αl , ∀l ∈ N, then
∞
(v − v , uα )H = lim (v − ∑ (uαi , v)H uαi , uα )H = 0.
N→∞
i=1
Hence
v − v⊥uα , ∀α ∈ L.
If
v − v = θ ,
then we could obtain an orthonormal set

v − v
uα , α ∈ L,
v − v H
1.13 Orthonormal Basis 39
which would properly contain the complete orthonormal set
{uα , α ∈ L},
a contradiction.
Therefore, v − v = θ , that is,
N
v = lim
N→∞
∑ (uαi , v)H uαi .
i=1
1.13.1 The Gram–Schmidt Orthonormalization
Let H be a Hilbert space and {un } ⊂ H be a sequence of linearly independent

vectors. Consider the procedure
w1
w1 = u1 , v1 = ,
w1 H
w2
w2 = u2 − (v1 , u2 )H v1 , v2 = ,
w2 H
and inductively,
n−1
wn
wn = un − ∑ (vk , un )H vk , vn = , ∀n ∈ N, n > 2.
k=1 wn H
Observe that clearly {vn } is an orthonormal set and for each m ∈ N, {vk }m
k=1 and
{uk }m
k=1 span the same vector subspace of H.
Such a process of obtaining the orthonormal set {vn } is known as the Gram–
Schmidt orthonormalization.
We finish this section with the following theorem.
Theorem 1.13.6. A Hilbert space H is separable if and only if it has a countable

orthonormal basis. If dim(H) = N < ∞, the H is isomorphic to CN . If dim(H) = +∞,
then H is isomorphic to l 2 , where

∞
l2 = {yn } | yn ∈ C, ∀n ∈ N and ∑ |yn|2 < +∞ .
n=1
Proof. Suppose H is separable and let {un } be a countable dense set in H. To ob-
tain an orthonormal basis it suffices to apply the Gram–Schmidt orthonormalization
procedure to the greatest linearly independent subset of {un }.
Conversely, if B = {vn } is an orthonormal basis for H, the set of all finite linear
combinations of elements of B with rational coefficients are dense in H, so that H is
separable.
Moreover, if dim(H) = +∞, consider the isomorphism F : H → l 2 given by
F(u) = {(un , u)H }n∈N .
Finally, if dim(H) = N < +∞, consider the isomorphism F : H → CN given by
F(u) = {(un , u)H }Nn=1 .

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 reflex-
ive spaces may be weakly convergent, and it is important to evaluate the average
behavior of such sequences in many practical applications. Finally, we emphasize
the main reference for this chapter is Brezis [16], where more details may be found.
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 Hahn–Banach theorem for real vector spaces,
which is sufficient for our purposes.
Theorem 2.2.1 (The Hahn–Banach Theorem). Consider a functional p : U → R
satisfying
p(λ u) = λ p(u), ∀u ∈ U, λ > 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)

42 2 The Hahn–Banach Theorems and Weak Topologies
and
f (u) ≤ p(u), ∀u ∈ U. (2.5)
Proof. Pick z ∈ U − V . Denote by Ṽ the space spanned by V and z, that is,
Ṽ = {v + α z | v ∈ V and α ∈ R}. (2.6)
We may define an extension of g to Ṽ , denoted by g̃, as
g̃(α z + v) = α g̃(z) + g(v), (2.7)
where g̃(z) will be appropriately defined. Suppose given v1 , v2 ∈ V , α > 0, β > 0.

Then
β g(v1 ) + α g(v2 ) = g(β v1 + α v2 )

β α
= (α + β )g( v1 + v2 )
α +β α +β
β α
≤ (α + β )p( (v1 − α z) + (v2 + β z))
α +β α +β
≤ β p(v1 − α z) + α p(v2 + β z) (2.8)
and therefore
1 1
[−p(v1 − α z)+g(v1 )] ≤ [p(v2 + β z)−g(v2 )], ∀v1 , v2 ∈ V, α , β > 0. (2.9)
α β
Thus, there exists a ∈ R such that
1 1
sup [ (−p(v − α z) + g(v))] ≤ a ≤ inf [ (p(v + α z) − g(v))]. (2.10)
v∈V,α >0 α v∈V,α >0 α
If we define g̃(z) = a, we obtain g̃(u) ≤ p(u), ∀u ∈ Ṽ . Define by E the set of ex-

tensions 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 {eα }α ∈A be a linearly ordered subset of
E . Let Vα be the subspace on which eα is defined. Define e on ∪α ∈AVα by setting
e(u) = eα on Vα . Clearly eα ≺ e so each linearly ordered set of E has an upper
bound. By Zorn’s lemma, E has a maximal element f defined on some set Ũ such
that f (u) ≤ p(u), ∀u ∈ Ũ. We can conclude that Ũ = U; otherwise, if there was an
z1 ∈ U − Ũ, as above, we could have a new extension f1 to the subspace spanned by
z1 and Ũ, contradicting the maximality of f .
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 ·, ·U : U × U ∗ → R (here we are referring to the standard
2.2 The Hahn–Banach Theorem 43
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) = u, u∗ U , ∀u ∈ U. (2.11)
The norm of f , denoted by f U ∗ , is defined as
f U ∗ = sup{|u, u∗ U | | u U ≤ 1}. (2.12)

u∈U
Corollary 2.2.3. Let V ⊂ U be a vector subspace of U and let g : V → R be a linear

continuous functional of norm
g V ∗ = sup{|g(u)| | u V ≤ 1}. (2.13)

u∈V
Then, there exists an u∗ in U ∗ such that
u, u∗ U = g(u), ∀u ∈ V, (2.14)
and
u∗ U ∗ = g V ∗ . (2.15)
Proof. Apply Theorem 2.2.1 with p(u) = g V ∗ u V .
Corollary 2.2.4. Given u0 ∈ U there exists u∗0 ∈ U ∗ such that
u∗0 U ∗ = u0 U and u0 , u∗0 U = u0 U2 . (2.16)
Proof. Apply Corollary 2.2.3 with V = {α u0 | α ∈ R} and g(tu0 ) = t u0 U2 so that

g V ∗ = u0 U .
Corollary 2.2.5. Given u ∈ U we have
u U = sup {|u, u∗U | | u∗ U ∗ ≤ 1}. (2.17)

u∗ ∈U ∗
Proof. Suppose u = θ . Since
|u, u∗ U | ≤ u U u∗ U ∗ , ∀u ∈ U, u∗ ∈ U ∗
we have
sup {|u, u∗U | | u∗ U ∗ ≤ 1} ≤ u U . (2.18)

u∗ ∈U ∗
However, from last corollary, we have that there exists u∗0 ∈ U ∗ such that u∗0 U ∗ =
u U and u, u∗0 U = u U2 . Define u∗1 = u U−1 u∗0 . Then u∗1 U = 1 and u, u∗1 U =
u U .
Definition 2.2.6 (Affine Hyperplane). Let U be a Banach space. An affine

hyperplane H is a set of the form
H = {u ∈ U | u, u∗ U = α } (2.19)
for some u∗ ∈ U ∗ and α ∈ R.

Proposition 2.2.7. A hyperplane H defined as above is closed.
Proof. The result follows from the continuity of u, u∗ U as a functional defined
in U.
Definition 2.2.8 (Separation). Given A, B ⊂ U we say that a hyperplane H, defined
as above, separates A and B if
u, u∗ U ≤ α , ∀u ∈ A, and u, u∗ U ≥ α , ∀u ∈ B. (2.20)
We say that H separates A and B strictly if there exists ε > 0 such that
u, u∗ U ≤ α − ε , ∀u ∈ A, and u, u∗ U ≥ α + ε , ∀u ∈ B, (2.21)
Theorem 2.2.9 (Hahn–Banach Theorem, Geometric Form). Consider A, B ⊂ U

two convex disjoint nonempty sets, where A is open. Then there exists a closed hy-
perplane that separates A and B.
We need the following lemma.
Lemma 2.2.10. Consider C ⊂ U a convex open set such that θ ∈ C. Given u ∈ U,
define
p(u) = inf{α > 0, α −1 u ∈ C}. (2.22)
Thus, p is such that there exists M ∈ R+ satisfying
0 ≤ p(u) ≤ M u U , ∀u ∈ U, (2.23)
and
C = {u ∈ U | p(u) < 1}. (2.24)
Also
p(u + v) ≤ p(u) + p(v), ∀u, v ∈ U.
Proof. Let r > 0 be such that B(θ , r) ⊂ C; thus
u U
p(u) ≤ , ∀u ∈ U (2.25)
r
which proves (2.23). Now suppose u ∈ C. Since C is open, (1 + ε )u ∈ C for ε is
sufficiently small. Therefore p(u) ≤ 1+1 ε < 1. Conversely, if p(u) < 1, there exists
0 < α < 1 such that α −1 u ∈ C and therefore, since C is convex, u = α (α −1 u) + (1 −
α )θ ∈ C.
2.2 The Hahn–Banach Theorem 45
Also, let u, v ∈ C and ε > 0. Thus u

p(u)+ε
∈ C and v
p(v)+ε
∈ C so that tu
p(u)+ε
+
(1−t)v p(u)+ε
p(v)+ε
∈ C, ∀t ∈ [0, 1]. Particularly for t = p(u)+p(v)+2ε
we obtain u+v
p(u)+p(v)+2ε
∈ C,
which means p(u + v) ≤ p(u) + p(v) + 2ε , ∀ε > 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 u, u∗ U < u0 , u∗ U , ∀u ∈ C
Proof. By a translation, we may assume θ ∈ C. Consider the functional p as in the

last lemma. Define V = {α u0 | α ∈ 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 exists a
linear functional f on U which extends g such that
f (u) ≤ p(u) ≤ M u U . (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).
Proof of Theorem 2.2.9. Define C = A + (−B) so that C is convex and θ ∈ C. From

Lemma 2.2.11, there exists u∗ ∈ U ∗ such that w, u∗ U < 0, ∀w ∈ C, which means
u, u∗ U < v, u∗ U , ∀u ∈ A, v ∈ B. (2.28)
Thus, there exists α ∈ R such that
supu, u∗ U ≤ α ≤ inf v, u∗ U , (2.29)

u∈A v∈B
which completes the proof.
Theorem 2.2.12 (Hahn–Banach Theorem, Second Geometric Form). Consider

A, B ⊂ U two convex disjoint nonempty sets. Suppose A is closed and B is compact.
Then there exists a hyperplane which separates A and B strictly.
Proof. There exists ε > 0 sufficiently small such that Aε = A + B(0, ε ) and Bε =
B + B(0, ε ) are convex disjoint sets. From Theorem 2.2.9, there exists u∗ ∈ U ∗ such
that u∗ = θ and
u + ε w1 , u∗ U ≤ u + ε w2 , u∗ U , ∀u ∈ A, v ∈ B, w1 , w2 ∈ B(0, 1). (2.30)
Thus, there exists α ∈ R such that
u, u∗ U + ε u∗ U ∗ ≤ α ≤ v, u∗ U − ε u∗ U ∗ , ∀u ∈ A, v ∈ B. (2.31)

Corollary 2.2.13. Suppose V ⊂ U is a vector subspace such that V̄ = U. Then there

exists u∗ ∈ U ∗ such that u∗ = θ and
u, u∗ U = 0, ∀u ∈ V. (2.32)
Proof. Consider u0 ∈ U such that u0 ∈ V . Applying Theorem 2.2.9 to A = V and

B = {u0 } we obtain u∗ ∈ U ∗ and α ∈ R such that u∗ = θ and
u, u∗ U < α < u0 , u∗ U , ∀u ∈ V. (2.33)
Since V is a subspace we must have u, u∗ U = 0, ∀u ∈ V .
2.3 Weak Topologies
Definition 2.3.1 (Weak Neighborhoods and Weak Topologies). For the topologi-
cal space U and u0 ∈ U, we define a weak neighborhood of u0 , denoted by Vw as
Vw = {u ∈ U | |u − u0, u∗i U | < ε , ∀i ∈ {1, . . . , m}}, (2.34)
for some m ∈ N, ε > 0, and u∗i ∈ U ∗ , ∀i ∈ {1, . . . , m}. Also, we define the weak
topology for U, denoted by σ (U,U ∗ ), as the set of arbitrary unions and finite inter-
sections of weak neighborhoods in U.
Proposition 2.3.2. Consider Z a topological vector space and ψ a function of Z

into U. Then ψ is continuous as U is endowed with the weak topology, if and only if
u∗ ◦ ψ is continuous, for all u∗ ∈ U ∗ .
Proof. It is clear that if ψ is continuous with U endowed with the weak topology,
then u∗ ◦ ψ is continuous for all u∗ ∈ U ∗ . Conversely, consider U a weakly open set
in U. We have to show that ψ −1 (U ) is open in Z. But observe that U = ∪λ ∈L Vλ ,
where each Vλ is a weak neighborhood. Thus ψ −1 (U ) = ∪λ ∈L ψ −1 (Vλ ). The result
follows considering that u∗ ◦ ψ is continuous for all u∗ ∈ U ∗ , so that ψ −1 (Vλ ) is
open, for all λ ∈ L.
Proposition 2.3.3. A Banach space U is Hausdorff as endowed with the weak topol-
ogy σ (U,U ∗ ).
Proof. Pick u1 , u2 ∈ U such that u1 = u2 . From the Hahn–Banach theorem, second

geometric form, there exists a hyperplane separating {u1 } and {u2 }. That is, there
exist u∗ ∈ U ∗ and α ∈ R such that
u1 , u∗ U < α < u2 , u∗ U . (2.35)
Defining
Vw1 = {u ∈ U | |u − u1, u∗ | < α − u1 , u∗ U }, (2.36)

2.3 Weak Topologies 47
and
Vw2 = {u ∈ U | |u − u2, u∗ U | < u2 , u∗ U − α }, (2.37)
we obtain u1 ∈ Vw1 , u2 ∈ Vw2 and Vw1 ∩ Vw2 = 0.

/
Remark 2.3.4. If {un } ∈ U is such that un converges to u in σ (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 σ (U,U ∗ ) ⇔ un , u∗ U → u, u∗ U , ∀u∗ ∈ U ∗ ,
2. if un → u strongly (in norm), then un u weakly,
3. if un u weakly, then { un U } is bounded and u U ≤ lim inf un U ,
n→∞
4. if un u weakly and u∗n → u∗ strongly in U ∗ , then un , u∗n U → u, u∗ U .
Proof.
1. The result follows directly from the definition of topology σ (U,U ∗ ).
2. This follows from the inequality
|un , u∗ U − u, u∗U | ≤ u∗ U ∗ un − u U . (2.38)
3. Since for every u∗ ∈ U ∗ the sequence {un , u∗ U } is bounded, from the uni-
form boundedness principle, we have that there exists M > 0 such that un U ≤
M, ∀n ∈ N. Furthermore, for u∗ ∈ U ∗ , we have
|un , u∗ U | ≤ u∗ U ∗ un U , (2.39)
and taking the limit, we obtain
|u, u∗ U | ≤ lim inf u∗ U ∗ un U . (2.40)

n→∞
Thus
u U = sup |u, u∗ U | ≤ lim inf un U . (2.41)

u U ∗ ≤1 n→∞
4. Just observe that
|un , u∗n U − u, u∗U | ≤ |un , u∗n − u∗ U |

+|u − un, u∗ U |
≤ u∗n − u∗ U ∗ un U
+|un − u, u∗U |
≤ M u∗n − u∗ U ∗
+|un − u, u∗U |. (2.42)
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 ∈ A. By the Hahn–Banach the-
orem there exists a closed hyperplane which separates u0 and A strictly. Therefore
there exists α ∈ R and u∗ ∈ U ∗ such that
u0 , u∗ U < α < v, u∗ U , ∀v ∈ A. (2.43)
Define
V = {u ∈ U | u, u∗ U < α }, (2.44)
so that u0 ∈ V , V ⊂ U − A. Since V is open for σ (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
u, u∗ U = u∗ , J(u)U ∗ , ∀u ∈ U, u∗ ∈ U ∗ , (2.45)
is onto.
The weak topology for U ∗ is denoted by σ (U ∗ ,U ∗∗ ). By analogy, we can define
the topology σ (U ∗ ,U), which is called the weak-star topology. A standard neighbor-
hood of u∗0 ∈ U ∗ for the weak-star topology, which we denoted by Vw∗ , is given by
Vw∗ = {u∗ ∈ U ∗ | |ui , u∗ − u∗0U | < ε , ∀i ∈ {1, . . . , m}} (2.46)
for some ε > 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.
2.5 Weak-Star Compactness
We start with an important theorem about weak-star compactness.

Theorem 2.5.1 (Banach–Alaoglu Theorem). The set BU ∗ = { f ∈ U ∗ | f U ∗ ≤ 1}
is compact for the topology σ (U ∗ ,U) (the weak-star topology).
2.5 Weak-Star Compactness 49
Proof. For each u ∈ U, we will associate a real number ωu and denote ω = ∏u∈U ωu .
We have that ω ∈ RU and let us consider the projections Pu : RU → R, where
Pu (ω ) = ωu . Consider the weakest topology σ for which the functions Pu (u ∈ U)
are continuous. For U ∗ , with the topology σ (U ∗ ,U), define φ : U ∗ → RU by
φ (u∗ ) = ∏ u, u∗U , ∀u∗ ∈ U ∗. (2.47)

u∈U
Since for each fixed u the mapping u∗ → u, u∗ U is weakly star continuous, we
see that φ is σ continuous, since weak-star convergence and convergence in σ are
equivalent in U ∗ . To prove that φ −1 is continuous, from Proposition 2.3.2, it suf-
fices to show that the function ω → u, φ −1 (ω )U is continuous on φ (U ∗ ). This is
true because u, φ −1 (ω )U = ωu on φ (U ∗ ). On the other hand, it is also clear that
φ (BU ∗ ) = K, where
K = {ω ∈ RU | |ωu | ≤ u U ,
ωu+v = ωu + ωv , ωλ u = λ ωu , ∀u, v ∈ U, λ ∈ R}. (2.48)
To finish the proof, it is sufficient, from the continuity of φ −1 , to show that K is

compact in RU , concerning the topology σ . Observe that K = K1 ∩ K2 , where
K1 = {ω ∈ RU | |ωu | ≤ u U , ∀u ∈ U}, (2.49)
and
K2 = {ω ∈ RU | ωu+v = ωu + ωv , ωλ u = λ ωu , ∀u, v ∈ U, λ ∈ R}. (2.50)
The set K3 ≡ ∏u∈U [− u U , u U ] is compact as a cartesian product of compact

intervals. Since K1 ⊂ K3 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 Bλ ,u as
Au,v = {ω ∈ RU | ωu+v − ωu − ωv = 0}, (2.51)
and
Bλ ,u = {ω ∈ RU |ωλ u − λ ωu = 0} (2.52)
we may write
K2 = (∩u,v∈U Au,v ) ∩ (∩(λ ,u)∈R×U Bλ ,u ). (2.53)
We recall that the K2 is closed because arbitrary intersections of closed sets are
closed. Finally, we have that K1 ∩ K2 is compact, which completes the proof.
Theorem 2.5.2 (Kakutani). Let U be a Banach space. Then U is reflexive if and

only if
BU = {u ∈ U | u U ≤ 1} (2.54)
is compact for the weak topology σ (U,U ∗ ).

Proof. Suppose U is reflexive, and then J(BU ) = BU ∗∗ . From the last theorem BU ∗∗ is
compact for the topology σ (U ∗∗ ,U ∗ ). Therefore it suffices to verify that J −1 : U ∗∗ →
U is continuous from U ∗∗ with the topology σ (U ∗∗ ,U ∗ ) to U, with the topology
σ (U,U ∗ ).
From Proposition 2.3.2 it is sufficient to show that the function u → J −1 u, f U
is continuous for the topology σ (U ∗∗ ,U ∗ ), for each f ∈ U ∗ . Since J −1 u, f U =
f , uU ∗ 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 α1 , . . . , αn ∈
R, and then 1 and 2 are equivalent, where:
1.
Given ε > 0, there exists uε ∈ U such that uε U ≤ 1 and
|uε , fi U − αi | < ε , ∀i ∈ {1, . . . , n}.
2.

n n

∑ βi αi ≤ ∑ βi fi , ∀β1 , . . . , βn ∈ R. (2.55)
i=1 i=1
U∗
Proof. 1 ⇒ 2: Fix β1 , . . . , βn ∈ R, ε > 0 and define S = ∑ni=1 |βi |. From 1, we have

n n

∑ βi uε , fi U − ∑ βi αi < ε S (2.56)
i=1 i=1

and therefore

n n

∑ βi αi − ∑ βi uε , fi U < ε S (2.57)
i=1 i=1
or

n n n

∑ βi αi < ∑ βi fi uε U + ε S ≤ ∑ βi fi + εS (2.58)
i=1 i=1 ∗

U i=1 U∗
so that

n n

∑ βi αi ≤ ∑ βi fi (2.59)
i=1 i=1
U∗
since ε is arbitrary. Now let us show that 2 ⇒ 1. Define α = (α1 , . . . , αn ) ∈ Rn and

consider the function ϕ (u) = (u, f1 U , . . . , u, fn U ). Item 1 implies that α belongs
2.5 Weak-Star Compactness 51
to the closure of ϕ (BU ). Let us suppose that α does not belong to the closure of
ϕ (BU ) and obtain a contradiction. Thus we can separate α and the closure of ϕ (BU )
strictly, that is, there exists β = (β1 , . . . , βn ) ∈ Rn and γ ∈ R such that
ϕ (u) · β < γ < α · β , ∀u ∈ BU (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 topol-
ogy σ (U ∗∗ ,U ∗ ).
Proof. Let u∗∗ ∈ BU ∗∗ and consider Vu∗∗ a neighborhood of u∗∗ for the topology
σ (U ∗∗ ,U ∗ ). It suffices to show that J(BU ) ∩ Vu∗∗ = 0.
/ As Vu∗∗ is a weak neighbor-
∗
hood, there exists f1 , . . . , fn ∈ U and ε > 0 such that
Vu∗∗ = {η ∈ U ∗∗ | fi , η − u∗∗U ∗ | < ε , ∀i ∈ {1, . . . , n}}. (2.61)
Define αi = fi , u∗∗ U ∗ and thus for any given β1 , . . . , βn ∈ R we have

n n n

∑ βi αi = ∑ βi fi , u ∗∗
≤ ∑ βi f i , (2.62)
i=1 i=1 ∗
i=1
∗
U U
so that from Helly lemma, there exists uε ∈ U such that uε U ≤ 1 and
|uε , fi U − αi | < ε , ∀i ∈ {1, . . . , n} (2.63)
or,
| fi , J(uε ) − u∗∗U ∗ | < ε , ∀i ∈ {1, . . . , n} (2.64)
and hence
J(uε ) ∈ Vu∗∗ . (2.65)
Now we will complete the proof of Kakutani theorem. Suppose BU is weakly com-
pact (i.e., compact for the topology σ (U,U ∗ )). Observe that J : U → U ∗∗ is weakly
continuous, that is, it is continuous with U endowed with the topology σ (U,U ∗ )
and U ∗∗ endowed with the topology σ (U ∗∗ ,U ∗ ). Thus as BU is weakly com-
pact, we have that J(BU ) is compact for the topology σ (U ∗∗ ,U ∗ ). From the last
lemma, J(BU ) is dense BU ∗∗ for the topology σ (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 σ (U,U ∗ )).
Since K is bounded, there exists α ∈ R+ such that K ⊂ α BU . Since K is weakly
closed and K = K ∩ α 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:
σ (M, M ∗ ) and the trace of σ (U,U ∗ ). (2.66)
It can be easily verified that these two topologies coincide (through restrictions and
extensions of linear forms). From Theorem 2.5.2, it suffices to show that BM is
compact for the topology σ (M, M ∗ ). But BU is compact for σ (U,U ∗ ) and M ⊂
U is closed (strongly) and convex so that it is weakly closed; thus, from the last
proposition, BM is compact for the topology σ (U,U ∗ ), and therefore it is compact
for σ (M, M ∗ ).
2.6 Separable Sets
Definition 2.6.1 (Separable Spaces). A metric space U is said to be separable if

there exists a set K ⊂ U such that K is countable and dense in U.
The next proposition is proved in [16].

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 {u∗n } a countable dense set in U ∗ . Observe that
u∗n U ∗ = sup{|u∗n , uU | | u ∈ U and u U = 1} (2.67)
so that for each n ∈ N, there exists un ∈ U such that un U = 1 and u∗n , un U ≥

∗ ∗
2 un U .
1
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. Suppose u∗ is such that u, u∗ U =
0, ∀u ∈ U1 . Since {u∗n } is dense in U ∗ , given ε > 0, there exists n ∈ N such that
u∗n − u∗ U ∗ < ε , so that
1 ∗
u U ∗ ≤ un , u∗n U = un , u∗n − u∗ U + un, u∗ U
2 n
≤ u∗n − u∗ U ∗ un U + 0 < ε (2.68)
or
u∗ U ∗ ≤ u∗n − u∗ U ∗ + u∗n U ∗ < ε + 2ε = 3ε . (2.69)

2.6 Separable Sets 53
Therefore, since ε is arbitrary, u∗ U ∗ = 0, that is, u∗ = θ . By Corollary 2.2.13 this

completes the proof.
Proposition 2.6.4. U is reflexive if and only if U ∗ is reflexive.
Proof. Suppose U is reflexive; as BU ∗ is compact for σ (U ∗ ,U) and σ (U ∗ ,U) =

σ (U ∗ ,U ∗∗ ), we have that BU ∗ is compact for σ (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 an 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.
Our final result in this section refers to the metrizability of BU ∗ .

Theorem 2.6.6. Let U be separable Banach space. Under such hypotheses BU ∗ is
metrizable with respect to the weak-star topology σ (U ∗ ,U). Conversely, if BU ∗ is
mertizable in σ (U ∗ ,U), then U is separable.
Proof. Let {un } be a dense countable set in BU . For each u∗ ∈ U ∗ define

∞
1
u∗ w = ∑ 2n | un , u∗ U |.
n=1
It may be easily verified that · w is a norm in U ∗ and
u∗ w ≤ u∗ U .
So, we may define a metric in U ∗ by
d(u∗ , v∗ ) = u∗ − v∗ w .
Now we shall prove that the topology induced by d coincides with σ (U ∗ ,U) in U ∗ .
Let u∗0 ∈ BU ∗ and let V be neighborhood of u∗0 in σ (U ∗ ,U).
We need to prove that there exists r > 0 such that
Vw = {u∗ ∈ BU ∗ | d(u∗0 , u∗ ) < r} ⊂ V.
Observe that for V we may assume the general format
V = {u∗ ∈ U ∗ | |vi , u∗ − u∗0 U | < ε , ∀i ∈ {1, ..., k}}
for some ε > 0 and v1 , . . . , vk ∈ U.

There is no loss in generality in assuming
vi U ≤ 1, ∀i ∈ {1, . . . , k}.
Since {un } is dense in U, for each i ∈ {1, . . . , k}, there exists ni ∈ N such that
ε
uni − vi U < .
4
Choose r > 0 small enough such that
ε
2ni r < , ∀i ∈ {1, . . . , k}.
2
We are going to show that Vw ⊂ V , where
Vw = {u∗ ∈ BU ∗ | d(u∗0 , u∗ ) < r} ⊂ V.
Observe that if u∗ ∈ Vw , then

d(u∗0 , u∗ ) < r,
so that
1
|uni , u∗ − u∗0U | < r, ∀i ∈ {1, . . . , k},
2 ni
so that
|vi , u∗ − u∗0 U | ≤ |vi − uni , u∗ − u∗0U | + |uni , u∗ − u∗0U |

≤ ( u∗ U ∗ + u∗0 U ∗ ) vi − uni U + |uni , u∗ − u∗0U |
ε ε
< 2 + = ε. (2.70)
4 2
Therefore, u∗ ∈ V , so that Vw ⊂ V .
Now let u0 ∈ BU ∗ and fix r > 0. We have to obtain a neighborhood V ∈ σ (U ∗U)
such that
V ⊂ Vw = {u∗ ∈ BU ∗ | d(u∗0 , u∗ ) < r}.
We shall define k ∈ N and ε > 0 in the next lines so that V ⊂ Vw , where
V = {u∗ ∈ BU ∗ | |ui , u∗ − u∗0U | < ε , ∀i ∈ {1, . . . , k}}.
For u∗ ∈ Vw we have
k
1
d(u∗ , u∗0 ) = ∑ 2n |un, u∗ − u∗0U |
n=1
∞
1
+ ∑ n
|un , u∗ − u∗0 U |
n=k+1 2
∞
1
< ε +2 ∑ 2 n
n=k+1
1
= ε+ . (2.71)
2k−1
2.7 Uniformly Convex Spaces 55
Hence, it suffices to take ε = r/2, and k sufficiently big such that
1
< r/2.
2k−1
The first part of the proof is finished.
Conversely, assume BU ∗ is metrizable in σ (U ∗ ,U). We are going to show that U
is separable.
Define,
1
Ṽn = u∗ ∈ BU ∗ | d(u∗ , θ ) < .
n
From the first part, we may find Vn a neighborhood of zero in σ (U ∗ ,U) such that
Vn ⊂ Ṽn .
Moreover, we may assume that Vn has the form
Vn = {u∗ ∈ BU ∗ | |u, u∗ − θ U | < εn , ∀u ∈ Cn },
where Cn is a finite set.

Define
D = ∪∞
i=1Cn .
Thus D is countable and we are going to prove that such a set is dense in U.
Suppose u∗ ∈ U ∗ is such that
u, u∗ U = 0, ∀u ∈ D.
Hence,
u∗ ∈ Vn ⊂ Ṽn , ∀n ∈ N,
so that u∗ = θ .
2.7 Uniformly Convex Spaces
Definition 2.7.1 (Uniformly Convex Spaces). A Banach space U is said to be uni-

formly convex if for each ε > 0, there exists δ > 0 such that:
If u, v ∈ U, u U ≤ 1, v U ≤ 1, and u − v U > ε , then u+v
2
U
< 1−δ.
Theorem 2.7.2 (Milman Pettis). Every uniformly convex Banach space is reflexive.
Proof. Let η ∈ U ∗∗ be such that η U ∗∗ = 1. It suffices to show that η ∈ J(BU ).

Since J(BU ) is closed in U ∗∗ , we have only to show that for each ε > 0 there exists
u ∈ U such that η − J(u) U ∗∗ < ε .
Thus, suppose given ε > 0. Let δ > 0 be the corresponding constant relating the
uniformly convex property.
Choose f ∈ U ∗ such that f U ∗ = 1 and
δ
f , η U ∗ > 1 − . (2.72)
2
Define
δ
V = ζ ∈ U ∗∗ | | f , ζ − η U ∗ | < .
2
Observe that V is neighborhood of η in σ (U ∗∗ ,U ∗ ). Since J(BU ) is dense in BU ∗∗
concerning the topology σ (U ∗∗ ,U ∗ ), we have that V ∩ J(BU ) = 0/ and thus there
exists u ∈ BU such that J(u) ∈ V. Suppose, to obtain contradiction, that
η − J(u) U ∗∗ > ε .
Therefore, defining
W = (J(u) + ε BU ∗∗ )c ,
we have that η ∈ W , where W is also a weak neighborhood of η in σ (U ∗∗ ,U ∗ ),
since BU ∗∗ is closed in σ (U ∗∗ ,U ∗ ).
Hence V ∩ W ∩ J(BU ) = 0, / so that there exists some v ∈ BU such that J(v) ∈
V ∩W. Thus, J(u) ∈ V and J(v) ∈ V , so that
δ
|u, f U − f , η U ∗ | < ,
2
and
δ
|v, f U − f , η U ∗ | < .
2
Hence,
2 f , η U ∗ 1−δ,
2
and thus from the definition of uniform convexity, we obtain
u − v U ≤ ε . (2.74)
On the other hand, since J(v) ∈ W , we have
J(u) − J(v) U ∗∗ = u − v U > ε ,
which contradicts (2.74). The proof is complete.

Chapter 3
Topics on Linear Operators
The main references for this chapter are Reed and Simon [52] and Bachman and
Narici [6].
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
A = sup{ Au Y | u U ≤ 1}.
The topology related to the metric induced by this norm is called the uniform oper-
ator topology.
Let us introduce now the strong operator topology, which is defined as the weak-
est topology for which the functions
Eu : L (U,Y ) → Y
are continuous where

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 ), Aui Y < ε , ∀i ∈ {1, . . . , n}},
where u1 , . . . , un ∈ U and ε > 0.

Observe that a sequence {An } ⊂ L (U,Y ) converges to A concerning this last
topology if
An u − Au Y → 0, as n → ∞, ∀u ∈ U.

58 3 Topics on Linear Operators
In the next lines we describe the weak operator topology in L (U,Y ). Such a topol-
ogy is weakest one such that the functions
Eu,v : L (U,Y ) → C
are continuous, where
Eu,v (A) = Au, vY , ∀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 ) | |Aui , v j Y | < ε , ∀i ∈ {1, . . ., n}, j ∈ {1, . . . , m}},
where ε > 0, u1 , . . . , un ∈ U, v1 , . . . , vm ∈ Y ∗ .
A sequence {An } ⊂ L (U,Y ) converges to A ∈ L (U,Y ) if
|An u, vY − Au, vY | → 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) = Au, v∗ Y is such that
|T (u)| ≤ Au Y · v∗ ≤ A v∗ Y ∗ u U .
Hence T (u) is a continuous linear functional on U and considering our fundamental

representation hypothesis, there exists u∗ ∈ U ∗ such that
T (u) = u, u∗ U , ∀u ∈ U.
We define A∗ by setting u∗ = A∗ v∗ , so that
T (u) = u, u∗ U = u, A∗ v∗ U
that is,
u, A∗ v∗ U = Au, v∗ Y , ∀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
A = A∗ .
3.2 Adjoint Operators 59
Proof. Observe that
A = sup{ Au | u U = 1}
u∈U

∗ ∗ ∗
= sup sup {Au, v Y | v Y = 1}, u U = 1
u∈U v∗ ∈Y ∗
= sup {Au, v∗ Y | v∗ Y ∗ = 1, u U = 1}
(u,v∗ )∈U×Y ∗
= sup {u, A∗ v∗ U | v∗ Y ∗ = 1, u U = 1}
(u,v∗ )∈U×Y ∗

∗ ∗ ∗
= sup sup{u, A v U | u U = 1}, v Y ∗ = 1
v∗ ∈Y ∗ u∈U
= sup { A v , v∗ Y ∗ = 1}
∗ ∗
v∗ ∈Y ∗
∗
= A . (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 A−1 , then A∗ has a bounded inverse and
(A∗ )−1 = (A−1 )∗ .
4. AA∗ = A 2 .
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 = AA−1 = A−1 A,
so that
I = I ∗ = (AA−1 )∗ = (A−1 )∗ A∗ = (A−1 A)∗ = A∗ (A−1 )∗ .
4. Observe that
A∗ A ≤ A A∗ = A 2 ,
and
A∗ A ≥ sup{(u, A∗ Au)H | u U = 1}
u∈U
= sup{(Au, Au)H | u U = 1}
u∈U
= sup{ Au 2H | u U = 1} = A 2, (3.2)
u∈U
and hence
A∗ A = A 2 .
Definition 3.2.4. Given A ∈ L (H) we say that A is self-adjoint if
A = 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 ∗ | Au, v∗ Y = 0, ∀u ∈ U}.
Proof. Let v∗ ∈ N(A∗ ). Choose v ∈ R(A). Thus there exists u in U such that Au = v
so that
v, v∗ Y = Au, v∗ Y = u, A∗ v∗ U = 0.
Since v ∈ R(A) is arbitrary we have obtained
N(A∗ ) ⊂ [R(A)]⊥ .
Suppose v∗ ∈ [R(A)]⊥ . Choose u ∈ U. Thus,
Au, v∗ Y = 0,
so that
u, A∗ v∗ U , ∀u ∈ U.
Therefore A∗ v∗ = θ , that is, v∗ ∈ N(A∗ ). Since v∗ ∈ [R(A)]⊥ is arbitrary, we get
[R(A)]⊥ ⊂ N(A∗ ).
The next result is relevant for subsequent developments.

Lemma 3.1. Let U,Y be Banach spaces and let A : U → Y be a bounded linear op-
erator. Suppose also that R(A) = {A(u) : u ∈ U} is closed. Under such hypotheses,
there exists K > 0 such that for each v ∈ R(A) there exists u0 ∈ U such that
A(u0 ) = v
and
u0 U ≤ K v Y .
Proof. Define L = N(A) = {u ∈ U : A(u) = θ } (the null space of A). Consider the
space U/L, where
U/L = {u : u ∈ U},
where
u = {u + w : w ∈ L}.
Define A : U/L → R(A), by
A(u) = A(u).
Observe that A is one-to-one, linear, onto, and bounded. Moreover R(A) is closed so
that it is a Banach space. Hence by the inverse mapping theorem we have that A has
a continuous inverse. Thus, for any v ∈ R(A), there exists u ∈ U/L such that
A(u) = v
so that
−1
u=A (v),
and therefore
−1
u ≤ A v Y .
Recalling that
u = inf { u + w U },
w∈L
we may find u0 ∈ u such that

−1
u0 U ≤ 2 u ≤ 2 A v Y ,
and so that
A(u0 ) = A(u0 ) = A(u) = v.
−1
Taking K = 2 A we have completed the proof.
Theorem 3.1. Let U,Y be Banach spaces and let A : U → Y be a bound linear
operator. Assume R(A) is closed. Under such hypotheses
R(A∗ ) = [N(A)]⊥ .
Proof. Let u∗ ∈ R(A∗ ). Thus there exists v∗ ∈ Y ∗ such that
u∗ = A∗ (v∗ ).
Let u ∈ N(A). Hence,
u, u∗ U = u, A∗ (v∗ )U = A(u), v∗ Y = 0.
Since u ∈ N(A) is arbitrary, we get u∗ ∈ [N(A)]⊥ , so that
R(A∗ ) ⊂ [N(A)]⊥ .
Now suppose u∗ ∈ [N(A)]⊥ . Thus
u, u∗ U = 0, ∀u ∈ N(A).
Fix v ∈ R(A). From the Lemma 3.1, there exists K > 0 (which does not depend on v)
and uv ∈ U such that
A(uv ) = v
and
uv U ≤ K v Y .
Define f : R(A) → R by
f (v) = uv , u∗ U .
Observe that
| f (v)| ≤ uv U u∗ U ∗ ≤ K v Y u∗ U ∗ ,
so that f is a bounded linear functional. Hence by a Hahn–Banach theorem corollary

there exists v∗ ∈ Y ∗ such that
f (v) = v, v∗ Y ≡ F(v), ∀v ∈ R(A),
that is, F is an extension of f from R(A) to Y .

In particular
f (v) = uv , u∗ U = v, v∗ Y = A(uv ), v∗ Y ∀v ∈ R(A),
where A(uv ) = v, so that
uv , u∗ U = A(uv ), v∗ Y ∀v ∈ R(A).
Now let u ∈ U and define A(u) = v0 . Observe that
u = (u − uv0 ) + uv0 ,
and
A(u − uv0 ) = A(u) − A(uv0 ) = v0 − v0 = θ .
Since u∗ ∈ [N(A)]⊥ we get

u − uv0 , u∗ U = 0
so that
u, u∗ U = (u − uv0 ) + uv0 , u∗ U
= uv0 , u∗ U
= A(uv0 ), v∗ Y
= A(u − uv0 ) + A(uv0 ), v∗ Y
= A(u), v∗ Y . (3.3)
Hence,
u, u∗ U = A(u), v∗ Y , ∀u ∈ U.
We may conclude that u∗ = A∗ (v∗ ) ∈ R(A∗ ). Since u∗ ∈ [N(A)]⊥ is arbitrary we
obtain
[N(A)]⊥ ⊂ R(A∗ ).
We finish this section with the following result.
Definition 3.2.6. Let U be a Banach space and S ⊂ U. We define the positive con-
jugate cone of S, denoted by S⊕ by
S⊕ = {u∗ ∈ U ∗ : u, u∗ U ≥ 0, ∀u ∈ S}.
Similarly, we define the negative cone of S, denoted by S by
S = {u∗ ∈ U ∗ : u, u∗ U ≤ 0, ∀u ∈ S}.
Theorem 3.2.7. Let U,Y be Banach spaces and A : U → Y be a bounded linear

operator. Let S ⊂ U. Then
[A(S)]⊕ = (A∗ )−1 (S⊕ ),
where
(A∗ )−1 = {v∗ ∈ Y ∗ : A∗ v∗ ∈ S⊕ }.
Proof. Let v∗ ∈ [A(S)]⊕ and u ∈ S. Thus,
A(u), v∗ Y ≥ 0,
so that
u, A∗ (v∗ )U ≥ 0.
Since u ∈ S is arbitrary, we get
v∗ ∈ (A∗ )−1 (S⊕ ).
From this
[A(S)]⊕ ⊂ (A∗ )−1 (S⊕ ).
Reciprocally, let v∗ ∈ (A∗ )−1 (S⊕ ). Hence A∗ (v∗ ) ∈ S⊕ so that for u ∈ S we obtain
u, A∗ (v∗ )U ≥ 0,
and therefore
A(u), v∗ Y ≥ 0.
Since u ∈ S is arbitrary, we get v∗ ∈ [A(S)]⊕ , that is,
(A∗ )−1 (S⊕ ) ⊂ [A(S)]⊕ .
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, { un } is bounded. Thus, given v∗ ∈ Y ∗ we
have
v∗ , Aun Y = A∗ v∗ , un U
→ A∗ v∗ , uU
= v∗ , AuY . (3.4)
Being v∗ ∈ Y ∗ arbitrary, we get that
Aun Au weakly in Y. (3.5)
Suppose Aun does not converge in norm to Au. Thus there exists ε > 0 and a subse-
quence {Aunk } such that
Aunk − Au Y ≥ ε , ∀k ∈ N.
As {unk } is bounded and A is compact, {Aunk } has a subsequence converging para
ṽ = Au. But then such a sequence converges weakly to ṽ = Au, which contradicts
(3.5). 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.
Proof. Let A be a compact operator in H. Let {φ j } an orthonormal basis in H. For
each n ∈ N define
λn = sup{ Aψ H | ψ ∈ [φ1 , . . . , φn ]⊥ and ψ H = 1}.

3.3 Compact Operators 65
It is clear that {λn } is a nonincreasing sequence that converges to a limit λ ≥ 0. We

will show that λ = 0. Choose a sequence {ψn } such that
ψn ∈ [φ1 , . . . , φn ]⊥ ,
ψn H = 1, and Aψn H ≥ λ /2. Now we will show that
ψn θ , weakly in H.
Let ψ ∗ ∈ H ∗ = H,; thus there exists a sequence {a j } ⊂ C such that

∞
ψ∗ = ∑ a jφ j.
j=1
Suppose given ε > 0. We may find n0 ∈ N such that

∞
∑ |a j |2 < ε .
j=n0
Choose n > n0 . Hence there exists {b j } j>n such that

∞
ψn = ∑ b jφ j,
j=n+1
and
∞
∑ |b j |2 = 1.
j=n+1
Therefore

∞

|(ψn , ψ )H | = ∑ (φ j , φ j )H a j · b j
∗
j=n+1

∞

= ∑ a j · b j
j=n+1

∞ ∞
≤ ∑ |a j |2 ∑ |b j |2
j=n+1 j=n+1
√
≤ ε, (3.6)
if n > n0 . Since ε > 0 is arbitrary,
(ψn , ψ ∗ )H → 0, as n → ∞.
Since ψ ∗ ∈ H is arbitrary, we get
ψn θ , weakly in H.
Hence, as A is compact, we have
Aψn → θ in norm ,
so that λ = 0. Finally, we may define {An } by

n n
An (u) = A ∑ (u, φ j )H φ j = ∑ (u, φ j )H Aφ j ,
j=1 j=1
for each u ∈ H. Thus

A − An = λn → 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
Ez 2H = x 2H ≤ x 2H + y 2H = z 2H ,
so that
E ≤ 1.
Definition 3.4.2. Let A, B ∈ L (H). We write
A≥θ
if
(Au, u)H ≥ 0, ∀u ∈ H,
3.4 The Square Root of a Positive Operator 67
and in this case we say that A is positive. Finally, we denote
A≥B
if
A−B ≥ θ.
Theorem 3.4.3. Let A and B be bounded self-adjoint operators such that A ≥ θ and
B ≥ θ . If AB = BA, then
AB ≥ θ .
Proof. If A = θ , the result is obvious. Assume A = θ and define the sequence
A
A1 = , An+1 = An − A2n, ∀n ∈ N.
A
We claim that
θ ≤ An ≤ I, ∀n ∈ N.
We prove the claim by induction.
For n = 1, it is clear that A1 ≥ θ . And since A1 = 1, we get
(A1 u, u)H ≤ A1 u H u H = (Iu, u)H , ∀u ∈ H,
so that
A1 ≤ I.
Thus
θ ≤ A1 ≤ I.
Now suppose θ ≤ An ≤ I. Since An is self-adjoint, we have
(A2n (I − An )u, u)H = ((I − An )An u, An u)H

= ((I − An )v, v)H ≥ 0, ∀u ∈ H, (3.7)
where v = An u. Therefore
A2n (I − An ) ≥ θ .
Similarly, we may obtain
An (I − An )2 ≥ θ ,
so that
θ ≤ A2n (I − An) + An (I − An )2 = An − A2n = An+1 .
So, also we have
θ ≤ I − An + A2n = I − An+1,
that is,
θ ≤ An+1 ≤ I,
so that
θ ≤ An ≤ I, ∀n ∈ N.
Observe that
A1 = A21 + A2
= A21 + A22 + A3
... ...................
= A21 + . . . + A2n + An+1. (3.8)
Since An+1 ≥ θ , we obtain
A21 + A22 + . . . + A2n = A1 − An+1 ≤ A1 . (3.9)
From this, for a fixed u ∈ H, we have

n n
∑ A j u 2 = ∑ (A j u, A j u)H
j=1 j=1
n
= ∑ (A2j u, u)H
j=1
≤ (A1 u, u)H . (3.10)
Since n ∈ N is arbitrary, we get ∞
∑ A j u 2
j=1
is a converging series, so that
An u → 0,
that is,
An u → θ , as n → ∞.
From this and (3.9), we get
n
∑ A2j u = (A1 − An+1)u → A1u, as n → ∞.
j=1
Finally, we may write

(ABu, u)H = A (A1 Bu, u)H
= A (BA1 u, u)H
= A (B lim ∑ = 1n A2j u, u)H
n..
j
= A lim ∑ = 1n (BA2j u, u)H

n...
j
= A lim ∑ = 1n (BA j u, BA j u)H

n...
j
≥ 0. (3.11)
Hence
(ABu, u)H ≥ 0, ∀u ∈ H.
Theorem 3.4.4. 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
CiC j = C jCi , ∀i, j ∈ N.
Also, if n > m, then

An − Am ≥ θ
so that
Cm = B − Am ≥ B − An = Cn .
Therefore, from Cm ≥ θ and Cm − Cn ≥ θ , we obtain
(Cm − Cn )Cm ≥ θ , if n > m
and also
Cn (Cm − Cn ) ≥ θ .
Thus,
(Cm2 u, u)H ≥ (CnCm u, u)H ≥ (Cn2 u, u)H ,
and we may conclude that
(Cn2 u, u)H
is a monotone nonincreasing sequence of real numbers, bounded below by 0, so that
there exists α ∈ R such that
lim (Cn2 u, u)H = α .

n→∞
Since each Cn is self-adjoint we obtain
(Cn − Cm )u 2H = ((Cn − Cm )u, (Cn − Cm )u)H

= ((Cn − Cm )(Cn − Cm )u, u)H
= (Cn2 u, u)H − 2(CnCm u, u) + (Cm2 u, u)H
→ α − 2α + α = 0, (3.12)
as
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→∞ n→∞
and hence there exists

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

lim An u, ∀u ∈ H
n→∞
exists we have that

sup{ Anu H }
n∈N
is finite for all u ∈ H. By the principle of uniform boundedness
sup{ An } < ∞
n∈N
so that there exists K > 0 such that
An ≤ K, ∀n ∈ N.
Therefore
An u H ≤ K u H ,
so that
Au = lim { An u H } ≤ K u H , ∀u ∈ H
n→∞
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→∞ 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 3.4.5. 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 ≥ θ , we denote
√
B = A.
Theorem 3.4.6. Suppose A ∈ L (H) is positive. Then there exists B ≥ θ such that
B2 = A.
Furthermore B commutes with any C ∈ L (H) such that commutes with A.
Proof. There is no loss of generality in considering
A ≤ 1,
which means θ ≤ A ≤ I, because we may replace A by

A
A
so that if
A
C2 =
A
then
B = A 1/2C.
Let
B0 = θ ,
and consider the sequence of operators given by
1
Bn+1 = Bn + (A − B2n), ∀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
Bi B j = B j Bi , ∀i, j ∈ N.
First we show that

Bn ≤ I, ∀n ∈ N ∪ {0}.
Since B0 = θ , and B1 = 12 A, the statement holds for n = 1. Suppose Bn ≤ I. Thus
1 1
I − Bn+1 = I − Bn − A + B2n
2 2
1 1
= (I − Bn ) + (I − A) ≥ θ
2
(3.13)
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
B0 ≤ B1 ,
and supposing
Bn−1 ≤ Bn ,
we have
1 1
Bn+1 − Bn = Bn + (A − B2n) − Bn−1 − (A − B2n−1)
2 2
1
= Bn − Bn−1 − (B2n − B2n−1)
2
1
= Bn − Bn−1 − (Bn + Bn−1)(Bn − Bn−1)
2
1
= (I − (Bn + Bn−1))(Bn − Bn−1)
2
1
= ((I − Bn−1) + (I − Bn))(Bn − Bn−1) ≥ θ .
2
The induction is complete, that is,
θ = B0 ≤ B1 ≤ B2 ≤ . . . ≤ Bn ≤ . . . ≤ I.
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 − B2n)u,
2
so that taking the limit in norm as n → ∞, we get
θ = (A − B2 )u.
3.5 About the Spectrum of a Linear Operator 73
Being u ∈ H arbitrary we obtain

A = B2 .
It is also clear that
B≥θ
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
λ is said to be in the resolvent set ρ (A) of A, if
λI −A
is a bijection with a bounded inverse. We call
Rλ (A) = (λ I − A)−1
the resolvent of A in λ .
If λ ∈ ρ (A), we write
λ ∈ σ (A) = C − ρ (A),
where σ (A) is said to be the spectrum of A.
Definition 3.5.2. Let A ∈ L (U).

1. If u = θ and Au = λ u for some λ ∈ C, then u is said to be an eigenvector of A
and λ the corresponding eigenvalue. If λ is an eigenvalue, then (λ I − A) is not
injective and therefore λ ∈ σ (A).
The set of eigenvalues is said to be the point spectrum of A.
2. If λ is not an eigenvalue but
R(λ I − A)
is not dense in U and therefore λ I −A is not a bijection, we have that λ ∈ σ (A). In
this case we say that λ is in the residual spectrum of A, or briefly λ ∈ Res[σ (A)].
Theorem 3.5.3. Let U be a Banach space and suppose that A ∈ L (U). Then ρ (A)
is an open subset of C and
F(λ ) = Rλ (A)
is an analytic function with values in L (U) on each connected component of ρ (A).
For λ , μ ∈ σ (A), Rλ (A), and Rμ (A) commute and
Rλ (A) − Rμ (A) = (μ − λ )Rμ (A)Rλ (A).
Proof. Let λ0 ∈ ρ (A). We will show that λ0 is an interior point of ρ (A).

Observe that symbolically we may write
1 1
=
λ −A λ − λ0 + (λ0 − A)
⎡ ⎤
1 ⎣ 1
= ⎦
λ 0 − A 1 − λ0 −λ
λ0 −A

∞
1 λ0 − λ n
= 1+ ∑ . (3.14)
λ0 − A n=1 λ0 − A
Define

∞
R̂λ (A) = Rλ0 (A) I + ∑ (λ − λ0) (Rλ0 ) n n
. (3.15)
n=1
Observe that
(Rλ0 )n ≤ Rλ0 n .
Thus, the series indicated in (3.15) will converge in norm if
|λ − λ0 | < Rλ0 −1 . (3.16)
Hence, for λ satisfying (3.16), R̂(A) is well defined and we can easily check that
(λ I − A)R̂λ (A) = I = R̂λ (A)(λ I − A).
Therefore
R̂λ (A) = Rλ (A), if |λ − λ0| < Rλ0 −1 ,
so that λ0 is an interior point. Since λ0 ∈ ρ (A) is arbitrary, we have that ρ (A) is
open. Finally, observe that
Rλ (A) − Rμ (A) = Rλ (A)(μ I − A)Rμ (A) − Rλ (A)(λ I − A)Rμ (A)

= Rλ (A)(μ I)Rμ (A) − Rλ (A)(λ I)Rμ (A)
= (μ − λ )Rλ (A)Rμ (A). (3.17)
Interchanging the roles of λ and μ we may conclude that Rλ and Rμ commute.
Corollary 3.5.4. Let U be a Banach space and A ∈ L (U). Then the spectrum of A
is nonempty.
Proof. Observe that if

A
<1
|λ |
we have
(λ I − A)−1 = [λ (I − A/λ )]−1

= λ −1 (I − A/λ )−1
3.5 About the Spectrum of a Linear Operator 75

∞ n
A
= λ −1 I+ ∑ . (3.18)
n=1 λ
Therefore we may obtain

n
∞
A
Rλ (A) = λ −1
I+ ∑ .
n=1 λ
In particular
Rλ (A) → 0, as |λ | → ∞. (3.19)
Suppose, to obtain contradiction, that
σ (A) = 0.
/
In such a case Rλ (A) would be an entire bounded analytic function. From Liouville’s
theorem, Rλ (A) would be constant, so that from (3.19) we would have
Rλ (A) = θ , ∀λ ∈ C,
which is a contradiction.
Proposition 3.5.5. Let H be a Hilbert space and A ∈ L (H).
1. If λ ∈ Res[σ (A)], then λ ∈ Pσ (A∗ ).
2. If λ ∈ Pσ (A), then λ ∈ Pσ (A∗ ) ∪ Res[σ (A∗ )].
Proof.
1. If λ ∈ Res[σ (A)], then
R(A − λ I) = H.
Therefore there exists v ∈ (R(A − λ I))⊥, v = θ such that
(v, (A − λ I)u)H = 0, ∀u ∈ H
that is,
((A∗ − λ I)v, u)H = 0, ∀u ∈ H
so that
(A∗ − λ I)v = θ ,
which means that λ ∈ Pσ (A∗ ).
2. Suppose there exists v = θ such that
(A − λ I)v = θ ,
and
λ ∈ Pσ (A∗ ).
Thus
(u, (A − λ I)v))H = 0, ∀u ∈ H,
so that
((A∗ − λ I)u, v)H , ∀u ∈ H.
Since
(A∗ − λ I)u = θ , ∀u ∈ H, u = θ ,
we get v ∈ (R(A∗ − λ I))⊥ , so that R(A∗ − λ I) = H.
Hence λ ∈ Res[σ (A∗ )].
Theorem 3.5.6. Let A ∈ L (H) be a self-adjoint operator, then

1. σ (A) ⊂ R.
2. Eigenvectors corresponding to distinct eigenvalues of A are orthogonal.
Proof. Let μ , λ ∈ R. Thus, given u ∈ H we have
(A − (λ + μ i))u 2 = (A − λ )u 2 + μ 2 u 2,
so that
(A − (λ + μ i))u 2 ≥ μ 2 u 2.
Therefore if μ = 0, A − (λ + μ i) has a bounded inverse on its range, which is closed.
If R(A − (λ + μ i)) = H, then by the last result (λ − μ i) would be in the point spec-
trum of A, which contradicts the last inequality. Hence, if μ = 0, then λ + μ i ∈ ρ (A).
To complete the proof, suppose
Au1 = λ1 u1 ,
and
Au2 = λ2 u2 ,
where
λ1 , λ2 ∈ R, λ1 = λ2 , and u1 , u2 = θ .
Thus
(λ1 − λ2)(u1 , u2 )H = λ1 (u1 , u2 )H − λ2 (u1 , u2 )H

= (λ1 u1 , u2 )H − (u1 , λ2 u2 )H
= (Au1 , u2 )H − (u1 , Au2 )H
= (u1 , Au2 )H − (u1 , Au2 )H
= 0. (3.20)
Since λ1 − λ2 = 0 we get
(u1 , u2 )H = 0.
3.6 The Spectral Theorem for Bounded Self-Adjoint Operators 77
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 | u H = 1},

u∈H
and
M = sup{(Au, u)H | u H = 1}.
u∈H
Remark 3.6.1. It is possible to prove that for a linear self-adjoint operator A : H → H

we have
A = sup{|(Au, u)H | | u ∈ H, u H = 1}.
This propriety, which prove in the next lines, is crucial for the subsequent results,
since, for example, for A, B linear and self-adjoint and ε > 0, we have
−ε I ≤ A − B ≤ ε I,
we also would have

A − B < ε .
So, we present the following basic result.
Theorem 3.6.2. Let A : H → H be a bounded linear self-adjoint operator. Define
α = max{|m|, |M|},
where
m = inf {(Au, u)H | u H = 1},
u∈H
and
M = sup{(Au, u)H | u H = 1}.
u∈H
Then
A = α .
Proof. Observe that
(A(u + v), u + v)H = (Au, u)H + (Av, v)H + 2(Au, v)H ,
and
(A(u − v), u − v)H = (Au, u)H + (Av, v)H − 2(Au, v)H .
Thus,
4(Au, v) = (A(u + v), u + v)H − (A(u − v), u − v)H ≤ M u + v U2 − m u − v U2 ,

so that
4(Au, v)H ≤ α ( u + v U2 + u − v U2 ).
Hence, replacing v by −v, we obtain
−4(Au, v)H ≤ α ( u + v U2 + u − v U2 ),
and therefore
4|(Au, v)H | ≤ α ( u + v U2 + u − v U2 ).
Replacing v by β v, we get
4|(A(u), v)H | ≤ 2α ( u U2 /β + β v U2 ).
Minimizing the last expression in β > 0, for the optimal
β = u U / v U ,
we obtain
|(Au, v)H | ≤ α u U v U , ∀u, v ∈ U.
Thus
A ≤ α .
On the other hand,
|(Au, u)H | ≤ A u U2 ,
so that
|M| ≤ A
and
|m| ≤ A ,
so that
α ≤ A .
At this point we start to develop the spectral theory. Define by P the set of all real
polynomials defined in R. Define
Φ1 : P → L (H),
by
Φ1 (p(λ )) = p(A), ∀p ∈ P.
Thus we have
1. Φ1 (p1 + p2 ) = p1 (A) + p2(A),
2. Φ1 (p1 · p2 ) = p1 (A)p2 (A),
3. Φ1 (α p) = α p(A), ∀α ∈ R, p ∈ P,
4. if p(λ ) ≥ 0, on [m, M], then p(A) ≥ θ .
We will prove (4):

Consider p ∈ P. Denote the real roots of p(λ ) less or equal to m by α1 , α2 , . . . , αn
and denote those that are greater or equal to M by β1 , β2 , . . . , βl . Finally denote all
the remaining roots, real or complex, by
v1 + i μ 1 , . . . , vk + i μ k .
Observe that if μi = 0, then vi ∈ (m, M). The assumption that p(λ ) ≥ 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 repre-
sentation for p(λ ) :
n l k
p(λ ) = a ∏(λ − αi ) ∏(βi − λ ) ∏((λ − vi )2 + μi2 ),
i=1 i=1 i=1
where a ≥ 0. Observe that

A − αi I ≥ θ ,
since
(Au, u)H ≥ m(u, u)H ≥ αi (u, u)H , ∀u ∈ H,
and by analogy
βi I − A ≥ θ .
On the other hand, since A − vk I is self-adjoint, its square is positive, and hence
since the sum of positive operators is positive, we obtain
(A − vk I)2 + μk2I ≥ θ .
Therefore,
p(A) ≥ θ .
The idea is now to extend the domain of Φ1 to the set of upper semicontinuous
functions, and such set we will denote by Cup .
Observe that if f ∈ Cup , there exists a sequence of continuous functions {gn }
such that
gn ↓ f , pointwise ,
that is,
gn (λ ) ↓ f (λ ), ∀λ ∈ R.
Considering the Weierstrass Theorem, since gn ∈ C([m, M]), we may obtain a se-
quence of polynomials {pn } such that

gn + 1 − pn < 1 ,
2 n 2n
∞
where the norm · ∞ refers to [m, M]. Thus
pn (λ ) ↓ f (λ ), on [m, M].
Therefore
p1 (A) ≥ p2 (A) ≥ p3 (A) ≥ . . . ≥ pn (A) ≥ . . .
Since pn (A) is self-adjoint for all n ∈ N, we have
p j (A)pk (A) = pk (A)p j (A), ∀ j, k ∈ N.
Then the lim pn (A) (in norm) exists, and we denote

n→∞
lim pn (A) = f (A).

n→∞
Now recall the Dini’s theorem.

Theorem 3.6.3 (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.
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).
n→∞ n→∞
First observe that being {pn } and {qn } sequences of continuous functions we have
that
ĥnk (λ ) = max{pn (λ ), qk (λ )}, ∀λ ∈ [m, M]
is also continuous, ∀n, k ∈ N. Now fix n ∈ N and define
hk (λ ) = max{pk (λ ), qn (λ )}.
Observe that
hk (λ ) ↓ qn (λ ), ∀λ ∈ R,
so that by Dini’s 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 (λ ) − qn(λ ) ≤ , ∀λ ∈ [m, M].
n
Since
pk (λ ) ≤ hk (λ ), ∀λ ∈ [m, M],
we obtain
1
pk (λ ) − qn (λ ) ≤ , ∀λ ∈ [m, M].
n
By analogy, we may show that for each n ∈ N there exists k̂n ∈ N such that if k > k̂n ,
then
1
qk (λ ) − pn(λ ) ≤ .
n
From above we obtain
1
lim pk (A) ≤ qn (A) + .
k→∞ n
Since the self-adjoint qn (A) + 1/n commutes with the
lim pk (A)
k→∞
we obtain

1
lim pk (A) ≤ lim qn (A) +
k→∞ n→∞ n
≤ lim qn (A). (3.21)
n→∞
Similarly we may obtain
lim qk (A) ≤ lim pn (A),

k→∞ n→∞
so that
lim qn (A) = lim pn (A) = f (A).
n→∞ n→∞
Hence, we may extend Φ1 : P → L (H) to Φ2 : Cup → L (H), where Cup , as earlier

indicated, denotes the set of upper semicontinuous functions, where
Φ2 ( f ) = f (A).
Observe that Φ2 has the following properties:

1. Φ2 ( f1 + f2 ) = Φ2 ( f1 ) + Φ2 ( f2 ),
2. Φ2 ( f1 · f2 ) = f1 (A) f2 (A),
3. Φ2 (α f ) = αΦ2 ( f ), ∀α ∈ R, α ≥ 0,
4. if f1 (λ ) ≥ f2 (λ ), ∀λ ∈ [m, M], then
f1 (A) ≥ f2 (A).
The next step is to extend Φ2 to Φ3 : C−

up
→ L (H), where
up
C− = { f − g | f , g ∈ Cup }.
up
For h = f − g ∈ C− we define
Φ3 (h) = f (A) − g(A).
Now we will show that Φ3 is well defined. Suppose that h ∈ C−

up
and
h = f1 − g1 and h = f2 − g2.
Thus
f 1 − g1 = f 2 − g2 ,
that is
f 1 + g2 = f 2 + g1 ,
so that from the definition of Φ2 we obtain
f1 (A) + g2(A) = f2 (A) + g1 (A),
that is,
f1 (A) − g1(A) = f2 (A) − g2 (A).
Therefore Φ3 is well defined. Finally observe that for α < 0
α ( f − g) = −α g − (−α ) f ,
where −α g ∈ Cup and −α f ∈ Cup . Thus
Φ3 (α f ) = α f (A) = αΦ3 ( f ), ∀α ∈ R.
3.6.1 The Spectral Theorem
Consider the upper semicontinuous function

1, if λ ≤ μ ,
hμ (λ ) = (3.22)
0, if λ > μ .
Denote
E(μ ) = Φ3 (h μ ) = hμ (A).
Observe that
h μ (λ )h μ (λ ) = h μ (λ ), ∀λ ∈ R,
so that
[E(μ )]2 = E(μ ), ∀μ ∈ R.
Therefore
{E(μ ) | μ ∈ R}
is a family of orthogonal projections. Also observe that if ν ≥ μ , we have
hν (λ )h μ (λ ) = hμ (λ )hν (λ ) = h μ (λ ),
so that
E(ν )E(μ ) = E(μ )E(ν ) = E(μ ), ∀ν ≥ μ .
If μ < m, then h μ (λ ) = 0, on [m, M], so that
E(μ ) = 0, if μ < m.
Similarly, if μ ≥ M, then hμ (λ ) = 1, on [m, M], so that
E(μ ) = I, if μ ≥ M.
Next we show that the family {E(μ )} is strongly continuous from the right. First
we will establish a sequence of polynomials {pn } such that
pn ↓ h μ
and
pn (λ ) ≥ hμ + 1 (λ ), on [m, M].
n
Observe that for any fixed n there exists a sequence of polynomials {pnj } such that
pnj ↓ hμ +1/n , point-wise.
Consider the monotone sequence
gn (λ ) = min{prs (λ ) | r, s ∈ {1, . . . , n}}.
Thus
gn (λ ) ≥ h μ + 1 (λ ), ∀λ ∈ R,
n
and we obtain
lim gn (λ ) ≥ lim hμ + 1 (λ ) = hμ (λ ).
n→∞ n→∞ n
On the other hand
gn (λ ) ≤ prn (λ ), ∀λ ∈ R, ∀r ∈ {1, . . . , n},
so that
lim gn (λ ) ≤ lim prn (λ ).
n→∞ n→∞
Therefore
lim gn (λ ) ≤ lim lim prn (λ )

n→∞ r→∞ n→∞
= hμ (λ ). (3.23)
Thus
lim gn (λ ) = h μ (λ ).
n→∞
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 (λ ) + 1/n − pn(λ )| < , ∀λ ∈ [m, M], n ∈ N,
2n
so that
pn (λ ) ≥ gn (λ ) + 1/n − 1/2n ≥ gn (λ ) ≥ h μ +1/n (λ ).
Thus
pn (A) → E(μ ),
and
pn (A) ≥ h μ + 1 (A) = E(μ + 1/n) ≥ E(μ ).
n
Therefore we may write
E(μ ) = lim pn (A) ≥ lim E(μ + 1/n) ≥ E(μ ).

n→∞ n→∞
Thus
lim E(μ + 1/n) = E(μ ).
n→∞
From this we may easily obtain the strong continuity from the right.
For μ ≤ ν we have
μ (hν (λ ) − hμ (λ )) ≤ λ (hν (λ ) − hμ (λ ))
≤ ν (hν (λ ) − hμ (λ )). (3.24)
To verify this observe that if λ < μ or λ > ν , then all terms involved in the above
inequalities are zero. On the other hand if
μ ≤λ ≤ν
then
hν (λ ) − hμ (λ ) = 1,
so that in any case (3.24) holds. From the monotonicity property we have
μ (E(ν ) − E(μ )) ≤ A(E(ν ) − E(μ ))

≤ ν (E(ν ) − E(μ )). (3.25)
Now choose a, b ∈ R such that
a < m and b ≥ M.
Suppose given ε > 0. Choose a partition P0 of [a, b], that is,
P0 = {a = λ0 , λ1 , . . . , λn = b},
such that
max {|λk − λk−1 |} < ε .
k∈{1,...,n}
Hence
λk−1 (E(λk ) − E(λk−1)) ≤ A(E(λk ) − E(λk−1))

≤ λk (E(λk ) − E(λk−1)). (3.26)
Summing up on k and recalling that

n
∑ E(λk ) − E(λk−1) = I,
k=1
we obtain
n
∑ λk−1(E(λk ) − E(λk−1)) ≤ A
k=1
n
≤ ∑ λk (E(λk ) − E(λk−1)). (3.27)
k=1
Let λk0 ∈ [λk−1 , λk ]. Since (λk − λk0) ≤ (λk − λk−1) from (3.26) we obtain
n n
A − ∑ λk0 (E(λk ) − E(λk−1)) ≤ ε ∑ (E(λk ) − E(λk−1))
k=1 k=1
= ε I. (3.28)
By analogy
n
− ε I ≤ A − ∑ λk0 (E(λk ) − E(λk−1)). (3.29)
k=1
Since
n
A − ∑ λk0 (E(λk ) − E(λk−1))
k=1
is self-adjoint we obtain
n
A − ∑ λk0 (E(λk ) − E(λk−1)) < ε .
k=1
Being ε > 0 arbitrary, we may write

b
A= λ dE(λ ),
a
that is, M
A= λ dE(λ ).
m−
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 ∗ .
Theorem 3.7.2. Every unitary transformation U has a spectral decomposition
2π
U= eiφ dE(φ ),
0−
where {E(φ )} is a spectral family on [0, 2π ]. Furthermore E(φ ) 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
p(eiφ ) = ∑ ck eikφ ,
k=−n
consider the transformation

n
p(U) = ∑ ck U k ,
k=−n
where ck ∈ C, ∀k ∈ {−n, . . ., 0, . . . , n}.

Observe that
n
p(eiφ ) = ∑ ck e−ikφ ,
k=−n
so that the corresponding operator is

n n
p(U)∗ = ∑ ckU −k = ∑ ck (U ∗ )k .
k=−n k=−n
Also if
p(eiφ ) ≥ 0
3.7 The Spectral Decomposition of Unitary Transformations 87
there exists a polynomial q such that
p(eiφ ) = |q(eiφ )|2 = q(eiφ )q(eiφ ),
so that
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 hμ (φ ) by

1, if 2kπ < φ ≤ 2kπ + μ ,
h μ (φ ) = (3.30)
0, if 2kπ + μ < φ ≤ 2(k + 1)π ,
for each k ∈ {0, ±1, ±2, ±3, . . .}. Define E(μ ) = hμ (U). Observe that the family
{E(μ )} are projections and in particular
E(0) = 0,
E(2π ) = I
and if μ ≤ ν , since
h μ (φ ) ≤ hν (φ ),
we have
E(μ ) ≤ E(ν ).
Suppose given ε > 0. Let P0 be a partition of [0, 2π ], that is,
P0 = {0 = φ0 , φ1 , . . . , φn = 2π }
such that
max {|φ j − φ j−1 |} < ε .
j∈{1,...,n}
For fixed φ ∈ [0, 2π ], let j ∈ {1, . . . , n} be such that
φ ∈ [φ j−1 , φ j ].
n
|eiφ − ∑ eiφk (hφk (φ ) − hφk−1 (φ ))| = |eiφ − eiφ j |
k=1
≤ |φ − φ j | < ε . (3.31)
Thus,
n
0 ≤ |eiφ − ∑ eiφk (hφk (φ ) − hφk−1 (φ ))|2 ≤ ε 2
k=1
so that, for the corresponding operators
n n
0 ≤ [U − ∑ eiφk (E(φk ) − E(φk−1 ))]∗ [U − ∑ eiφk (E(φk ) − E(φk−1 ))]
k=1 k=1
≤ ε 2I
and hence
n

U − ∑ e (E(φk ) − E(φk−1 ) < ε .
iφk
k=1

Being ε > 0 arbitrary, we may infer that
2π
U= eiφ dE(φ ).
0
3.8 Unbounded Operators
3.8.1 Introduction
Let H be a Hilbert space. Let A : D(A) → H be an operator, where unless indi-

cated 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 Γ (A), by
Γ (A) = {(u, Au) | u ∈ D}.
Definition 3.8.2. An operator A : D → H is said to be closed if Γ (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 denoted 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
Γ (A) = Γ (A).
3.8 Unbounded Operators 89
Proof. Suppose B is a closed extension of A. Then
Γ (A) ⊂ Γ (B) = Γ (B),
so that if (θ , φ ) ∈ Γ (A), then (θ , φ ) ∈ Γ (B), and hence φ = θ . Define the operator

C by
D(C) = {ψ | (ψ , φ ) ∈ Γ (A) for some φ },
and C(ψ ) = φ , where φ is the unique point such that (ψ , φ ) ∈ Γ (A). Hence
Γ (C) = Γ (A) ⊂ Γ (B),
so that
A ⊂ C.
However C ⊂ B and since B is an arbitrary closed extension of A we have
C=A
so that
Γ (C) = Γ (A) = Γ (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∗ ) = {φ ∈ H | (Aψ , φ )H = (ψ , η )H , ∀ψ ∈ D for some η ∈ H}.
In this case we denote

A∗ φ = η .
A∗ defined in this way is called the adjoint operator related to A.
Observe that by the Riesz lemma, φ ∈ D(A∗ ) if and only if there exists K > 0 such
that
|(Aψ , φ )H | ≤ K ψ H , ∀ψ ∈ D.
Also note that if
A ⊂ B then B∗ ⊂ A∗ .
Finally, as D is dense in H, then
η = A∗ (φ )
is uniquely defined. However the domain of A∗ may not be dense, and in some
situations we may have D(A∗ ) = {θ }.
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∗ .

Proof.
1. We define the operator V : H × H → H × H by
V (φ , ψ ) = (−ψ , φ ).
Let E ⊂ H × H be a subspace. Thus, if (φ1 , ψ1 ) ∈ V (E ⊥ ), then there exists

(φ , ψ ) ∈ E ⊥ such that
V (φ , ψ ) = (−ψ , φ ) = (φ1 , ψ1 ).
Hence
ψ = −φ1 and φ = ψ1 ,
so that for (ψ1 , −φ1 ) ∈ E⊥ and (w1 , w2 ) ∈ E we have
((ψ1 , −φ1 ), (w1 , w2 ))H×H = 0 = (ψ1 , w1 )H + (−φ1 , w2 )H .
Thus
(φ1 , −w2 )H + (ψ1 , w1 )H = 0,
and therefore
((φ1 , ψ1 ), (−w2 , w1 ))H×H = 0,
that is,
((φ1 , ψ1 ),V (w1 , w2 ))H×H = 0, ∀(w1 , w2 ) ∈ E.
This means that
(φ1 , ψ1 ) ∈ (V (E))⊥ ,
so that
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 (φ , η ) ∈ H × H. Thus, (φ , η ) ∈ V (Γ (A))⊥ if and only if
((φ , η ), (−Aψ , ψ ))H×H = 0, ∀ψ ∈ D,
which holds if and only if

3.9 Symmetric and Self-Adjoint Operators 91
(φ , Aψ )H = (η , ψ )H , ∀ψ ∈ D,
that is, if and only if

(φ , η ) ∈ Γ (A∗ ).
Thus
Γ (A∗ ) = V (Γ (A))⊥ .
Since (V (Γ (A))⊥ is closed, A∗ is closed.
2. Observe that Γ (A) is a linear subset of H × H so that
Γ (A) = [Γ (A)⊥ ]⊥
= V 2 [Γ (A)⊥ ]⊥
= [V [V (Γ (A))⊥ ]]⊥
= [V (Γ (A∗ )]⊥ (3.32)
so that from the proof of item 1, if A∗ is densely defined, we get
Γ (A) = Γ [(A∗ )∗ ].
Conversely, suppose D(A∗ ) is not dense. Thus there exists ψ ∈ [D(A∗ )]⊥ such
that ψ = θ . Let (φ , A∗ φ ) ∈ Γ (A∗ ). Hence
((ψ , θ ), (φ , A∗ φ ))H×H = (ψ , φ )H = 0,
so that
(ψ , θ ) ∈ [Γ (A∗ )]⊥ .
Therefore V [Γ (A∗ )]⊥ is not the graph of a linear operator. Since Γ (A) =
V [Γ (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∗ φ = Aφ , ∀φ ∈ D.
Equivalently, A is symmetric if and only if
(Aφ , ψ )H = (φ , Aψ )H , ∀φ , ψ ∈ 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 self-adjoint. If A is closed, a subset
E ⊂ D is said to be a core for A if A|E = A.
Theorem 3.9.4. Let A : D → H be a symmetric operator. Then the following state-

ments are equivalent:
1. A is self-adjoint,
2. A is closed and N(A∗ ± iI) = {θ },
3. R(A ± iI) = H.
Proof.
• 1 implies 2:
Suppose A is self-adjoint, let φ ∈ D = D(A∗ ) be such that
Aφ = iφ
so that
A∗ φ = iφ .
Observe that
− i(φ , φ )H = (iφ , φ )H
= (Aφ , φ )H
= (φ , Aφ )H
= (φ , iφ )H
= i(φ , φ )H , (3.33)
so that (φ , φ )H = 0, that is, φ = θ . Thus
N(A − iI) = {θ }.
Similarly we prove that N(A + iI) = {θ }. Finally, since A∗ = A∗ = A, we get that

A = A∗ is closed.
• 2 implies 3:
Suppose 2 holds. Thus the equation
A∗ φ = −iφ
has no nontrivial solution. We will prove that R(A − iI) is dense in H. If ψ ∈

R(A − iI)⊥, then
((A − iI)φ , ψ )H = 0, ∀φ ∈ D,
so that ψ ∈ D(A∗ ) and
(A − iI)∗ψ = (A∗ + iI)ψ = θ ,
and hence by above ψ = θ . Now we will prove that R(A − iI) is closed and
conclude that
R(A − iI) = H.
Given φ ∈ D we have
(A − iI)φ 2H = Aφ 2H + φ 2H . (3.34)
Let ψ0 ∈ H be a limit point of R(A − iI). Thus we may find {φn } ⊂ D such that
(A − iI)φn → ψ0 .
From (3.34)
φn − φm H ≤ (A − iI)(φn − φm ) H , ∀m, n ∈ N
so that {φn } is a Cauchy sequence, therefore converging to some φ0 ∈ H. Also

from (3.34)
Aφn − Aφm H ≤ (A − iI)(φn − φm ) H , ∀m, n ∈ N
so that {Aφn } is a Cauchy sequence, hence also a converging one. Since A is

closed, we get φ0 ∈ D and
(A − iI)φ0 = ψ0 .
Therefore R(A − iI) is closed, so that
R(A − iI) = H.
Similarly
R(A + iI) = H.
• 3 implies 1: Let φ ∈ D(A∗ ). Since R(A − iI) = H, there is an η ∈ D such that
(A − iI)η = (A∗ − iI)φ ,
and since D ⊂ D(A∗ ) we obtain φ − η ∈ D(A∗ ) and
(A∗ − iI)(φ − η ) = θ .
Since R(A + iI) = H we have N(A∗ − iI) = {θ }. Therefore φ = η , 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
2π
U= eiφ dF(φ ),
0
where {F(φ )} is a monotone family of orthogonal projections, strongly continuous

from the right and we may consider it such that

0, if φ ≤ 0,
F(φ ) = (3.35)
I, if φ ≥ 2π .
Since F(φ ) = 0, for all φ ≤ 0 and
F(0) = F(0+ )
we obtain
F(0+ ) = 0 = F(0− ),
that is, F(φ ) is continuous at φ = 0. We claim that F is continuous at φ = 2π .
Observe that F(2π ) = F(2π + ) so that we need only to show that
F(2π − ) = F(2π ).
Suppose
F(2π ) − F(2π − ) = θ .
Thus, there exists some u, v ∈ H such that
(F(2π ) − F(2(π − )))u = v = θ .
Therefore
F(φ )v = F(φ )[(F(2π ) − F(2π − ))u],
so that

0, if φ < 2π ,
F(φ )v = (3.36)
v, if φ ≥ 2π .
Observe that
2π
U −I = (eiφ − 1)dF(φ ),
0
and
2π
U∗ − I = (e−iφ − 1)dF(φ ).
0
Let {φn } be a partition of [0, 2π ]. From the monotonicity of [0, 2π ] and pairwise
orthogonality of
{F(φn ) − F(φn−1 )}
we can show that (this is not proved in details here)
2π
(U ∗ − I)(U − I) = (e−iφ − 1)(eiφ − 1)dF(φ ),
0
so that, given z ∈ H, we have

2π
((U ∗ − I)(U − I)z, z)H = |eiφ − 1|2 d F(φ )z 2 ,
0
thus, for v defined above
(U − I)v 2 = ((U − I)v, (U − I)v)H

= ((U − I)∗ (U − I)v, v)H
2π
= |eiφ − 1|2d F(φ )v 2
0
2π −
= |eiφ − 1|2d F(φ )v 2
0
= 0. (3.37)
The last two equalities result from e2π i − 1 = 0 and d F(φ )v = θ on [0, 2π ). Since
v = θ the last equation implies that 1 ∈ Pσ (U), which contradicts the existence of
(I − U)−1.
Thus, F is continuous at φ = 2π .
Now choose a sequence of real numbers {φn } such that φn ∈ (0, 2π ), n =
0, ±1, ±2, ±3, . . . such that
φn
− cot = n.
2
Now define Tn = F(φn ) − F(φn−1 ). Since U commutes with F(φ ), 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
∑ Tn = ∑(F(φn ) − F(φn−1 ))
n n
= lim F(φ ) − lim F(φ )
φ →2π φ →0
= I − θ = I. (3.38)
Hence
∑ R(Tn ) = H.
n
Also, for u ∈ H, we have that

⎧
⎨ 0, if φ < φn−1 ,
F(φ )Tn u = (F(φ ) − F(φn−1 ))u, if φn−1 ≤ φ ≤ φn , (3.39)
⎩
F(φn ) − F(φn−1 ))u, if φ > φn ,
so that
2π
(I − U)Tnu = (1 − eiφ )dF(φ )Tn u
0
φn
= (1 − eiφ )dF(φ )u. (3.40)
φn−1
Therefore
φn
(1 − eiφ )−1 dF(φ )(I − U)Tn u
φn−1
φn φn
= (1 − eiφ )−1 dF(φ ) (1 − eiφ )dF(φ )u
φn−1 φn−1
φn
= (1 − eiφ )−1 (1 − eiφ )dF(φ )u
φn−1
φn
= dF(φ )u
φn−1
2π
= dF(φ )Tn u = Tn u. (3.41)
0
Hence
−1 φn
(I − U)|R(Tn ) = (1 − eiφ )−1 dF(φ ).
φn−1
From this, from above, and as
A = i(I + U)(I − U)−1

we obtain
φn
ATn u = i(1 + eiφ )(1 − eiφ )−1 dF(φ )u.
φn−1
Therefore defining

φ
λ = − cot ,
2
and
E(λ ) = F(−2 cot−1 λ ),
we get
iφ iφ −1 φ
i(1 + e )(1 − e ) = − cot = λ.
2
Hence,
n
ATn u = λ dE(λ )u.
n−1
Finally, from
∞
u= ∑ Tn u,
n=−∞
we can obtain
∞
Au = A( ∑ Tn u)
n=−∞
∞
= ∑ ATn u
n=−∞
∞ n
= ∑ λ dE(λ )u. (3.42)
n=−∞ n−1
Being the convergence in question in norm, we may write

∞
Au = λ dE(λ )u.
−∞
Since u ∈ H is arbitrary, we may denote

∞
A= λ dE(λ ). (3.43)
−∞
Chapter 4
Basic Results on Measure and Integration
The main references for this chapter are Rudin [57], Royden [59], and Stein and
Shakarchi [62], where more details may be found. All these three books are excellent
and we strongly recommend their reading.
4.1 Basic Concepts
In this chapter U denotes a topological space.

Definition 4.1.1 (σ -algebra). A collection M of subsets of U is said to be a σ -
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 σ -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 0/ = U \ U so that from 1 and 2 in Definition 4.1.1, we have that
0/ ∈ 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}.
3. Since ∩∞ ∞
i=1 Ai = (∪i=1 Ai ) also from Definition 4.1.1, it is clear that M is closed
c c
under countable intersections.

4. Since A \ B = Bc ∩ A we obtain : if A, B ∈ M , then A \ B ∈ M .

100 4 Basic Results on Measure and Integration
Theorem 4.1.5. Let F be any collection of subsets of U. Then there exists a smallest
σ -algebra M0 in U such that F ⊂ M0 .
Proof. Let Ω be the family of all σ -algebras that contain F . Since the set of all
subsets in U is a σ -algebra, Ω is nonempty.
Let M0 = ∩Mλ ⊂Ω Mλ ; it is clear that M0 ⊃ F , and it remains to prove that in
fact M0 is a σ -algebra. Observe that:
1. U ∈ Mλ , ∀Mλ ∈ Ω , so that U ∈ M0 ,
2. A ∈ M0 implies A ∈ Mλ , ∀Mλ ∈ Ω , so that Ac ∈ Mλ , ∀Mλ ∈ Ω , which means
Ac ∈ M0 ,
3. {An } ⊂ M0 implies {An } ⊂ Mλ , ∀Mλ ∈ Ω , so that ∪∞
n=1 An ∈ Mλ , ∀Mλ ∈ Ω ,
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 σ -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 σ -algebra in U and V is a topological space. For

f : U → V , we have:
1. If Ω = {E ⊂ V | f −1 (E) ∈ M }, then Ω is a σ -algebra.
2. If V = [−∞, ∞], and f −1 ((α , ∞]) ∈ M , for each α ∈ R, then f is measurable.
Proof.
1.(a) V ∈ Ω since f −1 (V ) = U and U ∈ M .
(b) E ∈ Ω ⇒ f −1 (E) ∈ M ⇒ U \ f −1 (E) ∈ M ⇒ f −1 (V \ E) ∈ M ⇒ V \ E ∈ Ω .
(c) {Ei } ⊂ Ω ⇒ f −1 (Ei ) ∈ M , ∀i ∈ N ⇒ ∪∞ i=1 f
−1 (E ) ∈ M ⇒ f −1 (∪∞ E ) ∈
i i=1 i
∞
M ⇒ ∪i=1 Ei ∈ Ω .
Thus Ω is a σ -algebra.
2. Define Ω = {E ⊂ [−∞, ∞] | f −1 (E) ∈ M }. From above Ω is a σ -algebra. Given
α ∈ R, let {αn } be a real sequence such that αn → α as n → ∞, αn < α , ∀n ∈ N.
Since (αn , ∞] ∈ Ω for each n and
[−∞, α ) = ∪∞ ∞
n=1 [−∞, αn ] = ∪n=1 (αn , ∞] ,
C
(4.1)
we obtain [−∞, α ) ∈ Ω . Furthermore, we have (α , β ) = [−∞, β ) ∩ (α , ∞] ∈ Ω .

Since every open set in [−∞, ∞] may be expressed as a countable union of in-
tervals (α , β ) we have that Ω contains all the open sets. Thus, f −1 (E) ∈ M
whenever E is open, so that f is measurable.
Proposition 4.1.8. If { fn : U → [−∞, ∞]} is a sequence of measurable functions and

g = supn≥1 fn and h = lim sup fn , then g and h are measurable.
n→∞
Proof. Observe that g−1 ((α , ∞]) = ∪∞ −1

n=1 f n ((α , ∞]). From the last theorem g is
measurable. By analogy h = infk≥1 {supi≥k fi } is measurable.
4.3 Measures 101
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 {α1 , . . . , αn } = R( f )
and we set Ai = {u ∈ U | f (u) = αi }, clearly we have f = ∑ni=1 αi χAi , where

1, if u ∈ Ai ,
χAi (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 δn = 2−n . To each n ∈ N and each t ∈ R+ , there corresponds a unique

integer K = Kn (t) such that
K δn ≤ t ≤ (K + 1)δn . (4.3)
Defining

Kn (t)δn , i f 0 ≤ t < n,
ϕn (t) = (4.4)
n, i f t ≥ n,
we have that each ϕn is a Borel function on [0, ∞], such that

1. t − δn < ϕn (t) ≤ t if 0 ≤ t ≤ n,
2. 0 ≤ ϕ1 ≤ . . . ≤ t,
3. ϕn (t) → t as n → ∞, ∀t ∈ [0, ∞].
It follows that the sequence {sn = ϕn ◦ f } corresponds to the results indicated above.
4.3 Measures
Definition 4.3.1 (Measure). Let M be a σ -algebra on a topological space U. A

function μ : M → [0, ∞] is said to be a measure if μ (0) / = 0 and μ is countably
additive, that is, given {Ai } ⊂ U, a sequence of pairwise disjoint sets then
∞
i=1 Ai ) = ∑ μ (Ai ).
μ (∪∞ (4.5)
i=1
In this case (U, M , μ ) is called a measure space.

Proposition 4.3.2. Let μ : M → [0, ∞], where M is a σ -algebra of U. Then we

have the following:
1. μ (A1 ∪ . . . ∪ An ) = μ (A1 ) + . . . + μ (An ) for any given {Ai } of pairwise disjoint
measurable sets of M .
2. If A, B ∈ M and A ⊂ B, then μ (A) ≤ μ (B).
3. If {An } ⊂ M , A = ∪∞ n=1 An and
A1 ⊂ A2 ⊂ A3 ⊂ . . . (4.6)
then lim μ (An ) = μ (A).

n→∞
4. If {An } ⊂ M , A = ∩∞
n=1 An , A1 ⊃ A2 ⊃ A3 ⊃ . . .., and μ (A1 ) is finite, then
lim μ (An ) = μ (A). (4.7)

n→∞
Proof.
1. Take An+1 = An+2 = . . . = 0/ in Definition 4.1.1 item 1.
2. Observe that B = A ∪ (B − A) and A ∩ (B − A) = 0/ so that by the above, μ (A ∪
(B − A)) = μ (A) + μ (B − A) ≥ μ (A).
3. Let B1 = A1 and let Bn = An − An−1; then Bn ∈ M , Bi ∩ B j = 0/ if i = j, An =
B1 ∪ . . . ∪ Bn , and A = ∪∞
i=1 Bi . Thus
∞ n
μ (A) = μ (∪∞
i=1 Bi ) = ∑ μ (Bi ) = lim
n→∞
∑ μ (Bi ) = lim μ (An ).
n→∞
(4.8)
n=1 i=1
4. Let Cn = A1 \ An . Then C1 ⊂ C2 ⊂ . . ., μ (Cn ) = μ (A1 ) − μ (An ), A1 \ A = ∪∞

n=1Cn .
Thus by 3 we have
μ (A1 ) − μ (A) = μ (A1 \ A) = lim μ (Cn ) = μ (A1 ) − lim μ (An ). (4.9)

n→∞ 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,
n
s = ∑ αi χAi , (4.10)
i=1
where

1, if u ∈ Ai ,
χAi (u) = (4.11)
0, otherwise,

we define the integral of s over E ⊂ M , denoted by E s d μ as
4.4 Integration of Simple Functions 103
n
s d μ = ∑ αi μ (Ai ∩ E). (4.12)
E i=1
The convention 0.∞ = 0 is used here.

Definition 4.4.2 (Integral for Nonnegative Measurable Functions). If f : U →
[0, ∞] is measurable, for E ∈ M , we define the integral of f on E, denoted by
E f d μ , as

f d μ = sup sd μ , (4.13)
E s∈A E
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 , 0}, f = max{− f , 0} and the integral
of f on E, denoted by E f d μ , as

f dμ = f + dμ − f − dμ.
E E E
Theorem 4.4.4 (Lebesgue’s Monotone Convergence Theorem). Let { fn } be a se-

quence 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) f is measurable,

(b) U fn d μ → U f d μ as n → ∞.

Proof. Since U fn d μ ≤ U f n+1 d μ , ∀n ∈ N, there exists α ∈ [0, ∞] such that

fn d μ → α , as n → ∞, (4.15)
U
By Proposition 4.1.8, f is measurable, and since fn ≤ f , we have

fn d μ ≤ f dμ. (4.16)
U U
From (4.15) and (4.16), we obtain

α≤ f dμ. (4.17)
U
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 = ∪n∈N En . Observe that

fn d μ ≥ fn d μ ≥ c sd μ . (4.19)
U En En
Letting n → ∞ and applying Proposition 4.3.2, we obtain

α = lim fn d μ ≥ c sd μ , (4.20)
n→∞ U U
so that

α≥ sd μ , ∀s simple and measurable such that 0 ≤ s ≤ f . (4.21)
U
This implies

α≥ f dμ. (4.22)
U
From (4.17) and (4.22) the proof is complete.
We do not prove the next result (it is a direct consequence of the last theorem). For
a proof see [57].
Corollary 4.4.5. Let { fn } be a sequence of nonnegative measurable functions de-
fined on U ( fn : U → [0, ∞], ∀n ∈ N). Defining f (u) = ∑∞
n=1 f n (u), ∀u ∈ U, we have
∞
U
f dμ = ∑ fn d μ .
n=1 U
Theorem 4.4.6 (Fatou’s Lemma). If { fn : U → [0, ∞]} is a sequence of measurable

functions, then

lim inf fn d μ ≤ lim inf fn d μ . (4.23)
U n→∞ n→∞ U
Proof. For each k ∈ N define gk : U → [0, ∞] by
gk (u) = inf{ fi (u)}. (4.24)

i≥k
Then
gk ≤ f k (4.25)
so that

gk d μ ≤ fk d μ , ∀k ∈ N. (4.26)
U U
4.4 Integration of Simple Functions 105
Also 0 ≤ g1 ≤ g2 ≤ . . ., each gk is measurable, and
lim gk (u) = lim inf fn (u), ∀u ∈ U. (4.27)

k→∞ n→∞
From the Lebesgue monotone convergence theorem

lim inf gk d μ = lim gk d μ = lim inf fn d μ . (4.28)
k→∞ U k→∞ U U n→∞
From (4.26) we have

lim inf gk d μ ≤ lim inf fk d μ . (4.29)
k→∞ U k→∞ U
Thus, from (4.28) and (4.29), we obtain

lim inf fn d μ ≤ lim inf fn d μ . (4.30)
U n→∞ n→∞ U
Theorem 4.4.7 (Lebesgue’s Dominated Convergence Theorem). Suppose { fn } is

sequence of complex measurable functions on U such that
lim fn (u) = f (u), ∀u ∈ U. (4.31)

n→∞

If there exists a measurable function g : U → R+ such that U g d μ < ∞ and
| fn (u)| ≤ g(u), ∀u ∈ U, n ∈ N, then

1. U | f| d μ < ∞,
2. lim U | fn − f | d μ = 0.
n→∞
Proof.
1. This inequality holds since f is measurable and | f | ≤ g.
2. Since 2g − | fn − f | ≥ 0, we may apply Fatou’s lemma and obtain

2gd μ ≤ lim inf (2g − | fn − f |)d μ , (4.32)
U n→∞ U
so that

lim sup | fn − f | d μ ≤ 0. (4.33)
n→∞ U
Hence

lim | fn − f | d μ = 0. (4.34)
n→∞ U

We finish this section with an important remark:

Remark 4.4.8. In a measurable space U we say that a property holds almost every-
where (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 mea-
sure, the proprieties of items 1 and 2 of the Lebesgue monotone convergence may
be considered a.e. in U, instead of in all U. Similar remarks are valid for Fatou’s
lemma and the Lebesgue dominated convergence theorem.
4.5 Signed Measures
In this section we study signed measures. We start with the following definition.
Definition 4.5.1. Let (U, M ) be a measurable space. We say that a measure μ is
finite if μ (U) < ∞. On the other hand, we say that μ is σ -finite if there exists a
sequence {Un } ⊂ U such that U = ∪∞ n=1Un and μ (Un ) < ∞, ∀n ∈ N.
Definition 4.5.2 (Signed Measure). Let (U, M ) be a measurable space. We say that
ν : M → [−∞, +∞] is a signed measure if
• ν may assume at most one the values −∞, +∞,
• ν (0)
/ = 0,
• ν (∑∞ ∞
n=1 En ) = ∑n=1 ν (En ) for all sequence of measurable disjoint sets {En }.
We say that A ∈ M is a positive set with respect to ν if A is measurable and

ν (E) ≥ 0 for all E measurable such that E ⊂ A.
Similarly, We say that B ∈ M is a negative set with respect to ν if B is measurable
and ν (E) ≤ 0 for all E measurable such that E ⊂ B.
Finally, if A ∈ M is both positive and negative with respect to ν , it is said to be
a null set.
Lemma 4.5.3. Considering the last definitions, we have that a countable union of
positive measurable sets is positive.
Proof. Let A = ∪∞
n=1 An where An is positive, ∀n ∈ N. Choose a measurable set
E ⊂ A. Set
En = (E ∩ An ) \ (∪i=1
n−1
Ai ).
Thus, En is a measurable subset of An so that ν (En ) ≥ 0. Observe that
E = ∪∞
n=1 En ,
where {En } is a sequence of measurable disjoint sets.

Therefore ν (E) = ∑∞ n=1 ν (En ) ≥ 0.
Since E ⊂ A is arbitrary, A is positive.
4.5 Signed Measures 107
Lemma 4.5.4. Considering the last definitions, let E be a measurable set such that
0 < ν (E) < ∞.
Then there exists a positive set A ⊂ E such that ν (A) > 0.

Proof. Observe that if E is not positive then it contains a set of negative measure. In
such a case, let n1 be the smallest positive integer such that there exists a measurable
set E1 ⊂ E such that
ν (E1 ) < −1/n1.

Reasoning inductively, if E \ ∪k−1j=1 E j is not positive, let nk be the smallest positive
integer such that there exists a measurable set

Ek ⊂ E \ ∪k−1 j=1 E j
such that
ν (Ek ) < −1/nk .
Define
A = E \ (∪∞
k=1 Ek ) .
Then
E = A ∪ (∪∞
k=1 Ek ) .
Since such a union is disjoint, we have

∞
ν (E) = ν (A) + ∑ ν (Ek ),
k=1
so that since ν (E) < ∞, this last series is convergent.

Also, since
1/nk < −ν (Ek ),
we have that
∞
∑ 1/nk
k=1
is convergent so that nk → ∞ as k → ∞.
From ν (E) > 0 we must have ν (A) > 0.
Now, we will show that A is positive. Let ε > 0. Choose k sufficiently big such
that 1/(nk − 1) < ε .
Since
A ⊂ E \ ∪kj=1 E j ,
A contains no measurable set with measure less than
−1/(nk − 1) > −ε ,
that is, A contains no measurable set with measure less than −ε .

Since ε > 0 is arbitrary, A contains no measurable negative set. Thus, A is

positive.
Proposition 4.5.5 (Hahn Decomposition). Let ν be a signed measure on a mea-

surable space (U, M ). Then there exist a positive set A and a negative set B such
that U = A ∪ B and A ∩ B = 0.
/
Proof. Without losing generality, suppose ν does not assume the value +∞ (the
other case may be dealt similarly). Define
λ = sup{ν (A) | A is positive }.
Since the empty set 0/ is positive, we obtain λ ≥ 0.

Let {An } be a sequence of positive sets such that
lim ν (An ) = λ .
n→∞
Define
A = ∪∞
i=1 Ai .
From Lemma 4.5.3, A is a positive set, so that
λ ≥ ν (A).
On the other hand,

A \ An ⊂ A
so that
ν (A − An) ≥ 0, ∀n ∈ N.
Therefore
ν (A) = ν (An ) + ν (A \ An) ≥ ν (An ), ∀n ∈ N.
Hence
ν (A) ≥ λ ,
so that λ = ν (A).
Let B = U \ A. Suppose E ⊂ B, so that E is positive. Hence,
λ ≥ ν (E ∪ A)
= ν (E) + ν (A)
= ν (E) + λ , (4.35)
so that ν (E) = 0.
Thus, B contains no positive set of positive measure, so that by Lemma 4.5.4, B
contains no subsets of positive measure, that is, B is negative.
4.6 The Radon–Nikodym Theorem 109
Remark 4.5.6. Denoting the Hahn decomposition of U relating ν by {A, B}, we may
define the measures ν + and ν − by
ν + (E) = ν (E ∩ A),
and
ν − (E) = −ν (E ∩ B),
so that
ν = ν + − ν −.
We recall that two measures ν1 and ν2 are mutually singular if there are disjoint
measurable sets such that
U = A∪B
and
ν1 (A) = ν2 (B) = 0.
Observe that the measures ν+ and ν − above defined are mutually singular. The
decomposition
ν = ν+ − ν−
is called the Jordan one of ν . The measures ν + and ν − are called the positive and
negative parts of ν , respectively.
Observe that either ν + or ν − is finite since only one of the values +∞, −∞ may
be assumed by ν . We may also define
|ν |(E) = ν + (E) + ν − (E),
which is called the absolute value or total variation of ν .
4.6 The Radon–Nikodym Theorem
We start this section with the definition of absolutely continuous measures.
Definition 4.6.1 (Absolutely Continuous Measures). We say that a measure ν is

absolutely continuous with respect to a measure μ and write ν μ , if ν (A) = 0 for
all set such that μ (A) = 0. In case of a signed measure we write ν μ if |ν | |μ |.
Theorem 4.6.2 (The Radon–Nikodym Theorem). Let (U, M , μ ) be a σ -finite

measure space. Let ν be a measure defined on M which is absolutely continuous
with respect to μ , that is, ν μ .
Then there exists a nonnegative measurable function f such that

ν (E) = f d μ , ∀E ∈ M .
E
The function f is unique up to usual representatives.

Proof. First assume ν and μ are finite.

Define λ = ν + μ . Also define the functional F by

F( f ) = f dμ.
U
We recall that f ∈ L2 (μ ) if f is measurable and

| f |2 d μ < ∞.
U
The space L2 (μ ) is a Hilbert one with inner product

( f , g)L2 (μ ) = f g dμ.
U
Observe that from the Cauchy–Schwartz inequality, we may write
|F( f )| = |( f , 1)L2 (μ ) |
≤ f L2 (μ ) [μ (U)]1/2
≤ f L2 (λ ) [μ (U)]1/2 , (4.36)
since
f 2L2 (μ ) = | f |2 d μ ≤ | f |2 d λ = f 2L2 (λ ) .
U U
Thus, F is a bounded linear functional on L2 (λ ), where f ∈ L2 (λ ), if f is measurable

and
f 2 d λ < ∞.
U
Since L2 (λ ) is also a Hilbert space with the inner product

( f , g)L2 (λ ) = f g dλ ,
U
from the Riesz representation theorem, there exists g ∈ L2 (λ ), such that

F( f ) = f g dλ .
U
Thus,
f dμ = f g dλ ,
U U
and in particular,
f dμ = f g (d μ + d ν ).
U U
Hence
f (1 − g) d μ = f g dν . (4.37)
U U
Assume, to obtain contradiction, that g < 0 in a set A such that μ (A) > 0.
Thus
(1 − g) d μ > 0,
A
so that from this and (4.37) with f = χA we get

g d ν > 0.
A
Since g < 0 on A we have a contradiction. Thus g ≥ 0, a.e. [μ ] on U.

Now, assume, also to obtain contradiction, that g > 1 on set B such that μ (B) > 0.
Thus
(1 − g) d μ ≤ 0,
B
so that from this and (4.37) with f = χB we obtain

ν (B) ≤ g dν ≤ 0
B
and hence
ν (B) = 0.

Thus, Bg d ν = 0 so that
(1 − g) d μ = 0,
B
which implies that μ (B) = 0, a contradiction.
From above we conclude that
0 ≤ g ≤ 1, a.e. [μ ] in U.
On the other hand, for a fixed E μ -measurable again from (4.37) with f = χE , we
get
(1 − g) d μ = g dν ,
E E
so that
(1 − g) d μ = g dν − d ν + ν (E),
E E E
and therefore
ν (E) = (1 − g) (d μ + d ν ),
E
that is,
ν (E) = (1 − g) d λ , ∀E ∈ M .
E
Define
B = {u ∈ U : g(u) = 0}.

Hence, μ (B) = B g d λ = 0.
From this, since λ μ , we obtain
g−1 g = 1, a.e. [λ ].
Therefore, for a not relabeled E ∈ M , we have

λ (E) = g−1 g d λ = g−1 d μ .
E E
Finally, observe that
μ (E) + ν (E) = λ (E)

= dλ
E
= g−1 d μ . (4.38)
E
Thus,

ν (E) = g−1 d μ − μ (E)
E

= (g−1 − 1) d μ
E

= (1 − g)g−1 d μ , ∀E ∈ M . (4.39)
E
The proof for the finite case is complete. The proof for σ -finite is developed in the
next lines.
Since U is σ -finite, there exists a sequence {Un } such that U = ∪∞ n=1Un , and
μ (Un ) < ∞ and ν (Un ) < ∞, ∀n ∈ N.
Define
Fn = Un \ ∪n−1
j=1 U j ,
thus U = ∪∞ n=1 Fn and {Fn } is a sequence of disjoint sets, such that μ (Fn ) < ∞ and
ν (Fn ) < ∞, ∀n ∈ N.
Let E ∈ M . For each n ∈ N from above we may obtain fn such that

ν (E ∩ Fn ) = fn d μ , ∀E ∈ M .
E∩Fn
From this and the monotone convergence theorem corollary we may write
∞
ν (E) = ∑ ν (E ∩ Fn)
n=1
∞
= ∑ fn d μ
n=1 E∩Fn
∞
= ∑ fn χFn d μ
n=1 E
∞
= ∑ fn χFn d μ
E n=1

= f dμ, (4.40)
E
where
∞
f= ∑ fn χFn .
n=1
Theorem 4.6.3 (The Lebesgue Decomposition). Let (U, M , μ ) be a σ -finite mea-

sure space and let ν be a σ -finite measure defined on M .
Then we may find a measure ν0 , singular with respect to μ , and a measure ν1 ,
absolutely continuous with respect to μ , such that
ν = ν0 + ν1 .
Furthermore, the measures ν0 and ν1 are unique.
Proof. Since μ and ν are σ -finite measures, so is
λ = ν + μ.
Observe that ν and μ are absolutely continuous with respect to λ . Hence, by the
Radon–Nikodym theorem, there exist nonnegative measurable functions f and g
such that
μ (E) = f d λ , ∀E ∈ M
E
and
ν (E) = g d λ ∀E ∈ M .
E
Define
A = {u ∈ U | f (u) > 0},
and
B = {u ∈ U | f (u) = 0}.
Thus,
U = A ∪ B,
and
A ∩ B = 0.
/
Also define
ν0 (E) = ν (E ∩ B), ∀E ∈ M .
We have that ν0 (A) = 0 so that
ν0 ⊥ μ .
Define
ν1 (E) = ν (E ∩ A)

= g dλ . (4.41)
E∩A
Therefore,
ν = ν0 + ν1 .
To finish the proof, we have only to show that
ν1 μ .
Let E ∈ M such that μ (E) = 0. Thus

0 = μ (E) = f dλ ,
E
and in particular
f d λ = 0.
(E∩A)
Since f > 0 on A ∩ E we conclude that
λ (A ∩ E) = 0.
Therefore, since ν λ , we obtain
ν (E ∩ A) = 0,
so that
ν1 (E) = ν (E ∩ A) = 0.
From this we may infer that
ν1 μ .
The proof of uniqueness is left to the reader.
4.7 Outer Measure and Measurability
Let U be a set. Denote by P the set of all subsets of U. An outer measure

μ ∗ : P → [0, +∞] is a set function such that
4.7 Outer Measure and Measurability 115
1. μ ∗ (0)
/ = 0,
2. if A ⊂ B, then μ ∗ (A) ≤ μ ∗ (B), ∀A, B ⊂ U,
3. if E ⊂ ∪∞
n=1 En , then
∞
μ ∗ (E) ≤ ∑ μ ∗(En ).
n=1
The outer measure is called finite if μ ∗ (U) < ∞.

Definition 4.7.1 (Measurable Set). A set E ⊂ U is said to be measurable with re-
spect to μ ∗ if
μ ∗ (A) = μ ∗ (A ∩ E) + μ ∗(A ∩ E c ), ∀A ⊂ U.
Theorem 4.7.2. The set B of μ ∗ -measurable sets is a σ -algebra. If μ is defined to
be μ ∗ restricted to B, then μ is a complete measure on B.
Proof. Let E = 0/ and let A ⊂ U.
Thus,
μ ∗ (A) = μ ∗ (A ∩ 0)
/ + μ ∗(A ∩ 0/ c ) = μ ∗ (A ∩U) = μ ∗ (A).
Therefore 0/ is μ ∗ -measurable.
Let E1 , E2 ∈ U be μ ∗ -measurable sets. Let A ⊂ U. Thus,
μ ∗ (A) = μ ∗ (A ∩ E2 ) + μ ∗ (A ∩ E2c ),
so that from the measurability of E1 we get
μ ∗ (A) = μ ∗ (A ∩ E2 ) + μ ∗ (A ∩ E2c ∩ E1 ) + μ ∗ (A ∩ E2c ∩ E1c ). (4.42)
Since
A ∩ (E1 ∪ E2 ) = (A ∩ E2 ) ∪ (A ∩ E1 ∩ E2c ),
we obtain
μ ∗ (A ∩ (E1 ∪ E2 )) ≤ μ ∗ (A ∩ E2 ) + μ ∗ (A ∩ E2c ∩ E1 ). (4.43)
μ ∗ (A) ≥ μ ∗ (A ∩ (E1 ∪ E2 )) + μ ∗ (A ∩ E1c ∩ E2c )

= μ ∗ (A ∩ (E1 ∪ E2 )) + μ ∗ (A ∩ (E1 ∪ E2 )c ). (4.44)
Hence E1 ∪ E2 is μ ∗ -measurable.
By induction, the union of a finite number of μ ∗ -measurable sets is μ ∗ -
measurable.
Assume E = ∪∞ ∗
i=1 Ei where {Ei } is a sequence of disjoint μ -measurable sets.
Define Gn = ∪ni=1 Ei . Then Gn is μ ∗ -measurable and for a given A ⊂ U we have
μ ∗ (A) = μ ∗ (A ∩ Gn ) + μ ∗ (A ∩ Gcn )
≥ μ ∗ (A ∩ Gn ) + μ ∗ (A ∩ E c ), (4.45)
since E c ⊂ Gcn , ∀n ∈ N.
Observe that
Gn ∩ E n = E n
and
Gn ∩ Enc = Gn−1 .
Thus, from the measurability of En , we may get
μ ∗ (A ∩ Gn ) = μ ∗ (A ∩ Gn ∩ En ) + μ ∗(A ∩ Gn ∩ Enc )
= μ ∗ (A ∩ En) + μ ∗ (A ∩ Gn−1). (4.46)
By induction we obtain
n
μ (A ∩ Gn ) = ∑ μ ∗ (A ∩ Ei ),
i=1
so that
n
μ ∗ (A) ≥ μ ∗ (A ∩ E c ) + ∑ μ ∗ (A ∩ Ei ), ∀n ∈ N,
i=1
that is, considering that

A ∩ E ⊂ ∪∞
i=1 (A ∩ Ei ),
we get
∞
μ ∗ (A) ≥ μ ∗ (A ∩ E c ) + ∑ μ ∗ (A ∩ Ei )
i=1
≥ μ ∗ (A ∩ E c ) + μ ∗ (A ∩ E). (4.47)
Since A ⊂ U is arbitrary we may conclude that E = ∪∞ ∗

i=1 Ei is μ -measurable. There-
fore B is a σ -algebra.
Finally, we prove that μ is a measure.
Let E1 , E2 ⊂ U be two disjoint μ ∗ -measurable sets.
Thus
μ (E1 ∪ E2 ) = μ ∗ (E1 ∪ E2 )
= μ ∗ ((E1 ∪ E2 ) ∩ E2 ) + μ ∗((E1 ∪ E2 ) ∩ E2c )
= μ ∗ (E2 ) + μ ∗(E1 ). (4.48)
By induction we obtain the finite additivity.

Also, if
E = ∪∞ i=1 Ei ,
where {Ei } is a sequence of disjoint measurable sets.

Thus,
n
μ (E) ≥ μ (∪ni=1 Ei ) = ∑ μ (Ei ), ∀n ∈ N.
i=1
Therefore,
∞
μ (E) ≥ ∑ μ (Ei ).
i=1
Now observe that

∞ ∞
i=1 Ei ) ≤ ∑ μ (Ei ) = ∑ μ (Ei ),
μ (E) = μ ∗ (∪∞ ∗
i=1 i=1
and thus
∞
μ (E) = ∑ μ (Ei ).
i=1
Definition 4.7.3. A measure on an algebra A ⊂ U is a set function μ : A → [0, +∞)

such that
1. μ (0)
/ = 0,
2. if {Ei } is a sequence of disjoint sets in A so that E = ∪∞
i=1 Ei ∈ A , then
∞
μ (E) = ∑ μ (Ei ).
i=1
We may define an outer measure in U by

∞
μ ∗ (E) = inf ∑ μ (Ai ) | E ⊂ ∪∞i=1 Ai ,
i=1
where Ai ∈ A , ∀i ∈ N.
Proposition 4.7.4. Suppose A ∈ A and {Ai } ⊂ A is such that
A ⊂ ∪∞
i=1 Ai .
Under such hypotheses,

∞
μ (A) ≤ ∑ μ (Ai ).
i=1
Proof. Define
Bn = (A ∩ An ) \ (∪i=1
n−1
Ai ).
Thus
Bn ⊂ An , ∀n ∈ N,
Bn ∈ A , ∀n ∈ N, and
A = ∪∞
i=1 Bi .
Moreover, {Bn } is a sequence of disjoint sets, so that

∞ ∞
μ (A) = ∑ μ (Bi ) ≤ ∑ μ (Ai ).
i=1 i=1
Corollary 4.7.5. If A ∈ A , then μ ∗ (A) = μ (A).
Theorem 4.7.6. The set function μ ∗ is an outer measure.
Proof. The only not immediate property to be proven is the countably sub-
additivity.
Suppose E ⊂ ∪∞ ∗
i=1 Ei . If μ (Ei ) = +∞ for some i ∈ N, the result holds.
∗
Thus, assume μ (Ei ) < +∞, ∀i ∈ N.
Let ε > 0. Thus for each i ∈ N there exists {Ai j } ⊂ A such that Ei ⊂ ∪∞j=1 Ai j ,
and
∞
ε
∑ μ (Ai j ) ≤ μ ∗ (Ei ) + 2i .
j=1
Therefore,
∞ ∞ ∞
μ (E) ≤ ∑ ∑ μ (Ai j ) ≤ ∑ μ ∗ (Ei ) + ε .
i=1 j=1 i=1
Since ε > 0 is arbitrary, we get

∞
μ ∗ (E) ≤ ∑ μ ∗ (Ei ).
i=1
Proposition 4.7.7. Suppose A ∈ A . Then A is μ ∗ -measurable.
Proof. Let E ∈ U such that μ ∗ (E) < +∞. Let ε > 0.

Thus, there exists {Ai } ⊂ A such that E ⊂ ∪∞i=1 Ai and
∞
∑ μ (Ai) < μ ∗ (E) + ε .
i=1
Observe that
μ (Ai ) = μ (Ai ∩ A) + μ (Ai ∩ Ac ),
so that from the fact that
E ∩ A ⊂ ∪∞
i=1 (Ai ∩ A),
and
(E ∩ Ac ) ⊂ ∪∞
i=1 (Ai ∩ A ),
c
we obtain
∞ ∞
μ ∗ (E) + ε > ∑ (Ai ∩ A) + ∑ μ (Ai ∩ Ac )
i=1 i=1
≥ μ ∗ (E ∩ A) + μ ∗(E ∩ Ac ). (4.49)
μ ∗ (E) ≥ μ ∗ (E ∩ A) + μ ∗(E ∩ Ac ).
Proposition 4.7.8. Suppose μ is a measure on an algebra A ⊂ U, μ ∗ is the outer

measure induced by μ , and E ⊂ U is a set. Then, for each ε > 0, there is a set
A ∈ Aσ with E ⊂ A and
μ ∗ (A) ≤ μ ∗ (E) + ε .
Also, there is a set B ∈ Aσ δ such that E ⊂ B and
μ ∗ (E) = μ ∗ (B).
Proof. Let ε > 0. Thus, there is a sequence {Ai } ⊂ A such that
E ⊂ ∪∞
i=1 Ai
and
∞
∑ μ (Ai) ≤ μ ∗ (E) + ε .
i=1
Define A = ∪∞
i=1 Ai , then
∞
μ ∗ (A) ≤ ∑ μ ∗(Ai )
i=1
∞
= ∑ μ (Ai )
i=1
∗
≤ μ (E) + ε . (4.50)
Now, observe that we write A ∈ Aσ if A = ∪∞ i=1 Ai where Ai ∈ A , ∀i ∈ N.

Also, we write B ∈ Aσ δ if B = ∩∞
n=1 A n , where An ∈ Aσ , ∀n ∈ N.
From above, for each n ∈ N, there exists An ∈ Aσ such that
E ⊂ An
and
μ ∗ (An ) ≤ μ (E) + 1/n.
Define B = ∩∞
n=1 An . Thus, B ∈ Aσ δ , E ⊂ B and
μ ∗ (B) ≤ μ ∗ (An ) ≤ μ ∗ (E) + 1/n, ∀n ∈ N.
Hence
μ ∗ (B) = μ ∗ (E).
Proposition 4.7.9. Suppose μ is a σ -finite measure on a σ -algebra A , and let μ ∗

be the outer measure induced by μ .
Under such hypotheses, a set E is μ ∗ measurable if and only if E = A \ B where
A ∈ Aσ δ , B ⊂ A, μ ∗ (B) = 0.
Finally, for each set B such that μ ∗ (B) = 0, there exists C ∈ Aσ δ such that B ⊂ C
and μ ∗ (C) = 0.
Proof. The if part is obvious.
Now suppose E is μ ∗ -measurable. Let {Ui } be a countable collection of disjoint
sets of finite measure such that
U = ∪∞
i=1Ui .
Observe that
E = ∪∞
i=1 Ei ,
where
Ei = E ∩Ui ,
is μ ∗ -measurable for each i ∈ N.
Let ε > 0. From the last proposition for each i, n ∈ N there exists Ani ∈ Aσ such
that
1
μ (Ani ) < μ ∗ (Ei ) + i .
n2
Define
An = ∪∞ i=1 Ani .
Thus,
E ⊂ An
and
An \ E ⊂ ∪∞
i=1 (Ani \ Ei ),
and therefore,
∞ ∞
1 1
μ (An \ E) ≤ ∑ μ (Ani \ Ei ) ≤ ∑ i
= .
i=1 i=1 n2 n
Since An ∈ Aσ , defining
A = ∩∞
n=1 An ,
we have that A ∈ Aσ δ and

A \ E ⊂ An \ E
so that
1
μ ∗ (A \ E) ≤ μ ∗ (An \ E) ≤ , ∀n ∈ N.
n
Hence μ ∗ (A \ E) = 0.
Theorem 4.7.10 (Carathéodory). Let μ be a measure on analgebra A and μ ∗ the
respective induced outer measure.
Then the restriction μ of μ ∗ to the μ ∗ -measurable sets is an extension of μ to

a σ -algebra containing A . If μ is finite or σ -finite, so is μ . In particular, if μ is
σ -finite, then μ is the only measure on the smallest σ -algebra containing A which
is an extension of μ .
Proof. From the Theorem 4.7.2, μ is an extension of μ to a σ -algebra containing

A , that is, μ is a measure on such a set.
Observe that from the last results, if μ is σ -finite, so is μ .
Now assume μ is σ -finite. We will prove the uniqueness of μ .
Let B be the smallest σ -algebra containing A and let μ̃ be another measure on
B which extends μ on A .
Since each set Aσ may be expressed as a disjoint countable union of sets in A ,
the measure μ̃ equals μ on Aσ . Let B be a μ ∗ -measurable set such that μ ∗ (B) < ∞.
Let ε > 0. By Proposition 4.7.9 there exists an A ∈ Aσ such that B ⊂ A and
μ ∗ (A) < μ ∗ (B) + ε .
Since B ⊂ A, we obtain
μ̃ (B) ≤ μ̃ (A) = μ ∗ (A) ≤ μ ∗ (B) + ε .
Considering that ε > 0 is arbitrary, we get
μ̃ (B) ≤ μ ∗ (B).
Observe that the class of μ ∗ -measurable sets is a σ -algebra containing A .

Therefore, as above indicated, we have obtained A ∈ Aσ such that B ⊂ A and
μ ∗ (A) ≤ μ ∗ (B) + ε
so that
μ ∗ (A) = μ ∗ (B) + μ ∗(A \ B),
from this and above
μ̃ (A \ B) ≤ μ ∗ (A \ B) ≤ ε ,
if μ ∗ (B) < ∞.
Therefore,
μ ∗ (B) ≤ μ ∗ (A) = μ̃ (A)

= μ̃ (B) + μ̃ (A \ B) ≤ μ̃ (B) + ε . (4.51)
Since ε > 0 is arbitrary we have
μ ∗ (B) ≤ μ̃ (B),
so that μ ∗ (B) = μ̃ (B). Finally, since μ is σ -finite, there exists a sequence of count-
able disjoint sets {Ui } such that μ (Ui ) < ∞, ∀i ∈ N, and U = ∪∞ i=1Ui .
If B ∈ B, then
B = ∪∞
i=1 (Ui ∩ B).
Thus, from above,

∞
μ̃ (B) = ∑ μ̃ (Ui ∩ B)
i=1
∞
= ∑ μ (Ui ∩ B)
i=1
= μ (B). (4.52)
Remark 4.7.11. We may start the process of construction of a measure by the action
of a set function on a semi-algebra. Here, a semi-algebra C is a collection of subsets
of U such that the intersection of any two sets in C is in C and the complement of
any set in C is a finite disjoint union of sets in C .
If C is any semi-algebra of sets, then the collection consisting of the empty set
and all finite disjoint unions of sets in C is an algebra, which is said to be generated
by C . We denote such algebra by A .
If we have a set function acting on C , we may extend it to A by defining
n
μ (A) = ∑ μ (Ei ),
i=1
where A = ∪ni=1 Ei and Ei ∈ C , ∀i ∈ {1, . . ., n}, so that this last union is disjoint. We
recall that any A ∈ A admits such a representation.
4.8 The Fubini Theorem
We start this section with the definition of complete measure space.

Definition 4.8.1. We say that a measure space (U, M , μ ) is complete if M contains
all subsets of sets of zero measure. That is, if A ∈ M , μ (A) = 0 and B ⊂ A, then
B ∈ M.
In the next lines we recall the formal definition of semi-algebra.
Definition 4.8.2. We say (in fact recall) 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 .
4.8 The Fubini Theorem 123
4.8.1 Product Measures
Let (U, M1 , μ1 ) and (V, M2 , μ2 ) 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 × Bc) ∪ (Ac × Bc ).
We define λ : M1 × M2 → R+ by
λ (A × B) = μ1 (A)μ2 (B).
Lemma 4.8.3. Let {Ai × Bi }i∈N be a countable disjoint collection of measurable

rectangles whose union is the rectangle A × B. Then
∞
λ (A × B) = ∑ μ1 (Ai )μ2 (Bi ).
i=1
Proof. Let u ∈ A. Thus each v ∈ B is such that (u, v) is exactly in one Ai × Bi .

Therefore
∞
χA×B (u, v) = ∑ χAi (u)χBi (v).
i=1
Hence, for the fixed u in question, from the corollary of Lebesgue monotone con-
vergence theorem, we may write
∞
V
χA×B(u, v)d μ2 (v) = ∑ χAi (u)χBi (v)d μ2 (v)
i=1
∞
= ∑ χAi (u)μ2 (Bi ) (4.53)
i=1
so that also from the mentioned corollary

∞
d μ1 (u) χA×B(u, v)d μ2 (v) = ∑ μ1 (Ai )μ2 (Bi ).
U V i=1
Observe that

d μ1 (u) χA×B (u, v)d μ2 (v) = d μ1 (u) χA (u)χB (v)d μ2 (v)
U V U V
= μ1 (A)μ2 (B).
From the last two equations we may write

∞
λ (A × B) = μ1 (A)μ2 (B) = ∑ μ1 (Ai )μ2 (Bi ).
i=1
Definition 4.8.4. Let E ⊂ U × V . We define Eu and Ev by
Eu = {v | (u, v) ∈ E},
and
Ev = {u | (u, v) ∈ E}.
Observe that
χEu (v) = χE (u, v),
(E c )u = (Eu )c ,
and
(∪Eα )u = ∪(Eα )u ,
for any collection {Eα }.
We denote by Rσ as the collection of sets which are countable unions of mea-
surable rectangles. Also, Rσ δ will denote the collection of sets which are countable
intersections of elements of Rσ .
Lemma 4.8.5. Let u ∈ U and E ∈ Rσ δ . Then Eu is a measurable subset of V .
Proof. If E ∈ R, the result is trivial. Let E ∈ Rσ . Then E may be expressed as a

disjoint union
E = ∪∞ i=1 Ei ,
where Ei ∈ R, ∀i ∈ N. Thus,
χEu (v) = χE (u, v)

= sup χEi (u, v)
i∈N
= sup χ(Ei )u (v). (4.54)
i∈N
Since each (Ei )u is measurable we have that
χ(Ei )u (v)
is a measurable function of v, so that
χEu (v)
is measurable, which implies that Eu is measurable. Suppose now
E = ∩∞
i=1 Ei ,
where Ei+1 ⊂ Ei , ∀i ∈ N. Then
χEu (v) = χE (u, v)

= inf χEi (u, v)
i∈N
= inf χ(Ei )u (v). (4.55)
i∈N
Thus as from above χ(Ei )u (v) is measurable for each i ∈ N, we have that χEu is also
measurable so that Eu is measurable.
Lemma 4.8.6. Let E be a set in Rσ δ with (μ1 × μ2 )(E) < ∞. Then the function g
defined by
g(u) = μ2 (Eu )
is a measurable function and

g d μ1 (u) = (μ1 × μ2 )(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) = μ2 ((Ei )u ).
Then each gi is a nonnegative measurable function and

∞
g = ∑ gi .
i=1
Thus, g is measurable, and by the corollary of the Lebesgue monotone convergence

theorem, we have
∞
U
g(u)d μ1 (u) = ∑ gi (u)d μ1 (u)
i=1 U
∞
= ∑ (μ1 × μ2)(Ei )
i=1
= (μ1 × μ2 )(E). (4.56)
Let E be a set of finite measure in Rσ δ . Then there is a sequence in Rσ such that
Ei+1 ⊂ Ei
and
E = ∩∞
i=1 Ei .
Let gi (u) = μ2 ((Ei )u ), since

g1 (u) = (μ1 × μ2 )(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) = μ2 (Eu ) = lim μ2 ((Ei )u ) = lim gi (u), (4.57)

i→∞ i→∞
that is,
gi → g, a.e. in E.
We may conclude that g is also measurable. Since
0 ≤ gi ≤ g, ∀i ∈ N
the Lebesgue dominated convergence theorem implies that

g(u) d μ1 (u) = lim gi d μ1 (u) = lim (μ1 × μ2 )(Ei ) = (μ1 × μ2 )(E).
E i→∞ i→∞
Lemma 4.8.7. Let E be a set such that (μ1 × μ2 )(E) = 0. Then for almost all u ∈ U
we have
μ2 (Eu ) = 0.
Proof. Observe that there is a set in Rσ δ such that E ⊂ F and
(μ1 × μ2)(F) = 0.
From the last lemma

μ2 (Fu ) = 0
for almost all u. From Eu ⊂ Fu we obtain
μ2 (Eu ) = 0
for almost all u, since μ2 is complete.

Proposition 4.8.8. Let E be a measurable subset of U ×V such that (μ1 × μ2 )(E) is

finite. For almost all u the set Eu is a measurable subset of V . The function g defined
by
g(u) = μ2 (Eu )
is measurable and
g d μ1 (u) = (μ1 × μ2 )(E).
Proof. First observe that there is a set F ∈ Rσ δ such that E ⊂ F and
(μ1 × μ2 )(F) = (μ1 × μ2 )(E).
Let G = F \ E. Since F and E are measurable, G is measurable, and
(μ1 × μ2 )(G) = 0.
By the last lemma we obtain

μ2 (Gu ) = 0,
for almost all u so that
g(u) = μ2 (Eu ) = μ2 (Fu ) a.e. in U.
By Lemma 4.8.6 we may conclude that g is measurable and

g d μ1 (u) = (μ1 × μ2 )(F) = (μ1 × μ2 )(E).
Theorem 4.8.9 (Fubini). Let (U, M1 , μ1 ) and (V, M2 , μ2 ) 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,
3. h1 (u) = V f (u, v) d μ2 (v) is integrable on U,
4. h2 (v) = U f (u, v) d μ1 (u) is integrable on V ,
5.

f d μ2 (v) d μ1 (u) = f d μ1 (u) d μ2 (v)
U V V U

= f d(μ1 × μ2 ). (4.58)
U×V
Proof. It suffices to consider the case where f is nonnegative (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 Theorem 4.2.2, we may obtain a sequence of nonnegative
simple functions {φn } such that
φn ↑ f .
Observe that given u ∈ U, fu is such that
(φn )u ↑ fu , a.e. .
By the Lebesgue monotone convergence theorem we get

f (u, v) d μ2 (v) = lim φn (u, v) d μ2 (v),
V n→∞ V
so that this last resulting function is integrable in U. Again by the Lebesgue mono-
tone convergence theorem, we obtain

f d μ2 (v) d μ1 (u) = lim φn d μ2 (v) d μ1 (u)
U V n→∞ U V

= lim φn d(μ1 × μ2 )
n→∞ U×V

= f d(μ1 × μ2 ). (4.59)
U×V
Chapter 5
The Lebesgue Measure in Rn
5.1 Introduction
In this chapter we will define the Lebesgue measure and the concept of Lebesgue
measurable set. We show that the set of Lebesgue measurable sets is a σ -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). The main reference for this chapter is [62].
We start with the following theorems without proofs.
Theorem 5.1.1. Every open set A ⊂ R may be expressed as a countable union of
disjoint open intervals.
Remark 5.1.2. In this text Q j denotes a closed cube in Rn and |Q j | its volume, that
is, |Q j | = ∏ni=1 (bi − ai ), where Q j = ∏ni=1 [ai , bi ]. Also we assume that if two Q1 and
Q2 , closed or not, have the same interior, then |Q1 | = |Q2 | = |Q̄1 |. We recall that two
cubes Q1 , Q2 ⊂ Rn are said to be quasi-disjoint if their interiors are disjoint.
Theorem 5.1.3. Every open set A ⊂ Rn , where n ≥ 1 may be expressed as a count-

able union of quasi-disjoint closed cubes.
Definition 5.1.4 (Outer Measure). Let E ⊂ Rn . The outer measure of E, denoted

by m∗ (E), is defined by

∞
m∗ (E) = inf ∑ |Q j | : E ⊂ ∪∞j=1 Q j ,
j=1
where Q j is a closed cube, ∀ j ∈ N.
130 5 The Lebesgue Measure in Rn
5.2 Properties of the Outer Measure
First observe that given ε > 0, there exists a sequence {Q j } such that
E ⊂ ∪∞j=1 Q j
and
∞
∑ |Q j | ≤ m∗ (E) + ε .
j=1
1. Monotonicity: If E1 ⊂ E2 then m∗ (E1 ) ≤ m∗ (E2 ). This follows from the fact that
if E2 ⊂ ∪∞j=1 Q j then E1 ⊂ ∪∞j=1 Q j .
2. Countable sub-additivity : If E ⊂ ∪∞j=1 E j , then m∗ (E) ≤ ∑∞j=1 m∗ (E j ).
Proof. First assume that m∗ (E j ) < ∞, ∀ j ∈ N; otherwise, the result is obvious.

Thus, given ε > 0 for each j ∈ N, there exists a sequence {Qk, j }k∈N such that
E j ⊂ ∪∞
k=1 Qk, j
and
∞
ε
∑ |Qk, j | < m∗ (E j ) + 2 j .
k=1
Hence
E ⊂ ∪∞j,k=1 Qk, j
and therefore

∞ ∞ ∞
∗
m (E) ≤ ∑ |Qk, j | = ∑ ∑ |Qk, j |
j,k=1 j=1 k=1
∞
ε
≤ ∑ m∗ (E j ) + j
2
j=1
∞
= ∑ m∗ (E j ) + ε . (5.1)
j=1
Being ε > 0 arbitrary, we obtain

∞
m∗ (E) ≤ ∑ m∗(E j ).
j=1
3. If
E ⊂ Rn ,
and
α = inf{m∗ (A) | A is open and E ⊂ A},
5.2 Properties of the Outer Measure 131
then
m∗ (E) = α .
Proof. From the monotonicity, we have
m∗ (E) ≤ m∗ (A), ∀A ⊃ E, A open.
Thus
m∗ (E) ≤ α .
Suppose given ε > 0. Choose a sequence {Q j } of closed cubes such that
E ⊂ ∪∞j=1 Q j
and
∞
∑ |Q j | ≤ m∗ (E) + ε .
j=1
Let {Q̃ j } be a sequence of open cubes such that Q̃ j ⊃ Q j

ε
|Q̃ j | ≤ |Q j | + , ∀ j ∈ N.
2j
Define
A = ∪∞j=1 Q̃ j ;
hence A is open, A ⊃ E, and
∞
m∗ (A) ≤ ∑ |Q̃ j |
j=1
∞
ε
≤ ∑ |Q j | +
2j
j=1
∞
= ∑ |Q j | + ε
j=1
∗
≤ m (E) + 2ε . (5.2)
Therefore
α ≤ m∗ (E) + 2ε .
Being ε > 0 arbitrary, we have
α ≤ m∗ (E).

4. If E = E1 ∪ E2 and 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 ε > 0. Choose {Q j } a sequence of closed cubes such that
E ⊂ ∪∞j=1 Q j ,
and
∞
∑ |Q j | ≤ m∗ (E) + ε .
j=1
Let δ > 0 such that

d(E1 , E2 ) > δ > 0.
Dividing the cubes Q j if necessary, we may assume that the diameter of each
cube Q j is smaller than δ . Thus each Q j intersects just one of the sets E1 and E2 .
Denote by J1 and J2 the sets of indices j such that Q j intersects E1 and E2 ,
respectively. Thus,
E1 ⊂ ∪ j∈J1 Q j and E2 ⊂ ∪ j∈J2 Q j .
Hence,
m∗ (E1 ) + m∗ (E2 ) ≤ ∑ |Q j | + ∑ |Q j |
j∈J1 j∈J2
∞
≤ ∑ |Q j | ≤ m∗ (E) + ε . (5.3)
j=1
Being ε > 0 arbitrary,

m∗ (E1 ) + m∗ (E2 ) ≤ m∗ (E).
5. If a set E is a countable union of cubes quasi disjoints, that is,
E = ∪∞j=1 Q j ,
then
∞
m∗ (E) = ∑ |Q j |.
j=1
Proof. Let ε > 0.

Let {Q̃ j } be open cubes such that Q̃ j ⊂⊂ Q◦j (i.e., the closure of Q̃ j is contained
in the interior of Q j ) and
ε
|Q j | ≤ |Q̃ j | + j .
2
5.4 Properties of Measurable Sets 133
Thus, for each N ∈ N the cubes Q̃1 , . . . , Q̃N are disjoint and each pair have a finite
distance. Hence,
N N ε
m∗ (∪Nj=1 Q̃ j ) = ∑ |Q̃ j | ≥ ∑ |Q j | − j .
2
j=1 j=1
Being
∪Nj=1 Q̃ j ⊂ E,
we obtain
N N
m∗ (E) ≥ ∑ |Q̃ j | ≥ ∑ |Q j | − ε .
j=1 j=1
Therefore
∞
∑ |Q j | ≤ m∗ (E) + ε .
j=1
Being ε > 0 arbitrary, we may conclude that

∞
∑ |Q j | ≤ m∗(E).
j=1
5.3 The Lebesgue Measure
Definition 5.3.1. A set E ⊂ Rn is said to be Lebesgue measurable if for each ε > 0

there exists A ⊂ Rn open such that
E ⊂A
and
m∗ (A − E) ≤ ε .
If E is measurable, we define its Lebesgue measure, denoted by m(E), as
m(E) = m∗ (E).
5.4 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 ε > 0, thus there
exists A ⊂ Rn open such that E ⊂ A and m∗ (A) < ε . Therefore
m∗ (A − E) < ε .
3. A countable union of measurable sets is measurable.

Proof. Suppose
E = ∪∞j=1 E j
where each E j is measurable. Suppose given ε > 0. For each j ∈ N, there exists
A j ⊂ Rn open such that
Ej ⊂ Aj
and
ε
m∗ (A j − E j ) ≤ .
2j
Define A = ∪∞j=1 A j . Thus E ⊂ A and
(A − E) ⊂ ∪∞j=1 (A j − E j ).
From the monotonicity and countable sub-additivity of the outer measure we

have
∞
m∗ (A − E) ≤ ∑ m∗ (A j − E j ) < ε .
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) < ∞.
Let ε > 0; thus, there exists an open A ⊂ Rn such that F ⊂ A and
m∗ (A) ≤ m∗ (F) + ε .
Being F closed, A − F is open, and therefore, A − F may be expressed as a count-

able union of quasi disjoint closed cubes. Hence
A − F = ∪∞j=1 Q j .
For each N ∈ N
K = ∪Nj=1 Q j
is compact; therefore
d(K, F) > 0.
Being K ∪ F ⊂ A, we have
N
m∗ (A) ≥ m∗ (F ∪ K) = m∗ (F) + m∗ (K) = m∗ (F) + ∑ |Q j |.
j=1
Therefore
N
∑ |Q j | ≤ m∗ (A) − m∗(F) ≤ ε .
j=1
Finally,
∞
m∗ (A − F) ≤ ∑ |Q j | < ε .
j=1

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
Observe that Ack is closed and therefore measurable, ∀k ∈ N. Thus
S = ∪∞ c
k=1 Ak
is also measurable. On the other hand

S ⊂ Ec
and if x ∈ (E c − S), then x ∈ E c and x ∈ S, so that x ∈ E and x ∈ Ack , ∀k ∈ N.
Hence x ∈ E and x ∈ Ak , ∀k ∈ N and finally x ∈ (Ak − E), ∀k ∈ N, that is,
E c − S ⊂ Ak − E, ∀k ∈ N.
Therefore
1
m∗ (E c − S) ≤ , ∀k ∈ N.
k
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
∩∞j=1 E j = (∪∞j=1 E cj )c .
Theorem 5.4.1. If {Ei } is sequence of measurable pairwise disjoint sets and

E = ∪∞j=1 Ei , then
∞
m(E) = ∑ m(E j ).
j=1
Proof. First assume that E j is bounded. Being E cj measurable, given ε > 0, there
exists an open H j ⊃ E cj such that
ε
m∗ (H j − E cj ) < , ∀ j ∈ N.
2j
Denoting Fj = H cj we have that Fj ⊂ E j is closed and
ε
m∗ (E j − Fj ) < , ∀ j ∈ N.
2j
For each N ∈ N the sets F1 , . . . , FN are compact and disjoint, so that
N
m(∪Nj=1 Fj ) = ∑ m(Fj ).
j=1
As
∪Nj=1 Fj ⊂ E
we have
N N
m(E) ≥ ∑ m(Fj ) ≥ ∑ m(E j ) − ε .
j=1 j=1
Hence
∞
m(E) ≥ ∑ m(E j ) − ε .
j=1
Being ε > 0 arbitrary, we obtain

∞
m(E) ≥ ∑ m(E j ).
j=1
As the reverse inequality is always valid, we have

∞
m(E) = ∑ m(E j ).
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 and Sk = Qk − Qk−1 , ∀k ≥ 2. Also define
E j,k = E j ∩ Sk , ∀ j, k ∈ N.
Thus

E = ∪∞j=1 ∪∞ ∞
k=1 E j,k = ∪ j,k=1 E j,k ,
where such a union is disjoint and each E j,k is bounded. Through the last result,
we get
∞
m(E) = ∑ m(E j,k )
j,k=1
∞ ∞
= ∑ ∑ m(E j,k )
j=1 k=1
∞
= ∑ m(E j ). (5.4)
j=1

Theorem 5.4.2. Suppose E ⊂ Rn is a measurable set. Then for each ε > 0:
1. There exists an open set A ⊂ Rn such that E ⊂ A and
m(A − E) < ε .
2. There exists a closed set F ⊂ Rn such that F ⊂ E and
m(E − F) < ε .
3. If m(E) is finite, there exists a compact set K ⊂ E such that
m(E \ K) < ε .
4. If m(E) is finite, there exist a finite union of closed cubes
F = ∪Nj=1 Q j
such that
m(E F) ≤ ε ,
where
E F = (E \ F) ∪ (F \ E).
Proof.
1. This item follows from the definition of measurable set.
2. Being E c measurable, there exists an open B ⊂ Rn such that E c ⊂ B and
m∗ (B \ E c ) < ε .
Defining F = Bc , we have that F is closed, F ⊂ E, and E \ F = B \ E c . Therefore

m(E − F) < ε .
3. Choose a closed set such that F ⊂ E e
ε
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 ) < ε ,
for all n sufficiently big.
4. Choose a sequence of closed cubes {Q j } such that
E ⊂ ∪∞j=1 Q j
and
∞
ε
∑ |Q j | ≤ m(E) + 2 .
j=1
Being m(E) < ∞ the series converges and there exists N0 ∈ N such that
∞
ε
∑ |Q j | < .
2
N0 +1
N
Defining F = ∪ j=1
0
Q j , we have
m(E F) = m(E − F) + m(F − E)

≤ m ∪∞j=N0 +1 Q j + m ∪∞j=1 Q j − E
∞ ∞
≤ ∑ |Q j | + ∑ |Q j | − m(E)
j=N0 +1 j=1
ε ε
≤ + = ε. (5.5)
2 2
5.5 Lebesgue Measurable Functions
Definition 5.5.1. 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.
5.5 Lebesgue Measurable Functions 139
Observe that:
1. If
f −1 ([−∞, a))
is measurable for each a ∈ R, then
f −1 ([−∞, a]) = ∩∞
k=1 f
−1
([−∞, a + 1/k))
is measurable for each a ∈ R.

2. If
f −1 ([−∞, a])
is measurable for each a ∈ R, then
f −1 ([−∞, a)) = ∪∞
k=1 f
−1
([−∞, a − 1/k])
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 . (5.6)
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 5.5.2. If f is continuous in Rn , then f is measurable. If f is measurable

and real and φ is continuous, then φ ◦ f is measurable.
Proof. The first implication is obvious. For the second, being φ continuous
φ −1 ([−∞, a))
is open, and therefore
(φ ◦ f )−1 (([−∞, a)) = f −1 (φ −1 ([−∞, a)))
is measurable, ∀a ∈ R.
Proposition 5.5.3. Suppose { fk } is a sequence of measurable functions. Then
sup fk (x), inf fk (x)

k∈N k∈N
and
lim sup fk (x), lim inf fk (x)
k→∞ k→∞
are measurable.
Proof. We will prove only that supn∈N fn (x) is measurable. The remaining proofs
are analogous. Let
f (x) = sup fn (x).
n∈N
Thus
f −1 ((a, +∞]) = ∪∞ −1
n=1 f n ((a, +∞]).
Being each fn measurable, such a set is measurable, ∀a ∈ R. By analogy
inf fk (x)
k∈N
is measurable and
lim sup fk (x) = inf sup f j (x),
k→∞ k≥1 j≥k
and
lim inf fk (x) = sup inf f j (x)
k→∞ k≥1 j≥k
are measurable.
Proposition 5.5.4. Let { fk } be a sequence of measurable functions such that
lim fk (x) = f (x).

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

k→∞ k→∞
The next result we do not prove it. For a proof see [62].
Proposition 5.5.5. 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.
Proposition 5.5.6. 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) = 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, +∞]). (5.7)
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 5.5.7. Suppose f is a nonnegative measurable function on Rn . Then there
exists an increasing sequence of nonnegative simple functions {ϕk } such that
lim ϕk (x) = f (x), ∀x ∈ Rn .

k→∞
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,
FN (x) = N, if x ∈ QN and f (x) > N,
⎩
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
NM
l
FN,M = ∑ M χEl,M ,
l=0
we have that FN,M is a simple function and
1
0 ≤ FN (x) − FN,M (x) ≤ .
M
If ϕK (x) = FK,K (x), we obtain
1
0 ≤ |FK (x) − ϕK (x)| ≤ .
K
Hence
| f (x) − ϕK (x)| ≤ | f (x) − FK (x)| + |FK (x) − ϕK (x)|.
Therefore
lim | f (x) − ϕK (x)| = 0, ∀x ∈ Rn .
K→∞
Theorem 5.5.8. Suppose that f is a measurable function defined on Rn . Then there

exists a sequence of simple functions {ϕk } such that
|ϕk (x)| ≤ |ϕk+1 (x)|, ∀x ∈ Rn , k ∈ N
and
lim ϕk (x) = f (x), ∀x ∈ Rn .
k→∞
Proof. Write
f (x) = f + (x) − f − (x),
where
f + (x) = max{ f (x), 0}
and
f − (x) = max{− f (x), 0}.
Thus f + and f − are nonnegative measurable functions so that from the last theorem
there exist increasing sequences of nonnegative simple functions such that
(1)
ϕk (x) → f + (x), ∀x ∈ Rn ,
and
(2)
ϕk (x) → f − (x), ∀x ∈ Rn ,
as k → ∞. Defining
(1) (2)
ϕk (x) = ϕk (x) − ϕk (x),
we obtain
ϕk (x) → f (x), ∀x ∈ Rn
as k → ∞ and
(1) (2)
|ϕk (x)| = ϕk (x) + ϕk (x) " | f (x)|, ∀x ∈ Rn ,
as k → ∞.
Theorem 5.5.9. Suppose f is a measurable function in Rn . Then there exists a

sequence of step functions {ϕk } 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 χE may be approximated almost everywhere in E by step func-
tions. Suppose given ε > 0. Observe that from Proposition 5.4.2, there exist cubes
Q1 , . . . , QN such that
m(E ∪Nj=1 Q j ) < ε .
We may obtain almost disjoints rectangles R̃ j such that ∪M
j=1 R̃ j = ∪ j=1 Q j and dis-
N
joints rectangles R j ⊂ R̃ j such that
m(E j=1 R j ) < 2ε .

∪M
Thus
M
f (x) = ∑ χR j ,
j=1
possibly except in a set of measure < 2ε . Hence, for each k > 0, there exists a step
function ϕk such that m(Ek ) < 2−k where
Ek = {x ∈ Rn | f (x) = ϕk (x)}.
Defining
Fk = ∪∞j=k+1 E j
we have
∞
m(Fk ) ≤ ∑ m(E j )
j=k+1
∞
≤ ∑ 2− j
j=k+1
2−(k+1)
=
1 − 1/2
= 2−k . (5.8)
Therefore also defining

F = ∩∞
k=1 Fk
we have m(F) = 0 considering that
m(F) ≤ 2−k , ∀k ∈ N.

ϕk (x) → f (x), ∀x ∈ F c .
Theorem 5.5.10 (Egorov). Suppose that { fk } is a sequence of measurable func-
tions defined in a measurable set E such that m(E) < ∞. Assume that fk →
f , a.e in E. Thus given ε > 0 we may find a closed set Aε ⊂ E such that fk → f
uniformly in Aε and m(E − Aε ) < ε .
Proof. Without losing generality we may assume that
fk → f , ∀x ∈ E.
For each N, k ∈ N define
EkN = {x ∈ E | | f j (x) − f (x)| < 1/N, ∀ j ≥ k}.
Fixing N ∈ N, we may observe that
EkN ⊂ Ek+1
N
and that ∪∞
k=1 Ek = E. Thus we may obtain kN such that
N
1
m(E − EkNN ) < .
2N
Observe that
1
| f j (x) − f (x)| < , ∀ j ≥ kN , x ∈ EkNN .
N
Choose M ∈ N such that
∞
ε
∑ 2−k ≤ 2 .
k=M
Define
Ãε = ∩∞
N≥M EkN .
N
Thus
∞
ε
m(E − Ãε ) ≤ ∑ m(E − EkNN ) < .
2
N=M
Suppose given δ > 0. Let N ∈ N be such that N > M and 1/N < δ . Thus if x ∈ Ãε
then x ∈ EkNN so that
| f j (x) − f (x)| < δ , ∀ j > kN .
Hence fk → f uniformly in Ãε . Observe that Ãε is measurable and thus there exists
a closed set Aε ⊂ Ãε such that
ε
m(Ãε − Aε ) < .
2
That is
ε ε
m(E − Aε ) ≤ m(E − Ãε ) + m(Ãε − Aε ) < + = ε,
2 2
and
fk → f
uniformly in Aε . The proof is complete.
Definition 5.5.11. We say that f : Rn → [−∞, +∞] ∈ L1 (Rn if f is measurable and

| f | dx < ∞.
Rn
Definition 5.5.12. We say that a set A ⊂ L1 (Rn ) is dense in L1 (Rn ), if for each
f ∈ L1 (Rn ) and each ε > 0 there exists g ∈ A such that

f − g L1(Rn ) = | f − g| dx < ε .
Rn
Theorem 5.5.13. 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 func-
tions such that
ϕk (x) → f (x) a.e. in Rn .
Since {ϕk } may be also such that
|ϕk | ≤ | f |, ∀k ∈ N
from the Lebesgue dominated converge theorem, we have
ϕk − f L1 (Rn ) → 0,
as k → ∞.
2. From the last item, it suffices to show that simple functions may be 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 ε > 0 and a set
of finite measure, there exists ϕ a step function such that
χE − ϕ L1 (Rn ) < ε .
This may be made similar as in Theorem 5.5.9.

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 = χ[a,b] by g(x), where g is continuous, and be linear on (a − ε , a)
and (b, b + ε ) and

1, if a ≤ x ≤ b,
g(x) =
0, if x ≤ a − ε or x ≥ b + ε .
Thus
f − g L1 (Rn ) < 2ε .
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 approxi-
mate f by the product of n functions similar to g defined above.
Chapter 6
Other Topics in Measure and Integration
In this chapter we present some important results which may be found in similar
form at Chapters 2, 6, and 7 in the excellent book Real and Complex Analysis, [57]
by Rudin, where more details may be found.
6.1 Some Preliminary Results
In the next results μ is a measure on U. We start with the following theorem.

Theorem 6.1.1. Let f : U → [0, ∞] be a measurable function. If E ∈ M and

f d μ = 0,
E
then
f = 0, a.e. in E.
Proof. Define
An = {u ∈ E | f (u) > 1/n}, ∀n ∈ N.
Thus

μ (An )/n ≤ f dμ ≤ f d μ = 0.
An E
Therefore μ (An ) = 0, ∀n ∈ N.
Define
A = {u ∈ E | f (u) > 0}.
Hence,
A = ∪∞
n=1 An ,
so that μ (A) = 0.
148 6 Other Topics in Measure and Integration
Thus,
f = 0, a.e. in E.

Theorem 6.1.2. Assume f ∈ L1 (μ ) and E f d μ = 0, ∀E ∈ M . Under such hypothe-
ses, f = 0, a.e. in U.
Proof. Consider first the case f : U → [−∞, +∞]. Define
An = {u ∈ U | f (u) > 1/n}, ∀n ∈ N.
Thus,

μ (An )/n ≤ f d μ = 0.
An
Hence, μ (An ) = 0, ∀n ∈ N.
Define
A = {u ∈ E | f (u) > 0}.
Therefore,
A = ∪∞
n=1 An ,
so that μ (A) = 0.
Thus,
f ≤ 0, a.e. in U.
By analogy we get
f ≥ 0, a.e. in U,
so that
f = 0, a.e. in U.
To complete the proof, just apply this last result to the real and imaginary parts of a
complex f .
Theorem 6.1.3. Suppose μ (U) < ∞ and f ∈ L1 (μ ). Moreover, assume

E |f| dμ
≤ α ∈ [0, ∞), ∀E ∈ M .
μ (E)
Under such hypothesis we have
| f | ≤ α , a.e. in U.
Proof. Define
An = {u ∈ U | | f (u)| > α + 1/n}, ∀n ∈ N.
Thus, if μ (An ) > 0, we get

An (| f | − α ) d μ An | f |dμ
1/n ≤ = − α ≤ 0,
μ (An ) μ (An )
a contradiction. Hence, μ (An ) = 0, ∀n ∈ N.

6.1 Some Preliminary Results 149
Define
A = {u ∈ U | | f (u)| > α }.
Therefore,
A = ∪∞
n=1 An ,
so that μ (A) = 0.
Thus,
| f (u)| ≤ α , a.e. in U.
At this point we present some preliminary results to the development of the well-
known Urysohn’s lemma.
Theorem 6.1.4. Let U be a Hausdorff space and K ⊂ U compact. Let v ∈ K c . Then

there exist open sets V and W ⊂ U such that v ∈ V , K ⊂ W and V ∩W = 0.
/
Proof. For each u ∈ K there exist open sets Wu ,Vvu ⊂ U such that u ∈ Wu , v ∈ Wvu ,
and Wu ∩Vvu = 0.
/
Observe that K ⊂ ∪u∈K Wu so that, since K is compact, there exist u1 , u2 , . . . , un ∈
K such that
K ⊂ ∪ni=1Wui .
Finally, defining the open sets
V = ∩ni=1Vvui
and
W = ∪ni=1Wui ,
we get
V ∩W = 0,
/
v ∈ V , and K ⊂ W .
Theorem 6.1.5. Let {Kα , α ∈ L} be a collection of compact subsets of a Hausdorff

space U.
Assume ∩α ∈L Kα = 0./ Under such hypotheses some finite subcollection of
{Kα , α ∈ L} has empty intersection.
Proof. Define Vα = Kαc , ∀α ∈ L. Fix α0 ∈ L. From the hypotheses
Kα0 ∩ [∩α ∈L\{α0 } Kα ] = 0.

/
Hence
Kα0 ⊂ [∩α ∈L\{α0 } Kα ]c ,
that is,
Kα0 ⊂ ∪α ∈L\{α0 } Kαc = ∪α ∈L\{α0 }Vα .
Since Kα0 is compact, there exists α1 , . . . , αn ∈ L such that

Kα0 ⊂ Vα1 ∪ . . . ∪Vαn = (Kα1 ∩ . . . ∩ Kαn )c ,
so that
Kα0 ∩ Kα1 ∩ . . . ∩ Kαn = 0.
/
Definition 6.1.6. We say that a space U is locally compact if each u ∈ U has a
neighborhood whose closure is compact.
Theorem 6.1.7. Let U be a locally compact Hausdorff space. Suppose W ⊂ U is
open and K ⊂ W , where K is compact. Then there exists an open set V ⊂ U whose
closure is compact and such that
K ⊂ V ⊂ V ⊂ W.
Proof. Let u ∈ K. Since U is locally compact there exists an open Vu ⊂ U such that
u ∈ Vu and V u is compact.
Observe that
K ⊂ ∪u∈K Vu
and since K is compact there exist u1 , u2 , . . . , un ∈ K such that
K ⊂ ∪nj=1Vu j .
Hence, defining G = ∪nj=1Vu j , we get
K ⊂ G,
where G is compact.
If W = U define V = G and the proof would be complete.
Otherwise, if W = U define C = U \ W . From Theorem 6.1.4, for each v ∈ C,
there exists an open Wv such that K ⊂ Wv and v ∈ W v .
Hence {C ∩ G ∩ W v : v ∈ C} is a collection of compact sets with empty
intersection.
From Theorem 6.1.5 there are points v1 , . . . , vn ∈ C such that
C ∩ G ∩W v1 ∩ . . . ∩W vn = 0.
/
Defining
V = G ∩Wv1 ∩ . . . ∩Wvn
we obtain
V ⊂ G ∩W v1 ∩ . . . ∩W vn .
Also,
K ⊂ V ⊂ V ⊂ W.
6.1 Some Preliminary Results 151
Definition 6.1.8. Let f : U → [−∞, +∞] be a function on a topological space U.

We say that f is lower semicontinuous if Aα = {u ∈ U : f (u) > α } is open for all
α ∈ R. Similarly, we say that f is upper semicontinuous if Bα = {u ∈ U : f (u) < α }
is open for all α ∈ R.
Observe that from this last definition f is continuous if and only if it is both lower
and upper semicontinuous.
Here we state and prove a very important result, namely, the Uryshon’s lemma.
Lemma 6.1.9 (Urysohn’s Lemma). Assume U is a locally compact Hausdorff

space and V ⊂ U is an open set which contains a compact set K. Under such ass-
umptions, there exists a function f ∈ Cc (V ) such that
• 0 ≤ f (u) ≤ 1, ∀u ∈ V ,
• f (u) = 1, ∀u ∈ K.
Proof. Set r1 = 0 and r2 = 1, and let r3 , r4 , r5 , . . . be an enumeration of the rational

numbers in (0, 1). Observe that we may find open sets V0 and V1 such that V 0 is
compact and
K ⊂ V1 ⊂ V 1 ⊂ V0 ⊂ V 0 ⊂ V.
Reasoning by induction, suppose n ≥ 2 and that Vr1 , . . . ,Vrn have been chosen so
that if ri < r j then V r j ⊂ Vri . Denote
ri = max{rk | k ∈ {1, . . . , n} and rk < rn+1 }
and
r j = min{rk , | k ∈ {1, . . . , n} and rk > rn+1 }.
We may find again an open set Vrn+1 such that
V r j ⊂ Vrn+1 ⊂ V rn+1 ⊂ Vri .
Thus, we have obtained a sequence Vr of open sets such that for every r rational in
(0, 1), V r is compact and if s > r then V s ⊂ Vr . Define

r, if u ∈ Vr ,
fr (u) =
0, otherwise,
and
1, if u ∈ V s ,
gs (u) =
s, otherwise.
Also define
f (u) = sup fr (u), ∀u ∈ V
r∈Q∩(0,1)
and
g(u) = inf gs (u), ∀u ∈ V.
s∈Q∩(0,1)
Observe that f is lower semicontinuous and g is upper semicontinuous. Moreover,

0≤ f ≤1
and
f = 1, if u ∈ K.
Observe also that the support of f is contained in V 0 .
To complete the proof, it suffices to show that
f = g.
The inequality
fr (u) > gs (u)
is possible only if r > s, u ∈ Vr , and u ∈ V s .
But if r > s, then Vr ⊂ Vs , and hence fr ≤ gs , ∀r, s ∈ Q ∩ (0, 1), so that f ≤ g.
Suppose there exists u ∈ V such that
f (u) < g(u).
Thus there exist rational numbers r, s such that

f (u) < r < s < g(u).
Since f (u) < r, u ∈ Vr . Since g(u) > s, u ∈ V s .
As V s ⊂ Vr , we have a contradiction. Hence f = g, and such a function is
continuous.
Theorem 6.1.10 (Partition of Unity). Let U be a locally compact Hausdorff space.

Assume K ⊂ U is compact so that
K ⊂ ∪ni=1Vi ,
where Vi is open ∀i ∈ {1, . . . , n}. Under such hypotheses, there exists functions
h1 , . . . , hn such that
n
∑ hi = 1, on K,
i=1
hi ∈ Cc (Vi ) and 0 ≤ hi ≤ 1, ∀i ∈ {1, . . . , n}.
Proof. Let u ∈ K ⊂ ∪ni=1Vi . Thus there exists j ∈ {1, . . . , n} such that u ∈ V j . We

may select an open set Wu such that W u is compact and W u ⊂ V j .
Observe that
K ⊂ ∪u∈K Wu .
From this, since K is compact, there exist u1 , . . . , uN such that
K ⊂ ∪Nj=1Wu j .
For each i ∈ {1, . . . , n} define by W̃i the union of those Wu j , contained in Vi .

6.2 The Riesz Representation Theorem 153
By the Uryshon’s lemma we may find continuous functions gi such that

gi = 1, on W̃i ,
gi ∈ Cc (Vi ),
0 ≤ gi ≤ 1, ∀i ∈ {1, . . . , n}.
Define
h1 = g1
h2 = (1 − g1)g2
h3 = (1 − g1)(1 − g2)g3
... ...............
hn = (1 − g1)(1 − g2) . . . (1 − gn−1)gn . (6.1)
Thus,
0 ≤ hi ≤ 1 and hi ∈ Cc (Vi ), ∀i ∈ {1, . . ., n}.
Furthermore, by induction, we may obtain
h1 + h2 + . . . + hn = 1 − (1 − g1)(1 − g2) . . . (1 − gn).
Finally, if u ∈ K then u ∈ W̃i for some i ∈ {1, . . . , n}, so that gi (u) = 1 and hence
(h1 + . . . + hn )(u) = 1, ∀u ∈ K.
The set {h1 , . . . , hn } is said to be a partition of unity on K subordinate to the open
cover {V1, . . . ,Vn }.
6.2 The Riesz Representation Theorem
In the next lines we introduce the main result in this section, namely, the Riesz
representation theorem.
Theorem 6.2.1 (Riesz Representation Theorem). Let U be a locally compact
Hausdorff space and let F be a positive linear functional on Cc (U). Then there
exists a σ -algebra M in U which contains all the Borel sets and there exists a
unique positive measure μ on M such that

1. F( f ) = U f d μ , ∀ f ∈ Cc (U),
2. μ (K) < ∞, for every compact K ⊂ U,
3. μ (E) = inf{μ (V ) | E ⊂ V, V open }, ∀E ∈ M ,
4. μ (E) = sup{μ (K) | K ⊂ E, K compact} holds for all open E and all E ∈ M
such that μ (E) < ∞,
5. If E ∈ M , A ⊂ E and μ (E) = 0 then A ∈ M .
Proof. We start by proving the uniqueness of μ . If μ satisfies 3 and 4, then μ is

determined by its values on compact sets. Then, if μ1 and μ2 are two measures for
which the theorem holds, to prove uniqueness, it suffices to show that
μ1 (K) = μ2 (K)
for every compact K ⊂ U. Let ε > 0. Fix a compact K ⊂ U. By 2 and 3, there exists
an open V ⊃ K such that
μ2 (V ) < μ2 (K) + ε .
By the Urysohn’s lemma, there exists a f ∈ Cc (V ) such that
0 ≤ f (u) ≤ 1, ∀u ∈ V
and
f (u) = 1, ∀u ∈ K.
Thus,

μ1 (K) = χ K d μ1
U
≤ f d μ1
U
= F( f )

= f d μ2
U
≤ χV d μ2
U
= μ2 (V )
< μ2 (K) + ε . (6.2)
μ1 (K) ≤ μ2 (K).
Interchanging the roles of μ1 and μ2 we similarly obtain
μ2 (K) ≤ μ1 (K),
so that
μ1 (K) = μ2 (K).
The proof of uniqueness is complete.
Now for every open V ⊂ U, define
μ (V ) = sup{F( f ) | f ∈ Cc (V ) and 0 ≤ f ≤ 1}.

If V1 ,V2 are open and V1 ⊂ V2 , then
μ (V1 ) ≤ μ (V2 ).
Hence,
μ (E) = inf{μ (V ) | E ⊂ V, V open},
if E is an open set. Define
μ (E) = inf{μ (V ) | E ⊂ V, V open},
∀E ⊂ U. Define by MF the collection of all E ⊂ U such that μ (E) < ∞ and
μ (E) = sup{μ (K) | K ⊂ E, K compact}.
Finally, define by M the collection of all sets such that E ⊂ U and E ∩ K ∈ MF for
all compact K ⊂ U. Since
μ (A) ≤ μ (B),
if A ⊂ B we have that μ (E) = 0 implies E ∩ K ∈ MF for all K compact, so that
E ∈ M . Thus, 5 holds and so does 3 by definition.
Observe that if f ≥ 0, then F( f ) ≥ 0, that is, if f ≤ g then F( f ) ≤ F(g).
Now we prove that if {En } ⊂ U is a sequence, then
∞
μ (∪∞
n=1 En ) ≤ ∑ μ (En ). (6.3)
n=1
First we show that

μ (V1 ∪V2 ) ≤ μ (V1 ) + μ (V2),
if V1 ,V2 are open sets.
Choose g ∈ Cc (V1 ∪V2 ) such that
0 ≤ g ≤ 1.
By Theorem 6.1.10 there exist functions h1 and h2 such that hi ∈ Cc (Vi ) and
0 ≤ hi ≤ 1
and so that h1 + h2 = 1 on the support of g. Hence, hi ∈ Cc (Vi ), 0 ≤ hi g ≤ 1, and
g = (h1 + h2)g and thus
F(g) = F(h1 g) + F(h2 g) ≤ μ (V1 ) + μ (V2).
Since g is arbitrary, from the definition of μ , we obtain
μ (V1 ∪V2 ) ≤ μ (V1 ) + μ (V2).
Furthermore, if μ (En ) = ∞, for some n ∈ N, then (6.3) is obviously valid. Assume

then μ (En ) < ∞, ∀n ∈ N.
Let a not relabeled ε > 0. Therefore for each n ∈ N there exists an open Vn ⊃ En
such that
ε
μ (Vn ) < μ (En ) + n .
2
Define
V = ∪∞ n=1Vn ,
and choose f ∈ Cc (V ) such that 0 ≤ f ≤ 1. Since the support of f is compact, there

exists N ∈ N such that
spt( f ) ⊂ ∪Nn=1Vn .
Therefore

F( f ) ≤ μ ∪Nn=1Vn
N
≤ ∑ μ (Vn)
n=1
∞
≤ ∑ μ (En) + ε . (6.4)
n=1
Since this holds for any f ∈ Cc (V ) with 0 ≤ f ≤ 1 and ∪∞

n=1 En ⊂ V , we get
∞
n=1 En ) ≤ μ (V ) ≤ ∑ μ (En ) + ε .
μ (∪∞
i=1
Since ε > 0 is arbitrary, we have proven (6.3).

In the next lines we prove that if K is compact, then K ∈ MF and
μ (K) = inf{F( f ) | f ∈ Cc (U), f = 1 on K}. (6.5)
For if f ∈ Cc (U), f = 1 on K, and 0 < α < 1, define
Vα = {u ∈ U | f (u) > α }.
Thus, K ⊂ Vα and if g ∈ Cc (Vα ) and 0 ≤ g ≤ 1 we get

αg ≤ f .
Hence,
μ (K) ≤ μ (Vα )
= sup{F(g) | g ∈ Cc (Vα ), 0 ≤ g ≤ 1}
≤ α −1 F( f ). (6.6)
Letting α → 1 we obtain
μ (K) ≤ F( f ).
Thus μ (K) < ∞, and obviously K ∈ MF .
Also there exists an open V ⊃ K such that
μ (V ) < μ (K) + ε .
By the Urysohn’s lemma, we may find f ∈ Cc (V ) such that f = 1 on K and

0 ≤ f ≤ 1. Thus
F( f ) ≤ μ (V ) < μ (K) + ε .
Since ε > 0 is arbitrary, (6.5) holds.
At this point we prove that for every open V we have
μ (V ) = sup{μ (K) | K ⊂ V, K compact} (6.7)
and hence MF contains every open set such that μ (V ) < ∞.

Let V ⊂ U be an open set such that μ (V ) < ∞.
Let α ∈ R be such that α < μ (V ). Therefore there exists f ∈ Cc (V ) such that
0 ≤ f ≤ 1 and such that α < F( f ).
If W ⊂ U is an open set such that K = spt( f ) ⊂ W , we have that f ∈ Cc (W ) and
0 ≤ f ≤ 1 so that
F( f ) ≤ μ (W ).
Thus, since W ⊃ K is arbitrary, we obtain
F( f ) ≤ μ (K),
so that
α < μ (K),
where K ⊂ U is a compact set.
Hence (6.7) holds.
Suppose that
E = ∪∞
n=1 En ,
where {En } is a sequence of disjoint sets in MF .

We are going to show that
∞
μ (E) = ∑ μ (En). (6.8)
n=1
In addition if μ (E) < ∞, then also E ⊂ MF .

First we show that if K1 , K2 ⊂ U are compact disjoint sets, then
μ (K1 ∪ K2 ) = μ (K1 ) + μ (K2 ). (6.9)
From the Urysohn’s lemma there exists f ∈ Cc (U) such that f = 1 on K1 , f = 0 on
K2 , and
0 ≤ f ≤ 1.
From (6.5) there exists g ∈ Cc (U) such that g = 1 on K1 ∪ K2 and
F(g) < μ (K1 ∪ K2 ) + ε .

Observe that f g = 1 on K1 and (1 − f )g = 1 on K2 and also f g, (1 − f )g ∈ Cc (U)

and 0 ≤ f g ≤ 1 and 0 ≤ (1 − f )g ≤ 1 so that
μ (K1 ) + μ (K2 ) ≤ F( f g) + F((1 − f )g)

= F(g)
≤ μ (K1 ∪ K2 ) + ε . (6.10)
Since ε > 0 is arbitrary we obtain
μ (K1 ) + μ (K2 ) ≤ μ (K1 ∪ K2 ).
From this (6.9) holds.

Also if μ (E) = ∞, (6.8) follows from (6.3).
Thus assume μ (E) < ∞.
Since En ∈ MF , ∀n ∈ N we may obtain compact sets Hn ⊂ En such that
ε
μ (Hn ) > μ (En ) − , ∀n ∈ N.
2n
Defining KN = ∪Nn=1 Hn , by 3 we get
μ (E) ≥ μ (KN )
N
= ∑ μ (Hn )
n=1
N
≥ ∑ μ (En ) − ε , ∀N ∈ N. (6.11)
n=1
Since N ∈ N and ε > 0 are arbitrary we get

∞
μ (E) ≥ ∑ μ (En).
n=1

∞
μ (E) = ∑ μ (En). (6.12)
n=1
Let ε0 > 2ε . If μ (E) < ∞, there exists N0 ∈ N such that μ (KN0 ) > ∑∞
n=1 μ (En ) − ε0 .
μ (E) ≤ μ (KN0 ) + ε0 .
Therefore, since ε > 0 and ε0 > 2ε are arbitrary, we may conclude that E satisfies 4
so that E ∈ MF .
Now we prove the following.
If E ∈ MF there is a compact K ⊂ U and an open V ⊂ U such that K ⊂ E ⊂ V

and
μ (V \ K) < ε .
From above, there exists a compact K and an open V such that
K ⊂E ⊂V
and
ε ε
μ (V ) −
< μ (E) < μ (K) + .
2 2
Since V \ K is open and of finite measure, it is in MF . From the last chain of
inequalities we obtain
μ (K) + μ (V \ K) = μ (V ) < μ (K) + ε ,
so that
μ (V \ K) < ε .
In the next lines we prove that if A, B ∈ MF then
A \ B, A ∪ B and A ∩ B ∈ MF .
By above there exist compact sets K1 , K2 and open sets V1 ,V2 such that
K1 ⊂ A ⊂ V1 , K2 ⊂ B ⊂ V2
and
μ (Vi \ Ki ) < ε , ∀i ∈ {1, 2}.
Since
(A \ B) ⊂ (V1 \ K2 ) ⊂ (V1 \ K1 ) ∪ (K1 \ V2) ∪ (V2 \ K2 ),
we get
μ (A \ B) < ε + μ (K1 \ V2) + ε ,
Since K1 \ V2 ⊂ A \ B is compact and ε > 0 is arbitrary, we get
A \ B ∈ MF .
Since
A ∪ B = (A \ B) ∪ B,
we obtain
A ∪ B ∈ MF .
Since
A ∩ B = A \ (A \ B)
we get
A ∩ B ∈ MF .
At this point we prove that M is a σ -algebra in U which contains all the Borel sets.
Let K ⊂ U be a compact subset. If A ∈ M then
Ac ∩ K = K \ (A ∩ K),
so that Ac ∩ K ∈ MF considering that K ∈ MF and A ∩ K ∈ MF .
Thus if A ∈ M then Ac ∈ M .
Next suppose
A = ∪∞n=1 An ,
where An ∈ M , ∀n ∈ N.
Define B1 = A1 ∩ K and
Bn = (An ∩ K) \ (B1 ∪ B2 ∪ . . . ∪ Bn−1 ),
∀n ≥ 2, n ∈ N.
Then {Bn } is disjoint sequence of sets in MF .
Thus
A ∩ K = ∪∞ n=1 Bn ∈ MF .
Hence A ∈ M . Finally, if C ⊂ U is a closed subset, then C ∩ K is compact, so that

C ∩ K ∈ MF . Hence C ∈ M .
Therefore M is a σ -algebra which contains the closed sets, so that it contains
the Borel sets.
Finally, we will prove that
MF = {E ∈ M | μ (E) < ∞}.
For, if E ∈ MF then E ∩ K ∈ MF for all compact K ⊂ U, hence E ∈ M .
Conversely, assume E ∈ M and μ (E) < ∞. There is an open V ⊃ E such that
μ (V ) < ∞. Pick a compact K ⊂ V such that
μ (V \ K) < ε .
Since E ∩ K ∈ MF there is a compact K1 ⊂ (E ∩ K) such that
μ (E ∩ K) < μ (K1 ) + ε .
Since
E ⊂ (E ∩ K) ∪ (V \ K),
it follows that
μ (E) ≤ μ (E ∩ K) + μ (V \ K) < μ (K1 ) + 2ε .
This implies E ∈ MF .
To finish the proof, we show that

F( f ) = f d μ , ∀ f ∈ Cc (U).
U
From linearity it suffices to prove the result for the case where f is real.
Let f ∈ Cc (U). Let K be the support of f and let [a, b] ⊂ R be such that
R( f ) ⊂ (a, b),
where R( f ) denotes the range of f .

Suppose given a not relabeled ε > 0. Choose a partition of [a, b] denoted by
{yi } = {a = y0 < y1 < y2 < . . . < yn = b},
such that yi − yi−1 < ε , ∀i ∈ {1, . . . , n}.

Denote
Ei = {u ∈ K | yi−1 < f (u) ≤ yi },
∀i ∈ {1, . . . , n}.
Since f is continuous, it is Borel measurable, and the sets Ei are disjoint Borel
ones such that
∪ni=1 Ei = K.
Select open sets Vi ⊃ Ei such that
ε
μ (Vi ) < μ (Ei ) + , ∀i ∈ {1, . . . , n},
n
and such that
f (u) < yi + ε , ∀u ∈ Vi .
From Theorem 6.1.10 there exists a partition of unity subordinate to {Vi }ni=1 such
that hi ∈ Cc (Vi ), 0 ≤ hi ≤ 1 and
n
∑ hi = 1, on K.
i=1
Hence
n
f = ∑ hi f
i=1
and

n n
μ (K) ≤ F ∑ hi f = ∑ F(hi f ).
i=1 i=1
Observe that
ε
μ (Ei ) + > μ (Vi )
n
= sup{F( f ) | f ∈ Cc (Vi ), 0 ≤ f ≤ 1}
> F(hi ), ∀i ∈ {1, . . . , n}. (6.13)
Thus
n
F( f ) = ∑ F(hi f )
i=1
n
≤ ∑ F(hi (yi−1 + 2ε ))
i=1
n
= ∑ (yi−1 + 2ε )F(hi )
i=1
n ε
< ∑ (yi−1 + 2ε ) μ (Ei ) +
n
i=1
n n
ε n
< ∑ yi−1 μ (Ei ) + ∑ (yi−1 ) n + 2ε ∑ μ (Ei ) + 2ε 2
i=1 i=1 i=1

< f d μ + bε + 2ε μ (K) + 2ε . 2
(6.14)
U
Since ε > 0 is arbitrary, we obtain

F( f ) ≤ f d μ , ∀ f ∈ Cc (U).
U
From this
F(− f ) ≤ (− f ) d μ , ∀ f ∈ Cc (U),
U
that is,

F( f ) ≥ f d μ , ∀ f ∈ Cc (U).
U
Hence
F( f ) = f d μ , ∀ f ∈ Cc (U).
U
6.3 The Lebesgue Points
In this section we introduce a very important concept in analysis, namely, the

definition of Lebesgue points.
We recall that in Rn the open ball with center u and radius r is defined by
Br (u) = {v ∈ Rn | |v − u|2 < r}.
Consider a Borel measure μ on Rn . We may associate to μ , the function Fr μ (u),
denoted by
μ (Br (u))
Fr μ (u) = ,
m(Br (u))
where m denotes the Lebesgue measure.
6.3 The Lebesgue Points 163
We define the symmetric derivative of μ at u, by (Dμ )(u), by

(Dμ )(u) = lim Fr μ (u),
r→0
whenever such a limit exists.

We also define the function Gμ for a positive measure μ by
Gμ (u) = sup Fr μ (u).
0<r<∞
The function Gμ : Rn → [0, +∞] is lower semicontinuous and hence measurable.

Lemma 6.3.1. Let W = ∪Ni=1 Bri (ui ) be a finite union of open balls. Then there is a
set S ⊂ {1, 2, . . . , N} such that
1. The balls Bri (ui ), i ∈ S are disjoint.
2. W ⊂ ∪i∈S B3ri (ui ).
Proof. Let us first order the balls Bri (ui ) so that

r1 ≥ r2 ≥ . . . ≥ rN .
Set i1 = 1, and discard all balls such that
Bi1 ∩ B j = 0.
/
Let Bi2 be the first of the remaining balls, if any. Discard all B j such that j > i2 and
Bi2 ∩ B j = 0.
/
Let Bi3 the first of the remaining balls as long as possible. Such a process stops
after a finite number of steps. Define S = {i1 , i2 , . . .}. It is clear that 1 holds. Now
we prove that each discarded B j is contained in
{B3ri , i ∈ S}.
Just observe that if r < r and Br (u ) intersects Br (u), then Br (u ) ⊂ B3r (u).
Theorem 6.3.2. Suppose μ is a finite Borel measure on Rn and λ > 0. Then

m(Aλ ) ≤ 3n λ −1 μ ,
where
Aλ = {u ∈ U | Gμ (u) > λ }
and
μ = |μ |(Rn ).
Proof. Let K be a compact subset of the open set Aλ .

As Gμ (u) = sup0<r<∞ {Fr μ (u)}, each u ∈ K is the center of an open ball Bu such
that
μ (Bu ) > λ m(Bu ).
Since K is compact, there exists a finite number of such balls which covers K. By
Lemma 6.3.1, there exists a disjoint subcollection here denoted by {Br1 , . . . , BrN }
such that K ⊂ ∪Nk=1 B3rk , so that
N
m(K) ≤ 3n ∑ m(Brk )
k=1
N
≤ 3n λ −1 ∑ |μ |(Brk )
k=1
≤ 3n λ −1 μ . (6.15)
The result follows taking the supremum relating all compact K ⊂ Aλ .
Remark 6.3.3. Observe that, if f ∈ L1 (Rn ) and λ > 0, for Aλ = {u ∈ Rn | | f | > λ },

we have
m(Aλ ) ≤ λ −1 f 1 .
This follows from the fact that

λ m(Aλ ) ≤ | f | dm ≤ | f | dm = f 1 .
Aλ Rn
Observe also that defining d η = | f | dm, for every λ > 0, defining
η (Br (u))
G f (u) = sup
0<r<∞ m(B r (u))
and
Aλ = {u ∈ U | G f (u) > λ },
we have
m(Aλ ) ≤ 3n λ −1 f 1 .
6.3.1 Lebesgue Points
Finally in this section we present the main definition of Lebesgue points and
some relating results.
Definition 6.3.4. Let f ∈ L1 (Rn ). A point u ∈ L1 (Rn ) such that

1
lim | f (v) − f (u)| dm(v) = 0
r→0 m(Br (u)) Br (u)
is called a Lebesgue point of f .
Theorem 6.3.5. If f ∈ L1 (Rn ), then almost all u ∈ Rn is a Lebesgue point of f .

6.3 The Lebesgue Points 165
Proof. Define

1
Hr f (u) = | f − f (u)| dm, ∀u ∈ Rn , r > 0,
m(Br (u)) Br (u)
and also define

H f (u) = lim sup Hr f (u).
r→0
We have to show that H f = 0, a.e. [m].

Select y > 0 and fix k ∈ N. Observe that there exists g ∈ C(Rn ) such that
f − g 1 < 1/k.
Define h = f − g. Since g is continuous, Hg = 0 in Rn . Observe that

1
Hrh (u) = |h − h(u)| dm
m(Br (u)) Br (u)

1
≤ |h| dm + |h(u)|, (6.16)
m(Br (u)) Br (u)
so that
Hh < Gh + |h|.
Since
Hr f ≤ Hrg + Hrh ,
we obtain
H f ≤ Gh + |h|.
Define
Ay = {u ∈ Rn | H f (u) > 2y},
By,k = {u ∈ Rn | Gh (u) > y},
and
Cy,k = {u ∈ Rn | |h| > y}.
Observe that h 1 < 1/k, so that from Remark 6.3.3 we obtain
3n
m(By,k ) ≤
yk
and
1
m(Cy,k ) ≤
yk
and hence
3n + 1
m(By,k ∪Cy,k ) ≤ .
yk
Therefore
3n + 1
m(Ay ) ≤ m(By,k ∪Cy,k ) ≤ .
yk
Since k is arbitrary, we get m(Ay ) = 0, ∀y > 0 so that m{u ∈ Rn | H f (u) > 0} = 0.
We finish this section with the following result.

Theorem 6.3.6. Suppose μ is a complex Borel measure on Rn such that μ m.
Suppose f is the Radon–Nikodym derivative of μ with respect to m. Under such
assumptions,
Dμ = f , a.e. [m]
and
μ (E) = Dμ dm,
E
for all Borel set E ⊂ Rn .
Proof. From the Radon–Nikodym theorem we have

μ (E) = f dm,
E
for all measurable set E ⊂ Rn .

Observe that at any Lebesgue point u of f we have

1
f (u) = lim f dm
r→0 m(Br (u)) Br (u)
μ (Br (u))
= lim
r→0 m(Br (u))
= Dμ (u). (6.17)

Chapter 7
Distributions
The main reference for this chapter is Rudin [58].
7.1 Basic Definitions and Results
Definition 7.1.1 (Test Functions, the Space D(Ω )). Let Ω ⊂ Rn be a nonempty
open set. For each K ⊂ Ω compact, consider the space DK , the set of all C∞ (Ω )
functions with support in K. We define the space of test functions, denoted by
D(Ω ) as
D(Ω ) = ∪K⊂Ω DK , K compact. (7.1)
Thus φ ∈ D(Ω ) if and only if φ ∈ C∞ (Ω ) and the support of φ is a compact subset

of Ω .
Definition 7.1.2 (Topology for D(Ω )). Let Ω ⊂ Rn be an open set.

1. For every K ⊂ Ω compact, σK denotes the topology which a local base is defined
by {VN,k }, where N, k ∈ N,
VN,k = {φ ∈ DK | φ N < 1/k} (7.2)

and
φ N = max{|Dα φ (x)| | x ∈ Ω , |α | ≤ N}. (7.3)
2. σ̂ denotes the collection of all convex balanced sets W ∈ D(Ω ) such that W ∩
DK ⊂ σK for every compact K ⊂ Ω .
3. We define σ in D(Ω ) as the collection of all unions of sets of the form φ + W ,
for φ ∈ D(Ω ) and W ∈ σ̂ .
168 7 Distributions
Theorem 7.1.3. Concerning the last definition we have the following:

1. σ is a topology in D(Ω ).
2. Through σ , D(Ω ) is made into a locally convex topological vector space.
Proof.
1. From item 3 of Definition 7.1.2, it is clear that arbitrary unions of elements of
σ are elements of σ . Let us now show that finite intersections of elements of σ
also belong to σ . Suppose V1 ∈ σ and V2 ∈ σ ; if V1 ∩ V2 = 0,
/ we are done. Thus,
suppose φ ∈ V1 ∩ V2 . By the definition of σ there exist two sets of indices L1 and
L2 , such that
Vi = ∪λ ∈Li (φiλ + Wiλ ), for i = 1, 2, (7.4)
and as φ ∈ V1 ∩ V2 there exist φi ∈ D(Ω ) and Wi ∈ σ̂ such that
φ ∈ φi + Wi , for i = 1, 2. (7.5)
Thus there exists K ⊂ Ω such that φi ∈ DK for i ∈ {1, 2}. Since DK ∩ Wi ∈ σK ,

DK ∩ Wi is open in DK so that from (7.5) there exists 0 < δi < 1 such that
φ − φi ∈ (1 − δi)Wi , for i ∈ {1, 2}. (7.6)
From (7.6) and from the convexity of Wi we have
φ − φi + δi Wi ⊂ (1 − δi)Wi + δi Wi = Wi (7.7)
so that
φ + δi Wi ⊂ φi + Wi ⊂ Vi , for i ∈ {1, 2}. (7.8)
Define Wφ = (δ1 W1 ) ∩ (δ2 W2 ) so that
φ + Wφ ⊂ V i , (7.9)
and therefore we may write
V1 ∩ V2 = ∪φ ∈V1 ∩V2 (φ + Wφ ) ∈ σ . (7.10)

2. It suffices to show that single points are closed sets in D(Ω ) and the vector space
operations are continuous.
(a) Pick φ1 , φ2 ∈ D(Ω ) such that φ1 = φ2 and define
V = {φ ∈ D(Ω ) | φ 0 < φ1 − φ2 0 }. (7.11)

7.1 Basic Definitions and Results 169
Thus V ∈ σ̂ and φ1 ∈ φ2 + V . As φ2 + V is open and also is contained in

D(Ω ) \ {φ1 } and φ2 = φ1 is arbitrary, it follows that D(Ω ) \ {φ1 } is open, so
that {φ1 } is closed.
(b) The proof that addition is σ -continuous follows from the convexity of any
element of σ̂ . Thus given φ1 , φ2 ∈ D(Ω ) and V ∈ σ̂ we have
1 1
φ1 + V + φ2 + V = φ1 + φ2 + V . (7.12)
2 2
(c) To prove the continuity of scalar multiplication, first consider φ0 ∈ D(Ω ) and
α0 ∈ R. Then,
αφ − α0 φ0 = α (φ − φ0 ) + (α − α0 )φ0 . (7.13)
For V ∈ σ̂ there exists δ > 0 such that δ φ0 ∈ 12 V . Let us define

c = 12 (|α0 | + δ ). Thus if |α − α0 | < δ then (α − α0 )φ0 ∈ 12 V . Let φ ∈ D(Ω )
such that
1
φ − φ0 ∈ cV = V, (7.14)
2(|α0 | + δ )
so that
1
(|α0 | + δ )(φ − φ0 ) ∈ V . (7.15)
2
This means
1 1
α (φ − φ0 ) + (α − α0 )φ0 ∈ V + V = V . (7.16)
2 2
Therefore αφ − α0 φ0 ∈ V whenever |α − α0 | < δ and φ − φ0 ∈ cV .
For the next result the proof may be found in Rudin [58].
Proposition 7.1.4. A convex balanced set V ⊂ D(Ω ) is open if and only if V ∈ σ .
Proposition 7.1.5. The topology σK of DK ⊂ D(Ω ) coincides with the topology that
DK inherits from D(Ω ).
Proof. From Proposition 7.1.4 we have
V ∈ σ implies DK ∩ V ∈ σK . (7.17)
Now suppose V ∈ σK , we must show that there exists A ∈ σ such that V = A ∩ DK .

The definition of σK implies that for every φ ∈ V , there exist N ∈ N and δφ > 0
such that
{ϕ ∈ DK | ϕ − φ N < δφ } ⊂ V . (7.18)
170 7 Distributions
Define
Uφ = {ϕ ∈ D(Ω ) | ϕ N < δφ }. (7.19)
Then Uφ ∈ σ̂ and
DK ∩ (φ + Uφ ) = φ + (DK ∩ Uφ ) ⊂ V . (7.20)
Defining A = ∪φ ∈V (φ + Uφ ), we have completed the proof.

The proof for the next two results may also be found in Rudin [58].
Proposition 7.1.6. If A is a bounded set of D(Ω ), then A ⊂ DK for some K ⊂ Ω ,
and there are MN < ∞ such that φ N ≤ MN , ∀φ ∈ A, N ∈ N.
Proposition 7.1.7. If {φn } is a Cauchy sequence in D(Ω ), then {φn } ⊂ DK for some
K ⊂ Ω compact, and
lim φi − φ j N = 0, ∀N ∈ N. (7.21)
i, j→∞
Proposition 7.1.8. If φn → 0 in D(Ω ), then there exists a compact K ⊂ Ω which

contains the support of φn , ∀n ∈ N and Dα φn → 0 uniformly, for each multi-index α .
The proof follows directly from the last proposition.
Theorem 7.1.9. Suppose T : D(Ω ) → 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 φn → θ in D(Ω ), then T (φn ) → θ 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 φn → 0 in D(Ω ), by the last proposition φn → 0
in some DK so that {φn } is bounded and {T (φn )} is also bounded. Hence, by
Proposition 1.9.3, T (φn ) → 0 in V .
• 3 ⇒ 4. Assume 3 holds and consider {φn } ⊂ DK . If φn → θ , then by Proposi-
tion 7.1.5, φn → θ in D(Ω ), so that by above, T (φn ) → θ 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 7.1.5,
U is open in D(Ω ) if and only if DK ∩ U is open in DK for each compact
K ⊂ Ω ; thus, if the restrictions of T to each DK are continuous at θ , then T is
continuous at θ ; hence, 4 implies 1.
Definition 7.1.10 (Distribution). A linear functional in D(Ω ) which is continuous
with respect to σ is said to be a distribution.
7.2 Differentiation of Distributions 171
Proposition 7.1.11. Every differential operator is a continuous mapping from

D(Ω ) into D(Ω ).
Proof. Since Dα φ N ≤ φ |α |+N , ∀N ∈ N, Dα is continuous on each DK , so that

by Theorem 7.1.9, Dα is continuous on D(Ω ).
Theorem 7.1.12. Denoting by D (Ω ) the dual space of D(Ω ) we have that T :

D(Ω ) → R ∈ D (Ω ) if and only if for each compact set K ⊂ Ω there exists an
N ∈ N and c ∈ R+ such that
|T (φ )| ≤ c φ N , ∀φ ∈ DK . (7.22)
Proof. The proof follows from the equivalence of 1 and 4 in Theorem 7.1.9.
7.2 Differentiation of Distributions
Definition 7.2.1 (Derivatives for Distributions). Given T ∈ D (Ω ) and a multi-

index α , we define the Dα derivative of T as
Dα T (φ ) = (−1)|α | T (Dα φ ), ∀φ ∈ D(Ω ). (7.23)
Remark 7.2.2. Observe that if |T (φ )| ≤ c φ N , ∀φ ∈ D(Ω ) for some c ∈ R+ , then
|Dα T (φ )| ≤ c Dα φ N ≤ c φ N+|α | , ∀φ ∈ D(Ω ), (7.24)
thus Dα T ∈ D (Ω ). Therefore, derivatives of distributions are also distributions.
Theorem 7.2.3. Suppose {Tn } ⊂ D (Ω ). Let T : D(Ω ) → R be defined by
T (φ ) = lim Tn (φ ), ∀φ ∈ D(Ω ). (7.25)

n→∞
Then T ∈ D (Ω ), and
Dα Tn → Dα T in D (Ω ). (7.26)
Proof. Let K be an arbitrary compact subset of Ω . Since (7.25) holds for every
φ ∈ DK , the principle of uniform boundedness implies that the restriction of T to
DK is continuous. It follows from Theorem 7.1.9 that T is continuous in D(Ω ), that
is, T ∈ D (Ω ). On the other hand
(Dα T )(φ ) = (−1)|α | T (Dα φ ) = (−1)|α | lim Tn (Dα φ )

n→∞
= lim (Dα Tn (φ )), ∀φ ∈ D(Ω ). (7.27)
n→∞
172 7 Distributions
7.3 Examples of Distributions
7.3.1 First Example
Let Ω ⊂ Rn be an open bounded set. As a first example of distribution consider

the functional
T : D(Ω ) → R
given by
T (φ ) = f φ dx,
Ω
where f ∈ L1 (Ω ). Observe that

|T (φ )| ≤ | f φ | dx
Ω
≤ | f | dx φ ∞ , (7.28)
Ω
so that T is a bounded linear functional on D(Ω ), that is, T is a distribution.
7.3.2 Second Example
For the second example, define Ω = (0, 1) and T : D(Ω ) → R by

T (φ ) = φ (1/2) + φ (1/3).
Thus,
|T (φ )| = |φ (1/2) + φ (1/3)| ≤ φ ∞ + φ ∞ ≤ 2 φ 1 ,
so that T is also a distribution (bounded and linear).
7.3.3 Third Example
For the third example, consider an open bounded Ω ⊂ Rn and T : D(Ω ) → R by

T (φ ) = f φ dx,
Ω
where f ∈ L1 (Ω ).
Observe that the derivative of T for the multi-index α = (α1 , . . . , αn ) is defined by

Dα T (φ ) = (−1)|α | T (Dα φ ) = (−1)|α | f Dα φ dx.
Ω
7.3 Examples of Distributions 173
If there exists g ∈ L1 (Ω ), such that

(−1)|α | f Dα φ dx = gφ dx, ∀φ ∈ D(Ω ),
Ω Ω
we say that g is the derivative Dα of f in the distributional sense.

For example, for Ω = (0, 1) and f : Ω → R given by

0, if x ∈ [0, 1/2],
f (x) =
1, if x ∈ (1/2, 1],
and

T (φ ) = f φ dx,
Ω
where φ ∈ Cc∞ (Ω ) ,we have

dφ
Dx T (φ ) = − f dx
Ω dx
1
dφ
=− (1) dx
1/2 dx
= −φ (1) + φ (1/2) = φ (1/2), (7.29)
that is,
Dx T (φ ) = φ (1/2), ∀φ ∈ Cc∞ (Ω ).
Finally, defining f : Ω → R by

x, if x ∈ [0, 1/2],
f (x) =
−x + 1, if x ∈ (1/2, 1],
and

T (φ ) = f φ dx,
Ω
where φ ∈ Cc∞ (Ω ) we have

dφ
Dx T (φ ) = − f dx
Ω dx
1
dφ
=− f dx
0 dx
1
= gφ dx, (7.30)
0
where

1, if x ∈ [0, 1/2],
g(x) =
−1, if x ∈ (1/2, 1].
174 7 Distributions
In such a case we denote g = Dx f and say that g is the derivative of f in the

distributional sense.
We emphasize that in this last example the classical derivative of f is not defined,
since f is not differentiable ant x = 1/2.
Chapter 8
The Lebesgue and Sobolev Spaces
Here, we emphasize that the two main references for this chapter are Adams [2]
and Evans [26]. We start with the definition of Lebesgue spaces, denoted by L p (Ω ),
where 1 ≤ p ≤ ∞ and Ω ⊂ Rn is an open set. In this chapter, integrals always refer
to the Lebesgue measure.
8.1 Definition and Properties of L p Spaces
Definition 8.1.1 (L p Spaces). For 1 ≤ p < ∞, we say that u ∈ L p (Ω ) if u : Ω → R

is measurable and

|u| p dx < ∞. (8.1)
Ω

We also denote u p = [ Ω |u| p dx]1/p and will show that · p is a norm.
Definition 8.1.2 (L∞ Spaces). We say that u ∈ L∞ (Ω ) if u is measurable and there

exists M ∈ R+ , such that |u(x)| ≤ M, a.e. in Ω . We define
u ∞ = inf{M > 0 | |u(x)| ≤ M, a.e. in Ω }. (8.2)
We will show that · ∞ is a norm. For 1 ≤ p ≤ ∞, we define q by the relations

⎧
⎨ +∞, if p = 1,
p
q = p−1 , if 1 < p < +∞,
⎩
1, 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.
176 8 The Lebesgue and Sobolev Spaces
Theorem 8.1.3 (Hölder Inequality). Consider u ∈ L p (Ω ) and v ∈ Lq (Ω ), with 1 ≤

p ≤ ∞. Then uv ∈ L1 (Ω ) and

|uv|dx ≤ u p v q . (8.3)
Ω
Proof. The result is clear if p = 1 or p = ∞. You may assume u p, v q > 0;

otherwise the result is also obvious. Thus suppose 1 < p < ∞. From the concav-
ity of log function on (0, ∞) we obtain

1 p 1 q 1 1
log a + b ≥ log a p + log bq = log(ab). (8.4)
p q p q
Thus,
1 p 1
ab ≤ (a ) + (bq ), ∀a ≥ 0, b ≥ 0. (8.5)
p q
Therefore
1 1
|u(x)||v(x)| ≤ |u(x)| p + |v(x)|q , a.e. in Ω . (8.6)
p q
Hence |uv| ∈ L1 (Ω ) and

1 1
|uv|dx ≤ u pp + v qq . (8.7)
Ω p q
Replacing u by λ u in (8.7) λ > 0, we obtain

λ p−1 1
|uv|dx ≤ u pp + v qq . (8.8)
Ω p λq
For λ = u −1
q/p
p v q we obtain the Hölder inequality.
The next step is to prove that · p is a norm.
Theorem 8.1.4. L p (Ω ) is a vector space and · p is norm ∀p such that 1 ≤ p ≤ ∞.
Proof. The only nontrivial 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 ∈ L p (Ω ) we have
|u(x) + v(x)| p ≤ (|u(x)| + |v(x)|) p ≤ 2 p (|u(x)| p + |v(x)| p ), (8.9)
so that u + v ∈ L p (Ω ). On the other hand

u + v pp = |u + v| p−1|u + v|dx
Ω

≤ |u + v| p−1
|u|dx + |u + v| p−1|v|dx, (8.10)
Ω Ω
8.1 Definition and Properties of L p Spaces 177
and hence, from the Hölder inequality,
u + v pp ≤ u + v pp−1 u p + u + v pp−1 v p , (8.11)
that is,
u + v p ≤ u p + v p, ∀u, v ∈ L p (Ω ). (8.12)
Theorem 8.1.5. L p (Ω ) is a Banach space for any p such that 1 ≤ p ≤ ∞.
Proof. Suppose p = ∞. Suppose {un} is Cauchy sequence in L∞ (Ω ). Thus, given

k ∈ N, there exists Nk ∈ N such that if m, n ≥ Nk , then
1
um − un ∞ < . (8.13)
k
Therefore, for each k, there exist a set Ek such that m(Ek ) = 0, and
1
|um (x) − un (x)| < , ∀x ∈ Ω \ Ek , ∀m, n ≥ Nk . (8.14)
k
Observe that E = ∪∞ k=1 Ek is such that m(E) = 0. Thus {un (x)} is a real Cauchy
sequence at each x ∈ Ω \ E. Define u(x) = lim un (x) on Ω \ E. Letting m → ∞
n→∞
in (8.14) we obtain
1
|u(x) − un(x)| < , ∀x ∈ Ω \ E, ∀n ≥ Nk . (8.15)
k
Thus u ∈ L∞ (Ω ) and un − u ∞ → 0 as n → ∞.
Now suppose 1 ≤ p < ∞. Let {un } be a Cauchy sequence in L p (Ω ). We can
extract a subsequence {unk } such that
1
unk+1 − unk p ≤ , ∀k ∈ N. (8.16)
2k
To simplify the notation we write uk in place of unk , so that
1
uk+1 − uk p ≤ , ∀k ∈ N. (8.17)
2k
Defining
n
gn (x) = ∑ |uk+1 (x) − uk (x)|, (8.18)
k=1
we obtain
gn p ≤ 1, ∀n ∈ N. (8.19)
From the monotone convergence theorem and (8.19), gn (x) converges to a limit g(x)
with g ∈ L p (Ω ). On the other hand, for m ≥ n ≥ 2, we have
|um (x) − un (x)| ≤ |um (x) − um−1(x)| + . . . + |un+1(x) − un (x)|

≤ g(x) − gn−1(x), a.e. in Ω . (8.20)
Hence {un (x)} is Cauchy a.e. in Ω and converges to a limit u(x) so that
|u(x) − un(x)| ≤ g(x), a.e. in Ω , for n ≥ 2, (8.21)
which means u ∈ L p (Ω ). Finally from |un (x) − u(x)| → 0, a.e. in Ω , |un (x) −
u(x)| p ≤ |g(x)| p , and the Lebesgue dominated convergence theorem we get
un − u p → 0 as n → ∞. (8.22)
Theorem 8.1.6. Let {un } ⊂ L p (Ω ) and u ∈ L p (Ω ) such that un − u p → 0. Then

there exists a subsequence {unk } such that
1. unk (x) → u(x), a.e. in Ω ,
2. |unk (x)| ≤ h(x), a.e. in Ω , ∀k ∈ N, for some h ∈ L p (Ω ).
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 Ω . To complete the
proof, just take h = u + g, where g is defined in the proof of last theorem.
Theorem 8.1.7. L p (Ω ) is reflexive for all p such that 1 < p < ∞.
Proof. We divide the proof into 3 parts.
1. For 2 ≤ p < ∞ we have that

u + v p u − v p 1
+ ≤ ( u Lpp (Ω ) + v Lpp (Ω ) ), ∀u, v ∈ L p (Ω ).
2 p 2 p 2
L (Ω ) L (Ω )
(8.23)
Proof. Observe that
α p + β p ≤ (α 2 + β 2 ) p/2 , ∀α , β ≥ 0. (8.24)

Now taking α = a+b
2
and β = a−b in (8.24), we obtain (using the convexity
2
of t p/2 )
p/2 2 p/2
a + b p a − b p a + b 2 a − b 2 a b2
+ ≤ + = +
2 2 2 2 2 2
1 1
≤ |a| p + |b| p . (8.25)
2 2
The inequality (8.23) follows immediately.
2. L p (Ω ) is uniformly convex and therefore reflexive for 2 ≤ p < ∞.

Proof. Suppose given ε > 0 and suppose that
u p ≤ 1, v p ≤ 1 and u − v p > ε . (8.26)
From part 1, we obtain
p
u + v p
< 1− ε , (8.27)
2 2
p
and therefore

u + v

2 < 1−δ, (8.28)
p
for δ = 1 − (1 − (ε /2) p )1/p > 0. Thus L p (Ω ) is uniformly convex and from

Theorem 2.7.2 it is reflexive.
3. L p (Ω ) is reflexive for 1 < p ≤ 2. Let 1 < p ≤ 2; from 2 we can conclude that Lq
is reflexive. We will define T : L p (Ω ) → (Lq )∗ by

Tu, f Lq (Ω ) = u f dx, ∀u ∈ L p (Ω ), f ∈ Lq (Ω ). (8.29)
Ω
From the Hölder inequality, we obtain
|Tu, f Lq (Ω ) | ≤ u p f q , (8.30)
so that
Tu (Lq (Ω ))∗ ≤ u p. (8.31)
Pick u ∈ L p (Ω ) and define f0 (x) = |u(x)| p−2 u(x) ( f0 (x) = 0 if u(x) = 0). Thus,
we have that f0 ∈ Lq (Ω ), f0 q = u pp−1, and Tu, f0 Lq (Ω ) = u pp. Therefore,
Tu, f0 Lq (Ω )
Tu (Lq (Ω ))∗ ≥ = u p. (8.32)
f 0 q
Hence from (8.31) and (8.32) we have
Tu (Lq (Ω ))∗ = u p, ∀u ∈ L p (Ω ). (8.33)
Thus T is an isometry from L p (Ω ) to a closed subspace of (Lq (Ω ))∗ . Since

from the first part Lq (Ω ) is reflexive, we have that (Lq (Ω ))∗ is reflexive. Hence
T (L p (Ω )) and L p (Ω ) are reflexive.
Theorem 8.1.8 (Riesz Representation Theorem). Let 1 < p < ∞ and let f be a
continuous linear functional on L p (Ω ). Then there exists a unique u0 ∈ Lq such that

f (v) = vu0 dx, ∀v ∈ L p (Ω ). (8.34)
Ω
Furthermore
f (L p )∗ = u0 q . (8.35)
Proof. First we define the operator T : Lq (Ω ) → (L p (Ω ))∗ by

Tu, vL p (Ω ) = uv dx, ∀v ∈ L p (Ω ). (8.36)
Ω
Similarly to last theorem, we obtain
Tu (L p(Ω ))∗ = u q. (8.37)
We have to show that T is onto. Define E = T (Lq (Ω )). As E is a closed subspace, it

suffices to show that E is dense in (L p (Ω ))∗ . Suppose h ∈ (L p )∗∗ = L p is such that
Tu, hL p (Ω ) = 0, ∀u ∈ Lq (Ω ). (8.38)
Choosing u = |h| p−2 h 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.
p
Definition 8.1.9. Let 1 ≤ p ≤ ∞. We say that u ∈ Lloc (Ω ) if u χK ∈ L p (Ω ) for all
compact K ⊂ Ω .
8.1.1 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 Ω ⊂ Rn and for any nonnegative integer m we define by Cm (Ω ) the set of
all functions u which the partial derivatives Dα u are continuous on Ω for any α such
that |α | ≤ m, where if Dα = Dα1 1 Dα2 2 . . . Dαn n , we have |α | = α1 + . . . + αn . We define
C∞ (Ω ) = ∩∞ m=0C (Ω ) and denote C (Ω ) = C(Ω ). Given a function φ : Ω → R, its
m 0
support, denoted by spt(φ ), is given by
spt(φ ) = {x ∈ Ω | φ (x) = 0}.
Cc∞ (Ω ) denotes the set of functions in C∞ (Ω ) with compact support contained in Ω .

The sets C0 (Ω ) and C0∞ (Ω ) consist of the closure of Cc (Ω ) (which is the set of
functions in C(Ω ) with compact support in Ω ) and Cc∞ (Ω ), respectively, relating the
uniform convergence norm. On the other hand, CBm (Ω ) denotes the set of functions
u ∈ Cm (Ω ) for which Dα u is bounded on Ω for 0 ≤ |α | ≤ m. Observe that CBm (Ω )
is a Banach space with the norm denoted by · B,m given by
u B,m = max sup {|Dα u(x)|} .

0≤|α |≤m x∈Ω
Also, we define Cm (Ω̄ ) as the set of functions u ∈ Cm (Ω ) for which Dα u is bounded

and uniformly continuous on Ω for 0 ≤ |α | ≤ m. Observe that Cm (Ω̄ ) is a closed
subspace of CBm (Ω ) and is also a Banach space with the norm inherited from CBm (Ω ).
An important space is one of the Hölder continuous functions.
Definition 8.1.10 (Spaces of the Hölder Continuous Functions). If 0 < λ < 1, for
a nonnegative integer m, we define the space of the Hölder continuous functions
denoted by Cm,λ (Ω̄ ), as the subspace of Cm (Ω̄ ) consisting of those functions u for
which, for 0 ≤ |α | ≤ m, there exists a constant K such that
|Dα u(x) − Dα u(y)| ≤ K|x − y|λ , ∀x, y ∈ Ω .
Cm,λ (Ω̄ ) is a Banach space with the norm denoted by · m,λ given by
α
|D u(x) − Dα u(y)|
u m,λ = u B,m + max sup , x = y .
0≤|α |≤m x,y∈Ω |x − y|λ
From now on we say that f : Ω → R is locally integrable, if it is Lebesgue integrable

on any compact K ⊂ Ω . Furthermore, we say that f ∈ Lloc
p
(Ω ) if f ∈ L p (K) for any
compact K ⊂ Ω . Finally, given an open Ω ⊂ R , we denote W ⊂⊂ Ω whenever W
n
is compact and W ⊂ Ω .
Theorem 8.1.11. The space C0 (Ω ) is dense in L p (Ω ), for 1 ≤ p < ∞.
Proof. For the proof we need the following lemma:

Lemma 8.1.12. Let f ∈ L1loc (Ω ) such that

f u dx = 0, ∀u ∈ C0 (Ω ). (8.39)
Ω
Then f = 0 a.e. in Ω .
First suppose f ∈ L1 (Ω ) and Ω bounded, so that m(Ω ) < ∞. Given ε > 0, since
C0 (Ω ) is dense in L1 (Ω ), there exists f1 ∈ C0 (Ω ) such that f − f1 1 < ε and thus,
from (8.39), we obtain

f1 u dx ≤ ε u ∞ , ∀u ∈ C0 (Ω ). (8.40)
Ω

Defining
K1 = {x ∈ Ω | f1 (x) ≥ ε } (8.41)
and
K2 = {x ∈ Ω | f1 (x) ≤ −ε }, (8.42)
as K1 and K2 are disjoint compact sets, by the Urysohn theorem, there exists u0 ∈
C0 (Ω ) such that

+1, if x ∈ K1 ,
u0 (x) = (8.43)
−1, if x ∈ K2
and
|u0 (x)| ≤ 1, ∀x ∈ Ω . (8.44)
Also defining K = K1 ∪ K2 , we may write

f1 u0 dx = f1 u0 dx + f1 u0 dx. (8.45)
Ω Ω −K K
Observe that, from (8.40),

| f1 | dx ≤ | f1 u0 | dx ≤ ε (8.46)
K Ω
so that

| f1 | dx = | f1 | dx + | f1 | dx ≤ ε + ε m(Ω ). (8.47)
Ω K Ω −K
Hence
f 1 ≤ f − f1 1 + f1 1 ≤ 2ε + ε m(Ω ). (8.48)
Since ε > 0 is arbitrary, we have that f = 0 a.e. in Ω . Finally, if m(Ω ) = ∞, define
Ωn = {x ∈ Ω | dist(x, Ω c ) > 1/n and |x| < n}. (8.49)
It is clear that Ω = ∪∞ n=1 Ω n and from above f = 0 a.e. on Ω n , ∀n ∈ N, so that f = 0

a.e. in Ω .
Finally, to finish the proof of Theorem 8.1.11, suppose h ∈ Lq (Ω ) is such that

hu dx = 0, ∀u ∈ C0 (Ω ). (8.50)
Ω

Observe that h ∈ L1loc (Ω ) since K |h| dx ≤ h qm(K)1/p < ∞. From last lemma
h = 0 a.e. in Ω , which by Corollary 2.2.13 completes the proof.
Theorem 8.1.13. L p (Ω ) is separable for any 1 ≤ p < ∞.
Proof. The result follows from last theorem and from the fact that C0 (K) is sep-
arable for each K ⊂ Ω compact [from the Weierstrass theorem, polynomials with
rational coefficients are dense C0 (K)]. Observe that Ω = ∪∞
n=1 Ω n , Ω n defined as in
(8.49), where Ω̄n is compact, ∀n ∈ N.
8.2 The Sobolev Spaces 183
8.2 The Sobolev Spaces
Now we define the Sobolev spaces, denoted by W m,p (Ω ).

Definition 8.2.1 (Sobolev Spaces). We say that u ∈ W m,p (Ω ) if u ∈ L p (Ω ) and
Dα u ∈ L p (Ω ), for all α such that 0 ≤ |α | ≤ m, where the derivatives are understood
in the distributional sense.
Definition 8.2.2. We define the norm · m,p for W m,p (Ω ), where m ∈ N and 1 ≤
p ≤ ∞, as

1/p
u m,p = ∑ Dα u pp , if 1 ≤ p < ∞, (8.51)
0≤|α |≤m
and
u m,∞ = max Dα u ∞ . (8.52)

0≤|α |≤m
Theorem 8.2.3. W m,p (Ω ) is a Banach space.
Proof. Consider {un } a Cauchy sequence in W m,p (Ω ). Then {Dα un } is a Cauchy

sequence for each 0 ≤ |α | ≤ m. Since L p (Ω ) is complete there exist functions u and
uα , for 0 ≤ |α | ≤ m, in L p (Ω ) such that un → u and Dα un → uα in L p (Ω ) as n → ∞.
From above L p (Ω ) ⊂ L1loc (Ω ) and so un determines a distribution Tun ∈ D (Ω ). For
any φ ∈ D(Ω ) we have, by the Hölder inequality,

|Tun (φ ) − Tu (φ )| ≤ |un (x) − u(x)||φ (x)|dx ≤ φ q un − u p. (8.53)
Ω
Hence Tun (φ ) → Tu (φ ) for every φ ∈ D(Ω ) as n → ∞. Similarly TDα un (φ ) → Tuα (φ )

for every φ ∈ D(Ω ). We have that
Tuα (φ ) = lim TDα un (φ ) = lim (−1)|α | Tun (Dα φ )

n→∞ n→∞
= (−1)|α | Tu (Dα φ ) = TDα u (φ ), (8.54)
for every φ ∈ D(Ω ). Thus uα = Dα u in the sense of distributions, for 0 ≤ |α | ≤ m,

and u ∈ W m,p (Ω ). As lim u − un m,p = 0, W m,p (Ω ) is complete.
n→∞
Remark 8.2.4. Observe that distributional and classical derivatives coincide when
the latter exist and are continuous. We define S ⊂ W m,p (Ω ) by
S = {φ ∈ Cm (Ω ) | φ m,p < ∞}. (8.55)
Thus, the completion of S concerning the norm · m,p is denoted by H m,p (Ω ).

Corollary 8.2.5. H m,p (Ω ) ⊂ W m,p (Ω ).

Proof. Since W m,p (Ω ) is complete we have that H m,p (Ω ) ⊂ W m,p (Ω ).
Theorem 8.2.6. W m,p (Ω ) is separable if 1 ≤ p < ∞ and is reflexive and uniformly
convex if 1 < p < ∞. Particularly, W m,2 (Ω ) is a separable Hilbert space with the
inner product
(u, v)m = ∑ Dα u, Dα vL2 (Ω ) . (8.56)

0≤|α |≤m
Proof. We can see W m,p (Ω ) as a subspace of L p (Ω , RN ), where N = ∑0≤|α |≤m 1.

From the relevant properties for L p (Ω ), we have that L p (Ω ; RN ) is a reflexive and
uniformly convex for 1 < p < ∞ and separable for 1 ≤ p < ∞. Given u ∈ W m,p (Ω ),
we may associate the vector Pu ∈ L p (Ω ; RN ) defined by
Pu = {Dα u}0≤|α |≤m . (8.57)
Since Pu pN = u m,p, we have that W m,p is closed subspace of L p (Ω ; RN ). Thus,

from Theorem 1.21 in Adams [1], we have that W m,p (Ω ) is separable if 1 ≤ p < ∞
and reflexive and uniformly convex if 1 < p < ∞.
Lemma 8.2.7. Let 1 ≤ p < ∞ and define U = L p (Ω ; RN ). For every continuous
linear functional f on U, there exists a unique v ∈ Lq (Ω ; RN ) = U ∗ such that
N
f (u) = ∑ ui , vi , ∀u ∈ U. (8.58)
i=1
Moreover,
f U ∗ = v qN , (8.59)
where · qN = · Lq (Ω ,RN ) .
Proof. For u = (u1 , . . . , un ) ∈ L p (Ω ; RN ), we may write

f (u) = f ((u1 , 0, . . . , 0)) + . . . + f ((0, . . . , 0, u j , 0, . . . , 0))
+ . . . + f ((0, . . . , 0, un )), (8.60)
and since f ((0, . . . , 0, u j , 0, . . . , 0)) is continuous linear functional on u j ∈
L p (Ω ), there exists a unique v j ∈ Lq (Ω ) such that f (0, . . . , 0, u j , 0, . . . , 0) =
u j , v j L2 (Ω ) , ∀u j ∈ L p (Ω ), ∀ 1 ≤ j ≤ N, so that
N
f (u) = ∑ ui , vi , ∀u ∈ U. (8.61)
i=1
From the Hölder inequality we obtain

N
| f (u)| ≤ ∑ u j p v j q ≤ u pN v qN , (8.62)
j=1
8.2 The Sobolev Spaces 185
and hence f U ∗ ≤ v qN . The equality in (8.62) is achieved for u ∈ L p (Ω , RN ),

1 0, there is a mea-

surable set A such that m(A) > 0 and |vk (x)| ≥ vk ∞ − ε , ∀x ∈ A. Defining u(x)
as

v̄k /vk , if i = k, x ∈ A and vk (x) = 0
ui (x) = (8.64)
0, otherwise,
we have

f (uk ) = u, vk L2 (Ω ) = |vk |dx ≥ ( (vk ∞ − ε ) uk 1
A
= ( v ∞N − ε ) u 1N . (8.65)
Since ε is arbitrary, the proof is complete.
Theorem 8.2.8. Let 1 ≤ p < ∞. Given a continuous linear functional f on W m,p (Ω ),

there exists v ∈ Lq (Ω , RN ) such that
f (u) = ∑ Dα u, vα L2 (Ω ) . (8.66)

0≤|α |≤m
Proof. Consider f a continuous linear operator on U = W m,p (Ω ). By the Hahn–

Banach theorem, we can extend f to f˜, on L p (Ω ; RN ), so that f˜ qN = f U ∗ and
by the last theorem there exists {vα } ∈ Lq (Ω ; RN ) such that
f˜(û) = ∑ ûα , vα L2 (Ω ) , ∀v ∈ L p (Ω ; RN ). (8.67)

0≤|α |≤m
In particular for u ∈ W m,p (Ω ), defining û = {Dα u} ∈ L p (Ω ; RN ), we obtain
f (u) = f˜(û) = ∑ Dα u, vα L2 (Ω ) . (8.68)

1≤|α |≤m
Finally, observe that, also from the Hahn–Banach theorem, f U ∗ = f˜ qN = v qN .
Definition 8.2.9. Let Ω ⊂ Rn be a domain. For m a positive integer and 1 ≤ p < ∞we
define W0m,p (Ω ) as the closure in · m,p of Cc∞ (Ω ), where we recall that Cc∞ (Ω )
denotes the set of C∞ (Ω ) functions with compact support contained in Ω . Finally,
we also recall that the support of φ : Ω → R, denoted by spt(φ ), is given by
spt(φ ) = {x ∈ Ω |φ (x) = 0}.

8.3 The Sobolev Imbedding Theorem
8.3.1 The Statement of the 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
u Y ≤ K u X , ∀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
· Y . At this point, we first introduce the following definition.
Definition 8.3.1. Let Ω ⊂ Rn be an open bounded set. We say that ∂ Ω is Ĉ1 if for
each x0 ∈ ∂ Ω , denoting x̂ = (x1 , . . . , xn−1 ) for a local coordinate system, there exist
r > 0 and a function f (x1 , . . . , xn−1 ) = f (x̂) such that
W = Ω ∩ Br (x0 ) = {x ∈ Br (x0 ) | xn ≥ f (x1 , . . . , xn−1 )}.
Moreover, f (x̂) is a Lipschitz continuous function, so that

| f (x̂) − f (ŷ)| ≤ C1 |x̂ − ŷ|2 , on its domain,
for some C1 > 0. Finally, we assume
n−1
∂ f (x̂)
∂ xk k=1
is classically defined, almost everywhere also on its concerning domain, so that

f ∈ W 1,2 .
Theorem 8.3.2 (The Sobolev Imbedding Theorem). Let Ω be an open bounded

set in Rn such that ∂ Ω is Ĉ1 . 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 (Ω ) → CBj (Ω ). (8.69)
Moreover,
W j+m,p (Ω ) → W j,q (Ω ), for p ≤ q ≤ ∞, (8.70)
and, in particular,
W m,p (Ω ) → Lq (Ω ), for p ≤ q ≤ ∞. (8.71)

8.4 The Proof of the Sobolev Imbedding Theorem 187
(b) Case B If mp = n, then

W j+m,p (Ω ) → W j,q (Ω ), for p ≤ q < ∞, (8.72)
and, in particular,
W m,p (Ω ) → Lq (Ω ), for p ≤ q < ∞. (8.73)
(c) Case C If mp < n and p = 1, then
np
W j+m,p (Ω ) → W j,q (Ω ), for p ≤ q ≤ p∗ = , (8.74)
n − mp
and, in particular,
np
W m,p (Ω ) → Lq (Ω ), for p ≤ q ≤ p∗ = . (8.75)
n − mp
2. Part II If mp > n > (m − 1)p, then
W j+m,p → C j,λ (Ω ), for 0 < λ ≤ m − (n/p), (8.76)

and if n = (m − 1)p, then
W j+m,p → C j,λ (Ω ), for 0 < λ < 1. (8.77)

Also, if n = m − 1 and p = 1, then (8.77) holds for λ = 1 as well.
3. Part III All imbeddings in Parts A and B are valid for arbitrary domains Ω
if the W -space undergoing the imbedding is replaced with the corresponding
W0 -space.
8.4 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 8.4.1. Let Ω ⊂ Rn be an open bounded set. For each ε > 0 define
Ωε = {x ∈ Ω | dist(x, ∂ Ω ) > ε }.
Definition 8.4.2. Define η ∈ Cc∞ (Rn ) by

C exp |x|21−1 , if |x|2 < 1,
η (x) = 2
0, if |x|2 ≥ 1,
where | · |2 refers to the Euclidean norm in Rn , that is, for x = (x1 , . . . , xn ) ∈ Rn , we

have !
|x|2 = x21 + . . . + x2n .
Moreover, C > 0 is chosen so that

η dx = 1.
Rn
For each ε > 0, set

1 x
ηε (x) = η .
εn ε
The function η is said to be the fundamental mollifier. The functions ηε ∈ Cc∞ (Rn )
and satisfy

ηε dx = 1,
Rn
and spt(ηε ) ⊂ B(0, ε ).
Definition 8.4.3. If f : Ω → Rn is locally integrable, we define its mollification,

denoted by fε : Ωε → R as
fε = ηε ∗ f ,
that is,

fε (x) = ηε (x − y) f (y) dy
Ω
= ηε (y) f (x − y) dy. (8.78)
B(0,ε )
Theorem 8.4.4 (Properties of Mollifiers). The mollifiers have the following prop-
erties:
1. fε ∈ C∞ (Ωε ),
2. fε → f a.e. as ε → 0,
3. If f ∈ C(Ω ), then fε → f uniformly on compact subsets of Ω .
Proof.
1. Fix x ∈ Ωε , i ∈ {1, . . . , n} and a h small enough such that
x + hei ∈ Ωε .
Thus

fε (x + hei) − fε (x) 1 1 x + hei − y x−y
= n η −η
h ε Ω h ε ε
× f (y) dy

1 1 x + hei − y x−y
= n η −η
ε Vh ε ε
× f (y) dy, (8.79)
for an appropriate compact V ⊂⊂ Ω . As

1 x + hei − y x−y 1 ∂η x−y
η −η → ,
h ε ε ε ∂ xi ε
as h → 0, uniformly on V , we obtain

∂ fε (x) ∂ ηε (x − y)
= f (y) dy.
∂ xi Ω ∂ xi
By analogy, we may show that

Dα fε (x) = Dα ηε (x − y) f (y) dy, ∀x ∈ Ωε .
Ω
2. From the Lebesgue differentiation theorem we have

1
lim | f (y) − f (x)| dy = 0, (8.80)
r→0 |B(x, r)| B(x,r)
for almost all x ∈ Ω . Fix x ∈ Ω such that (8.80) holds. Hence,

| fε (x) − f (x)| = ηε (x − y)[ f (x) − f (y)] dy
B(x,ε )

1 x−y
≤ n η [ f (x) − f (y)] dy
ε B(x,ε ) ε

C
≤ | f (y) − f (x)| dy (8.81)
|B(x, ε )| B(x,ε )
for an appropriate constant C > 0. From (8.80), we obtain fε → f as ε → 0.

3. Assume f ∈ C(Ω ). Given V ⊂⊂ Ω choose W such that
V ⊂⊂ W ⊂⊂ Ω ,
and note that f is uniformly continuous on W . Thus the limit indicated in (8.80)
holds uniformly on V , and therefore fε → f uniformly on V .
Theorem 8.4.5. Let u ∈ L p (Ω ), where 1 ≤ p < ∞. Then
ηε ∗ u ∈ L p (Ω ),
ηε ∗ u p ≤ u p, ∀ε > 0
and
lim ηε ∗ u − u p = 0.
ε →0+
Proof. Suppose u ∈ L p (Ω ) and 1 < p < ∞. Defining q = p/(p − 1), from the Hölder
inequality, we have

|ηε ∗ u(x)| = ηε (x − y)u(y) dy
Rn

= [ηε (x − y)](1−1/p)[ηε (x − y)]1/pu(y) dy
Rn
1/q 1/p
≤ ηε (x − y) dy ηε (x − y)|u(y)| dy
p
Rn Rn
1/p
= ηε (x − y)|u(y)| p dy . (8.82)
Rn
From this and the Fubini theorem, we obtain

|ηε ∗ u(x)| dx ≤
p
ηε (x − y)|u(y)| p dy dx
Ω Rn Rn

= |u(y)| p ηε (x − y) dx dy
Rn Rn
= u pp. (8.83)
Suppose given ρ > 0. As C0 (Ω ) is dense in L p (Ω ), there exists φ ∈ C0 (Ω ) such that
u − φ p < ρ /3.
From the fact that

ηε ∗ φ → φ
as ε → 0, uniformly in Ω we have that there exists δ > 0 such that
ηε ∗ φ − φ p < ρ /3
if ε < δ . Thus, for any ε < δ (ρ ), we get
ηε ∗ u − u p = ηε ∗ u − ηε ∗ φ + ηε ∗ φ − φ + φ − u p
≤ ηε ∗ u − ηε ∗ φ p + ηε ∗ φ − φ p + φ − u p
≤ ρ /3 + ρ /3 + ρ /3 = ρ . (8.84)
Since ρ > 0 is arbitrary, the proof is complete.
For the next theorem we denote

u(x), if x ∈ Ω ,
ũ(x) =
0, if x ∈ Rn \ Ω .
8.4.1 Relatively Compact Sets in L p (Ω )
Theorem 8.4.6. Consider 1 ≤ p < ∞. A bounded set K ⊂ L p (Ω ) is relatively com-

pact if and only if for each ε > 0, there exist δ > 0 and G ⊂⊂ Ω (we recall that
G ⊂⊂ Ω means that G is compact and G ⊂ Ω ) such that for each u ∈ K and h ∈ Rn
such that |h| < δ we have
1.

|ũ(x + h) − ũ(x)| p dx < ε p , (8.85)
Ω
2.

|u(x)| p dx < ε p . (8.86)
Ω −G
Proof. Suppose K is relatively compact in L p (Ω ). Suppose given ε > 0. As K is

compact we may find a finite ε /6-net for K. Denote such a ε /6-net by N where
N = {v1 , . . . , vm } ⊂ L p (Ω ).
Since Cc (Ω ) is dense in L p (Ω ), for each k ∈ {1, . . . , m}, there exists φk ∈ Cc (Ω )

such that
ε
φk − vk p < .
6
Thus defining
S = {φ1 , . . . , φm },
given u ∈ K, we may select vk ∈ N such that
ε
u − vk p < ,
6
so that
φk − u p ≤ φk − vk p + vk − u p
ε ε ε
≤ + = . (8.87)
6 6 3
Define
k=1 spt(φk ),
G = ∪m
where
spt(φk ) = {x ∈ Rn | φk (x) = 0}.
We have that
G ⊂⊂ Ω ,
where as abovementioned this means G ⊂ Ω . Observe that

ε p > u − φk pp ≥ |u(x)| p dx.
Ω −G
Since u ∈ K is arbitrary, (8.86) is proven. Since φk is continuous and spt(φk ) is

compact we have that φk is uniformly continuous, that is, for the ε given above,
there exists δ̃ > 0 such that if |h| < min{δ̃ , 1}, then
ε
|φk (x + h) − φk(x)| < , ∀x ∈ G.
3(|G| + 1)
Thus,
ε p
|φk (x + h) − φk (x)| p dx < .
Ω 3
Also observe that since
ε
u − φk p < ,
3
we have that
ε
Th u − Thφk p < ,
3
where Th u = u(x + h). Thus, if |h| < δ = min{δ̃ , 1}, we obtain
Th ũ − ũ p ≤ Th ũ − Th φk p + Thφk − φk p
+ φk − u p
ε ε ε
< + + = ε. (8.88)
3 3 3
For the converse, it suffices to consider the special case Ω = Rn , because for the
general Ω we can define K̃ = {ũ | u ∈ K}. Suppose given ε > 0 and choose G ⊂⊂ Rn
such that for all u ∈ K we have

ε
|u(x)| p dx < .
Rn −G 3
For each ρ > 0 the function ηρ ∗ u ∈ C∞ (Rn ), and in particular ηρ ∗ u ∈ C(G).
Suppose φ ∈ C0 (Rn ). Fix ρ > 0. By the Hölder inequality we have
p

|ηρ ∗ φ (x) − φ (x)| =
p ηρ (y)(φ (x − y) − φ (x)) dy
Rn
p

= (ηρ (y)) 1−1/p
(ηρ (y)) (T−y φ (x) − φ (x)) dy
1/p
Rn

≤ (ηρ (y))|T−y φ (x) − φ (x)| p dy. (8.89)
B ρ (θ )
Hence, from the Fubini theorem, we may write

|ηρ ∗ φ (x) − φ (x)| p dx
Rn

≤ (ηρ (y)) |T−y φ (x) − φ (x)| p dx dy, (8.90)
B ρ (θ ) Rn
so that we may write
ηρ ∗ φ − φ p ≤ sup { Th φ − φ p}. (8.91)

h∈Bρ (θ )
Fix u ∈ L p (Rn ). We may obtain a sequence {φk } ⊂ Cc (Rn ) such that
φk → u, in L p (Rn ).
Observe that
ηρ ∗ φk → ηρ ∗ u, in L p (Rn ),
as k → ∞. Also
Th φk → Th u, in L p (Rn ),
as k → ∞. Thus
Th φk − φk p → Th u − u p,
in particular

" #
lim sup sup { Th φk − φk ≤ sup Th u − u p .
k→∞ h∈Bρ (θ ) h∈Bρ (θ )
Therefore as
ηρ ∗ φk − φk p → ηρ ∗ u − u p,
as k → ∞, from (8.91) we get
ηρ ∗ u − u p ≤ sup { Th u − u p}.
h∈Bρ (θ )
ηρ ∗ u − u p → 0, uniformly in K as ρ → 0.
Fix ρ0 > 0 such that

ε
|ηρ0 ∗ u − u| p dx < , ∀u ∈ K.
G 3 · 2 p−1
Observe that

|ηρ0 ∗ u(x)| = ηρ0 (x − y)u(y) dy
Rn

= [ηρ0 (x − y)](1−1/p)[ηρ0 (x − y)]1/pu(y) dy
Rn
1/q 1/p
≤ ηρ0 (x − y) dy ηρ0 (x − y)|u(y)| dy
p
Rn Rn
1/p
= ηρ0 (x − y)|u(y)| p dy . (8.92)
Rn
From this, we may write

1/p
|ηρ0 ∗ u(x)| ≤ sup ηρ0 (y) u p ≤ K1 , ∀x ∈ Rn , u ∈ K,
y∈Rn
where K1 = K2 K3 , 1/p
K2 = sup ηρ0 (y) ,
y∈Rn
and K3 is any constant such that

u p < K3 , ∀u ∈ K.
Similarly
1/p
|ηρ0 ∗ u(x + h) − ηρ0 u(x)| ≤ sup ηρ0 (y) Th u − u p,
y∈Rn
and thus from (8.85) we obtain

ηρ0 ∗ u(x + h) → ηρ0 ∗ u(x), as h → 0
uniformly in Rn and for u ∈ K.
By the Arzela–Ascoli theorem
{ηρ0 ∗ u | u ∈ K}
is relatively compact in C(G), and it is totally bounded so that there exists a ε0 -net
N = {v1 , . . . , vm } where
1/p
ε
ε0 = .
3 · 2 p−1|G|
Thus for some k ∈ {1, . . . , m} we have
vk − ηρ0 ∗ u ∞ < ε0 .
Hence,

|u(x) − ṽk (x)| p dx = |u(x)| p dx + |u(x) − vk (x)| p dx
Rn Rn −G G

ε
≤ + 2 p−1 (|u(x) − (ηρ0 ∗ u)(x)| p
3 G
+|ηρ0 ∗ u(x) − vk (x)| p ) dx

ε ε ε |G|
≤ + 2 p−1 +
3 3 · 2 p−1 3 · 2 p−1|G|
= ε. (8.93)
Thus K is totally bounded and therefore it is relatively compact.

8.4.2 Some Approximation Results
Theorem 8.4.7. Let Ω ⊂ Rn be an open set. Assume u ∈ W m,p (Ω ) for some 1 ≤

p < ∞, and set
uε = ηε ∗ u in Ωε .
Then,
1. uε ∈ C∞ (Ωε ), ∀ε > 0,
m,p
2. uε → u in Wloc (Ω ), as ε → 0.
Proof. Assertion 1 has been already proved. Let us prove 2. We will show that if
|α | ≤ m, then
Dα uε = ηε ∗ Dα u, in Ωε .
For that, let x ∈ Ωε . Thus,

Dα uε (x) = Dα ηε (x − y)u(y) dy
Ω

= Dαx ηε (x − y)u(y) dy
Ω

= (−1)|α | Dαy (ηε (x − y))u(y) dy. (8.94)
Ω
Observe that for fixed x ∈ Ωε the function
φ (y) = ηε (x − y) ∈ Cc∞ (Ω ).
Therefore,

Dαy (ηε (x − y)) u(y) dy = (−1)|α | ηε (x − y)Dαy u(y) dy,
Ω Ω
and hence,

Dα uε (x) = (−1)|α |+|α | ηε (x − y)Dα u(y) dy
Ω
= (ηε ∗ Dα u)(x). (8.95)
Now choose any open bounded set such that V ⊂⊂ Ω . We have that
Dα uε → Dα u, in L p (V ) as ε → 0,
for each |α | ≤ m.
Thus,
p
uε − u m,p,V = ∑ Dα uε − Dα u p,V → 0,
|α |≤m
as ε → 0.
Theorem 8.4.8. Let Ω ⊂ Rn be a bounded open set and suppose u ∈ W m,p (Ω ) for
some 1 ≤ p < ∞. Then there exists a sequence {uk } ⊂ C∞ (Ω ) such that
uk → u in W m,p (Ω ).
Proof. Observe that

Ω = ∪∞
i=1 Ω i ,
where
Ωi = {x ∈ Ω | dist(x, ∂ Ω ) > 1/i}.
Define
Vi = Ωi+3 − Ω̄i+1,
and choose any open set V0 such that V0 ⊂⊂ Ω , so that
Ω = ∪∞
i=0Vi .
Let {ζi }∞ ∞
i=0 be a smooth partition of unit subordinate to the open sets {Vi }i=0 .
That is,
0 ≤ ζi ≤ 1, ζi ∈ Cc∞ (Vi )
∑∞ i=0 ζi = 1, on Ω .
Now suppose u ∈ W m,p (Ω ). Thus ζi u ∈ W m,p (Ω ) and spt(ζi u) ⊂ Vi ⊂ Ω . Choose

δ > 0. For each i ∈ N choose εi > 0 small enough so that
ui = ηεi ∗ (ζi u)
satisfies
δ
ui − ζi u m,p,Ω ≤ ,
2i+1
and spt(ui ) ⊂ Wi where Wi = Ωi+4 − Ω̄i ⊃ Vi . Define

∞
v = ∑ ui .
i=0
Thus such a function belongs to C∞ (Ω ), since for each open V ⊂⊂ Ω there are at
most finitely many nonzero terms in the sum. Since
∞
u = ∑ ζi u,
i=0
we have that for a fixed V ⊂⊂ Ω ,

∞
v − u m,p,V ≤ ∑ ui − ζiu m,p,V
i=0
∞
1
≤δ∑ i+1
= δ. (8.96)
i=0 2
Taking the supremum over sets V ⊂⊂ Ω we obtain
v − u m,p,Ω < δ .
Since δ > 0 is arbitrary, the proof is complete.
The next result is also relevant. For a proof see Evans [26], p. 232.
Theorem 8.4.9. Let Ω ⊂ Rn be a bounded set such that ∂ Ω is C1 . Suppose u ∈
W m,p (Ω ) where 1 ≤ p < ∞. Thus there exists a sequence {un } ⊂ C∞ (Ω ) such that
un → u in W m,p (Ω ), as n → ∞.
Anyway, now we prove a more general result.

Theorem 8.4.10. Let Ω ⊂ Rn be an open bounded set such that ∂ Ω is Ĉ1 . Let u ∈
W m,p (Ω ) where m is a nonnegative integer and 1 ≤ p < ∞.
Under such assumptions, there exists {uk } ⊂ C∞ (Ω ) such that
uk − u m,p,Ω → 0, as k → ∞.
Proof. Fix x0 ∈ ∂ Ω . Since ∂ Ω is Ĉ1 , denoting x̂ = (x1 , . . . , xn−1 ) for a local

coordinate system, there exists r > 0 and a function f (x1 , . . . , xn−1 ) = f (x̂) such that
W = Ω ∩ Br (x0 ) = {x ∈ Br (x0 ) | xn ≥ f (x1 , . . . , xn−1 )}.
We emphasize f (x̂) is a Lipschitz continuous function, so that
| f (x̂) − f (ŷ)| ≤ C1 |x̂ − ŷ|2 , on its domain,

for some C1 > 0. Furthermore

n−1
∂ f (x̂)
∂ xk k=1
is classically defined, almost everywhere also on its concerning domain.

Let ε > 0. For each δ > 0 define xδ = x +Cδ en , where C > 1 is a fixed constant.
Define uδ = u(xδ ). Now choose δ > 0 sufficiently small such that
uδ − u m,p,W < ε /2.
For each n ∈ N, x ∈ W define
vn (x) = (η1/n ∗ uδ )(x).
Observe that
vn − u m,p,W ≤ vn − uδ m,p,W + uδ − u m,p,W .
For the fixed δ > 0, there exists Nε ∈ N such that if n > Nε we have
vn − uδ m,p,W < ε /2,
and
vn ∈ C∞ (W ).
Hence
vn − u m,p,W ≤ vn − uδ m,p,W + uδ − u m,p,W < ε /2 + ε /2 = ε .
Clarifying the dependence of r on x0 ∈ ∂ Ω we denote r = rx0 . Observe that
∂ Ω ⊂ ∪x0 ∈∂ Ω Brx0 (x0 )
so that since ∂ Ω is compact, there exists x1 , . . . , xM ∈ ∂ Ω such that
∂ Ω ⊂ ∪M
i=1 Bri (xi ).
We denote Bri (xi ) = Bi and Wi = Ω ∩ Bi , ∀i ∈ {1, . . . , M}. We also choose an

appropriate open set B0 ⊂⊂ Ω such that
Ω ⊂ ∪M
i=0 Bi .
Let {ζi }M
i=0 be a concerned partition of unity relating {Bi }i=0 .
M
∞
Thus ζi ∈ Cc (Bi ) and 0 ≤ ζi ≤ 1, ∀i ∈ {0, . . . , M} and also
M
∑ ζi = 1 on Ω .
i=0
From above, we may find vi ∈ C∞ (W i ) such that vi − u m,p,Wi < ε , ∀i ∈ {1, . . ., M}.
Define u0 = v0 = u on B0 ≡ W0 ,
ui = ζi u, ∀i ∈ {0, . . . , M}
and
M
v = ∑ ζi vi .
i=0
We emphasize
v ∈ C∞ (Ω ).
Therefore

M

v − u m,p,Ω = ∑ (ζi u − ζi vi )
i=0
m,p,Ω
M
≤ C2 ∑ u − vi m,p,(Ω ∩Bi )
i=0
M
= C2 ∑ u − vi m,p,Wi
i=0
< C2 M ε . (8.97)
Since neither C2 nor M depends on ε > 0, the proof is complete.
8.4.3 Extensions
In this section we study extensions of the Sobolev spaces from a domain Ω ⊂ Rn

to Rn . First we enunciate a result found in Evans [26].
Theorem 8.4.11. Assume Ω ⊂ Rn is an open bounded set and that ∂ Ω is C1 . Let

V be a bounded open set such that Ω ⊂⊂ V . Then there exists a bounded linear
operator
E : W 1,p (Ω ) → W 1,p (Rn ),
such that for each u ∈ W 1,p (Ω ) we have:
1. Eu = u, a.e. in Ω ,
2. Eu has support in V ,
3. Eu 1,p,Rn ≤ C u 1,p,Ω , where the constant depends only on p, Ω , and V.
The next result, which we prove, is a more general one.
Theorem 8.4.12. Assume Ω ⊂ Rn is an open bounded set and that ∂ Ω is Ĉ1 . Let
V be a bounded open set such that Ω ⊂⊂ V . Then there exists a bounded linear
operator
E : W 1,p (Ω ) → W 1,p (Rn ),
such that for each u ∈ W 1,p (Ω ) we have:

1. Eu = u, a.e. in Ω ,
2. Eu has support in V ,
3. Eu 1,p,Rn ≤ C u 1,p,Ω , where the constant depends only on p, Ω , and V.
Proof. Let u ∈ W 1,p (Ω ). Fix N ∈ N and select φN ∈ C∞ (Ω ) such that

φN − u 1,p,Ω < 1/N.
Choose x0 ∈ ∂ Ω . From the hypothesis we may write
Ω ∩ Br (x0 ) = {x ∈ Br (x0 ) | xn ≥ f (x1 , . . . , xn−1 )},
for some r > 0 and so that denoting x̂ = (x1 , . . . , xn−1 ), f (x1 , . . . , xn−1 ) = f (x̂) is a
Lipschitz continuous function such that

∂ f (x̂) n−1
∂ xk k=1
is classically defined almost everywhere on its domain and
| f (x̂) − f (ŷ)| ≤ C1 |x̂ − ŷ|2 , ∀x̂, ŷ on its domain,
for some C1 > 0.

Define the variable y ∈ Rn by yi = xi , ∀i ∈ {1, . . . , n − 1}, and yn = xn −
f (x1 , . . . , xn−1 ).
Thus
φN (x1 , . . . , xn ) = φN (y1 , . . . , yn−1 , yn + f (y1 , .., yn−1 )) = φ N (y1 , . . . , yn ).
Observe that defining ψ (x) = y from the continuity of ψ −1 , there exists r1 > 0
such that
ψ −1 (B+
r1 (y0 )) ⊂ Ω ∩ Br (x0 ),
+
where y0 = (x01 , . . . , x0n−1 , 0). We define W + = ψ −1 (Br1 (y0 )) and W − = ψ −1
−
(Br1 (y0 )) where we denote
B+ = B+
r1 (y0 ) = {y ∈ Br1 (y0 ) | yn ≥ 0},
and
B− = B−
r1 (y0 ) = {y ∈ Br1 (y0 ) | yn < 0}.
We emphasize that locally about x0 we have that ∂ Ω and ψ (∂ Ω ) correspond to the

equations xn − f (x1 , . . . , xn−1 ) = 0 and yn = 0, respectively.
Moreover, φ N is Lipschitz continuous on B+ so that φ ∈ W 1,p (B+ ), and therefore
+
there exists φ̃N ∈ C∞ (B ) such that
φ̃N − φ N 1,p,B+ < 1/N.

Define φ̂N : B → R by
+
φ̃N (y) if y ∈ B
φ̂N (y) =
−3φ̃N (y1 , . . . , yn−1 , −yn ) + 4φ̃N (y1 , . . . , yn−1 , −yn /2) if y ∈ B− .
It may be easily verified that φ̂N ∈ C1 (B). Also, there exists C2 > 0 such that
φ̂N 1,p,B ≤ C2 φ̂N 1,p,B+

= C2 φ̃N 1,p,B+
≤ C2 φ N 1,p,B+ + C2 /N, (8.98)
where C2 depends only on Ω and p.

We claim that {φ̂N } is a Cauchy sequence in W 1,p (B).
For N1 , N2 ∈ N we have
φ̂N1 − φ̂N2 1,p,B ≤ C1 φ̂N1 − φ̂N2 1,p,B+

≤ C1 φ̃N1 − φ N1 + φ N1 − φ N2 + φ N2 − φ̂N2 1,p,B+
≤ C1 φ̃N1 − φ N1 1,p,B+ + C1 φ N1 − φ N2 1,p,B+
+C1 φ N2 − φ̂N2 1,p,B+
≤ C1 /N1 + C1 φ N1 − φ N2 1,p,B+ + C1 /N2
→ 0, as N1 , N2 → ∞. (8.99)
Also, since φ̂N → u(x(y)), in W 1,p (B+ ), up to a subsequence not relabeled,
φ̂N (y) → u(x(y)), a.e. in B+ .
Define û = limN→∞ φ̂N in W 1,p (B). Therefore
û(y(x)) = u(x), a.e. in W + .
Now denoting simply

û(y(x)) = u(x)
we obtain
u = u, a.e. in W + .
Now choose ε > 0. Thus there exists N0 ∈ N such that if N > N0 we have
u 1,p,W ≤ φ̂N (y(x)) 1,p,W + ε

≤ C3 φ̂N (y) 1,p,B + ε
≤ C4 φ̂N 1,p,B+ + ε
≤ C5 φ̂N 1,p,W + + ε (8.100)
so that letting N → ∞, since ε > 0 is arbitrary, we get
u 1,p,W ≤ C5 u 1,p,W + .
Now denoting W = Wx0 we have that ∂ Ω ⊂ ∪x0 ∈∂ Ω Wx0 and since ∂ Ω is compact,
there exist x1 , . . . , xM ∈ ∂ Ω , such that
∂ Ω ⊂ ∪M
i=1Wxi .
Hence for an appropriate open W0 ⊂⊂ Ω , we get
Ω ⊂ ∪M
i=0Wi .
where we have denoted Wi = Wxi , ∀i ∈ {0, . . ., M}.

Let {ζi }M
i=0 be a concerned partition of unity relating {Wi }i=0 , so that
M
M
∑ ζi = 1, in Ω ,
i=0
and ζi ∈ Cc∞ (Wi ), 0 ≤ ζi ≤ 1, ∀i ∈ {0, . . . , M}.

Define
ui = ζi u, ∀i ∈ {0, . . . , M}.
For each i we denote the extension of u from Wi+ to Wi by ui . Also define u0 = u, ∈
W0 , and u = ∑Mi=0 ζi ui .
Recalling that u = ui , a.e. on Wi+ and that Ω = ∪M +
i=1Wi ∪W0 , we obtain
u = ∑i=0 ζi ui = ∑i=0 ζi u = u, a.e. in Ω . Furthermore
M M
M
u 1,p,Rn ≤ ∑ ζi ui 1,p,Rn
i=0
M
≤ C5 ∑ ui 1,p,Wi
i=0
M
≤ C5 u 1,p,W0 + C5 ∑ ui 1,p,W +
i
i=1
≤ (M + 1)C5 u 1,p,Ω
= C u 1,p,Ω , (8.101)
where C = (M + 1)C5 .
We recall that the partition of unity may be chosen so that its support is on V .
Finally, we denote Eu = u.
8.4.4 The Main Results

np
Definition 8.4.13. For 1 ≤ p < n we define r = n−p .
Theorem 8.4.14 (Gagliardo–Nirenberg–Sobolev Inequality). Let 1 ≤ p < n. Thus

there exists a constant K > 0 depending only p and n such that
u r,Rn ≤ K Du p,Rn , ∀u ∈ Cc1 (Rn ).
Proof. Suppose p = 1. Let u ∈ Cc1 (Rn ). From the fundamental theorem of calculus
we have
xi
∂ u(x1 , . . . , xi−1 , yi , xi+1 , . . . , xn )
u(x) = dyi ,
−∞ ∂ xi
so that
∞
|u(x)| ≤ |Du(x1 , . . . , xi−1 , yi , xi+1 , . . . , xn )| dyi .
−∞
Therefore,
n ∞
1/(n−1)
|u(x)| n/(n−1)
≤∏ |Du(x1 , . . . , xi−1 , yi , xi+1 , . . . , xn )| dyi .
i=1 −∞
From this, we get

∞
|u(x)|n/(n−1) dx1
−∞
∞ n ∞ 1/(n−1)
≤
−∞ i=1
∏ −∞
|Du| dyi dx1
∞
1/(n−1)
≤ |Du| dy1
−∞
∞
1/(n−1)
n ∞
×
−∞
∏ −∞
|Du| dyi dx1 . (8.102)
i=2
From this and the generalized Hölder inequality, we obtain

∞
|u(x)|n/(n−1) dx1
−∞
∞
1/(n−1)
≤ |Du| dy1
−∞
n ∞ ∞ 1/(n−1)
×∏ |Du| dx1 dyi . (8.103)
i=2 −∞ −∞
Integrating in x2 we obtain
∞ ∞
|u(x)|n/(n−1) dx1 dx2
−∞ −∞
∞

∞ 1/(n−1)
≤ |Du| dy1
−∞ −∞
n ∞ ∞ 1/(n−1)
×∏ |Du| dx1 dyi dx2 ,
i=2 −∞ −∞
so that
∞ ∞
|u(x)|n/(n−1) dx1 dx2
−∞ −∞
∞ ∞ 1/(n−1)
≤ |Du| dy2 dx1
−∞ −∞
∞
1/(n−1)
∞
× |Du| dy1
−∞ −∞
n ∞ ∞ 1/(n−1)
×∏ |Du| dx1 dyi dx2 . (8.104)
i=3 −∞ −∞
By applying the generalized Hölder inequality we get

∞ ∞
|u(x)|n/(n−1) dx1 dx2
−∞ −∞
∞ ∞ 1/(n−1)
≤ |Du| dy2 dx1
−∞ −∞
∞ ∞ 1/(n−1)
× |Du| dy1 dx2
−∞ −∞
n ∞ ∞ ∞ 1/(n−1)
×∏ |Du| dx1 dx2 dyi . (8.105)
i=3 −∞ −∞ −∞
Therefore, reasoning inductively, after n steps, we get

|u(x)|n/(n−1) dx
Rn
n ∞ ∞ 1/(n−1)
≤∏ ... |Du| dx
i=1 −∞ −∞
n/(n−1)
= |Du| dx . (8.106)
Rn
This is the result for p = 1. Now suppose 1 1 apply the above result for
v = |u|γ ,
to obtain
(n−1)/n
|u(x)|γ n/(n−1) dx ≤ |D|u|γ |; dx
Rn Rn

≤γ |u|γ −1 |Du|; dx
Rn
(p−1)/p
≤γ |u|(γ −1)p/(p−1) dx
Rn
1/p
× |Du| dx
p
. (8.107)
Rn
In particular for γ such that

γn (γ − 1)p
= ,
n−1 p−1
that is, γ = p(n−1)
n−p , so that
γn (γ − 1)p np
= = ,
n−1 p−1 n− p
we get
((n−1)/n−(p−1)/p) 1/p
|u| dx
r
≤C |Du| dx
p
.
Rn Rn
From this and considering that

n−1 p−1 n− p 1
− = = ,
n p np r
we finally obtain
1/r 1/p
|u|r dx ≤C |Du| p dx .
Rn Rn
Theorem 8.4.15. Let Ω ⊂ Rn be a bounded open set. Suppose ∂ Ω is Ĉ1 , 1 ≤ p < n,

and u ∈ W 1,p (Ω ).
Then u ∈ Lr (Ω ) and
u r,Ω ≤ K u 1,p,Ω ,
where the constant depends only on p, n, and Ω .
Proof. Since ∂ Ω is Ĉ1 , from Theorem 8.4.12, there exists an extension Eu = ū ∈

W 1,p (Rn ) such that ū = u in Ω the support of ū is compact and
ū 1,p,Rn ≤ C u 1,p,Ω ,
where C does not depend on u. As ū has compact support, from Theorem 8.4.10,
there exists a sequence {uk } ∈ Cc∞ (Rn ) such that
uk → ū in W 1,p (Rn ),
from the last theorem
uk − ul r,Rn ≤ K Duk − Dul p,Rn .
Hence,
uk → ū in Lr (Rn ),
also from the last theorem
uk r,Rn ≤ K Duk p,Rn , ∀k ∈ N,
so that
ū r,Rn ≤ K Dū p,Rn .
Therefore, we may get
u r,Ω ≤ ū r,Rn
≤ K Dū p,Rn
≤ K1 ū 1,p,Rn
≤ K2 u 1,p,Ω . (8.108)
Theorem 8.4.16. Let Ω ⊂ Rn be a bounded open set such that ∂ Ω ∈ Ĉ1 . If mp < n,
then W m,p (Ω ) → Lq (Ω ) for p ≤ q ≤ (np)/(n − mp).
Proof. Define q0 = np/(n − mp). We first prove by induction on m that
W m,p → Lq0 (Ω ).
The last result is exactly the case for m = 1. Assume
W m−1,p → Lr1 (Ω ), (8.109)
where
r1 = np/(n − (m − 1)p) = np/(n − np + p),
whenever n > (m − 1)p. If u ∈ W m,p (Ω ) where n > mp, then u and D j u are in
W m−1,p (Ω ), so that from (8.109) we have u ∈ W 1,r1 (Ω ) and
u 1,r1,Ω ≤ K u m,p,Ω . (8.110)
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
u q0 ,Ω ≤ K2 u 1,r1,Ω ,
where the constant K2 does not depend on u, and therefore from this and (8.110) we
obtain
u q0 ,Ω ≤ K2 u 1,r1,Ω ≤ K3 u m,p,Ω . (8.111)
The induction is complete. Now suppose p ≤ q ≤ q0 . Define
s = (q0 − q)p/(q0 − p) and t = p/s = (q0 − p)/(q0 − q).
Through the Hölder inequality, we get

u qq,Ω = |u(x)|s |u(x)|q−s dx
Ω
1/t 1/t
(q−s)t
≤ |u(x)| dx
st
|u(x)| dx
Ω Ω
p/t q /t
= u p,Ω u q00,Ω
q /t
≤ u p,Ω (K3 )q0 /t u m,p,
p/t 0
Ω
q /t
≤ (K3 )q0 /t u m,p,Ω u m,p,
0 p/t
Ω

= (K3 )q0 /t u qm,p,Ω , (8.112)
since
p/t + q0/t = q.
Corollary 8.4.17. If mp = n, then W m,p (Ω ) → Lq for p ≤ q < ∞.
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 (Ω ) → W m,s (Ω )
with the imbedding constant depending only on |Ω |. Since ms < n, by the last the-
orem, we obtain
W m,p (Ω ) → W m,s (Ω ) → Lq (Ω ).

Now if p ≤ q ≤ p , from above we have W m,p (Ω ) → L p (Ω ) and the obvious imbed-
ding W m,p (Ω ) → L p (Ω ). 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 (8.112).
About the next theorem, note that its hypotheses are satisfied if ∂ Ω is Ĉ1 (here we
do not give the details).
Theorem 8.4.18. Let Ω ⊂ Rn be an open bounded set, such that for each x ∈ Ω
there exists a convex set Cx ⊂ Ω whose shape depends on x but such that |Cx | > α ,
for some α > 0 that does not depend on x. Thus, if mp > n, then
W m,p (Ω ) → CB0 (Ω ).
Proof. Suppose first m = 1 so that p > n. Fix x ∈ Ω and pick y ∈ Cx . For φ ∈ C∞ (Ω ),

from the fundamental theorem of calculus, we have
1
d(φ (x + t(y − x))
φ (y) − φ (x) = dt.
0 dt
Thus, 1
d(φ (x + t(y − x))
dt,
|φ (x)| ≤ |φ (y)| +
0 dt
and hence
1
d(φ (x + t(y − x))
dt dy,
|φ (x)| dy ≤ |φ (y)| dy +
Cx Cx Cx 0 dt
so that, from the Hölder inequality and the Fubini theorem, we get
|φ (x)|α ≤ |φ (x)| · |Cx |

1
d(φ (x + t(y − x))
≤ φ p,Ω |Cx |1/p + dy dt.
0 Cx
dt
Therefore 1

|φ (x)|α ≤ φ p,Ω |Ω |1/p + |∇φ (z)|δ t −n dz dt,
0 V
where |V | = t n |Cx | and δ denote the diameter of Ω . From the Hölder inequality
again, we obtain
1 1/p

|φ (x)|α ≤ φ p,Ω |Ω |1/p + δ |∇φ (z)| p dy t −n (t n |Cx |)1/p dt,
0 V
and thus
1
1/p 1/p
|φ (x)|α ≤ φ p,Ω |Ω | + δ |Cx | ∇φ p,Ω t −n(1−1/p ) dt.
0
Since p > n we obtain

1 1
1
t −n(1−1/p ) dt = t −n/p dt = .
0 0 1 − n/p
From this, the last inequality and from the fact that |Cx | ≤ |Ω |, we have that there
exists K > 0 such that
|φ (x)| ≤ K φ 1,p,Ω , ∀x ∈ Ω , φ ∈ C∞ (Ω ). (8.113)
Here the constant K depends only on p, n, and Ω . Consider now u ∈ W 1,p (Ω ).

Thus there exists a sequence {φk } ⊂ C∞ (Ω ) such that
φk → u, in W 1,p (Ω ).
Up to a not relabeled subsequence, we have
φk → u, a.e. in Ω . (8.114)
Fix x ∈ Ω such that the limit indicated in (8.114) holds. Suppose given ε > 0. There-
fore, there exists k0 ∈ N such that
|φk0 (x) − u(x)| ≤ ε /2
and
φk0 − u 1,p,Ω < ε /(2K).
Thus,
|u(x)| ≤ |φk0 (x)| + ε /2

≤ K φk0 1,p,Ω + ε /2
≤ K u 1,p,Ω + ε . (8.115)
Since ε > 0 is arbitrary, the proof for m = 1 is complete, because for {φk } ∈ C∞ (Ω )
such that φk → u in W 1,p (Ω ), from (8.113), we have that {φk } is a uniformly Cauchy
sequence, so that it converges to a continuous u∗ , where u∗ = u, a.e. in Ω .
For m > 1 but p > n we still have
|u(x)| ≤ K u 1,p,Ω ≤ K1 u m,p,Ω , a.e. in Ω , ∀u ∈ W m,p (Ω ).
If p ≤ n ≤ mp, there exists j satisfying 1 ≤ j ≤ m − 1 such that jp ≤ n ≤ ( j + 1)p.

If jp < n, set
r̂ = np/(n − jp).
Let 1 ≤ p1 ≤ n such that
r̂ = np1 /(n − p1).
Thus we have that

np/(n − jp) = np1 /(n − p1),
so that
p1 = np/(n − ( j − 1)p),
so that by above and the last theorem:
u ∞ ≤ K1 u 1,r̂,Ω ≤ K1 u m− j,r̂,Ω ≤ K2 u m−( j−1),p1,Ω .
Now define
r̂1 = p1 = np/(n − ( j − 1)p)
and 1 ≤ p2 ≤ n such that
r̂1 = np2 /(n − p2),
so that
np/(n − ( j − 1)p) = np2 /(n − p2).
Hence p2 = np/(n − ( j − 2)p) so that by the last theorem
u m−( j−1),p1,Ω = u m−( j−1),r̂1,Ω ≤ K3 u m−( j−2),p2,Ω .
Proceeding inductively in this fashion, after j steps, observing that p j = p, we get
u ∞ ≤ K1 u 1,r̂,Ω ≤ K1 u m− j,r̂,Ω ≤ K j u m,p,Ω ,
for some appropriate K j . Finally, if jp = n, choosing r̂ = max{n, p} also by the

last theorem we obtain the same last chain of inequalities. For that, assume r̂ =
max{n, p} = n > p. Let p1 be such that
np1
r1 = = n,
n − p1
that is,
n
p1 = .
2
Since n > p, we have that n ≥ 2 so that 1 ≤ p1 < n. From the last theorem we obtain
u ∞ ≤ C u m− j,r1 ,Ω ≤ C1 u m−( j−1),p1,Ω .
Let r2 = p1 = n/2, and define p2 such that

np2
r2 = n/2 = ,
n − p2
that is, p2 = n/3.

Hence, again by the last theorem, we get
u ∞ ≤ C1 u m−( j−1),r2,Ω ≤ C2 u m−( j−2),p2,Ω .

Reasoning inductively, after j − 1 steps, we get p j−1 = n/ j = p, so that
u ∞ ≤ C u m−( j−1),r1,Ω ≤ C3 u m−( j−( j−1)),p j−1,Ω ≤ C4 u m,p,Ω .
Finally, if r1 = max{n, p} = p ≥ n, define p1 such that

np1
r1 = p = ,
n − p1
that is,
np
p1 = ≤ p,
n+ p
so that by last theorem
u ∞ ≤ u m− j,r1,Ω ≤ C5 u m−( j−1),p1,Ω ≤ C6 u m,p,Ω .
Theorem 8.4.19. Let Ω ⊂ Rn be a set with a boundary Ĉ1 . If mp > n, then

W m,p (Ω ) → Lq (Ω ) for p ≤ q ≤ ∞.
Proof. From the proof of the last theorem, we may obtain
u ∞,Ω ≤ K u m,p,Ω , ∀u ∈ W m,p (Ω ).
If p ≤ q < ∞, we have

u qq,Ω = |u(x)| p |u(x)|q−p dx
Ω
q−p
≤ |u(x)| p K u m,p,Ω dx
Ω
≤K q−p
u pp,Ω u m,p,
q−p
Ω
p q−p
≤ K q−p u m,p,Ω u m,p,Ω
q
= K q−p u m,p,Ω . (8.116)
Theorem 8.4.20. Let S ⊂ Rn be an n-dimensional ball of radius bigger than 3. If

n < p, then there exists a constant C, depending only on p and n, such that
u C0,λ (S) ≤ C u 1,p,S, ∀u ∈ C1 (S),
where 0 < λ ≤ 1 − n/p.
Proof. First consider λ = 1 − n/p and u ∈ C1 (S). Let x, y ∈ S such that |x − y| < 1
and define σ = |x − y|. Consider a fixed cube denoted by Rσ ⊂ S such that |Rσ | = σ n
and x, y ∈ R̄σ . For z ∈ Rσ , we may write
1
du(x + t(z − x))
u(x) − u(z) = − dt,
0 dt
that is,
1
u(x)σ n = u(z) dz − ∇u(x + t(z − x)) · (z − x) dt dz.
Rσ Rσ 0
Thus, denoting in the next lines V by an appropriate set such that |V | = t n |Rσ |, we
obtain
1
√
|u(x) − u(z) dz/σ n | ≤ nσ 1−n |∇u(x + t(z − x))| dt dz
Rσ Rσ 0
1
√
≤ nσ 1−n t −n |∇u(z)| dz dt
0 V
1
√
≤ nσ 1−n t −n ∇u p,S|V |1/p dt
0
1
√
≤ nσ 1−n σ n/p ∇u p,S t −nt n/p dt
0
1
√
≤ nσ 1−n/p ∇u p,S t −n/p dt
0
≤σ 1−n/p u 1,p,SK, (8.117)
where
√ 1 −n/p √
K= n t dt = n/(1 − n/p).
0
A similar inequality holds with y in place of x, so that
|u(x) − u(y)| ≤ 2K|x − y|1−n/p u 1,p,S, ∀x, y ∈ Rσ .
Now consider 0 < λ < 1 − n/p. Observe that, as |x− y|λ ≥ |x− y|1−n/p, if |x− y| < 1,
we have

|u(x) − u(y)|
sup | x =
y, |x − y| < 1
x,y∈S |x − y|λ

|u(x) − u(y)|
≤ sup | x = y, |x − y| < 1 ≤ K u 1,p,S. (8.118)
x,y∈S |x − y|1−n/p
Also,

|u(x) − u(y)|
sup | |x − y| ≥ 1 ≤ 2 u ∞,S ≤ 2K1 u 1,p,S
x,y∈S |x − y|λ
so that

|u(x) − u(y)|
sup | x = y ≤ (K + 2K1) u 1,p,S , ∀u ∈ C1 (S).
x,y∈S |x − y|λ
Theorem 8.4.21. Let Ω ⊂ Rn be an open bounded set such that ∂ Ω is Ĉ1 . Assume
n < p ≤ ∞.
Then
W 1,p (Ω ) → C0,λ (Ω ),
for all 0 < λ ≤ 1 − n/p.
Proof. Fix 0 < λ ≤ 1 − n/p and let u ∈ W 1,p (Ω ). Since ∂ Ω is Ĉ1 , from Theo-
rem 8.4.12, there exists an extension Eu = ū such that ū = u, a.e. in Ω , and
ū 1,p,Rn ≤ K u 1,p,Ω ,
where the constant K does not depend on u. From the proof of this same theorem,
we may assume that spt(ū) is on an n-dimensional sphere S ⊃ Ω with sufficiently
big radius and such sphere does not depend on u. Thus, in fact, we have
ū 1,p,S ≤ K u 1,p,Ω .
Since C∞ (S) is dense in W 1,p (S), there exists a sequence {φk } ⊂ C∞ (S) such that
uk → ū, in W 1,p (S). (8.119)
Up to a not relabeled subsequence, we have
uk → ū, a.e. in Ω .
From last theorem we have
uk − ul C0,λ (S) ≤ C uk − ul 1,p,S ,
so that {uk } is a Cauchy sequence in C0,λ (S), and thus uk → u∗ for some u∗ ∈
C0,λ (S). Hence, from this and (8.119), we have
u∗ = ū, a.e. in S.
Finally, from above and last theorem, we may write
u∗ C0,λ (Ω ) ≤ u∗ C0,λ (S) ≤ K1 ū 1,p,S ≤ K2 u 1,p,Ω .
8.5 The Trace Theorem
In this section we state and prove the trace theorem.

Theorem 8.5.1. Let 1 < p < ∞ and let Ω ⊂ Rn be an open bounded set such that
∂ Ω is Ĉ1 . Then there exists a bounded linear operator
T : W 1,p (Ω ) → L p (∂ Ω ),
such that
• Tu = u|∂ Ω if u ∈ W 1,p (Ω ) ∩C(Ω ),
•
Tu p,∂ Ω ≤ C u 1,p,Ω , ∀u ∈ W 1,p (Ω ),
where the constant C depends only on p and Ω .
Proof. Let u ∈ W 1,p (Ω ) ∩C(Ω ). Choose x0 ∈ ∂ Ω .
Since ∂ Ω is Ĉ1 , there exists r > 0 such that for a local coordinate system we may
write
Ω ∩ Br (x0 ) = {x ∈ Br (x0 ) | xn ≥ f (x1 , . . . , xn−1 )},
where denoting x̂ = (x1 , . . . , xn−1 ), f (x̂) is continuous and such that its partial deriva-
tives are classically defined a.e. and bounded on its domain. Furthermore
| f (x̂) − f (ŷ)| ≤ K|x̂ − ŷ|2 , ∀x̂, ŷ
for some K > 0 also on its domain.

Define the coordinates y by
yi = xi , ∀i ∈ {1, . . . , n − 1},
and
yn = xn − f (x1 , . . . , xn−1 ).
Define û(y) by
u(x1 , . . . , xn ) = u(y1 , . . . , yn−1 , yn + f (y1 , . . . , yn−1 ) = û(y).
Also define y0 = (x01 , . . . , x0n−1 , x0n − f (x01 , . . . , x0n−1 ) = (y01 , . . . , y0n−1 , 0) and
choose r1 > 0 such that
Ψ −1 (B+
r1 (y0 )) ⊂ Ω ∩ Br (x0 ).
Observe that this is possible since Ψ and Ψ −1 are continuous, where y = Ψ (x). Here
B+
r1 (y0 ) = {y ∈ Br1 (y0 ) | yn > 0}.
+
For each N ∈ N, choose, by mollification, for example, φN ∈ C∞ (Br1 (y0 )) such that
8.5 The Trace Theorem 215
1
φN − û ∞,B+ (y ) < .
r1 0 N
Denote B = Br1 /2 (y0 ), and B+ = B+ (y ). Now choose η ∈ Cc∞ (Br1 (y0 )), such that
r1 /2 0
η > 0 and η ≡ 1 on B. Also denote
Γ̃ = {y ∈ B | yn = 0},
and
Γ̃1 = {y ∈ Br1 (y0 ) | yn = 0}.
Observe that

|φN | p dΓ ≤ η |φN | p dΓ
Γ̃ Γ̃1

=− (η |φN | p )yn dy
B+
r1

≤− (ηyn |φN | p ) dy
B+
r1

+ (p|φN | p−1 |(φN )yn |η ) dy. (8.120)
B+
r1
Here we recall the Young inequality

a p bq 1 1
ab ≤ + , ∀a, b ≥ 0, where + = 1.
p q p q
Thus,
|(φN )yn | p η p |φN |(p−1)q
(|φN | p−1 |)(|(φN )yn |η ) ≤ + ,
p q
so that replacing such an inequality in (8.120), since (p − 1)q = p, we get

|φN | p dΓ ≤ C1 |φN | p dy + |DφN | p dy . (8.121)
Γ̃ B+
r1 B+
r1
Letting N → +∞ we obtain

|u(x)| dΓ ≤ C2
p
|û(y)| p dΓ
Γ Γ̃

≤ C3 |û| dy +
p
|Dû| dy
p
B+
r1 B+
r1

≤ C4 |u| dx +
p
|Du| dx ,
p
(8.122)
W+ W+
where Γ = ψ −1 (Γ̃ ) and W + = Ψ −1 (B+

r1 ).
Observe that denoting W = Wx0 we have that ∂ Ω ⊂ ∪x∈∂ Ω Wx , and thus, since
∂ Ω is compact, we may select x1 , . . . , xM such that ∂ Ω ⊂ ∪M
i=1Wi . We emphasize to
have denoted Wxi = Wi , ∀i ∈ {1, . . . , M}. Denoting Wi+ = Wi ∩ Ω we may obtain
M
∂Ω
|u(x)| p dΓ ≤ ∑ |u(x)| p dΓ
i=1 Γi
M
≤ ∑ C4i Wi+
|u| dx +
p
Wi+
|Du| dx
p
i=1

≤ C5 M |u| p dx + |Du| p dx
Ω Ω

=C |u| p dx + |Du| p dx . (8.123)
Ω Ω
At this point we denote Tu = u|∂ Ω .

Finally, for the case u ∈ W 1,p (Ω ), select {uk } ⊂ C∞ (Ω ) such that
uk − u 1,p,Ω → 0, as k → ∞.
From above
Tuk − Tul p,∂ Ω ≤ C uk − ul 1,p,Ω ,
so that
{Tuk }
is a Cauchy sequence. Hence we may define
Tu = lim Tuk , in L p (∂ Ω ).
k→∞
Remark 8.5.2. Similar results are valid for W0m,p ; however, in this case the traces
relative to derivatives of order up to m − 1 are involved.
8.6 Compact Imbeddings
Theorem 8.6.1. Let m be a nonnegative integer and let 0 < ν < λ ≤ 1. Then the
following imbeddings exist:
Cm+1 (Ω ) → Cm (Ω ), (8.124)
Cm,λ (Ω ) → Cm (Ω ), (8.125)
8.6 Compact Imbeddings 217
Cm,λ (Ω ) → Cm,ν (Ω ). (8.126)
If Ω is bounded, then Imbeddings (8.125) and (8.126) are compact.
Proof. Imbeddings (8.124) and (8.125) follow from the inequalities
φ Cm (Ω ) ≤ φ Cm+1 (Ω ) ,
φ Cm (Ω ) ≤ φ Cm,λ (Ω ) .
To establish (8.126) note that for |α | ≤ m
α
|D φ (x) − Dα φ (y)|
sup | x =
y, |x − y| < 1
x,y∈Ω |x − y|ν
α
|D φ (x) − Dα φ (y)|
≤ sup | x = y, |x − y| < 1 , (8.127)
x,y∈Ω |x − y|λ
and also,

|Dα φ (x) − Dα φ (y)|
sup | |x − y| ≥ 1 ≤ 2 sup {|Dα φ |}. (8.128)
x,y∈Ω |x − y|ν x∈Ω
Therefore, we may conclude that
φ Cm,ν (Ω̄ ) ≤ 3 φ Cm,λ (Ω ) , ∀φ ∈ Cm,ν (Ω ).
Now suppose Ω is bounded. If A is a bounded set in C0,λ (Ω ), then there exists

M > 0 such that
φ C0,λ (Ω ) ≤ M, ∀φ ∈ A.
But then
|φ (x) − φ (y)| ≤ M|x − y|λ , ∀x, y ∈ Ω , φ ∈ A,
so that by the Ascoli–Arzela theorem, A is pre-compact in C(Ω ). This proves the
compactness of (8.125) for m = 0.
If m ≥ 1 and A is bounded in Cm,λ (Ω ), then A is bounded in C0,λ (Ω ). Thus, by
above there is a sequence {φk } ⊂ A and φ ∈ C0,λ (Ω ) such that
φk → φ in C(Ω ).
However, {Di φk } is also bounded in C0,λ (Ω ), so that there exists a not relabeled
subsequence, also denoted by {φk } and ψi such that
Di φk → ψi , in C(Ω ).
The convergence in C(Ω̄ ) being the uniform one, we have ψi = Di φ . We can proceed
extracting (not relabeled) subsequences until obtaining
Dα φk → Dα φ , in C(Ω ), ∀ 0 ≤ |α | ≤ m.
This completes the proof of compactness of (8.125). For (8.126), let S be a bounded
set in Cm,λ (Ω ). Observe that
ν /λ
|Dα φ (x) − Dα φ (y)| |Dα φ (x) − Dα φ (y)|
=
|x − y|ν |x − y|λ
·|Dα φ (x) − Dα φ (y)|1−ν /λ
≤ K|Dα φ (x) − Dα φ (y)|1−ν /λ , (8.129)
for all φ ∈ S. From (8.125), S has a converging subsequence in Cm (Ω ). From (8.129)

such a subsequence is also converging in Cm,ν (Ω ). The proof is complete.
Theorem 8.6.2 (Rellich–Kondrachov). Let Ω ⊂ Rn be an open bounded set such
that ∂ Ω is Ĉ1 . 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 (Ω ) → W j,q (Ω ),
if 0 < n − mp < n and 1 ≤ np/(n − mp), (8.130)
W j+m,p (Ω ) → W j,q (Ω ), if n = mp, 1 ≤ q < ∞. (8.131)
2. Part II If mp > n, then the following imbeddings are compact:
W j+m,p → CB (Ω ),
j
(8.132)
W j+m,p (Ω ) → W j,q (Ω ), if 1 ≤ q ≤ ∞. (8.133)
3. Part III The following imbeddings are compact:
W j+m,p (Ω ) → C j (Ω ), if mp > n, (8.134)
W j+m,p (Ω ) → C j,λ (Ω ),
if mp > n ≥ (m − 1)p and 0 < λ < m − n/p. (8.135)
4. Part IV All the above imbeddings are compact if we replace W j+m,p (Ω ) by

j+m,p
W0 (Ω ).
Remark 8.6.3. 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 → ũ where ũ(x) = u(x) if x ∈ Ω
and ũ(x) = 0 if x ∈ Rn − Ω defines an imbedding W0j+m,p (Ω ) → 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 Ω bounded, the general results may be found in
Adams [1]).
Remark 8.6.4. 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 (Ω ) we have that
{Dα ui } is bounded in W m,p (Ω ) for each α such that |α | ≤ j. From the case j = 0 it
is possible to extract a subsequence (similarly to a diagonal process) {uik } for which
{Dα uik } converges in Lq (Ω ) for each α such that |α | ≤ j, so that {uik } converges
in W j,q (Ω ).
Remark 8.6.5. Since Ω is bounded, CB0 (Ω ) → Lq (Ω ) for 1 ≤ q ≤ ∞. In fact
u 0,q,Ω ≤ u C0 [vol(Ω )]1/q . (8.136)

B
Thus the compactness of (8.133) (for j = 0) follows from that of (8.132).
Proof of Parts II and III. If mp > n > (m − 1)p and 0 < λ < (m − n)/p, then
there exists μ such that λ < μ < m − (n/p). Since Ω is bounded, the imbedding
C0,μ (Ω ) → C0,λ (Ω ) is compact by Theorem 8.6.1. Since by the Sobolev imbed-
ding theorem we have W m,p (Ω ) → C0,μ (Ω ), we have that Imbedding (8.135) is
compact.
If mp > n, let j∗ be the nonnegative integer satisfying (m − j∗ )p > n ≥
(m − j∗ − 1)p. Thus we have the chain of imbeddings
∗ ,p
W m,p (Ω ) → W m− j (Ω ) → C0,μ (Ω ) → C(Ω ), (8.137)
where 0 < μ < m − j∗ − (n/p). The last imbedding in (8.137) is compact by Theo-
rem 8.6.1, so that (8.134) is compact for j = 0. By analogy (8.132) is compact for
j = 0. Therefore from the above remarks (8.133) is also compact. For the proof of
Part I, we need the following lemma:
Lemma 8.6.6. Let Ω be a bounded domain in Rn . Let 1 ≤ q1 ≤ q0 and suppose
W m,p (Ω ) → Lq0 (Ω ), (8.138)
W m,p (Ω ) → Lq1 . (8.139)
Suppose also that (8.139) is compact. If q1 ≤ q < q0 , then the imbedding
W m,p → Lq (Ω ) (8.140)
is compact.
Proof. Define λ = q1 (q0 − q)/(q(q0 − q1 )) and μ = q0 (q − q1)/(q(q0 − q1 )). We

have that λ > 0 and μ ≥ 0. From the Hölder inequality and (8.138) there exists
K ∈ R+ such that
μ μ
u 0,q,Ω ≤ u λ0,q1,Ω u 0,q0,Ω ≤ K u λ0,q1,Ω u m,p,Ω , ∀u ∈ W m,p (Ω ). (8.141)
Thus considering a sequence {ui } bounded in W m,p (Ω ), since (8.139) is compact

there exists a subsequence {unk } that converges and is therefore a Cauchy sequence
in Lq1 (Ω ). From (8.141), {unk } is also a Cauchy sequence in Lq (Ω ), so that (8.140)
is compact.
Proof of Part I. Consider j = 0. Define q0 = np/(n − mp). To prove the imbedding
W m,p (Ω ) → Lq (Ω ), 1 ≤ q < q0 , (8.142)
is compact, by last lemma it suffices to do so only for q = 1. For k ∈ N, define
Ωk = {x ∈ Ω | dist(x, ∂ Ω ) > 2/k}. (8.143)
Suppose A is a bounded set of functions in W m,p (Ω ), that is, suppose there exists
K1 > 0 such that
u W m,p (Ω ) < K1 , ∀u ∈ A.
Also, suppose given ε > 0, and define, for u ∈ W m,p (Ω ), ũ(x) = u(x) if x ∈ Ω ,
ũ(x) = 0, if x ∈ Rn \ Ω . Fix u ∈ A. From the Hölder inequality and considering that
W m,p (Ω ) → Lq0 (Ω ), we have
1/q0 1−1/q0
|u(x)|dx ≤ |u(x)|q0 dx 1dx
Ω −Ω k Ω −Ω k Ω −Ω k
≤ K1 u m,p,Ω [vol(Ω − Ωk )] 1−1/q0

. (8.144)
Thus, since A is bounded in W m,p (Ω ), there exists K0 ∈ N such that if k ≥ K0 , then

|u(x)|dx < ε , ∀u ∈ A, (8.145)
Ω −Ω k
and, now fixing a not relabeled k > K0 , we get

|ũ(x + h) − ũ(x)|dx < 2ε , ∀u ∈ A, ∀h ∈ Rn . (8.146)
Ω −Ω k
Observe that if |h| < 1/k, then x + th ∈ Ω2k provided x ∈ Ωk and 0 ≤ t ≤ 1. If

u ∈ C∞ (Ω ), we have that
1
du(x + th)
|u(x + h) − u(x)| ≤ dx | |dt
Ωk Ωk 0 dt
1
≤ |h| dt |∇u(y)|dy ≤ |h| u 1,1,Ω
0 Ω2k
≤ K2 |h| u m,p,Ω . (8.147)
Since C∞ (Ω ) is dense in W m,p (Ω ), from above for |h| sufficiently small,

|ũ(x + h) − ũ(x)|dx < 3ε , ∀u ∈ A. (8.148)
Ω
From Theorem 8.4.6, A is relatively compact in L1 (Ω ), and therefore the imbedding

indicated (8.142) is compact for q = 1. This completes the proof.
Part II
Variational Convex Analysis
Chapter 9
Basic Concepts on the Calculus of Variations
9.1 Introduction to the Calculus of Variations
We emphasize the main references for this chapter are [37, 38, 68].
Here we recall that a functional is a function whose 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:

1. F(u) = Ω f (x, u, ∇u) dx, where Ω ⊂ Rn is an open, bounded, and connected set.
2. F(u) = Ω f (x, u, ∇u, D2 u) dx, here

∂ ui
Du = ∇u =
∂xj
and
∂ 2 ui
D u = {D ui } =
2 2
,
∂ xk ∂ xl
for i ∈ {1, . . . , N} and j, k, l ∈ {1, . . ., n}.
Also, f : Ω × RN × RN×n → R is denoted by f (x, s, ξ ) and we assume
1.
∂ f (x, s, ξ )
∂s
and
2.
∂ f (x, s, ξ )
∂ξ
are continuous ∀(x, s, ξ ) ∈ Ω × RN × RN×n .
Remark 9.1.1. We also recall that the notation ∇u = Du may be used.
Now we define our general problem, namely problem P where
226 9 Basic Concepts on the Calculus of Variations
Problem P : minimize F(u) on U,
that is, to find u0 ∈ U such that
F(u0 ) = min{F(u)}.
u∈U
At this point, we introduce some essential definitions.

Definition 9.1.2 (Space of Admissible Variations). Given F : U → R we define the
space of admissible variations for F, denoted by V as
V = {ϕ | u + ϕ ∈ U, ∀u ∈ U}.
For example, for F : U → R given by

1
F(u) = ∇u · ∇u dx − u, f U ,
2 Ω
where Ω ⊂ R3 and
U = {u ∈ W 1,2 (Ω ) | u = û on ∂ Ω }
we have
V = W01,2 (Ω ).
Observe that in this example U is a subset of a Banach space.
Definition 9.1.3 (Local Minimum). Given F : U → R, we say that u0 ∈ U is a local
minimum for F if there exists δ > 0 such that
F(u) ≥ F(u0 ), ∀u ∈ U, such that u − u0 U < δ ,
or equivalently
F(u0 + ϕ ) ≥ F(u0 ), ∀ϕ ∈ V , such that ϕ U < δ .
Definition 9.1.4 (Gâteaux Variation). Given F : U → R we define the Gâteaux

variation of F at u ∈ U on the direction ϕ ∈ V , denoted by δ F(u, ϕ ) as
F(u + εϕ ) − F(u)
δ F(u, ϕ ) = lim ,
ε →0 ε
if such a limit is well defined. Furthermore, if there exists u∗ ∈ U ∗ such that
δ F(u, ϕ ) = ϕ , u∗ U , ∀ϕ ∈ U,
we say that F is Gâteaux differentiable at u ∈ U, and u∗ ∈ U ∗ is said to be the

Gâteaux derivative of F at u. Finally we denote
9.2 Evaluating the Gâteaux Variations 227
∂ F(u)
u∗ = δ F(u) or u∗ = .
∂u
9.2 Evaluating the Gâteaux Variations
Consider F : U → R such that

F(u) = f (x, u, ∇u) dx
Ω
where the hypothesis indicated in the last section is assumed. Consider u ∈

C1 (Ω̄ ; RN ) and ϕ ∈ Cc1 (Ω̄ ; RN ) and let us evaluate δ F(u, ϕ ):
From Definition 9.1.4,
F(u + εϕ ) − F(u)
δ F(u, ϕ ) = lim .
ε →0 ε
Observe that
f (x, u + εϕ , ∇u + ε ∇ϕ ) − f (x, u, ∇u)

lim
ε →0 ε
∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
= ·ϕ + · ∇ϕ .
∂s ∂ξ
Define
f (x, u + εϕ , ∇u + ε ∇ϕ ) − f (x, u, ∇u)
G(x, u, ϕ , ε ) = ,
ε
and
∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
G̃(x, u, ϕ ) = ·ϕ + · ∇ϕ .
∂s ∂ξ
Thus we have
lim G(x, u, ϕ , ε ) = G̃(x, u, ϕ ).
ε →0
Now we will show that

lim G(x, u, ϕ , ε ) dx = G̃(x, u, ϕ ) dx.
ε →0 Ω Ω
Suppose to obtain contradiction that we do not have

ε →0 Ω Ω
Hence, there exists ε0 > 0 such that for each n ∈ N there exists 0 < εn < 1/n such
that

G(x, u, ϕ , εn ) dx − G̃(x, u, ϕ ) dx ≥ ε0 . (9.1)
Ω Ω

Define
cn = max{|G(x, u(x), ϕ (x), εn ) − G̃(x, u(x), ϕ (x))|}.
x∈Ω
Since the function in question is continuous on the compact set Ω , {xn } is well
defined. Also from the fact that Ω is compact, there exists a subsequence {xn j } and
x0 ∈ Ω̄ such that
lim xn j = x0 .
j→+∞
Thus
lim cn j = c0
j→+∞
= lim {|G(xn j , u(xn j ), ϕ (xn j ), εn j ) − G̃(x0 , u(x0 ), ϕ (x0 ))|} = 0.

j→+∞
Therefore there exists j0 ∈ N such that if j > j0 , then
cn j < ε0 /|Ω |.
Thus, if j > j0 , we have

G(x, u, ϕ , εn ) dx − G̃(x, u, ϕ ) dx
Ω j
Ω

≤ |G(x, u, ϕ , εn j ) − G̃(x, u, ϕ )| dx ≤ cn j |Ω | < ε0 , (9.2)
Ω
which contradicts (9.1). Hence, we may write

lim G(x, u, ϕ , ε ) dx = G̃(x, u, ϕ ) dx,
ε →0 Ω Ω
that is,
∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
δ F(u, ϕ ) = ·ϕ + · ∇ϕ dx.
Ω ∂s ∂ξ
Theorem 9.2.1 (Fundamental Lemma of Calculus of Variations). Consider an

open set Ω ⊂ Rn and u ∈ L1loc (Ω ) such that

uϕ dx = 0, ∀ϕ ∈ Cc∞ (Ω ).
Ω
Then u = 0, a.e. in Ω .
Remark 9.2.2. Of course a similar result is valid for the vectorial case. A proof of
such a result was given in Chap. 8.
Theorem 9.2.3 (Necessary Conditions for a Local Minimum). Suppose u ∈ U is

a local minimum for a Gâteaux differentiable F : U → R. Then
9.3 The Gâteaux Variation: A More General Case 229
δ F(u, ϕ ) = 0, ∀ϕ ∈ V .
Proof. Fix ϕ ∈ V . Define φ (ε ) = F(u + εϕ ). Since by hypothesis φ is differentiable
and attains a minimum at ε = 0, from the standard necessary condition φ (0) = 0,
we obtain φ (0) = δ F(u, ϕ ) = 0.
Theorem 9.2.4. Consider the hypotheses stated in Section 9.1 on F : U → R. Sup-
pose F attains a local minimum at u ∈ C2 (Ω̄ ; RN ) and additionally assume that
f ∈ C2 (Ω , RN , RN×n ). 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)
− div = θ , in Ω .
∂s ∂ξ
Proof. From Theorem 9.2.3, the necessary condition stands for δ F(u, ϕ ) = 0, ∀ϕ
∈ V . From the above this implies, after integration by parts

∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
− div · ϕ dx = 0,
Ω ∂s ∂ξ
∀ϕ ∈ Cc∞ (Ω , RN ).
The result then follows from the fundamental lemma of calculus of variations.
9.3 The Gâteaux Variation: A More General Case
Theorem 9.3.1. Consider the functional F : U → R, where
U = {u ∈ W 1,2 (Ω , RN ) | u = u0 in ∂ Ω }.
Suppose
F(u) = f (x, u, ∇u) dx,
Ω
where f : Ω × RN × RN×n is such that for each K > 0 there exists K1 > 0 which does
not depend on x such that
| f (x, s1 , ξ1 ) − f (x, s2 , ξ2 )| < K1 (|s1 − s2 | + |ξ1 − ξ2 |)

∀s1 , s2 ∈ RN , ξ1 , ξ2 ∈ RN×n , such that |s1 | < K, |s2 | < K,
|ξ1 | < K, |ξ2 | < K.
Also assume the hypotheses of Section 9.1 except for the continuity of derivatives
of f . Under such assumptions, for each u ∈ C1 (Ω ; RN ) and ϕ ∈ Cc∞ (Ω ; RN ), we
have
∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
δ F(u, ϕ ) = ·ϕ + · ∇ϕ dx.
Ω ∂s ∂ξ
Proof. From Definition 9.1.4,
F(u + εϕ ) − F(u)
δ F(u, ϕ ) = lim .
ε →0 ε
Observe that
f (x, u + εϕ , ∇u + ε ∇ϕ ) − f (x, u, ∇u)

lim
ε →0 ε
∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
= ·ϕ + · ∇ϕ , a.e in Ω .
∂s ∂ξ
Define
f (x, u + εϕ , ∇u + ε ∇ϕ ) − f (x, u, ∇u)
G(x, u, ϕ , ε ) = ,
ε
and
∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
G̃(x, u, ϕ ) = ·ϕ + · ∇ϕ .
∂s ∂ξ
Thus we have
lim G(x, u, ϕ , ε ) = G̃(x, u, ϕ ), a.e in Ω .
ε →0
Now we will show that

ε →0 Ω Ω
It suffices to show that (we do not provide details here)

lim G(x, u, ϕ , 1/n) dx = G̃(x, u, ϕ ) dx.
n→∞ Ω Ω
Observe that for an appropriate K > 0, we have
|G(x, u, ϕ , 1/n)| ≤ K(|ϕ | + |∇ϕ |), a.e. in Ω . (9.3)
By the Lebesgue dominated convergence theorem, we obtain

lim G(x, u, ϕ , 1/(n)) dx = G̃(x, u, ϕ ) dx,
n→+∞ Ω Ω
that is,
∂ f (x, u, ∇u) ∂ f (x, u, ∇u)
δ F(u, ϕ ) = ·ϕ + · ∇ϕ dx.
Ω ∂s ∂ξ
9.5 Elementary Convexity 231
9.4 Fréchet Differentiability
In this section we introduce a very important definition, namely, Fréchet

differentiability.
Definition 9.4.1. Let U,Y be Banach spaces and consider a transformation T :
U→Y . We say that T is Fréchet differentiable at u ∈ U if there exists a bounded
linear transformation T (u) : U→Y such that
T (u + v) − T(u) − T (u)(v) Y
lim = 0, v = θ .
v→θ v U
In such a case T (u) is called the Fréchet derivative of T at u ∈ U.
9.5 Elementary Convexity
In this section we develop some proprieties concerning elementary convexity.

Definition 9.5.1. A function f : Rn → R is said to be convex if
f (λ x + (1 − λ )y) ≤ λ f (x) + (1 − λ ) f (y), ∀x, y ∈ Rn , λ ∈ [0, 1].
Proposition 9.5.2. If f : Rn → R is convex and differentiable, then
f (y) − f (x) ≥ f (x), y − xRn , ∀x, y ∈ Rn .
Proof. Pick x, y ∈ Rn . By hypothesis
f ((1 − λ )x + λ y) ≤ (1 − λ ) f (x) + λ f (y), ∀λ ∈ [0, 1].
Thus
f (x + λ (y − x)) − f (x)
≤ f (y) − f (x), ∀λ ∈ (0, 1].
λ
Letting λ → 0+ we obtain
f (y) − f (x) ≥ f (x), y − xRn .
Since x, y ∈ Rn are arbitrary, the proof is complete.
Proposition 9.5.3. Let f : Rn → R be a differentiable function. If
f (y) − f (x) ≥ f (x), y − xRn , ∀x, y ∈ Rn ,
then f is convex.
Proof. Define f ∗ (x∗ ) by
f (x∗ ) = sup {x, x∗ Rn − f (x)}.

x∈Rn
Such a function f ∗ is called the Fenchel conjugate of f . Observe that by hypothesis,
f ∗ ( f (x)) = sup {y, f (x)Rn − f (y)} = x, f (x)Rn − f (x). (9.4)

y∈Rn
On the other hand
f ∗ (x∗ ) ≥ x, x∗ Rn − f (x), ∀x, x∗ ∈ Rn ,
that is,
f (x) ≥ x, x∗ Rn − f ∗ (x∗ ), ∀x, x∗ ∈ Rn .
Observe that from (9.4)
f (x) = x, f (x)Rn − f ∗ ( f (x))
and thus
f (x) = sup {x, x∗ Rn − f (x∗ )}, ∀x ∈ Rn .
x∗ ∈Rn
Pick x, y ∈ Rn and λ ∈ [0, 1]. Thus, we may write
f (λ x + (1 − λ )y) = sup {λ x + (1 − λ )y, x∗Rn − f ∗ (x∗ )}

x∗ ∈Rn
= sup {λ x, x∗ Rn + (1 − λ )y, x∗Rn − λ f ∗ (x∗ )
x∗ ∈Rn
− (1 − λ ) f ∗(x∗ )}
≤λ { sup {x, x∗ Rn − f ∗ (x∗ )}}
x∗ ∈Rn
+ (1 − λ ){ sup {y, x∗ Rn − f ∗ (x∗ )}}
x∗∈Rn
=λ f (x) + (1 − λ ) f (y). (9.5)
Since x, y ∈ Rn and λ ∈ [0, 1] are arbitrary, we have that f is convex.

Corollary 9.5.4. Let f : Rn → R be twice differentiable and
2
∂ f (x)
,
∂ xi ∂ x j
positive definite, for all x ∈ Rn . Then f is convex.

Proof. Pick x, y ∈ Rn . Using Taylor’s expansion we obtain
n n
∂ 2 f (x̄)
f (y) = f (x) + f (x), y − xRn + ∑ ∑ (yi − xi )(y j − x j ),
i=1 j=1 ∂ xi ∂ x j
9.5 Elementary Convexity 233
for x̄ = λ x + (1 − λ )y (for some λ ∈ [0, 1]). From the hypothesis we obtain
f (y) − f (x) − f (x), y − xRn ≥ 0.
Since x, y ∈ Rn are arbitrary, the proof is complete.

Similarly we may obtain the following result.
Corollary 9.5.5. Let U be a Banach space. Consider F : U → R Gâteaux differen-
tiable. Then F is convex if and only if
F(v) − F(u) ≥ F (u), v − uU , ∀u, v ∈ U.
Definition 9.5.6 (The Second Variation). Let U be a Banach space. Suppose F :

U → R is a Gâteaux differentiable functional. Given ϕ , η ∈ V , we define the second
variation of F at u, relating the directions ϕ , η , denoted by
δ 2 F(u, ϕ , η ),
by
δ F(u + εη , ϕ ) − δ F(u, ϕ )
δ 2 F(u, ϕ , η ) = lim .
ε →0 ε
If such a limit exists ∀ϕ , η ∈ V , we say that F is twice Gâteaux differentiable at u.
Finally, if η = ϕ , we denote δ 2 F(u, ϕ , η ) = δ 2 F(u, ϕ ).
Corollary 9.5.7. Let U be a Banach space. Suppose F : U → R is a twice Gâteaux
differentiable functional and that
δ 2 F(u, ϕ ) ≥ 0, ∀u ∈ U, ϕ ∈ V .
Then, F is convex.
Proof. Pick u, v ∈ U. Define φ (ε ) = F(u + ε (v − u)). By hypothesis, φ is twice
differentiable, so that
φ (1) = φ (0) + φ (0) + φ (ε̃ )/2,
where |ε̃ | ≤ 1. Thus
F(v) = F(u) + δ F(u, v − u) + δ 2F(u + ε̃ (v − u), v − u)/2.
Therefore, by hypothesis,
F(v) ≥ F(u) + δ F(u, v − u).
Since F is Gâteaux differentiable, we obtain
F(v) ≥ F(u) + F (u), v − uU .
Being u, v ∈ U arbitrary, the proof is complete.

Corollary 9.5.8. Let U be a Banach space. Let F : U → R be a convex Gâteaux

differentiable functional. If F (u) = θ , 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) + F (u), v − uU , ∀u, v ∈ U.
Therefore, from F (u) = θ , we obtain
F(v) ≥ F(u), ∀v ∈ U.
Theorem 9.5.9 (Sufficient Condition for a Local Minimum). Let U be a Banach

space. Suppose F : U → R is a twice Gâteaux differentiable functional at a neigh-
borhood of u0 , so that
δ F(u0 ) = θ
and
δ 2 F(u, ϕ ) ≥ 0, ∀u ∈ Br (u0 ), ϕ ∈ V ,
for some r > 0. Under such hypotheses, we have
F(u0 ) ≤ F(u0 + εϕ ), ∀ ε , ϕ such that |ε | < min{r, 1}, ϕ U < 1.
Proof. Fix ϕ ∈ V such that ϕ U < 1. Define
φ (ε ) = F(u0 + εϕ ).
Observe that for |ε | < min{r, 1}, for some ε̃ such that |ε̃ | ≤ |ε |, we have
φ (ε ) = φ (0) + φ (0)ε + φ (ε̃ )ε 2 /2

= F(u0 ) + ε ϕ , δ F(u0 )U + (ε 2 /2)δ 2 F(u0 + ε̃ϕ , ϕ )
= F(u0 ) + (ε 2 /2)δ 2 F(u0 + ε̃ϕ , ϕ ) ≥ F(u0 ).
Hence,
F(u0 ) ≤ F(u0 + εϕ ), ∀ ε , ϕ such that |ε | < r, ϕ U < 1.
9.6 The Legendre–Hadamard Condition
Theorem 9.6.1. If u ∈ C1 (Ω̄ ; RN ) is such that
δ 2 F(u, ϕ ) ≥ 0, ∀ϕ ∈ Cc∞ (Ω , RN ),
9.6 The Legendre–Hadamard Condition 235
then
fξαi ξ k (x, u(x), ∇u(x))ρ i ρ k ηα ηβ ≥ 0, ∀x ∈ Ω , ρ ∈ RN , η ∈ Rn .
β
Such a condition is known as the Legendre-Hadamard condition.
Proof. Suppose
δ 2 F(u, ϕ ) ≥ 0, ∀ϕ ∈ Cc∞ (Ω ; RN ).
We denote δ 2 F(u, ϕ ) by

δ 2 F(u, ϕ ) = a(x)Dϕ (x) · Dϕ (x) dx
Ω

+ b(x)ϕ (x) · Dϕ (x) dx + c(x)ϕ (x) · ϕ (x) dx, (9.6)
Ω Ω
where
a(x) = fξ ξ (x, u(x), Du(x)),
b(x) = 2 fsξ (x, u(x), Du(x)),
and
c(x) = fss (x, u(x), Du(x)).
Now consider v ∈ Cc∞(B1 (0), RN ). Thus given x0 ∈ Ω for λ sufficiently small we
have that ϕ (x) = λ v x−x λ
0
is an admissible direction. Now we introduce the new
coordinates y = (y , . . . , y ) by setting y = λ −1 (x − x0 ) and multiply (9.6) by λ −n to
1 n
obtain

{a(x0 + λ y)Dv(y) · Dv(y) + 2λ b(x0 + λ y)v(y) · Dv(y)
B1 (0)
+ λ 2 c(x0 + λ y)v(y) · v(y)} dy > 0,
αβ β
where a = {ai j }, b = {b jk } and c = {c jk }. Since a, b and c are continuous, we have
a(x0 + λ y)Dv(y) · Dv(y) → a(x0 )Dv(y) · Dv(y),
λ b(x0 + λ y)v(y) · Dv(y) → 0,

and
λ 2 c(x0 + λ y)v(y) · v(y) → 0,
uniformly on Ω̄ as λ → 0. Thus this limit gives us

αβ
f˜jk Dα v j Dβ vk dx ≥ 0, ∀v ∈ Cc∞ (B1 (0); RN ), (9.7)
B1 (0)
where
αβ αβ
f˜jk = a jk (x0 ) = fξαi ξ k (x0 , u(x0 ), ∇u(x0 )).
β
Now define v = (v1 , . . . , vN ), where

v j = ρ j cos((η · y)t)ζ (y)
ρ = (ρ 1 , . . . , ρ N ) ∈ RN
and
η = (η1 , . . . , ηn ) ∈ Rn
and ζ ∈ Cc∞ (B1 (0)). From (9.7) we obtain

αβ
0 ≤ f˜jk ρ j ρ k (ηα t(−sin((η · y)t)ζ + cos((η · y)t)Dα ζ )
B1 (0)

· ηβ t(−sin((η · y)t)ζ + cos((η · y)t)Dβ ζ dy (9.8)
By analogy for
v j = ρ j sin((η · y)t)ζ (y)
we obtain

αβ
0 ≤ f˜jk ρ j ρ k (ηα t(cos((η · y)t)ζ + sin((η · y)t)Dα ζ )
B1 (0)

· ηβ t(cos((η · y)t)ζ + sin((η · y)t)Dβ ζ dy (9.9)
Summing up these last two equations, dividing the result by t 2 , and letting t → +∞
we obtain
αβ
0 ≤ f˜jk ρ j ρ k ηα ηβ ζ 2 dy,
B1 (0)
for all ζ ∈ Cc∞ (B1 (0)), which implies

αβ
0 ≤ f˜jk ρ j ρ k ηα ηβ .
9.7 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 9.7.1. We say that u ∈ Ĉ1 ([a, b]; RN ) if u : [a, b] → RN is continuous in
[a, b] and Du is continuous except on a finite set of points in [a, b].
Theorem 9.7.2 (Weierstrass). Let Ω = (a, b) and f : Ω̄ × RN × RN → R be such

that fs (x, s, ξ ) and fξ (x, s, ξ ) are continuous on Ω̄ × RN × RN .
Define F : U → R by
b
F(u) = f (x, u(x), u (x)) dx,
a
9.7 The Weierstrass Condition for n = 1 237
where
U = {u ∈ Ĉ1 ([a, b]; RN ) | u(a) = α , u(b) = β }.
Suppose u ∈ U minimizes locally F on U, that is, suppose that there exists ε0 > 0
such that
F(u) ≤ F(v), ∀v ∈ U, such that u − v ∞ < ε0 .
Under such hypotheses, we have
E(x, u(x), u (x+), w) ≥ 0, ∀x ∈ [a, b], w ∈ RN ,
and
E(x, u(x), u (x−), w) ≥ 0, ∀x ∈ [a, b], w ∈ RN ,
where
u (x+) = lim u (x + h),
h→0+
u (x−) = lim u (x + h),

h→0−
and
E(x, s, ξ , w) = f (x, s, w) − f (x, s, ξ ) − fξ (x, s, ξ )(w − ξ ).
Remark 9.7.3. The function E is known as the Weierstrass excess function.
Proof. Fix x0 ∈ (a, b) and w ∈ RN . Choose 0 < ε < 1 and h > 0 such that u + v ∈ U
and
v ∞ < ε0
where v(x) is given by
⎧
⎨ (x − x0 )w, if 0 ≤ x − x0 ≤ ε h,
v(x) = ε̃ (h − x + x0)w, if ε h ≤ x − x0 ≤ h,
⎩
0, otherwise,
where
ε
ε̃ = .
1−ε
From
F(u + v) − F(u) ≥ 0
we obtain
x0 +h
f (x, u(x) + v(x), u (x) + v (x)) dx
x0
x0 +h
− f (x, u(x), u (x)) dx ≥ 0. (9.10)
x0
Define
x − x0
x̃ = ,
h
so that
dx
d x̃ = .
h
1
h f (x0 + x̃h, u(x0 + x̃h) + v(x0 + x̃h), u (x0 + x̃h) + v(x0 + x̃h) d x̃
0
1
−h f (x0 + x̃h, u(x0 + x̃h), u (x0 + x̃h)) d x̃ ≥ 0. (9.11)
0
where the derivatives are related to x.

Therefore
ε
f (x0 + x̃h, u(x0 + x̃h) + v(x0 + x̃h), u (x0 + x̃h) + w) d x̃
0
ε
− f (x0 + x̃h, u(x0 + x̃h), u (x0 + x̃h)) d x̃
0
1
+ f (x0 + x̃h, u(x0 + x̃h) + v(x0 + x̃h), u (x0 + x̃h) − ε̃ w) d x̃
ε
1
− f (x0 + x̃h, u(x0 + x̃h), u (x0 + x̃h)) d x̃
ε
≥ 0. (9.12)
Letting h → 0 we obtain
ε ( f (x0 , u(x0 ), u (x0 +) + w) − f (x0, u(x0 ), u (x0 +))

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

−(1 − ε )ε̃ ( fξ (x0 , u(x0 ), u (x0 +) + ρ (ε̃ )w)) · w ≥ 0. (9.13)
Dividing by ε and letting ε → 0, so that ε̃ → 0 and ρ (ε̃ ) → 0, we finally obtain
f (x0 , u(x0 ), u (x0 +) + w) − f (x0, u(x0 ), u (x0 +))

− fξ (x0 , u( x0 ), u (x0 +)) · w ≥ 0.
Similarly we may get
f (x0 , u(x0 ), u (x0 −) + w) − f (x0, u(x0 ), u (x0 −))

− fξ (x0 , u( x0 ), u (x0 −)) · w ≥ 0.
Since x0 ∈ [a, b] and w ∈ RN are arbitrary, the proof is complete.

9.8 The Weierstrass Condition: The General Case 239
9.8 The Weierstrass Condition: The General Case
In this section we present a proof for the Weierstrass necessary condition for
N ≥ 1, n ≥ 1. Such a result may be found in similar form in [37].
Theorem 9.1. Assume u ∈ C1 (Ω ; RN ) is a point of strong minimum for a Fréchet

differentiable functional F : U → R that is, in particular, there exists ε > 0 such that
F(u + ϕ ) ≥ F(u),
for all ϕ ∈ Cc∞ (Ω ; Rn ) such that
ϕ ∞ < ε .
Here
F(u) = f (x, u, Du) dx,
Ω
where we recall to have denoted

∂ ui
Du = ∇u = .
∂xj
Under such hypotheses, for all x ∈ Ω and each rank-one matrix η = {ρi β α } =
{ρ ⊗ β }, we have that
E(x, u(x), Du(x), Du(x) + ρ ⊗ β ) ≥ 0,
where
E(x, u(x), Du(x), Du(x) + ρ ⊗ β )

= f (x, u(x), Du(x) + ρ ⊗ β ) − f (x, u(x), Du(x))
−ρ i βα fξαi (x, u(x), Du(x)). (9.14)
Proof. Since u is a point of local minimum for F, we have that
δ F(u; ϕ ) = 0, ∀ϕ ∈ Cc∞ (Ω ; RN ),
that is,
(ϕ · fs (x, u(x), Du(x)) + Dϕ · fξ (x, u(x), Du(x)) dx = 0,
Ω
and hence,

( f (x, u(x), Du(x) + Dϕ (x)) − f (x, u(x), Du(x)) dx
Ω

− (ϕ (x) · fs (x, u(x), Du(x)) − Dϕ (x) · fξ (x, u(x), Du(x)) dx
Ω
≥ 0, (9.15)
∀ϕ ∈ V , where
V = {ϕ ∈ Cc∞ (Ω ; RN ) : ϕ ∞ < ε }.
Choose a unit vector e ∈ Rn and write
x = (x · e)e + x,
where
x · e = 0.
Denote De v = Dv · e and let ρ = (ρ1 , . . . ., ρN ) ∈ RN .
Also, let x0 be any point of Ω . Without loss of generality assume x0 = 0.
Choose λ0 ∈ (0, 1) such that Cλ0 ⊂ Ω , where
Cλ0 = {x ∈ Rn : |x · e| ≤ λ0 and x ≤ λ0 }.
Let λ ∈ (0, λ0 ) and

φ ∈ Cc ((−1, 1); R)
and choose a sequence
φk ∈ Cc∞ ((−λ 2 , λ ); R)
which converges uniformly to the Lipschitz function φλ given by
⎧
⎨ t + λ 2, if − λ 2 ≤ t ≤ 0,
φλ = λ (λ − t), if 0 < t < λ (9.16)
⎩
0, otherwise
and such that φk converges uniformly to φλ on each compact subset of
Aλ = {t : −λ 2 < t < λ , t = 0}.
We emphasize the choice of {φk } may be such that for some K > 0 we have φ ∞ <
K, φk ∞ < K and φk ∞ < K, ∀k ∈ N.
Observe that for any sufficiently small λ > 0 we have that ϕk defined by
ϕk (x) = ρφk (x · e)φ (|x|2 /λ 2 ) ∈ V , ∀k ∈ N
so that letting k → ∞ we obtain that
ϕ (x) = ρφλ (x · e)φ (|x|2 /λ 2 ),
is such that (9.15) is satisfied.

Moreover,
De ϕ (x) = ρφλ (x · e)φ (|x|2 /λ 2 ),
and
Dϕ (x) = ρφλ (x · e)φ (|x|2 /λ 2 )2λ −2 x,
where D denotes the gradient relating the variable x.
9.8 The Weierstrass Condition: The General Case 241
Note that for such a ϕ (x), the integrand of (9.15) vanishes if x ∈ Cλ , where
Cλ = {x ∈ Rn : |x · e| ≤ λ and x ≤ λ }.
Define Cλ+ and Cλ− by
Cλ− = {x ∈ Cλ : x · e ≤ 0},
and
Cλ+ = {x ∈ Cλ : x · e > 0}.
Hence, denoting
gk (x) = ( f (x, u(x), Du(x) + Dϕk (x)) − f (x, u(x), Du(x))

−(ϕk (x) · fs (x, u(x), Du(x) + Dϕk (x) · fξ (x, u(x), Du(x)) (9.17)
and
g(x) = ( f (x, u(x), Du(x) + Dϕ (x)) − f (x, u(x), Du(x))

−(ϕ (x) · fs (x, u(x), Du(x) + Dϕ (x) · fξ (x, u(x), Du(x)) (9.18)
letting k → ∞, using the Lebesgue dominated converge theorem, we obtain

−
gk (x) dx + gk (x) dx
Cλ Cλ+

→ −
g(x) dx + g(x) dx ≥ 0, (9.19)
Cλ Cλ+
Now define
y = ye e + y,
where
x·e
ye = ,
λ2
and
x
y= .
λ
The sets Cλ− and Cλ+ correspond, concerning the new variables, to the sets B−
λ and
B+
λ, where
B−λ = {y : y ≤ 1, and − λ
−1
≤ ye ≤ 0},
B+ −1
λ = {y : y ≤ 1, and 0 < y ≤ λ }.
e
Therefore, since dx = λ n+1 dy, multiplying (9.19) by λ −n−1 , we obtain

−
g(x(y)) dy + g(x(y)) dy + g(x(y)) dy ≥ 0, (9.20)
B1 B−
λ
−
\B1 B+
λ
where
x = (x · e)e + x = λ 2 ye + λ y ≡ x(y).
Observe that
⎧
⎨ ρφ ( y 2 ) if − 1 ≤ ye ≤ 0,
De ϕ (x) = ρφ ( y )(−λ ) if 0 ≤ ye ≤ λ −1 ,
2 (9.21)
⎩
0, otherwise.
Observe also that !

|g(x(y))| ≤ o( |ϕ (x)|2 + |Dϕ (x)|2 ),
so that from the expression of ϕ (x) and Dϕ (x) we obtain, for
y ∈ B+ − −
λ , or y ∈ Bλ \ B1 ,
that
|g(x(y))| ≤ o(λ ), as λ → 0.
Since the Lebesgue measures of B− +
λ and Bλ are bounded by
2n−1 /λ
the second and third terms in (9.20) are of o(1) where
lim o(1)/λ = 0,
λ →0+
so that letting λ → 0+ , considering that
x(y) → 0,
and on B−
1 (up to the limit set B)
g(x(y)) → f (0, u(0), Du(0) + ρφ ( y 2 )e)

− f (0, u(0), Du(0)) −
ρφ ( y 2 )e fξ (0, u(0), Du(0)) (9.22)
we get

[ f (0, u(0), Du(0) + ρφ ( y 2 )e) − f (0, u(0), Du(0))
B
−ρφ ( y 2 )e fξ (0, u(0), Du(0))] dy2 . . . dyn
≥ 0, (9.23)
where B is an appropriate limit set (we do not provide more details here) such that
B = {y ∈ Rn : ye = 0 and y ≤ 1}.
9.9 The du Bois–Reymond Lemma 243
Here we have used the fact that on the set in question,
Dϕ (x) → ρφ ( y 2 )e, as λ → 0+ .
Finally, inequality (9.23) is valid for a sequence {φn } (in place of φ ) such that
0 ≤ φn ≤ 1 and φn (t) = 1, if |t| < 1 − 1/n,
∀n ∈ N.
Letting n → ∞, from (9.23), we obtain
f (0, u(0), Du(0) + ρ ⊗ e) − f (0, u(0), Du(0))

−ρ · e fξ (0, u(0), Du(0)) ≥ 0. (9.24)
9.9 The du Bois–Reymond Lemma
We present now a simpler version of the fundamental lemma of calculus of varia-

tions. The result is specific for n = 1 and is known as the du Bois–Reymond lemma.
Lemma 9.9.1 (du Bois–Reymond). If u ∈ C([a, b]) and

b
uϕ dx = 0, ∀ϕ ∈ V ,
a
where
V = {ϕ ∈ C1 [a, b] | ϕ (a) = ϕ (b) = 0},
then there exists c ∈ R such that
u(x) = c, ∀x ∈ [a, b].
Proof. Define
b
1
c= u(t) dt,
b−a a
and x
ϕ (x) = (u(t) − c) dt.
a
Thus we have ϕ (a) = 0 and
b
ϕ (b) = u(t) dt − c(b − a) = 0.
a
Moreover ϕ ∈ C1 ([a, b]) so that

ϕ ∈V.
Therefore
b
0≤ (u(x) − c)2 dx
a
b
= (u(x) − c)ϕ (x) dx
a
b
= u(x)ϕ (x) dx − c[ϕ (x)]ba = 0. (9.25)
a
Thus b
(u(x) − c)2 dx = 0,
a
and being u(x) − c continuous, we finally obtain
u(x) − c = 0, ∀x ∈ [a, b].
Proposition 9.9.2. If u, v ∈ C([a, b]) and

b
(u(x)ϕ (x) + v(x)ϕ (x)) dx = 0,
a
∀ϕ ∈ V , where
V = {ϕ ∈ C1 [a, b] | ϕ (a) = ϕ (b) = 0},
then
v ∈ C1 ([a, b])
and
v (x) = u(x), ∀x ∈ [a, b].
Proof. Define x
u1 (x) = u(t) dt, ∀x ∈ [a, b].
a
Thus u1 ∈ C1 ([a, b]) and
u1 (x) = u(x), ∀x ∈ [a, b].
Hence, for ϕ ∈ V , we have

b
0= (u(x)ϕ (x) + v(x)ϕ (x) dx
a
b
= (−u1 (x)ϕ (x) + vϕ (x)) dx + [u1(x)ϕ (x)]ba
a
b
= (v(x) − u1 (x))ϕ (x) dx. (9.26)
a
9.10 The Weierstrass–Erdmann Conditions 245
That is,
b
(v(x) − u1 (x))ϕ (x) dx, ∀ϕ ∈ V .
a
By the du Bois–Reymond lemma, there exists c ∈ R such that
v(x) − u1(x) = c, ∀x ∈ [a, b].
Hence
v = u1 + c ∈ C1 ([a, b]),
so that
v (x) = u1 (x) = u(x), ∀x ∈ [a, b].
9.10 The Weierstrass–Erdmann Conditions
We start with a definition.

Definition 9.10.1. Define I = [a, b]. A function u ∈ Ĉ1 ([a, b]; RN ) is said to be a
weak extremal of b
F(u) = f (x, u(x), u (x)) dx,
a
if
b
( fs (x, u(x), u (x)) · ϕ + fξ (x, u(x), u (x)) · ϕ (x)) dx = 0,
a
∀ϕ ∈ Cc∞ ([a, b]; RN ).
Proposition 9.10.2. For any weak extremal of

b
F(u) = f (x, u(x), u (x)) dx
a
there exists a constant c ∈ RN such that

x
fξ (x, u(x), u (x)) = c + fs (t, u(t), u (t)) dt, ∀x ∈ [a, b]. (9.27)
a
Proof. Fix ϕ ∈ Cc∞ ([a, b]; RN ). Integration by parts of the extremal condition
δ F(u, ϕ ) = 0,
implies that
b
fξ (x, u(x), u (x)) · ϕ (x) dx
a
b x
− fs (t, u(t), u (t)) dt · ϕ (x) dx = 0.
a a
Since ϕ is arbitrary, considering the du Bois-Reymond lemma is valid also for u ∈

L1 ([a, b]) and the respective N-dimensional version (see [37], page 32 for details),
there exists, c ∈ RN such that
x
fξ (x, u(x), u (x)) − fs (t, u(t), u (t)) dt = c, ∀x ∈ [a, b].
a
Theorem 9.10.3 (Weierstrass–Erdmann Corner Conditions). Let I = [a, b]. Sup-

pose u ∈ Ĉ1 ([a, b]; RN ) is such that
F(u) ≤ F(v), ∀v ∈ Cr ,
for some r > 0 where
Cr = {v ∈ Ĉ1 ([a, b]; RN ) | v(a) = u(a), v(b) = u(b),

and u − v ∞ < r}.
Let x0 ∈ (a, b) be a corner point of u. Denoting u0 = u(x0 ), ξ0+ = u (x0 + 0), and
ξ0− = u (x0 − 0), then the following relations are valid:
1. fξ (x0 , u0 , ξ0− ) = fξ (x0 , u0 , ξ0+ ),
2.
f (x0 , u0 , ξ0− ) − ξ0− fξ (x0 , u0 , ξ0− )

= f (x0 , u0 , ξ0+ ) − ξ0+ fξ (x0 , u0 , ξ0+ ).
Remark 9.10.4. The conditions above are known as the Weierstrass–Erdmann corner
conditions.
Proof. Condition (1) is just a consequence of (9.27). For (2), define
τε (x) = x + ελ (x),
where λ ∈ Cc∞ (I). Observe that τε (a) = a and τε (b) = b, ∀ε > 0. Also τ0 (x) = x.
Choose ε0 > 0 sufficiently small such that for each ε satisfying |ε | < ε0 , we have
τε (x) > 0 and
ũε (x) = (u ◦ τε−1)(x) ∈ Cr .
9.10 The Weierstrass–Erdmann Conditions 247
Define
φ (ε ) = F(x, ũε , ũε (x)).
Thus φ has a local minimum at 0, so that φ (0) = 0, that is,
d(F(x, ũε , ũε (x)))

|ε =0 = 0.
dε
Observe that
d ũε d τ −1 (x)
= u (τε−1 (x)) ε ,
dx dx
and
d τε−1 (x) 1
= .
dx 1 + ελ (τε−1 (x))
Thus,
b
1
F(ũε ) = f x, u(τε−1 (x)), u (τε−1 (x)) dx.
a 1 + ελ (τε−1 (x))
Defining
x̄ = τε−1 (x),
we obtain
1
d x̄ = dx,
1 + ελ (x̄)
that is,
dx = (1 + ελ (x̄)) d x̄.
Dropping the bar for the new variable, we may write
b

u (x)
F(ũε ) = f x + ελ (x), u(x), 1 + ελ (x) dx.
a 1 + ελ (x)

From
dF(ũε )
|ε =0 ,
dε
we obtain
b
(λ fx (x, u(x), u (x)) + λ (x)( f (x, u(x), u (x))
a
− u (x) fξ (x, u(x), u (x)))) dx = 0. (9.28)
Since λ is arbitrary, from Proposition 9.9.2, (in fact from its version for u ∈ L1 ([a, b])
and respective extension for the N dimensional case, please see [37] for details), we
obtain
x
f (x, u(x), u (x)) − u (x) fξ (x, u(x), u (x)) − fx (t, u(t), u (t)) dt = c1
a
for some c1 ∈ RN .

Since ax fx (t, u(t), u (t)) dt + c1 is a continuous function (in fact absolutely con-
tinuous), the proof is complete.
9.11 Natural Boundary Conditions
Consider the functional f : U → R, where

F(u) f (x, u(x), ∇u(x)) dx,
Ω
f (x, s, ξ ) ∈ C1 (Ω̄ , RN , RN×n ),

and Ω ⊂ Rn is an open bounded connected set.
Proposition 9.11.1. Assume
U = {u ∈ W 1,2 (Ω ; RN ); u = u0 on Γ0 },
where Γ0 ⊂ ∂ Ω is closed and ∂ Ω = Γ = Γ0 ∪ Γ1 being Γ1 open in Γ and Γ0 ∩ Γ1 = 0.

/
Thus if ∂ Ω ∈ C1 , f ∈ C2 (Ω̄ , RN , RN×n ) and u ∈ C2 (Ω̄ ; RN ), and also
δ F(u, ϕ ) = 0, ∀ϕ ∈ C1 (Ω̄ ; RN ), such that ϕ = 0 on Γ0 ,
then u is a extremal of F which satisfies the following natural boundary conditions:
nα fξαi (x, u(x)∇u(x)) = 0, a.e. on Γ1 , ∀i ∈ {1, . . . , N}.
Proof. Observe that δ F(u, ϕ ) = 0, ∀ϕ ∈ Cc∞ (Ω ; RN ); thus u is an extremal of F and

through integration by parts and the fundamental lemma of calculus of variations,
we obtain
L f (u) = 0, in Ω ,
where
L f (u) = fs (x, u(x), ∇u(x)) − div( fξ (x, u(x), ∇u(x)).
Defining
V = {ϕ ∈ C1 (Ω ; RN ) | ϕ = 0 on Γ0 },
for an arbitrary ϕ ∈ V , we obtain

δ F(u, ϕ ) = L f (u) · ϕ dx
Ω

+ nα fξαi (x, u(x), ∇u(x))ϕ i (x) d Γ
Γ1

= nα fξαi (x, u(x), ∇u(x))ϕ i (x) d Γ
Γ1
= 0, ∀ϕ ∈ V . (9.29)
9.11 Natural Boundary Conditions 249
Suppose, to obtain contradiction, that
nα fξαi (x0 , u(x0 ), ∇u(x0 )) = β > 0,
for some x0 ∈ Γ1 and some i ∈ {1, . . ., N}. Defining
G(x) = nα fξαi (x, u(x), ∇u(x)),
by the continuity of G, there exists r > 0 such that
G(x) > β /2, in Br (x0 ),
and in particular
G(x) > β /2, in Br (x0 ) ∩ Γ1 .
Choose 0 < r1 < r such that Br1 (x0 ) ∩ Γ0 = 0.
/ This is possible since Γ0 is closed and
x0 ∈ Γ1 .
Choose ϕ i ∈ Cc∞ (Br1 (x0 )) such that ϕ i ≥ 0 in Br1 (x0 ) and ϕ i > 0 in Br1 /2 (x0 ).
Therefore
β
G(x)ϕ i (x) dx > ϕ i dx > 0,
Γ1 2 Γ1
and this contradicts (9.29). Thus
G(x) ≤ 0, ∀x ∈ Γ1 ,
and by analogy
G(x) ≥ 0, ∀x ∈ Γ1 ,
so that
G(x) = 0, ∀x ∈ Γ1 .
Chapter 10
Basic Concepts on Convex Analysis
For this chapter the most relevant reference is Ekeland and Temam, [25].
10.1 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
λ u + (1 − λ )v ∈ S, ∀λ ∈ [0, 1]. (10.1)
Definition 10.1.1 (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
Co(S) = ∑ λi ui | n ∈ N,
i=1

n
∑ λi = 1, λi ≥ 0, ui ∈ S, ∀i ∈ {1, . . . , n} . (10.2)
i=1
Definition 10.1.2 (Convex Functional). Let S be convex subset of the vector space
U. A functional F : S → R̄ = R ∪ {+∞, −∞} is said to be convex if
F(λ u + (1 − λ )v) ≤ λ F(u) + (1 − λ )F(v), ∀u, v ∈ S, λ ∈ [0, 1]. (10.3)
10.1.1 Lower Semicontinuity
We start with the definition of epigraph.
252 10 Basic Concepts on Convex Analysis
Definition 10.1.3 (Epigraph). Given F : U → R̄ we define its epigraph, denoted by

E pi(F) as
E pi(F) = {(u, a) ∈ U × R | a ≥ F(u)}.
Definition 10.1.4. Let U be a Banach space. Consider the weak topology σ (U,U ∗ )
and let F : U → R ∪ {+∞}. Such a function is said to be weakly lower semicontinu-
ous if ∀λ such that λ < F(u), there exists a weak neighborhood Vλ (u) ∈ σ (U,U ∗ )
such that
F(v) > λ , ∀v ∈ Vλ (u).
Theorem 10.1.5. Let U be a Banach space and let F : U → R ∪ {+∞}. The follow-
ing statements are equivalent:
1. F is weakly lower semicontinuous (w-l.s.c.).
2. E pi(F) is closed in U × R with the product topology between σ (U,U ∗ ) and the
usual topology in R.
3. HγF = {u ∈ U | F(u) ≤ γ } is closed in σ (U,U ∗ ), ∀γ ∈ R.
4. The set GFγ = {u ∈ U | F(u) > γ } is open in σ (U,U ∗ ), ∀γ ∈ R.
5.
lim inf F(v) ≥ F(u), ∀u ∈ U.
vu
Proof. Assume that F is w-l.s.c..We will show that E pi(F)c is open in σ (U,U ∗ )
product with the usual topology in R. Choose (u, r) ∈ E pi(F)c . Then (u, r) ∈
E pi(F), so that r < F(u). Select λ such that r < λ < F(u). Since F is w-l.s.c.,
there exists a weak neighborhood Vλ (u) such that
F(v) > λ , ∀v ∈ Vλ (u).
Thus
Vλ × (−∞, λ ) ⊂ E pi(F)c
so that (u, r) is an interior point of E pi(F)c , and hence, since such a point in E pi(F)c
is arbitrary, we may conclude that E pi(F)c is open so that E pi(F) is closed in
σ (U,U ∗ ) product with the usual topology in R.
Now assume (2). Observe that
HγF × {γ } = E pi(F) ∩ (U × {γ }).
Since from the hypothesis E pi(F) is closed, we have that HγF × {γ } is closed and
hence HγF is closed.
Now assume (3). To obtain (4) just take the complement of HγF . Suppose (4) is
valid. Let γ ∈ R such that
γ < F(u).
Since GFγ is open in σ (U,U ∗ ) there exists a weak neighborhood V (u) such that
V (u) ⊂ GFγ ,
10.1 Convex Sets and Convex Functions 253
so that
F(v) > γ , ∀v ∈ V (u),
and hence
inf F(v) ≥ γ ,
v∈V (u)
and hence in particular

lim inf F(v) ≥ γ .
vu
Letting γ → F(u), we get

lim inf F(v) ≥ F(u).
vu
Finally assume that

lim inf F(v) ≥ F(u).
vu
Let λ < F(u). Thus there exists a weak neighborhood V (u) such that F(v) ≥ F(u) >
λ , ∀v ∈ V (u). The proof is complete.
Similar result is valid for the strong topology of the Banach space U so that a func-
tional F : U → R ∪ {+∞} is strongly lower semicontinuous (l.s.c.) at u ∈ U, if
lim inf F(v) ≥ F(u). (10.4)

v→u
Corollary 10.1.6. Every convex l.s.c. function F : U → R is also w-l.s.c. (weakly

lower semicontinuous).
Proof. The result follows from the fact that the epigraph of F is convex and closed
convex sets are weakly closed.
Definition 10.1.7 (Affine Continuous Function). Let U be a Banach space. A func-
tional F : U → R is said to be affine continuous if there exist u∗ ∈ U ∗ and α ∈ R
such that
F(u) = u, u∗ U + α , ∀u ∈ U. (10.5)
Definition 10.1.8 (Γ (U)). Let U be a Banach space. We say that F : U → R̄ belongs

to Γ (U) and write F ∈ Γ (U) if F can be represented as the point-wise supremum of
a family of affine continuous functions. If F ∈ Γ (U) and F(u) ∈ R for some u ∈ U,
then we write F ∈ Γ0 (U).
The next result is proven in [25].
Proposition 10.1.9. Let U be a Banach space, then F ∈ Γ (U) if and only if F is
convex and l.s.c., and if F takes the value −∞, then F ≡ −∞.
Definition 10.1.10 (Convex Envelope). Let U be a Banach space. Given F : U → R̄,
we define its convex envelope, denoted by CF : U → R̄ by
CF(u) = sup {u, u∗ + α }, (10.6)

(u∗ ,α )∈A∗
where
A∗ = {(u∗ , α ) ∈ U ∗ × R | v, u∗ U + α ≤ F(v), ∀v ∈ U} (10.7)
Definition 10.1.11 (Polar Functionals). Given F : U → R̄, we define the related

polar functional, denoted by F ∗ : U ∗ → R̄, by
F ∗ (u∗ ) = sup{u, u∗ U − F(u)}, ∀u∗ ∈ U ∗ . (10.8)

u∈U
Definition 10.1.12 (Bipolar Functional). Given F : U → R̄, we define the related

bipolar functional, denoted by F ∗∗ : U → R̄, as
F ∗∗ (u) = sup {u, u∗ U − F ∗ (u∗ )}, ∀u ∈ U. (10.9)

u∗ ∈U ∗
Proposition 10.1.13. Given F : U → R̄, then F ∗∗ (u) = CF(u) and in particular if

F ∈ Γ (U), then F ∗∗ (u) = F(u).
Proof. By definition, the convex envelope of F is the supremum of all affine contin-
uous minorants of F. We can consider only the maximal minorants, which functions
of the form
u → u, u∗ U − F ∗ (u∗ ). (10.10)
Thus,
CF(u) = sup {u, u∗U − F ∗ (u∗ )} = F ∗∗ (u). (10.11)

u∗ ∈U ∗
Corollary 10.1.14. Given F : U → R̄, we have F ∗ = F ∗∗∗ .
Proof. Since F ∗∗ ≤ F we obtain
F ∗ ≤ F ∗∗∗ . (10.12)
On the other hand, we have
F ∗∗ (u) ≥ u, u∗ U − F ∗ (u∗ ), (10.13)
so that
F ∗∗∗ (u∗ ) = sup{u, u∗U − F ∗∗ (u)} ≤ F ∗ (u∗ ). (10.14)

u∈U
From (10.12) and (10.14) we obtain F ∗ (u∗ ) = F ∗∗∗ (u∗ ).
Here we recall the definition of Gâteaux differentiability.

Definition 10.1.15 (Gâteaux Differentiability). A functional F : U → R̄ is said to
be Gâteaux differentiable at u ∈ U if there exists u∗ ∈ U ∗ such that
F(u + λ h) − F(u)
lim = h, u∗ U , ∀h ∈ U. (10.15)
λ →0 λ
The vector u∗ is said to be the Gâteaux derivative of F : U → R at u and may be
denoted as follows:
∂ F(u)
u∗ = or u∗ = δ F(u) (10.16)
∂u
Definition 10.1.16 (Sub-gradients). Given F : U → R̄, we define the set of sub-
gradients of F at u, denoted by ∂ F(u) as
∂ F(u) = {u∗ ∈ U ∗ , such that v − u, u∗ U + F(u) ≤ F(v), ∀v ∈ U}. (10.17)
Here we recall the definition of adjoint operator.

Definition 10.1.17 (Adjoint Operator). Let U and Y be Banach spaces and Λ :
U → Y a continuous linear operator. The adjoint operator related to Λ , denoted by
Λ ∗ : Y ∗ → U ∗ , is defined through the equation
u, Λ ∗ v∗ U = Λ u, v∗ Y , ∀u ∈ U, v∗ ∈ Y ∗ . (10.18)
Lemma 10.1.18 (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 = θ and
F(u) = 0. Let V be a neighborhood of origin such that F(v) ≤ a < +∞, ∀v ∈ V .
Define W = V ∩ (−V ) (which is a symmetric neighborhood of origin). Pick ε ∈
(0, 1). If v ∈ ε W , since F is convex and
v
∈V (10.19)
ε
we may infer that
F(v) ≤ (1 − ε )F(θ ) + ε F(v/ε ) ≤ ε a. (10.20)
Also
−v
∈V. (10.21)
ε
Thus,
F(v) ε
F(θ ) ≤ + F(−v/ε ),
1+ε 1+ε
so that
F(v) ≥ (1 + ε )F(θ ) − ε F(−v/ε ) ≥ −ε a. (10.22)

Therefore
|F(v)| ≤ ε a, ∀v ∈ ε W , (10.23)
that is, F is continuous at u = θ .
Proposition 10.1.19. Let F : U → R̄ be a convex function finite and continuous at

u ∈ U. Then ∂ F(u) = 0.
/
Proof. Since F is convex, E pi(F) is convex, as F is continuous at u, we have that

E pi(F) is nonempty. Observe that (u, F(u)) belongs to the boundary of E pi(F), so
that denoting A = E pi(F), we may separate (u, F(u)) from Å by a closed hyperplane
H, which may be written as
H = {(v, a) ∈ U × R | v, u∗ U + α a = β }, (10.24)
for some fixed α , β ∈ R and u∗ ∈ U ∗ , so that
v, u∗ U + α a ≥ β , ∀(v, a) ∈ E pi(F), (10.25)
and
u, u∗ U + α F(u) = β , (10.26)
where (α , β , u∗ ) = (0, 0, θ ). Suppose α = 0. Thus we have
v − u, u∗U ≥ 0, ∀v ∈ U, (10.27)
and thus we obtain u∗ = θ and β = 0. Therefore we may assume α > 0 (consider-

ing (10.25)) so that ∀v ∈ U we have
β
− v, u∗/α U ≤ F(v), (10.28)
α
and
β
− u, u∗/α U = F(u), (10.29)
α
or
v − u, −u∗/α U + F(u) ≤ F(v), ∀v ∈ U, (10.30)
so that
−u∗ /α ∈ ∂ F(u). (10.31)

Definition 10.1.20 (Carathéodory Mapping ). Let S ⊂ Rn be an open set. We say

that g : S × Rl → R is a Carathéodory mapping if
∀ξ ∈ Rl , x → g(x, ξ ) is a measurable function,
and
for almost all x ∈ S, ξ → g(x, ξ ) is a continuous function.
The proof of next results may be found in Ekeland and Temam [25].
Proposition 10.1.21. Let E and F be two Banach spaces, S a Borel subset of Rn and
g : S × E → F a Carathéodory mapping. For each measurable function u : S → E,
let G1 (u) be the measurable function x → g(x, u(x)) ∈ F.
If G1 maps L p (S, E) into Lr (S, F) for 1 ≤ p, r < ∞, then G1 is continuous in the
norm topology.

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 special case of the more general hypothesis presented in
[25]) we have the following result.
Proposition 10.1.22. Considering the last proposition we can express G∗ : U ∗ →
R̄ as

G∗ (u∗ ) = g∗ (x, u∗ (x))dS, (10.32)
S
where g∗ (x, y) = sup (y · η − g(x, η )), almost everywhere in S.

η ∈Rl
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 10.1.23 (Legendre’s Transform and Associated Functional). Con-
sider a differentiable function g : Rn → R. Its Legendre transform, denoted by
g∗L : RnL → R, is expressed as
g∗L (y∗ ) = x0i · y∗i − g(x0), (10.33)
where x0 is the solution of the system:
∂ g(x0 )
y∗i = , (10.34)
∂ xi
and RnL = {y∗ ∈ Rn such that (10.34) has a unique solution}.
Furthermore, considering the functional G : Y → R defined as G(v) = S g(v)dS,
we define the associated Legendre transform functional, denoted by G∗L : YL∗ → R as

G∗L (v∗ ) = g∗L (v∗ )dS, (10.35)
S
where YL∗ = {v∗ ∈ Y ∗ | v∗ (x) ∈ RnL , a.e. in S}.

About the Legendre transform we still have the following results:

Proposition 10.1.24. Considering the last definitions, suppose that for each y∗ ∈ RnL
at least in a neighborhood (of y∗ ) it is possible to define a differentiable function by
the expression
∂ g −1 ∗
x0 (y∗ ) = [ ] (y ). (10.36)
∂x
Then, ∀ i ∈ {1, . . . , n} we may write
∂ g(x0 ) ∂ g∗ (y∗ )
y∗i = ⇔ x0 i = L ∗ (10.37)
∂ xi ∂ yi
Proof. Suppose firstly that
∂ g(x0 )
y∗i = , ∀ i ∈ {1, . . . , n}, (10.38)
∂ xi
thus,
g∗L (y∗ ) = y∗i x0i − g(x0 ) (10.39)
and taking derivatives for this expression we have
∂ g∗L (y∗ ) ∂x j ∂ g(x0 ) ∂ x0 j

∗ = y∗j 0∗ + x0i − , (10.40)
∂ yi ∂ yi ∂ x j ∂ y∗i
or
∂ g∗L (y∗ ) ∂ g(x0 ) ∂ x0 j
= (y∗j − ) + x0 i (10.41)
∂ y∗i ∂ x j ∂ y∗i
which from (10.38) implies that
∂ g∗L (y∗ )
= x0i , ∀ i ∈ {1, . . . , n}. (10.42)
∂ y∗i
This completes the first half of the proof. Conversely, suppose now that
∂ g∗L (y∗ )
x0 i = , ∀i ∈ {1, . . . , n}. (10.43)
∂ y∗i
As y∗ ∈ RnL there exists x̄0 ∈ Rn such that
∂ g(x̄0 )
y∗i = ∀i ∈ {1, . . . , n}, (10.44)
∂ xi
and
g∗L (y∗ ) = y∗i x̄0i − g(x̄0 ) (10.45)

and therefore taking derivatives for this expression we can obtain
∂ g∗L (y∗ ) ∂ x̄ j ∂ g(x̄0 ) ∂ x̄0 j

= y∗j 0∗ + x̄0i − , (10.46)
∂ y∗i ∂ yi ∂ x j ∂ y∗i
∀ i ∈ {1, . . . , n}, so that
∂ g∗L (y∗ ) ∂ g(x̄0 ) ∂ x̄0 j

= (y∗j − ) + x̄0i (10.47)
∂ y∗i ∂ x j ∂ y∗i
∀ i ∈ {1, . . . , n}, which from (10.43) and (10.44) implies that
∂ g∗L (y∗ )
x̄0i = = x0i , ∀ i ∈ {1, . . . , n}, (10.48)
∂ y∗i
from this and (10.44) we have
∂ g(x̄0 ) ∂ g(x0 )
y∗i = = ∀ i ∈ {1, . . . , n}. (10.49)
∂ xi ∂ xi
Theorem 10.1.25. Consider the functional J : U → R̄ defined as J(u) = (G ◦

Λ )(u) − u, f U where Λ (= {Λi }) : U → Y (i ∈ {1, . . . , n}) is a continu-
operator and G : Y → R is a functional that can be expressed as
ous linear
G(v) = S g(v)dS, ∀v ∈ Y (here g : Rn → R is a differentiable function that admits
Legendre transform denoted by g∗L : RnL → R. That is, the hypothesis mentioned at
Proposition 10.1.24 is satisfied).
Under these assumptions we have
δ J(u0 ) = θ ⇔ δ (−G∗L (v∗0 ) + u0, Λ ∗ v∗0 − f U ) = θ , (10.50)
∂ G(Λ (u0 ))
where v∗0 = ∂v is supposed to be such that v∗0 (x) ∈ RnL , a.e. in S and in this
case
J(u0 ) = −G∗L (v∗0 ). (10.51)
Proof. Suppose first that δ J(u0 ) = θ , that is,
∂ G(Λ u0 )
Λ∗ − f =θ (10.52)
∂v
∂ G(Λ u0 )
which, as v∗0 = ∂v , implies that
Λ ∗ v∗0 − f = θ , (10.53)
and
∂ g(Λ u0 )
v∗0i = . (10.54)
∂ xi
Thus from the last proposition we can write
∂ g∗L (v∗0 )
Λi (u0 ) = , for i ∈ {1, .., n} (10.55)
∂ y∗i
which means
∂ G∗L (v∗0 )
Λ u0 = . (10.56)
∂ v∗
Therefore from (10.53) and (10.56) we have
δ (−G∗L (v∗0 ) + u0 , Λ ∗ v∗0 − f U ) = θ . (10.57)
This completes the first part of the proof.

Conversely, suppose now that
δ (−G∗L (v∗0 ) + u0 , Λ ∗ v∗0 − f U ) = θ , (10.58)
that is,
Λ ∗ v∗0 − f = θ (10.59)
and
∂ G∗L (v∗0 )
Λ u0 = . (10.60)
∂ v∗
Clearly, from (10.60), the last proposition and (10.59), we can write
∂ G(Λ (u0 ))
v∗0 = (10.61)
∂v
and
∂ G(Λ u0 )
Λ∗ − f = θ, (10.62)
∂v
which implies
δ J(u0 ) = θ . (10.63)
Finally, we have
J(u0 ) = G(Λ u0 ) − u0, f U (10.64)
From this, (10.59) and (10.61), we have
J(u0 ) = G(Λ u0 ) − u0 , Λ ∗ v∗0 U = G(Λ u0 ) − Λ u0 , v∗0 Y (10.65)

= −G∗L (v∗0 ). (10.66)
10.2 Duality in Convex Optimization 261
10.2 Duality in Convex Optimization
Let U be a Banach space. Given F : U → R̄ (F ∈ Γ0 (U)) we define the problem

P as
P : minimize F(u) on U. (10.67)
We say that u0 ∈ U is a solution of problem P if F(u0 ) = infu∈U F(u). Consider a

function φ (u, p) (φ : U × Y → R̄) such that
φ (u, 0) = F(u). (10.68)
We define the problem P ∗ as
P ∗ : maximize − φ ∗ (0, p∗ ) on Y ∗ . (10.69)
Observe that
φ ∗ (0, p∗ ) = sup {0, uU + p, p∗ Y − φ (u, p)} ≥ −φ (u, 0), (10.70)
(u,p)∈U×Y
or
inf {φ (u, 0)} ≥ sup {−φ ∗ (0, p∗ )}. (10.71)

u∈U p∗ ∈Y ∗
Proposition 10.2.1. Consider φ ∈ Γ0 (U × Y ). If we define
h(p) = inf {φ (u, p)}, (10.72)

u∈U
then h is convex.
Proof. We have to show that given p, q ∈ Y and λ ∈ (0, 1), we have
h(λ p + (1 − λ )q) ≤ λ h(p) + (1 − λ )h(q). (10.73)
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) ≤ φ (u, p) ≤ a, (10.74)
and if b > h(q), there exists v ∈ U such that
h(q) ≤ φ (v, q) ≤ b. (10.75)

Thus
h(λ p + (1 − λ )q) ≤ inf {φ (w, λ p + (1 − λ )q)}

w∈U
≤ φ (λ u + (1 − λ )v, λ p + (1 − λ )q) ≤ λ φ (u, p) + (1 − λ )φ (v, q)
≤ λ a + (1 − λ )b. (10.76)
Letting a → h(p) and b → h(q) we obtain
h(λ p + (1 − λ )q) ≤ λ h(p) + (1 − λ )h(q). (10.77)
Proposition 10.2.2. For h as above, we have h∗ (p∗ ) = φ ∗ (0, p∗ ), ∀p∗ ∈ Y ∗ , so that
h∗∗ (0) = sup {−φ ∗ (0, p∗ )}. (10.78)

p∗ ∈Y ∗
Proof. Observe that
h∗ (p∗ ) = sup{p, p∗ Y − h(p)} = sup{p, p∗ Y − inf {φ (u, p)}}, (10.79)

p∈Y p∈Y u∈U
so that
h∗ (p∗ ) = sup {p, p∗ Y − φ (u, p)} = φ ∗ (0, p∗ ). (10.80)

(u,p)∈U×Y
Proposition 10.2.3. The set of solutions of the problem P ∗ (the dual problem) is
identical to ∂ h∗∗ (0).
Proof. Consider p∗0 ∈ Y ∗ a solution of problem P ∗ , that is,
−φ ∗ (0, p∗0 ) ≥ −φ ∗ (0, p∗ ), ∀p∗ ∈ Y ∗ , (10.81)
which is equivalent to
−h∗ (p∗0 ) ≥ −h∗ (p∗ ), ∀p∗ ∈ Y ∗ , (10.82)
− h(p∗0 ) = sup {0, p∗ Y − h∗(p∗ )} ⇔ −h∗ (p∗0 ) = h∗∗ (0)

p∗ ∈Y ∗
⇔ p∗0 ∈ ∂ h∗∗ (0). (10.83)
Theorem 10.2.4. Consider φ : U ×Y → R̄ convex. Assume infu∈U {φ (u, 0)} ∈ R and

there exists u0 ∈ U such that p → φ (u0 , p) is finite and continuous at 0 ∈ Y . Then
inf {φ (u, 0)} = sup {−φ ∗ (0, p∗ )}, (10.84)

u∈U p∗ ∈Y ∗
and the dual problem has at least one solution.

10.2 Duality in Convex Optimization 263
Proof. By hypothesis h(0) ∈ R and as was shown above, h is convex. As the func-
tion p → φ (u0 , p) is convex and continuous at 0 ∈ Y , there exists a neighborhood V
of zero in Y such that
φ (u0 , p) ≤ M < +∞, ∀p ∈ V , (10.85)
for some M ∈ R. Thus, we may write
h(p) = inf {φ (u, p)} ≤ φ (u0 , p) ≤ M, ∀p ∈ V . (10.86)

u∈U
Hence, from Lemma 10.1.18, h is continuous at 0. Thus, by Proposition 10.1.19,

h is sub-differentiable at 0, which means h(0) = h∗∗ (0). Therefore, by Proposi-
tion 10.2.3, the dual problem has solutions and
h(0) = inf {φ (u, 0)} = sup {−φ ∗ (0, p∗ )} = h∗∗ (0). (10.87)
u∈U p∗ ∈Y ∗
Now we apply the last results to φ (u, p) = G(Λ u + p) + F(u), where Λ : U → Y is a

continuous linear operator whose adjoint operator is denoted by Λ ∗ : Y ∗ → U ∗ . We
may enunciate the following theorem.
Theorem 10.2.5. Suppose U is a reflexive Banach space and define J : U → R by
J(u) = G(Λ u) + F(u) = φ (u, 0), (10.88)
where lim J(u) = +∞ as u U → ∞ and F ∈ Γ0 (U), G ∈ Γ0 (Y ). Also suppose there

exists û ∈ U such that J(û) < +∞ with the function p → G(p) continuous at Λ û.
Under such hypothesis, there exist u0 ∈ U and p∗0 ∈ Y ∗ such that
J(u0 ) = min{J(u)} = max

∗ ∗
{−G∗ (p∗ ) − F ∗ (−Λ ∗ p∗ )}
u∈U p ∈Y
∗
= −G (p0 ) − F (−Λ ∗ p∗0 ).
∗ ∗
(10.89)
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 conver-
gent subsequence to some u0 ∈ U. Finally, from the lower semicontinuity of primal
formulation, we may conclude that u0 is a minimizer. The other conclusions follow
from Theorem 10.2.4 just observing that
φ ∗ (0, p∗ ) = sup {p, p∗ Y − G(Λ u + p) − F(u)}

u∈U,p∈Y
= sup {q, p∗ − G(q) − Λ u, p∗ − F(u)}, (10.90)

u∈U,q∈Y
so that
φ ∗ (0, p∗ ) = G∗ (p∗ ) + sup{−u, Λ ∗ p∗ U − F(u)}

u∈U
= G (p ) + F (−Λ ∗ p∗ ).
∗ ∗ ∗
(10.91)
Thus,
inf {φ (u, 0)} = sup {−φ ∗ (0, p∗ )} (10.92)

u∈U p∗ ∈Y ∗
and solutions u0 and p∗0 for the primal and dual problems, respectively, imply that

∗ ∗
{−G∗ (p∗ ) − F ∗ (−Λ ∗ p∗ )}
u∈U p ∈Y
∗
= −G (p0 ) − F (−Λ ∗ p∗0 ).
∗ ∗
(10.93)
10.3 The Min–Max Theorem
Our main objective in this section is to state and prove the min–max theorem.
Definition 10.1. Let U,Y be Banach spaces, A ⊂ U and B ⊂ Y and let L : A × B → R

be a functional. We say that (u0 , v0 ) ∈ A × B is a saddle point for L if
L(u0 , v) ≤ L(u0 , v0 ) ≤ L(u, v0 ), ∀u ∈ A, v ∈ B.
Proposition 10.1. Let U,Y be Banach spaces, A ⊂ U and B ⊂ Y . A functional L :

U × Y → R has a saddle point if and only if
max inf L(u, v) = min sup L(u, v).

v∈B u∈A u∈A v∈B
Proof. Suppose (u0 , v0 ) ∈ A × B is a saddle point of L.

Thus,
L(u0 , v) ≤ L(u0 , v0 ) ≤ L(u, v0 ), ∀u ∈ A, v ∈ B. (10.94)
Define
F(u) = sup L(u, v).
v∈B
Observe that
inf F(u) ≤ F(u0 ),
u∈A
so that
inf sup L(u, v) ≤ sup L(u0 , v). (10.95)
u∈A v∈B v∈B
Define
G(v) = inf L(u, v).
u∈A
Thus
sup G(v) ≥ G(v0 ),
v∈B
10.3 The Min–Max Theorem 265
so that
sup inf L(u, v) ≥ inf L(u, v0 ). (10.96)
v∈B u∈A u∈A
From (10.94), (10.95), and (10.96) we obtain
inf sup L(u, v) ≤ sup L(u0 , v)

u∈A v∈B v∈B
≤ L(u0 , v0 )
≤ inf L(u, v0 )
u∈A
≤ sup inf L(u, v). (10.97)
v∈B u∈A
Hence
inf sup L(u, v) ≤ L(u0 , v0 )

u∈A v∈B
≤ sup inf L(u, v). (10.98)

v∈B u∈A
On the other hand
inf L(u, v) ≤ L(u, v), ∀u ∈ A, v ∈ B,

u∈A
so that
sup inf L(u, v) ≤ sup L(u, v), ∀u ∈ A,
v∈B u∈A vInB
and hence
sup inf L(u, v) ≤ inf sup L(u, v). (10.99)
inf sup L(u, v) = sup L(u0 , v)

u∈A v∈B v∈B
= L(u0 , v0 )
= inf L(u, v0 )
u∈A
= sup inf L(u, v). (10.100)
v∈B u∈A
Conversely suppose
max inf L(u, v) = min sup L(u, v).
As above defined,
F(u) = sup L(u, v),
v∈B
and
G(v) = inf L(u, v).
u∈A
From the hypotheses, there exists (u0 , v0 ) ∈ A × B such that
sup G(v) = G(v0 ) = F(u0 ) = inf F(u).

v∈B u∈A
so that
F(u0 ) = sup L(u0 , v) = inf L(u, v0 ) = G(v0 ).
v∈B u∈U
In particular
L(u0 , v0 ) ≤ sup L(u0 , v) = inf L(u, v0 ) ≤ L(u0 , v0 ).

v∈B u∈U
Therefore
sup L(u0 , v) = L(u0 , v0 ) = inf L(u, v0 ).
v∈B u∈U
Proposition 10.2. Let U,Y be Banach spaces, A ⊂ U, B ⊂ Y and let L : A × B → R

be a functional. Assume there exist u0 ∈ A, v0 ∈ B, and α ∈ R such that
L(u0 , v) ≤ α , ∀v ∈ B,
and
L(u, v0 ) ≥ α , ∀u ∈ A.
Under such hypotheses (u0 , v0 ) is a saddle point of L, that is,
L(u0 , v) ≤ L(u0 , v0 ) ≤ L(u, v0 ), ∀u ∈ A, v ∈ B.
Proof. Observe, from the hypotheses, that we have
L(u0 , v0 ) ≤ α ,
and
L(u0 , v0 ) ≥ α ,
so that
L(u0 , v) ≤ α = L(u0 , v0 ) ≤ L(u, v0 ), ∀u ∈ A, v ∈ B.
In the next lines we state and prove the min-max theorem.
Theorem 10.1. Let U,Y be reflexive Banach spaces, A ⊂ U, B ⊂ Y and let L : A ×

B → R be a functional.
Suppose that:
1. A ⊂ U is convex, closed, and nonempty.
2. B ⊂ Y is convex, closed, and nonempty.
3. For each u ∈ A, Fu (v) = L(u, v) is concave and upper semicontinuous.
4. For each v ∈ B, Gv (u) = L(u, v) is convex and lower semicontinuous.

5. The sets A and B are bounded.
Under such hypotheses L has at least one saddle point (u0 , v0 ) ∈ A × B such that
L(u0 , v0 ) = min max L(u, v)

u∈A v∈B
= max min L(u, v). (10.101)
v∈B u∈A
Proof. Fix v ∈ B. Observe that Gv (u) = L(u, v) is convex and lower semicontinuous.
Therefore it is weakly lower semicontinuous on the weak compact set A. At first we
assume the additional hypothesis that Gv (u) is strictly convex, ∀v ∈ B. Hence Gv (u)
attains a unique minimum on A. We denote the optimal u ∈ A by u(v).
Define
G(v) = min Gv (u) = min L(u, v).
u∈A u∈U
Thus,
G(v) = L(u(v), v).
The function G(v) is expressed as the minimum of a family of concave
weakly upper semicontinuous functions, and hence it is also concave and upper
semicontinuous.
Moreover, G(v) is bounded above on the weakly compact set B, so that there
exists v0 ∈ B such that
G(v0 ) = max G(v) = max min L(u, v).

v∈B v∈B u∈A
Observe that
G(v0 ) = min L(u, v0 ) ≤ L(u, v0 ), ∀u ∈ U.
u∈A
Observe that from the concerned concavity, for u ∈ A, v ∈ B, and λ ∈ (0, 1), we have
L(u, (1 − λ )v0 + λ v) ≥ (1 − λ )L(u, v0) + λ L(u, v).
In particular denote u((1 − λ )v0 + λ v) = uλ , where uλ is such that
G((1 − λ )v0 + λ v) = min L(u, (1 − λ )v0 + λ v)

u∈A
= L(uλ , (1 − λ )v0 + λ v). (10.102)
Therefore,
G(v0 ) = max G(v)

v∈B
≥ G((1 − λ )v0 + λ v)
= L(uλ , (1 − λ )v0 + λ v)
≥ (1 − λ )L(uλ , v0 ) + λ L(uλ , v)
≥ (1 − λ ) min L(u, v0 ) + λ L(uλ , v)

u∈A
= (1 − λ )G(v0) + λ L(uλ , v). (10.103)
From this, we obtain

G(v0 ) ≥ L(uλ , v). (10.104)
Let {λn } ⊂ (0, 1) be such that λn → 0.
Let {un } ⊂ A be such that
G((1 − λn)v0 + λnv) = min L(u, (1 − λn)v0 + λnv)

u∈A
= L(un , (1 − λn)v0 + λn v). (10.105)
Since A is weakly compact, there exists a subsequence {unk } ⊂ {un } ⊂ A and

u0 ∈ A such that
unk u0 , weakly in U, as k → ∞.
Observe that
(1 − λnk )L(unk , v0 ) + λnk L(unk , v) ≤ L(unk , (1 − λnk )v0 + λnk v)

= min L(u, (1 − λnk )v0 + λnk v)
u∈A
≤ L(u, (1 − λnk )v0 + λnk v), (10.106)
∀u ∈ A, k ∈ N.
Recalling that λnk → 0, from this and (10.106), we obtain
L(u0 , v0 ) ≤ lim inf L(unk , v0 )

k→∞
= lim inf((1 − λnk )L(unk , v0 ) + λnk L(u, v))
k→∞
≤ lim sup L(u, (1 − λnk )v0 + λnk v)
k→∞
≤ L(u, v0 ), ∀u ∈ U. (10.107)
Hence, L(u0 , v0 ) = minu∈A L(u, v0 ).

Observe that from (10.104) we have
G(v0 ) ≥ L(unk , v),
so that
G(v0 ) ≥ lim inf L(unk , v) ≥ L(u0 , v), ∀v ∈ B.
k→∞
Denoting α = G(v0 ) we have
α = G(v0 ) ≥ L(u0 , v), ∀v ∈ B,
and
α = G(v0 ) = min L(u, v0 ) ≤ L(u, v0 ), ∀u ∈ A.
u∈U
From these last two results and Proposition 10.2 we have that (u0 , v0 ) is a saddle
point for L. Now assume that
Gv (u) = L(u, v)
is convex but not strictly convex ∀v ∈ B.

For each n ∈ N define Ln by
Ln (u, v) = L(u, v) + u U /n.
In such a case
(Gv )n (u) = Ln (u, v)
is strictly convex for all n ∈ N.
From above we mainly obtain (un , vn ) ∈ A × B such that
L(un , v) + un U /n ≤ L(un , vn ) + un U /n
≤ L(u, vn ) + u /n. (10.108)
Since A × B is weakly compact and {(un , vn )} ⊂ A × B, up to subsequence not

relabeled, there exists (u0 , v0 ) ∈ A × B such that
un u0 , weakly in U,
vn v0 , weakly in Y,
so that
L(u0 , v) ≤ lim inf(L(un , v) + un U /n)

n→∞
≤ lim sup L(u, vn ) + u U /n
n→∞
≤ L(u, v0 ). (10.109)
Hence,
L(u0 , v) ≤ L(u, v0 ), ∀u ∈ A, v ∈ B,
so that
L(u0 , v) ≤ L(u0 , v0 ) ≤ L(u, v0 ), ∀u ∈ A, v ∈ B.
In the next result we deal with more general situations.
Theorem 10.2. Let U,Y be reflexive Banach spaces, A ⊂ U, B ⊂ Y and let L : A ×
B → R be a functional.
Suppose that
1. A ⊂ U is convex, closed, and nonempty.
2. B ⊂ Y is convex, closed, and nonempty.
3. For each u ∈ A, Fu (v) = L(u, v) is concave and upper semicontinuous.
4. For each v ∈ B, Gv (u) = L(u, v) is convex and lower semicontinuous.

5. Either the set A is bounded or there exists ṽ ∈ B such that
L(u, ṽ) → +∞, as u → +∞, u ∈ A.
6. Either the set B is bounded or there exists ũ ∈ A such that
L(ũ, v) → −∞, as v → +∞, v ∈ B.
Under such hypotheses L has at least one saddle point (u0 , v0 ) ∈ A × B.

Proof. We prove the result just for the special case such that there exists ṽ ∈ B such
that
L(u, ṽ) → +∞, as u → +∞, u ∈ A,
and B is bounded. The proofs of remaining cases are similar.
For each n ∈ N denote
An = {u ∈ A : u U ≤ n}.
Fix n ∈ N. The sets An and B are closed, convex, and bounded, so that from the last
Theorem 10.1 there exists a saddle point (un , vn ) ∈ An × B for
L : An × B → R.
Hence,
L(un , v) ≤ L(un , vn ) ≤ L(u, vn ), ∀u ∈ An , v ∈ B.
For a fixed ũ ∈ A1 we have
L(un , ṽ) ≤ L(un , vn )
≤ L(ũ, vn )
≤ sup L(ũ, v) ≡ b ∈ R. (10.110)
v∈B
On the other hand, from the hypotheses,
Gṽ (u) = L(u, ṽ)
is convex, lower semicontinuous, and coercive, so that it is bounded below. Thus

there exists a ∈ R such that
−∞ < a < Gṽ (u) = L(u, ṽ), ∀u ∈ A.
Hence
a ≤ L(un , ṽ) ≤ L(un , vn ) ≤ b, ∀n ∈ N.
Therefore {L(un , vn )} is bounded.
Moreover, from the coercivity hypotheses and
a ≤ L(un , ṽ) ≤ b, ∀n ∈ N,
10.4 Relaxation for the Scalar Case 271
we may infer that {un } is bounded.

Summarizing, {un }, {vn }, and {L(un , vn )} are bounded sequences, and thus there
exists a subsequence {nk }, u0 ∈ A, v0 ∈ B, and α ∈ R such that
unk u0 , weakly in U,
vnk v0 , weakly in Y,
L(unk , vnk ) → α ∈ R,
as k → ∞. Fix (u, v) ∈ A × B. Observe that if nk > n0 = u U , then
L(unk , v) ≤ L(unk , vnk ) ≤ L(u, vnk ),
so that letting k → ∞, we obtain
L(u0 , v) ≤ lim inf L(unk , v)

k→∞
≤ lim L(unk , vnk ) = α
k→∞
≤ lim sup L(u, vnk )
k→∞
≤ L(u, v0 ), (10.111)
that is,
L(u0 , v) ≤ α ≤ L(u, v0 ), ∀u ∈ A, v ∈ B.
From this and Proposition 10.2 we may conclude that (u0 , v0 ) is a saddle point
for L : A × B → R.
10.4 Relaxation for the Scalar Case
In this section, Ω ⊂ RN denotes a bounded open set with a locally Lipschitz

boundary. That is, for each point x ∈ ∂ Ω there exists a neighborhood Ux whose
intersection with ∂ Ω is the graph of a Lipschitz continuous function.
We start with the following definition.
Definition 10.4.1. A function u : Ω → R is said to be affine if ∇u is constant on Ω .
Furthermore, we say that u : Ω → R is piecewise affine if it is continuous and there
exists a partition of Ω 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 [25].
Theorem 10.4.2. Let r ∈ N and let uk 1 ≤ k ≤ r be piecewise affine functions from
Ω into R and {αk } such that αk > 0, ∀k ∈ {1, . . . , r} and ∑rk=1 αk = 1. Given ε > 0,
there exists a locally Lipschitz function u : Ω → R and r disjoint open sets Ωk ,

1 ≤ k ≤ r, such that
|m(Ωk ) − αk m(Ω )| < αk ε , ∀k ∈ {1, . . . , r}, (10.112)
∇u(x) = ∇uk (x), a.e. on Ωk , (10.113)
|∇u(x)| ≤ max {|∇uk (x)|}, a.e. on Ω , (10.114)

1≤k≤r

r

u(x) − ∑ αk uk < ε , ∀x ∈ Ω , (10.115)
k=1

r
u(x) = ∑ αk uk (x), ∀x ∈ ∂ Ω . (10.116)
k=1
The next result is also found in [25].

Proposition 10.4.3. Let r ∈ N and let uk 1 ≤ k ≤ r be piecewise affine functions
from Ω into R. Consider a Carathéodory function f : Ω × RN → R and a positive
function c ∈ L1 (Ω ) which satisfy
c(x) ≥ sup{| f (x, ξ )| | |ξ | ≤ max { ∇uk ∞ }}. (10.117)

1≤k≤r
Given ε > 0, there exists a locally Lipschitz function u : Ω → R such that

r

f (x, ∇u)dx − ∑ αk f (x, ∇uk )dx < ε , (10.118)
Ω k=1 Ω
|∇u(x)| ≤ max {|∇uk (x)|}, a.e. in Ω , (10.119)

1≤k≤r
r
|u(x) − ∑ αk uk (x)| < ε , ∀x ∈ Ω (10.120)
k=1
r
u(x) = ∑ αk uk (x), ∀x ∈ ∂ Ω . (10.121)
k=1
Proof. It is sufficient to establish the result for functions uk affine over Ω , since
Ω can be divided into pieces on which uk are affine, and such pieces can be put
together through (10.121). Let ε > 0 be given. We know that simple functions are
dense in L1 (Ω ), concerning the L1 norm. Thus there exists a partition of Ω into a

finite number of open sets Oi , 1 ≤ i ≤ N1 , and a negligible set, and there exists f¯k
constant functions over each Oi such that

| f (x, ∇uk (x)) − f¯k (x)|dx < ε , 1 ≤ k ≤ r. (10.122)
Ω
Now choose δ > 0 such that

ε
δ≤ (10.123)
N1 (1 + max1≤k≤r { f¯k ∞ })
and if B is a measurable set

m(B) < δ ⇒ c(x)dx ≤ ε /N1 . (10.124)
B
Now we apply Theorem 10.4.2, to each of the open sets Oi ; therefore there exists
a locally Lipschitz function u : Oi → R and there exist r open disjoints spaces Ωki ,
1 ≤ k ≤ r, such that
|m(Ωki ) − αk m(Oi )| ≤ αk δ , for 1 ≤ k ≤ r, (10.125)
∇u = ∇uk , a.e. in Ωki , (10.126)
|∇u(x)| ≤ max {|∇uk (x)|}, a.e. Oi , (10.127)

1≤k≤r

r

u(x) − ∑ αk uk (x) ≤ δ , ∀x ∈ Oi (10.128)
k=1

r
u(x) = ∑ αk uk (x), ∀x ∈ ∂ Oi . (10.129)
k=1
We can define u = ∑rk=1 αk uk on Ω − ∪Ni=1

1
Oi . Therefore u is continuous and locally
Lipschitz. Now observe that
r
f (x, ∇u(x))dx − ∑ f (x, ∇uk (x))dx
k=1 Ωk
Oi i

= f (x, ∇u(x))dx. (10.130)
Oi −∪rk=1 Ω ki
From | f (x, ∇u(x))| ≤ c(x), m(Oi − ∪rk=1 Ωki ) ≤ δ and (10.124) we obtain

r

f (x, ∇u(x))dx − ∑ f (x, ∇uk (x)dx
Oi k=1 Ω i
k

= f (x, ∇u(x))dx ≤ ε /N1 . (10.131)
Oi −∪rk=1 Ω ki
Considering that f¯k is constant in Oi , from (10.123), (10.124), and (10.125), we

obtain
r
∑| Ωki
f¯k (x)dx − αk
Oi
f¯k (x)dx| < ε /N1 . (10.132)
k=1
We recall that Ωk = ∪Ni=1

1
Ωki so that

r

f (x, ∇u(x))dx − ∑ αk f (x, ∇uk (x))dx
Ω k=1 Ω

r

≤ f (x, ∇u(x))dx − ∑ f (x, ∇uk (x))dx
Ω k=1 Ω k

r
+∑ | f (x, ∇uk (x) − f¯k (x)|dx
k=1 Ωk

r
+∑
fk (x)dx − αk
¯ fk (x)dx
¯
Ω
k=1 k Ω
r
+ ∑ αk | f¯k (x) − f (x, ∇uk (x))|dx. (10.133)
k=1 Ω
From (10.131), (10.122),(10.132), and (10.122) again, we obtain

r

f (x, ∇u(x))dx − ∑ αk f (x, ∇uk )dx < 4ε . (10.134)
Ω k=1 Ω
We do not prove the next result. It is a well-known result from the finite element
theory.
Proposition 10.4.4. If u ∈ W01,p (Ω ), there exists a sequence {un } of piecewise affine

functions over Ω , null on ∂ Ω , such that
un → u, in L p (Ω ) (10.135)
and
∇un → ∇u, in L p (Ω ; RN ). (10.136)

Proposition 10.4.5. For p such that 1 0 such that
a2 (x) + c2 |ξ | p ≤ f (x, ξ ) ≤ a1 (x) + c1 |ξ | p , ∀x ∈ Ω , ξ ∈ RN . (10.137)
Then, given u ∈ W 1,p (Ω ) piecewise affine, ε > 0, and a neighborhood V of zero in

the topology σ (L p (Ω , RN ), Lq (Ω , RN )), there exists a function v ∈ W 1,p (Ω ) such
that
∇v − ∇u ∈ V , (10.138)
u = v on ∂ Ω ,
v − u ∞ < ε , (10.139)
and

f (x, ∇v(x))dx − f ∗∗
(x, ∇u(x))dx < ε. (10.140)
Ω Ω

Proof. Suppose given ε > 0, u ∈ W 1,p (Ω ) piecewise affine continuous, and a neigh-
borhood V of zero, which may be expressed as

V = {w ∈ L p (Ω , RN ) | hm · wdx < η ,
Ω
∀m ∈ {1, . . . , M}}, (10.141)
where M ∈ N, hm ∈ Lq (Ω , RN ), η ∈ R+ . By hypothesis, there exists a partition of

Ω into a negligible set Ω0 and open subspaces Δi , 1 ≤ i ≤ r, over which ∇u(x) is
constant. From standard results of convex analysis in RN , for each i ∈ {1, . . . , r}, we
can obtain {αk ≥ 0}1≤k≤N+1 and ξk such that ∑N+1k=1 αk = 1 and
N+1
∑ αk ξk = ∇u, ∀x ∈ Δi , (10.142)
k=1
and
N+1
∑ αk f (x, ξk ) = f ∗∗ (x, ∇u(x)). (10.143)
k=1
Define βi = maxk∈{1,...,N+1} {|ξk | on Δi }, and ρ1 = maxi∈{1,...,r} {βi }, and ρ =

max{ρ1 , ∇u ∞ }. Now, observe that we can obtain functions ĥm ∈ C0∞ (Ω ; RN ) such
that
η
max ĥm − hm Lq (Ω ,RN ) < . (10.144)
m∈{1,...,M} 4ρ m(Ω )
Define C = maxm∈{1,...,M} div(ĥm ) Lq (Ω ) and we can also define
ε1 = min{ε /4, 1/(m(Ω )1/p), η /(2Cm(Ω )1/p ), 1/m(Ω )} (10.145)
We recall that ρ does not depend on ε . Furthermore, for each i ∈ {1, . . . , r}, there
exists a compact subset Ki ⊂ Δi such that

ε1
[a1 (x) + c1(x) max {|ξ | p}]dx < . (10.146)
Δ i −Ki |ξ |≤ρ r
Also, observe that the sets Ki may be obtained such that the restrictions of f and f ∗∗
to Ki × ρ B are continuous, so that from this and from the compactness of ρ B, for all
x ∈ Ki , we can find an open ball ωx with center in x and contained in Ω , such that
ε1
| f ∗∗ (y, ∇u(x)) − f ∗∗ (x, ∇u(x))| < , ∀y ∈ ωx ∩ Ki , (10.147)
m(Ω )
and
ε1
| f (y, ξ ) − f (x, ξ )| < , ∀y ∈ ωx ∩ Ki , ∀ξ ∈ ρ B. (10.148)
m(Ω )
Therefore, from this and (10.143), we may write

N+1 2ε1
∗∗
f (y, ∇u(x)) − ∑ αk f (y, ξk ) < , ∀y ∈ ωx ∩ Ki . (10.149)
k=1
m(Ω)
We can cover the compact set Ki with a finite number of those open ball ωx , de-
noted by ω j , 1 ≤ j ≤ l. Consider the open sets ω j = ω j − ∪i=1 ω̄i . We have that
j−1
∪lj=1 ω̄ j = ∪lj=1 ω̄ j . Defining functions uk , for 1 ≤ k ≤ N + 1 such that ∇uk = ξk and

k=1 αk uk , we may apply Proposition 10.4.3 to each of the open sets ω j , so that
u = ∑N+1
we obtain functions vi ∈ W 1,p (Ω ) such that

N+1 ε

f (x, ∇vi (x)dx − ∑ αk f (x, ξk )dx < ,
1
(10.150)
ωj k=1 ω j
rl
|∇vi | < ρ , ∀x ∈ ω j , (10.151)
|vi (x) − u(x)| < ε1 , ∀x ∈ ω j , (10.152)

and
vi (x) = u(x), ∀x ∈ ∂ ω j . (10.153)
Finally we set
vi = u on Δi − ∪lj=1ω j . (10.154)
We may define a continuous mapping v : Ω → R by
v(x) = vi (x), if x ∈ Δi , (10.155)
v(x) = u(x), if x ∈ Ω0 . (10.156)
We have that v(x) = u(x), ∀x ∈ ∂ Ω , and ∇v ∞ < ρ . Also, from (10.146)

ε1
| f ∗∗ (x, ∇u(x)|dx < (10.157)
Δ i −Ki r
and

ε1
| f (x, ∇v(x)|dx < . (10.158)
Δ i −Ki r
On the other hand, from (10.149) and (10.150)

f (x, ∇v(x))dx − f ∗∗ (x, ∇u(x))dx
Ki ∩ω j
Ki ∩ω j
ε1 ε1 m(ω j ∩ Ki )
≤ + (10.159)
rl m(Ω )
so that

| f (x, ∇v(x))dx − f ∗∗ (x, ∇u(x))dx|
Ki Ki
ε1 ε1 m(Ki )
≤ + . (10.160)
r m(Ω )
Now summing up in i and considering (10.157) and (10.158) we obtain (10.140),

that is,

| f (x, ∇v(x))dx − f ∗∗ (x, ∇u(x))dx| < 4ε1 ≤ ε . (10.161)
Ω Ω
Also, observe that from above, we have
v − u ∞ < ε1 , (10.162)
and thus

ĥm · (∇v(x) − ∇u(x))dx = − div(ĥm )(v(x) − u(x))dx
Ω Ω
≤ div(ĥm ) Lq (Ω ) v − u L p(S)
≤ Cε1 m(Ω )1/p
η
< . (10.163)
2
Also we have that

(ĥm − hm ) · (∇v − ∇u)dx
Ω

η
≤ ĥm − hm Lq (Ω ,RN ) ∇v − ∇u L p(Ω ,RN ) ≤ . (10.164)
2
Thus

hm · (∇v − ∇u)dx < η , ∀m ∈ {1, . . . , M}. (10.165)
Ω
Theorem 10.4.6. Assuming the hypothesis of last theorem, given a function u ∈

W01,p (Ω ), given ε > 0, and a neighborhood of zero V in σ (L p (Ω , RN ), Lq (Ω , RN )),
we have that there exists a function v ∈ W01,p (Ω ) such that
∇v − ∇u ∈ V , (10.166)
and

f (x, ∇v(x))dx − f ∗∗
(x, ∇u(x))dx < ε. (10.167)
Ω Ω

Proof. We can approximate u by a function w which is piecewise affine and null

on the boundary. Thus, there exists δ > 0 such that we can obtain w ∈ W0 (Ω )
1,p
piecewise affine such that
u − w 1,p < δ (10.168)
so that
1
∇w − ∇u ∈ V , (10.169)
2
and

ε
f ∗∗ (x, ∇w(x))dx − f ∗∗
(x, ∇u(x))dx < . (10.170)
Ω Ω
2
From Proposition 10.4.5 we may obtain v ∈ W01,p (Ω ) such that
1
∇v − ∇w ∈ V , (10.171)
2
and

ε
f ∗∗ (x, ∇w(x))dx − f (x, ∇v(x))dx < . (10.172)
Ω Ω 2
From (10.170) and (10.172)

f ∗∗ (x, ∇u(x))dx − f (x, ∇v(x))dx < ε. (10.173)
Ω Ω

Finally, from (10.169), (10.171), and from the fact that weak neighborhoods are
convex, we have
∇v − ∇u ∈ V . (10.174)
To finish this chapter, we present two theorems which summarize the last results.
Theorem 10.4.7. Let f be a Carathéodory function from Ω × RN into R which sat-
isfies
a2 (x) + c2 |ξ | p ≤ f (x, ξ ) ≤ a1 (x) + c1 |ξ | p (10.175)
where a1 , a2 ∈ L1 (Ω ), 1 0. Under such assumptions,

defining Û = W01,p (Ω ), we have

∗∗
inf f (x, ∇u)dx = min f (x, ∇u)dx (10.176)
u∈Û Ω u∈Û Ω
The solutions of relaxed problem are weak cluster points in W01,p (Ω ) of the mini-
mizing sequences of primal problem.
Proof. The existence of solutions for the convex relaxed formulation is a conse-
quence of the reflexivity of U and coercivity hypothesis, which allows an applica-
tion 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 û ∈ W 1,p (Ω ). Finally, from the lower
semicontinuity of relaxed formulation, we may conclude that û is a minimizer. The
relation (10.176) follows from last theorem.
Theorem 10.4.8. Let f be a Carathéodory function from Ω × RN into R which

satisfies
a2 (x) + c2 |ξ | p ≤ f (x, ξ ) ≤ a1 (x) + c1 |ξ | p (10.177)

where a1 , a2 ∈ L1 (Ω ), 1 0. Let u0 ∈ W 1,p (Ω ).

Under such assumptions, defining Û = {u | u − u0 ∈ W01,p (Ω )}, we have

∗∗
inf f (x, ∇u)dx = min f (x, ∇u)dx (10.178)
The solutions of relaxed problem are weak cluster points in W 1,p (Ω ) of the min-
imizing sequences of primal problem.
Proof. Just apply the last theorem to the integrand g(x, ξ ) = f (x, ξ + ∇u0 ). For
details see [25].
10.5 Duality Suitable for the Vectorial Case
10.5.1 The Ekeland Variational Principle
In this section we present and prove the Ekeland variational principle. This proof
may be found in Giusti, [39], pp. 160–161.
Theorem 10.5.1 (Ekeland Variational Principle). Let (U, d) be a complete metric
space and let F : U → R be a lower semicontinuous bounded below functional
taking a finite value at some point.
Let ε > 0. Assume for some u ∈ U we have
F(u) ≤ inf {F(u)} + ε .

u∈U
Under such hypotheses, there exists v ∈ U such that

1. d(u, v) ≤ 1,
2. F(v) ≤ F(u),
3. F(v) ≤ F(w) + ε d(v, w), ∀w ∈ U.
Proof. Define the sequence {un } ⊂ U by
u1 = u,
and having u1 , . . . , un , select un+1 as specified in the next lines. First, define
Sn = {w ∈ U | F(w) ≤ F(un ) − ε d(un, w)}.
Observe that un ∈ Sn so that Sn in nonempty.

On the other hand, from the definition of infimum, we may select un+1 ∈ Sn such
that
1
F(un+1 ) ≤ F(un ) + inf {F(w)} . (10.179)
2 w∈Sn
10.5 Duality Suitable for the Vectorial Case 281
Since un+1 ∈ Sn we have
ε d(un+1 , un ) ≤ F(un ) − F(un+1 ). (10.180)
and hence
m
ε d(un+m , un ) ≤ ∑ d(un+i , un+i−1 ) ≤ F(un ) − F(un+m ). (10.181)
i=1
From (10.180) {F(un )} is decreasing sequence bounded below by infu∈U F(u)

so that there exists α ∈ R such that
F(un ) → α as n → ∞.
From this and (10.181), {un } is a Cauchy sequence , converging to some v ∈ U.

Since F is lower semicontinuous we get
α = lim inf F(un+m ) ≥ F(v),

m→∞
so that letting m → ∞ in (10.181) we obtain
ε d(un , v) ≤ F(un ) − F(v), (10.182)
and, in particular, for n = 1 we get
0 ≤ ε d(u, v) ≤ F(u) − F(v) ≤ F(u) − inf F(u) ≤ ε .

u∈U
Thus, we have proven 1 and 2.

Suppose, to obtain contradiction, that 3 does not hold.
Hence, there exists w ∈ U such that
F(w) < F(v) − ε d(w, v).
In particular we have
w = v. (10.183)
Thus, from this and (10.182), we have
F(w) < F(un ) − ε d(un , v) − ε d(w, v) ≤ F(un ) − ε d(un , w), ∀n ∈ N.
Now observe that w ∈ Sn , ∀n ∈ N so that
inf {F(w)} ≤ F(w), ∀n ∈ N.

w∈Sn
2F(un+1 ) − F(un ) ≤ F(w) < F(v) − ε d(v, w),

so that
2 lim inf{F(un+1 )} ≤ F(v) − ε d(v, w) + lim inf{F(un )}.
n→∞ n→∞
Hence,
F(v) ≤ lim inf{F(un+1 )} ≤ F(v) − ε d(v, w),
n→∞
so that
0 ≤ −ε d(v, w),
which contradicts (10.183).
Thus 3 holds.
Remark 10.5.2. We may introduce in U a new metric given by d1 = ε 1/2 d. We
highlight that the topology remains the same and also F remains lower semicon-
tinuous. Under the hypotheses of the last theorem, if there exists u ∈ U such that
F(u) < infu∈U F(u) + ε , then there exists v ∈ U such that
1. d(u, v) ≤ ε 1/2 ,
2. F(v) ≤ F(u),
3. F(v) ≤ F(w) + ε 1/2 d(u, w), ∀w ∈ U.
Remark 10.5.3. Observe that if U is a Banach space,
F(v) − F(v + tw) ≤ ε 1/2t w U , ∀t ∈ [0, 1], w ∈ U, (10.184)
so that if F is Gâteaux differentiable, we obtain
−δ F(v), wU ≤ ε 1/2 w U . (10.185)
Similarly
F(v) − F(v + t(−w)) ≤ ε 1/2t w U ≤, ∀t ∈ [0, 1], w ∈ U, (10.186)
so that if F is Gâteaux differentiable, we obtain
δ F(v), wU ≤ ε 1/2 w U . (10.187)
Thus
δ F(v) U ∗ ≤ ε 1/2 . (10.188)
We have thus obtained, from the last theorem and remarks, the following result.
Theorem 10.5.4. Let U be a Banach space. Let F : U → R be a lower semicon-
tinuous Gâteaux differentiable functional. Given ε > 0 suppose that u ∈ U is such
that
F(u) ≤ inf {F(u)} + ε . (10.189)

u∈U
Then there exists v ∈ U such that

F(v) ≤ F(u), (10.190)
√
u − v U ≤ ε, (10.191)
and
√
δ F(v) U ∗ ≤ ε. (10.192)
The next theorem easily follows from above results.

Theorem 10.5.5. Let J : U → R be defined by
J(u) = G(∇u) − f , uL2 (S;RN ) , (10.193)
where
U = W01,2 (S; RN ), (10.194)
We suppose G is a l.s.c and Gâteaux differentiable so that J is bounded below. Then,

given ε > 0, there exists uε ∈ U such that
J(uε ) < inf {J(u)} + ε , (10.195)

u∈U
and
√
δ J(uε ) U ∗ < ε. (10.196)
We finish this chapter with an important result for vectorial problems in the calculus
of variations.
Theorem 10.5.6. Let U be a reflexive Banach space. Consider (G ◦ Λ ) : U → R and
(F ◦ Λ1 ) : U → R l.s.c. functionals such that J : U → R defined as
J(u) = (G ◦ Λ )(u) − (F ◦ Λ1 )(u) − u, f U
is below bounded. (Here Λ : U → Y and Λ1 : U → Y1 are continuous linear oper-
ators whose adjoint operators are denoted by Λ ∗ : Y ∗ → U ∗ and Λ1∗ : Y ∗ → U ∗ ,
respectively). Also we suppose the existence of L : Y1 → Y continuous and linear
operator such that L∗ is onto and
Λ (u) = L(Λ1 (u)), ∀u ∈ U.
Under such assumptions, we have
inf {J(u)} ≥ sup { ∗inf ∗ {F ∗ (L∗ z∗ ) − G∗(v∗ + z∗ )}},
u∈U v∗ ∈A∗ z ∈Y1
where
A∗ = {v∗ ∈ Y ∗ | Λ ∗ v∗ = f }.
In addition we assume (F ◦ Λ1 ) : U → R is convex and Gâteaux differentiable, and

suppose there exists a solution (v∗0 , z∗0 ) of the dual formulation, so that
∗ ∗ ∗
∂ F (L z0 )
L ∈ ∂ G∗ (v∗0 + z∗0 ),
∂ v∗
Λ ∗ v∗0 − f = 0.
Suppose u0 ∈ U is such that
∂ F ∗ (L∗ z∗0 )
= Λ1 u 0 ,
∂ v∗
so that
Λ u0 ∈ ∂ G∗ (v∗0 + z∗0 ).
Also we assume that there exists a sequence {un } ⊂ U such that un u0 weakly in
U and
G(Λ un ) → G∗∗ (Λ u0 ) as n → ∞.
Under these additional assumptions we have
inf {J(u)} = max

∗
{ inf ∗ {F ∗ (L∗ z∗ ) − G∗ (v∗ + z∗ )}}
∗ ∗
u∈U v ∈A z ∈Y1
∗
=F (L∗ z∗0 ) − G∗(v∗0 + z∗0 ).
Proof. Observe that
G∗ (v∗ + z∗ ) ≥ Λ u, v∗ Y + Λ u, z∗ Y − G(Λ u), ∀u ∈ U,
that is,
− F ∗ (L∗ z∗ ) + G∗ (v∗ + z∗ ) ≥ u, f U − F ∗ (L∗ z∗ ) + Λ1 u, L∗ z∗ Y1

− G(Λ u), ∀u ∈ U, v∗ ∈ A∗
so that
sup {−F ∗ (L∗ z∗ ) + G∗ (v∗ + z∗ )}

z∗ ∈Y1∗
≥ sup {u, f U − F ∗ (L∗ z∗ ) + Λ1 u, L∗ z∗ Y1 − G(Λ u)},

z∗ ∈Y1∗
∀v∗ ∈ A∗ , u ∈ U and therefore
G(Λ u) − F(Λ1 u) − u, f U ≥ ∗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∗ )}},

u∈U v∗ ∈A∗ z ∈Y1
where
A∗ = {v∗ ∈ Y ∗ | Λ ∗ v∗ = f }.
Now suppose
∂ F ∗ (L∗ z∗0 )
L ∈ ∂ G∗ (v∗0 + z∗0 ),
∂ v∗
and u0 ∈ U is such that
∂ F ∗ (L∗ z∗0 )
= Λ1 u 0 .
∂ v∗
Observe that
Λ u0 = L(Λ1 u0 ) ∈ ∂ G(v∗0 + z∗0 )
implies that
G∗ (v∗0 + z∗0 ) = Λ u0 , v∗0 Y + Λ u0 , z∗0 Y − G∗∗(Λ u0 ).
From the hypothesis

un u0 weakly in U
and
G(Λ un ) → G∗∗ (Λ u0 ) as n → ∞.
Thus, given ε > 0, there exists n0 ∈ N such that if n ≥ n0 then
G∗ (v∗0 + z∗0 ) − Λ un , v∗0 Y − Λ un , z∗0 Y + G(Λ un ) < ε /2.
On the other hand, since F(Λ1 u) is convex and l.s.c., we have
lim sup{−F(Λ1 un )} ≤ −F(Λ1 u0 ).

n→∞
Hence, there exists n1 ∈ N such that if n ≥ n1 , then

ε ε
Λ un , z∗0 Y − F(Λ1 un ) ≤ Λ u0 , z∗0 Y − F(Λ1 u0 ) + = F ∗ (L∗ z∗0 ) + ,
2 2
so that for all n ≥ max{n0 , n1 } we obtain
G∗ (v∗0 + z∗0 ) − F ∗ (L∗ z∗0 ) − un , f U − F(Λ1 un ) + G(Λ un ) < ε .
Since ε is arbitrary, the proof is complete.

Chapter 11
Constrained Variational Optimization
11.1 Basic Concepts
For this chapter the most relevant reference is the excellent book of Luenberger
[47], where more details may be found. Other relevant references are [15, 40–42].
We start with the definition of cone.
Definition 11.1.1 (Cone). Given a Banach space U, we say that C ⊂ U is a cone
with the vertex at the origin; if given u ∈ C, we have that λ u ∈ C, ∀λ ≥ 0. By
analogy we define a cone with the vertex at p ∈ U as P = p + C, where C is any
cone with the vertex at the origin. From now on we consider only cones with vertex
at origin, unless otherwise indicated.
Definition 11.1.2. 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 ≥ θ if and only if u ∈ C. Also
P+ = {u∗ ∈ U ∗ | u, u∗ U ≥ 0, ∀u ∈ P}. (11.1)
If u∗ ∈ P+ , we write u∗ ≥ θ ∗ .
Proposition 11.1.3. Let U be a Banach space and P be a closed cone in U. If u ∈ U
satisfies u, u∗ U ≥ 0, ∀u∗ ≥ θ ∗ , then u ≥ θ .
Proof. We prove the contrapositive. Assume u ∈ P. Then by the separating hyper-
plane theorem there is an u∗ ∈ U ∗ such that u, u∗ U < p, u∗ U , ∀p ∈ P. Since P is
a cone we must have p, u∗ U ≥ 0; otherwise we would have u, u∗ > α p, u∗ U
for some α > 0. Thus u∗ ∈ P+ . Finally, since inf p∈P{p, u∗ U } = 0, we obtain
u, u∗ U < 0 which completes the proof.
Definition 11.1.4 (Convex Mapping). Let U, Z be vector spaces. Let P ⊂ Z be a
cone. A mapping G : U → Z is said to be convex if the domain of G is convex and
G(α u1 + (1 − α )u2) ≤ α G(u1 ) + (1 − α )G(u2),

∀u1 , u2 ∈ U, α ∈ [0, 1]. (11.2)
288 11 Constrained Variational Optimization
Consider the problem P, defined as
Problem P : Minimize F : U → R subject to u ∈ Ω , and G(u) ≤ θ
Define
ω (z) = inf{F(u) | u ∈ Ω and G(u) ≤ z}. (11.3)
For such a functional we have the following result.

Proposition 11.1.5. If F is a real convex functional and G is convex, then ω is
convex.
Proof. Observe that
ω (α z1 + (1 − α )z2) = inf{F(u) | u ∈ Ω
and G(u) ≤ α z1 + (1 − α )z2}
(11.4)
≤ inf{F(u) | u = α u1 + (1 − α )u2 u1 , u2 ∈ Ω
and G(u1 ) ≤ z1 , G(u2 ) ≤ z2 }
(11.5)
≤α inf{F(u1 ) | u1 ∈ Ω , G(u1 ) ≤ z1 }
+ (1 − α ) inf{F(u2 ) | u2 ∈ Ω , G(u2 ) ≤ z2 }
(11.6)
≤αω (z1 ) + (1 − α )ω (z2). (11.7)
Now we establish the Lagrange multiplier theorem for convex global optimization.
Theorem 11.1.6. Let U be a vector space, Z a Banach space, Ω a convex subset of
U, and P a positive cone of Z. Assume that P contains an interior point. Let F be
a real convex functional on Ω and G a convex mapping from Ω into Z. Assume the
existence of u1 ∈ Ω such that G(u1 ) < θ . Defining
μ0 = inf {F(u) | G(u) ≤ θ }, (11.8)

u∈Ω
then there exists z∗0 ≥ θ , z∗0 ∈ Z ∗ such that
μ0 = inf {F(u) + G(u), z∗0 Z }. (11.9)

u∈Ω
Furthermore, if the infimum in (11.8) is attained by u0 ∈ U such that G(u0 ) ≤ θ , it

is also attained in (11.9) by the same u0 and also G(u0 ), z∗0 Z = 0. We refer to z∗0 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 ∈ Ω }, (11.10)

11.1 Basic Concepts 289
and
B = {(r, z) ∈ R × Z | r ≤ μ0 , z ≤ θ }, (11.11)
where μ0 = infu∈Ω {F(u) | G(u) ≤ θ }. 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 nonzero element w∗0 = (r0 , z∗0 ) ∈ W ∗ such that
r0 r1 + z1 , z∗0 Z ≥ r0 r2 + z2 , z∗0 Z , ∀(r1 , z1 ) ∈ A, (r2 , z2 ) ∈ B. (11.12)
From the nature of B it is clear that w∗0 ≥ θ . That is, r0 ≥ 0 and z∗0 ≥ θ . We will show
that r0 > 0. The point (μ0 , θ ) ∈ B; hence
r0 r + z, z∗0 Z ≥ r0 μ0 , ∀(r, z) ∈ A. (11.13)
If r0 = 0, then G(u1 ), z∗0 Z ≥ 0 and z∗0 = θ . Since G(u1 ) < θ and z∗0 ≥ θ we have
a contradiction. Therefore r0 > 0 and, without loss of generality, we may assume
r0 = 1. Since the point (μ0 , θ ) is arbitrarily close to A and B, we have
μ0 = inf {r + z, z∗0 Z } ≤ inf {F(u) + G(u), z∗0 Z }

(r,z)∈A u∈Ω
≤ inf{F(u) | u ∈ Ω , G(u) ≤ θ } = μ0 . (11.14)
Also, if there exists u0 such that G(u0 ) ≤ θ , μ0 = F(u0 ), then
μ0 ≤ F(u0 ) + G(u0), z∗0 Z ≤ F(u0 ) = μ0 . (11.15)
Hence
G(u0 ), z∗0 Z = 0. (11.16)
Corollary 11.1.7. Let the hypothesis of the last theorem hold. Suppose
F(u0 ) = inf {F(u) | G(u) ≤ θ }. (11.17)

u∈Ω
Then there exists z∗0 ≥ θ such that the Lagrangian L : U × Z ∗ → R defined by
L(u, z∗ ) = F(u) + G(u), z∗ Z (11.18)
has a saddle point at (u0 , z∗0 ). That is
L(u0 , z∗ ) ≤ L(u0 , z∗0 ) ≤ L(u, z∗0 ), ∀u ∈ Ω , z∗ ≥ θ . (11.19)
Proof. For z∗0 obtained in the last theorem, we have
L(u0 , z∗0 ) ≤ L(u, z∗0 ), ∀u ∈ Ω . (11.20)

As G(u0 ), z∗0 Z = 0, we have
L(u0 , z∗ ) − L(u0 , z∗0 ) = G(u0 ), z∗ Z − G(u0), z∗0 Z

= G(u0 ), z∗ Z ≤ 0. (11.21)
We now prove two theorems relevant to develop the subsequent section.

Theorem 11.1.8. Let F : Ω ⊂ U → R and G : Ω → Z. Let P ⊂ Z be a cone. Suppose
there exists (u0 , z∗0 ) ∈ U × Z ∗ where z∗0 ≥ θ and u0 ∈ Ω are such that
F(u0 ) + G(u0 ), z∗0 Z ≤ F(u) + G(u), z∗0 Z , ∀u ∈ Ω . (11.22)
Then
F(u0 ) + G(u0 ), z∗0 Z

= inf{F(u) | u ∈ Ω and G(u) ≤ G(u0 )}. (11.23)
Proof. Suppose there is a u1 ∈ Ω such that F(u1 ) < F(u0 ) and G(u1 ) ≤ G(u0 ). Thus
G(u1 ), z∗0 Z ≤ G(u0 ), z∗0 Z (11.24)
so that
F(u1 ) + G(u1), z∗0 Z < F(u0 ) + G(u0 ), z∗0 Z , (11.25)
which contradicts the hypothesis of the theorem.

Theorem 11.1.9. Let F be a convex real functional and G : Ω → Z convex and let
u0 and u1 be solutions to the problems P0 and P1 respectively, where
P0 : minimize F(u) subject to u ∈ Ω and G(u) ≤ z0 , (11.26)
and
P1 : minimize F(u) subject to u ∈ Ω and G(u) ≤ z1 . (11.27)
Suppose z∗0 and z∗1 are the Lagrange multipliers related to these problems. Then
z1 − z0 , z∗1 Z ≤ F(u0 ) − F(u1 ) ≤ z1 − z0 , z∗0 Z . (11.28)
Proof. For u0 , z∗0 we have
F(u0 ) + G(u0 ) − z0 , z∗0 Z ≤ F(u) + G(u) − z0, z∗0 Z , ∀u ∈ Ω , (11.29)
and, particularly for u = u1 and considering that G(u0 ) − z0 , z∗0 Z = 0, we obtain
F(u0 ) − F(u1 ) ≤ G(u1 ) − z0 , z∗0 Z ≤ z1 − z0 , z∗0 Z . (11.30)
A similar argument applied to u1 , z∗1 provides us the other inequality.

11.2 Duality 291
11.2 Duality
Consider the basic convex programming problem:
Minimize F(u) subject to G(u) ≤ θ , u ∈ Ω , (11.31)
where F : U → R is a convex functional, G : U → Z is convex mapping, and Ω is a

convex set. We define ϕ : Z ∗ → R by
ϕ (z∗ ) = inf {F(u) + G(u), z∗ Z }. (11.32)

u∈Ω
Proposition 11.2.1. ϕ is concave and
ϕ (z∗ ) = inf {ω (z) + z, z∗ Z }, (11.33)

z∈Γ
where
ω (z) = inf {F(u) | G(u) ≤ z}, (11.34)

u∈Ω
and
Γ = {z ∈ Z | G(u) ≤ z f or some u ∈ Ω }.
Proof. Observe that
ϕ (z∗ ) = inf {F(u) + G(u), z∗ Z }

u∈Ω
≤ inf {F(u) + z, z∗ Z | G(u) ≤ z}
u∈Ω
= ω (z) + z, z∗ Z , ∀z∗ ≥ θ , z ∈ Γ . (11.35)
On the other hand, for any u1 ∈ Ω , defining z1 = G(u1 ), we obtain
F(u1 ) + G(u1), z∗ Z ≥ inf {F(u) + z1 , z∗ Z | G(u) ≤ z1 }

u∈Ω
= ω (z1 ) + z1 , z∗ Z , (11.36)
so that
ϕ (z∗ ) ≥ inf {ω (z) + z, z∗ Z }. (11.37)

z∈Γ
Theorem 11.2.2 (Lagrange Duality). Consider F : Ω ⊂ U → R is a convex func-

tional, Ω a convex set, and G : U → Z a convex mapping. Suppose there exists a u1
such that G(u1 ) < θ and that infu∈Ω {F(u) | G(u) ≤ θ } < ∞. Under such assump-
tions, we have
inf {F(u) | G(u) ≤ θ } = max

∗
{ϕ (z∗ )}. (11.38)
u∈Ω z ≥θ
If the infimum on the left side in (11.38) is achieved at some u0 ∈ U and the max on
the right side at z∗0 ∈ Z ∗ , then
G(u0 ), z∗0 Z = 0 (11.39)
and u0 minimizes F(u) + G(u), z∗0 Z on Ω .

Proof. For z∗ ≥ θ we have
inf {F(u) + G(u), z∗ Z } ≤ inf {F(u) + G(u), z∗ Z }

u∈Ω u∈Ω ,G(u)≤θ
≤ inf F(u) ≤ μ0 . (11.40)

u∈Ω ,G(u)≤θ
or
ϕ (z∗ ) ≤ μ0 . (11.41)
The result follows from Theorem 11.1.6.
11.3 The Lagrange Multiplier Theorem
Remark 11.3.1. This section was published in similar form by the journal Computa-
tional and Applied Mathematics, SBMAC-Springer, reference [15].
In this section we develop a new and simpler proof of the Lagrange multiplier
theorem in a Banach space context. In particular, we address the problem of min-
imizing a functional F : U → R subject to G(u) = θ , where θ denotes the zero
vector and G : U → Z is a Fréchet differentiable transformation. Here U, Z are Ba-
nach spaces. General results on Banach spaces may be found in [1, 26], for example.
For the theorem in question, among others, we would cite [13, 40, 47]. Specially the
proof given in [47] is made through the generalized inverse function theorem. We
emphasize such a proof is extensive and requires the continuous Fréchet differentia-
bility of F and G. Our approach here is different and the results are obtained through
other hypotheses.
The main result is summarized by the following theorem.
Theorem 11.3.2. Let U and Z be Banach spaces. Assume u0 is a local minimum of
F(u) subject to G(u) = θ , where F : U → R is a Gâteaux differentiable functional
and G : U → Z is a Fréchet differentiable transformation such that G (u0 ) maps U
onto Z. Finally, assume there exist α > 0 and K > 0 such that if ϕ U < α , then
G (u0 + ϕ ) − G(u0 ) ≤ K ϕ U .
Under such assumptions, there exists z∗0 ∈ Z ∗ such that
F (u0 ) + [G (u0 )]∗ (z∗0 ) = θ ,

11.3 The Lagrange Multiplier Theorem 293
that is,
ϕ , F (u0 )U + G (u0 )ϕ , z∗0 Z = 0, ∀ϕ ∈ U.
Proof. First observe that there is no loss of generality in assuming 0 < α < 1. Also
from the generalized mean value inequality and our hypothesis, if ϕ U < α , then
G(u0 + ϕ ) − G(u0) − G (u0 ) · ϕ

" #
≤ sup G (u0 + hϕ ) − G(u0 ) ϕ U
h∈[0,1]
≤ K sup { hϕ U } ϕ U ≤ K ϕ U2 . (11.42)
h∈[0,1]
For each ϕ ∈ U, define H(ϕ ) by
G(u0 + ϕ ) = G(u0 ) + G (u0 ) · ϕ + H(ϕ ),
that is,
H(ϕ ) = G(u0 + ϕ ) − G(u0) − G (u0 ) · ϕ .

Let L0 = N(G (u0 )) where N(G (u0 )) denotes the null space of G (u0 ). Observe
that U/L0 is a Banach space for which we define A : U/L0 → Z by
A(ū) = G (u0 ) · u,
where ū = {u + v | v ∈ L0 }.
Since G (u0 ) is onto, so is A, so that by the inverse mapping theorem A has a
continuous inverse A−1 .
Let ϕ ∈ U be such that G (u0 ) · ϕ = θ . For a given t such that 0 < |t| < 1+ αϕ U ,
let ψ0 ∈ U be such that
H(t ϕ )
G (u0 ) · ψ0 + 2 = θ ,
t
Observe that from (11.42),
H(t ϕ ) ≤ Kt 2 ϕ U2 ,
and thus from the boundedness of A−1 , ψ0 as a function of t may be chosen

uniformly bounded relating t (i.e., despite the fact that ψ0 may vary with t, there
exists K1 > 0 such that ψ0 U < K1 , ∀t such that 0 < |t| < 1+ αϕ U ).
Now choose 0 < r < 1/4 and define g0 = θ .
Also define
r
ε= .
4( A−1 + 1)(K + 1)(K1 + 1)( ϕ U + 1)
Since from the hypotheses G (u) is continuous at u0 , we may choose 0 < δ < α
such that if v U < δ then
G (u0 + v) − G(u0 ) < ε .
Fix t ∈ R such that

δ
0 < |t| < .
2(1 + ϕ U + K1 )
Observe that ψ ∈ U is such that G(u0 + t ϕ + t 2 ψ ) = θ if and only if
H(t ϕ + t 2ψ )
G (u0 ) · ψ + = θ.
t2
Define

−1 H(t ϕ + t 2(ψ0 − g0))
L1 = A G (u0 ) · (ψ0 − g0 ) + ,
t2
so that

H(t ϕ + t 2(ψ0 − g0))
L1 = A−1 [A(ψ0 − g0)] + A−1
t2
= ψ0 − g0 + w1
= ψ0 + w1
= {ψ0 + w1 + v | v ∈ L0 }.
Here w1 ∈ U is such that

−1 H(t ϕ + t 2 (ψ0 − g0 ))
w1 = A ,
t2
that is,
H(t ϕ + t 2 (ψ0 − g0 ))
A(w1 ) = ,
t2
so that
H(t ϕ + t 2(ψ0 − g0))
G (u0 ) · w1 = .
t2
Select g1 ∈ L1 such that
g1 − g0 U ≤ 2 L1 − L0 .
This is possible since

L1 − L0 = inf { g − g0 U }.
g∈L1
So we have that

−1 H(t ϕ ) H(t ϕ + t 2(ψ0 − g0))
L1 = A − 2 + . (11.43)
t t2
However
H(t ϕ + t 2(ψ0 − g0)) − H(t ϕ )

= G(u0 + t ϕ + t 2(ψ0 )) − G(u0 )
−G (u0 ) · (t ϕ + t 2 (ψ0 ))
−G(u0 + t ϕ ) + G(u0)
+G (u0 ) · (t ϕ )
= G(u0 + t ϕ + t 2(ψ0 )) − G(u0 + t ϕ )
−G (u0 ) · (t 2 (ψ0 )), (11.44)
so that by the generalized mean value inequality we may write
H(t ϕ + t 2(ψ0 − g0)) − H(t ϕ )

≤ sup G (u0 + t ϕ + ht 2(ψ0 )) − G (u0 ) t 2 ψ0 U
h∈[0,1]
< ε t 2 ψ0 U . (11.45)
From this and (11.43) we get
L1 ≤ A−1 H(t ϕ + t 2 (ψ0 − g0 )) − H(t ϕ ) /t 2

< A−1 ε ψ0 U
r
< A−1 K1
4( A−1 + 1)(K + 1)(K1 + 1)( ϕ U + 1)
r
< . (11.46)
4
Hence
g1 U < 2 L1 < r/2.
Now reasoning by induction, for n ≥ 2, assume that gn−1 U <r and gn−2 U <r
and define Ln by

−1 H(t ϕ + t 2 (ψ0 − gn−1))
Ln − Ln−1 = A G (u0 ) · (ψ0 − gn−1) + .
t2
Observe that

−1 H(t ϕ + t 2(ψ0 − gn−1))
Ln = A G (u0 ) · (ψ0 − gn−1) + + Ln−1
t2

H(t ϕ + t 2(ψ0 − gn−1))
= A−1 A(ψ0 − gn−1) + A−1 + gn−1
t2

H(t ϕ + t 2 (ψ0 − gn−1))
= ψ0 − gn−1 + A−1 + gn−1
t2

H(t ϕ + t 2(ψ0 − gn−1))
−1
= ψ0 + A
t2
= {ψ0 + wn + v | v ∈ L0 }.
Here wn ∈ U is such that

H(t ϕ + t 2 (ψ0 − gn−1))
wn = A−1 ,
t2
that is,

H(t ϕ + t 2 (ψ0 − gn−1))
A(wn ) = ,
t2
so that

H(t ϕ + t 2(ψ0 − gn−1))
G (u0 ) · wn = .
t2
Choose gn ∈ Ln such that
gn − gn−1 U ≤ 2 Ln − Ln−1 .
This is possible since
Ln − Ln−1 = inf { g − gn−1 U }.

g∈Ln
Observe that we may write
Ln−1 = A−1 [A(ḡn−1)] = A−1 [G (u0 ) · gn−1].
Thus

−1 H(t ϕ + t 2 (ψ0 − gn−1))
Ln = A G (u0 ) · (ψ0 − gn−1) + + G (u0 ) · gn−1 .
t2
By analogy

H(t ϕ + t 2(ψ0 − gn−2))
Ln−1 = A−1 G (u0 ) · (ψ0 − gn−2) + + G
(u 0 ) · g n−2 .
t2
Observe that
H(t ϕ + t 2(ψ0 − gn−1)) − H(t ϕ + t 2(ψ0 − gn−2))

= G(u0 + t ϕ + t 2(ψ0 − gn−1)) − G(u0)
−G (u0 ) · (t ϕ + t 2 (ψ0 − gn−1))
−G(u0 + t ϕ + t 2 (ψ0 − gn−2)) + G(u0 )
+G (u0 ) · (t ϕ + t 2 (ψ0 − gn−2))
= G(u0 + t ϕ + t 2(ψ0 − gn−1)) − G(u0 + t ϕ + t 2 (ψ0 − gn−2))

−G (u0 ) · (t 2 (−gn−1 + gn−2)), (11.47)
so that by the generalized mean value inequality we may write
H(t ϕ + t 2(ψ0 − gn−1)) − H(t ϕ + t 2(ψ0 − gn−2))

≤ sup G (u0 + t ϕ + t 2ψ0 − t 2 (h(gn−1 ) + (1 − h)gn−2)) − G (u0 )
h∈[0,1]
× t 2(−gn−1 + gn−2) U
< ε t 2 gn−1 − gn−2 U .
Therefore, similarly as above,
A−1
Ln − Ln−1 ≤ H(t ϕ + t 2 (ψ0 − gn−1)) − H(t ϕ + t 2(ψ0 − gn−2))
t2
< ε A−1 gn−1 − gn−2 U
< (r/4) gn−1 − gn−2 U
1
< gn−1 − gn−2 U . (11.48)
4
Thus,
1
gn − gn−1 U ≤ 2 Ln − Ln−1 < gn−1 − gn−2 U .
2
Finally
gn U = gn − gn−1 + gn−1 − gn−2 + gn−2 − . . . + g1 − g0 U

1 1
≤ g1 U 1 + + . . . + n < 2 g1 U < r. (11.49)
2 2
Thus gn U < r and
1
gn − gn−1 U < gn−1 − gn−2 U , ∀n ∈ N,
2
so that {gn } is a Cauchy sequence, and since U is a Banach space there exists g ∈ U
such that
gn → g, in norm, as n → ∞.
Hence
Ln → L = ḡ, in norm, as n → ∞,
so that

H(t ϕ + t 2 (ψ0 − g))
Ln − Ln−1 → L − L = θ = A−1 G (u0 ) · (ψ0 − g) + .
t2
Since A−1 is a bijection, denoting ψ̃0 = (ψ0 − g), we get
H(t ϕ + t 2(ψ˜0 ))
G (u0 ) · ψ˜0 + =θ
t2
Clarifying the dependence on t we denote ψ˜0 = ψ̃0 (t) where as above mentioned,
t ∈ R is such that
δ
0 < |t| < .
2(1 + ϕ U + K1 )
Therefore
G(u0 + t ϕ + t 2 ψ̃0 (t)) = θ .
Observe also that t 2 ψ̃0 (t) U = t 2 (ψ0 (t) − g) U ≤ t 2 (K1 + r) ≤ t 2 (K1 + 1) so
that t 2 ψ̃0 (t) → θ as t → 0. Thus, by defining t 2 ψ̃0 (t)|t=0 = θ (observe that in prin-
ciple such a function would not be defined at t = 0), we obtain
2
d(t 2 ψ̃0 (t)) t ψ̃0 (t) − θ
|t=0 = lim = θ,
dt t→0 t
considering that
t ψ̃0 (t) U ≤ |t|(K1 + 1) → 0, as t → 0.
Finally, defining
φ (t) = F(u0 + t ϕ + t 2 ψ̃0 (t)),
from the hypotheses, we have that there exists a suitable t˜2 > 0 such that
φ (0) = F(u0 ) ≤ F(u0 + t ϕ + t 2ψ̃0 (t)) = φ (t), ∀|t| < t˜2 ,
also from the hypothesis we get
φ (0) = δ F(u0 , ϕ ) = 0,
that is,
ϕ , F (u0 )U = 0, ∀ϕ such that G (u0 ) · ϕ = θ .
In the next lines as usual N[G (u0 )] and R[G (u0 )] denote the null space and the
range of G (u0 ), respectively. Thus F (u0 ) is orthogonal to the null space of G (u0 ),
which we denote by
F (u0 ) ⊥ N[G (u0 )].
Since R[G (u0 )] is closed, we get F (u0 ) ∈ R([G (u0 )]∗ ), that is, there exists z∗0 ∈ Z ∗
such that
F (u0 ) = [G (u0 )]∗ (−z∗0 ).
11.4 Some Examples Concerning Inequality Constraints 299
11.4 Some Examples Concerning Inequality Constraints
In this section we assume the hypotheses of the last theorem for F and G spec-
ified below. As an application of this same result, consider the problem of locally
minimizing F(u) subject to G1 (u) = θ and G2 (u) ≤ θ , where F : U → R, U being a
function Banach space, G1 : U → [L p (Ω )]m1 , G2 : U → [L p (Ω )]m2 where 1 < p < ∞,
and Ω is an appropriate subset of RN . We refer to the simpler case in which the par-
tial order in [L p (Ω )]m2 is defined by u = {ui } ≥ θ if and only if ui ∈ L p (Ω ) and
ui (x) ≥ 0 a.e. in Ω , ∀i ∈ {1, . . . , m2 }.
Observe that defining
F̃(u, v) = F(u),

G(u, v) = {(G1 )i (u)}m1 ×1 , {(G2 )i (u) + v2i }m2 ×1
it is clear that (locally) minimizing F̃(u, v) subject to G(u, v) = (θ , θ ) is equivalent
to the original problem. We clarify the domain of F̃ is denoted by U × Y , where
Y = {v measurable such that v2i ∈ L p (Ω ), ∀i ∈ {1, . . . , m2 }}.
Therefore, if u0 is a local minimum for the original constrained problem, then

for an appropriate and easily defined v0 , we have that (u0 , v0 ) is a point of local
minimum for the extended constrained one, so that by the last theorem there exists
a Lagrange multiplier z∗0 = (z∗1 , z∗2 ) ∈ [Lq (Ω )]m1 × [Lq (Ω )]m2 where 1/p + 1/q = 1
and
F̃ (u0 , v0 ) + [G (u0 , v0 )]∗ (z∗0 ) = (θ , θ ),
that is,
F (u0 ) + [G1 (u0 )]∗ (z∗1 ) + [G2(u0 )]∗ (z∗2 ) = θ , (11.50)
and
(z∗2 )i v0i = θ , ∀i ∈ {1, . . . , m2 }.
In particular for almost all x ∈ Ω , if x is such that v0i (x)2 > 0, then z∗2i (x) = 0,
and if v0i (x) = 0, then (G2 )i (u0 (x)) = 0, so that (z∗2 )i (G2 )i (u0 ) = 0, a.e. in Ω , ∀i ∈
{1, . . . , m2 }.
Furthermore, consider the problem of minimizing F1 (v) = F̃(u0 , v) = F(u0 ) sub-
ject {G2i (u0 ) + v2i } = θ . From the above such a local minimum is attained at v0 .
Thus, from the stationarity of F1 (v) + z∗2 , {(G2 )i (u0 ) + v2i }[L p (Ω )]m2 at v0 and the
standard necessary conditions for the case of convex (in fact quadratic) constraints
we get (z∗2 )i ≥ 0 a.e. in Ω , ∀i ∈ {1, . . . , m2 }, that is, z∗2 ≥ θ .
Summarizing, for the order in question, the first-order necessary optimality con-
ditions are given by (12.37), z∗2 ≥ θ and (z∗2 )i (G2 )i (u0 ) = θ , ∀i ∈ {1, . . . , m2 } (so that
z∗2 , G2 (u0 )[L p (Ω )]m2 = 0), G1 (u0 ) = θ , and G2 (u0 ) ≤ θ .
Remark 11.4.1. For the case U = Rn and Rmk replacing [L p (Ω )]mk , for k ∈ {1, 2},
the conditions (z∗2 )i vi = θ mean that for the constraints not active (e.g., vi = 0)
the corresponding coordinate (z∗2 )i of the Lagrange multiplier is 0. If vi = 0, then
(G2 )i (u0 ) = 0, so that in any case (z∗2 )i (G2 )i (u0 ) = 0.
Summarizing, for this last mentioned case, we have obtained the standard
necessary optimality conditions: (z∗2 )i ≥ 0, and (z∗2 )i (G2 )i (u0 ) = 0, ∀i ∈ {1, . . . , m2 }.
11.5 The Lagrange Multiplier Theorem for Equality

and Inequality Constraints
In this section we develop more rigorous results concerning the Lagrange multi-
plier theorem for the case involving equalities and inequalities.
Theorem 11.1. Let U, Z1 , Z2 be Banach spaces. Consider a cone C in Z2 as specified

above and such that if z1 ≤ θ and z2 < θ , then z1 + z2 < θ , where z ≤ θ means that
z ∈ −C and z < θ means that z ∈ (−C)◦ . The concerned order is supposed to be also
that if z < θ , z∗ ≥ θ ∗ and z∗ = θ , then z, z∗ Z2 < 0. Furthermore, assume u0 ∈ U is a
point of local minimum for F : U → R subject to G1 (u) = θ and G2 (u0 ) ≤ θ , where
G1 : U → Z1 , G2 : U → Z2 and F are Fréchet differentiable at u0 ∈ U. Suppose also
G1 (u0 ) is onto and that there exist α > 0, K > 0 such that if ϕ U < α , then
G1 (u0 + ϕ ) − G1(u0 ) ≤ K ϕ U .
Finally, suppose there exists ϕ0 ∈ U such that
G1 (u0 ) · ϕ0 = θ
and
G2 (u0 ) · ϕ0 < θ .
Under such hypotheses, there exists a Lagrange multiplier z∗0 = (z∗1 , z∗2 ) ∈ Z1∗ × Z2∗
such that
F (u0 ) + [G1 (u0 )]∗ (z∗1 ) + [G2(u0 )]∗ (z∗2 ) = θ ,
z∗2 ≥ θ ∗ ,
and
G2 (u0 ), z∗2 Z2 = 0.
Proof. Let ϕ ∈ U be such that
G1 (u0 ) · ϕ = θ
and
G2 (u0 ) · ϕ = v − λ G2(u0 ),
for some v ≤ θ and λ ≥ 0.
Select α ∈ (0, 1) and define
ϕα = αϕ0 + (1 − α )ϕ .
11.5 The Lagrange Multiplier Theorem for Equality and Inequality Constraints 301
Observe that G1 (u0 ) = θ and G1 (u0 ) · ϕα = θ so that as in the proof of the La-
grange multiplier Theorem 11.3.2 we may find K1 > 0, ε > 0 and ψ0 (t) such that
G1 (u0 + t ϕα + t 2 ψ0 (t)) = θ , ∀|t| < ε ,
and
ψ0 (t) U < K1 , ∀|t| < ε .
Observe that
G2 (u0 ) · ϕα
= α G2 (u0 ) · ϕ0 + (1 − α )G2(u0 ) · ϕ
= α G2 (u0 ) · ϕ0 + (1 − α )(v − λ G2(u0 ))
= α G2 (u0 ) · ϕ0 + (1 − α )v − (1 − α )λ G2(u0 ))
= v0 − λ0G2 (u0 ), (11.51)
where
λ0 = (1 − α )λ ,
and
v0 = α G2 (u0 ) · ϕ0 + (1 − α )v < θ .
Hence, for t > 0,
G2 (u0 + t ϕα + t 2 ψ0 (t)) = G2 (u0 ) + G2 (u0 ) · (t ϕα + t 2ψ0 (t)) + r(t),
where
r(t)
lim = 0.
t→0+ t
Therefore from (11.51) we obtain
G2 (u0 + t ϕα + t 2 ψ0 (t)) = G2 (u0 ) + tv0 − t λ0G2 (u0 ) + r1 (t),
where
r1 (t)
lim = 0.
t→0+ t
Observe that there exists ε1 > 0 such that if 0 < t < ε1 < ε , then
r1 (t)
v0 + < θ,
t
and
G2 (u0 ) − t λ0G2 (u0 ) = (1 − t λ0)G2 (u0 ) ≤ θ .
Hence
G2 (u0 + t ϕα + t 2ψ0 (t)) < θ , if 0 < t < ε1 .
From this there exists 0 < ε2 < ε1 such that
F(u0 + t ϕα + t 2ψ0 (t)) − F(u0 )

= t ϕα + t 2ψ0 (t), F (u0 )U + r2 (t) ≥ 0, (11.52)
where
|r2 (t)|
lim = 0.
t→0+ t
Dividing the last inequality by t > 0 we get
ϕα + t ψ0 (t), F (u0 )U + r2 (t)/t ≥ 0, ∀0 < t < ε2 .
Letting t → 0+ we obtain
ϕα , F (u0 )U ≥ 0.
Letting α → 0+ , we get
ϕ , F (u0 )U ≥ 0,
if
G1 (u0 ) · ϕ = θ ,
and
G2 (u0 ) · ϕ = v − λ G2(u0 ),
for some v ≤ θ and λ ≥ 0. Define
A = {(ϕ , F (u0 )U + r, G1 (u0 ) · ϕ , G2 (u0 )ϕ − v + λ G(u0)),

ϕ ∈ U, r ≥ 0, v ≤ θ , λ ≥ 0}. (11.53)
Observe that A is a convex set with a nonempty interior.

If
G1 (u0 ) · ϕ = θ ,
and
G2 (u0 ) · ϕ − v + λ G2(u0 ) = θ ,
with v ≤ θ and λ ≥ 0 then
ϕ , F (u0 )U ≥ 0,
so that
ϕ , F (u0 )U + r ≥ 0.
Moreover, if
ϕ , F (u0 ) + r = 0,
with r ≥ 0,
G1 (u0 ) · ϕ = θ ,
and
G2 (u0 ) · ϕ − v + λ G2(u0 ) = θ ,
11.5 The Lagrange Multiplier Theorem for Equality and Inequality Constraints 303
with v ≤ θ and λ ≥ 0, then we have
ϕ , F (u0 )U ≥ 0,
so that
ϕ , F (u0 )U = 0,
and r = 0. Hence (0, θ , θ ) is on the boundary of A. Therefore, by the Hahn–Banach
theorem, geometric form, there exists
(β , z∗1 , z∗2 ) ∈ R × Z1∗ × Z2∗
such that
(β , z∗1 , z∗2 ) = (0, θ , θ )
and
β (ϕ , F (u0 )U + r) + G1 (u0 ) · ϕ , z∗1 Z1

+ G2 (u0 ) · ϕ − v + λ G2(u0 ), z∗2 Z2 ≥ 0, (11.54)
∀ ϕ ∈ U, r ≥ 0, v ≤ θ , λ ≥ 0. Suppose β = 0. Fixing all variable except v we get

z∗2 ≥ θ . Thus, for ϕ = cϕ0 with arbitrary c ∈ R, v = θ , λ = 0, if z∗2 = θ , then G2 (u0 )·
ϕ0 , z∗2 Z2 < 0, so that we get z∗2 = θ . Since G1 (u0 ) is onto, a similar reasoning lead
us to z∗1 = θ , which contradicts (β , z∗1 , z∗2 ) = (0, θ , θ ).
Hence, β = 0, and fixing all variables except r we obtain β > 0. There is no loss
of generality in assuming β = 1.
Again fixing all variables except v, we obtain z∗2 ≥ θ . Fixing all variables except
λ , since G2 (u0 ) ≤ θ we get
G2 (u0 ), z∗2 Z2 = 0.
Finally, for r = 0, v = θ , λ = 0, we get
ϕ , F (u0 )U + G1 (u0 )ϕ , z∗1 Z1 + G2 (u0 ) · ϕ , z∗2 Z2 ≥ 0, ∀ϕ ∈ U,
that is, since obviously such an inequality is valid also for −ϕ , ∀ϕ ∈ U, we obtain
ϕ , F (u0 )U + ϕ , [G1 (u0 )]∗ (z∗1 )U + ϕ , [G2 (u0 )]∗ (z∗2 )U = 0, ∀ϕ ∈ U,
so that
F (u0 ) + [G1 (u0 )]∗ (z∗1 ) + [G2(u0 )]∗ (z∗2 ) = θ .
11.6 Second-Order Necessary Conditions
In this section we establish second-order necessary conditions for a class of con-

strained problems in Banach spaces. We highlight the next result is particularly ap-
plicable to optimization in Rn .
Theorem 11.2. Let U, Z1 , Z2 be Banach spaces. Consider a cone C in Z2 as above
specified and such that if z1 ≤ θ and z2 < θ , then z1 +z2 < θ , where z ≤ θ means that
z ∈ −C and z < θ means that z ∈ (−C)◦ . The concerned order is supposed to be also
that if z < θ , z∗ ≥ θ ∗ and z∗ = θ , then z, z∗ Z2 < 0. Furthermore, assume u0 ∈ U is a
point of local minimum for F : U → R subject to G1 (u) = θ and G2 (u0 ) ≤ θ , where
G1 : U → Z1 , G2 : U → (Z2 )k , and F are twice Fréchet differentiable at u0 ∈ U.
Assume G2 (u) = {(G2 )i (u)} where (G2 )i : U → Z2 , ∀i ∈ {1, . . . , k} and define
A = {i ∈ {1, . . . , k} : (G2 )i (u0 ) = θ },
and also suppose that (G2 )i (u0 ) < θ , if i ∈ A. Moreover, suppose {G1 (u0 ), {(G2 )i
(u0 )}i∈A } is onto and that there exist α > 0, K > 0 such that if ϕ U < α , then
G̃ (u0 + ϕ ) − G̃(u0 ) ≤ K ϕ U ,
where
G̃(u) = {G1 (u), {(G2 )i (u)}i∈A }.
Finally, suppose there exists ϕ0 ∈ U such that
G1 (u0 ) · ϕ0 = θ
and
G2 (u0 ) · ϕ0 < θ .
Under such hypotheses, there exists a Lagrange multiplier z∗0 = (z∗1 , z∗2 ) ∈ Z1∗ × (Z2∗ )k
such that
F (u0 ) + [G1 (u0 )]∗ (z∗1 ) + [G2(u0 )]∗ (z∗2 ) = θ ,
z∗2 ≥ (θ ∗ , . . . , θ ∗ ) ≡ θk∗ ,
and
(G2 )i (u0 ), (z∗2 )i Z = 0, ∀i ∈ {1, . . . , k},
(z∗2 )i = θ ∗ , if i ∈ A,
Moreover, defining
L(u, z∗1 , z∗2 ) = F(u) + G1(u), z∗1 Z1 + G2 (u), z∗2 Z2 ,
we have that
δuu
2
L(u0 , z∗1 , z∗2 ; ϕ ) ≥ 0, ∀ϕ ∈ V0 ,
where
11.6 Second-Order Necessary Conditions 305
V0 = {ϕ ∈ U : G1 (u0 ) · ϕ = θ , (G2 )i (u0 ) · ϕ = θ , ∀i ∈ A}.
Proof. Observe that A is defined by
A = {i ∈ {1, . . . , k} : (G2 )i (u0 ) = θ }.
Observe also that (G2 )i (u0 ) < θ , if i ∈ A.

Hence the point u0 ∈ U is a local minimum for F(u) under the constraints
G1 (u) = θ , and (G2 )i (u) ≤ θ , ∀i ∈ A.
From the last Theorem 11.1 for such an optimization problem there exists a La-
grange multiplier (z∗1 , {(z∗2 )i∈A }) such that (z∗2 )i ≥ θ ∗ , ∀i ∈ A, and
F (u0 ) + [G1(u0 )]∗ (z∗1 ) + ∑ [(G2 )i (u0 )]∗ ((z∗2 )i ) = θ . (11.55)
i∈A
The choice (z∗2 )i = θ , if i ∈ A leads to the existence of a Lagrange multiplier

(z∗1 , z∗2 ) =
(z∗1 , {(z∗2 )i∈A , (z∗2 )i∈A }) such that
z∗2 ≥ θk∗
and
(G2 )i (u0 ), (z∗2 )i Z = 0, ∀i ∈ {1, . . . , k}.
Let ϕ ∈ V0 , that is, ϕ ∈ U,
G1 (u0 )ϕ = θ
and
(G2 )i (u0 ) · ϕ = θ , ∀i ∈ A.
Recall that G̃(u) = {G1 (u), (G2 )i∈A (u)} and therefore, similarly as in the proof
of the Lagrange multiplier Theorem 11.3.2, we may obtain ψ0 (t), K > 0 and ε > 0
such that
G̃(u0 + t ϕ + t 2 ψ0 (t)) = θ , ∀|t| < ε ,
and
ψ0 (t) ≤ K, ∀|t| < ε .
Also, if i ∈ A, we have that (G2 )i (u0 ) < θ , so that
(G2 )i (u0 + t ϕ + t 2 ψ0 (t)) = (G2 )i (u0 ) + Gi (u0 ) · (t ϕ + t 2ψ0 (t)) + r(t),
where
r(t)
lim = 0,
t→0 t
that is,
(G2 )i (u0 + t ϕ + t 2 ψ0 (t)) = (G2 )i (u0 ) + t(G2 )i (u0 ) · ϕ + r1 (t),

where
r1 (t)
= 0,
lim
t t→0
and hence there exists 0 < ε1 < ε such that
(G2 )i (u0 + t ϕ + t 2 ψ0 (t)) < θ , ∀|t| < ε1 < ε .
Therefore, since u0 is a point of local minimum under the constraint G(u) ≤ θ ,

there exists 0 < ε2 < ε1 such that
F(u0 + t ϕ + t 2ψ0 (t)) − F(u0 ) ≥ 0, ∀|t| < ε2 ,
so that
F(u0 + t ϕ + t 2ψ0 (t)) − F(u0 )

= F(u0 + t ϕ + t 2ψ0 (t)) − F(u0 )
" #
+G1 (u0 + t ϕ + t 2ψ0 (t)), z∗1 Z1 + ∑ (G2 )i (u0 + t ϕ + t 2 ψ0 (t)), (z∗2 )i Z2
i∈A
−G1 (u0 ), z∗1 Z1 ∑
− {(G2 )i (u0 ), (z∗2 )i Z2 }
i∈A
= F(u0 + t ϕ + t 2ψ0 (t)) − F(u0 )
+G1 (u0 + t ϕ + t 2ψ0 (t)), z∗1 Z1 − G1 (u0 ), z∗1 Z1
+G2 (u0 + t ϕ + t 2ψ0 (t)), z∗2 Z2 − G2 (u0 ), z∗2 Z2
= L(u0 + t ϕ + t 2ψ0 (t)), z∗1 , z∗2 ) − L(u0 , z∗1 , z∗2 )
1 2
= δu L(u0 , z∗1 , z∗2 ;t ϕ + t 2 ψ0 (t)) + δuu L(u0 , z∗1 , z∗2 ;t ϕ + t 2 ψ0 (t)) + r2 (t)
2
t2 2
= δuu L(u0 , z∗1 , z∗2 ; ϕ + t ψ0(t)) + r2 (t) ≥ 0, ∀|t| < ε2 .
2
where
lim |r2 (t)|/t 2 = 0.
t→0
To obtain the last inequality we have used
δu L(u0 , z∗1 , z∗2 ;t ϕ + t 2ψ0 (t)) = 0
Dividing the last inequality by t 2 > 0 we obtain

1 2
δ L(u0 , z∗1 , z∗2 ; ϕ + t ψ0(t)) + r2 (t)/t 2 ≥ 0, ∀0 < |t| < ε2 ,
2 uu
and finally, letting t → 0, we get
1 2
δ L(u0 , z∗1 , z∗2 ; ϕ ) ≥ 0.
2 uu
11.7 On the Banach Fixed Point Theorem 307
11.7 On the Banach Fixed Point Theorem
Now we recall a classical definition, namely, the Banach fixed theorem also
known as the contraction mapping theorem.
Definition 11.7.1. 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 ≤ α < 1 such
that
T (u) − T (v) U ≤ α u − v U , ∀u, v ∈ C.
Remark 11.7.2. Observe that if T (u) U ≤ α < 1 on a convex set C, then T is a

contraction mapping, since by the mean value inequality,
T (u) − T (v) U ≤ sup{ T (u) } u − v U , ∀u, v ∈ C.

u∈C
The next result is the base of our generalized method of lines.

Theorem 11.7.3 (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
u0 ∈ C such that u0 = T (u0 ). Moreover, for an arbitrary u1 ∈ C defining the sequence
u2 = T (u1 ) and un+1 = T (un ), ∀n ∈ N
we have
un → u0 , in norm, as n → +∞.
Proof. Let u1 ∈ C. Let {un} ⊂ C be defined by
un+1 = T (un ), ∀n ∈ N.
Hence, reasoning inductively
un+1 − un U = T (un ) − T (un−1 ) U

≤ α un − un−1 U
≤ α 2 un−1 − un−2 U
≤ ......
≤ α n−1 u2 − u1 U , ∀n ∈ N. (11.56)
Thus, for p ∈ N, we have
un+p − un U
= un+p − un+p−1 + un+p−1 − un+p−2 + . . . − un+1 + un+1 − un U
≤ un+p − un+p−1 U + un+p−1 − un+p−2 U + . . . + un+1 − un U
≤ (α n+p−2 + α n+p−3 + . . . + α n−1) u2 − u1 U
≤ α n−1 (α p−1 + α p−2 + . . . + α 0 ) u2 − u1 U

∞
≤ α n−1 ∑ αk u2 − u1 U
k=0
α n−1
≤ u2 − u1 U (11.57)
1−α
Denoting n + p = m, we obtain
α n−1
um − un U ≤ u2 − u1 U , ∀m > n ∈ N.
1−α
Let ε > 0. Since 0 ≤ α < 1, there exists n0 ∈ N such that if n > n0 then
α n−1
0≤ u2 − u1 U < ε ,
1−α
so that
um − un U < ε , if m > n > n0 .
From this we may infer that {un } is a Cauchy sequence, and since U is a Banach
space, there exists u0 ∈ U such that
un → u0 , in norm, as n → ∞.
Observe that
u0 − T (u0 ) U = u0 − un + un − T (u0 ) U
≤ u0 − un U + un − T (u0 ) U
≤ u0 − un U + α un−1 − u0 U
→ 0, as n → ∞. (11.58)
Thus u0 − T (u0 ) U = 0.
Finally, we prove the uniqueness. Suppose u0 , v0 ∈ C are such that
u0 = T (u0 ) and v0 = T (v0 ).
Hence,
u0 − v0 U = T (u0 ) − T (v0 ) U
≤ α u0 − v0 U . (11.59)
From this we get

u0 − v0 ||U ≤ 0,
that is,
u0 − v0 U = 0.
11.8 Sensitivity Analysis 309
11.8 Sensitivity Analysis
11.8.1 Introduction
In this section we state and prove the implicit function theorem for Banach
spaces. A similar result may be found in Ito and Kunisch [40], page 31.
We emphasize the result found in [40] is more general; however, the proof present
here is almost the same for a simpler situation. The general result found in [40] is
originally from Robinson [53].
Theorem 11.3 (Implicit Function Theorem). Let V,U,W be Banach spaces. Given
a function F̂ : V × U → W , suppose (x0 , u0 ) ∈ V × U is such that F̂(x0 , u0 ) = θ .
Assume F̂ is Fréchet differentiable, F̂x (x0 , u0 ) is continuous, and [F̂x (x0 , u0 )]−1 is a
(single-valued) bounded linear operator so that we denote [F̂x (x0 , u0 )]−1 = ρ > 0.
Under such hypotheses, for each ε > 0, there exist a neighborhood Uε of u0 , a
neighborhood Vε of x0 , and a function x : Uε → Vε such that for each u ∈ Uε , x(u)
is the unique solution of
F̂(x, u) = θ ,
that is,
F̂(x(u), u) = θ .
Moreover, for each u, v ∈ Uε , we have
x(u) − x(v) ≤ (ρ + ε ) F̂(x(v), u) − F̂(x(v), v) .
Finally, if F̂(x, u) − F̂(x, v) ≤ K u − v , ∀(x, u) ∈ Vε × Uε , then
x(u) − x(v) ≤ K0 u − v , ∀u, v ∈ Uε ,
where K0 = K(ρ + ε ).
Proof. Let ε > 0. Choose δ > 0 such that

ε
ρδ < .
(ρ + ε )
Define
T (x) = F̂(x0 , u0 ) + F̂x(x0 , u0 )(x − x0 )

= F̂x (x0 , u0 )(x − x0 ), (11.60)
and
h(x, u) = F̂(x0 , u0 ) + F̂x(x0 , u0 )(x − x0 ) − F̂(x, u)

= F̂x (x0 , u0 )(x − x0 ) − F̂(x, u). (11.61)
Select a ball Uε about u0 and a closed ball Vε of radius r > 0 about x0 such that
for each u ∈ Uε and x ∈ Vε we have
F̂x (x, u) − F̂x (x0 , u0 ) ≤ δ ,
ρ F̂(x0 , u) − F̂(x0 , u0 ) ≤ (1 − ρδ )r.

For each u ∈ Uε define
φu (x) = T −1 (h(x, u)).
Fix u ∈ U. Observe that for x1 , x2 ∈ Vε we have
φu (x1 ) − φu (x2 ) ≤ T −1 h(x1, u) − h(x2, u)

= ρ h(x1 , u) − h(x2, u)
1

= ρ h
0 x 1 (x + t(x 2 − x 1 ), u) · (x 1 − x 2 ) dt

≤ ρδ x1 − x2 , (11.62)
so that since 0 < ρδ < 1 we may infer that φu (x) is a contractor. Observe also that
x0 = T −1 (θ ), so that
φu (x0 ) − x0 ≤ ρ h(x0, u) − θ
= ρ F̂(x0 , u) − F̂(x0 , u0 )
≤ (1 − ρδ )r. (11.63)
Hence, for x ∈ Vε , we obtain
φu (x) − x0 ≤ φu (x) − φu (x0 )

+ φu (x0 ) − x0
≤ ρδ x − x0 + (1 − ρδ )r ≤ r. (11.64)
Therefore φu (x) ∈ Vε , ∀x ∈ Vε so that from this, (11.62) and the Banach fixed
point theorem, φu has a unique fixed point in Vε , which we denote by x(u).
Thus,
x(u) = φu (x(u))
= T −1 (h(x(u), u))
= T −1 (F̂x (x0 , u0 )(x(u) − x0 ) − F̂(x(u), u))
= [F̂x (x0 , u0 )]−1 (F̂x (x0 , u0 )(x(u) − x0 ) − F̂(x(u), u)) + x0
= x(u) − x0 + x0 − [F̂x (x0 , u0 )]−1 (F̂(x(u), u))
= x(u) − [F̂x (x0 , u0 )]−1 (F̂(x(u), u)). (11.65)
From this,
[F̂x (x0 , u0 )]−1 (F̂(x(u), u)) = θ ,

so that
F̂(x(u), u) = F̂x (x0 , u0 )[F̂x (x0 , u0 )]−1 (F̂(x(u), u)) = F̂x (x0 , u0 )θ = θ ,
that is,
F̂(x(u), u) = θ .
Also as a consequence of the Banach fixed point theorem, we have that
x(u) − x ≤ (1 − ρδ )−1 φu (x) − x .
Now observe that for u, v ∈ Uε , with x = x(v) in the last inequality, we get
x(u) − x(v) ≤ (1 − ρδ )−1 φu (x(v)) − x(v) .
However, x(v) = φv (x(v)), so that from this and the last inequality, we obtain
x(u) − x(v) ≤ (1 − ρδ )−1 φu (x(v)) − φv (x(v))

≤ (1 − ρδ )−1ρ h(x(v), u) − h(x(v), v)
= ρ (1 − ρδ )−1 F̂(x(v), u) − F̂(x(v), v) . (11.66)
Since ρ (1 − ρδ )−1 ≤ ρ + ε , the proof is complete.
11.8.2 The Main Results About Gâteaux Differentiability
Again let V,U be Banach spaces and let F : V × U → R be a functional. Fix

u ∈ U and consider the problem of minimizing F(x, u) subject to G(x, u) ≤ θ and
H(x, u) = θ . Here the order and remaining details on the primal formulation are the
same as those indicated in Section 11.4.
Hence, for the specific case in which
G : V × U → [L p (Ω )]m1
and
H : V × U → [L p (Ω )]m2 ,
(the cases in which the co-domains of G and H are Rm1 and Rm2 , respectively, are
dealt similarly), we redefine the concerned optimization problem, again for a fixed
u ∈ U, by minimizing F(x, u) subject to
{Gi (x, u) + v2i } = θ ,
and
H(x, u) = θ .
At this point we assume F(x, u), G̃(x, u, v) = {Gi (x, u) + v2i } ≡ G(u) + v2 (from
now on we use this general notation) and H(x, u) satisfy the hypotheses of the La-
grange multiplier Theorem 11.3.2.
Hence, for the fixed u ∈ U, we assume there exists an optimal x ∈ V which locally
minimizes F(x, u) under the mentioned constraints.
From Theorem 11.3.2 there exist Lagrange multipliers λ1 , λ2 such that denoting
[L p (Ω )]m1 and [L p (Ω )]m2 simply by L p and defining
F̃(x, u, λ1 , λ2 , v) = F(x, u) + λ1, G(u) + v2 L p + λ2, H(x, u)L p ,
the following necessary conditions hold:
F̃x (x, u) = Fx (x, u) + λ1 · Gx (x, u) + λ2 · Hx (x, u) = θ , (11.67)
G(x, u) + v2 = θ , (11.68)
λ1 · v = θ , (11.69)
λ1 ≥ θ , (11.70)
H(x, u) = θ . (11.71)
Clarifying the dependence on u, we denote the solution x, λ1 , λ2 , v by x(u),
λ1 (u), λ2 (u), v(u), respectively. In particular, we assume that for a u0 ∈ U,
x(u0 ), λ1 (u0 ), λ2 (u0 ), v(u0 ) satisfy the hypotheses of the implicit function theo-
rem. Thus, for any u in an appropriate neighborhood of u0 , the corresponding
x(u), λ1 (u), λ2 (u), v(u) are uniquely defined.
We emphasize that from now on the main focus of our analysis is to evaluate vari-
ations of the optimal x(u), λ1 (u), λ2 (u), v(u) with variations of u in a neighborhood
of u0 .
For such an analysis, the main tool is the implicit function theorem and its main
hypothesis is satisfied through the invertibility of the matrix of Fréchet second
derivatives.
Hence, denoting x0 = x(u0 ), (λ1 )0 = λ1 (u0 ), (λ2 )0 = λ2 (u0 ), v0 = v(u0 ), and
A1 = Fx (x0 , u0 ) + (λ1 )0 · Gx (x0 , u0 ) + (λ2)0 · Hx (x0 , u0 ),
A2 = G(x0 , u0 ) + v20
A3 = H(x0 , u0 ),
A4 = (λ1 )0 · v0 ,
we reiterate to assume that
A1 = θ , A2 = θ , A3 = θ , A4 = θ ,
and M −1 to represent a bounded linear operator, where

⎡ ⎤
(A1 )x (A1 )λ1 (A1 )λ2 (A1 )v
⎢ (A2 )x (A2 )λ1 (A2 )λ2 (A2 )v ⎥
M=⎢
⎣ (A3 )x
⎥ (11.72)
(A3 )λ1 (A3 )λ2 (A3 )v ⎦
(A4 )x (A4 )λ1 (A4 )λ2 (A4 )v
where the derivatives are evaluated at (x0 , u0 , (λ1 )0 , (λ2 )0 , v0 ) so that

⎡ ⎤
A Gx (x0 , u0 ) Hx (x0 , u0 ) θ
⎢ Gx (x0 , u0 ) θ θ 2v0 ⎥
M=⎢ ⎣ Hx (x0 , u0 ) θ
⎥
⎦ (11.73)
θ θ
θ v0 θ (λ1 )0
where
A = Fxx (x0 , u0 ) + (λ1 )0 · Gxx (x0 , u0 ) + (λ2 )0 · Hxx (x0 , u0 ).
Moreover, also from the implicit function theorem,
(x(u), λ1 (u), λ2 (u), v(u)) − (x(u0), λ1 (u0 ), λ2 (u0 ), v(u0 )) ≤ K u − u0 , (11.74)
for some appropriate K > 0, ∀u ∈ Br (u0 ), for some r > 0.

Beyond assuming F̃ to be twice Fréchet differentiable we suppose
F̃xx
is continuous in a neighborhood of u0 so that from (11.74) there exists K1 > 0 such

that
F̃xx (x(u), u, λ (u), v(u)) ≤ K1 , ∀u ∈ Br1 (u0 ), (11.75)
for some appropriate K1 > 0, r1 > 0. We highlight to have denoted λ (u) =
(λ1 (u), λ2 (u)).
Let ϕ ∈ [C∞ (Ω )]k ∩U, where k depends on the vectorial expression of U.
At this point we will be concerned with the following Gâteaux variation
evaluation:
δu F̃(x(u0 ), u0 , λ (u0 ), v(u0 ); ϕ ).
Observe that
δu F̃(x(u0 ), u0 , λ (u0 ), v(u0 ); ϕ )

F̃(x(u0 + εϕ ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ ))
= lim
ε →0 ε

F̃(x(u0 ), u0 , λ (u0 ), v(u0 ))
− ,
ε
so that
δu F̃(x(u0 ), u0 , λ (u0 ), v(u0 ); ϕ )


= lim
ε →0 ε
F̃(x(u0 ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ ))
−
ε
F̃(x(u0 ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ ))
+
ε

F̃(x(u0 ), u0 , λ (u0 ), v(u0 ))
− .
ε
However,


ε

F̃(x(u0 ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ ))
−
ε

≤ F̃x (x(u0 + εϕ ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ )) K ϕ

+ sup F̃xx (x(u0 + t εϕ ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ )) K 2 ϕ 24 ε
t∈[0,1]
≤ K1 K 2 ϕ 24 ε
→ 0, as ε → 0.
In these last inequalities we have used

x(u0 + εϕ ) − x(u0)

lim sup ≤ K ϕ ,
ε
ε →0
and
F̃x (x(u0 + εϕ ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ )) = θ .
On the other hand,

F̃(x(u0 ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ ))
ε

F̃(x(u0 ), u0 , λ (u0 ), v(u0 ))
−
ε

F̃(x(u0 ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ ))
=
ε
F̃(x(u0 ), u0 + εϕ , λ (u0 ), v(u0 ))
−
ε
F̃(x(u0 ), u0 + εϕ , λ (u0 ), v(u0 ))

+
ε

F̃(x(u0 ), u0 , λ (u0 ), v(u0 ))
−
ε
Now observe that

F̃(x(u0 ), u0 + εϕ , λ (u0 + εϕ ), v(u0 + εϕ ))
ε
F̃(x(u0 ), u0 + εϕ , λ (u0 ), v(u0 ))
−
ε
λ1 (u0 + εϕ ), G(x(u0 ), u0 + εϕ ) + v(u0 + εϕ )2 L p
=
ε
λ1 (u0 ), G(x(u0 ), u0 + εϕ ) + v(u0)2 L p
−
ε
λ2 (u0 + εϕ ) − λ2(u0 ), H(x(u0 ), u0 + εϕ )L p
+ . (11.76)
ε
Also,


ε

λ1 (u0 ), G(x(u0 ), u0 + εϕ ) + v(u0)2 L p
−
ε

≤
ε

λ1 (u0 ), G(x(u0 ), u0 + εϕ ) + v(u0 + εϕ )2 L p
−
ε

λ1 (u0 ), G(x(u0 ), u0 + εϕ ) + v(u0 + εϕ )2 L p
+
ε

λ1 (u0 ), G(x(u0 ), u0 + εϕ ) + v(u0)2 L p
−
ε
K ϕ
≤ε G(x(u0 ), u0 + εϕ ) + v(u0 + εϕ )2
ε
K ϕ ε
+ λ1 (u0 )(v(u0 + εϕ ) + v(u0))
ε
→ 0 as ε → 0.
To obtain the last inequalities we have used

λ1 (u0 + εϕ ) − λ1(u0 )

lim sup ≤ K ϕ ,
ε
ε →0
λ1 (u0 )v(u0 ) = θ ,
λ1 (u0 )v(u0 + εϕ ) → θ , as ε → 0,
and

λ1 (u0 )(v(u0 + εϕ )2 − v(u0)2 )

ε

λ1 (u0 )(v(u0 + εϕ ) + v(u0))(v(u0 + εϕ ) − v(u0))
=
ε
λ1 (u0 )(v(u0 + εϕ ) + v(u0)) K ϕ ε
≤
ε
→ 0, as ε → 0. (11.77)
Finally,

λ2 (u0 + εϕ ) − λ2(u0 ), H(x(u0 ), u0 + εϕ )L p

ε
K ε ϕ
≤ H(x(u0 ), u0 + εϕ )
ε
→ 0, as ε → 0.
To obtain the last inequalities we have used

λ2 (u0 + εϕ ) − λ2(u0 )
lim sup

≤ K ϕ ,

ε →0 ε
and
H(x(u0 ), u0 + εϕ ) → θ , as ε → 0.
From these last results, we get
δu F̃(x(u0 ), u0 , λ (u0 ), v(u0 ); ϕ )

F̃(x(u0 ), u0 + εϕ , λ (u0 ), v(u0 ))
= lim
ε →0 ε

F̃(x(u0 ), u0 , λ (u0 ), v(u0 ))
−
ε
= Fu (x(u0 ), u0 ), ϕ U + λ1 (u0 ) · Gu (x(u0 ), u0 ), ϕ L p
+λ2 (u0 ) · Hu (x(u0 ), u0 ), ϕ L p .
In the last lines we have proven the following corollary of the implicit function
theorem.
Corollary 11.1. Suppose (x0 , u0 , (λ1 )0 , (λ2 )0 , v0 ) is a solution of the system (11.67),
(11.68),(11.69), (11.71), and assume the corresponding hypotheses of the implicit
function theorem are satisfied. Also assume F̃(x, u, λ1 , λ2 , v) is such that the Fréchet
second derivative F̃xx (x, u, λ1 , λ2 ) is continuous in a neighborhood of
(x0 , u0 , (λ1 )0 , (λ2 )0 ).

Under such hypotheses, for a given ϕ ∈ [C∞ (Ω )]k , denoting
F1 (u) = F̃(x(u), u, λ1 (u), λ2 (u), v(u)),
we have
δ (F1 (u); ϕ )|u=u0

= Fu (x(u0 ), u0 ), ϕ U + λ1 (u0 ) · Gu (x(u0 ), u0 ), ϕ L p
+λ2 (u0 ) · Hu (x(u0 ), u0 ), ϕ L p .
Part III
Applications
Chapter 12
Duality Applied to Elasticity
12.1 Introduction
The first part of the present work develops a new duality principle applicable to
nonlinear elasticity. The proof of existence of solutions for the model in question
has been obtained in Ciarlet [21]. In earlier results (see [65] for details) the concept
of complementary energy is equivalently developed under the hypothesis of positive
definiteness of the stress tensor at a critical point. In more recent works, Gao [33,
34, 36] applied his triality theory to similar models obtaining duality principles for
more general situations, including the case of negative definite optimal stress tensor.
We emphasize our main objective is to establish a new and different duality
principle which allows the local optimal stress tensor to not be either positive or
negative definite. Such a result is a kind of extension of a more basic one obtained
in Toland [67]. Despite the fact we do not apply it directly, we follow a similar idea.
The optimality conditions are also new. We highlight the basic tools on convex anal-
ysis here used may be found in [25, 54, 67] for example. For related results about
the plate model presented in Ciarlet [22], see Botelho [11, 13].
In a second step, we present other two duality principles which qualitatively agree
with the triality theory proposed by Gao (see again [33, 34], for details).
However, our proofs again are obtained through more traditional tools of convex
analysis. Finally, in the last section, we provide a numerical example in which the
optimal stress field is neither positive nor negative definite.
At this point we start to describe the primal formulation.
Consider Ω ⊂ R3 an open, bounded, connected set, which represents the
reference volume of an elastic solid under the loads f ∈ L2 (Ω ; R3 ) and the boundary
loads fˆ ∈ L2 (Γ ; R3 ), where Γ denotes the boundary of Ω . The field of displace-
ments resulting from the actions of f and fˆ is denoted by u ≡ (u1 , u2 , u3 ) ∈ U,
where u1 , u2 , and u3 denote the displacements relating the directions x, y, and z,
respectively, in the Cartesian system (x, y, z).
Here U is defined by
U = {u = (u1 , u2 , u3 ) ∈ W 1,4 (Ω ; R3 ) | u = (0, 0, 0) ≡ θ on Γ0 } (12.1)
322 12 Duality Applied to Elasticity
and Γ = Γ0 ∪ Γ1 , Γ0 ∩ Γ1 = 0/ (for details about the Sobolev space U see [2]). We

assume |Γ0 | > 0 where |Γ0 | denotes the Lebesgue measure of Γ0 .
The stress tensor is denoted by {σi j }, where

1
σi j = Hi jkl (uk,l + ul,k + um,k um,l ) , (12.2)
2
{Hi jkl } = {λ δi j δkl + μ (δik δ jl + δil δ jk )},
{δi j } is the Kronecker delta and λ , μ > 0 are the Lamé constants (we assume they
are such that {Hi jkl } is a symmetric constant positive definite fourth-order tensor).
The boundary value form of the nonlinear elasticity model is given by
⎧
⎨ σi j, j + (σm j ui,m ), j + fi = 0, in Ω ,
u = θ, on Γ0 , (12.3)
⎩
σi j n j + σm j ui,m n j = fî , on Γ1 ,
where n denotes the outward normal to the surface Γ .
The corresponding primal variational formulation is represented by J : U → R,
where

1 1 1
J(u) = Hi jkl (ui, j + u j,i + um,i um, j ) (uk,l + ul,k + um,k um,l ) dx
2 Ω 2 2

−u, f L2 (Ω ;R3 ) − fî ui dΓ (12.4)
Γ1
where

u, f L2 (Ω ;R3 ) = fi ui dx.
Ω
Remark 12.1.1. Derivatives must be always understood in the distributional sense,

whereas boundary conditions are in the sense of traces. Moreover, from now on by
a regular boundary Γ of Ω , we mean regularity enough so that the standard Gauss–
Green formulas of integrations by parts and the well-known Sobolev imbedding and
trace theorems hold. Finally, we denote by θ the zero vector in appropriate function
spaces, the standard norm for L2 (Ω ) by · 2, and L2 (Ω ; R3×3 ) simply by L2 .
12.2 The Main Duality Principle
Now we prove the main result.

Theorem 12.2.1. Assume the statements of last section. In particular, let Ω ⊂ R3 be
an open, bounded, connected set with a regular boundary denoted by Γ = Γ0 ∪ Γ1 ,
where Γ0 ∩ Γ1 = 0/ and |Γ0 | > 0. Consider the functional (G◦ Λ ) : U → R expressed by
(G ◦ Λ )(u)

1 ui, j + u j,i um,i um, j uk,l + ul,k um,k um,l
= Hi jkl + + dx,
2 Ω 2 2 2 2
12.2 The Main Duality Principle 323
where Λ : U → Y × Y is given by
Λ u = {Λ1 u, Λ2 u},

ui, j + u j,i
Λ1 u =
2
and
Λ2 u = {um,i }.
Here
U = {u ∈ W 1,4 (Ω ; R3 ) | u = (u1 , u2 , u3 ) = θ on Γ0 }.
Define (F ◦ Λ2 ) : U → R, (GK ◦ Λ ) : U → R, and (G1 ◦ Λ2) : U → R by
K
(F ◦ Λ2 )(u) = um,i , um,i L2 (Ω ) ,
2
K
GK (Λ u) = GK (Λ1 u, Λ2 u) = G(Λ u) + um,i , um,i L2 (Ω ) ,
4
and
K
(G1 ◦ Λ2 )(u) = um,i , um,i L2 (Ω ) ,
4
respectively.
Also define
C = {u ∈ U | G∗∗
K (Λ u) = GK (Λ u)},
where K > 0 is an appropriate constant to be specified.

For f ∈ L2 (Ω ; R3 ), fˆ ∈ L2 (Γ ; R3 ), let J : U → R be expressed by

J(u) = G(Λ u) − u, f L2 (Ω ;R3 ) − fî ui dΓ . (12.5)
Γ1
inf {J(u)}
u∈C1

" ∗ ∗ ∗ ∗ ∗ ∗
#
≥ sup inf ∗ F (z ) − G̃K (σ , z , v) − G̃1(σ̃ , σ , z , v) ,
∗
(σ̃ ,σ ,v)∈Ỹ z ∈Y
where Ỹ = A∗ × Y ∗ × Ŷ ∗ , Y = Y ∗ = L2 (Ω ; R3×3 ) ≡ L2 ,
Ŷ ∗ = {v ∈ Y ∗ such that W ∗ (z∗ ) is positive definite in Ω }, (12.6)
and
3 (z∗ v )2
z∗mi z∗mi
− H i jkl z∗i j z∗kl − ∑
ij mj
W ∗ (z∗ ) = . (12.7)
K m,i=1 K/2
Here C1 = C2 ∩C, where
C2 = {u ∈ U | {ui, j } ∈ Ŷ ∗ }.
Furthermore,
A∗ = {σ̃ ∈ Y ∗ | σ̃i j, j + fi = 0 in Ω and σ̃i j n j = fî on Γ1 }.
Also
F ∗ (z∗ ) = sup {v2 , z∗ Y − F(v2 )}

v2 ∈Y
1 ∗ ∗
= z , z 2 , (12.8)
2K mi mi L (Ω )
where we recall that z∗i j = z∗ji . Through the relations
Qmi = (σi j + z∗i j )vm j + (K/2)vmi ,
we define
G̃∗K (σ , z∗ , v) = G∗K (σ + z∗ , Q)
= sup {v1 , σ + z∗ Y + v2 , QY − GK (v1 , v2 )}, (12.9)
(v1 ,v2 )∈Y ×Y
so that in particular,
G̃∗K (σ , z∗ , v) = G∗K (σ + z∗ , Q)

1
= H i jkl (σi j + z∗i j )(σkl + z∗kl ) dx
2 Ω

1 K
+ (σi j + z∗i j )vmi vm j dx + vmi , vmi L2 (Ω )
2 Ω 4
if (σ̃ , σ , v, z∗ ) ∈ B∗ . We emphasize to denote
B∗ = {(σ̃ , σ , v, z∗ ) ∈ [Y ∗ ]4 | σK (σ , z∗ ) is positive definite in Ω },
⎧ ⎫
⎨ σ11 + z∗11 + K/2 σ12 + z∗12 σ13 + z∗13 ⎬
σK (σ , z∗ ) = σ21 + z∗21 σ22 + z∗22 + K/2 σ23 + z∗23 , (12.10)
⎩ ⎭
σ31 + z∗31 σ32 + z∗32 σ33 + z∗33 + K/2
and
{H i jkl } = {Hi jkl }−1 .
Moreover,
G̃∗1 (σ̃ , σ , z∗ , v) = G∗1 (σ̃ , −σ , −Q)

= sup {v2 , σ̃ − σ − QY − G1 (v2 )}
v2 ∈Y
1 3
= ∑ σ̃mi − σmi − Qmi 22
K m,i=1
1 3
= ∑ σ̃mi − σmi − (σi j + z∗i j )vm j − (K/2)vmi 22 .
K m,i=1
Finally, if there exists a point (u0 , σ̃0 , σ0 , v0 , z∗0 ) ∈ C1 × ((Ỹ × Y ∗ ) ∩ B∗ ), such that

δ u0i , −σ̃0i j, j − fi L2 (Ω ) − u0i ( fî − σ̃0i j n j ) dΓ
Γ1

∗ ∗ ∗ ∗ ∗ ∗
+F (z0 ) − G̃K (σ0 , z0 , v0 ) − G̃1(σ̃0 , σ0 , z0 , v0 ) = θ , (12.11)
we have
J(u0 ) = min {J(u)}

u∈C1

" ∗ ∗ ∗ ∗ ∗ ∗
#
= sup inf
∗ ∗
F (z ) − G̃ K (σ , z , v) − G̃ 1 (σ̃ , σ , z , v)
(σ̃ ,σ ,v)∈Ỹ z ∈Y
∗
=F (z∗0 ) − G̃∗K (σ0 , z∗0 , v0 ) − G̃∗1 (σ̃0 , σ0 , z∗0 , v0 ). (12.12)
Proof. We start by proving that G∗K (σ + z∗ , Q) = G∗KL (σ + z∗ , Q) if σK (σ , z∗ ) is

positive definite in Ω , where

G∗KL (σ , Q) = g∗KL (σ , Q) dx
Ω
is the Legendre transform of GK : Y × Y → R. To simplify the notation we denote

(σ , Q) = y∗ = (y∗1 , y∗2 ). We first formally calculate g∗KL (y∗ ), the Legendre transform
of gK (y), where

1 1
gK (y) = Hi jkl y1i j + y2mi y2m j y1kl + y2mk y2ml
2 2
K
+ y2mi y2mi . (12.13)
4
We recall that
g∗KL (y∗ ) = y, y∗ R18 − gK (y) (12.14)

where y ∈ R18 is the solution of equation
∂ gK (y)
y∗ = . (12.15)
∂y
Thus

1
y∗1i j = σi j = Hi jkl y1kl + y2mk y2ml (12.16)
2
and

1
y∗2mi = Qmi = Hi jkl y1kl + y2ok y2ol y2m j + (K/2)y2mi (12.17)
2
so that
Qmi = σi j y2m j + (K/2)y2mi . (12.18)
Inverting these last equations, we have
y2mi = σ Kij Qm j (12.19)
where {σ Kij } = σK−1 (σ ),

⎧ ⎫
⎨ σ11 + K/2 σ12 σ13 ⎬
σK (σ ) = σ21 σ22 + K/2 σ23 (12.20)
⎩ ⎭
σ31 σ32 σ33 + K/2
and also
1
y1i j = H i jkl σkl − y2mi y2m j . (12.21)
2
Finally
1 1
g∗KL (σ , Q) = H i jkl σi j σkl + σ̄iKj Qmi Qm j . (12.22)
2 2
Now we will prove that g∗KL (y∗ ) = g∗K (y∗ ) if σK (y∗1 ) = σK (σ ) is positive definite.
First observe that
g∗K (y∗ ) = sup {y1 , σ R9 + y2 , QR9 − gK (y)}
y∈R18

= sup y1 , σ R9 + y2 , QR9
y∈R18

1 1 1
− Hi jkl y1i j + y2mi y2m j y1kl + y2mk y2ml
2 4 2

K
− y2mi y2mi
4

1
= sup ȳ1i j − y2mi y2m j , σi j R + y2 , QR9
(ȳ1 ,y2 )∈R9 ×R9 2

1 K
− Hi jkl [ȳ1i j ][ȳ1kl ] − y2mi y2mi .
2 4
The result follows just observing that

1 1
sup ȳ1i j , σi j R − Hi jkl [ȳ1i j ][ȳ1kl ] = H i jkl σi j σkl (12.23)
ȳ1 ∈R9 2 2
and

1 K
sup − y2mi y2m j , σi j R + y2 , QR9 − y2mi y2mi
y2 ∈R9 2 4
1
= σ Kij Qmi Qm j (12.24)
2
if σK (y∗1 ) = σK (σ ) is positive definite.
Now observe that using the relation
Qmi = (σi j + z∗i j )vm j + (K/2)vmi ,
we have
G̃∗K (σ , z∗ , v) = G∗K (σ + z∗ , Q)

= g∗KL (σ + z∗ , Q) dx, (12.25)
Ω
if σK (σ + z∗ ) is positive definite.
Also, considering the concerned symmetries, we may write
G̃∗K (σ , z∗ , v) + G̃∗1 (σ̃ , σ , z∗ , v) = G∗K (σ + z∗ , Q) + G∗1 (σ̃ , −σ , −Q)

≥ Λ1 u, σ L2 + Λ2 u, z∗ + QL2
+Λ1 u, σ̃ − σ L2 − Λ2 u, QL2
−G∗∗
K (Λ u) − G1 (Λ2 u), (12.26)
∀u ∈ U, z∗ ∈ Y ∗ , (σ̃ , σ , v) ∈ Ỹ , so that
G̃∗K (σ , z∗ , v) + G̃∗1(σ̃ , σ , z∗ , v)
≥ Λ2 u, z∗ L2 + Λ1 u, σ̃ L2
−GK (Λ u) − G1(Λ2 u)
= Λ2 u, z∗ L2 + u, f L2 (Ω ;R3 )

+ fî ui dΓ − GK (Λ u) − G1 (Λ2 u), (12.27)
Γ1
∀u ∈ C1 , z∗ ∈ Y ∗ , (σ̃ , σ , v) ∈ Ỹ . Hence
−F ∗ (z∗ ) + G̃∗K (σ , z∗ , v) + G̃∗1 (σ̃ , σ , z∗ , v)

≥ −F ∗ (z∗ ) + Λ2 u, z∗ L2 + u, f L2 (Ω ;R3 )

+ fî ui dΓ − GK (Λ u) − G1 (Λ2 u), (12.28)
Γ1
∀u ∈ C1 , z∗ ∈ Y ∗ , (σ̃ , σ , v) ∈ Ỹ , and thus
sup {−F ∗ (z∗ ) + G̃∗K (σ , z∗ , v) + G̃∗1(σ̃ , σ , z∗ , v)}

z∗ ∈Y ∗
≥ sup {−F ∗ (z∗ ) + Λ2u, z∗ L2 + u, f L2 (Ω ;R3 )
z∗ ∈Y ∗

+ fî ui dΓ − GK (Λ u) − G1(Λ2 u)}, (12.29)
Γ1
∀u ∈ C1 , (σ̃ , σ , v) ∈ Ỹ .
Therefore,
sup {−F ∗ (z∗ ) + G̃∗K (σ , z∗ , v) + G̃∗1 (σ̃ , σ , z∗ , v)}

z∗ ∈Y ∗

≥ F(Λ2 u) + u, f L2 (Ω ;R3 ) + fî ui dΓ
Γ1
−GK (Λ u) − G1 (Λ2 u), (12.30)
∀u ∈ C1 , (σ̃ , σ , v) ∈ Ỹ , that is,
sup {−F ∗ (z∗ ) + G̃∗K (σ , z∗ , v) + G̃∗1 (σ̃ , σ , z∗ , v)}

z∗ ∈Y ∗
≥ −J(u), (12.31)
∀u ∈ C1 , (σ̃ , σ , v) ∈ Ỹ . Finally,
inf {J(u)} (12.32)

u∈C1

" ∗ ∗ ∗ ∗ ∗ ∗
#
≥ sup inf
∗ ∗
F (z ) − G̃ K (σ , z , v) − G̃ 1 (σ̃ , σ , z , v) .
(σ̃ ,σ ,v)∈Ỹ z ∈Y
Now suppose there exists a point (u0 , σ̃0 , σ0 , z∗0 , v0 ) ∈ C1 × ((Ỹ ×Y ∗ )∩B∗ ), such that

Γ1

+F ∗ (z∗0 ) − G̃∗K (σ0 , z∗0 , v0 ) − G̃∗1(σ̃0 , σ0 , z∗0 , v0 ) = θ , (12.33)
that is,

Γ1

1
+F ∗ (z∗0 ) − H̄i jkl (σ0i j + z∗0i j )(σ0kl + z∗0kl ) dx
2 Ω

1 K
− (σ0i j + z∗0i j )v0mi v0m j dx − v mi , v0mi L2 (Ω )
2 Ω 4 0

3
1
− ∑ σ̃0mi − σ0mi − (σ0i j + z∗0i j )v0m j − K/2v0mi 22 = θ.
m,i=1 K
Observe that the variation in σ̃ gives us
σ̃0mi − σ0mi − (σ0i j + z∗0i j )v0m j − (K/2)v0mi = (K/2)u0m,i in Ω . (12.34)
From this and recalling that σ̃i j = σ̃ ji , so that we may use the replacement
σ̃i j + σ̃ ji
σ̃i j = = σ̃ ji
2
(observe that a similar remark is valid for σ0i j + z∗0i j ), the variation in σ gives us
−H̄i jkl (σ0kl + z∗0kl ) − v0mi v0m j /2

u i, j + u0 j,i
+ 0 + u0m,i v0m j = 0, (12.35)
2
in Ω . From (12.34) and the variation in v we get
−(σ0i j + z∗0i j )vm j − (K/2)v0mi

+(σ0i j + z∗0i j )u0m, j + (K/2)u0m,i = 0, (12.36)
so that
{v0i j } = {u0i, j }, in Ω . (12.37)

u0k,l + u0l,k u0m,k u0m,l
σ0i j + z∗0i j = Hi jkl + . (12.38)
2 2
Through such relations the variation in z∗ gives us

K
z∗0i j = (u i, j + u0 j,i ) in Ω . (12.39)
2 0
Finally, from the variation in u, we get
σ̃0i j, j + fi = 0, in Ω , (12.40)
u0 = θ on Γ0 ,
and
σ̃0i j n j = fî on Γ1 ,
where from (12.34), (12.37), and (12.39), we have

σ̃0i j = Hi jkl +
2 2

u0k,l + u0l,k u0 p,k u0 p,l
+Hm jkl + u0i,m . (12.41)
2 2
Replacing such results in the dual formulation we obtain
J(u0 ) = F ∗ (z∗0 ) − G̃∗K (σ0 , z∗0 , v0 ) − G̃∗1 (σ̃0 , σ0 , z∗0 , v0 ). (12.42)
From the hypothesis indicated in (12.6), the extremal relation through which z∗0
is obtained is in fact a global one.
From this, (12.2) and (12.42), the proof is complete.
Remark 12.2.2. About the last theorem, there is no duality gap between the primal
and dual problems, if K is big enough so that for the optimal dual point, σK (σ0 , z∗0 )
is positive definite in Ω , where
⎧ ⎫
⎨ σ11 + z∗11 + K/2 σ12 + z∗12 σ13 + z∗13 ⎬
σK (σ , z∗ ) = σ21 + z∗21 σ22 + z∗22 + K/2 σ23 + z∗23 , (12.43)
⎩ ⎭
σ31 + z∗31 σ32 + z∗32 σ33 + z∗33 + K/2
and
σ0i j + z∗0i j = Hi jkl + , (12.44)
2 2
and, at the same time, K is small enough so that for the fixed point {v0m j } = {u0m, j }
the quadratic form (in z∗ ) W ∗ (z∗ ) is also positive definite in Ω , where
∗
3 (z v m j )
z∗mi z∗mi
2
W ∗ (z∗ ) = − H̄i jkl z∗i j z∗kl − ∑
ij 0
. (12.45)
K m,i=1 K/2
For K ≈ O(min{H1111 /2, H2222/2, H1212/2}) there is a large class of external

loads for which such a K satisfies the conditions above, including to some extent the
large deformation context.
Finally, we have not formally proven, but one may obtain from the relation be-
tween the primal and dual variables that
C = {u ∈ U | G∗∗
K (Λ u) = GK (Λ u)}
= {u ∈ U | σK (σ (u), θ ) is positive definite in Ω }, (12.46)
12.3 Other Duality Principles 331
where as above indicated

1
σi j (u) = Hi jkl (uk,l + ul,k + um,k um,l ) . (12.47)
2
12.3 Other Duality Principles
At this point we present another main result, which is summarized by the follow-
ing theorem.
Theorem 12.3.1. Let Ω ⊂ R3 be an open, bounded, connected set with a regu-

lar boundary denoted by Γ = Γ0 ∪ Γ1 , where Γ0 ∩ Γ1 = 0.
/ Consider the functional
(G ◦ Λ ) : U → R expressed by
(G ◦ Λ )(u)

= Hi jkl + + dx,
2 Ω 2 2 2 2
where
U = {u = (u1 , u2 , u3 ) ∈ W 1,4 (Ω ; R3 ) | u = (0, 0, 0) ≡ θ on Γ0 }, (12.48)
and Λ : U → Y = Y ∗ = L2 (Ω ; R3×3 ) ≡ L2 is given by

1
Λ u = {Λi j (u)} = (ui, j + u j,i + um,i um, j ) .
2
Define J : U → R by

Γ1
Also define
JK : U × Y → R
by

K
JK (u, p) = G(Λ u + p) + Kp, pL2 − u, f L2 (Ω ;R3 ) − fî ui dΓ − p, pL2 ,
Γ1 2
and assume that K > 0 is sufficiently big so that JK (u, p) is bounded below.
Also define
2
∂ G∗ (σ )
JK∗ (σ , u) = ∗
Ff (σ ) − G (σ ) + K Λ u − + 1 σ , σ 2 , (12.50)
∂ σ L2 2K L
where
G∗ (σ ) = sup{v, σ L2 − G(v)}

v∈Y

1
= H i jkl σi j σkl dx, (12.51)
2 Ω
{H i jkl } = {Hi jkl }−1

and
Ff (σ ) = sup Λ u, σ L2 − u, f L2 (Ω ;R3 ) − fî ui dΓ .
u∈U Γ1
inf {JK (u, p)} ≤ inf {JK∗ (σ , u)}. (12.52)

(u,p)∈U (σ ,u)∈Y ∗ ×U
Finally, assume that Γ0 , f ∈ L2 (Ω ; R3 ) and fˆ ∈ L2 (Γ ; R3 ) are such that a local

minimum of JK over V0 = Br (u0 ) × Br (p0 ) is attained at some (u0 , p0 ) ∈ U ×Y such
that
∂ G(Λ u0 + p0 )
σ0 = (12.53)
∂v
is negative definite.
Here
Br (u0 ) = {u ∈ U | u − u0 U < r},
and
Br (p0 ) = {p ∈ Y | p − p0 Y < r},
for some appropriate r > 0.
Under such hypotheses, there exists a set Ṽ0 ⊂ Y ∗ × U, such that
JK (u0 , p0 ) = inf {JK (u, p)}

(u,p)∈V0
≤ inf {JK∗ (σ , u)}
(σ ,u)∈Ṽ0
≤ JK∗ (σ0 , u0 )
= JK (u0 , p0 )
≈ J(u0 ) + O(1/K). (12.54)
Proof. Define
G1 (u, p) = G(Λ u + p) + Kp, pL2 ,
and
K
G2 (u, p) = u, f L2 (Ω ;R3 ) + fî ui dΓ + p, pL2 .
Γ1 2
Observe that αK = inf(u,p)∈U×Y {JK (u, p)} ∈ R is such that
JK (u, p) = G1 (u, p) − G2(u, p) ≥ αK , ∀u ∈ U, p ∈ Y.
Thus,
−G2 (u, p) ≥ −G1 (u, p) + αK , ∀u ∈ U, p ∈ Y,
so that
Λ u + p, σ L2 − G2 (u, p) ≥ Λ u + p, σ L2 − G1 (u, p) + αK , ∀u ∈ U, p ∈ Y.
Hence,
sup {Λ u + p, σ L2 − G2 (u, p)}≥Λ u+p, σ L2 − G1 (u, p)+ αK , ∀u ∈ U, p ∈ Y.

(u,p)∈U×Y
(12.55)
In particular for u, p such that
∂ G(Λ u + p)
σ= ,
∂v
we get
∂ G∗ (σ )
p +Λu = ,
∂σ
that is,
∂ G∗ (σ )
p= − Λ u,
∂σ
and
G∗ (σ ) = Λ u + p, σ L2 − G(Λ u + p).
Hence
∗ 2
∂ G (σ )
Λ u + p, σ L2 − G1 (u, p) = G∗ (σ ) − K
∂σ − Λ u .
2
L
On the other hand,

1
sup {Λ u + p, σ L2 − G2(u, p)} = Ff (σ ) + σ , σ L2 .
(u,p)∈U×Y 2K
Replacing such results in (12.55), we get

∗ 2
∂ G (σ ) 1
∗
Ff (σ ) − G (σ ) + K − Λ u
∂σ 2 + 2K σ , σ L2 ≥ αK ,
L
∀σ ∈ Y ∗ , u ∈ U.
Thus,
αK = inf {JK (u, p)} ≤ inf {JK∗ (σ , u)}. (12.56)

(u,p)∈U×Y (σ ,u)∈Y ∗ ×U
Now, let (u0 , p0 ) ∈ U × Y be such that
J(u0 , p0 ) = min {JK (u, p)}.

(u,p)∈V0
Defining
∂ G(Λ u0 + p0 )
σ0 = , (12.57)
∂v
since for the extremal point, we have

δu G(Λ u + p0) − u, f L2 (Ω ;R3 ) − fi ui dΓ |u=u0 = θ ,
ˆ
Γ1
from this and (12.57), we also have

δu Λ u, σ0 L2 − u, f L2 (Ω ;R3 ) − fî ui dΓ |u=u0 = θ ,
Γ1
and therefore, since σ0 is negative definite, we obtain

Ff (σ0 ) = Λ u0 , σ0 L2 − u0 , f L2 (Ω ;R3 ) − fî u0i dΓ . (12.58)
Γ1
From (12.57), we get
G∗ (σ0 ) = Λ u0 + p0 , σ L2 − G(Λ u0 + p0), (12.59)
so that, from (12.58) and (12.59), we obtain

∗ 2
∂ G (σ0 ) 1
∗
Ff (σ0 ) − G (σ0 ) + K − Λ u0
∂σ 2 + 2K σ0 , σ0 L2
L
K
= G(Λ u0 + p0) + p0 , p0 L2 − u0, f L2 (Ω ;R3 )
2
− fî u0 dΓ ,
i (12.60)
Γ1
that is,
JK∗ (σ0 , u0 ) = JK (u0 , p0 ). (12.61)
Observe that, from the hypotheses,
JK (u, p) ≥ JK (u0 , p0 ), ∀(u, p) ∈ V0 .
At this point we develop a reasoning similarly to the lines above but now for the
specific case of a neighborhood around the local optimal point. We repeat some
analogous details for the sake of clarity.
From above,
G1 (u, p) = G(Λ u + p) + Kp, pL2 ,
and
K
G2 (u, p) = u, f L2 (Ω ;R3 ) + fî ui dΓ + p, pL2 .
Γ1 2
Observe that α = inf(u,p)∈V0 {JK (u, p)} ∈ R is such that
JK (u, p) = G1 (u, p) − G2(u, p) ≥ α , ∀(u, p) ∈ V0 .
Thus,
−G2 (u, p) ≥ −G1 (u, p) + α , ∀(u, p) ∈ V0 ,
so that
Λ u + p, σ L2 − G2(u, p) ≥ Λ u + p, σ L2 − G1 (u, p) + α , ∀(u, p) ∈ V0 .
Hence,
sup {Λ u + p, σ L2 − G2 (u, p)}

(u,p)∈U
≥ sup {Λ u + p, σ L2 − G2 (u, p)}

(u,p)∈V0
≥ Λ u + p, σ L2 − G1 (u, p) + α , ∀(u, p) ∈ V0 . (12.62)
In particular, if (σ , u) ∈ Ṽ0 , where such a set is defined by the points (σ , u) such

that u ∈ Br (u0 ) and for the σ in question there exists p ∈ Br (p0 ) such that
∂ G(Λ u + p)
σ= ,
∂v
that is,
∂ G∗ (σ )
p +Λu = ,
∂σ
we get
∂ G∗ (σ )
p= − Λ u,
∂σ
and
G∗ (σ ) = Λ u + p, σ L2 − G(Λ u + p).
Hence
∗ 2
∂ G (σ )
Λ u + p, σ L2 − G1 (u, p) = G∗ (σ ) − K
∂σ − Λ u .
2 (12.63)
L
On the other hand
sup {Λ u + p, σ L2 − G2 (u, p)}

(u,p)∈V0
≤ sup {Λ u + p, σ L2 − G2 (u, p)}

(u,p)∈U×Y
1
= Ff (σ ) + σ , σ L2 . (12.64)
2K
Observe that σ0 ∈ Ṽ0 . We do not provide details here, but from the generalized
inverse function theorem, also an appropriate neighborhood of σ0 belongs to Ṽ0 .
Replacing the last relations (12.63) and (12.64) into (12.62), we get
∗ 2
∂ G (σ )
Ff (σ ) − G∗ (σ ) + K
∂σ − Λ u
2
L
1
+ σ , σ L2 ≥ α , (12.65)
2K
∀(σ , u) ∈ Ṽ0 .
Thus,
α= inf {JK (u, p)} ≤ inf {JK∗ (σ , u)}. (12.66)

(u,p)∈V0 (σ ,u)∈Ṽ0
Finally, since
1 ∂ G(Λ u0 + p0 )
p0 = − , (12.67)
K ∂p
we get
1
p0 Y ≈ O ,
K
so that from this, (12.61), and (12.65), we may finally write
α = JK (u0 , p0 ) = inf {JK (u, p)}

(u,p)∈V0
≤ inf {JK∗ (σ , u)}
(σ ,u)∈Ṽ0
≤ JK∗ (σ0 , u0 )
= JK (u0 , p0 )
≈ J(u0 ) + O(1/K). (12.68)
Remark 12.3.2. Of particular interest is the model behavior as K → +∞. From

(12.68) it seems to be clear that the duality gap between the original primal and
dual formulations goes to zero as K goes to +∞.
Our final result is summarized by the next theorem. It refers to a duality principle
for the case of a local maximum for the primal formulation.
lar boundary denoted by Γ = Γ0 ∪ Γ1 , where Γ0 ∩ Γ1 = 0.
/ Consider the functional
(G ◦ Λ ) : U → R expressed by
(G ◦ Λ )(u)

= Hi jkl + + dx,
2 Ω 2 2 2 2
where
U = {u = (u1 , u2 , u3 ) ∈ W 1,4 (Ω ; R3 ) | u = (0, 0, 0) ≡ θ on Γ0 }, (12.69)
and Λ : U → Y = Y ∗ = L2 (Ω ; R3×3 ) ≡ L2 is given by

1
Λ u = {Λi j (u)} = (ui, j + u j,i + um,i um, j ) .
2

Γ1
Assume that Γ0 , f ∈ L2 (Ω ; R3 ), and fˆ ∈ L2 (Γ ; R3 ) are such that a local maximum

of J over V0 = Br (u0 ) is attained at some u0 ∈ U such that
∂ G(Λ u0 )
σ0 = (12.71)
∂v
is negative definite.
Also define
J ∗ (σ ) = Ff (σ ) − G∗ (σ ), (12.72)
where
G∗ (σ ) = sup{v, σ L2 − G(v)}

v∈Y

1
= H i jkl σi j σkl dx, (12.73)
2 Ω
{H i jkl } = {Hi jkl }−1

and
Ff (σ ) = sup Λ u, σ L2 − u, f L2 (Ω ;R3 ) − fî ui dΓ .
u∈U Γ1
Under such assumptions, there exists a set Ṽ0 ⊂ Y ∗ such that
− J ∗(σ0 ) = max{−J ∗ (σ )} = max{J(u)} = J(u0 ). (12.74)

σ ∈Ṽ0 u∈V0
Proof. Define α = J(u0 ).

Thus,

J(u) = G(Λ u) − u, f L2 (Ω ;R3 ) − fî ui dΓ ≤ J(u0 ) = α ,
Γ1
∀u ∈ V0 .
Hence,

−u, f L2 (Ω ;R3 ) − fî ui dΓ ≤ −G(Λ u) + α , ∀u ∈ V0 ,
Γ1
so that

Λ u, σ L2 − u, f L2 (Ω ;R3 ) − fî ui dΓ
Γ1
≤ Λ u, σ L2 − G(Λ u) + α , ∀u ∈ V0 , σ ∈ Y ∗ . (12.75)
Therefore,

sup Λ u, σ L2 − u, f L2 (Ω ;R3 ) − fî ui dΓ
u∈V0 Γ1
≤ sup {v, σ L2 − G(v)} + α , ∀σ ∈ Y ∗ . (12.76)

v∈Y
We define Ṽ0 by the points σ ∈ Y ∗ such that

sup Λ u, σ L2 − u, f L2 (Ω ;R3 ) − f i ui d Γ
ˆ
u∈V0 Γ1

= sup Λ u, σ L2 − u, f L2 (Ω ;R3 ) − fî ui dΓ
u∈U Γ1
= Ff (σ ). (12.77)
We highlight that σ0 ∈ Ṽ0 , and from the generalized inverse function theorem,
any σ in an appropriate neighborhood of σ0 also belongs to Ṽ0 (we do not provide
the details here).
From this and (12.76), we get
Ff (σ ) − G∗ (σ ) ≤ α = J(u0 ), ∀σ ∈ Ṽ0 . (12.78)

12.4 A Numerical Example 339

Ff (σ0 ) − G∗(σ0 ) = G(Λ u0 ) − u0, f L2 (Ω ;R3 ) − fî u0i dΓ
Γ1
= J(u0 ). (12.79)
From this and (12.78), the proof is complete.
12.4 A Numerical Example
Consider the functional J : U → R defined by

2 1
H 1 1
J(u) = ux + u2x dx − Pu dx,
2 0 2 0
where
U = {u ∈ W 1,4 ([0, 1]) | u(0) = u(1) = 0} = W01,4 ([0, 1]),
H = 105
P = −1000
where the units refer to the international system. The condition indicated in (12.45)
here stands for W ∗ (z∗ ) to be positive definite in a critical point u0 ∈ U, where
(z∗ )2 (z∗ )2 (u0 (x))2 (z∗ )2

W ∗ (z∗ ) = − − ,
K H K/2
∂ 2W ∗ (z∗ )
≥ 0,
∂ (z∗ )2
so that, for K = H/2, we get
(u0 (x))2 ≤ 0.25, a.e. in [0, 1],
that is,
|u0 (x)| ≤ 0.5, a.e. in [0, 1].
We have computed a critical point through the primal formulation, again denoted by
u0 ∈ U. Please see Fig. 12.1. For u0 (x), see Fig. 12.2.
We may observe that
|u0 (x)| ≤ 0.5,
in [0, 1], so that by the main duality, such a point is a local minimum on the set
C1 = C ∩C2 , where
C = {u ∈ U | G∗∗
K (ux ) = GK (ux )}
= {u ∈ U | H(ux + u2x /2) + K/2 > 0, in [0, 1]}, (12.80)
x 10−3
0
−0.2
−0.4
−0.6
−0.8
−1
−1.2
−1.4
0 0.2 0.4 0.6 0.8 1
Fig. 12.1 The solution u0 (x) through the primal formulation
x 10−3
5
4
3
2
1
0
−1
−2
−3
−4
−5
0 0.2 0.4 0.6 0.8 1
Fig. 12.2 The solution u0 (x) through the primal formulation
C2 = {u ∈ U | ux ∈ Ŷ ∗ }, where
1 1
H K
GK (ux ) = (ux + u2x /2)2 dx + u2x dx,
2 0 4 0
and
Ŷ ∗ = {v ∈ L2 ([0, 1]) | W ∗ (z∗ ) is positive definite in [0, 1]}.
In fact, plotting the function F(x) = H(x + x2 /2)2 /2, we may observe that inside
the set [−0.5, 0.5] there is a local minimum, that is, in a close set, the Legendre
necessary condition for a local minimum is satisfied. Please see Fig. 12.3.
We emphasize on the concerned sets there is no duality gap between the primal
and dual formulations. Also, from the graphic of u0 (x), it is clear that the stress
H(u0 + 1/2(u0)2 )
is not exclusively positive or negative in [0, 1].
12.5 Conclusion 341
x 104
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6
Fig. 12.3 The function F(x) = H(x + x2 /2)2 /2
12.5 Conclusion
In this chapter we develop new duality principles applicable to nonlinear finite

elasticity. The results are obtained through the basic tools of convex analysis and
include sufficient conditions of restricted optimality. It is worth mentioning that the
methods developed here may be applied to many other situations, such as nonlin-
ear models of plates and shells. Applications to related areas (specially to the shell
model presented in [23]) are planned for future works.
Chapter 13
Duality Applied to a Plate Model
13.1 Introduction
In the present work we develop dual variational formulations for the Kirchhoff–
Love thin plate model. Earlier results establish the complementary energy under
the hypothesis of positive definiteness of the membrane force tensor at a critical
point (please see [30–32, 36, 65] for details). In more recent works Gao has applied
his triality theory to models in elasticity (see [33–35] for details) obtaining duality
principles which allow the optimal stress tensor to be negative definite. Here for the
present case we have obtained a dual variational formulation which allows the global
optimal point in question to be not only positive definite (for related results see
Botelho [11, 13]) but also not necessarily negative definite. The approach developed
also includes sufficient conditions of optimality for the primal problem. Moreover,
a numerical example concerning the main duality principle application is presented
in the last section.
It is worth mentioning that the standard tools of convex analysis used in this text
may be found in [13, 25, 54], for example. Another relating result may be found
in [14].
Let Ω ⊂ R2 be an open, bounded, connected set which represents the middle
surface of a plate of thickness h. The boundary of Ω , which is assumed to be regular,
is denoted by ∂ Ω . The vectorial basis related to the Cartesian system {x1 , x2 , x3 } is
denoted by (aα , a3 ), where α = 1, 2 (in general Greek indices stand for 1 or 2)
and where a3 is the vector normal to Ω , whereas a1 and a2 are orthogonal vectors
parallel to Ω . Also, n is the outward normal to the plate surface.
The displacements will be denoted by
û = {ûα , û3 } = ûα aα + û3 a3 .
The Kirchhoff–Love relations are
ûα (x1 , x2 , x3 ) = uα (x1 , x2 ) − x3w(x1 , x2 ),α and û3 (x1 , x2 , x3 ) = w(x1 , x2 ).
344 13 Duality Applied to a Plate Model
Here −h/2 ≤ x3 ≤ h/2 so that we have u = (uα , w) ∈ U where

∂w
U = (uα , w) ∈ W 1,2 (Ω ; R2 ) × W 2,2 (Ω ), uα = w = = 0 on ∂ Ω
∂n
= W01,2 (Ω ; R2 ) × W02,2 (Ω ).
It is worth emphasizing that the boundary conditions here specified refer to a

clamped plate.
We define the operator Λ : U → Y × Y , where Y = Y ∗ = L2 (Ω ; R2×2 ), by
Λ (u) = {γ (u), κ (u)},

uα ,β + uβ ,α w,α w,β
γαβ (u) = + ,
2 2
καβ (u) = −w,αβ .
The constitutive relations are given by
Nαβ (u) = Hαβ λ μ γλ μ (u), (13.1)
Mαβ (u) = hαβ λ μ κλ μ (u), (13.2)

h2
where {Hαβ λ μ } and {hαβ λ μ = 12 Hαβ λ μ } are symmetric positive definite fourth-
order tensors. From now on, we denote {H αβ λ μ } = {Hαβ λ μ }−1 and {hαβ λ μ } =
{hαβ λ μ }−1 .
Furthermore {Nαβ } denote the membrane force tensor and {Mαβ } the moment
one. The plate stored energy, represented by (G ◦ Λ ) : U → R, is expressed by

1 1
(G ◦ Λ )(u) = Nαβ (u)γαβ (u) dx + Mαβ (u)καβ (u) dx, (13.3)
2 Ω 2 Ω
and the external work, represented by F : U → R, is given by
F(u) = w, PL2 (Ω ) + uα , Pα L2 (Ω ) , (13.4)
where P, P1 , P2 ∈ L2 (Ω ) are external loads in the directions a3 , a1 , and a2 , respec-

tively. The potential energy, denoted by J : U → R, is expressed by
J(u) = (G ◦ Λ )(u) − F(u).
It is important to emphasize that the existence of a minimizer (here denoted by

u0 ) related to J(u) has been proven in Ciarlet [22]. Some inequalities of Sobolev
type are necessary to prove the above result. In particular, we assume the boundary
∂ Ω of Ω is regular enough so that the standard Gauss–Green formulas of integration
by parts and the well-known Sobolev imbedding and trace theorems hold. Details
about such results may be found in [1].
Finally, we also emphasize from now on, as their meaning is clear, we may denote
L2 (Ω ) and L2 (Ω ; R2×2 ) simply by L2 and the respective norms by · 2. Moreover
derivatives are always understood in the distributional sense, θ denotes the zero vec-
tor in appropriate Banach spaces, and the following and relating notations are used:
∂ 2w
w,αβ = ,
∂ xα ∂ xβ
∂ uα
uα ,β = ,
∂ xβ
∂ Nαβ
Nαβ ,1 = ,
∂ x1
and
∂ Nαβ
Nαβ ,2 = .
∂ x2
In this section, we develop a duality principle presented in similar form in

[13, 14]. The novelty here is its suitability for the Kirchhoff–Love plate model.
Theorem 13.2.1. Let Ω ⊂ R2 be an open, bounded, connected set with a regular

boundary denoted by Γ . Suppose (G ◦ Λ ) : U → R is defined by

1
G(Λ u) = Hαβ λ μ γαβ (u)γλ μ (u) dx
2 Ω

1
+ h κ (u)κλ μ (u) dx
2 Ω αβ λ μ αβ

K K
+ (wx )2 dx + (wy )2 dx, (13.5)
2 Ω 2 Ω
and let (F ◦ Λ2 ) : U → R be expressed by

K K
F(Λ2 u) = (wx )2 dx + (wy )2 dx,
2 Ω 2 Ω
where
1
γαβ (u) = Λ1αβ (u) + Λ2α (u)Λ2β (u),
2

uα ,β + uβ ,α
{Λ1αβ (u)} = ,
2
{Λ2α (u)} = {w,α },
{καβ (u)} = {−Λ3αβ (u)} = {−w,αβ },
where u = (uα , w) ∈ U = W01,2 (Ω ; R2 ) × W02,2 (Ω ).

Also, define F1 : U → R by
F1 (u) = w, PL2 + uα , Pα L2 ≡ u, P̃L2 ,
where P̃ = (P, Pα ), and let J : U → R be expressed by
J(u) = (G ◦ Λ )(u) − F(Λ2 u) − F1(u).

∗ ∗ ∗ ∗ ∗
inf {J(u)} ≥ sup ∗inf ∗ {F̃ (z ) − G (v , z )} , (13.6)
u∈U v∗ ∈A∗ z ∈Y0
where, when the meaning is clear, denoting Y = Y ∗ = L2 (Ω , R2×2 ) ≡ L2 , Y1 = Y1∗ =

L2 (Ω , R2 ) ≡ L2 , v∗ = (v∗1 , v∗2 , v∗3 ), and v∗1 = {Nαβ }, v∗2 = {Qα }, and v∗3 = {Mαβ },
we have
A∗ = {v∗ ∈ Y ∗ | Λ ∗ v∗ = P̃},
that is, recalling that P̃ = (P, Pα ), we may write
A∗ = A1 ∩ A2 .
Here
A1 = {v∗ ∈ Y ∗ | Mαβ ,αβ + Qα ,α + P = 0, in Ω },
A2 = {v∗ ∈ Y ∗ | Nαβ ,β + Pα = 0, in Ω },
and
Y0∗ = {z∗ ∈ Y1∗ | z∗11 = z∗22 = 0 and (z∗11 )x n1 + (z∗22 )y n2 = 0 on Γ }.
Also,
F̃ ∗ (z∗ ) = sup {−v1 , (z∗11 )x L2 (Ω ) − v2 , (z∗22 )y L2 (Ω )

v∈Y1

K K
− (v1 )2 dx − (v2 )2 dx}
2 Ω 2 Ω

1 1
= ((z∗11 )x )2 dx + ((z∗ )y )2 dx, (13.7)
2K Ω 2K Ω 22
and
G∗ (v∗ , z∗ ) = sup{v1 , v∗1 L2 + v2 , v∗2 L2
v∈Y
+v3 , v∗3 + z∗ L2 − G(v)}

1
= h (M + z∗αβ )(Mλ μ + z∗λ μ ) dx

1
+ H N N dx
2 Ω αβ λ μ αβ λ μ

1 K
+ N αβ Qα Qβ dx, (13.8)
2 Ω
if N K is positive definite, where

N11 + K N12
NK = , (13.9)
N21 N22 + K
K K −1
{N αβ } = {Nαβ } .
Moreover,
{H αβ λ μ } = {Hαβ λ μ }−1 ,
and
{hαβ λ μ } = {hαβ λ μ }−1 .
Here we recall that z∗12 = z∗21 = 0 in Ω .
Finally, if there is a point (u0 , v∗0 , z∗0 ) ∈ U × A∗ × Y0∗ such that
δ {F̃ ∗ (z∗0 ) − G∗ (v∗0 , z∗0 ) + u0, Λ ∗ v∗0 − P̃L2 } = θ ,
where K > 0 is such that
F̃ ∗ (z∗ ) − G∗ (z∗ ) > 0, ∀z∗ ∈ Y0∗ such that z∗ = θ , (13.10)
we have that
J(u0 ) = min{J(u)}
u∈U

" #
= sup ∗inf ∗ F̃ ∗ (z∗ ) − G∗ (v∗ , z∗ )
v∗ ∈A∗ z ∈Y0
= F̃ ∗ (z∗0 ) − G∗ (v∗0 , z∗0 ). (13.11)
Proof. Observe that
G∗ (v∗ , z∗ ) ≥ Λ3 u, v∗3 + z∗ L2 + Λ1 u, v∗1 L2

+Λ1 u, v∗2 L2 − G(Λ u), (13.12)
∀u ∈ U, v∗ ∈ A∗ , z∗ ∈ Y0∗ .
Thus,
− F̃ ∗ (z∗ ) + G∗ (v∗ , z∗ ) ≥ −F̃ ∗ (z∗ ) + Λ3 u, z∗ L2

−G(Λ u) + u, P̃L2 , (13.13)
∀u ∈ U, v∗ ∈ A∗ , z∗ ∈ Y0∗ , so that, taking the supremum in z∗ at both sides of last

inequality, we obtain
sup {−F̃ ∗ (z∗ ) + G∗ (v∗ , z∗ )}

z∗ ∈Y0∗
≥ F(Λ2 u) − G(Λ u) + u, P̃L2

= F(Λ2 u) − G(Λ u) + F1(u)

= J(u), (13.14)
∀u ∈ U, v∗ ∈ A∗ . Hence,

∗ ∗ ∗ ∗ ∗
inf {J(u)} ≥ sup inf
∗ ∗
{ F̃ (z ) − G (v , z )} . (13.15)
u∈U v∗ ∈A∗ z ∈Y0
Finally, suppose that (u0 , v∗0 , z∗0 ) ∈ U × A∗ × Y0∗ is such that
δ {F̃ ∗ (z∗0 ) − G∗ (v∗0 , z∗0 ) + u0, Λ ∗ v∗0 − P̃L2 } = θ .
From the variation in v∗ we obtain

∂ G∗ (v∗0 , z∗0 )
− Λ (u0 ) = θ , (13.16)
∂ v∗
that is,
∂ G(Λ u0 )
v∗03 + z∗0 = ,
∂ v3
∂ G(Λ u0 )
v∗01 = ,
∂ v1
∂ G(Λ u0 )
v∗02 = ,
∂ v2
and
G∗ (v∗0 , z∗0 ) = Λ3 u0 , v∗03 + z∗0 L2 + Λ1 u0 , v∗01 L2

Λ2 u0 , v∗02 L2 − G(Λ u0). (13.17)
From the variation in z∗ we get
∂ G∗ (v∗0 , z∗0 )
− ((z0 )∗11 )xx /K − = θ, (13.18)
∂ z∗11
∂ G∗ (v∗0 , z∗0 )
− ((z0 )∗22 )yy /K − = θ, (13.19)
∂ z∗22
that is, from this and (13.16) we get
−((z0 )∗11 )x = K(w0 )x ,
−((z0 )∗22 )y = K(w0 )y ,

so that
F̃ ∗ (z∗0 ) = (z0 )∗11 , (w0 )xx L2 (Ω ) + (z0 )∗22 , (w0 )yy L2 (Ω ) − F(Λ2 u0 ). (13.20)
13.3 Another Duality Principle 349
From the variation in u, we get
Λ ∗ v∗0 − P̃ = 0, (13.21)
so that v∗0 ∈ A∗ . Therefore, from (13.17), (13.20), and (13.21) we have
F̃ ∗ (z∗0 ) − G∗(v∗0 , z∗0 )

= G(Λ u0 ) − F2(Λ u0 ) − u0, P̃L2
= J(u0 ). (13.22)
To complete the proof just observe that from the condition indicated in (13.10), the
extremal relations (13.18) and (13.19) refer to a global optimization (in z∗ , for a
fixed v∗0 ), so that the infimum indicated in the dual formulation is attained for the z∗0
in question.
13.3 Another Duality Principle
In this section we present another result, which is summarized by the following

theorem.
Theorem 13.3.1. Considering the introduction statements, let (G ◦ Λ ) : U → R be

expressed by

1
G(Λ u) = Hαβ λ μ γαβ (u)γλ μ (u) dx
2 Ω

1
+ hαβ λ μ καβ (u)κλ μ (u) dx, (13.23)
2 Ω
where
Λ (u) = {γ (u), κ (u)},
uα ,β + uβ ,α w,α w,β
γαβ (u) = + ,
2 2
καβ (u) = −w,αβ ,
where u = (uα , w) ∈ U = W01,2 (Ω ; R2 ) × W02,2 (Ω ).

As above, define F : U → R by
F(u) = w, PL2 + uα , Pα L2 ,
and J : U → R by
J(u) = (G ◦ Λ )(u) − F(u).
inf {J(u)} ≥ sup {−G∗1 (N) − G̃∗2 (−N)}, (13.24)

u∈U N∈A∗
where
1
G1 (γ (u)) = H γ (u)γλ μ (u) dx,

1
G2 (u) = h κ (u)κλ μ (u) dx − F(u),
so that
J(u) = G1 (γ (u)) + G2 (u).
Moreover,
G∗1 (N) = sup{v, NL2 − G1(v)}

v∈Y

1
= H αβ λ μ Nαβ Nλ μ dx, (13.25)
2 Ω
and
G̃∗2 (−N) = sup{γ (u), −NL2 − G2 (u)},

u∈U

1 1
= hαβ λ μ ŵ,αβ ŵ,λ μ dx + Nαβ ŵ,α ŵ,β dx,
2 Ω 2 Ω
if
N = {Nαβ } ∈ A∗ = A1 ∩ A2 ,
where ŵ ∈ W02,2 (Ω ) is the solution of equation
(hαβ λ μ ŵ,λ μ ),αβ − (Nαβ ŵ,α ),β − P = 0, in Ω .
Also,
A1 = {N ∈ Y ∗ | J(w)
˜ > 0, ∀w ∈ W02,2 (Ω ) such that w = θ },

1 1
˜
J(w) = hαβ λ μ w,αβ w,λ μ dx + Nαβ w,α w,β dx,
2 Ω 2 Ω
and
A2 = {N ∈ Y ∗ | Nαβ ,β + Pα = 0 in Ω }.
Finally, if there exists u0 ∈ U such that δ J(u0 ) = θ and N0 = {Nαβ (u0 )} ∈ A1 ,
where Nαβ (u0 ) = Hαβ λ μ γλ μ (u0 ), then
J(u0 ) = min{J(u)}
u∈U
= max∗ {−G∗1 (N) − G̃∗2 (−N)}
N∈A
= −G∗1 (N0 ) − G̃∗2 (−N0 )}. (13.26)
13.3 Another Duality Principle 351
Proof. Clearly
J(u) = G1 (γ (u)) + G2 (u)

= −γ (u), NL2 + G1 (γ (u)) + γ (u), NL2 + G2 (u)
≥ inf {−v, NL2 + G1 (v)} + inf {−γ (u), −NL2 + G2 (u)}
v∈Y u∈U
= −G∗1 (N) − G̃∗2 (−N), ∀u ∈ U, N ∈ Y ∗ . (13.27)
Hence,
inf {J(u)} ≥ sup {−G∗1 (N) − G̃∗2 (−N)}. (13.28)
u∈U N∈A∗
Now suppose there exists u0 ∈ U such that δ J(u0 ) = θ and N0 = {Nαβ (u0 )} ∈ A1 .
First, note that from δ J(u0 ) = θ , the following extremal equation is satisfied:
(N0 )αβ ,β + Pα = 0 in Ω ,
that is, N0 ∈ A2 , so that N0 ∈ A1 ∩ A2 = A∗ .

Thus, from N0 ∈ A1 , we obtain
G̃∗2 (−N0 ) = sup{γ (u), −N0 L2 − G2 (u)},

u∈U
= γ (û), −N0 L2 − G2 (û)
) *
ŵ,α ŵ,β 1
= , −(N0 )αβ − h ŵ ŵ dx
2 L2 2 Ω αβ λ μ ,αβ ,λ μ
+ŵ, PL2 , (13.29)
where ŵ ∈ W02,2 (Ω ) is the solution of equation

(N0 )αβ ŵ,α ,β − (hαβ λ μ ŵ,λ μ ),αβ + P = 0, in Ω . (13.30)
Replacing such a relation in (13.29), we obtain

1 1
G̃∗2 (−N0 ) = hαβ λ μ ŵ,αβ ŵ,λ μ dx + (N0 )αβ ŵ,α ŵ,β dx.
2 Ω 2 Ω
Hence, also from the equation δ J(u0 ) = θ and (13.30), we may get ŵ = w0 , so
that from this and (13.29), we obtain
G̃∗2 (−N0 ) = γ (u0 ), −N0 L2 − G2 (u0 ). (13.31)
Finally, considering that
(N0 )αβ = Hαβ λ μ γλ μ (u0 ),
we get
G∗1 (N0 ) = γ (u0 ), N0 L2 − G1 (γ (u0 )),
so that
− G∗1(N0 ) − G̃∗2(−N0 ) = −γ (u0 ), N0 L2 + G1 (γ (u0 ))

−γ (u0 ), −N0 L2 + G2 (u0 )
= G1 (γ (u0 )) + G2 (u0 )
= J(u0 ). (13.32)
From this and (13.28) and also from the fact that N0 ∈ A∗ , the proof is complete.
Remark 13.3.2. From the last duality principle, we may write
inf {J(u)} ≥ sup {−J˜∗(M, N, u)},

u∈U
(M,N,u)∈Â
where
J˜∗ (M, N, u) = G∗1 (N) + G̃∗2 (−N)

1 1
= H N N dx + h M M dx
2 Ω αβ λ μ αβ λ μ 2 Ω αβ λ μ αβ λ μ

1
+ N w,α w,β dx, (13.33)
2 Ω αβ
Â = A1 ∩ A2 ∩ A3 ∩ A4 ,
A3 = {(M, N, u) ∈ Y ∗ × Y ∗ × U | Mαβ ,αβ + (Nαβ w,α ),β + P = 0, in Ω },
A4 = {(M, N, u) ∈ Y ∗ × Y ∗ × U | {Mαβ } = {hαβ λ μ (−w,λ μ )}, in Ω },

and A1 and A2 as above specified, that is,
A1 = {N ∈ Y ∗ | J(w)
˜ > 0, ∀w ∈ W02,2 (Ω ) such that w = θ },
where
1 1
˜
J(w) = hαβ λ μ w,αβ w,λ μ dx + Nαβ w,α w,β dx,
2 Ω 2 Ω
and
A2 = {(M, N, u) ∈ Y ∗ × Y ∗ × U | Nαβ ,β + Pα = 0 in Ω }.
Finally, we could suggest as a possible approximate dual formulation the problem
of maximizing −JK∗ (M, N, u) on A1 ∩ A2 ∩ A3 , where K > 0 and

1 1
JK∗ (M, N, u) = H αβ λ μ Nαβ Nλ μ dx + hαβ λ μ Mαβ Mλ μ dx
2 Ω 2 Ω

1
+ N w,α w,β dx
2 Ω αβ
K 2
2 α ,∑
+ Mαβ − hαβ λ μ (−w,λ μ ) 22 . (13.34)
β =1
13.4 An Algorithm for Obtaining Numerical Results 353
A study about the system behavior as K → +∞ is planned for a future work. Any-
way, big values for K > 0 allow the gap function 12 Ω Nαβ w,α w,β dx to be nonposi-
tive at a possible optimal point inside the region of convexity of JK∗ .
13.4 An Algorithm for Obtaining Numerical Results
In this section we develop an algorithm which we prove, under certain mild hy-
potheses; it is convergent up to a subsequence (the result stated in the next lines
must be seen as an existence one and, of course, it is not the full proof of conver-
gence from a numerical analysis point of view). Such a result is summarized by the
following theorem.
Theorem 13.4.1. Consider the system of equations relating the boundary value
form of the Kirchhoff–Love plate model, namely
⎧
⎪
⎪ Mαβ ,αβ + (Nαβ w,α ),β + P = 0, in Ω
⎪
⎪
⎨
Nαβ ,β + Pα = 0 in Ω (13.35)
⎪
⎪
⎪
⎪
⎩
uα = w = ∂∂ wn = 0 on ∂ Ω
where
Nαβ (u) = Hαβ λ μ γλ μ (u), (13.36)
Mαβ (u) = hαβ λ μ κλ μ (u). (13.37)
Define, as above,

1
(G ◦ Λ )(u) = Nαβ (u)γαβ (u) dx
2 Ω

1
+ Mαβ (u)καβ (u) dx, (13.38)
2 Ω
F(u) = w, PL2 + uα , Pα L2 (13.39)

and J : U → R by
J(u) = G(Λ u) − F(u), ∀u ∈ U. (13.40)
Assume { Pα 2 } are small enough so that (from [22] pages 285–287) if either
uα W 1,2 (Ω ) → ∞
or
w W 2,2 (Ω ) → ∞,
then
J(u) → +∞.
Let {un = ((un )α , wn )} ⊂ U be the sequence obtained through the following al-
gorithm:
1. Set n = 1.
2. Choose (z∗1 )1 , (z∗2 )1 ∈ L2 (Ω ).
3. Compute un by

K K
un = argminu∈U G(Λ u) + (wx )2 dx + (wy )2 dx
2 Ω 2 Ω
−wx , (z∗1 )n L2 − wy , (z∗2 )n L2

1 ∗ 2 1 ∗ 2
+ (z ) dx + (z ) dx − F(u) ,
2K Ω 1 n 2K Ω 2 n
which means to solve the equation

⎧
⎪
⎪ Mαβ ,αβ + (Nαβ w,α ),β + P + Kw,αα − (z∗n )α ,α = 0, in Ω
⎪
⎪
⎨
Nαβ ,β + Pα = 0 in Ω (13.41)
⎪
⎪
⎪
⎪
⎩
uα = w = ∂∂ wn = 0 on ∂ Ω
4. Compute z∗n+1 = ((z∗1 )n+1 , (z∗2 )n+1 ) by

K K
z∗n+1 = argminz∗ ∈L2 ×L2 G(Λ un ) + (wn )2x dx + (wn )2y dx
2 Ω 2 Ω
−(wn )x , z∗1 L2 − (wn )y , z∗2 L2

1 ∗ 2 1 ∗ 2
+ (z ) dx + (z ) dx − F(un ) ,
2K Ω 1 2K Ω 2
that is,
(z∗1 )n+1 = K(wn )x ,
and
(z∗2 )n+1 = K(wn )y .
5. Set n → n + 1 and go to step 3 till the satisfaction of a suitable approximate
convergence criterion.
Assume {un = ((un )α , wn )} ⊂ U is such that for a sufficiently big K > 0 we have
N11 (un ) + K > 0, N22 (un ) + K > 0,

and (N11 (un ) + K)(N22 (un ) + K) − N12
2
(un ) > 0,
in Ω , ∀n ∈ N. (13.42)
Under such assumptions, the sequence {un } is uniquely defined (depending

only on (z∗ )1 ), and such that, up to a subsequence not relabeled, for some u0 =
((u0 )α , w0 ) ∈ U, we have
(un )α (u0 )α , weakly in W01,2 (Ω ),
(un )α → (u0 )α , strongly in L2 (Ω ),

wn w0 , weakly in W02,2 (Ω ),
and
wn → w0 , strongly in W01,2 (Ω ),
where
u0 ∈ U
is a solution for the system of equations indicated in (13.35).
Proof. Since J : U → R is defined by
J(u) = G(Λ u) − F(u), (13.43)
we have

K K
J(u) = G(Λ u) + (wx )2 dx + (wy )2 dx
2 Ω 2 Ω
−wx , z∗1 L2 − wy , z∗2 L2

K K
− (wx )2 dx − (wy )2 dx
2 Ω 2 Ω
+wx , z∗1 L2 + wy , z∗2 L2 − F(u)

K K
≤ G(Λ u) + (wx )2 dx (wy )2 dx
2 Ω 2 Ω
−wx , z∗1 L2 − wy , z∗2 L2

K K
+ sup − v1 dx −
2
v22 dx
v∈L ×L
2 2 2 Ω 2 Ω
+v1 , z∗1 L2 + v2, z∗2 L2 } − F(u)

K K
= G(Λ u) + (wx )2 dx + (wy )2 dx +
2 Ω 2 Ω
−wx , z∗1 L2 − wy , z∗2 L2

1 1
+ (z∗1 )2 dx + (z∗ )2 dx − F(u), (13.44)
2K Ω 2K Ω 2
∀u ∈ U, z∗ ∈ L2 (Ω ) × L2 (Ω ).
From the hypotheses, {un} is inside the region of strict convexity of the func-
tional in U (for z∗ fixed) in question, so that it is uniquely defined for each z∗n
(through the general results in [11] we may infer the region of convexity of the
functional

K K
J(u) = G(Λ u) + (wx )2 dx + (wy )2 dx +
2 Ω 2 Ω
−wx , (z∗1 )n L2 − wy , (z∗2 )n L2

1 1
+ (z∗1 )2n dx + (z∗ )2 dx − F(u), (13.45)
2K Ω 2K Ω 2 n
corresponds to the satisfaction of constraints
N11 (u) + K > 0, N22 (u) + K > 0,

and (N11 (u) + K)(N22(u) + K) − N12
2
(u) > 0, in Ω ). (13.46)
Denoting

K K
αn = G(Λ un ) + (wn )2x dx + (wn )2y dx +
2 Ω 2 Ω
−(wn )x , (z∗1 )n L2 − (wn )y , (z∗2 )n L2

1 1
+ (z∗1 )2n dx + (z∗ )2 dx − F(un ), (13.47)
2K Ω 2K Ω 2 n
we may easily verify that {αn } is a real nonincreasing sequence bounded below by
infu∈U {J(u)}; therefore, there exists α ∈ R such that
lim αn = α . (13.48)
n→∞
From the hypotheses

J(u) → +∞,
if either uα W 1,2 (Ω ) → ∞ or w W 2,2 (Ω ) → ∞.
0 0
From this, (13.44), (13.47), and (13.48) we may infer there exists C > 0 such that
wn W 2,2 (Ω ) < C, ∀n ∈ N,
0
and
(uα )n W 1,2 (Ω ) < C, ∀n ∈ N.
0
Thus, from the Rellich–Kondrachov theorem, up to a subsequence not relabeled,

there exists u0 = ((u0 )α , w0 ) ∈ U such that
(un )α → (u0 )α , strongly in L2 (Ω ),

wn w0 , weakly in W02,2 (Ω ),
and
wn → w0 , strongly in W01,2 (Ω ),
so that, considering the algorithm in question,
zn → z∗0 strongly in L2 (Ω ; R2 ),
where
(z∗0 )α = K(w0 ),α .
From these last results, the Sobolev imbedding theorem and relating results (more
specifically, Korn’s inequality and its consequences; details may be found in [21]),
we have that there exist K1 , K2 > 0 such that
(un )α ,β + (un)β ,α 2 < K1 , ∀n ∈ N, α , β ∈ {1, 2},
and
(wn ),α 4 < K2 , ∀n ∈ N, α ∈ {1, 2}.
On the other hand, un ∈ U such that

K K
un = argminu∈U G(Λ u) + (wx )2 dx + (wy )2 dx
2 Ω 2 Ω
−wx , (z∗1 )n L2 − wy , (z∗2 )n L2

1 1
+ (z∗1 )2n dx + (z∗2 )2n dx − F(u) (13.49)
2K Ω 2K Ω
is also such that

hαβ λ μ (wn ),λ μ ,αβ

(un )λ ,μ + (un )λ ,μ (wn ),λ (wn ),μ
− Hαβ λ μ + (wn ),β
2 2 ,α
−K(wn ),αα + (z∗n )α ,α − P = 0 in Ω , (13.50)
and

(un )λ ,μ + (un )λ ,μ (wn ),λ (wn ),μ
Hαβ λ μ +
2 2 ,β
+Pα = 0 in Ω , (13.51)
in the sense of distributions (theoretical details about similar results may be found
in [25]).
Fix φ ∈ Cc∞ (Ω ). In the next lines, we will prove that

) *
(un )α ,β + (un )β ,α
+ (wn ),α (wn ),β (wn ),α , φ,β
2 L2
) *
(u0 )α ,β + (u0 )β ,α
→ + (w0 ),α (w0 ),β (w0 ),α , φ,β , (13.52)
2 L2
as n → ∞, ∀α , β ∈ {1, 2} (here the repeated indices do not sum).

Observe that, since
from the Hölder inequality, we obtain

) *
(un )α ,β + (un)β ,α
(wn ),α , φ,β
2 L2
) *
(u0 )α ,β + (u0 )β ,α
− (w0 ),α , φ,β
2 2

L
)
(un )α ,β + (un)β ,α (un )α ,β + (un )β ,α
= (wn ),α − (w0 ),α
2 2
*
(un )α ,β + (un )β ,α (u0 )α ,β + (u0)β ,α
+ (w0 ),α − (w0 ),α , φ,β
2 2
L2

(un )α ,β + (un)β ,α
≤
(wn ),α − (w0 ),α 2 φ,β ∞

2 2
) *
(un )α ,β + (un )β ,α (u0 )α ,β + (u0 )β ,α
+ (w0 ),α − (w0 ),α , φ,β
2 2
L2
≤ K1 (wn ),α − (w0 ),α 2 φ,β ∞
) *
(un )α ,β + (un )β ,α (u0 )α ,β + (u0 )β ,α
+ − , (w0 ),α φ,β

2 2 L2
→ 0, as n → ∞.
Moreover, from the generalized Hölder inequality, we get

, - , -
(wn ),α (wn ),β (wn ),α , φ,β 2 − (w0 ),α (w0 ),β (w0 ),α , φ,β 2
, L L
-

= (wn ),α (wn ),β − (wn ),α (w0 ),β + (wn ),α (w0 ),β − (w0 )2,α (w0 ),β , φ,β L2
2 2 2
, - , -

≤ (wn )2,α ((wn ),β − (w0 ),β ), φ,β L2 + (wn )2,α − (w0 )2,α )(w0 ),β , φ,β L2
, -

≤ (wn )2,α ((wn ),β − (w0 ),β ), φ,β L2
, -
+ ((wn ),α + (w0 ),α )((wn ),α − (w0 ),α )(w0 ),β , φ,β 2 L
≤ (wn ),α 24 (wn ),α − (w0 ),α 2 φ,β ∞

+ (wn ),α + (w0 ),α 4 (wn ),α − (w0 ),α 2 (w0 ),β 4 φ,β ∞
≤ (K22 + 2K2 K2 ) (wn ),α − (w0 ),α 2 φ,β ∞
→ 0, as n → ∞,
where also up to here the repeated indices do not sum.

Thus (13.52) has been proven, so that we may infer that

hαβ λ μ (w0 ),λ μ ,αβ

(u0 )λ ,μ + (u0)λ ,μ (w0 ),λ (w0 ),μ
− Hαβ λ μ + (w0 ),β − P, φ
2 2 ,α L2
./
= lim hαβ λ μ (wn ),λ μ ,αβ
n→∞

(un )λ ,μ + (un)λ ,μ (wn ),λ (wn ),μ
− Hαβ λ μ + (wn ),β
2 2 ,α

−K(wn ),αα + (z∗n )α ,α − P, φ
L2
= lim 0 = 0.
n→∞
Since φ ∈ Cc∞ (Ω ) is arbitrary, we obtain
(hαβ λ μ (w0 ),λ μ ),αβ

(u0 )λ ,μ + (u0 )λ ,μ (w0 ),λ (w0 ),μ
− Hαβ λ μ + (w0 ),β − P = 0, in Ω
2 2 ,α
in the distributional sense.

Similarly,

(u0 )λ ,μ + (u0 )λ ,μ (w0 ),λ (w0 ),μ
Hαβ λ μ +
2 2 ,β
+Pα = 0, in Ω , (13.53)
for α ∈ {1, 2}, also in the distributional sense.

From the convergence in question, we also get in a weak sense
∂ w0
(u0 )α = w0 = = 0, on ∂ Ω .
∂n
Remark 13.4.2. We emphasize that for each n ∈ N, from the condition indicated in
(13.42), {un } is obtained through the minimization of a convex functional. There-
fore the numerical procedure translates into the solution of a sequence of convex
optimization problems.
13.5 Numerical Results
In this section we present some numerical results. Let Ω = [0, 1] × [0, 1] and
consider the problem of minimizing J : U → R where
J(u) = G(Λ u) − F(u),

1 1
(G ◦ Λ )(u) = Nαβ (u)γαβ (u) dx + Mαβ (u)καβ (u) dx, (13.54)
2 Ω 2 Ω
{Nαβ } denote the membrane force tensor and {Mαβ } the moment one, so that
from the constitutive relations,
Nαβ (u) = Hαβ λ μ γλ μ (u), (13.55)
Mαβ (u) = hαβ λ μ κλ μ (u). (13.56)

Also, F : U → R is given by
F(u) = w, PL2 + uα , Pα L2 . (13.57)
Here
U = {(uα , w) ∈ W 1,2 (Ω ; R2 ) × W 2,2 (Ω ) | uα = 0 on Γ0 , w = 0 on ∂ Ω },
Γ0 = {[0, y] ∪ [x, 0], 0 ≤ x, y ≤ 1}, and P, P1 , P2 ∈ L2 denote the external loads in the
directions a3 , a1 , and a2 , respectively.
We consider the particular case where all entries of {Hαβ λ μ } and {hαβ λ μ } are
zero, except for H1111 = H2222 = H1212 = 105 and h1111 = h2222 = h1212 = 104 .
Moreover P = 1000, P1 = −100, and P2 = −100 (units refer to the international
system). In a first moment, define the trial functions w : Ω → R, u1 : Ω → R, and
u2 : Ω → R by
w(x, y) = a1 sin(π x) sin(π y) + a2 sin(2π x) sin(2π y),
u1 (x, y) = a3 sin(π x/2) sin(π y/2),

u2 (x, y) = a4 sin(π x/2) sin(π y/2),
respectively.
The coefficients {a1 , a2 , a3 , a4 } will be found through the extremal points of J.
We have obtained only one real critical point, namely,
13.5 Numerical Results 361
(a0 )1 = 0.000832
(a0 )2 = −1.038531 ∗ 10−8

(a0 )3 = −0.000486
(a0 )4 = −0.000486
so that the candidate to optimal point is ((u0 )1 , (u0 )2 , w0 ) where
w0 (x, y) = (a0 )1 sin(π x) sin(π y) + (a0)2 sin(2π x) sin(2π y),
(u0 )1 (x, y) = (a0 )3 sin(π x/2) sin(π y/2),

(u0 )2 = (a0 )4 sin(π x/2) sin(π y/2).
With such values for the coefficients, it is clear that N11 (u0 ) and N22 (u0 ) are negative
in Ω so that {Nαβ (u0 )} is not positive definite. Even so, as we shall see in the
next lines, the optimality criterion of the second duality principle developed may be
applied. Let
w(x, y) = a1 sin(π x) sin(π y) + a2 sin(2π x) sin(2π y).
We have that

1 1
W (a1 , a2 ) = h w w dx + (N(u0 ))αβ w,α w,β dx
2 Ω αβ λ μ ,αβ ,λ μ 2 Ω
= 360319.a21 + 191.511a1a2 + 5.7668 ∗ 106a22 . (13.58)
Now observe that

∂ 2W (a1 , a2 )
s11 = = 720638.0,
∂ a21
∂ 2W (a1 , a2 )
s22 = = 1.15336 ∗ 107,
∂ a22
∂ 2W (a1 , a2 )
s12 = = 191.511.
∂ a1 ∂ a2
Therefore s11 > 0, s22 > 0, and s11 s22 − s212 = 8.31155 ∗ 1012 > 0 so that
W (a1 , a2 ) is a positive definite quadratic form.
Hence, from the second duality principle, we may conclude that ((u0 )1 , (u0 )2 , w0 )
is indeed the optimal solution (approximate global minimizer for J).
Refining the results through finite differences using the algorithm of last section,
we obtain again the field of displacements w0 (x, y) (please see Fig. 13.1).
x 10−4
6
5
4
3
2
1
0
1.5
1 1.5
1
0.5 0.5
0 0
Fig. 13.1 Vertical axis: w0 (x, y)-field of displacements
13.6 Conclusion
In this chapter, we develop duality principles for the Kirchhoff–Love plate model.
The results are obtained through the basic tools of convex analysis and include suf-
ficient conditions of optimality. It is worth mentioning that earlier results require
the membrane force tensor to be positive definite in a critical point in order to
guarantee global optimality, whereas from the new results here presented, we are
able to guarantee global optimality for a critical point such that N11 (u0 ) < 0 and
N22 (u0 ) < 0, in Ω . Finally, the methods developed here may be applied to many
other nonlinear models of plates and shells. Applications to related areas (specially
to the shell models found in [23]) are planned for future works.
Chapter 14
About Ginzburg–Landau-Type Equations:
The Simpler Real Case
14.1 Introduction
In this chapter, our first objectives are to show existence and develop dual for-
mulations concerning the real semi-linear Ginzburg–Landau equation. We start by
describing the primal formulation.
By S ⊂ R3 we denote an open connected bounded set with a sufficiently regular
boundary Γ = ∂ S (regular enough so that the Sobolev imbedding theorem holds).
The Ginzburg–Landau equation is given by
2
−∇2 u + α ( u2 − β )u − f = 0 in S, (14.1)
u = 0 on Γ ,
where u : S → R denotes the primal field and f ∈ L2 (S). Moreover, α , β are real
positive constants.
Remark 14.1.1. The complex Ginzburg–Landau equation plays a fundamental role
in the theory of superconductivity (see [4], 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

1 α u2
J(u) = |∇u|2 dx + ( − β )2 dx − f u dx (14.2)
2 S 2 S 2 S
where U = {u ∈ W 1,2 (S) | u = 0 on Γ } = W01,2 (S).

We are particularly concerned with the fact that equations indicated in (14.1) are
necessary conditions for the solution of problem P, where
Problem P : to find u0 ∈ U such that J(u0 ) = min{J(u)}.

u∈U
364 14 About Ginzburg–Landau-Type Equations: The Simpler Real Case
14.1.1 Existence of Solution for the Ginzburg–Landau Equation
We start with a remark.
Remark 14.1.2. From the Sobolev imbedding theorem for

mp < n, n − mp < n, p ≤ q ≤ p∗ = np/(n − mp),
we have
W j+m,p (Ω ) → W j,q (Ω ).
Therefore, considering n = 3, m = 1, j = 0, p = 2, and q = 4, we obtain
W 1,2 (Ω ) ⊂ L4 (Ω ) ⊂ L2 (Ω )
and thus
u L4 (Ω ) → +∞ ⇒ u W 1,2 (Ω ) → +∞.
Furthermore, from the above and the Poincaré inequality, it is clear that for J given
by (14.2), we have
J(u) → +∞ as u W 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 14.1.3. For α , β ∈ R+ , f ∈ L2 (S) there exists at least one u0 ∈ U such that
J(u0 ) = min{J(u)}
u∈U
where
1 α u2
J(u) = |∇u| dx +
2
( − β )2 dx − f u dx
2 S 2 S 2 S
and U = {u ∈ W 1,2 (S) | u = 0 on Γ } = W01,2 (S).
Proof. From Remark 14.1.2 we have
J(u) → +∞ as u U → +∞.
Also from the Poincaré inequality, there exists α1 ∈ R such that α1 =
infu∈U {J(u)} so that for {un } minimizing sequence, in the sense that
J(un ) → α1 as n → +∞ (14.3)
we have that un U is bounded, and thus, as W01,2 (S) is reflexive, there exists u0 ∈
W01,2 (S) and a subsequence {un j } ⊂ {un } such that
un j u0 , weakly in W01,2 (S). (14.4)

14.2 A Concave Dual Variational Formulation 365
From (14.4), by the Rellich–Kondrachov theorem, up to a subsequence, which is

also denoted by {un j }, we have
un j → u0 , strongly in L2 (S). (14.5)
Furthermore, defining J1 : U → R as

1 α
J1 (u) = |∇u|2 dx + u4 dx − f u dx
2 S 8 S S
we have that J1 : U → R is convex and strongly continuous, therefore weakly lower

semicontinuous, so that
lim inf{J1 (un j )} ≥ J1 (u0 ). (14.6)

j→+∞
On the other hand, from (14.5),

(un j )2 dx → u20 dx, as j → +∞ (14.7)
S S
and thus, from (14.6) and (14.7), we may write
α1 = inf {J(u)} = lim inf{J(un j )} ≥ J(u0 ).

u∈U j→+∞
14.2 A Concave Dual Variational Formulation
We start this section by enunciating the following theorem which has been proven
in [11].
Theorem 14.2.1. Let U be a reflexive Banach space, (G ◦ Λ ) : U → R̄ a convex

Gâteaux differentiable functional, and (F ◦ Λ1 ) : U → R̄ convex, coercive, and lower
semicontinuous (l.s.c.) such that the functional
J(u) = (G ◦ Λ )(u) − F(Λ1 u) − u, u∗0U
is bounded from below, where Λ : U → Y and Λ1 : U → Y1 are continuous linear

operators.
Then we may write
inf sup {F ∗ (z∗ ) − G∗ (v∗ )} ≥ inf {J(u)}

z∗ ∈Y1∗ v∗ ∈B∗ (z∗ ) u∈U
where B∗ (z∗ ) = {v∗ ∈ Y ∗ such that Λ ∗ v∗ − Λ1∗ z∗ − u∗0 = 0}.
Our next result refers to a convex dual variational formulation, through which we
obtain sufficient conditions for optimality.
Theorem 14.2.2. Consider J : U → R, where

1 α u2
J(u) = |∇u|2 dx + ( − β )2 dx − f u dx,
S2 S 2 2 S
and U = W01,2 (S). For K = 1/K0 , where K0 stands for the constant related to the
Poincaré inequality, we have the following duality principle:
inf {J(u)} ≥ sup {−G∗L (v∗ , z∗ )}

u∈U (z∗ ,v∗1 ,v∗0 )∈B∗
where

1 1 1 (v∗1 )2
G∗L (v∗ , z∗ ) = |∇z∗ |2 dx − (z∗ )2 dx + ∗ dx
2K 2 S 2K S 2 S v0 + K

1
+ (v∗0 )2 dx + β v∗0 dx, (14.8)
2α S S
and
B∗ = {(z∗ , v∗1 , v∗0 ) ∈ L2 (S; R3 ) |

1
− ∇2 z∗ + v∗1 − z∗ = f , v∗0 + K > 0, a.e. in S, z∗ = 0 on Γ }.
K
If in addition there exists u0 ∈ U such that δ J(u0 ) = θ and v̄∗0 + K = (α /2)u20 − β +
K > 0, a.e. in S, then
J(u0 ) = min{J(u)} = max {−G∗L (v∗ , z∗ )} = −G∗L (v̄∗ , z̄∗ ),

u∈U (z∗ ,v∗1 ,v∗0 )∈B∗
where
α 2
v̄∗0 = u −β,
2 0
v̄∗1 = (v̄∗0 + K)u0
and
z̄∗ = Ku0 .
Proof. Observe that we may write

J(u) = G(Λ u) − F(Λ1 u) − f u dx,
S
where
1 α u2 K
G(Λ u) = |∇u|2 dx + ( − β + 0)2 dx + u2 dx,
S2 S 2 2 2 S

K
F(Λ1 u) = u2 dx,
2 S
14.2 A Concave Dual Variational Formulation 367
where
Λ u = {Λ0 u, Λ1 u, Λ2 u},
and
Λ0 u = 0, Λ1 u = u, Λ2 u = ∇u.
From Theorem 14.2.1 (here this is an auxiliary theorem through which we obtain
A∗ , below indicated), we have
inf {J(u)} = ∗inf ∗ sup {F ∗ (z∗ ) − G∗ (v∗ )},

u∈U z ∈Y1 v∗ ∈A∗
where

1
F ∗ (z∗ ) = (z∗ )2 dx,
2K S
and

∗ ∗ 1 1 (v∗1 )2 1
G (v ) = |v∗2 |2 dx + ∗ + K dx + 2α (v∗0 )2 dx + β v∗0 dx,
2 S 2 v
S 0 S S
if v∗0 + K > 0, a.e. in S, and
A∗ = {v∗ ∈ Y ∗ | Λ ∗ v∗ − Λ1∗ z∗ − f = 0},
or
A∗ = {(z∗ , v∗ ) ∈ L2 (S) × L2(S; R5 ) |

− div(v∗2 ) + v∗1 − z∗ − f = 0, a.e. in S}.
Observe that
G∗ (v∗ ) ≥ Λ u, v∗ Y − G(Λ u), ∀u ∈ U, v∗ ∈ A∗ ,
and thus
−F ∗ (z∗ ) + G∗ (v∗ ) ≥ −F ∗ (z∗ ) + Λ1u, z∗ L2 (S) + u, f U − G(Λ u), (14.9)
and hence, making z∗ an independent variable through A∗ , from (14.9), we may write

∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
sup {−F (z ) + G (v2 (v1 , z ), v1 , v0 )} ≥ sup −F ∗ (z∗ )
z∗ ∈L2 (S) z∗ ∈L2 (S)

+Λ1 u, z∗ L2 (S) + f u dx − G(Λ u) , (14.10)
S
so that

1 1 1 (v∗1 )2
sup − (z∗ )2 dx + (v∗2 (z∗ , v∗1 ))2 dx + ∗ dx
z∗ ∈L2 (S) 2K S 2 S 2 S v0 + K

1
+ (v∗0 )2 dx + β v∗0 dx
2α S S

≥ F(Λ1 u) + f udx − G(Λ u). (14.11)
S
Therefore, if K ≤ 1/K0 (here K0 denotes the constant concerning the Poincaré

inequality), the supremum in the left side of (14.11) is attained through the relations
∇z∗
v∗2 = and z∗ = 0 on Γ ,
K
so that the final format of our duality principle is given by

1 1
inf {J(u)} ≥ sup − 2 |∇z∗ |2 dx + (z∗ )2 dx
u∈U ∗ ∗ ∗
(z ,v ,v )∈B ∗ 2K S 2K S
1 0

1 (v∗1 )2 1
− ∗ + K dx − 2α (v∗0 )2 dx − β v∗0 dx , (14.12)
2 v
S 0 S S
where
B∗ = {(z∗ , v∗1 , v∗0 ) ∈ L2 (S; R3 ) |

1
− ∇2 z∗ + v∗1 − z∗ = f , v∗0 + K > 0, a.e. in S, z∗ = 0 on Γ }.
K
The remaining conclusions follow from an application (with little changes) of
Theorem 10.1.25.
Remark 14.2.3. The relations

∇z∗
v∗2 = and z∗ = 0 on Γ ,
K
are sufficient for the attainability of the supremum indicated in (14.11) but just par-
tially necessary; however, we assume them because the expression of dual problem
is simplified without violating inequality (14.12) (in fact the difference between the
primal and dual functionals even increases under such relations).
In this section we present numerical results for a one-dimensional example

originally due to Bolza (see [50] for details about the primal formulation).
Consider J : U → R expressed as
1 1
1 1
J(u) = ((u,x )2 − 1)2 dx + (u − f )2 dx
2 0 2 0
or, defining S = [0, 1],

1
1
G(Λ u) = ((u,x )2 − 1)2 dx
2 0
and 1
1
F(u) = (u − f )2 dx
2 0
we may write
J(u) = G(Λ u) + F(u)
where, for convenience, we define Λ : U → Y ≡ L4 (S) × L2(S) as
Λ u = {u,x , 0}.
Furthermore, we have
U = {u ∈ W 1,4 (S) | u(0) = 0 and u(1) = 0.5}
For Y = Y ∗ = L4 (S) × L2(S), defining

1
G(Λ u + p) = ((u,x + p1 )2 − 1.0 + p0)2 dx
2 S
for v∗0 > 0, we obtain
G(Λ u) + F(u) ≥ inf {−p0 , v∗0 L2 (S) − p1, v∗1 L2 (S) + G(Λ u + p) + F(u)}
p∈Y
or
G(Λ u) + F(u) ≥ inf {−q0 , v∗0 L2 (S) − q1, v∗1 L2 (S) + G(q)
p∈Y
+ 0, v0L2 (S) + u, v∗1 L2 (S) + F(u)}.

∗
Here q = Λ u + p so that
G(Λ u) + F(u) ≥ −G∗ (v∗ ) + 0, v∗0 L2 (S) + u,x , v∗1 L2 (S) + F(u).
That is,
G(Λ u) + F(u) ≥ −G∗ (v∗ ) + inf {0, v∗0 L2 (S) + u,x , v∗1 L2 (S) + F(u)},
u∈U
or
inf {G(Λ u) + F(u)} ≥ sup {−G∗ (v∗ ) − F ∗ (−Λ ∗ v∗ )}
u∈U v∗ ∈A∗
where
1 (v∗1 )2 1
G∗ (v∗ ) = ∗ dx + (v∗0 )2 dx,
2 S v0 2 S
if v∗0 > 0, a.e. in S. Also

1
F ∗ (−Λ ∗ v∗ ) = [(v∗1 ),x ]2 dx + f , (v∗1 ),x L2 (S) − v∗1 (1)u(1)
2 S
and
A∗ = {v∗ ∈ Y ∗ | v∗0 > 0, a.e. in S}.
Remark 14.3.1. Through the extremal condition v∗0 = ((u,x )2 − 1) and Weierstrass
condition (u,x )2 − 1.0 ≥ 0 we can see that the dual formulation is convex for v∗0 > 0;
however, it is possible that the primal formulation has no minimizers, and we could
expect a microstructure formation through v∗0 = 0 (i.e., u,x = ±1, depending on f(x)).
To allow v∗0 = 0 we will redefine the primal functional as indicated below.
Define G1 : U → R and F1 : U → R by

K
G1 (u) = G(Λ u) + F(u) + (u,x )2 dx
2 S
and
K
F1 (u) = (u,x )2 dx.
2 S

Also defining Ĝ(Λ u) = G(Λ u) + K2 S (u,x ) dx,
2 from Theorem 14.2.1, we can write
inf {J(u)} ≤ ∗inf ∗ sup {F1∗ (z∗ ) − Ĝ∗(v∗0 , v∗2 ) − F ∗ (v∗1 )} (14.13)
u∈U z ∈Y v∗ ∈B∗ (z∗ )
where
1
F1∗ (z∗ ) = (z∗ )2 dx,
2K S

1 (v∗2 )2 1
Ĝ∗ (v∗0 , v∗2 ) = ∗ dx + (v∗0 )2 dx,
2 S v0 + K 2 S

1
F ∗ (v∗1 ) = (v∗1 )2 dx + f , v∗1 L2 (S) − v∗2 (1)u(1)
2 S
and
B∗ (z∗ ) = {v∗ ∈ Y ∗ | − (v∗2 ),x + v∗1 − z∗ = 0 and v∗0 ≥ 0 a.e. in S}.
We developed an algorithm based on the dual formulation indicated in (14.13).

It is relevant to emphasize that such a dual formulation is convex if the supremum
indicated is evaluated under the constraint v∗0 ≥ 0 a.e. in S (this result follows from
the traditional Weierstrass condition, so that there is no duality gap between the
primal and dual formulations and the inequality indicated in (14.13) is in fact an
equality).
0.6
0.5
0.4
0.3
0.2
0.1
−0.1
0 0.2 0.4 0.6 0.8 1
Fig. 14.1 Vertical axis: u0 (x)-weak limit of minimizing sequences for f(x)=0
We present numerical results for f (x) = 0 (see Fig. 14.1), f (x) = 0.3 ∗ Sin
(π ∗ x) (Fig. 14.2), and f (x) = 0.3 ∗ Cos(π ∗ x) (Fig. 14.3). The solutions indicated
as optimal through the dual formulations (denoted by u0 ) are in fact weak cluster
points of minimizing sequences for the primal formulations.
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 0.2 0.4 0.6 0.8 1
Fig. 14.2 Vertical axis: u0 (x)-weak limit of minimizing sequences for f (x) = 0.3 ∗ Sin(π ∗ x)
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 0.2 0.4 0.6 0.8 1
Fig. 14.3 Vertical axis: u0 (x)-weak limit of minimizing sequences for f (x) = 0.3 ∗Cos(π ∗ x)
Chapter 15
The Full Complex Ginzburg–Landau System
15.1 Introduction
Remark 15.1.1. This chapter was published in an article form by Applied Mathe-
matics and Computation-Elsevier, reference [12].
We recall that about the year 1950 Ginzburg and Landau introduced a theory to
model the superconducting behavior of some types of materials below a critical
temperature Tc , which depends on the material in question. They postulated that the
free-energy density may be written close to Tc as

h̄ α (T ) β (T )
Fs (T ) = Fn (T ) + |∇ψ |22 dx + |ψ |4 dx − |ψ |2 dx,
4m Ω 4 Ω 2 Ω
where ψ is a complex parameter and Fn (T ) and Fs (T ) are the normal and super-
conducting free-energy densities, respectively. (see [4, 9, 45, 46] for details). Here
Ω ⊂ R3 denotes the superconducting sample with a boundary denoted by ∂ Ω = Γ .
The complex function ψ ∈ W 1,2 (Ω ; C) is intended to minimize Fs (T ) for a fixed
temperature T .
Denoting α (T ) and β (T ) simply by α and β , the corresponding Euler–Lagrange
equations are given by
⎧ h̄ 2
⎨ − 2m ∇ ψ + α |ψ |2 ψ − β ψ = 0, in Ω
(15.1)
⎩ ∂ψ
∂ n = 0, on ∂ Ω .
This last system of equations is well known as the Ginzburg–Landau (G-L) one.
In the physics literature, it is also well known the G-L energy in which a magnetic
potential here denoted by A is included. The functional in question is given by
374 15 The Full Complex Ginzburg–Landau System
2

1 h̄2 ∇ψ − 2ie Aψ dx
J(ψ , A) = | curl A − B0 |22 dx +
8π R3 4m Ω h̄c 2

α β
+ |ψ |4 dx − |ψ |2 dx (15.2)
4 Ω 2 Ω
Considering its minimization on the space U, where
U = W 1,2 (Ω ; C) × W 1,2 (R3 ; R3 ),
through the physics notation, the corresponding Euler–Lagrange equations are

⎧ 2
c A ψ + α |ψ | ψ − β ψ = 0, in Ω
⎨ 2m
1
−ih̄∇ − 2e 2
(15.3)
⎩
ih̄∇ψ + c Aψ · n = 0,
2e
on ∂ Ω ,
and
⎧
⎨ curl (curl A) = curl B0 + 4cπ J,˜ in Ω
(15.4)
⎩
curl (curl A) = curl B0 , in R3 − Ω ,
where
ieh̄ ∗ 2e 2
J˜ = − (ψ ∇ψ − ψ ∇ψ ∗ ) − |ψ |2 A.
2m mc
and
B0 ∈ L2 (R3 ; R3 )
is a known applied magnetic field.
15.2 Global Existence for the Ginzburg–Landau System
The existence of a global minimizer for the Ginzburg–Landau energy for a

system in superconductivity in the presence of a magnetic field is proven in the
next lines. The key hypothesis is the boundedness of infinity norm of the magnetic
potential. It is worth emphasizing that such a hypothesis is physically observed. We
start with the following remark:
Remark 15.2.1. For an open bounded subset Ω ⊂ R3 , we denote the L2 (Ω ) norm
by · L2 (Ω ) or simply by · 2. A similar remark is valid for the L2 (Ω ; R3 ) norm,
which is denoted by · L2 (Ω ;R3 ) or simply by · 2 , when its meaning is clear, and
for the L4 (Ω ) one, which is denoted by · L4 (Ω ) or simply by · 4. On the other
hand, by | · |2 , we denote the standard Euclidean norm in R3 or C3 , |Ω | denotes the
Lebesgue measure of Ω , and n is the outward normal to its boundary.
Moreover derivatives are always understood in the distributional sense. Finally,
by a regular boundary ∂ Ω = Γ of Ω , we mean regularity enough so that the standard
15.2 Global Existence for the Ginzburg–Landau System 375
Sobolev imbedding theorems, the trace theorem, and the Gauss–Green formulas of
integration by parts hold. Details about such results may be found in [1, 26].
Theorem 15.2.2. Let Ω ⊂ R3 be an open, bounded, and connected set with a

regular boundary denoted by ∂ Ω = Γ . Consider the functional J : U → R given by
2

1 h̄2 2ie
J(ψ , A) = | curl A − B0 |2 dx +
2
∇ψ − Aψ dx
8π R3 4m Ω h̄c 2

α β
+ |ψ | dx −
4
|ψ | dx
2
(15.5)
4 Ω 2 Ω
where h̄, m, c, e, α , β are positive constants, i is the imaginary unit, and
U = W 1,2 (Ω ; C) × W 1,2 (R3 ; R3 ).

Assume there exists a minimizing sequence {(ψn , An )} ⊂ U such that
An ∞ < K, ∀n ∈ N
for some finite K > 0. Under such a hypothesis, there exists (ψ0 , A0 ) ∈ U such that
J(ψ0 , A0 ) = min {J(ψ , A)}.

(ψ ,A)∈U
Proof. Suppose {(ψn , An )} ⊂ U is a minimizing sequence for J, that is,
lim J(ψn , An ) = inf {J(ψ , A)}, (15.6)

n→∞ (u,A)∈U
such that
An ∞ < K, ∀n ∈ N
for some finite K > 0.
Observe that

1
J(ψn , An ) ≥ | curl An − B0 |22 dx
8π R3

h̄2 h̄2 2ie
+ |∇ψn |2 dx − K
2
∇ψn 2 ψn 2
4m Ω 2m h̄c
2

+
h̄2 2ie An ψn dx + α |ψn |4 dx
4m Ω h̄c
2 4 Ω

β
− |ψn | dx, ∀n ∈ N.
2
(15.7)
2 Ω
Suppose, to obtain contradiction, that there exists a subsequence {ψnk } such that
either ψnk 4 → ∞ or ∇ψnk 2 → ∞, as k → ∞. In such a case from (15.7) we
would obtain J(ψnk , Ank ) → +∞, as k → ∞, which contradicts (15.6).
Therefore, there exists K1 > 0 such that
ψn 4 < K1 and ∇ψn 2 < K1 ,
∀n ∈ N. From this we may conclude that there exists K2 > 0 such that
ψn 2 < K2 , ∀n ∈ N. Hence by the Rellich–Kondrachov theorem, there exists
ψ0 ∈ W 1,2 (Ω ; C) and a subsequence not relabeled such that
∇ψn ∇ψ0 , weakly in L2 (Ω ; C3 )
and
ψn → ψ0 strongly in L2 (Ω ; C). (15.8)
On the other hand, since { An ∞ } is uniformly bounded, there exists A0 ∈
L∞ (Ω ; R3 ) such that up to a subsequence not relabeled we have
An A0 , weakly star in L∞ (Ω ; R3 ).
Fix v ∈ L2 (Ω , C3 ), since

|vψ0 | dx ≤ v 2 ψ0 2
Ω
we have that vψ0 ∈ L1 (Ω ; C3 ), so that

(An ψn − A0 ψ0 ) · v dx
Ω

= (An ψn − An ψ0 + An ψ0 − A0 ψ0 ) · v dx
Ω

≤ |(An ψn − An ψ0 ) · v| dx
Ω

+ (An ψ0 − A0 ψ0 ) · v dx
Ω

≤ An · v 2 ψn − ψ0 2 + (An − A0 ) · vψ0 dx

Ω

≤ K v 2 ψn − ψ0 2 + (An − A0 ) · vψ0 dx

Ω
→ 0 as n → ∞. (15.9)
Thus, since v ∈ L2 (Ω , C3 ) is arbitrary, we obtain

2ie 2ie
∇ψn − An ψn ∇ψ0 − A0 ψ0 , weakly in L2 (Ω ; C3 )
h̄c h̄c
so that
2
2

2ie 2ie
lim inf ∇ψn − An ψn dx ≥ ∇ψ0 − A0 ψ0 dx. (15.10)
n→∞ Ω h̄c Ω h̄c2 2
Also it is clear that

| curl An |22 dx < K3 , ∀n ∈ N,
R3
15.3 A Related Optimal Control Problem 377
for some K3 > 0, so that there exists v0 ∈ L2 (R3 ; R3 ) such that up to a subsequence
not relabeled, we have
curl An v0 , weakly in L2 (R3 ; R3 ).
Fix φ ∈ Cc∞ (R3 ; R3 ). Hence, we get
A0 , curl∗ φ L2 (R3 ;R3 ) = lim An , curl∗ φ L2 (R3 ;R3 )
n→∞
= lim curl An , φ L2 (R3 ;R3 )
n→∞
= v0 , φ L2 (R3 ;R3 ) . (15.11)
Since φ ∈ Cc∞ (R3 ; R3 ) is arbitrary we have that
v0 = curl A0 ,
in the distributional sense, so that
curl An curl A0 , weakly in L2 (R3 ; R3 ).
Therefore, considering the convexity of the functional in question, we obtain

1 α
lim inf | curl An − B0 |2 dx +
2
|ψn | dx
4
n→∞ 8π R3 4 Ω

1 α
≥ | curl A0 − B0 |22 dx + |ψ0 |4 dx. (15.12)
8π R3 4 Ω
From this, (15.10) and (15.8) we get
inf {J(ψ , A)} = lim inf{J(ψn , An )} ≥ J(ψ0 , A0 ).

(ψ ,A)∈U n→∞
15.3 A Related Optimal Control Problem
In this section we study the existence of solutions for a closely related optimal
control problem. In particular the state equation is of Ginzburg–Landau type. It is
worth mentioning that the present case refers to the simpler real one. In the next
lines we describe such a problem.
Let Ω ⊂ R3 be an open, bounded, and connected set with a regular boundary
denoted by ∂ Ω = Γ . Let ψd : Ω → R be a function such that ψd ∈ L2 (Ω ). Consider
the problem P, that is, the problem of minimizing J : U → R given by

1 1
J(ψ , u) = |ψ − ψd |2 dx + |u|2 dΓ
2 Ω 2 ∂Ω
subject to

−∇2 ψ + αψ 3 − β ψ = f , in Ω
∂ψ (15.13)
∂ n = u, on ∂ Ω .
Here U = W 1,2 (Ω ) × L2 (∂ Ω ) and f ∈ L2 (Ω ).
We say that the set of admissible fields for problem P is nonempty if there exists
(ψ , u) ∈ U satisfying (15.13).
A similar problem is studied in [40] through a different approach. We will prove
that such a problem has a solution. We start with the following proposition:
Proposition 15.3.1. The set of admissible fields for problem P is nonempty.
Proof. From reference [13], Chap. 13, there exists ψ̃ ∈ W01,2 (Ω ) which minimizes
J˜ on W01,2 (Ω ), where

˜ ψ) = 1 α β
J( |∇ψ |22 dx + |ψ |4 dx − |ψ |2 dx − ψ , f L2 (Ω ) ,
2 Ω 4 Ω 2 Ω
so that
−∇2 ψ̃ + α ψ̃ 3 − β ψ̃ = f , in Ω
(15.14)
ψ̃ = 0, on ∂ Ω .
Therefore (ψ , u) = (ψ̃ , ∂∂ψ̃n ) is an admissible field for problem P.

Theorem 15.3.2. Problem P has at least one solution.
Proof. Let {(ψn , un )} ⊂ U be a minimizing sequence for problem P. Clearly there
exists K > 0 such that
ψn 2 < K and un L2 (∂ Ω ) < K,
∀n ∈ N. Therefore, there exist ψ0 ∈ L2 (Ω ) and u0 ∈ L2 (∂ Ω ) such that up to a sub-

sequence not relabeled we have
ψn ψ0 , weakly in L2 (Ω ),
and
un u0 , weakly in L2 (∂ Ω ),
We claim that there exists K1 > 0 such that
ψn 4 < K1 , ∀n ∈ N.
Suppose, to obtain contradiction, that there exists a subsequence {ψnk } such that
ψnk 4 → ∞, as k → ∞. (15.15)
Observe that for each k ∈ N we have

−∇2 ψnk + αψn3k − β ψnk = f , in Ω

∂ ψnk (15.16)
∂n = u nk , on ∂ Ω ,
so that

|∇ψnk |22 dx + α |ψnk |4 dx − β |ψnk |2 dx
Ω Ω Ω
) *
∂ ψn k
−ψnk , f L2 (Ω ) − , ψn k = 0, (15.17)
∂n L2 (∂ Ω )
Hence

β |ψnk |2 dx ≥ |∇ψnk |22 dx + α |ψnk |4 dx
Ω Ω Ω
− ψnk 2 f 2 − unk L2 (∂ Ω ) ψnk L2 (∂ Ω )

≥ |∇ψnk |22 dx + α |ψnk |4 dx
Ω Ω
−K f L2 (Ω ) − K ψnk L2 (∂ Ω ) (15.18)
and thus, from the trace theorem, there exists C1 > 0 such that

β |ψnk |2 dx + ψnk 22 ≥ |∇ψnk |22 dx + α |ψnk |4 dx
Ω Ω Ω
+ ψnk 22 − K f 2
−KC1 ψnk W 1,2 (Ω ) . (15.19)

|ψnk |2 dx → ∞ as k → ∞,
Ω
which is a contradiction. Hence, there exists K1 > 0 such that

ψn 4 < K1 , ∀n ∈ N.
Thus, up to a subsequence not relabeled, there exists ψ̃ such that
ψn ψ̃ , weakly in L4 (Ω ),
so that from
ψn ψ0 , weakly in L2 (Ω ),
we get
ψ̃ = ψ0 .
From the last results and (15.17) we may obtain

|∇ψn |22 dx ≤ α K14 + β K 2 + K f 2 + K C1 ψn W 1,2 (Ω )
Ω
!
≤ α K14 + β K 2 + K f 2 + K C1 ψn 22 + ∇ψn 22
!
≤ α K14 + β K 2 + K f 2 + K C1 K 2 + ∇ψn 22 ,
so that there exists K2 > 0 such that
∇ψn 2 < K2 , ∀n ∈ N.
So, we may conclude that there exists K3 > 0 such that
ψn W 1,2 (Ω ) < K3 , ∀n ∈ N.
Therefore, from the Rellich–Kondrachov theorem, up to a subsequence not rela-

beled, we may infer that there exists ψ̂ ∈ W 1,2 (Ω ) such that
∇ψn ∇ψ̂ , weakly in L2 (Ω )
and
ψn → ψ̂ , strongly in L2 (Ω )
so that as
ψn ψ0 , weakly in L2 (Ω )
we can get
ψ̂ = ψ0 ,
that is,
∇ψn ∇ψ0 , weakly in L2 (Ω )
and
ψn → ψ0 , strongly in L2 (Ω ).
Choose φ ∈ Cc∞ (Ω ). Clearly we have
ψn , −∇2 φ L2 (Ω ) → ψ0 , −∇2 φ L2 (Ω ) , (15.20)
and
ψn , φ L2 (Ω ) → ψ0 , φ L2 (Ω ) , (15.21)
and in the next lines, we will prove that
ψn3 , φ L2 (Ω ) → ψ03 , φ L2 (Ω ) , (15.22)
as n → ∞. Observe that

(ψ 3 − ψ 3 )φ dx ≤ (ψ 3 − ψ 2ψ0 + ψ 2 ψ0 − ψ 3 )φ dx
Ω n 0 Ω n n n 0

≤ ψn2 (ψn − ψ0 )φ dx
Ω

+ ψ0 (ψ 2 − ψ 2)φ dx. (15.23)
n 0
Ω
Also observe that

ψn2 (ψn − ψ0)φ dx ≤ ψn 2 4 (ψn − ψ0)φ 2
L (Ω ) L (Ω )
Ω
≤ K12 ψn − ψ0 L2 (Ω ) φ ∞
→ 0, as n → ∞. (15.24)
On the other hand, from the generalized Hölder inequality, we get

ψ0 (ψn2 − ψ02 )φ dx = |ψ0 (ψn + ψ0 )(ψ − ψ0 )φ | dx
Ω Ω
≤ ψ0 4 ψn + ψ0 4 (ψn − ψ0 )φ 2
≤ ψ0 4 ( ψn 4 + ψ0 4 ) (ψn − ψ0 ) 2 φ ∞
≤ K1 (K1 + K1 ) φ ∞ ψn − ψ0 2
→ 0, as n → ∞. (15.25)
Summarizing the last results we get
ψn3 , φ L2 (Ω ) → ψ03 , φ L2 (Ω ) , (15.26)
as n → ∞. Therefore
ψ0 , −∇2 φ L2 (Ω ) + ψ03 − ψ0 − f , φ L2 (Ω )

= lim ψn , −∇2 φ L2 (Ω ) + ψn3 − ψn − f , φ L2 (Ω )
n→∞
= lim 0 = 0. (15.27)
n→∞
Since φ ∈ Cc∞ (Ω ) is arbitrary we get
−∇2 ψ0 + αψ03 − β ψ0 − f = 0, in Ω ,
in the distributional sense. From the weak convergence and
∂ ψn
= un , on ∂ Ω , ∀n ∈ N,
∂n
we may also obtain
∂ ψ0
= u0 , on ∂ Ω .
∂n
Finally, from the convexity of the functional in question,
lim inf J(ψn , un ) ≥ J(ψ0 , u0 ).

n→∞
Therefore (ψ0 , u0 ) is a solution of Problem P.

15.4 The Generalized Method of Lines
In this section we prepare a route to obtain numerical results. We reintroduce the

generalized method of lines, originally presented in Botelho [13]. In the present con-
text we add new theoretical and applied results to the original presentation. Specially
the computations are almost all completely new. Consider first the equation
∇2 u = 0, in Ω ⊂ R2 , (15.28)
with the boundary conditions
u = 0 on Γ0 and u = u f , on Γ1 .
From now on we assume that u f is a smooth function, unless otherwise specified.
Here Γ0 denotes the internal boundary of Ω and Γ1 the external one. Consider the
simpler case where
Γ1 = 2Γ0 ,
and suppose there exists r(θ ), a smooth function such that
Γ0 = {(θ , r(θ )) | 0 ≤ θ ≤ 2π },
being r(0) = r(2π ).
Also assume (0, 0) ∈ Ω and
min {r(θ )} O(1).
θ ∈[0,2π ]
We emphasize this is a crucial assumption for the application of the contraction

mapping theorem, which is the base of this method.
In polar coordinates the above equation may be written as
∂ 2u 1 ∂ u 1 ∂ 2u
+ + = 0, in Ω , (15.29)
∂ r2 r ∂ r r2 ∂ θ 2
and
u = 0 on Γ0 and u = u f , on Γ1 .
Define the variable t by
r
t= .
r(θ )
Also defining ū by
u(r, θ ) = ū(t, θ ),
dropping the bar in ū, (15.28) is equivalent to
∂ 2u 1 ∂u
+ f2 (θ )
∂ t2 t ∂t
1 ∂ 2u f4 (θ ) ∂ 2 u
+ f3 (θ ) + 2 = 0, (15.30)
t ∂θ∂t t ∂θ2
in Ω . Here f2 (θ ), f3 (θ ), and f4 (θ ) are known functions.
15.4 The Generalized Method of Lines 383
More specifically, denoting
−r (θ )
f1 (θ ) = ,
r(θ )
we have
f1 (θ )
f2 (θ ) = 1 + ,
1 + f1(θ )2
2 f1 (θ )
f3 (θ ) = ,
1 + f1 (θ )2
and
1
f4 (θ ) = .
1 + f1 (θ )2
Observe that t ∈ [1, 2] in Ω . Discretizing in t (N equal pieces which will generate
N lines) we obtain the equation
un+1 − 2un + un−1 (un − un−1) 1

+ f2 (θ )
d2 d tn
∂ (un − un−1) 1 ∂ 2 un f4 (θ )
+ f3 (θ ) + = 0, (15.31)
∂θ tn d ∂ θ 2 tn2
∀n ∈ {1, . . . , N − 1}. Here, un (θ ) corresponds to the solution on the line n. Thus we

may write
un = T (un−1 , un , un+1 ),
where

un+1 + un−1 d 2 (un − un−1) 1
T (un−1 , un , un+1 ) = + f2 (θ )
2 2 d tn

∂ (un − un−1) 1 ∂ 2 un f4 (θ )
+ f3 (θ ) + . (15.32)
∂θ tn d ∂ θ 2 tn2
Now we recall a classical definition.
Definition 15.4.1. 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 ≤ α < 1 such
that
T (x1 ) − T (x2 ) U ≤ α x1 − x2 U , ∀x1 , x2 ∈ C.
Remark 15.4.2. Observe that if T (x) U ≤ α < 1 on a convex set C, then T is a

contraction mapping, since by the mean value inequality,
T (x1 ) − T (x2 ) U ≤ sup{ T (x) } x1 − x2 U , ∀x1 , x2 ∈ C.

x∈C
The next result is the base of our generalized method of lines. For a proof
see [47].
Theorem 15.4.3 (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
x̃ ∈ C such that x̃ = T (x̃). Moreover, for an arbitrary x0 ∈ C defining the sequence
x1 = T (x0 ) and xk+1 = T (xk ), ∀k ∈ N

we have
xk → x̃, in norm, as k → +∞.
From (15.32), if d = 1/N is small enough and if un−1 ≈ un , it is clear that for a
fixed un+1 , G(un ) = T (un−1 , un , un+1 ) is a contraction mapping, considering that d
may be chosen so that G (un ) ≤ α < 1, for some 0 < α < 1 in a set that contains
the solution of the 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 ), and
4. 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 the 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→∞
We proceed in this fashion until obtaining

uN−1 = gN−1 (uN ) = gN−1 (u f ).
u f being 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 ).
Thus the problem is solved.
Remark 15.4.4. Here we consider some points concerning the convergence of the
method.
In the next lines the norm indicated as in xk refers to W 2,2 ([0, 2π ]). In particular
for n = 1 from the above we have
u1 = T (0, u1 , u2 ).
We will construct the sequence xk (in a little different way as above) by defining
x1 = u2 /2,
and
xk+1 = T (0, xk , u2 ) = u2 /2 + d T̃(xk ),
where the operator T̃ is properly defined from the expression of T . Observe that
xk+2 − xk+1 ≤ d T̃ xk+1 − xk ,
and if
0 ≤ α = d T̃ < 1,
we have that {xk } is (Cauchy) convergent. Through a standard procedure for this
kind of sequence, we may obtain
1
xk+1 − x1 ≤ x2 − x1 ,
1−α
so that denoting u1 = lim xk , we get
k→∞
1
u1 − u2/2 ≤ d T̃ u2 /2 ,
1−α
Having such an estimate, we may similarly obtain
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.
15.4.1 About the Approximation Error
Consider again the equation in finite differences for the example in question:
un+1 − 2un + un−1 (un − un−1) 1

+ f2 (θ )
d2 d t
∂ (un − un−1) 1 ∂ 2 un f4 (θ )
+ f3 (θ ) + = 0, (15.33)
∂θ td ∂ θ 2 t2
∀n ∈ {1, . . . , N − 1}. Here, un (θ ) corresponds to the solution on the line n. Thus, as
above, we may write
un = T (un−1 , un , un+1 ),
where

un+1 + un−1 d 2 (un − un−1) 1
T (un−1 , un , un+1 ) = + f2 (θ )
2 2 d t

∂ (un − un−1) 1 ∂ 2 un f4 (θ )
+ f3 (θ ) + . (15.34)
∂θ td ∂ θ 2 t2
For n = 1, we evaluate u1 = g1 (u2 ) through the contraction mapping theorem

obtaining
u1 (x) ≈ 0.5u2 (x) + 0.25du2(x) f2 (x)

+0.25d f3(x)u2 (x) + 0.25d 2 f4 (x)u2 (x). (15.35)
We can also obtain un (x) = g̃n (un+1 , un−1 ), that is,
un (x) ≈ 0.5un−1(x) + 0.5un+1(x) − 0.25d un−1 (x) f2 (x)/tn

+0.25d un+1(x) f2 (x)/tn − 0.25d f3 (x)un−1 (x)/tn
+0.25d f3 (x)un+1 (x)/tn + 0.25d 2 f4 (x)un−1 (x)/tn2
+0.25d 2 f4 (x)un+1 (x)/tn2 . (15.36)
The approximation error in (15.35) is of order O(d 3 ) plus the error concerning the
application of the contraction mapping theorem, which is well known and, if d is
small enough, may be made arbitrarily small in a reasonable number of iterations.
Also we may infer that the approximation error in (15.36) is also of order O(d 3 ).
The discretization error in this case is known to be of order O(d) (see [63] for
details).
15.4.2 The Solution of Laplace Equation for a Special Class

of Domains
As an example, we compute by the generalized method of lines the solution of

the equation
∇2 u = 0, in Ω ⊂ R2 , (15.37)
u = u0 on Γ0 and u = u f , on Γ1 .
We assume u0 and u f are smooth functions. As above Γ0 denotes the internal bound-
ary of Ω and Γ1 the external one. We consider the simpler case where
Γ1 = 2Γ0 .
Suppose there exists r(θ ), a smooth function such that
Γ0 = {(θ , r(θ )) | 0 ≤ θ ≤ 2π },
being r(0) = r(2π ).
Also assume (0, 0) ∈ Ω and
min {r(θ )} O(1).

θ ∈[0,2π ]
Denoting x = θ , particularly for N = 10, truncating the series up the terms in d 2 , we

obtain the following expression for the lines:
Line 1
u1 (x) = 0.1u f (x) + 0.9u0(x) − 0.034u0(x) f2 (x) + 0.034 f2(x)u f (x)
−0.034 f3(x)u0 (x) + 0.034 f3(x)uf (x)
+0.018 f4(x)u0 (x) + 0.008 f4(x)uf (x)
Line 2
−0.058 f3(x)u0 (x) + 0.058 f3(x)uf (x)
+0.029 f4(x)u0 (x) + 0.015 f4(x)uf (x)
Line 3
−0.075 f3(x)u0 (x) + 0.075 f3(x)uf (x)
+0.034 f4(x)u0 (x) + 0.020 f4(x)uf (x)
Line 4
−0.083 f3(x)u0 (x) + 0.083 f3(x)uf (x)
+0.035 f4(x)u0 (x) + 0.024 f4(x)uf (x)
Line 5
−0.085 f3(x)u0 (x) + 0.085 f3(x)uf (x)
+0.033 f4(x)u0 (x) + 0.026 f4(x)uf (x)
Line 6
−0.080 f3(x)u0 (x) + 0.080 f3(x)uf (x)
+0.028 f4(x)u0 (x) + 0.026 f4(x)uf (x)
Line 7
−0.068 f3(x)u0 (x) + 0.068 f3(x)uf (x)
+0.023 f4(x)u0 (x) + 0.023 f4(x)uf (x)
Line 8
−0.051 f3(x)u0 (x) + 0.051 f3(x)uf (x)
+0.015 f4(x)u0 (x) + 0.018 f4(x)uf (x)
Line 9
−0.028 f3(x)u0 (x) + 0.028 f3(x)uf (x)
+0.008 f4(x)u0 (x) + 0.010 f4(x)uf (x)
Remark 15.4.5. Here a word of caution is necessary. Consider for example the
equation
ε ∇2 u + G(u) = 0, in Ω ⊂ R2 , (15.38)
u = u0 on Γ0 and u = u f , on Γ1 .
We assume G, u0 , and u f are smooth functions.
15.5 A First Numerical Example 389
If ε is too small, for example, about 0.001 or 0.0001, the error just truncating
the series up the order d 2 is big. It seems that higher-order approximations or even
discretizing more does not solve the problem. However, for example, for G(u) = u,
by solving the equation with ε = 1, we can infer that the solution at each line has
the general format
un (x) ≈ a1 [n]u f (x) + a2 [n]u0 (x) + a3[n]u0 (x) f2 (x) + a4[n] f2 (x)u f (x)
a5 [n] f3 (x)u0 (x) + a6[n] f3 (x)uf (x)
+a7 [n] f4 (x)u0 (x) + a8 [n] f4 (x)uf (x)
+a9 [n] f5 (x)u0 (x) + a10[n] f5 (x)u f (x),
where f5 (x) = r2 (x) f4 (x).

This expression we get from the series that would represent the exact solution
obtained through an application of contraction mapping theorem for the concerned
inversions (which the first terms are qualitatively known up to the exact coefficient
values).
Thus, we just have to calculate the optimal real coefficients {ak [n]} which min-
imize the error concerning the original differential equation. Here derivatives must
be understood as matrices acting on vectors. A similar remark is valid as
max {r(θ )} % O(2).

θ ∈[0,2π ]
So, to summarize, we emphasize that through the problem solution with ε = 1 we

may discover its general format for smaller values of ε , up to constants which may
be easily evaluated (e.g., by the error minimization). Such a procedure has worked
very well in all examples we have so far developed. Of course, for this specific
example, other procedures are possible.
15.5 A First Numerical Example
Just to illustrate the possibilities of the generalized method of lines, we apply it

to the equation
∇2 u = ∇2 ū, in Ω ,
where
Ω = {(r, θ ) | 1 ≤ r ≤ 2, 0 ≤ θ ≤ 2π },
u = ū on Γ0 and Γ1 ,
where Γ0 and Γ1 are boundaries of the circles with centers at the origin and radius 1
and 2, respectively. Finally, in polar coordinates (here x stands for θ ),
ū = r2 cos(x).
See below the approximate values for the 9 lines (N = 10) obtained by the
generalized method of lines (un (x)) and the exact values ( ūn (x) for the same lines):
Line 1
u1 (x) = 1.21683 cos(x), ū1 (x) = 1.21 cos(x)
Line 2
u2 (x) = 1.44713 cos(x), ū2 (x) = 1.44 cos(x)
Line 3
u3 (x) = 1.69354 cos(x), ū3 (x) = 1.69 cos(x)
Line 4
u4 (x) = 1.95811 cos(x) ū4 (x) = 1.96 cos(x)
Line 5
u5 (x) = 2.24248 cos(x), ū5 (x) = 2.25 cos(x)
Line 6
u6 (x) = 2.54796 cos(x), ū6 (x) = 2.56 cos(x)
Line 7
u7 (x) = 2.87563 cos(x), ū7 (x) = 2.89 cos(x)
Line 8
u8 (x) = 3.22638 cos(x), ū8 (x) = 3.24 cos(x)
Line 9
u9 (x) = 3.60096 cos(x), ū9 (x) = 3.61 cos(x)
15.6 A Numerical Example Concerning the Optimal

Control Problem
We compute the solution of problem P, that is, the problem of minimizing J :

U → R, which similarly as above stated, is given by

1 1
J(ψ , u) = |ψ − ψd |2 dx + |u|2 dΓ
2 Ω 2 Γ1
15.6 A Numerical Example Concerning the Optimal Control Problem 391
subject to
⎧
⎨ −∇ ψ + αψ − β ψ = 0, in Ω
2 3
ψ = 0, on Γ0 , (15.39)
⎩ ∂ψ
∂ n = u, on Γ1 .
Also U = W 1,2 (Ω ) × L2 (Γ1 ). In this example we consider in polar coordinates
Ω = {(r, θ ) | 1 ≤ r ≤ 2, 0 ≤ θ ≤ 2π },
Γ0 = {(1, θ ) | 0 ≤ θ ≤ 2π },
Γ1 = {(2, θ ) | 0 ≤ θ ≤ 2π },
and
ψd (r, θ ) = (r − 1)2 sin θ .
We discretize the domain in lines (in fact curves). We divide the interval [1, 2] into
10 pieces (corresponding to the discretization in r) obtaining the following system
of equations:
ψn+1 − 2ψn + ψn−1 1 ψn − ψn−1 1 ∂ 2 ψn

+ + 2
d2 rn d rn ∂ θ 2
− αψn3 + β ψn = 0, ∀n ∈ {1, . . . , N − 1}, (15.40)
where N = 10 , d = 1/N, and rn = 1 + nd.

Thus ψn corresponds to the solution on the line n.
For α = β = 1 our procedure was first to compute ψ through the generalized
method of lines as a function of its value on the boundary, which we have denoted
by u f (x) (where as above x stands for θ ), obtaining the following approximate ex-
pressions for the lines (we have truncated the series up the terms in d 2 ):
Line 1
ψ1 (x) = 0.150254u f (x) − 0.004917u f (x)3 + 0.0078404uf (x)
Line 2
ψ2 (x) = 0.290418u f (x) − 0.009824u f (x)3 + 0.0148543uf (x)
Line 3
ψ3 (x) = 0.420248u f (x) − 0.014651u f (x)3 + 0.0204794uf (x)
Line 4
ψ4 (x) = 0.539385u f (x) − 0.019208u f (x)3 + 0.0243293uf (x)
Line 5
ψ5 (x) = 0.64738u f (x) − 0.023125u f (x)3 + 0.0261384uf (x)
Line 6
ψ6 (x) = 0.743709u f (x) − 0.025792u f (x)3 + 0.0257253uf (x)
Line 7
ψ7 (x) = 0.827787u f (x) − 0.026299u f (x)3 + 0.0229684uf (x)
Line 8
ψ8 (x) = 0.898983u f (x) − 0.023376u f (x)3 + 0.0177894uf (x)
Line 9
ψ9 (x) = 0.956623u f (x) − 0.015333u f (x)3 + 0.0101412uf (x)
The second step is to replace the field ψ obtained in J and then to compute
through a numerical minimization of J the optimal u f . For the candidate to optimal
u f (x) see Fig. 15.1. Finally, we have computed a critical point, but we cannot guar-
antee it is the global optimal solution.
0.8
0.6
0.4
0.2
0
−0.2
−0.4
−0.6
−0.8
0 20 40 60 80 100
Fig. 15.1 The optimal (candidate) u f (x)-units in x: 2π /100
15.7 Conclusion
In this chapter, we have presented global existence results concerning the

Ginzburg–Landau system in the presence of a magnetic field. In the second step
we prove the existence of solution for a closely related optimal control problem,
introducing the generalized method of lines (or briefly the GMOL) as an efficient
tool for computing its solution. It seems to be clear that the generalized method of
lines may used for solving a large class of nonlinear problems, specially when we
apply its matrix version. It is our objective in the future to develop applications of
GMOL to three-dimensional and time-dependent problems.
Chapter 16
More on Duality and Computation
for the Ginzburg–Landau System
16.1 Introduction
We recall again here (more details may be found in the introduction at Chap. 15)
that close to a critical temperature Tc , the Ginzburg–Landau energy would be
expressed by

h̄ α (T ) β (T )
Fs (T ) = Fn (T ) + |∇ψ |22 dx + |ψ |4 dx − |ψ |2 dx,
4m Ω 4 Ω 2 Ω
where ψ is a complex parameter and Fn (T ) and Fs (T ) are the normal and super-
conducting free-energy densities, respectively (see [4, 9, 45, 46] for details). Here
Ω ⊂ R3 denotes the superconducting sample with a boundary denoted by ∂ Ω = Γ .
The complex function ψ ∈ W 1,2 (Ω ; C) is intended to minimize Fs (T ) for a fixed
temperature T .
Denoting α (T ) and β (T ) simply by α and β , the corresponding Euler–Lagrange
equations are given by
⎧ h̄ 2
⎨ − 2m ∇ ψ + α |ψ |2 ψ − β ψ = 0, in Ω
(16.1)
⎩ ∂ψ
∂n = 0, on ∂ Ω .
This last system of equations is well known as the Ginzburg–Landau (G-L) one.
In the physics literature, it is also well known the G-L energy in which a magnetic
potential here denoted by A is included. The functional in question is given by
2

J(ψ , A) = | curl A − B0 |22 dx +
8π R3 4m Ω h̄c 2

α β
+ |ψ |4 dx − |ψ |2 dx (16.2)
4 Ω 2 Ω
394 16 More on Duality and Computation for the Ginzburg–Landau System
Considering its minimization on the space U, where
U = W 1,2 (Ω ; C) × W 1,2 (R3 ; R3 ),
through the physics notation, the corresponding Euler–Lagrange equations are

⎧ 2
c A ψ + α |ψ | ψ − β ψ = 0, in Ω
⎨ 2m
1
−ih̄∇ − 2e 2
(16.3)
⎩
ih̄∇ψ + 2ec Aψ · n = 0, on ∂Ω,
and ⎧
⎨ curl (curl A) = curl B0 + 4cπ J,˜ in Ω
(16.4)
⎩
curl (curl A) = curl B0 , in R3 \ Ω ,
where
ieh̄ ∗ 2e 2
J˜ = − (ψ ∇ψ − ψ ∇ψ ∗ ) − |ψ |2 A.
2m mc
and
B0 ∈ L2 (R3 ; R3 )
is a known applied magnetic field.
16.2 The Duality Principle
In this section we develop a duality principle for the Ginzburg–Landau system

in the presence of a magnetic field. Such a result includes sufficient conditions of
global optimality and is summarized by the next theorem.
boundary denoted by ∂ Ω = Γ . Consider the functional J : U → R given by
2

J(ψ , A) = | curl A − B0 |22 dx +
8π R3 4m Ω h̄c 2

α β
+ |ψ |4 dx − |ψ |2 dx (16.5)
4 Ω 2 Ω
where h̄, m, c, e, α , β are positive constants, i is the imaginary unit, and
U = W 1,2 (Ω ; C) × W01,2 (R3 ; R3 ).
Under such a hypothesis, we have
inf {J(ψ , A)} ≥ sup {−J ∗(v∗ )},

(ψ ,A)∈U v∗ ∈A∗
16.2 The Duality Principle 395
where v∗ = (v∗1 , v∗2 , σ ) and

J ∗ (v∗ ) = G∗0 (v∗2 ) + G∗1(v∗1 ) + G∗2 (σ ) + G∗3 (v∗1 , v∗2 , σ ). (16.6)
Furthermore,

h̄2
G1 (v1 ) = |v1 |22 dx,
4m Ω

α β
G2 (v3 ) = (v3 )2 dx − v3 dx
4 Ω 2 Ω
and
) *
2ie
G3 (ψ , A) = −ψ , div(v∗1 )L2 (Ω ;C) − Aψ , v∗1 . (16.7)
h̄c L2 (Ω ;C3 )
Also,

1
G∗0 (v∗2 ) = sup 2 ∗
Ak , v2k L2 (Ω ) − |curl A − B0 |2 dx ,
2
(ψ ,A)∈U 8π R3
G∗1 (v∗1 ) = sup {v1 , v∗1 L2 (Ω ;C3 ) − G1 (v1 )}

v1 ∈L2 (Ω ;C3 )
m ∗ ∗
= v1 , v1 L2 (Ω ;C3 ) . (16.8)
h̄2
Despite that we are dealing with complex and real variables, we highlight the func-
tionals in question are real, so that we denote, for u, v ∈ L2 (Ω ; C),

u, vL2 (Ω ;C) = u1 v1 dx + u2 v2 dx,
Ω Ω
where u1 , v1 are the real parts and u2 , v2 are the imaginary ones of u, v, respectively.
A similar remark is valid for L2 (Ω ; C3 ).
Moreover,
G∗2 (σ ) = sup {v3 , σ L2 − G2 (v3 )}
v3 ∈L2 (Ω )

1
= (σ + β /2)2 dx (16.9)
α Ω
and
G∗3 (v∗1 , v∗2 , σ ) = sup {−|ψ |2 , σ L2 (Ω ) − A2k , v∗2k L2 (Ω ) − G3 (ψ , A)}
(ψ ,A)∈U
.
= sup −|ψ |2 , σ L2 (Ω ) − A2k , v∗2k L2 (Ω )
(ψ ,A)∈U
) *
2ie
+ψ , div(v∗1 )L2 (Ω ;C) + Aψ , v∗1
h̄c L2 (Ω ;C3 )
so that if v∗ ∈ A∗ , we have
G∗3 (v∗1 , v∗2 , σ ) = −|ψ̃ |2 , σ L2 (Ω )
−Ãk , v∗2k L2 (Ω ) − G3 (ψ̃ , Ã),
2
(16.10)
where (ψ̃ , Ã) ∈ U is the only critical point of the quadratic functional indicated
in (16.10).
Here,
Y ∗ = W 1,2 (Ω ; C3 ) × L2 (Ω ; R3 ) × L2 (Ω ),
and defining
) *
2ie
G̃(ψ , A) = |ψ | 2
, σ L2 (Ω ) + A2k , v∗2k L2 (Ω ) − Aψ , v∗1 ,
h̄c L2 (Ω ;C3 )
we also define
A∗ = A1 ∩ A2 ,
A1 = {v∗ ∈ Y ∗ | G̃(ψ , A) > 0 ∀(ψ , A) ∈ U such that

(ψ , A) = (θ , θ ), and v∗1 · n = 0 on ∂ Ω }.
Furthermore,
A2 = {v∗ ∈ Y ∗ | J(A)
˜ > 0, ∀A ∈ W01,2 (R3 ; R3 ) such that A = θ },
and

1
˜
J(A) = |curl A|22 dx − A2k , v∗2k L2 (Ω ) .
8π R3
Finally, define Ind0 : U → R ∪ {+∞} by

0, if ih̄∇ψ + 2e
c Aψ · n = 0, on ∂ Ω ,
Ind0 (ψ , A) = (16.11)
+∞, otherwise.
Assume (ψ0 , A0 ) ∈ U is such that δ J(ψ0 , A0 ) = θ and Ind0 (ψ0 , A0 ) = 0 and also
such that
v∗0 = (v∗01 , v∗02 , σ0 ) ∈ A∗ ,
where such a point is the solution of the following relations:
∂ G1 (∇ψ0 − 2ie
h̄c A0 ψ0 )
v∗01 =
∂ v1

h̄2 2ie
= ∇ψ0 − A0 ψ0 , in Ω , (16.12)
2m h̄c
3
curl curl A0 − curl B0
− ∑ 2(v∗02 )k A0k ek + = θ in Ω , (16.13)
k=1 4π
where {e1 , e2 , e3 } denotes the canonical basis of R3 and
∂ G2 (|ψ0 |2 )
σ0 =
∂ v3
α β
= |ψ0 |2 − , in Ω . (16.14)
2 2
J(ψ0 , A0 ) = min J(ψ , A) = max

∗ ∗
{−J ∗ (v∗ )} = −J ∗ (v∗0 ).
(ψ ,A)∈U v ∈A
Proof. Observe that

1
J(ψ , A) = |curl A − B0 |22 dx
8π R3
2

h̄2 2ie
+ ∇ψ − Aψ dx
4m Ω h̄c 2

α β
+ |ψ |4 dx − |ψ |2 dx
4 Ω 2 Ω
) * ) *
2ie 2ie
+ ∇ψ − Aψ , v∗1 − ∇ψ − Aψ , v∗1
h̄c L2 (Ω ;C3 ) h̄c L2 (Ω ;C3 )
+A2k , v∗2k L2 (Ω ) − A2k , v∗2k L2 (Ω )

+|ψ |2 , σ L2 (Ω ) − |ψ |2 , σ L2 (Ω ) , (16.15)
∀(ψ , A) ∈ U, v∗ ∈ A∗ .
Hence,

1
J(ψ , A) = −A2k , v∗2k L2 (Ω ) + | curl A − B0 |22 dx
8π R3
) * 2

2ie h̄2 2ie
− ∇ψ − ∗
Aψ , v 1 + ∇ψ − Aψ dx
h̄c L2 (Ω ;C3 ) 4m Ω h̄c 2

α β
−|ψ |2 , σ L2 (Ω ) + |ψ |4 dx − |ψ |2 dx
4Ω 2 Ω
) *
2 ∗ 2ie ∗
+Ai , v2i L2 (Ω ) + |ψ | , σ L2 (Ω ) + ∇ψ −
2
Aψ , v 1 (16.16)
h̄c L2 (Ω ;C3 )
∀(ψ , A) ∈ U, v∗ ∈ A∗ , so that

1
J(ψ , A) ≥ inf −A2k , v∗2k L2 (Ω ) + | curl A − B0 |22 dx
A∈U 8π R3

∗ h̄2
+ inf −v1 , v1 L2 (Ω ;C3 ) + |v1 |2 dx
2
v1 ∈L2 (Ω ;C3 ) 4m Ω

α β
+ inf −v3 , σ L2 (Ω ) + (v3 ) dx −
2
v3 dx
v3 ∈L2 (Ω ) 4 Ω 2 Ω

) *
∗ 2ie ∗
+ inf −ψ , div(v1 )L2 (Ω ;C) − Aψ , v 1
(ψ ,A)∈U h̄c L2 (Ω ;C3 )
0
+|ψ |2 , σ L2 (Ω ) + A2k , v∗2k L2 (Ω ) ,
∀(ψ , A) ∈ U, v∗ ∈ A∗ .
Therefore,
J(ψ , A) ≥ −G∗0 (v∗2 ) − G∗1 (v∗1 ) − G∗2(σ ) − G∗3 (v∗1 , v∗2 , σ )

= −J ∗ (v∗ ), (16.17)
∀(ψ , A) ∈ U, v∗ ∈ A∗ .
Now observe that since
δ J(ψ0 , A0 ) = θ ,
from the variation in ψ , we get
2
h̄ 2ie h̄2 2ie −2ieA0
−div ∇ψ0 − A0 ψ0 + ∇ψ0 − A0 ψ0 ·
2m h̄c 2m h̄c h̄c
+α |ψ0 |2 ψ0 − β ψ0 = θ , in Ω .
Hence, from this, (16.12) and (16.14), we obtain

∗ ∗ 2ieA0
− div(v01 ) − v01 · + 2σ0ψ0 = θ , in Ω . (16.18)
h̄c
On the other hand, the variation in A gives us
curl curl A0 curl B0

−
4π 4π
2
h̄ 2ie −2ieψ0
+Re ∇ψ0 − A0 ψ0 = θ , in Ω , (16.19)
2m h̄c h̄c
where Re[v] denotes the real part of v and
curl curl A0 curl B0

− = θ , in R3 \ Ω . (16.20)
4π 4π
From (16.12), (16.13), and (16.19), we obtain

3
−2ieψ0
∑ 02 k 0 k k
2(v ∗
) (A ) e + Re v ∗
01 h̄c
= θ , in Ω . (16.21)
k=1
From (16.18) and (16.21) we get
G∗3 ((v∗0 )1 , (v∗0 )2 , σ0 ) = −|ψ0 |2 , σ L2 (Ω )

−(A0 )2k , (v∗0 )2k L2 (Ω ) − G3 (ψ0 , A0 ), (16.22)
) *
2ie
G∗1 ((v∗0 )1 ) = ∇ψ0 − A0 ψ0 , (v∗0 )1
h̄c L2 (Ω ;C3 )
2ie
−G1 (∇ψ0 − A0 ψ0 )
) h̄c *
2ie ∗
= ∇ψ0 − A 0 ψ0 , (v0 )1
h̄c L2 (Ω ;C3 )
2
h̄2 2ie
− ∇ψ0 − A0 ψ0 dx. (16.23)
4m Ω h̄c 2
By (16.14) we may infer that
G∗2 (σ0 ) = |ψ0 |2 , σ0 L2 − G2 (|ψ0 |2 ). (16.24)
From (16.13) and (16.20) we get
G∗0 ((v∗0 )2 ) = (A0 )2k , (v∗0 )2k L2 (Ω )

1
− |curl A0 − B0 |22 dx. (16.25)
8π R3
Finally, by (16.22), (16.23), (16.24), (16.25), and from the fact that v∗0 ∈ A∗ , we
obtain
G∗0 ((v∗0 )2 ) + G∗1((v∗0 )1 ) + G∗2(σ0 ) + G∗3 ((v∗0 )1 , (v∗0 )2 , σ0 )

= (A0 )2k , (v∗0 )2k L2 (Ω )

1
− |curl A0 − B0 |22 dx
8π R3
) *
2ie 2ie
+ ∇ψ0 − A0 ψ0 , (v∗0 )1 − G1 ∇ψ0 − A0 ψ0
h̄c L2 (Ω ;C3 ) h̄c
+|ψ0 |2 , σ0 L2 − G2(|ψ0 |2 )
) *
2ie ∗
− ∇ψ0 − A0 ψ0 , (v0 )1
h̄c L2 (Ω ;C3 )
−|ψ0 |2 , σ0 L2 (Ω ) − (A0 )2k , (v∗0 )2k L2 (Ω )

1 2ie
=− |curl A0 − B0 |22 dx − G1 ∇ψ0 − A0 ψ0 − G2 (|ψ0 |2 )
8π R3 h̄c
= −J(ψ0 , A0 ),
that is,
J(ψ0 , A0 ) = −J ∗ (v∗0 ).
16.3 On the Numerical Procedures for Ginzburg–Landau-Type

Equations
We first apply Newton’s method. The solution here is obtained similarly as for
the generalized method of lines procedure. See the next sections for details on such
a method for PDEs.
Consider again the equation.
⎧
⎨ u + f (u) + g = 0, in [0, 1]
(16.26)
⎩
u(0) = u0 , u(1) = u f ,
As above, in finite differences, we have
un+1 − 2un + un−1 + f (un )d 2 + gnd 2 = 0.
Assume such an equation is nonlinear. Linearizing it about a first solution {ũ}, we

have (in fact this is an approximation)
un+1 − 2un + un−1 + f (ũn )d 2 + f (ũn )(un − ũn )d 2 + gnd 2 = 0.
Thus we may write
un+1 − 2un + un−1 + Anun d 2 + Bn d 2 = 0,
where
An = f (ũn ),
and
Bn = f (ũn ) − f (ũn )ũn + gn .
In particular for n = 1 we get
u2 − 2u1 + u0 + A1 u1 d 2 + B1 d 2 = 0.
Solving such an equation for u1 , we get
u 1 = a 1 u 2 + b 1 u 0 + c1 ,
where
a1 = (2 − A1d 2 )−1 , b1 = a1 , c1 = a1 B1 .
16.3 On the Numerical Procedures for Ginzburg–Landau-Type Equations 401
Reasoning inductively, having
un−1 = an−1un + bn−1u0 + cn−1,
and
un+1 − 2un + un−1 + Anun d 2 + Bn d 2 = 0,
we get
un+1 − 2un + an−1un + bn−1u0 + cn−1 + An un d 2 + Bn d 2 = 0,
so that
un = an un+1 + bnu0 + cn ,
where
an = (2 − an−1 − An d 2 )−1 ,
bn = an bn−1 ,
and
cn = an (cn−1 + Bnd 2 ),
∀n ∈ 1, . . . , N − 1.
We have thus obtained
un = an un+1 + bn u0 + cn ≡ Hn (un+1 ), ∀n ∈ {1, . . . , N − 1},
and in particular
uN−1 = HN−1 (u f ),
so that we may calculate
uN−2 = HN−2 (uN−1 ),
uN−3 = HN−3 (uN−2 ),

u1 = H1 (u2 ).
The next step is to replace {ũn} by the {un } calculated and repeat the process
up to the satisfaction of an appropriate convergence criterion. We present numerical
results for the equation
⎧ 3
⎨ u − uε + uε + g = 0, in [0, 1]
(16.27)
⎩
u(0) = 0, u(1) = 0,
where
1
g(x) = ,
ε
The results are obtained for ε = 1.0, ε = 0.1, ε = 0.01, and ε = 0.001. Please
see Figs. 16.1, 16.2, 16.3, and 16.4, respectively. For the other two solutions for
ε = 0.01 see Figs. 16.5 and 16.6.
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0 0.2 0.4 0.6 0.8 1
Fig. 16.1 The solution u(x) by Newton’s method for ε = 1
Other solutions through Newton’s method are also shown.

Remark 16.3.1. We highlight that the results obtained through Newton’s method are
consistent with problem physics.
1.4
1.2
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
Fig. 16.2 The solution u(x) by Newton’s method for ε = 0.1

16.4 Numerical Results for Related PDEs 403
1.4
1.2
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
16.4 Numerical Results for Related PDEs
16.4.1 A Related PDE on a Special Class of Domains
We start by describing a similar equation, but now in a two-dimensional context.

Let Ω ⊂ R2 be an open, bounded, connected set with a regular boundary denoted by
∂ Ω . Consider a real Ginzburg–Landau-type equation (see [4, 9, 45, 46] for details
about such an equation), given by
⎧ 2
⎨ ε ∇ u − α u3 + β u = f , in Ω
(16.28)
⎩
u = 0, on ∂ Ω ,
1.4
1.2
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1

−1
−2
−3
−4
0 0.2 0.4 0.6 0.8 1
Fig. 16.5 Other solution u(x) by Newton’s method for ε = 0.01
where α , β , ε > 0, u ∈ U = W01,2 (Ω ), and f ∈ L2 (Ω ). The corresponding primal

variational formulation is represented by J : U → R, where

ε α β
J(u) = ∇u · ∇u dx + u4 dx − u2 dx + f u dx.
2 Ω 4 Ω 2 Ω Ω
18
16
14
12
10
8
6
4
2
0
0 0.2 0.4 0.6 0.8 1
Fig. 16.6 Other solution u(x) by Newton’s method for ε = 0.01
16.4.2 About the Matrix Version of GMOL
The generalized method of lines was originally developed in [13]. In this work
we address its matrix version. Consider the simpler case where Ω = [0, 1] × [0, 1].
We discretize the domain in x, that is, in N + 1 vertical lines obtaining the following
equation in finite differences (see [63] for details about finite differences schemes).
ε (un+1 − 2un + un−1)

+ ε M2 un /d12 − α u3n + β un = fn , (16.29)
d2
∀n ∈ {1, . . . , N − 1}, where d = 1/N and un corresponds to the solution on the line
n. The idea is to apply Newton’s method. Thus choosing an initial solution {(u0 )n }
we linearize (16.29) about it, obtaining the linear equation
3α d 2
un+1 − 2un + un−1 + M̃2 un − (u0 )2n un
ε
2α β d2 d2
+ (u0 )3n d 2 + un − fn = 0, (16.30)
ε ε ε
2
where M̃2 = M2 dd 2 and
1
⎡ ⎤
−2 1 0 0 ... 0
⎢ 1 −2 1 0 ... 0 ⎥
⎢ ⎥
⎢ 0 1 −2 1 0.. 0 ⎥
M2 = ⎢
⎢... ... ... ...
⎥, (16.31)
⎢ ... ... ⎥
⎥
⎣ 0 0 ... 1 −2 1 ⎦
0 0 ... ... 1 −2
with N1 lines corresponding to the discretization in the y axis. Furthermore

d1 = 1/N1 .
In particular for n = 1 we get
3α d 2
u2 − 2u1 + M̃2 u1 − (u0 )21 u1
ε
2α β d2 d2
+ (u0 )31 d 2 + u1 − f1 = 0. (16.32)
ε ε ε
Denoting
α d2 β d2
M12 [1] = 2Id − M̃2 + 3 (u0 )21 Id − Id ,
ε ε
where Id denotes the (N1 − 1) × (N1 − 1) identity matrix,
2α d 2 d2
Y0 [1] = (u0 )31 − f1 ,
ε ε
and M50 [1] = M12 [1]−1 , we obtain
u1 = M50 [1]u2 + z[1].
where
z[1] = M50 [1] ·Y0 [1].
Now for n = 2 we get
3α d 2
u3 − 2u2 + u1 + M̃2 u2 − (u0 )22 u2
ε
2α β d2 d2
+ (u0 )32 d 2 + u2 − f2 = 0, (16.33)
ε ε ε
that is,
3α d 2
u3 − 2u2 + M50 [1]u2 + z[1] + M̃2 u2 − (u0 )22 u2
ε
2α β d2 d2
+ (u0 )32 d 2 + u2 − f2 = 0, (16.34)
ε ε ε
so that denoting
α d2 β d2
M12 [2] = 2Id − M̃2 − M50 [1] + 3 (u0 )22 Id − Id ,
ε ε
2α d 2 d2
Y0 [2] = (u0 )32 − f2 ,
ε ε
and M50 [2] = M12 [2]−1 , we obtain
u2 = M50 [2]u3 + z[2],
where
z[2] = M50 [2] · (Y0 [2] + z[1]).
Proceeding in this fashion, for the line n, we obtain
3α d 2
un+1 − 2un + M50 [n − 1]un + z[n − 1] + M̃2un − (u0 )2n un
ε
2α β d2 d2
+ (u0 )3n d 2 + un − fn = 0, (16.35)
ε ε ε
so that denoting
α d2 β d2
M12 [n] = 2Id − M̃2 − M50[n − 1] + 3 (u0 )2n Id − Id ,
ε ε
and also denoting
2α d 2 d2
Y0 [n] = (u0 )3n − fn ,
ε ε
and M50 [n] = M12 [n]−1 , we obtain
un = M50 [n]un+1 + z[n],
where
z[n] = M50 [n] · (Y0[n] + z[n − 1]).
Observe that we have
uN = θ ,
where θ denotes the zero matrix (N1 − 1) × 1, so that we may calculate
uN−1 = M50 [N − 1] · uN + z[N − 1],
and
uN−2 = M50 [N − 2] · uN−1 + z[N − 2],
and so on, up to obtaining
u1 = M50 [1] · u2 + z[1].
The next step is to replace {(u0 )n } by {un } and thus to repeat the process until
convergence is achieved.
This is Newton’s method; what seems to be relevant is the way we inverted the
big matrix ((N1 −1)·(N −1))×((N1 −1)·(N −1)), and in fact instead of inverting it
directly we have inverted N − 1 matrices (N1 − 1) × (N1 − 1) through an application
of the generalized method of lines.
So far we cannot guarantee convergence; however, through the next theorem,
we describe a procedure that always leads to a solution. Anyway, we highlight the
next result is not a formal proof of convergence in a numerical analysis context. In
fact, such a result must be seen as an existence of one of the critical points for the
equation in question.

lar boundary denoted by ∂ Ω . Consider the real Ginzburg–Landau-type equation,
given by
⎧ 2
⎨ ε ∇ u − α u3 + β u = f , in Ω
(16.36)
⎩
u = 0, on ∂ Ω ,
where α , β , ε > 0, u ∈ U = W01,2 (Ω ), and f ∈ L2 (Ω ).

Consider the sequence obtained through the algorithm:
1. Set n = 1.
2. Choose z∗1 ∈ L2 (Ω ).
3. Compute un by

ε α
un = argminu∈U ∇u · ∇u dx + u4 dx
2 Ω 4 Ω

1
−u, z∗n L2 + (z∗n )2 dx + f u dx , (16.37)
2β Ω Ω
which means to solve the equation

⎧ 2
⎨ ε ∇ u − α u3 + z∗n = f , in Ω
(16.38)
⎩
u = 0, on ∂ Ω .
4. Compute z∗n+1 by

ε α
z∗n+1 = argminz∗ ∈L2 (Ω ) ∇un · ∇un dx + u4 dx
2 Ω 4 Ω n

∗ 1 ∗ 2
−un , z L2 + (z ) dx + f un dx , (16.39)
2β Ω Ω
that is,
z∗n+1 = β un .
5. Set n → n + 1 and go to step 3 (up to the satisfaction of an appropriate conver-
gence criterion).
The sequence {un } is such that up to a subsequence not relabeled
un → u0 , strongly in L2 (Ω ),
where
u0 ∈ W01,2 (Ω )
is a solution of equation (16.36).
Proof. Observe that defining J : U → R, by

ε α
J(u) = ∇u · ∇u dx + u4 dx
2 Ω 4 Ω

β
− u2 dx + f u dx , (16.40)
2 Ω Ω
we have

ε α β
J(u) = ∇u · ∇u dx + u dx −
4
u2 dx
2 Ω 4 Ω

2 Ω
+u, z∗ L2 − u, z∗ L2 + f u dx

Ω

ε α
≤ ∇u · ∇u dx + u4 dx
Ω 2 4 Ω

β
−u, z∗ L2 + sup u, z∗ L2 − u2 dx + f u dx
u∈U 2 Ω Ω

ε α
≤ ∇u · ∇u dx + u4 dx
2 Ω 4 Ω

1
−u, z∗ L2 + (z∗ )2 dx + f u dx. (16.41)
2β Ω Ω
Denoting

ε α
αn = ∇un · ∇un dx + u4n dx
2 Ω 4 Ω

1
−un , z∗n L2 + (z∗n )2 dx + f un dx, (16.42)
2β Ω Ω
we may easily verify that {αn } is a real nonincreasing sequence bounded below by
infu∈U {J(u)}; therefore there exists α ∈ R such that
lim αn = α . (16.43)
n→∞
From the Poincaré inequality (see [1] for details) we have that J(u) → +∞ if
u W 1,2 (Ω ) → ∞. From this, (16.41), (16.42), and (16.43), we may infer that
un W 1,2 (Ω ) < C, ∀n ∈ N
for some C > 0.

Thus, from the Rellich–Kondrachov theorem, up to a not relabeled subsequence,
there exists u0 ∈ W 1,2 (Ω ) such that
∇un ∇u0 weakly in L2 (Ω ),
un → u0 strongly in L2 (Ω ),
so that considering the algorithm in question
zn → z∗0 strongly in L2 (Ω ),
where
z∗0 = β u0 .
Observe that the unique un ∈ U such that

ε α
un = argminu∈U ∇u · ∇u dx + u4 dx
2 Ω 4 Ω

1
−u, z∗n L2 + (z∗n )2 dx + f u dx , (16.44)
2β Ω Ω
is also such that

ε ∇2 un − α u3n + z∗n + f = 0 in Ω ,
in the sense of distributions (details about this result may be found in [25]).
Fix φ ∈ Cc∞ (Ω ). In the next lines, we will prove that
u3n , φ L2 (Ω ) → u30 , φ L2 (Ω ) , (16.45)

as n → ∞. First observe that from (16.41), (16.42), and (16.43), it is clear that there
exists K1 > 0 such that
un 4 < K1 , ∀n ∈ N.
Observe also that

(u3n − u3 )φ dx ≤ (u3n − u2n u0 + u2n u0 − u3)φ dx
Ω 0 Ω 0

≤ u2n (un − u0)φ dx
Ω

+ u0 (u2n − u20)φ dx. (16.46)
Ω
Furthermore,

u2 (un − u0 )φ dx ≤ un 2 (un − u0 )φ 2
n 4
Ω
≤ K12 un − u0 2 φ ∞
→ 0, as n → ∞. (16.47)
On the other hand, from the generalized Hölder inequality, we get

u0 (u2n − u20 )φ dx = |u0 (un + u0 )(un − u0)φ | dx
Ω Ω
≤ u0 4 un + u0 4 (un − u0)φ 2
≤ u0 4 ( un 4 + u0 4 ) (un − u0) 2 φ ∞
≤ K1 (K1 + K1 ) φ ∞ un − u0 2
→ 0, as n → ∞. (16.48)
Summarizing the last results we get
u3n , φ L2 (Ω ) → u30 , φ L2 (Ω ) , (16.49)
as n → ∞. So, we may write
0 = lim {un , ε ∇2 φ L2 + −α u3n + z∗n − f , φ L2 }
n→∞
= u0 , ε ∇2 φ L2 + −α u30 + z∗0 − f , φ L2
= u0 , ε ∇2 φ L2 + −α u30 + β u0 − f , φ L2
(16.50)
that is,
ε ∇2 u0 − α u30 + β u0 = f in Ω ,
in the sense of distributions. From un = 0 on ∂ Ω , ∀n ∈ N we also obtain in a weak
sense
u0 = 0, on ∂ Ω .
16.5 Numerical Results 411
Remark 16.4.2. Observe that for each n, the procedure of evaluating un stands for
the solution of a convex optimization problem with unique solution, given by the
one of equation
ε ∇2 un − α u3n + z∗n + f = 0 in Ω ,
which may be easily obtained, due to convexity, through the generalized method of
lines (matrix version) associated with Newton’s method as above described.
16.5 Numerical Results
We solve the equation

⎧ 2
⎨ ε ∇ u − α u3 + β u + 1 = 0, in Ω = [0, 1] × [0, 1]
(16.51)
⎩
u = 0, on ∂ Ω ,
through the algorithm specified in the last theorem. We consider α = β = 1. For
ε = 1.0 see Fig. 16.7, and for ε = 0.0001 see Fig. 16.8.
0.08
0.06
0.04
0.02
0
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 16.7 The solution u(x, y) for ε = 1.0
1.6
1.4
1.2
0.8
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 16.8 The solution u(x, y) for ε = 0.0001

16.6 A New Procedure to Obtain Approximate PDE Solutions
In this section we describe a procedure to obtain an approximate solution of a

class of PDE. We start with the following theorem.
Theorem 16.6.1. Consider the equation given by

⎧ 2
⎨ ∇ u + G(u) = f , in Ω
(16.52)
⎩
u = 0, on ∂ Ω ,
where G : R → R is a smooth function with bounded first derivatives in bounded

sets, u ∈ U = W01,2 (Ω ), and f ∈ L2 (Ω ). Consider the simpler case where Ω =
[0, 1] × [0, 1]. We discretize the domain in x, that is, in N + 1 vertical lines obtaining
the following equation in finite differences:
(un+1 − 2un + un−1)

+ M2 un /d12 + G(un ) = fn , (16.53)
d2
∀n ∈ {1, . . . , N − 1}, where d = 1/N and un corresponds to the solution on the line
n. We rewrite equation (16.53), obtaining
un+1 − 2un + un−1 + M̃2 un + G(un )d 2 − fn d 2 = 0, (16.54)
2
where M̃2 = M2 dd 2 and
1
⎡ ⎤
−2 1 0 0 ... 0
⎢ 1 −2 1 0 . . . 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 1 −2 1 0.. 0 ⎥
M2 = ⎢
⎢...
⎥, (16.55)
⎢ ... ... ... ... ... ⎥
⎥
⎢ ⎥
⎣ 0 0 . . . 1 −2 1 ⎦
0 0 . . . . . . 1 −2
with N1 lines corresponding to the discretization in the y axis. Furthermore,

d1 = 1/N1 . Then, for such a system, we have the following relations:
un = M50 [n]un+1 + M60[n]G(un+1 )d 2 + z[n] + Er[n],

16.6 A New Procedure to Obtain Approximate PDE Solutions 413
where
M12 [1] = 2Id − M̃2 ,
Id denotes the (N1 − 1) × (N1 − 1) identity matrix, M50 [1] = M12 [1]−1 ,
M60 [1] = M50 [1],
z[1] = M50 [1] · (− f1 d 2 ),

Er[1] = M50 [1](G(u1 ) − G(u2))d 2 ,
M12 [n] = 2Id − M̃2 − M50[n − 1],
M50 [n] = M12 [n]−1 ,
M60 [n] = M50 [n] · (M60 [n − 1] + Id),
z[n] = M50 [n] · (z[n − 1] − fnd 2 ),

and
Er[n] = M50 [n](Er[n − 1]) + M60[n](G(un ) − G(un+1))d 2 ,
∀n ∈ {1, . . . , N − 1}.
Proof. In particular for n = 1 we get
u2 − 2u1 + M̃2 u1 + G(u1)d 2 − f1 d 2 = 0. (16.56)
Denoting
M12 [1] = 2Id − M̃2 ,
where Id denotes the (N1 − 1) × (N1 − 1) identity matrix and M50 [1] = M12 [1]−1 , we
obtain
u1 = M50 [1](u2 + G(u1 )d 2 − f1 d 2 ),
so that
u1 = M50 [1](u2 ) + M60 [1]G(u2 )d 2 + z[1] + Er[1],
where
M60 [1] = M50 [1],
z[1] = M50 [1] · (− f1 d 2 ),
and
Er[1] = M50 [1](G(u1 ) − G(u2))d 2 .
Now for n = 2 we get
u3 − 2u2 + u1 + M̃2 u2 + G(u2 )d 2 − f2 d 2 = 0, (16.57)
that is,
u3 − 2u2 + M50 [1]u2 + M60[1]G(u2 )d 2 + z[1]
+Er[1] + M̃2u2 + G(u2 )d 2 − f2 d 2 = 0, (16.58)
so that denoting
M12 [2] = 2Id − M̃2 − M50 [1],
and also denoting M50 [2] = M12 [2]−1 , we obtain
u2 = M50 [2]u3 + M60[2]G(u3 )d 2 + z[2] + Er[2],
where
M60 [2] = M50 [2] · (M60[1] + Id ),
z[2] = M50 [2] · (z[1] − f2d 2 ).
and
Er[2] = M50 [2](Er[1]) + M60[2](G(u2 ) − G(u3 ))d 2 .
Proceeding in this fashion, for the line n, we obtain
un+1 − 2un + M50[n − 1]un + M60 [n − 1]G(un)d 2

+z[n − 1] + Er[n − 1] + M̃2un + G(un )d 2 − fn d 2 = 0, (16.59)
so that denoting
M12 [n] = 2Id − M̃2 − M50[n − 1],
and M50 [n] = M12 [n]−1 , we obtain
un = M50 [n]un+1 + M60[n]G(un+1 )d 2 + z[n] + Er[n],

where
M60 [n] = M50 [n] · (M60 [n − 1] + Id),
z[n] = M50 [n] · (z[n − 1] − fnd 2 ),
and
Er[n] = M50 [n](Er[n − 1]) + M60[n](G(un ) − G(un+1))d 2 .
Remark 16.6.2. We may use, as a first approximation for the solution, the relations
un ≈ M50 [n]un+1 + M60[n]G(un+1 )d 2 + z[n].
Observe that we have
uN = θ ,
where θ denotes the zero matrix (N1 − 1) × 1, so that we may calculate
uN−1 ≈ M50 [N − 1] · uN + M60 [N − 1] · G(uN )d 2 + z[N − 1],
and
uN−2 ≈ M50 [N − 2] · uN−1 + M60 [N − 2] · G(uN−1)d 2 + z[N − 2],
and so on, up to obtaining
u1 ≈ M50 [1] · u2 + M60[1] · G(u2 )d 2 + z[1].

16.6 A New Procedure to Obtain Approximate PDE Solutions 415
The next step is to use the {un } obtained as the initial solution for Newton’s
method.
What is relevant is that in general, the first approximation is a good one for the
exact solution.
We have computed the first approximation using such a method, for
−u3 u
G(u) = + ,
ε ε
1
f (x, y) = − , ∀(x, y) ∈ Ω
ε
and ε = 0.01. Please see Fig. 16.9.
This first approximation is close to the solution obtained through Newton’s
method. For the solution through the earlier approach, see Fig. 16.10.
1.5
0.5
0
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 16.9 The first approximation for u(x, y) for ε = 0.01
1.5
0.5
0
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 16.10 The solution u(x, y) by Newton’s method for ε = 0.01

16.7 Final Results, Newton’s Method for a First-Order System
Consider the first-order system and respective boundary conditions

⎧
⎪
⎪ u + f1 (u, v) + g1 = 0, in [0, 1]
⎪
⎪
⎨
v + f2 (u, v) + g2 = 0, in [0, 1] (16.60)
⎪
⎪
⎪
⎪
⎩
u(0) = u0 , v(1) = v f ,
Linearizing the equations about the first solutions ũ, and ṽ, we obtain
∂ f1 (ũ, ṽ)
u + f1 (ũ, ṽ) + (u − ũ)
∂u
∂ f1 (ũ, ṽ)
+ (v − ṽ) + g1 = 0, (16.61)
∂v
∂ f2 (ũ, ṽ)
v + f2 (ũ, ṽ) + (u − ũ)
∂u
∂ f2 (ũ, ṽ)
+ (v − ṽ) + g2 = 0. (16.62)
∂v
In finite differences, we could write
∂ f1 (ũn−1 , ṽn−1 )
un − un−1 + f1 (ũn−1 , ṽn−1 )d + (un−1 − ũn−1)d
∂u
∂ f1 (ũn−1 , ṽn−1 )
+ (vn−1 − ṽn−1)d + (g1 )n−1 d = 0, (16.63)
∂v
∂ f2 (ũn−1 , ṽn−1 )
vn − vn−1 + f2 (ũn−1 , ṽn−1 )d + (un−1 − ũn−1)d
∂u
∂ f2 (ũn−1 , ṽn−1 )
+ (vn−1 − ṽn−1)d + (g2)n−1 d = 0. (16.64)
∂v
Hence, we may write
un = an un−1 + bnvn−1 + cn ,
vn = dn un−1 + en vn−1 + fn ,
where
∂ f1 (ũn−1 , ṽn−1 )
an = − d + 1,
∂u
∂ f1 (ũn−1 , ṽn−1 )
bn = − d,
∂v
16.7 Final Results, Newton’s Method for a First-Order System 417
∂ f1 (ũn−1 , ṽn−1 )
cn = − f1 (ũn−1 , ṽn−1 )d + ũn−1 d
∂u
∂ f1 (ũn−1 , ṽn−1 )
+ ṽn−1 d − (g1 )n−1 d, (16.65)
∂v
and
∂ f2 (ũn−1 , ṽn−1 )
dn = − d,
∂u
∂ f2 (ũn−1 , ṽn−1 )
en = − d + 1,
∂v
∂ f2 (ũn−1 , ṽn−1 )
fn = − f2 (ũn−1 , ṽn−1 )d + ũn−1d
∂u
∂ f2 (ũn−1 , ṽn−1 )
+ ṽn−1 d − (g2)n−1 d. (16.66)
∂v
In particular, for n = 1, we get
u 1 = a 1 u 0 + b 1 v0 + c1 , (16.67)
and
v1 = d 1 u 0 + e1 v0 + f 1 . (16.68)
From this last equation,
v0 = (v1 − d1 u0 − f1 )/e1 ,
so that from this and Eq. (16.67), we get
u1 = a1 u0 + b1 (v1 − d1 u0 − f1 )/e1 + c1 = F1 v1 + G1,
where
F1 = b1 /e1 , G1 = a1 u0 − b1 (d1 u0 + f1 )/e1 + c1.
un−1 = Fn−1 vn−1 + Gn−1,

we also have
un = an un−1 + bnvn−1 + cn ,
vn = dn un−1 + en vn−1 + fn ,
vn = dn (Fn−1 vn−1 + Gn−1 ) + envn−1 + fn ,

that is,
vn−1 = Hn vn + Ln ,
where
Hn = 1/(dn Fn−1 + en ),
Ln = −Hn (dn Gn−1 + fn ).
Hence
un = an (Fn−1 vn−1 + Gn−1) + bnvn−1 + cn−1,
so that
un = an (Fn−1 (Hn vn + Ln ) + Gn−1) + bn(Hn vn + Ln ) + cn−1,
and hence
Fn = an Fn−1 Hn + bn Hn ,
and
Gn = an (Fn−1 Ln + Gn−1) + bnLn + cn−1.
Thus,
un = Fn vn + Gn ,
so that, in particular,
uN = FN v f + GN ,
vN−1 = HN v f + LN ,
and hence
uN−1 = FN−1 vN−1 + GN−1 ,
vN−2 = HN−1 vN−1 + LN−1 ,
and so on, up to finding,
u1 = F1 v1 + G1 ,
and
v0 = H0 v1 + L0 ,
where H0 = 1/e1 and L0 = −(d1 u0 + f1 )/e1 .
The next step is to replace {ũn} and {ṽn } by {un } and {vn }, respectively, and then
to repeat the process up to the satisfaction of an appropriate convergence criterion.
16.7.1 An Example in Nuclear Physics
As an application of the method above exposed we develop numerical results for

the system of equations relating the neutron kinetics of a nuclear reactor. Following
[61], the system in question is given by
⎧
⎪ (ρ (T )−β )
⎨ n (t) = L n(t) + λ C(t)
β
⎪ C (t) = L n(t) − λ C(t) (16.69)
⎩ T (t) = Hn(t),
where n(t) is the neutron population, C(t) is the concentration of delayed neutrons,
T (t) is the core temperature, ρ (T ) is the reactivity (which depends on the tempera-
ture T ), β is the delayed neutron fraction, L is the prompt reactors generation time,
λ is the average decay constant of the precursors, and H is the inverse of the reactor
thermal capacity.
For our numerical examples we consider T (0s) = 300 K and T (100 s) = T f =
350 K. Moreover we assume the relation
1 (β − ρ (0))
C(0) = n(0),
λ L
where n(0) is unknown (to be numerically calculated by our method such that we
have T (100 s) = T f ).
Also we consider
ρ (T ) = ρ (0) − α (T − T (0)).
The remaining values are β = 0.0065, L = 0.0001 s, λ = 0.00741 s−1 , H =
0.05 K/(MWs), α = 5 · 10−5 K−1 , and ρ (0) = 0.2β .
First we linearize the system in question about (ñ, T̃ ) obtaining (in fact it is a first
approximation)
ρ (T̃ ) − β ρ (T ) − β
n (t) = n(t) + ñ(t)
L L
ρ (T̃ ) − β
− ñ(t) + λ C(t), (16.70)
L
β
C (t) = n(t) − λ C(t),
L
T (t) = Hn(t),
where ρ (T ) = ρ (0) − α (T − T (0)).
Discretizing such a system in finite differences, we get
ρ (T̃i ) − β ρ (Ti ) − β
(ni+1 − ni )/d = ni + ñi
L L
ρ (T̃i ) − β
− ñi + λ Ci , (16.71)
L
β
(Ci+1 − Ci )/d = ni − λ Ci ,
L
(Ti+1 − Ti )/d = Hni ,
where d = 100 s/N, where N is the number of nodes.
Hence, we may write
ni+1 = ai ni + bi Ti + diCi + ei , (16.72)
Ci+1 = f ni + gCi , (16.73)

Ti+1 = hTi + mni, (16.74)
where
ρ (T̃i ) − β
ai = 1 + d,
L
−α
bi = ñi d,
L
di = λ d,
(ρ (0) + α T (0) − β ) (ρ (T̃i ) − β )
ei = ñi d − ñi d,
L L
β
f = d,
L
g = 1 − λ d,
h = 1,
m = Hd.
Observe that
C0 = α̃ n0 ,
where
β − ρ (0)
α̃ = .
Lλ
For i = 0 from (16.74) we obtain
T1 − hT0
n0 = = α1 T1 + β1, (16.75)
m
where α1 = 1/m and β1 = −(h/m)T0 .
Therefore,
C0 = α̃ n0 = α̃ (α1 T1 + β1).
Still for i = 0, replacing this last relation and (16.75) into (16.72), we get
n1 = a0 (α1 T1 + β1 ) + b0T0 + d0α̃ (α1 T1 + β1) + e0 ,
so that
n1 = α̃1 T1 + β̃1, (16.76)
where
α̃1 = a0 α0 + d0α̃α1 ,
and
β̃1 = a0 β1 + b0 T0 + d0α̃β1 + e0 .
Finally, from (16.73),
C1 = f (α1 T1 + β1) + gα̃ (α1 T1 + β1)
= α̂1 T1 + β̂1 , (16.77)
where
α̂1 = f α1 + gα̃α1 ,
and
β̂1 = f β1 + gα̃β1 .
ni = α̃i Ti + β̃i , (16.78)
ni−1 = αi Ti + βi , (16.79)
Ci = α̂i Ti + β̂i , (16.80)
we are going to obtain the corresponding relations for i + 1, i ≥ 1. From (16.74)
and (16.78) we obtain
Ti+1 = hTi + m(α̃i Ti + β̃i),
so that
Ti = ηi Ti+1 + ξi , (16.81)
where
ηi = (h + mα̃i )−1 ,
and
ξi = −(mβ̃i )ηi .
On the other hand, from (16.72), (16.78), and (16.80), we have
ni+1 = ai (α̃i Ti + B̃i ) + biTi + di (α̂i Ti + β̂i ) + ei ,
so that from this and (16.81), we obtain
ni+1 = α̃i Ti+1 + β̃i+1 ,

where
α̃i+1 = ai α̃i ηi + bi ηi + diα̂i ηi ,
and
β̃i+1 = ai (α̃i ξi + β̃i ) + biξi + di (α̂i ξi + β̂i ) + ei .
Also from (16.78) and (16.81) we have
ni = α̃i (ηi Ti+1 + ξi ) + β̃i = αi+1 Ti+1 + βi+1,
where
αi+1 = α̃i ηi ,
and
βi+1 = α̃i ξi + β̃i .
Moreover,
Ci+1 = f ni + gCi
= f (αi+1 Ti+1 + βi+1)
+g(α̂i Ti + β̂i )
= f (αi+1 Ti+1 + βi+1)
+g(α̂i (ηi Ti + ξi ) + β̂i )
= α̂i+1 Ti+1 + β̂i+1, (16.82)
where
α̂i+1 = f αi+1 + gα̂i ηi ,
and
β̂i = f βi+1 + gα̂i ξi + gβ̂i.
Summarizing, we have obtained linear functions (F0 )i , (F1 )i , and (F2 )i such that
Ti = (F0 )i (Ti+1 ),
ni = (F1 )i (Ti+1 ),
Ci = (F2 )i (Ti+1 ),
∀i ∈ {1, . . . , N − 1}.
Thus, considering the known value TN = T f , we obtain
TN−1 = (F0 )N−1 (T f ),
nN−1 = (F1 )N−1 (T f ),

CN−1 = (F2 )N−1 (T f ),
and having TN−1 , we get
TN−2 = (F0 )N−2 (TN−1 ),
nN−2 = (F1 )N−2 (TN−1 ),

CN−2 = (F2 )N−1 (TN−1 ),
T1 = (F0 )1 (T2 ),
n1 = (F1 )1 (T2 ),
C1 = (F2 )1 (T2 ),
and n0 = (F0 )1 (T1 ).
16.8 Conclusion 423
The next step is to replace (ñ, T̃ ) by the last calculated (n, T ) and then to repeat
the process until an appropriate convergence criterion is satisfied.
Concerning our numerical results through such a method, for the solution n(t)
obtained, please see Fig. 16.11. For the solution T (t), see Fig. 16.12.
12
11
10
6
0 20 40 60 80 100
Fig. 16.11 Solution n(t) for 0 s ≤ t ≤ 100 s
We emphasize the numerical results here obtained are consistent with the current
literature (see [61] for details).
16.8 Conclusion
In this chapter, first we have presented a duality principle for the Ginzburg–
Landau system in the presence of a magnetic field. We highlight to have obtained
sufficient conditions of optimality, similarly to the canonical duality procedure int-
roduced by Gao [36]. It is worth mentioning the dual formulation is concave and
amenable to numerical computation.
In a second step, we have introduced the matrix version of the generalized
method of lines. Also we develop a convergent algorithm suitable for equations
that present strong variational formulation and, in particular, suitable for Ginzburg–
Landau-type equations. The results are rigorously proven and numerical examples
are provided. We emphasize that even as the parameter ε is very small, namely,
ε = 0.0001, the results are consistent and the convergence is very fast. Finally, in the
350
345
340
335
330
325
320
315
310
305
300
0 20 40 60 80 100
Fig. 16.12 Solution T (t) for 0 s ≤ t ≤ 100 s
last section, we develop in details Newton’s method combined with the generalized
method of lines main idea, with numerical results relating an example in nuclear
physics.
Chapter 17
On Duality Principles for Scalar and Vectorial
Multi-well Variational Problems
17.1 Introduction
Remark 17.1.1. This chapter was published in an article form by Nonlinear

Analysis-Elsevier, reference [14]
In this chapter, our first objective is the establishment of a duality principle suitable
for the variational form of some nonlinear vectorial problems in physics and
engineering. The results are based on standard tools of convex analysis. As a first
example we apply them to a phase transition model, which may be found in a similar
format in Chenchiah and Bhattacharya [18]. It is relevant to observe that the study
developed in [18] is restricted to the two-well problem, whereas our new duality
principle is applicable to vectorial multi-well formulations in general, not restricted
to two or three wells.
In Sect. 17.4 we discuss how the standard tools of convex analysis can be used
to study the scalar case.
In Sect. 17.3 we present the main theorem in this chapter, namely Theorem 17.3.1,
which corresponds, as mentioned above, to a new duality principle. It is important
to emphasize that this principle stands for relaxation for a vectorial phase transition
problem. In the next lines, we describe such a result.
Consider (G ◦ Λ ) : U → R and (F ◦ Λ ) : U → R, F being a convex Gâteaux
differentiable functional such that J : U → R defined as
J(u) = (G ◦ Λ )(u) − (F ◦ Λ )(u) − u, f U
is bounded below. Here Λ : U → Y is a continuous linear injective operator whose
the respective adjoint is denoted by Λ ∗ : Y ∗ → U ∗ . Under such assumptions, we
have
inf {G∗∗ (Λ u) − F(Λ u) − u, f U }
u∈U

∗ ∗ ∗ ∗ ∗ ∗
≥ sup ∗inf ∗ {(F ◦ Λ ) (Λ z ) − G (v + z )} .
v∗ ∈A∗ z ∈Y
426 17 On Duality Principles for Scalar and Vectorial Multi-well Variational Problems
where
A∗ = {v∗ ∈ Y ∗ | Λ ∗ v∗ = f }.
Furthermore, under additional assumptions to be specified, we have
inf {G∗∗ (Λ u) − F(Λ u) − u, f U }
u∈U

= sup ∗inf ∗ {(F ◦ Λ )∗ (Λ ∗ z∗ ) − G∗(v∗ + z∗ )} .
v∗ ∈A∗ z ∈Y
Remark 17.1.2. Henceforth by a regular boundary we mean a condition sufficient

for the standard Green–Gauss theorems of integration by parts holds. Moreover, all
derivatives in this text are understood in distributional sense.
Now we also present a summary of our main applied result, namely, a duality
principle applied to a vectorial phase transition problem.
For an open bounded connected set S ⊂ R3 with a regular boundary denoted by
∂ S, consider the functional J : U → R, where

1
J(u) = min {gk (ε (u)) + βk } dx − u, f U ,
2 S k∈{1,...,N}
gk (ε (u)) = (εi j (u) − ekij )Cikjlm (εlm (u) − eklm ).

The operator ε : U → Y = Y ∗ = L2 (S; R9 ) is defined by
1
εi j (u) = (ui, j + u j,i ), for i, j ∈ {1, 2, 3}.
2
Furthermore {Cikjlm } are positive definite matrices and βk ∈ R for each k ∈
{1, . . . , N}, and f ∈ L2 (S; R3 ) is an external load. Here ek ∈ R3×3 for k ∈ {1, . . . , N}
represent the stress-free configurations or phases presented by a solid with field of
displacements u = (u1 , u2 , u3 ) ∈ W 1,2 (S; R3 ) (due to f ). Also
U = {u ∈ W 1,2 (S; R3 ) | u = (0, 0, 0) ≡ θ on ∂ S} = W01,2 (S; R3 ).
J(u) = G(ε (u)) − F(ε (u)) − u, f U
where

1 K
G(ε (u)) = min {gk (ε (u)) + βk } dx + (εi j (u))Hi jlm (εlm (u)) dx,
2 S k∈{1,...,N} 2 S

K
F(ε (u)) = (εi j (u))Hi jlm (εlm (u)) dx,
2 S
17.2 Preliminaries 427
{Hi jlm } is a positive definite matrix (the identity for example) and K > 0 is an
appropriate constant. The final duality principle is expressed by
inf {G∗∗ (ε (u)) − F(ε (u)) − u, f U }

u∈U

1
≥ sup inf z∗i j (v∗ ,t)Ĥi jlm z∗lm (v∗ ,t) dx
∗
v ∈A ∗ t∈B 2K S
N
1
∑ − S 2 tk (v∗i j + z∗i j (v∗ ,t))Dkijlm (v∗lm + z∗lm (v∗ ,t)) dx
k=1

∗ ∗ ∗
+ (−(vi j + zi j (v ,t))tkĈi jlm elm + tk βk ) dx
k k
, (17.1)
S
where z∗ (v∗ ,t) is obtained through Eq. (17.44), that is,

N N
1
Ĥi jlm z∗lm − ∑ {tk Dkijlm (v∗lm + z∗lm )} − ∑ tkĈikjlm eklm = 0, in S.
K k=1 k=1
Finally,
A∗ = {v∗ ∈ Y ∗ | v∗i j, j + fi = 0 in S},
and

B= (t1 , . . . ,tN ) measurable |

N
tk (x) ∈ [0, 1], ∀k ∈ {1, . . . , N}, ∑ tk (x) = 1, a.e. in S .
k=1
17.2 Preliminaries
We denote by U and Y Banach spaces which the topological dual spaces are
identified with U ∗ and Y ∗ , respectively. Unless otherwise indicated, Y is assumed to
be reflexive. The canonical duality pairing between U and U ∗ is denoted by ·, ·U :
U × U ∗ → R, through which the linear continuous functionals defined on U are
represented.
Given F : U → R̄ = R ∪ {+∞} its polar F ∗ : U ∗ → R̄ is defined as
F ∗ (u∗ ) = sup{u, u∗ U − F(u)}.

u∈U
Recall that the sub-differential ∂ F(u) is the subset of U ∗ given by

∂ F(u) = {u∗ ∈ U ∗ , such that v − u, u∗U + F(u) ≤ F(v), ∀v ∈ U}.
Also relevant is the next definition.
Definition 17.2.1 (Adjoint Operator). Let U and Y be Banach spaces and

Λ : U → Y a continuous linear operator. The adjoint operator related to Λ , den-
oted by Λ ∗ : Y ∗ → U ∗ , is defined through the equation:
u, Λ ∗ v∗ U = Λ u, v∗ Y , ∀u ∈ U, v∗ ∈ Y ∗ .
Finally, the following duality principle found in [25] will be used in this text.
Theorem 17.2.2. Let G : Y → R̄ = R ∪ {+∞} and F : U → R be two convex l.s.c.
(lower semicontinuous) functionals so that J : U → R̄ defined as
J(u) = (G ◦ Λ )(u) − F(u)
is bounded below, where Λ : U → Y is a continuous linear operator which the res-

pective adjoint is denoted by Λ ∗ : Y ∗ → U ∗ . Thus if there exists û ∈ U such that
F(û) < +∞, G(Λ û) < +∞ being G continuous at Λ û, we have
inf {J(u)} = sup {−G∗ (v∗ ) − F ∗ (−Λ ∗ v∗ )}
u∈U v∗ ∈Y ∗
and there exists at least one v∗0 ∈ Y ∗ which maximizes the dual formulation. If in
addition U is reflexive and
lim J(u) = +∞
u →+∞
then both primal and dual formulations have global extremals so that there exist
u0 ∈ U and v∗0 ∈ Y ∗ such that

∗ ∗
{−G∗ (v∗ ) − F ∗ (−Λ ∗ v∗ )} = −G∗ (v∗0 ) − F ∗ (−Λ ∗ v∗0 ).
u∈U v ∈Y
Also
G(Λ u0 ) + G∗(v∗0 ) = Λ u0 , v∗0 Y ,
F(u0 ) + F ∗ (−Λ ∗ v∗0 ) = u0 , −Λ ∗ v∗0 U ,
so that
G(Λ u0 ) + F(u0 ) = −G∗ (v∗0 ) − F ∗ (−Λ ∗ v∗0 ).
Also fundamental for the construction of the main duality principle is a result found
in Toland [67] (despite we have not used it directly we have followed a similar idea)
which is as follows.
Theorem 17.2.3. Consider the functionals F, G : U → R through which we define
J : U → R as
J(u) = G(u) − F(u). (17.2)
Suppose there exists u0 ∈ U such that
J(u0 ) = inf {J(u)} (17.3)
u∈U
and ∂ F(u0 ) = 0.
/
Under such assumptions we can write
inf {J(u)} = ∗inf ∗ {F ∗ (u∗ ) − G∗(u∗ )} (17.4)

u∈U u ∈U
and for u∗0 ∈ ∂ F(u0 ) we have
F ∗ (u∗0 ) − G∗ (u∗0 ) = ∗inf ∗ {F ∗ (u∗ ) − G∗(u∗ )}. (17.5)

u ∈U
Now we present the main theoretical result in this chapter.

Theorem 17.3.1. Consider the functionals (G ◦ Λ ) : U → R and (F ◦ Λ ) : U → R
being F convex and Gâteaux differentiable and also such that J : U → R defined as
J(u) = (G ◦ Λ )(u) − (F ◦ Λ )(u) − u, f U
is bounded below. Here Λ : U → Y is a continuous linear injective operator whose

respective adjoint is denoted by Λ ∗ : Y ∗ → U ∗ .
inf {G∗∗ (Λ u) − F(Λ u) − u, f U } ≥

u∈U

sup ∗inf ∗ {(F ◦ Λ )∗ (Λ ∗ z∗ ) − G∗(v∗ + z∗ )} .
v∗ ∈A∗ z ∈Y
where
A∗ = {v∗ ∈ Y ∗ | Λ ∗ v∗ = f }.
Furthermore, assuming that G∗ : Y ∗ → R is Lipschitz continuous, there exists
v∗0 ∈ Y ∗ such that
Jˆ∗ (v∗0 ) = max
∗ ∗
{Jˆ∗ (v∗ )},
v ∈Y
where
J ∗ (v∗ ) = ∗inf ∗ {(F ◦ Λ )∗ (Λ ∗ z∗ ) − G∗ (v∗ + z∗ )} ,
z ∈Y

0, if Λ ∗ v∗ + f = 0,
Ind(v∗ ) =
+∞, otherwise,
and
Jˆ∗ (v∗ ) = J ∗ (v∗ ) − Ind(v∗).
In addition we suppose that defining
J1∗ (v∗ , z∗ ) = (F ◦ Λ )∗ (Λ ∗ z∗ ) − G∗ (v∗ + z∗ ),

we have that
J1∗ (v∗0 , z∗ ) → +∞
as
z∗ Y ∗ → +∞, or Λ ∗ z∗ L2 → +∞.
Furthermore, suppose that if {z∗n } ⊂ Y ∗ is such that Λ ∗ z∗n L2 < K, ∀n ∈ N for some
K > 0, then there exists z̃∗ ∈ Y ∗ such that for a not relabeled subsequence, we have
Λ ∗ z∗n Λ ∗ z̃∗ , weakly in L2 ,
and
z∗n → z̃∗ , strongly in Y ∗ .
Under such additional assumptions, there exist z∗0 ∈ Y ∗ and u0 ∈ U such that
inf {G∗∗ (Λ u) − F(Λ u) − u, f U }

u∈U
= G∗∗ (Λ u0 ) − F(Λ u0 ) − u0 , f U
= (F ◦ Λ )∗ (Λ ∗ z∗0 ) − G∗(v∗0 + z∗0 )

∗ ∗ ∗ ∗ ∗ ∗
= sup ∗inf ∗ {(F ◦ Λ ) (Λ z ) − G (v + z )} .
v∗ ∈A∗ z ∈Y
Proof. Observe that

G∗ (v∗ + z∗ ) ≥ Λ u, v∗ Y + Λ u, z∗ Y − G∗∗ (Λ u),
∀u ∈ U, v∗ ∈ Y ∗ , z∗ ∈ Y ∗ , that is,
−(F ◦ Λ )∗ (Λ ∗ z∗ ) + G∗ (v∗ + z∗ )
≥ u, f U − (F ◦ Λ )∗ (Λ ∗ z∗ ) + Λ u, z∗ Y
−G∗∗ (Λ u), (17.6)
∀u ∈ U, v∗ ∈ A∗ , z∗ ∈ Y ∗ , and hence,
sup {−(F ◦ Λ )∗ (Λ ∗ z∗ ) + G∗ (v∗ + z∗ )}
z∗ ∈Y ∗
≥ u, f U + F(Λ u) − G∗∗ (Λ u), (17.7)
∀u ∈ U, v∗ ∈ A∗ and thus
inf {G∗∗ (Λ u) − F(Λ u) − u, f U }
u∈U

∗ ∗ ∗ ∗ ∗ ∗
≥ sup ∗inf ∗ {(F ◦ Λ ) (Λ z ) − G (v + z )} .
v∗ ∈A∗ z ∈Y
inf {G∗∗ (Λ u) − F(Λ u) − u, f U } ≥ sup {Jˆ∗ (v∗ )}. (17.8)

u∈U v∗ ∈Y ∗
Moreover the functional Jˆ∗ (v∗ ) is concave, as the infimum in z∗ of a family of

concave functionals in v∗ . On the other hand it is clear that
Jˆ∗ (v∗ ) ≤ −G∗ (v∗ ) − Ind(v∗ ) ∀v∗ ∈ Y ∗ ,
so that
Jˆ∗ (v∗ ) → −∞, as v∗ Y ∗ → +∞ or Λ ∗ v∗ L2 → +∞.
Therefore if {v∗n } is a maximizing sequence of Jˆ∗ (v∗ ) we have that there exists a
constant K0 > 0 such that
v∗n Y ∗ < K0 , and Λ ∗ v∗ L2 < K0 , ∀n ∈ N.
Since Y ∗ and L2 are reflexive Banach spaces, there exist v∗0 ∈ Y ∗ and ṽ∗0 ∈ L2 , such
that up to a not relabeled subsequence we have
v∗n v∗0 , weakly in Y ∗ ,
and
Λ ∗ v∗n ṽ∗0 , weakly in L2 .
Observe that given ϕ ∈ Cc∞ we have
v∗0 , Λ ϕ L2 = lim v∗n , Λ ϕ L2

n→∞
= lim Λ ∗ v∗n , ϕ L2
n→∞
= ṽ∗0 , ϕ L2 . (17.9)
Thus
ṽ∗0 = Λ ∗ v∗0 ,
in distributional sense, so that
Λ ∗ v∗n Λ ∗ v∗0 , weakly in L2 .
Hence, as Jˆ∗ (v∗ ) is concave and strongly continuous, it is also upper semicontinu-
ous, so that
lim sup{Jˆ∗ (v∗n )} ≤ Jˆ∗ (v∗0 ).
n→∞
Therefore, since {v∗n } is a maximizing sequence, we may conclude that

Jˆ∗ (v∗0 ) = max
∗ ∗
{Jˆ∗ (v∗ )}.
v ∈Y
Consider now the infimum

inf {J1 (v∗0 , z∗ )},
z∗ ∈Y ∗
where
J1∗ (v∗0 , z∗ ) = (F ◦ Λ )∗ (Λ ∗ z∗ ) − G∗ (v∗0 + z∗ ).
From the coercivity hypothesis, if {z∗n } is a minimizing sequence, there exists K1 > 0
such that
z∗n Y ∗ < K1 and Λ ∗ z∗n L2 < K1 , ∀n ∈ N.
Also from the hypothesis, up to a not relabeled subsequence, there exist z∗0 ∈ Y ∗
such that
Λ z∗n Λ ∗ z∗0 , weakly in L2 ,
and
z∗n → z∗0 , strongly in Y ∗ .
As G∗ (v∗ ) is strongly continuous, we obtain
G(v∗0 + z∗n ) → G∗ (v∗0 + z∗0 ).
On the other hand, as (F ◦ Λ )∗ is convex and strongly continuous, it is also weakly

lower semicontinuous, so that
lim inf{(F ◦ Λ )∗ (Λ ∗ z∗n )} ≥ (F ◦ Λ )∗ (Λ ∗ z∗0 ).

n→∞
Therefore
lim inf{(F ◦ Λ )∗ (Λ ∗ z∗n ) − G∗ (v∗0 + z∗n )} ≥ (F ◦ Λ )∗ (Λ ∗ z∗0 ) − G∗ (v∗0 + z∗0 ).

n→∞
{z∗n } being a minimizing sequence, we obtain
inf {J1∗ (v∗0 , z∗ )} = (F ◦ Λ )∗ (Λ ∗ z∗0 ) − G∗ (v∗0 + z∗0 ) = Jˆ∗ (v∗0 ).

z∗ ∈Y ∗
To complete the proof observe that the extremal equation is satisfied

∂ (F ◦ Λ )∗ (Λ ∗ z∗0 )
Λ ∈ ∂ G∗ (v∗0 + z∗0 ), (17.10)
∂ ẑ∗
where
ẑ∗ = Λ ∗ z∗ .
Defining
∂ (F ◦ Λ )∗ (Λ ∗ z∗0 )
u0 = ,
∂ ẑ∗
we obtain
Λ u0 ∈ ∂ G∗ (v∗0 + z∗0 ). (17.11)
From these two last equations we obtain respectively
(F ◦ Λ )∗ (Λ ∗ z∗0 ) = u0 , Λ ∗ v∗0 U − F(Λ u0 ),

17.4 The Scalar Multi-well Problem 433
and
G∗ (v∗0 + z∗0 ) = Λ u0 , v∗0 Y + Λ u0 , z∗0 Y − G∗∗(Λ u0 ). (17.12)
From the fact that v∗0 ∈ A∗ , we have
Λ ∗ v∗0 = f .
Note that from the last three equations, we obtain
(F ◦ Λ )∗ (Λ ∗ z∗0 ) − G∗ (v∗0 + z∗0 )
= G∗∗ (Λ u0 ) − F(Λ u0 ) − u0, f U . (17.13)
Therefore, we may conclude that
inf {G∗∗ (Λ u) − F(Λ u) − u, f U }
u∈U
= G∗∗ (Λ u0 ) − F(Λ u0 ) − u0 , f U
= (F ◦ Λ )∗ (Λ ∗ z∗0 ) − G∗(v∗0 + z∗0 )

∗ ∗ ∗ ∗ ∗ ∗
= sup ∗inf ∗ {(F ◦ Λ ) (Λ z ) − G (v + z )} .
v̂∗ ∈A∗ z ∈Y
17.4 The Scalar Multi-well Problem
This section is dedicated to the analysis of the scalar multi-well problem via
duality.
17.4.1 The Primal Variational Formulation
Consider an open bounded connected set S ⊂ Rn with a regular boundary Γ . Also

consider the convex and differentiable functions gi : Rn → R for each i ∈ {1, . . . , N}
and (G ◦ ∇) : U → R non-convex defined by
g(∇u) = min {gi (∇u)}, (17.14)

i∈{1,...,N}
and

G(∇u) = min {gi (∇u)}dS = g(∇u)dS. (17.15)
S i∈{1,...,N} S
We also assume
G(∇u)
→ +∞ as u U → ∞, (17.16)
u U
and
U = {u ∈ W 1,2 (S) | u = u0 on Γ }. (17.17)
As a preliminary result, we present Corollary 3.8, at p. 339 of Ekeland and Temam
[25] (here Ω stands for S).
Theorem 17.4.1. Let f be a Carathéodory function from Ω × (R × Rn ) into R which
satisfies
a2 (x) + c2 |ξ |α ≤ f (x, s, ξ ) ≤ a1 (x) + b|s|α + c1 |ξ |α (17.18)
where a1 , a2 ∈ L1 (Ω ), 1 < α < +∞, b ≥ 0 and c1 ≥ c2 > 0. Let u0 ∈ W 1,α (Ω ).
Under such assumptions, defining Û = {u | u − u0 ∈ W01,2 (Ω )}, we have

∗∗
inf f (x, u, ∇u)dx = min f (x, u; ∇u) dx (17.19)
The solutions of relaxed problem are weak cluster points in W 1,α (Ω ) of the min-
imizing sequences of primal problem.
Now we can enunciate the following result.
Theorem 17.4.2. Consider the definition and assumptions about (G ◦ ∇) : U → R
indicated in (17.14), (17.15), and (17.16). Also assuming the hypothesis of
Theorem 17.4.1, we have
inf {G(∇u) − u, f L2 (S) } = inf {G∗∗ (∇u) − u, f L2 (S) } (17.20)
u∈U u∈U
and there exists u0 ∈ U such that

min{G∗∗ (∇u) − u, f L2 (S) } = G∗∗ (∇u0 ) − u0, f L2 (S) . (17.21)
u∈U
The proof follows directly from Theorem 17.4.1. Our next proposition is very im-
portant to establish the subsequent results. It is simple so that we do not prove it.
Proposition 17.4.3. Consider g : Rn → R defined as
g(v) = min {gi (v)} (17.22)

i∈{1,...,N}
where gi : Rn → R are not necessarily convex functions. Under such assumptions,

we have
g∗ (v∗ ) = max {g∗i (v∗ )}. (17.23)
i∈{1,...,N}
Now we present the main duality principle for the scalar case.
Theorem 17.4.4. For (G ◦ ∇) : U → R defined as above, that is,

G(∇u) = min {gi (∇u)}dS, (17.24)
S i∈{1,...,N}
17.4 The Scalar Multi-well Problem 435
where here gi : Rn → R is convex, for all i ∈ {1, . . . , N} and F : U → R, defined as
F(u) = u, f L2 (S) , (17.25)
we have
min{G∗∗ (∇u) − F(u)} = sup {−G∗ (v∗ ) + u0, v∗ · nL2 (Γ ) }, (17.26)

u∈U v∗ ∈C∗
where
G∗ (v∗ ) = max {g∗i (v∗ )}dS (17.27)
S i∈{1,...,N}
and
C∗ = {v∗ ∈ Y ∗ | div(v∗ ) + f (x) = 0, in S}. (17.28)
Proof. We have that
G∗ (v∗ ) = G∗∗∗ (v∗ ) = sup{v, v∗ Y − G∗∗ (v)} (17.29)

v∈Y
that is,
G∗ (v∗ ) ≥ ∇u, v∗ Y − G∗∗(∇u)

= u, −div(v∗ )L2 (S) + u0, v∗ · nL2 (Γ ) − G∗∗ (∇u),
∀u ∈ U, v∗ ∈ Y ∗ and thus, for v∗ ∈ C∗ , we can write
G∗ (v∗ ) ≥ u0 , v∗ · nL2 (Γ ) + u, f L2 (S) − G∗∗ (∇u), ∀u ∈ U, (17.30)
or
inf {G∗∗ (∇u) − u, f L2 (S) } ≥ sup {−G∗ (v∗ ) + u0, v∗ · nL2 (Γ ) }. (17.31)
u∈U v∗ ∈C∗
The equality in (17.31) follows from the hypothesis indicated in (17.16) and
Theorem 17.2.2.
Observe that the dual formulation is convex but non-smooth. It is through the
points of non-smoothness that the microstructure is formed, specially when the orig-
inal primal formulation has no minimizers in the classical sense.
17.4.2 A Scalar Multi-well Formulation
To start this section, we present duality for the solution of a standard scalar
multi-well problem. Consider an open bounded connected set S ⊂ R3 , with a regular
boundary Γ , and the function (W ◦ ∇) defined as

1
W (∇u) = min |∇u − ai|2 (17.32)
i∈{1,...,N} 2
where
U = {u ∈ W 1,2 (S) | u = u0 on Γ } (17.33)
ai are known matrices, for all i ∈ {1, . . . , N}, so that the energy of the system is
modeled by J : U → R, where

J(u) = W (∇u) dx − u, f L2 (S) (17.34)
S
or
" #
1
J(u) = min |∇u − ai|2 dx − u, f L2 (S) . (17.35)
2 S i∈{1,...,N}
From Theorem 17.4.4 we have

1 ∗2
inf {J(u)} = sup − max |v | + v∗T ai dx + u0 , v∗ · nL2 (Γ )
u∈U v∗ ∈C∗ S i∈{1,...,N} 2
(17.36)
or
inf {J(u)}
u∈U

1 ∗2 N
= sup inf − |v | + ∑ λi v ai dx + u0 , v · nL2 (Γ )
∗T ∗
v∗ ∈C∗ λ ∈B S 2 i=1
where
B = {λ = (λ1 , . . . , λN ) measurable |
λi (x) ∈ [0, 1], ∀i ∈ {1, . . . , N}
N
and ∑ λi (x) = 1}, (17.37)
i=1
and
C∗ = {v∗ ∈ Y ∗ | div(v∗ ) + f = 0, in S} (17.38)
It is important to emphasize that, in general, this kind of problem does not present
minimizers in the classical sense. The solution of the dual problem (which is well
posed and convex) reflects the average behavior of minimizing sequences as weak
cluster points (of such sequences).
17.5 Duality for a Vectorial Multi-well Model Applicable to Phase Transition Problems 437
17.5 Duality for a Vectorial Multi-well Model Applicable

to Phase Transition Problems
In this section we consider duality for another class of multi-well problems simi-
lar as those found in [18]. However, here the format of our problem is more general,
not restricted to two-well formulations. Observe that the relaxation for the case of
three or more wells was so far an open question in the current literature. Our main
result is summarized by the next theorem.
Theorem 17.5.1. Consider an open bounded connected set S ⊂ R3 with a regular

boundary denoted by ∂ S, and the functional J : U → R where

1
J(u) = min {gk (ε (u)) + βk } dx − u, f U ,
2 S k∈{1,...,N}
gk (ε (u)) = (εi j (u) − ekij )Cikjlm (εlm (u) − eklm ).

The operator ε : U → Y = Y ∗ = L2 (S; R9 ) is defined by
1
εi j (u) = (ui, j + u j,i ), for i, j ∈ {1, 2, 3}.
2
Furthermore {Cikjlm } are positive definite matrices and βk ∈ R for each k ∈
{1, . . . , N}, and f ∈ L2 (S; R3 ) is a external load. Here ek ∈ R3×3 for k ∈ {1, . . . , N}
represent the stress-free configurations or phases presented by a solid with field of
displacements u = (u1 , u2 , u3 ) ∈ W 1,2 (S; R3 ) (due to f ). Also
U = {u ∈ W 1,2 (S; R3 ) | u = (0, 0, 0) ≡ θ on ∂ S} = W01,2 (S; R3 ).
We may write
J(u) = G(ε (u)) − F(ε (u)) − u, f U
where

1 K
G(ε (u)) = min {gk (ε (u)) + βk } dx + (εi j (u))Hi jlm (εlm (u)) dx,
2 S k∈{1,...,N} 2 S

K
F(ε (u)) = (εi j (u))Hi jlm (εlm (u)) dx,
2 S
{Hi jlm } is a positive definite matrix and K > 0. Under such assumptions, observing
that

∗ ∗ ∗ 1 ∗
G (v + z ) = max (vi j + z∗i j )Dkijlm (v∗lm + z∗lm )
S k∈{1,...,N} 2
0
+(v∗i j + z∗i j )Ĉikjlm eklm − βk dx, (17.39)
and

1
F ∗ (z∗ ) = z∗i j Ĥi jlm z∗lm dx,
2K S
we have

∗∗ ∗ ∗ ∗ ∗ ∗
inf {G (ε (u)) − F(ε (u)) − u, f U } ≥ sup inf {F (z ) − G (z + v )} .
u∈U v∗ ∈A∗ z∗ ∈Y ∗
Furthermore, this last duality principle may be written as
inf {G∗∗ (ε (u)) − F(ε (u)) − u, f U }

u∈U

1
v∗ ∈A∗ t∈B 2K S
N
1
k=1

∗ ∗ ∗
k k
, (17.40)
S
where
{Ĥi jlm } = {Hi jlm }−1 ,
{Dkijlm } = {Cikjlm + KHi jlm }−1 ,
and
{Ĉikjlm } = {Cikjlm + KHi jlm }−1 {Cikjlm }.
Moreover, z∗ (v∗ ,t) is obtained through Eq. (17.44), that is,
N N
1
Ĥi jlm z∗lm − ∑ {tk Dkijlm (v∗lm + z∗lm )} − ∑ tkĈikjlm eklm = 0, in S.
K k=1 k=1
Also,
A∗ = {v∗ ∈ Y ∗ | v∗i j, j + fi = 0 in S},
and
B = {(t1 , . . . ,tN ) measurable |

N
tk (x) ∈ [0, 1], ∀k ∈ {1, . . . , N}, ∑ tk (x) = 1, a.e. in S .
k=1
Finally, assuming the hypotheses of the main duality principle, there exist u0 ∈ U
and (v∗0 , z∗0 ) ∈ Ŷ ∗ such that
G∗∗ (ε (u0 )) − F(ε (u0 )) − u0 , f U

= inf {G∗∗ (ε (u)) − F(ε (u)) − u, f U }
u∈U

= sup inf {(F ◦ ε )∗ (ε ∗ z∗ ) − G∗ (v∗ + z∗ )}
v∗ ∈A∗ z∗ ∈Y ∗
= (F ◦ ε )∗ (ε ∗ z∗0 ) − G∗ (v∗0 + z∗0 ). (17.41)
Proof. Observe that

G∗ (v∗ + z∗ ) ≥ ε (u), v∗ Y + ε (u), z∗ Y − G∗∗ (ε (u)), ∀u ∈ U, v∗ , z∗ ∈ Y ∗ ,
that is,
−F ∗ (z∗ ) + G∗(v∗ + z∗ ) ≥ u, f U + ε (u), z∗ Y − F ∗ (z∗ ) − G∗∗(ε (u)),
∀u ∈ U, v∗ ∈ A∗ , z∗ ∈ Y ∗ .
Taking the supremum in z∗ in both sides of last inequality, we obtain
sup {−F ∗ (z∗ ) + G∗(z∗ + v∗ )} ≥ u, f U + F(ε (u)) − G∗∗ (ε (u)),
z∗ ∈Y ∗
∀u ∈ U, v∗ ∈ A∗ . Therefore

∗∗ ∗ ∗ ∗ ∗ ∗
inf {G (ε (u)) − F(ε (u)) − u, f U } ≥ sup inf {F (z ) − G (z + v )} ,
u∈U v∗ ∈A∗ z∗ ∈Y ∗
where

1 ∗
G∗ (v∗ + z∗ ) = max (v + z∗i j )Dkijlm (v∗lm + z∗lm )
S k∈{1,...,N} 2 ij
0
+(v∗i j + z∗i j )Ĉikjlm eklm − βk dx, (17.42)
{Dkijlm } = {Cikjlm + KHi jlm }−1 ,

and
{Ĉikjlm } = {Cikjlm + KHi jlm }−1 {Cikjlm }.
Hence the concerned duality principle is expressed as
inf {G∗∗ (ε (u)) − F(ε (u)) − u, f U }

u∈U

1
≥ sup ∗inf ∗ z∗ Ĥi jlm z∗lm dx
v∗ ∈A∗ z ∈Y 2K S i j

1
+ min − (v∗i j + z∗i j )Dkijlm (v∗lm + z∗lm )
S k∈{1,...,N} 2
0 00
∗ ∗
−(vi j + zi j )Ĉi jlm elm + βk dx ,
k k
so that
inf {G∗∗ (ε (u)) − F(ε (u)) − u, f U )}

u∈U

1
≥ sup inf inf z∗i j Ĥi jlm z∗lm dx
v∗ ∈A∗ t∈B z∗ ∈Y ∗ 2K S
N
1
+∑ − tk (v∗i j + z∗i j )Dkijlm (v∗lm + z∗lm )
k=1 S 2
1000
−(v∗i j + z∗i j )tkĈikjlm eklm + tk βk dx ,
where
B = {(t1 , . . . ,tN ) measurable |

N
tk (x) ∈ [0, 1], ∀k ∈ {1, . . . , N}, ∑ tk (x) = 1, a.e. in S . (17.43)
k=1
Observe that the infimum in z∗ is attained for functions satisfying

N N
1
Ĥi jlm z∗lm − ∑ {tk Dkijlm (v∗lm + z∗lm )} − ∑ tkĈikjlm eklm = 0, in S. (17.44)
K k=1 k=1
The final format of the concerned duality principle is given by
inf {G∗∗ (ε (u)) − F(ε (u)) − u, f U }

u∈U

1
v∗ ∈A∗ t∈B 2K S
N
1
k=1

∗ ∗ ∗
k k
, (17.45)
S
where z∗ (v∗ ,t) is obtained through Eq. (17.44).

The remaining conclusions follow from the main duality principle.
Remark 17.5.2. In fact the only hypothesis difficult to verify, concerning the main
duality principle, is the coercivity in z∗ . It is worth noting that to satisfy such a
hypothesis and obtain a finite value for J ∗ (v∗ ), where
J ∗ (v∗ ) = ∗inf ∗ {(F ◦ ε )(ε ∗ z∗ ) − G∗(v∗ + z∗ )},

z ∈Y
we may, if necessary to replace it by
J˜∗ (v∗ ) = inf {F ∗ (z∗ ) − G∗(v∗ + z∗ )},

z∗ ∈Ỹ
where z∗ ∈ Ỹ if z∗ ∈ Y ∗ and
∂ F(ε (u))
z∗ = ,
∂v
for some u ∈ U. We also recall that Ỹ may be defined through linear equations
of compatibility analogously to those of linear elasticity, considering that in the
present case F is a quadratic functional. We do not work these elementary details
here, postponing a more extensive analysis for a future work.
Remark 17.5.3. To illustrate how the dual formulation depends on K we analyze a

simple variational problem. Fix A = 10 and c1 , c2 ∈ R. Define J : U → R by
J(u) = G(u ) − u, f U ,
where
1
G(u ) = min{g1 (u ), g2 (u )} dx,
0
A A
g1 (u ) = (u − c1 )2 and g2 (u ) = (u − c2)2 ,
2 2
and
U = W01,2 ([0, 1]).
From Theorem 17.4.4, denoting Y ∗ = L2 ([0, 1]), we obtain
inf {J(u)} = sup {−G∗(v∗ )},

u∈u v∗ ∈C∗
where
1
G∗ (v∗ ) = max{g∗1 (v∗ ), g∗2 (v∗ )} dx,
0
1 ∗ 2
g∗1 (v∗ ) = (v ) + c1 v∗
2A
and
1 ∗ 2
g∗2 (v∗ ) = (v ) + c2 v∗ .
2A
Also
C∗ = {v∗ ∈ Y ∗ | (v∗ ) + f = 0 in [0, 1]}.
In this case there is no duality gap between the primal and dual problems. Anyway,
let us analyze the dual problem obtained as we redefine the primal formulation as
indicated in the next lines.
Define

K 1 2
Ĝ(u ) = G(u ) + (u ) dx,
2 0
and

K 1 2
F(u ) = (u ) dx.
2 0
From Theorem 17.3.1 (the main duality principle) we have
inf {J(u)} ≥ sup { ∗inf ∗ {F ∗ (z∗ ) − Ĝ∗ (z∗ + v∗)}}.

u∈U v∗ ∈C∗ z ∈Y
where
1
∗ ∗
Ĝ (v ) = max{ĝ∗1 (v∗ ), ĝ∗2 (v∗ )} dx,
0
1 Ac1 ∗ AKc21
ĝ∗1 (v∗ ) = (v∗ )2 + v −
2(A + K) A+K 2(A + K)
and
1 Ac2 ∗ AKc22
ĝ∗2 (v∗ ) = (v∗ )2 + v − .
2(A + K) A+K 2(A + K)
Also
1
1
F ∗ (z∗ ) = (z∗ )2 dx.
2K 0
Now defining
JK∗ (v∗ ) = ∗inf ∗ {F ∗ (z∗ ) − Ĝ∗(z∗ + v∗ )},
z ∈Y
we may write
inf {J(u)} ≥ sup {JK∗ (v∗ )},
u∈U v∗ ∈C∗
and
1
JK∗ (v∗ ) = inf{ ∗inf ∗ {F ∗ (z∗ ) − (t ĝ∗1 (v∗ + z∗ ) + (1 − t)ĝ∗2(v∗ + z∗ )) dx}},
t∈B z ∈Y 0
that is,
1 1
1
JK∗ (v∗ ) = inf{ ∗inf ∗ { (z∗ )2 dx − (t ĝ∗1 (v∗ + z∗ ) + (1 − t)ĝ∗2(v∗ + z∗ )) dx}},
t∈B z ∈Y 2K 0 0
where
B = {t measurable | t ∈ [0, 1] a.e. in [0, 1]}.
Evaluating the infimum in z∗ we obtain the final expression for JK∗ , namely,
1
JK∗ (v∗ ) = inf{ (a · (c1 − c2)2t − a · (c1 − c2)2 · t 2
t∈B 0
1
−tc1 v∗ − (1 − t)c2v∗ − (v∗ )2 ) dx}. (17.46)
2 · 10
For different values of K we have different values of a, for example:
1. for K = 5, we have a = 1.66667,
2. for K = 50, we have a = 4.166667,
3. for K = 500, a = 4.90196,

4. for K = 5, 000, a = 4.99000,
5. for K = 50, 000, a = 4.99999.
It seems that
a → 5, as K → ∞.
Also observe that
1
JK∗ (v∗ ) = inf (a · (c1 − c2 )2t − a · (c1 − c2 )2 · t 2
t∈B 0

∗ ∗ 1 ∗ 2
−tc1v − (1 − t)c2v − (v ) ) dx
2 · 10
1
1
≥ inf (−tc1 v∗ − (1 − t)c2v∗ − (v∗ )2 ) dx .
t∈B 0 2 · 10
Considering the vectorial case in question, for the analogous final value of a
(in this case a = 5), the difference observed through (17.47) will result in no duality
gap between the primal and dual problems. The difference is noted for the intermedi-
ate values of t, that is, for 0 < t < 1, and it is particularly relevant in a microstructural
context.
Finally, denoting α = a · (c1 − c2 )2 , through a Lagrange multiplier λ , we may
write
1
∗ ∗ (α − λ ) c1 + c2 ∗
JK (v ) = sup { − v } dx
λ ∈C 0 4 2
1 1
(c1 − c2 )2 (v∗ )2 1
− dx − (v∗ )2 dx ,
0 4(λ − α ) 2 · 10 0
where
C = {λ ∈ Y ∗ | λ − α > 0, a.e. in [0, 1]}.
Thus the final expression of the duality principle would be
1
(α − λ ) c1 + c2 ∗
inf {J(u)} ≥ sup { − v } dx
u∈U (v∗ ,λ )∈C∗ ×C 0 4 2
1 1
(c1 − c2 )2 (v∗ )2 1 ∗ 2
− dx − (v ) dx .
0 4(λ − α ) 2 · 10 0
It seems to be clear that different values of K and corresponding a may produce

different optimal microstructures for the dual problem. In the next result we prove
that the duality gap may become arbitrarily small as K → ∞.
Our final result is concerned with the evaluation of duality gap between the
primal and dual problems.
Theorem 17.5.4. Let ε > 0 be a small constant. Denote g : R9 → R by
g(y) = min {gk (y) + βk },

k∈{1,...,N}
where gk is as in Theorem 17.5.1. Consider U, Y , S ⊂ R3 and ε : U → Y also as in

Theorem 17.5.1.
J(u) = GK (ε (u)) − FK (ε (u)) − u, f U ,
where, for each K ∈ N,

1 K
GK (ε (u)) = min {gk (ε (u)) + βk } dx + (εi j (u))Hi jlm (εlm (u)) dx,
2 S k∈{1,...,N} 2 S

K
FK (ε (u)) = (εi j (u))Hi jlm (εlm (u)) dx,
2 S
Also define JK∗∗ : U → R by
JK∗∗ (u) = G∗∗

K (ε (u)) − FK (ε (u)) − u, f U .
From the last theorem we may select a sequence {u0K } ⊂ U such that

∗∗ ∗ ∗ ∗ ∗ ∗ ∗
JK (u0K ) = sup ∗inf ∗ {(FK ◦ ε ) (ε z ) − GK (v + z )} (17.47)
v∗ ∈A∗ z ∈Y
∀K ∈ N.
Suppose there exists K̃ such that
ε (u0K ) ∞ < K̃, ∀K ∈ N.
Under such assumptions, there exists Kε ∈ N such that if K > Kε then
|J(u0K ) − JK∗∗(u0K )| < ε ,
and also

J(u0 ) − sup inf {(F ◦ ε )∗ ∗ ∗
(ε z ) − G ∗ ∗
(v + z∗
) < ε.
K ∗ ∗
K
z ∈Y
K
v∗ ∈A∗
Proof. Choose K1 > K̃ sufficiently big so that defining

g(y), if |y| < K1 ,
g̃(y) =
+∞, otherwise,
we have
GK (ε (u0K )) = (G̃K )(ε (u0K )), ∀K ∈ N, (17.48)
and
G∗∗ ∗∗
K (ε (u0K )) = (G̃K ) (ε (u0K )), ∀K ∈ N, (17.49)
(this is possible since g is the minimum of N quadratic positive definite functions
and ε (u0K ) ∞ is uniformly bounded), where

1 K
(G̃K )(ε (u)) = g̃(ε (u)) dx + (εi j (u))Hi jlm (εlm (u)) dx,
2 S 2 S
From a standard mollification, there exists gδ ∈ C∞ (R9 ) such that

ε
|g(y) − gδ (y)| < , ∀|y| < K1 . (17.50)
|S|
Define
gδ (y), if |y| < K1 ,
g̃δ (y) =
+∞, otherwise.
Observe that from (17.50)
ε
|G̃K (ε (u0K )) − (G̃K )δ (ε (u0K ))| < ,
2
and
ε
|G̃∗∗ ∗∗
K (ε (u0K )) − (G̃K )δ (ε (u0K ))| < , ∀K ∈ N,
2
where

1 K
(G̃K )δ (ε (u)) = g̃δ (ε (u)) dx + (εi j (u))Hi jlm (εlm (u)) dx.
2 S 2 S
Since gδ ∈ C∞ (R9 ) we have that its matrix of second derivatives is bounded in

bounded sets. Thus, as
ε (u0K ) ∞ < K̃, ∀K ∈ N,
there exists Kε > 0 such that if K > Kε then
(G̃K )δ (ε (u0K )) − (G̃K )∗∗

δ (ε (u0K )) = 0.
Hence, if K > Kε , we obtain
|G̃K (ε (u0K )) − G̃∗∗

K (ε (u0K ))| ≤ |G̃K (ε (u0K )) − (G̃K )δ (ε (u0K ))|
+|(G̃K )δ (ε (u0K )) − (G̃K )∗∗ δ (ε (u0K ))|
+|(G̃K )∗∗
δ (ε (u 0K )) − G̃ ∗∗
K ε (u0K ))|
(
ε ε
< + = ε. (17.51)
2 2
From (17.48) and (17.49) we obtain
|GK (ε (u0K )) − G∗∗

K (ε (u0K ))| < ε ,
so that from this last inequality and (17.47) we finally obtain

J(u0 ) − sup inf {(F ◦ ε )∗ ∗ ∗
(ε z ) − G ∗ ∗
(v + z∗
)} < ε,
K ∗ ∗
z ∈Y
K K
v∗ ∈A∗
if K > Kε .
Remark 17.5.5. Through this final result we have shown that for the problem in
question, the duality gap between the primal and dual formulations becomes arbi-
trarily small as K goes to ∞.
17.6 Conclusion
In this chapter, we have developed dual variational approaches for multi-well

formulations, introducing duality as an efficient tool to tackle this kind of problem.
The standard results of convex analysis can be used to clarify the understanding of
mixture of phases. What is new and relevant is the duality principle for vectorial
multi-well models, applicable to phase transition problems, such as those studied in
the article by Chenchiah and Bhattacharya, [18]. For such problems, duality is an
interesting alternative for relaxation and computation.
In our view, the importance of duality for the theoretical and numerical analysis
of multi-well and related phase transition problems seems to have been clarified.
Chapter 18
More on Duality Principles for Multi-well
Problems
18.1 Introduction
In this chapter we develop dual variational formulations for a more general class
of non-convex multi-well variational models. Such models appear in similar form
in phase transition and related problems. Please see [5, 7, 18–20, 28, 48, 64] for
details. We also address problems of conductivity in composites and optimal design
and control in elasticity.
In this section we state and prove the main result in this chapter, which is sum-
marized by the next theorem. From now on, by a regular boundary ∂ Ω of Ω ⊂ R3 ,
we mean regularity enough so that the standard Gauss–Green formulas of integra-
tions by parts, the Sobolev imbedding theorem, and the trace theorem to hold. Also,
n denotes the outward normal to ∂ Ω and derivatives must be understood always in
the distributional sense.
Remark 18.2.1. At some points of our analysis we refer to the problems in question
after discretization. In such a case, we are referring to their approximations in a
finite element or finite differences context. Finally, to simplify the notation in some
parts of the text, we may denote L2 (Ω ), L2 (Ω ; R3 ) and L2 (Ω ; R3×3 ) simply by L2 .

boundary denoted by ∂ Ω = Γ . Denote J : U × B → R by
J(u,t) = G(ε (u),t) − u, f L2 ,
where

1
G(ε (u),t) = tk gk (ε (u)) dx,
2 Ω
448 18 More on Duality Principles for Multi-well Problems
gk (ε (u)) = Hikjlm (εi j (u) − ekij )(εlm (u) − eklm ) + βk , ∀k ∈ {1, . . . , N}.
Here
1
ε (u) = {εi j (u)} = (ui, j + u j,i ) ,
2
B = {t = (t1 , . . . ,tN ) measurable |

N
∑ tk (x) = 1, 0 ≤ tk (x) ≤ 1, in Ω , ∀k ∈ {1, . . . , N}}, (18.1)
k=1
and
U = {u ∈ W 1,2 (Ω ; R3 ) | u = u0 on ∂ Ω }.
Also, βk ∈ R, {ekij } ∈ L2 (Ω ; R3×3 ), f ∈ L2 (Ω ; R3 ) and
{Hikjlm }
are fourth-order positive definite constant tensors ∀k ∈ {1, . . . , N}.
inf {J(u,t)} ≥ sup {−J ∗(v∗ )},

(u,t)∈U×B v∗ ∈A∗
where
J ∗ (v∗ ) = sup{G∗ (v∗ ,t)} − v∗i j n j , (u0 )i L2 (Γ ) ,
t∈B
G (v ,t) = sup {v, v∗ L2 − G(v,t)},

∗ ∗
v∈L2
and
A∗ = {v∗ ∈ Y ∗ | v∗i j, j + fi = 0 in Ω }.
Furthermore, there exists v∗0 ∈ A∗ ⊂ Y ∗ = Y = L2 (Ω ; R3×3 ) such that
−J ∗ (v∗0 ) = max
∗ ∗
{−J ∗ (v∗ )}.
v ∈A
Finally, for the discretized version of the problem, assume t0 ∈ B such that
G∗ (v∗0 ,t0 ) = sup{G∗ (v∗0 ,t)},
t∈B
is also such that the hypotheses of Corollary 11.1 are satisfied.

Under such hypotheses, for û ∈ U such that
ε (û) ∈ ∂ (J ∗ (v∗0 )),
we have
J(û,t0 ) = min {J(u,t)} = max

∗ ∗
{−J ∗ (v∗ )} = −J ∗ (v∗0 ).
(u,t)∈U×B v ∈A
Proof. Observe that
J(u,t) = G(ε (u),t) − u, f L2

= −ε (u), v∗ L2 + G(ε (u),t)
+ε (u), v∗ L2 − u, f L2
≥ inf {−v, v∗L2 + G(v,t)}
v∈Y
+ inf {ε (u), v∗ L2 − u, f L2 }
u∈U
= −G∗ (v∗ ,t) + v∗i j n j , (u0 )i L2 (Γ )
≥ inf {−G∗(v∗ ,t) + v∗i j n j , (u0 )i L2 (Γ ) }
t∈B
= −J ∗ (v∗ ), (18.2)
∀(u,t) ∈ U × B, v∗ ∈ A∗ .
Thus,
inf {J(u,t)} ≥ sup {−J ∗ (v∗ )}. (18.3)
(u,t)∈U×B v∗ ∈A∗
The dual functional is concave and upper semicontinuous; therefore, it is weakly

upper semicontinuous, so that from the direct method of calculus of variations (since
it is a standard procedure we do not provide details here), there exists v∗0 ∈ A∗ such
that
−J ∗ (v∗0 ) = max
∗ ∗
{−J ∗ (v∗ )}.
v ∈A
At this point and on, we consider the discretized version of the problem.
From the hypotheses, t0 ∈ B such that
G∗ (v∗0 ,t0 ) = sup{G∗ (v∗0 ,t)},
t∈B
is also such the hypotheses of Corollary 11.1 are satisfied.

From such a corollary, we may infer that for the extended functional
−(û)i , v∗i j, j + fi L2 (Ω ) − J ∗ (v∗ )
the optimal extremal relation
ε (û) ∈ ∂ (J ∗ (v∗0 )),
stands for
∂ G∗ (v∗0 ,t0 )
ε (û) = .
∂ v∗
Hence, since for a fixed t0 ∈ B G(ε (u),t0 ) is convex in u, we get
G∗ (v∗0 ,t0 ) = ε (û), v∗0 L2 − G(ε (û),t0 )
= −ûi , (v0 )∗i j, j L2 (Ω ) + (v0 )∗i j n j , (u0 )i L2 (Γ )
−G(ε (û),t0 ). (18.4)
Therefore,
G∗ (v∗0 ,t0 ) − (v0 )∗i j n j , (u0 )i L2 (Γ )
= −G(ε (û),t0 ) + ûi, fi L2 (Ω ) . (18.5)
From this we get
J ∗ (v∗0 ) = −J(û,t0 ).
From this last equation and (18.3), the proof is complete.
18.3 Another Duality Principle for Phase Transition Models
In this section we state and prove another relevant result, which is summarized
by the next theorem.
boundary denoted by ∂ Ω = Γ . Denote J : U × B → R by
J(u,t) = G(ε (u),t),
where

1
G(ε (u),t) = tk gk (ε (u)) dx,
2 Ω
gk (ε (u)) = Hikjlm (εi j (u) − ekij )(εlm (u) − eklm ) + βk , ∀k ∈ {1, . . . , N}.
Here

1
ε (u) = {εi j (u)} = (ui, j + u j,i ) ,
2
B = {t = (t1 , . . . ,tN ) measurable |

N
∑ tk (x) = 1, 0 ≤ tk (x) ≤ 1, in Ω ,
k=1
∀k ∈ {1, . . . , N}}, (18.6)
and
U = {u ∈ W 1,2 (Ω ; R3 ) | u = u0 on ∂ Ω }.
Also, βk ∈ R, {ekij } ∈ L2 (Ω ; R3×3 ) ≡ L2 , and as above,
{Hikjlm }
are fourth-order positive definite constant tensors, ∀k ∈ {1, . . . , N}.
inf {J(u,t)} = inf {J ∗ (v∗ ,t)},

(u,t)∈U×B (v∗ ,t)∈A∗ ×B
18.4 Duality for a Problem on Conductivity in Composites 451
where
J ∗ (v∗ ,t) = G∗1 (v∗1 ,t) + G∗2(v∗2 ,t)

−(v∗1 )i j n j , (u0 )i L2 (Γ ) − (v∗2 )i j n j , (u0 )i L2 (Γ ) , (18.7)
G∗1 (v∗1 ,t) = sup {v, v∗1 L2 − G1 (v,t)},

v∈L2
G∗2 (v∗2 ,t) = sup {v, v∗2 L2 − G2 (v,t)},

v∈L2

1 1
−G1 (v,t) = G(v,t) − tk Hikjlm (vi j − ekij )(vlm − eklm ) dx,
2 2 Ω

1 1
−G2 (v,t) = G(−v,t) + tk Hikjlm (vi j − ekij )(−vlm − eklm ) dx
2 2 Ω
and
A∗ = {v∗ ∈ Y ∗ | (v∗1 )i j, j + (v∗2 )i j, j = 0 in Ω }.
Proof. Observe that
inf {J(u,t)} = inf{ inf {G(ε (u),t)}}

(u,t)∈U×B t∈B u∈U

1
= inf{ inf {− tk Hikjlm (εi j (û) − ekij )(εlm (u) − eklm ) dx
t∈B u∈U 2 Ω
1
+ G(ε (u),t)
2
1
+ tk Hikjlm (εi j (û) − ekij )(εlm (u) − eklm ) dx
2 Ω
1
+ G(ε (u),t)}}
2
= inf{sup{−G1 (ε (û),t) − G2 (ε (û),t)}}
t∈B û∈U
= inf{ ∗inf ∗ {G∗1 (v∗1 ,t) + G∗2(v∗2 ,t)

t∈B v ∈A
−(v∗1 )i j n j , (u0 )i L2 (Γ )
−(v∗2 )i j n j , (u0 )i L2 (Γ ) }}
∗ ∗
= inf {J (v ,t)}. (18.8)
(v∗ ,t)∈A∗ ×B
18.4 Duality for a Problem on Conductivity in Composites
For the primal formulation we repeat the statements found in reference [27].
Consider a material confined into a bounded domain Ω ⊂ RN , N > 1. The medium
is obtained by mixing two constituents with different electric permittivity and
conductivity. Let Q0 and Q1 denote the two N × N symmetric matrices of electric

permittivity corresponding to each phase. For each phase, we also denote by L j ,
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 Ω which we denote by E ⊂ Ω . Regarding the set E as our
design variable, we introduce the characteristic function χ : Ω → {0, 1}:

1, if x ∈ E,
χ (x) = (18.9)
0, otherwise,
Thus,

dx = χ (x)dx = t1 dx = t1 |Ω |. (18.10)
E Ω Ω
The matrix of conductivity corresponding to the material as a whole is L = χ L1 +
(1 − χ )L0.
Finally, the electrostatic potential, denoted by u : Ω → R, is supposed to satisfy
the equation
div[χ L1 ∇u + (1 − χ )L0∇u] + f (x) = 0, in Ω , (18.11)
u = u0 , on ∂ Ω (18.12)
where f : Ω → R is a given source or sink of current (we assume f ∈ L2 (Ω )). From
now on we assume N = 3. Consider the slightly different problem of minimizing
the cost functional

χ (1 − χ )
I(χ , u) = (∇u)T Q1 ∇u + (∇u)T Q0 ∇u dx (18.13)
Ω 2 2
subject to
div[χ L1 ∇u + (1 − χ )L0∇u] + f (x) = 0 (18.14)
and
χ dx ≤ t1 |Ω |,
Ω
where u ∈ U, here U = {u ∈ W 1,2 (Ω ) | u = u0 on ∂ Ω }.

Our main duality principle for such a non-convex optimization problem is sum-
marized by the next theorem.

boundary denoted by ∂ Ω = Γ . Redefine without relabeling it, J : U × B → R =
R ∪ {+∞} by
J(u,t) = G(∇u,t) + Ind(u,t),
where
U = {u ∈ W 1,2 (Ω ) | u = u0 on ∂ Ω },
Y = Y ∗ = L2 (Ω ; R3 ) ≡ L2 ,
B = {t measurable | t ∈ {0, 1}, a.e. in Ω

and t dx ≤ t1 |Ω |}, (18.15)
Ω
and
0 < t1 < 1.
Also,
" #
1
G(∇u,t) = t(∇uT Q1 ∇u) + (1 − t)(∇uT Q0 ∇u) dx,
2 Ω

0 if (u,t) ∈ A
Ind(u,t) = (18.16)
+∞ otherwise,
where
A = {(u,t) ∈ U × B |
div((tL1 + (1 − t)L0)∇u) + f = 0 in Ω }. (18.17)
inf {J(u,t)} ≥ sup {−J ∗ (v∗ , λ )}.

(u,t)∈U×B (v∗ ,λ )∈A∗ ×U1
Here,
J ∗ (v∗ , λ ) = sup{G∗ (v∗ − (tL1 + (1 − t)L0)∇λ )} + λ , f L2 (Ω )

t∈B

− (v∗ · n)u0 dΓ , (18.18)
Γ
G∗ (v∗ − (tL1 + (1 − t)L0)∇λ ) = sup{v, v∗ L2 − G(v,t)

v∈Y
−(tL1 + (1 − t)L0)∇λ , vL2 },
and
A∗ = {v∗ ∈ Y ∗ | div(v∗ ) = 0 in Ω }.
Moreover, there exists (v∗0 , λ0 ) ∈ A∗ × U1 such that
−J ∗ (v∗0 , λ0 ) = max {−J ∗ (v∗ , λ )},

(v∗ ,λ )∈A∗ ×U1
where U1 = W01,2 (Ω ).
Assume, after discretization, that t0 ∈ B such that
G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 ) = sup{G∗ (v∗0 − (tL1 + (1 − t)L0)∇λ0 )},

t∈B
is also such that for an appropriate û ∈ U the optimal inclusions
∇û ∈ ∂v∗ J(v∗0 , λ0 ),
and
θ ∈ ∂λ J ∗ (v∗0 , λ0 ),
stand for
∂ G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 )
∇û = ,
∂ v∗
and

∂ G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 )
θ = div (t0 L1 + (1 − t0)L0 ) + f.
∂ v∗
Under such additional hypotheses we have
J(û,t0 ) = inf {J(u,t)}

(u,t)∈U×B
= max {−J ∗ (v∗ , λ )}
(v∗ ,λ )∈A∗ ×U1
= −J ∗ (v∗0 , λ0 ). (18.19)
Proof. Observe that
J(u,t) = G(∇u,t) + Ind(u,t)

≥ G(∇u,t) − λ , div((tL1 + (1 − t)L0)∇u + f L2 (Ω )
= G(∇u,t) + (tL1 + (1 − t)L0)∇λ , ∇uL2 − λ , f L2
= −∇u, v∗ L2 + G(∇u,t) + (tL1 + (1 − t)L0)∇λ , ∇uL2
−λ , f L2 + ∇u, v∗ L2
≥ inf {−v, v∗ L2 + G(v,t) + (tL1 + (1 − t)L0)∇λ , vL2
v∈Y
+ inf {−λ , f L2 + ∇u, v∗ L2 }
u∈U
= −G∗ (v∗ − (tL1 + (1 − t)L0)∇λ ) − λ , f L2

+ (v∗ · n)u0 dΓ
Γ
≥ inf{−G∗ (v∗ − (tL1 + (1 − t)L0)∇λ )} − λ , f L2
t∈B

+ (v∗ · n)u0 dΓ
Γ
= −J (v, λ ), ∀(u,t) ∈ U × B, (v∗ , λ ) ∈ A∗ × U1.
∗
(18.20)
Thus,
inf {J(u,t)} ≥ sup {−J ∗(v∗ , λ )}. (18.21)
(u,t)∈U×B (v∗ ,λ )∈A∗ ×U1
From the remaining hypotheses, after discretization, we have that t0 ∈ B such that
G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 ) = sup{G∗ (v∗0 − (tL1 + (1 − t)L0)∇λ0 ),

t∈B
is also such that for an appropriate û ∈ U the optimal inclusions
∇û ∈ ∂v∗ J(v∗0 , λ0 ),
and
θ ∈ ∂λ J ∗ (v∗0 , λ0 ),
stand for
∂ G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 )
∇û = ,
∂ v∗
and

∂ G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 )
θ = div (t0 L1 + (1 − t0)L0 ) + f.
∂ v∗
Hence
div((t0 L1 + (1 − t0)L0 )∇û) + f = θ , in Ω ,
and
G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 ) = ∇û, v∗0 L2 − G(∇û,t0 )

−(t0 L1 + (1 − t0)L0 )∇λ0 , ∇ûL2

= (v∗0 · n)u0 dΓ − G(∇û,t0 )
Γ
−(t0 L1 + (1 − t0)L0 )∇λ0 , ∇ûL2

= (v∗0 · n)u0 dΓ − G(∇û,t0 )
Γ
−λ0 , f L2 . (18.22)
Therefore,
G(∇û,t0 ) + Ind(û,t0 ) = −G∗ (v∗0 − (t0 L1 + (1 − t0)L0 )∇λ0 )

−λ0 , f L2 + (v∗0 · n)u0 dΓ , (18.23)
Γ
so that
J(û,t0 ) = −J ∗ (v∗0 , λ0 ).
18.5 Optimal Design and Control for a Plate Model
In this section, we develop duality for the optimal control of a two-phase plate
model. The control variable is t1 and is related to the elastic constant distribution
K(t1 ) given by
K(t1 ) = t1 K1 + (1 − t1)K2
where K1 % K2 > 0. Moreover, the plate stiffness is given by the tensor Hαβ λ μ (t) as
specified in the next theorem. We are concerned with the calculation of the optimal
t,t1 which minimizes the plate compliance (or, equivalently, its inner work), under
appropriate constraints. The plate model in question is the Kirchhoff one, where
Ω ⊂ R2 denotes the plate mid-surface. Moreover, w ∈ W 1,2 (Ω ) denotes the field of
displacements resulting from a vertical load f ∈ L2 (Ω ) action. Please see the next
theorem for details.
18.5.1 The Duality Principle for the Plate Model
We start this subsection with the following theorem.

Theorem 18.5.1. Let Ω ⊂ R2 be an open bounded connected set with a regular
boundary denoted by ∂ Ω = Γ . Define J : U × B × B1 → R = R ∪ {+∞}, by
J(w,t,t1 ) = G1 (Λ w,t) + G2 (w,t1 ) + Ind(w,t,t1),
where U = W02,2 (Ω ),
B = {t measurable | t ∈ {0, 1}, a.e. in Ω

and t dx ≤ t˜|Ω |}, (18.24)
Ω
0 < t˜ < 1,
B1 = {t1 measurable | 0 ≤ t1 ≤ 1, a.e. in Ω

and t1 dx ≤ t˜1 |Ω |}, (18.25)
Ω
and
0 < t˜1 < 1.
Also, Λ :→ Y = Y ∗ = L2 (Ω ; R2×2 ) ≡ L2 is given by
Λ w = {w,αβ },

1
G1 (Λ w,t) = Hαβ λ μ (t)w,αβ w,λ μ dx,
2 Ω
18.5 Optimal Design and Control for a Plate Model 457

1
G2 (w,t1 ) = K(t1 )w2 dx,
2 Ω
Hαβ λ μ (t) = t(H0 )αβ λ μ + (1 − t)(H1)αβ λ μ ,
where {(H0 )αβ λ μ } and {(H1 )αβ λ μ } are fourth-order positive definite constant
tensors. Also,
K(t1 ) = t1 K1 + (1 − t1)K2 ,
where
K1 % K2 > 0.
Moreover,

0 if (w,t,t1 ) ∈ A
Ind(w,t,t1 ) = (18.26)
+∞ otherwise,
where
A = {(w,t,t1 ) ∈ U × B × B1 |
(Hαβ λ μ (t)w,λ μ ),αβ + K(t1 )w − f = 0 in Ω }. (18.27)
inf {J(w,t,t1 )} = inf {J ∗ (v∗ ,t,t1 )}.

(w,t,t1 )∈U×B×B1 (v∗ ,t,t1 )∈Y ∗ ×B×B1
Here,
J ∗ (v∗ ,t,t1 ) = G∗1 (v∗ ,t) + G∗2 (Λ ∗ v∗ − f ,t1 ),
where
G∗1 (v∗ ,t) = sup{v, v∗ L2 − G1 (v,t)}

v∈Y

1
= H αβ λ μ (t)v∗αβ v∗λ μ dx, (18.28)
2 Ω
H αβ λ μ (t) = {Hαβ λ μ (t)}−1 ,
G∗2 (v∗ ,t1 ) = sup {−Λ w, v∗ L2 − G2 (w,t1 ) + w, f L2 (Ω ) }

w∈U

1
= K(t1 )((v∗αβ ),αβ − f )2 ; dx, (18.29)
2 Ω
K(t) = {K(t)}−1 .
Furthermore, also under the mentioned hypotheses, we have
inf {J(w,t,t1 )} ≥ sup {−J(

ˆ ŵ)},
(w,t,t1 )∈U×B×B1 ŵ∈U
where
ˆ ŵ) =
J( sup {G1 (Λ ŵ,t) + G2 (ŵ,t1 ) − ŵ, f L2 }.
(t,t1 )∈B×B1
Finally, there exists ŵ0 ∈ U such that

−J(
ˆ ŵ0 ) = max{−J(
ˆ ŵ)}.
ŵ∈U
Suppose, after discretization, that (t0 , (t1 )0 ) ∈ B × B1 such that

ˆ ŵ0 ) =
J( sup {G1 (Λ ŵ0 ,t) + G2(ŵ0 ,t1 )
(t,t1 )∈B×B1
−ŵ0 , f L2 }
= G1 (Λ ŵ0 ,t0 ) + G2(ŵ0 , (t1 )0 )
−ŵ0 , f L2 . (18.30)
is also such that the optimal inclusion

θ ∈ ∂ J(
ˆ ŵ0 )
stands for
δŵ {G1 (Λ ŵ0 ,t0 ) + G2 (ŵ0 , (t1 )0 ) − ŵ0 , f L2 } = θ .
inf {J(w,t,t1 )} = J(ŵ0 ,t0 , (t1 )0 ) = −J(

ˆ ŵ0 ) = max{−J(
ˆ ŵ)}.
(w,t,t1 )∈U×B×B1 ŵ∈U
Proof. Observe that
inf {J(w,t,t1 )} = inf { inf {J(w,t,t1 )}

(w,t,t1 )∈U×B×B1 (t,t1 )∈B×B1 w∈U
= inf {sup { inf {G1 (Λ w,t) + G2 (w,t1 )
(t,t1 )∈B×B1 ŵ∈U w∈U
−ŵ, (Hαβ λ μ (t)w,λ μ ),αβ + K(t1 )w − f L2 }}
= inf {sup {−G2 (Λ ŵ,t) − G2(ŵ,t1 )
(t,t1 )∈B×B1 ŵ∈U
+ŵ, f L2 }}
= inf { ∗inf ∗ {G∗1 (v∗ ,t)
(t,t1 )∈B×B1 v ∈Y
+G∗2 (Λ ∗ v∗ − f ,t1 )}}
∗ ∗
= inf {J (v ,t,t1 )}. (18.31)
(v∗ ,t,t1 )∈Y ∗ ×B×B1
Moreover,
J(w,t,t1 ) ≥ inf {J(w,t,t1 )}

w∈U
≥ inf {G1 (Λ w,t) + G2 (w,t1 )
w∈U
18.5 Optimal Design and Control for a Plate Model 459
−ŵ, (Hαβ λ μ (t)w,λ μ ),αβ + K(t1 )w − f L2 (Ω ) }

= −G1 (Λ ŵ,t) − G2(ŵ,t1 ) + ŵ, f L2
≥ inf {−G1 (Λ ŵ,t) − G2 (ŵ,t1 ) + ŵ, f L2 }
(t,t1 )∈B×B1
= −J(
ˆ ŵ), ∀(w,t,t1 ) ∈ U × B × B1, ŵ ∈ U. (18.32)
Hence,
inf {J(w,t,t1 )} ≥ sup {−J(
ˆ ŵ)}. (18.33)
(w,t,t1 )∈U×B×B1 ŵ∈U
Also, since −Jˆ : U → R is concave and continuous, it is weakly upper semicontin-

uous, so that by the direct method of calculus of variations (since it is a standard
procedure, we do not provide more detail here), there exists ŵ0 ∈ U such that
−J(
ˆ ŵ0 ) = max{−J(
ˆ ŵ)}.
ŵ∈U
Finally, from the hypotheses, after discretization, we have that (t0 , (t1 )0 ) ∈ B× B1
such that
ˆ ŵ) =
J( sup {G1 (Λ ŵ0 ,t) + G2(ŵ0 ,t1 )
(t,t1 )∈B×B1
−ŵ0 , f L2 }
= G1 (Λ ŵ0 ,t0 ) + G2 (ŵ0 , (t1 )0 )
−ŵ0 , f L2 . (18.34)
θ ∈ ∂ J(
ˆ ŵ0 )
stands for
δŵ {G1 (Λ ŵ0 ,t0 ) + G2 (ŵ0 , (t1 )0 ) − ŵ0 , f L2 } = θ .
Thus,
ˆ ŵ0 ) = −G1 (Λ ŵ0 ,t0 ) − G2 (ŵ0 , (t1 )0 )
− J(
+ŵ0 , f L2
= G1 (Λ ŵ0 ,t0 ) + G2(ŵ0 , (t1 )0 ) + Ind(ŵ0,t0 , (t1 )0 )
= J(ŵ0 ,t0 , (t1 )0 ). (18.35)
From this (18.33), the proof is complete.

In this section we develop a numerical example. In fact we address the problem

of establishing the optimal distribution of springs on a beam similarly as above
specified, in order to minimize its compliance. Here the control variable is t, where
the function relating to the constant spring distribution is given by K(t) = t(x)K1 +
(1 − t(x))K2 , where K1 % K2 > 0 are the constants related to a strong and a weak
spring, respectively. Our main result is summarized by the following theorem, which
may be proven similarly as the last one (we do not provide a proof here).
Theorem 18.6.1. Consider a straight beam with rectangular cross section in which
the axis corresponds to the set Ω = [0, l]. Define J˜ : U × B → R = R ∪ {+∞} by
l l
1 1
˜
J(w,t) = EIw2,xx dx + K(t)w2 dx,
2 0 2 0
subject to
EIwxxxx + K(t)w − f = 0, in [0, l], (18.36)
where
U = {w ∈ W 2,2 (Ω ) | w(0) = w(l) = 0},
1
B = {t measurable | 0 ≤ t ≤ 1, a.e. in Ω and t dx ≤ c0 Ω },
0
and 0 < c0 < 1, K1 % K2 > 0.
Now, define J : U × B → R = R ∪ {+∞} by
J(w,t) = G(Λ w) + F(w,t) + Ind(w,t)
where
Λ : U → Y = Y ∗ = L2 (Ω ) ≡ L2 ,
is expressed by
Λ w = w,xx ,
l
1
G(Λ w) = EIw2,xx dx,
2 0

1 l
F(w,t) = K(t)w2 dx,
2 0

0, if (w,t) ∈ B0
Ind(w,t) = (18.37)
+∞, otherwise,
and
B0 = {(w,t) ∈ U × B such that (18.36) is satisfied }.
inf {J(w,t)} = inf {J ∗ (v∗ ,t)},

(w,t)∈U×B (v∗ ,t)∈A∗ ×B
where A∗ = U,
J ∗ (v∗ ,t) = G∗ (v∗ ) + F ∗ (Λ ∗ v∗ − f ,t),
G∗ (v∗ ) = sup{v, v∗ L2 − G(v)}

v∈Y
l
1
= (v∗ )2 dx, (18.38)
2EI 0
F ∗ (Λ ∗ v∗ ,t) = sup {−wxx , v∗ L2 − F(w,t)

w∈U
+w, f L2 }
l
1 ((v∗ )xx − f )2
= dx. (18.39)
2 0 K(t)
Furthermore, also under the mentioned hypotheses, we have
inf {J(w,t)} ≥ sup {−J(

ˆ ŵ)},
(w,t)∈U×B ŵ∈U
where
ˆ ŵ) = G(Λ ŵ) + F̂(ŵ) − ŵ, f L2 ,
J(
and
F̂(ŵ) = sup{F(ŵ,t)}.
t∈B
Finally, there exists ŵ0 ∈ U such that
−J(
ˆ ŵ0 ) = max{−J(
ˆ ŵ)}.
ŵ∈U
Suppose, after discretization, that t0 ∈ B such that
F̂(ŵ0 ) = F(ŵ0 ,t0 ),
θ ∈ ∂ J(
ˆ ŵ0 )
stands for
δŵ {G(Λ ŵ0 ) + F(ŵ0 ,t0 ) − ŵ0 , f L2 } = θ .
inf {J(w,t)} = J(ŵ0 ,t0 ) = −J(

ˆ ŵ0 ) = max{−J(
ˆ ŵ)}.
(w,t)∈U×B ŵ∈U
We have computed, through the dual formulations, the solution for the numerical
values EI = 105 , K1 = 990, 000, K2 = 10, and P = 1, 000, t1 = 0.5, l = 1.0 with units
relating the international system. We consider two cases: first, for t, the constraint
0 ≤ t ≤ 1 and, in a second step, the constraint t ∈ {0, 1}.
For the results, please see Figs. 18.1, 18.2, 18.3, and 18.4, below indicated.
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 0.2 0.4 0.6 0.8 1
Fig. 18.1 Parameter 0 ≤ t ≤ 1 relating the distribution of spring constants
18.7 Conclusion
At some point of our analysis, for almost all results, we have considered the dis-
cretized problem version. The reason is that we may not guarantee the attainability
of a measurable t0 ∈ B such that
G∗ (v∗0 ,t0 ) = sup{G∗ (v∗0 ,t)},

t∈B
before discretization. It seems to be clear that after discretization, we are more

likely to satisfy the hypotheses of Corollary 11.1. We emphasize again that if such
hypotheses are satisfied by (v∗0 ,t0 ) ∈ A∗ × B, then the duality gap between the primal
and dual formulations is zero for the different problems addressed. For the second
result we obtain directly a duality principle with no duality gap between the primal
and dual problems and such that the dual formulation computation is relatively easy.
We also highlight again that for some of the results developed, the dual formulations
are concave and also useful for relaxation and to obtain numerical results. In the last
section we present a numerical example. We also emphasize the results obtained are
consistent with the problem physics. Finally, for this same numerical example, it is
worth mentioning that through the theoretical results developed, we may assert that
18.7 Conclusion 463
120
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8 1
Fig. 18.2 Moments v∗ (x) relating the case 0 ≤ t ≤ 1
1.2
0.8
0.6
0.4
0.2
−0.2
0 0.2 0.4 0.6 0.8 1
Fig. 18.3 Parameter t ∈ {0, 1} relating the distribution of spring constants
the solution related to Figs. 18.3 and 18.4 is the global optimal one. On the other
hand, the solution related to Figs. 18.1 and 18.2 is just a critical point (anyway, a
possible candidate to global optimal solution).
120
100
80
60
40
20
0
0 0.2 0.4 0.6 0.8 1
Fig. 18.4 Moments v∗ (x) relating the case t ∈ {0, 1}

Chapter 19
Duality and Computation for Quantum
Mechanics Models
Fabio Botelho and Anderson Ferreira
19.1 Introduction
Our first objective is to obtain a duality principle for a class of nonlinear

eigenvalue problems. The results are closely related to the canonical duality
procedure; for details see Gao [36]. In a second step we apply the method to
compute examples of nonlinear Schrödinger equation. We highlight the nonlinear
Hamiltonian part refers to a kind of wave function self-interacting term, which
models a great variety of physical phenomena. Among others, we would mention
the nonlinear dynamics of superfluids, the Ginzburg–Landau theory of phase tran-
sitions, and the propagation of electromagnetic waves in plasmas.
At this point we start to describe the equation in question and respective varia-
tional formulation.
Let Ω ⊂ R3 be an open, bounded, connected set with a regular boundary denoted
by ∂ Ω . By a regular boundary we mean regularity enough so that the Sobolev
imbedding theorem, the trace theorem, and the standard Gauss–Green formulas of
integration by parts hold.
Moreover, we define
U = {ϕ ∈ W 1,2 (Ω ) | ϕ = 0 on ∂ Ω } = W01,2 (Ω ).
We emphasize the derivatives must be understood in the distributional sense,

whereas the boundary conditions are in the sense of traces.
F. Botelho ()
Department of Mathematics and Statistics, Federal University of Pelotas, Pelotas, RS-Brazil
e-mail: [email protected]
A. Ferreira
Department of Physics, Federal University of Pelotas, Brazil
466 19 Duality and Computation for Quantum Mechanics Models
Consider the eigenvalue problem given by the solution in U of equation
h̄2 2
− ∇ ϕ + 2α |ϕ |2 ϕ + V (x)ϕ − μϕ = 0 in Ω , (19.1)
2m
where μ ∈ R is a suitable Lagrange multiplier to be obtained such that the following
constraint is satisfied:
|ϕ |2 dx = c. (19.2)
Ω
Here c > 0 is appropriate constant to be specified.
Furthermore, α , h̄, m are positive constants and V ∈ L2 (Ω ).
In case ϕ ∈ U satisfies (19.1) and (19.2), the function ψ (x,t), given by
iμ t
ψ (x,t) = e− h̄ ϕ (x),
solves the well-known nonlinear Schrödinger equation given by
∂ ψ (x,t) h̄2
ih̄ = − ∇2 ψ (x,t) + 2α |ψ (x,t)|2 ψ (x,t) + V (x)ψ (x,t) in Ω × [0, ∞),
∂t 2m
with the boundary condition ψ = 0 on ∂ Ω × [0, ∞), so that

|ψ (x,t)|2 dx = c, ∀t ∈ [0, ∞).
Ω
Remark 19.1.1. About the references, we highlight that details on the Sobolev
spaces involved may be found in [1, 26]. Duality principles for related problems are
addressed in [13]. Also, an extensive study on Lagrange multipliers may be found in
[40, 47]. For the numerical results, details on finite difference schemes are presented
in [63]. Finally, details on related physics problems are developed in [4, 45, 46].
The corresponding primal variational formulation of the system above described

is given by the functional J : U → R = R ∪ {+∞}, where unless otherwise indicated,
we denote x = (x1 , x2 , x3 ) ∈ R3 , dx = dx1 dx2 dx3 and

h̄2 α
J(ϕ ) = ∇ϕ · ∇ϕ dx + |ϕ |4 dx
4m Ω 2 Ω

1
+ V (x)|ϕ |2 dx + Ind(ϕ ), (19.3)
2 Ω
where
0, if Ω |ϕ |2 dx = c,
Ind(ϕ ) = (19.4)
+∞, otherwise.
For such a nonlinear optimization problem, we have the following duality

principle.
Theorem 19.2.1. Redefine J : U → R by
J(ϕ ) = G1 (∇ϕ ) + G2 (|ϕ |2 ) + G3(ϕ ) + Ind(ϕ ),
where

h̄2
G1 (∇ϕ ) = ∇ϕ · ∇ϕ dx,
4m Ω

α
G2 (|ϕ |2 ) = |ϕ |4 dx,
2 Ω

1
G3 (ϕ ) = V (x)|ϕ |2 dx,
2 Ω
and, as above indicated,

0, if Ω |ϕ |2 dx = c,
Ind(ϕ ) = (19.5)
+∞, otherwise.
inf {J(ϕ )} ≥ sup {−J ∗(v∗ )},

ϕ ∈U v∗ ∈Y ∗
where Y = Y ∗ = L2 (Ω ),
J ∗ (v∗ ) = G∗2 (v∗ ) + F̃ ∗ (v∗ ),
G∗2 (v∗ ) = sup{v, v∗ L2 (Ω ) − G2 (v)}

v∈Y

1
= |v∗ |2 dx. (19.6)
2α Ω
Moreover,
F̃ ∗ (v∗ ) = sup {−|ϕ |2 , v∗ L2 (Ω )

ϕ ∈U
−G1 (∇ϕ ) − G3 (ϕ ) − Ind(ϕ )}. (19.7)
Finally, assume there exists (ϕ0 , μ0 ) ∈ U × R, such that
δϕ {Jμ0 (ϕ0 )} = θ ,
where
Jμ0 (ϕ ) = G1 (∇ϕ ) + G2 (|ϕ |2 ) + G3 (ϕ )

μ0
− |ϕ |2 dx − c , (19.8)
2 Ω
and so that
|ϕ0 |2 dx − c = 0.
Ω
Also, suppose that
J˜μ0 (ϕ ) ≥ 0, ∀ϕ ∈ U,
where
J˜μ0 (ϕ ) = |ϕ |2 , v∗0 L2 (Ω ) + G1(∇ϕ )

μ0
+G3 (ϕ ) − |ϕ |2 dx, (19.9)
2 Ω
and
∂ G2 (|ϕ0 |2 )
v∗0 =
∂v
= α |ϕ0 |2 . (19.10)
min{J(ϕ )} = J(ϕ0 ) = −J ∗ (v∗0 ) = max

∗ ∗
{−J ∗ (v∗ )}.
ϕ ∈U v ∈Y
Proof. Observe that
J(ϕ ) = −|ϕ |2 , v∗ L2 (Ω ) + G2 (|ϕ |2 )

+|ϕ |2 , v∗ L2 (Ω ) + G1 (∇ϕ )
+G3 (ϕ ) + Ind(ϕ )
≥ inf {−v, v∗ L2 (Ω ) + G2 (v)}
v∈Y
+ inf {|ϕ |2 , v∗ L2 (Ω ) + G1(∇ϕ )
ϕ ∈U
+G3 (ϕ ) + Ind(ϕ )}
= −G∗2 (v∗ ) − F̃ ∗ (v∗ )
= −J ∗ (v∗ ), (19.11)
∀ϕ ∈ U, v∗ ∈ Y ∗ . Thus,
inf {J(ϕ )} ≥ sup {−J ∗ (v∗ )}. (19.12)

ϕ ∈U v∗ ∈Y ∗
Finally, from the additional hypotheses, we may infer that
− F̃ ∗ (v∗0 ) = |ϕ0 |2 , v∗0 L2 (Ω ) + G1(∇ϕ0 )

μ0
+G3 (ϕ0 ) − |ϕ0 |2 dx − c , (19.13)
2 Ω
19.3 Numerical Examples 469
that is,
− F̃ ∗ (v∗0 ) = |ϕ0 |2 , v∗0 L2 (Ω ) + G1 (∇ϕ0 )

+G3 (ϕ0 ) + Ind(ϕ0). (19.14)
From the definition of v∗0 we get
G∗2 (v∗0 ) = |ϕ0 |2 , v∗0 L2 (Ω ) − G2(|ϕ0 |2 ),
so that from these last two equations, we obtain
−G∗2 (v∗0 ) − F̃ ∗ (v∗0 ) = G1 (∇ϕ0 ) + G2 (|ϕ0 |2 ) + G3(ϕ0 ) + Ind(ϕ0 ).
Hence,
−J ∗ (v∗0 ) = J(ϕ0 ).
From this and (19.12) the proof is complete.
19.3 Numerical Examples
Before presenting the numerical examples, we introduce the following remark.
Remark 19.3.1. Consider the function f : R2 \ {(0, 0)} → R2 , where

x y
f (x, y) = , .
x2 + y2 x2 + y2
We recall that, for the Euclidean norm in question,

x y
∇ f (x, y) = ∇ ,∇ ,
x2 + y2 x2 + y2
so that
2 2 2
3
3

∇ f (x, y) = 4∇
x y
+ ∇ .
x +y
2 2 x +y
2 2
We may compute
1
∇ f (x, y) = . (19.15)
x + y2
2
A similar result is valid for a N > 2-dimensional vector representing the dis-
cretized version of a function.
Thus denoting φ = (x1 , . . . , xN ) ∈ RN we may symbolically write

φ
δ = 1 . (19.16)
φ φ
We present numerical results, first, for the following one-dimensional closely

related eigenvalue problem with α = 1/2
1 d 2 ϕ (x)
− + ϕ 3 (x) + V (x)ϕ (x) − μϕ (x) = 0, in Ω = [0, 1],
λ 2 dx2
ϕ (0) = ϕ (1) = 0.
Here μ ∈ R is such that

ϕ 2 (x) dx = 1.
Ω
Moreover, the potential V (x) is given by

0, if x ∈ (0, 1),
V (x) = (19.17)
+∞, otherwise.
19.3.1 The Algorithm
We denote by I(ϕ ) the diagonal matrix which the diagonal is the vector ϕ .
Furthermore, if ϕ ∈ RN is such that
ϕ = [ϕ1 , . . . , ϕN ],
we denote
ϕ p = [ϕ1p , . . . , ϕNp ], ∀p ∈ N.
The above equation, after discretization, in finite differences may be expressed by
1
− M2 ϕ + I(ϕ 2 )ϕ − μϕ = 0,
λ 2d 2
where ⎡ ⎤
−2 1 0 0 ... 0
⎢ 1 −2 1 0 ... 0 ⎥
⎢ ⎥
⎢ 0 1 −2 1 0.. 0 ⎥
M2 = ⎢
⎢... ... ... ...
⎥. (19.18)
⎢ ... ... ⎥
⎥
⎣ 0 0 ... 1 −2 1 ⎦
0 0 ... ... 1 −2
Here M2 is a N × N matrix, where N is the number of nodes and d = 1/N.
In the next lines we describe the algorithm:

1. Set n = 1.
2. Choose ϕ̃1 ∈ W 1,2 (Ω ), so that ϕ̃1 = 0, a.e. in Ω .
3. Define

∗ |ϕ̃n |2
vn = α .
ϕ̃n 2
4. Obtain ϕ̃n+1 ∈ U as the solution of the linear equation
1 ϕ̃n
− M2 ϕ̃n+1 + 2I(v∗n)ϕ̃n+1 − = 0.
λ 2d 2 ϕ̃n
5. Set n → n + 1, and go to step 3, up to the satisfaction of an appropriate conver-

gence criterion.
Here we present a rather informal discussion about the algorithm convergence.
First, observe that the equation in question may be written as
5 6
ϕ̃n 2 ϕ̃n 2 2
−M2 ϕ̃n+1 + I ϕ̃n+1 d 2 λ 2 − d λ = 0,
ϕ̃n ϕ̃n
so that
5 6
ϕ̃n 2 2 2

−M2 + I d λ ϕ̃n+1 ≥ d 2 λ 2 ,
ϕ̃n
that is,
5 6
1 1 ϕ̃n 2 2 2
≤ −M2 + I d λ , ∀n ∈ N. (19.19)
ϕ̃n+1 d 2 λ 2 ϕ̃n
Also, we may denote

ϕ̃n+1 = G(ϕ̃n ),
where 5 2 6 −1
ϕ̃n ϕ̃n
G(ϕ̃n ) = −M2 + I d λ
2 2
· d2λ 2.
ϕ̃n ϕ̃n
Therefore
ϕ̃n+1 = G(ϕ̃n )
and
ϕ̃n+2 = G(ϕ̃n+1 ),
so that, denoting for t ∈ [0, 1],
(ϕ̃t )n = t ϕ̃n + (1 − t)ϕ̃n+1

from the generalized mean value inequality, we obtain

ϕ̃n+2 − ϕ̃n+1 ≤ sup G ((ϕ̃t )n ) ϕ̃n+1 − ϕ̃n , (19.20)
t∈[0,1]
Computing the derivative and from (19.19), we get the estimate

5 −2
2 6 5 26
ϕ̃ ϕ̃

G (ϕ̃n ) ≤ −M2 − I
n
d λ · −M2 + I
2 2 n
d λ
2 2
ϕ̃n ϕ̃n

ϕ̃n
×δ ϕ̃n d λ
2 2
5 6
ϕ̃n 2 2 2

≤ −M2 − I d λ
ϕ̃n
5 6 −2
2
ϕ̃n
× −M 2 + I d λ
2 2

ϕ̃n

1
× d2λ 2
ϕ̃n
5 6
ϕ̃n 2 2 2

≤ −M2 − I d λ
ϕ̃n
5 6 −2
2
ϕ̃

× −M2 + I
n
d λ
2 2
ϕ̃
n
5 6
ϕ̃n 2 2 2

× −M2 + I d λ , (19.21)
ϕ̃n
where we have used equality (19.16), as it is indicated at Remark 19.3.1, that is, for
a vector ϕ̃n ∈ RN ,

ϕ̃n
δ = 1 .
ϕ̃n ϕ̃n
Recalling that −M2 is positive definite, at this point, we assume that for an ap-
propriate choice of d, there exist n0 ∈ N and 0 < β < 1 such that
5 6
ϕ̃n 2 2 2

−M2 − I d λ
ϕ̃n
5 −2
2 6
ϕ̃

× −M2 + I
n
d λ
2 2
ϕ̃n

5 6
ϕ̃n 2 2 2

× −M2 + I d λ ≤ β < 1, ∀n > n0 .
ϕ̃n
We have thus obtained
G (ϕ̃n ) ≤ β , ∀n > n0 , (19.22)
where 0 < β < 1.

As we have mentioned above, this is just an informal discussion.
Suppose that from (19.22) we may obtain that (in fact we do not provide details
here)
G ((ϕ̃t )n ) ≤ β , ∀t ∈ [0, 1], n > n0 . (19.23)
From this and (19.20), we have
ϕ̃n+2 − ϕ̃n+1 ≤ β ϕ̃n+1 − ϕ̃n , ∀n > n0 .
So, we may infer that

{ϕ̃n }
is a Cauchy sequence, that is, it is strongly converging to some ϕ̃0 .
Thus, we may write
5 6
ϕ̃0 2 ϕ̃0 2 2
−M2 ϕ̃0 + I ϕ̃0 d 2 λ 2 − d λ = 0,
ϕ̃0 ϕ̃0
so that denoting
ϕ̃0
ϕ0 = ,
ϕ̃0
we get
ϕ0 2 2
−M2 ϕ0 + I(ϕ02 )ϕ0 d 2 λ 2 − d λ = 0.
ϕ̃0
That is,
1
− M2 ϕ0 + I(ϕ02 )ϕ0 − μ0 ϕ0 = 0,
λ 2d 2
where
1
μ0 = .
ϕ̃0
Clearly we have

|ϕ0 |2 dx = 1.
Ω
Remark 19.3.2. Observe that we have not formally proven that the algorithm con-
verges. However, from the analysis developed, we have a strong indication that such
a convergence, under mild hypotheses, holds. Indeed, all numerical examples so far
worked have converged very easily.
An analogous analysis is valid for the two-dimensional case.
About the numerical results for this one-dimensional example, the ground state
is given in Fig. 19.1. Other solution with a greater eigenvalue is plotted in Fig. 19.2.
We highlight the numerical results obtained perfectly agree with the well-known
analytic solutions for this one-dimensional model. See [17] for details.
Finally, we present an analogous two-dimensional example, that is, we develop
results for the eigenvalue problem
1.4
1.2
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
Fig. 19.1 Ground state wave function ϕ (x) for λ = 25
1.5
0.5
−0.5
−1
−1.5
0 0.2 0.4 0.6 0.8 1
Fig. 19.2 Other wave function ϕ (x) for λ = 25

1 2
− ∇ ϕ (x, y) + ϕ 3 (x, y) + V (x, y)ϕ (x, y) − μϕ (x, y) = 0, in Ω = [0, 1] × [0, 1],
λ2
with the boundary condition
ϕ = 0 on ∂ Ω .
Here μ ∈ R is such that

ϕ 2 (x, y) dxdy = 1.
Ω
Moreover, the potential V (x, y) is given by

0, if (x, y) ∈ (0, 1) × (0, 1),
V (x, y) = (19.24)
+∞, otherwise.
For such a two-dimensional case, the ground state for λ = 1 is given in Fig. 19.3.
The ground state, for λ = 25, is given in Fig. 19.4.
1.5
0.5
0
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 19.3 Ground state wave function ϕ (x, y) for λ = 1
Other solution, also for λ = 25, is plotted in Fig. 19.5.

1.5
0.5
0
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 19.4 Ground state wave function ϕ (x, y) for λ = 25
1.5
1
0.5
0
−0.5
−1
−1.5
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 19.5 Other wave function ϕ (x, y) for λ = 25
19.4 Conclusion
In this chapter we develop a concave dual variational formulation for a nonlin-

ear model in quantum mechanics. In practice, the results may be applied to obtain
the ground state for the Schrödinger equation, which may be verified through the
optimality conditions obtained. Finally, we emphasize the approach here developed
may be applied to many other situations, such as for problems in quantum mechan-
ics involving a large number of variables.
Chapter 20
Duality Applied to the Optimal Design
in Elasticity
Fabio Botelho and Alexandre Molter
20.1 Introduction
In this chapter we develop duality for an optimal design problem in elasticity. We

start by describing the primal formulation.
Consider Ω ⊂ R3 , an open, bounded, and connected set with a regular boundary
denoted by ∂ Ω = Γ0 ∪ Γ1 , where Γ0 ∩ Γ1 = 0. / By a regular boundary ∂ Ω we mean
regularity enough so that the Sobolev imbedding theorem and relating results, the
trace theorem, and the standard Gauss–Green formulas of integration by parts hold.
Here Ω stands for the volume of an elastic solid under the action of a load
P ∈ L2 (Ω ; R3 ). We assume |Γ0 | > 0, where |Γ0 | denotes the Lebesgue measure of
Γ0 . Also, we denote by n the outward normal to the solid surface. The field of dis-
placements is denoted by u = (u1 , u2 , u3 ) ∈ U, where
U = {u ∈ W 1,2 (Ω ; R3 ) | u = (0, 0, 0) on Γ0 }. (20.1)
The strain tensor, given by e = {ei j }, is defined by
1
ei j (u) = (ui, j + u j,i ). (20.2)
2
Denoting by Hi1jkl and Hi0jkl two symmetric positive definite fourth-order constant
tensors, first we define the optimization problem of minimizing J(u,t) where

1
J(u,t) = tHi1jkl ei j (u)ekl (u) + (1 − t)Hi0jkl ei j (u)ekl (u) dx,
2 Ω
F. Botelho ()
Department of Mathematics and Statistics, Federal University of Pelotas, Pelotas, RS-Brazil
A. Molter
Department of Mathematics and Statistics, Federal University of Pelotas, Brazil
478 20 Duality Applied to the Optimal Design in Elasticity
subject to

tHi1jkl ekl (u) + (1 − t)Hi0jkl ekl (u) + Pi = 0, in Ω , (20.3)
,j

tHi1jkl ekl (u) + (1 − t)Hi0jkl ekl (u) n j = 0, on Γ1 , (20.4)
∀i ∈ {1, 2, 3}, u ∈ U, t ∈ {0, 1}, a.e. in Ω , and

t dx ≤ t1 |Ω |. (20.5)
Ω
Here 0 < t1 < 1 and |Ω | denote the Lebesgue measure of Ω .

We relax such an original problem now allowing the parameter t to assume values
in [0, 1] on Ω . Also, a penalization constant p ≥ 1 is introduced in order to replace
t by t p in the energy functional, with the objective of approximating the resulting
design variable to the set {0, 1}. A standard value for p would be p = 3, for example.
Thus, we rewrite J : U × B → R = R ∪ {+∞} as

Hi jkl (t p )
J(u,t) = ei j (u)ekl (u) dx + Ind(u,t), (20.6)
Ω 2
where
Hi jkl (t p ) = t p Hi1jkl + (1 − t p)Hi0jkl ,
Ind(u,t) = Ind1(u,t) + Ind2(u,t),

0, if (Hi jkl (t p )ekl (u)), j + Pi = 0, in Ω , ∀i ∈ {1, 2, 3},
Ind1 (u,t) =
+∞, otherwise,

0, if (Hi jkl (t p )ekl (u))n j = 0, on Γ1 , ∀i ∈ {1, 2, 3}
Ind2 (u,t) =
+∞, otherwise,
and

B = t measurable | 0 ≤ t(x) ≤ 1, a.e. in Ω , t(x) dx ≤ t1 |Ω | .
Ω
Also Y = Y ∗ = L2 (Ω ; R3×3 ) and from now on we denote
{H i jkl (t p )} = {Hi jkl (t p )}−1 .
20.2 On the Duality Principle
In this section we develop a duality principle for the problem in question.

Similar problems are addressed in [3, 8, 13, 14, 29]. Details on general Sobolev
spaces theory may be found in [1, 26]. We start with the next theorem. It is worth
20.2 On the Duality Principle 479
emphasizing the result indicated in (20.8) is well known, whereas the chain of equal-
ities indicated in (20.15) and (20.17) we believe they are both new for the relaxed
problem, that is, for the case p > 1 where the min–max theorem does not apply.
Theorem 20.2.1. Considering the above expressed function Ind(u,t), defining

Hi jkl (t p )
G(e(u),t) = ei j (u)ekl (u) dx, (20.7)
Ω 2
and
J(u,t) = G(e(u),t) + Ind(u,t), ∀(u,t) ∈ U × B,
we have that
inf {J(u,t)} = inf {G̃∗ (σ ,t)}

(u,t)∈U×B (σ ,t)∈A∗ ×B
≥ sup{−J (û)},˜∗ (20.8)

û∈U
where
− J˜∗(û) = inf {−G(e(û),t) + û, PL2 } , ∀û ∈ U. (20.9)

t∈B
Under such definitions, there exists û0 ∈ U such that
− J˜∗(û0 ) = max{−J˜∗ (û)}. (20.10)

u∈U
Moreover,
G̃∗ (σ ,t) = sup{v, σ L2 − G(v,t)}

v∈Y

1
= H i jkl (t p )σi j σkl dx, (20.11)
2 Ω
and
A∗ = {σ ∈ Y ∗ | σi j, j + Pi = 0, in Ω , σi j n j = 0 on Γ1 }.
Also, the following representation holds:

1
G(e(û),t) = sup − Hi jkl (t )vi j vkl dx + ei j (û), Hi jkl (t )vkl L2
p p
v∈Y 2 Ω

1
= Hi jkl (t p )ei j (û)ekl (û) dx. (20.12)
2 Ω
Assume there exists (σ0 , t˜0 , ũ0 ) ∈ A∗ × B × U such that
δ {Jλ∗ (σ0 , t˜0 , ũ0 )} = θ ,

where

Jλ∗ (σ ,t, û) = G̃∗ (σ ,t) + ûi , −σi j, j − PiL2 + ûi σi j n j dΓ
Γ1

+ λ1 (t 2 − t) dx + λ2 t dx − t1 |Ω | . (20.13)
Ω Ω
Here (λ1 , λ2 ) = λ are appropriate Lagrange multipliers.

Furthermore, suppose
− J˜∗(ũ0 ) = inf {−G(e(ũ0 ),t) + ũ0, PL2 }

t∈B
= −G(e(ũ0 ), t˜0 ) + ũ0 , PL2 . (20.14)
Under such hypotheses we have
min {J(u,t)} = J(ũ0 , t˜0 )

(u,t)∈U×B
= G̃∗ (σ0 , t˜0 )

= min {G̃∗ (σ ,t)}
(σ ,t)∈A∗ ×B
= max{−J˜∗ (û)}
û∈U
= −J˜∗ (ũ0 ), (20.15)
where
σ0 = {(σ0 )i j } = {Hi jkl (t˜0p )ekl (ũ0 )}.
Finally, considering the same problem after discretization,
for the optimal û0 ∈ U satisfying (20.10), assume t0 ∈ B defined by
− J˜∗(û0 ) = inf {−G(e(û0 ),t) + û0, PL2 }

t∈B
= −G(e(û0 ),t0 ) + û0 , PL2 , (20.16)
is such that (û0 ,t0 ) are also such that the hypotheses of Corollary 11.1 to hold.
min {J(u,t)} = J(û0 ,t0 ) = −J˜∗ (û0 ) = max{−J˜∗ (û)} (20.17)

(u,t)∈U×B û∈U
We emphasize to have denoted L2 (Ω ), L2 (Ω ; R3 ), or L2 (Ω ; R3×3 ) simply by L2 ,

as their meaning is clear.
Proof. First observe that
inf {J(u,t)}
(u,t)∈U×B
= inf{ inf {J(u,t)}}
t∈B u∈U
= inf{ inf {G(e(u),t) + Ind(u,t)}}
t∈B u∈U

= inf sup inf {G(e(u),t) + ûi, (Hi jkl (t p )ekl (u)), j + Pi L2
t∈B û∈U u∈U

− ûi Hi jkl (t )ekl (u)n j dΓ }
p
,
Γ1
so that
inf {J(u,t)}
(u,t)∈U×B

" #
= inf sup inf G(e(u),t) − ei j (û), Hi jkl (t p )ekl (u)L2 + ûi , Pi L2
t∈B û∈U u∈U

= inf sup {−G(e(û),t) + ûi , Pi L2 }
t∈B û∈U

" #
= inf inf∗ G̃∗ (σ ,t)
t∈B σ ∈A
= inf {G̃∗ (σ ,t)}.

(σ ,t)∈A∗ ×B
(20.18)
On the other hand,
J(u,t) = G(e(u),t) + Ind(u,t)

≥ G(e(u),t) + ûi , (Hi jkl (t p )ekl (u)), j + Pi L2

− ûi Hi jkl (t p )ekl (u)n j dΓ
Γ1
= G(e(u),t) − ei j (û), Hi jkl (t p )ekl (u)L2 + ûi , Pi L2
" #
≥ inf G(v,t) − ei j (û), Hi jkl (t p )vkl L2 + ûi , Pi L2
v∈Y
= −G(e(û),t) + ûi , Pi L2
≥ inf {−G(e(û),t) + ûi , Pi L2 }
t∈B
= −J˜∗ (û), (20.19)
∀u, û ∈ U.
Therefore,
inf {J(u,t)} ≥ sup{−J˜∗ (û)}, (20.20)

(u,t)∈U×B û∈U
so that from (20.18) and (20.19) we obtain
inf {J(u,t)} = inf {G̃∗ (σ ,t)}

(u,t)∈U×B (σ ,t)∈A∗ ×B
˜∗
≥ sup{−J (û)}, (20.21)
û∈U
From the hypotheses, there exists (σ0 , t˜0 , ũ0 ) ∈ A∗ × B × U such that
δ {Jλ∗ (σ0 , t˜0 , ũ0 )} = θ ,
where

Jλ∗ (σ ,t, û) = G̃∗ (σ ,t) + ûi , −σi j, j − PiL2 + ûi σi j n j dΓ
Γ1

+ λ1 (t 2 − t) dx + λ2 t dx − t1 |Ω | , (20.22)
Ω Ω
where (λ1 , λ2 ) = λ are appropriate Lagrange multipliers and so that
− J˜∗(ũ0 ) = inf {−G(e(ũ0 ),t) + ũ0, PL2 }

t∈B
= −G(e(ũ0 ), t˜0 ) + ũ0 , PL2 . (20.23)
By this last equation, we get
− J˜∗(ũ0 ) = −G(e(ũ0 ), t˜0 ) + ũ0, PL2

1 p
=− Hi jkl (t˜0 )ei j (ũ0 )ekl (ũ0 ) dx + ũ0, PL2
2 Ω

1
= Hi jkl (t˜0p )ei j (ũ0 )ekl (ũ0 ) dx
2 Ω
= G(e(ũ0 ), t˜0 ) + Ind(ũ0, t˜0 )
= J(ũ0 , t˜0 )
= −G(e(ũ0 ), t˜0 ) + ũ0, PL2
= −G(e(ũ0 ), t˜0 ) + (ũ0)i , −(σ0 )i j, j L2
= −G(e(ũ0 ), t˜0 ) + ei j (ũ0 ), (σ0 )i j L2
= G̃∗ (σ0 , t˜0 ). (20.24)
From this and (20.21), we have proven (20.15).

For proving (20.10) and (20.17), observe that from Korn’s inequality (in fact it

may be shown that u U and |e(u)|0 = ( Ω ei j (u)ei j (u) dx)1/2 are equivalent norms;
for details see [21]) we have
−J˜∗ (û) → −∞, as û U → ∞.
Furthermore, −J˜∗ (û) is concave since it is the infimum of a family of concave func-
tions. From the coerciveness above verified, if {ûn} is a maximizing sequence, there
exists K1 > 0 such that
ûn U ≤ K1 , ∀n ∈ N.
Hence, there exists û0 ∈ U, such that up to a subsequence not relabeled, we have
e(ûn ) e(û0 ), weakly in L2 .

ûn → û0 , strongly in L2 ,

as n → ∞.
From the concavity and weak upper semicontinuity of −J˜∗ (û) we get
sup{−J˜∗(û)} = lim sup{−J˜∗ (ûn )} ≤ −J˜∗ (û0 ).

û∈U n→∞
Thus,
max{−J˜∗ (û)} = −J˜∗ (û0 ).
û∈U
At this point and on we consider the problem in question after discretization.

Recall we have assumed that t0 ∈ B such that
− J˜∗(û0 ) = inf {−G(e(û0 ),t) + û0, PL2 }

t∈B
= −G(e(û0 ),t0 ) + û0 , PL2 , (20.25)
is such that (û0 ,t0 ) are also such that the hypotheses of Corollary 11.1 are satisfied.
From such a corollary, the optimal equation
δ J˜∗ (û0 ) = θ
stands for
∂ {G(e(û0 ),t0 ) − û0 , PL2 }
= θ.
∂u
Hence,
(Hi jkl (t0p )ekl (û0 )), j + Pi = 0 in Ω , (20.26)
and
(Hi jkl (t0p )ekl (û0 ))n j = 0 on Γ1 , (20.27)
∀i ∈ {1, 2, 3}.
By (20.26) and (20.27), we obtain

1
− J˜∗(û0 ) = − Hi jkl (t0p )ei j (û0 )ekl (û0 ) dx + û0, PL2
2 Ω

1
= Hi jkl (t0p )ei j (û0 )ekl (û0 ) dx + Ind(û0,t0 )
2 Ω
= G(e(û0 ),t0 ) + Ind(û0,t0 ). (20.28)
Observe that
G(e(û0 ),t0 ) + Ind(û0,t0 ) ≥ inf {G(e(u),t) + Ind(u,t)}

(u,t)∈U×B
≥ −J˜∗ (û0 ). (20.29)

By (20.28) and (20.29), we obtain
G(e(û0 ),t0 ) + Ind(û0,t0 ) = inf {G(e(u),t) + Ind(u,t)}

(u,t)∈U×B
= −J˜∗ (û0 ), (20.30)
so that from this and (20.20) we finally have
min {J(u,t)} = J(û0 ,t0 )

(u,t)∈U×B
= G(e(û0 ),t0 ) + Ind(û0,t0 )
= −J˜∗ (û0 )
= max{−J˜∗ (û)}. (20.31)
û∈U
Remark 20.2.2. After discretization, for the case p = 1, the chain of equalities in-
dicated in (20.31) may be easily obtained from the min–max theorem. However,
to improve the numerical results in practical situations, it is desirable to use p > 1
and in such a case the min–max theorem does not apply (p = 3, e.g., is a standard
choice).
As an example, we present the problem of finding the optimal plate thickness

distribution relating the structural inner work minimization. For, consider a plate
which the middle surface is denoted by Ω ⊂ R2 , where Ω is an open, bounded,
connected set with a sufficiently regular boundary denoted by ∂ Ω . As mentioned
above the design variable, the plate thickness h(x), is such that h0 ≤ h(x) ≤ h1 ,
where x = (x1 , x2 ) ∈ Ω ⊂ R2 . The field of normal displacements to Ω , due to a
external load P ∈ L2 (Ω ), is denoted by w : Ω → R.
Such an optimization problem is represented by the minimization of J : U × B →
R = R ∪ {+∞}, where

Hαβ λ μ (t)
J(w,t) = w,αβ w,λ μ dx + Ind(w,t), (20.32)
Ω 2
Ind(w,t) = Ind1 (w,t) + Ind2(w,t),

0, if (Hαβ λ μ (t)w,λ μ ),αβ − P = 0, in Ω ,
Ind1(w,t) =
+∞, otherwise,

0, if Hαβ λ μ (t)w,λ μ nα nβ = 0, on ∂ Ω ,
Ind2 (w,t) =
+∞, otherwise,
and
B = {t measurable | t(x) ∈ [0, 1], a.e. in Ω

and (th31 + (1 − t)h30) dx ≤ t1 h31 |Ω |}. (20.33)
Ω
Moreover, |Ω | denotes the Lebesgue measure of Ω and

" #
U = w ∈ W 2,2 (Ω ) | w = 0 on ∂ Ω . (20.34)
We develop numerical results for the particular case, where t1 = 0.52,
Hαβ λ μ (t) = H(t) = ĥ(t)E, (20.35)
ĥ(t) = th31 + (1 − t)h30, h1 = 0.1, h0 = 10−4, and E = 107 , with the units related to
the international system (we emphasize to denote x = (x1 , x2 )). Observe that 0 ≤
t(x) ≤ 1, a.e. in Ω . Similarly, as in the last section (see [13] for details), we may
obtain the following duality principle (here in a slightly different version):

1
inf {J(w,t)} = inf H αβ λ μ (t)Mαβ Mλ μ dx ,
(w,t)∈U×B (t,{Mαβ })∈B×D∗ 2 Ω
where
{H αβ λ μ (t)} = {Hαβ λ μ (t)}−1 ,
and
D∗ = {{Mαβ } ∈ Y ∗ | Mαβ ,αβ + P = 0, in Ω , Mαβ nα nβ = 0 on ∂ Ω }.
We have computed the dual problem for Ω = [0, 1] × [0, 1] and a vertical load acting
on the plate given by P(x) = 10, 000, obtaining the results indicated in the respective
Figs. 20.1 and 20.2, for t0 (x) and w0 (x). We emphasize they are critical points, that
is, just candidates to optimal points. Observe that for the concerned critical point

Hαβ λ μ (t0 )
J(w0 ,t0 ) = (w0 ),αβ (w0 ),λ μ dx + Ind(w0,t0 )
Ω 2

Hαβ λ μ (t0 )
= (w0 ),αβ (w0 ),λ μ dx
Ω 2

1
= H (t0 )(M0 )αβ (M0 )λ μ dx
2 Ω αβ λ μ
= G̃∗ (σ0 ,t0 ), (20.36)
where the moments {(M0 )αβ } are given by
{(M0 )αβ } = {−Hαβ λ μ (t0 )(w0 ),λ μ }.

0.8
0.6
0.4
0.2
0
1
1
0.8
0.5 0.6
0.4
0.2
0 0
Fig. 20.1 t0 (x)-function relating the plate thickness distribution
x 10−3
2
1.5
0.5
0
1
1
0.5 0.8
0.6
0.4
0.2
0 0
Fig. 20.2 w0 (x)-field of displacements
20.4 Another Numerical Example
In the paper [60] Sigmund presents an algorithm for shape optimization. Such
an algorithm refers to find critical points of the Lagrangian functional Lλ (slightly
changed considering the development established in this text) given by

Lλ (u, û,t) = u, PL2 − Hi jkl (t p )ei j (u)ekl (û) dx + û, PL2
Ω

+ λ1+ (t(x) − 1) dx + λ1− (tmin − t(x)) dx
Ω
Ω
+λ2 t(x) dx − t1|Ω | . (20.37)
Ω
20.4 Another Numerical Example 487
Dual
0.5
0.4
0.3
m
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m
Fig. 20.3 The cantilever beam density through the dual formulation, t1 = 0.50
Primal
0.5
0.4
m
0.3
F
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m
Fig. 20.4 The cantilever beam density through the primal formulation, t1 = 0.50
Dual F
0.2
0.18
0.16
0.14
0.12
0.1
m
0.08
0.06
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m
Fig. 20.5 Clamped beam density through the dual formulation, t1 = 0.50
Primal F
0.2
0.18
0.16
0.14
0.12
0.1
m
0.08
0.06
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m
Fig. 20.6 Clamped beam density through the primal formulation, t1 = 0.50
20.4 Another Numerical Example 489
Dual F
0.2
0.18
0.16
0.14
0.12
0.1
m
0.08
0.06
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m
Fig. 20.7 Simply supported beam density through the dual formulation, t1 = 0.50
Primal F
0.2
0.18
0.16
0.14
0.12
0.1
m
0.08
0.06
0.04
0.02
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m
Fig. 20.8 Simply supported beam density through the primal formulation, t1 = 0.50
We made a change in the 99-line Sigmund algorithm, now updating the density t
through a critical point of the dual functional. The optimal equation would be

δt H i jkl (t p )σi j σkl dx
Ω

+ λ1+ (t(x) − 1) dx + λ1− (tmin − t(x)) dx
Ω Ω

+λ2 t(x) dx − t1 |Ω | = θ. (20.38)
Ω
For a fixed σ we update t through this last equation and update σ , for a fixed t,
through the equilibrium equation.
In a more practical fashion, concerning the 99-line finite element algorithm and
respective notation, the heuristic equation for each element through which te is up-
dated up to the OC function, that is,
η
p(te ) p−1 uTe K0 ue
te = te
new
, (20.39)
λ2 ∂∂ Vte
is replaced by

∂ (tep K0 ue )T (tenew )−p K0−1 (tep K0 ue )
+ λ2 = 0, (20.40)
∂ tenew
so that
1
p(te2p )uTe K0 ue
1+p
tenew = .
λ2
Hence,
1
p(tep−1)uTe K0 ue
1+p
tenew = te , (20.41)
λ2
It is worth emphasizing in such a procedure we must have 0 < te ,tenew < 1. To in-
clude the general case we have used an analogous OC function as found in the
original Sigmund algorithm. We may observe that Eqs. (20.41) and (20.39) are very
similar, so that in fact this last heuristic equation (20.39) almost corresponds, for
different values of η , to the Euler–Lagrange equations for the dual problem, which,
to some extent, justify the algorithm convergence. Based on the 99-line O. Sigmund
software with inclusion of the mentioned change (which corresponds, as above in-
dicated, to optimization through the dual formulation), the second author of this
chapter (Alexandre Molter) designed a code suitable for the different situations here
addressed, through which we have developed numerical examples.
20.5 Conclusion 491
We develop examples concerning a two-dimensional elastic beam with different

types of boundary conditions and in all cases we have initially obtained a good
approximation for the restriction t ∈ {0, 1}, a.e.. However, it is worth mentioning,
for the figures in question, in the gray scale obtained in [0, 1] we have post-processed
such figures qualitatively by setting t ≡ 0 if for the original result 0 ≤ t < 0.5 and
setting t ≡ 1 if 0.5 ≤ t ≤ 1.
The numerical values for such examples are E1 = 210 ∗ 109 (Young modulus),
E0 E1 (for simulating absence of material), ν = 0.3 (Poisson coefficient), and
F = 10, 000, where the position of applied F is indicated in each figure, the volume
fraction is 0.5, and units refer to the international system. Please see Figs. 20.3,
20.5, and 20.7 for the solutions through the dual formulation and Figs. 20.4, 20.6,
and 20.8 for solutions through the primal one.
We highlight the beam dimensions are:
1 m × 0.6 m for the cantilever ones indicated in Figs. 20.3 and 20.4.
1 m × 0.4 m for the beams clamped at both extremals indicated in Figs. 20.5 and
20.6.
1 m × 0.2 m for the beams simply supported at both extremals (in fact boundary
conditions refer just to zero vertical displacement at x = 0 and x = 1.0) indicated
in Figs. 20.7 and 20.8. Finally, we may observe the results are qualitatively similar
through the dual and primal formulations. The main differences are found just in
Figs. 20.5 and 20.6. For related results, see [8, 49].
20.5 Conclusion
In this chapter we have developed a dual variational formulation for an optimal

design problem in elasticity. The infimum in t indicated in the dual formulation rep-
resents the structure search for stiffness in the optimization process, which implies
the internal work minimization. In some cases the primal problem, before discretiza-
tion, may not have solutions, so that the solution of dual problem is a weak cluster
point of minimizing sequences for the primal one. After discretization, it has been
established conditions for the duality gap between the primal and dual problems be
zero. We expect the results obtained can be used as engineering project tools.
Chapter 21
Duality Applied to Micro-Magnetism
21.1 Introduction
In this chapter we develop dual variational formulations for models in micro-

magnetism. For the primal formulation we refer to references [10, 43, 44, 51] for
details. In particular we refer to the original results presented in [10], emphasizing
that the present work is their natural continuation and extension.
Let Ω ⊂ R3 be an open bounded set with a finite Lebesgue measure and a regular
boundary denoted by ∂ Ω . By a regular boundary ∂ Ω we mean regularity enough so
that the Sobolev imbedding theorem and relating results, the trace theorem and the
standard Gauss–Green formulas of integration by parts to hold. The corresponding
outward normal is denoted by n. Also, we denote by θ either the zero vector in R3
or the zero in an appropriate function space.
Under such assumptions and notations, consider problem of finding the magne-
tization m : Ω → R3 , which minimizes the functional

α
J(m, f ) = |∇m|22 dx + ϕ (m(x)) dx − H(x) · m dx
2 Ω Ω Ω

1
+ | f (x)|22 dx, (21.1)
2 R3
where
m = (m1 , m2 , m3 ) ∈ W 1,2 (Ω ; R3 ) ≡ Y1 , |m(x)|2 = 1, a.e. in Ω (21.2)
and f ∈ L2 (R3 ; R3 ) ≡ Y2 is the unique field determined by the simplified Maxwell’s

equations
Curl( f ) = θ , div(− f + mχΩ ) = 0, a.e. in R3 . (21.3)
Here H ∈ L2 (Ω ; R3 ) is a known external field and χΩ is a function defined by
494 21 Duality Applied to Micro-Magnetism

1, if x ∈ Ω ,
χΩ (x) = (21.4)
0, otherwise.
The term
α
|∇m|22 dx
2 Ω
is called the exchange energy, where
2
3 3
3
|m|2 = 4 ∑ m2k
k=1
and
3
|∇m|22 = ∑ |∇mk |22 .
k=1
Finally, ϕ (m) represents the anisotropic contribution and is given by a multi-well

functional whose minima establish the preferred directions of magnetization.
Remark 21.1.1. Here are some brief comments on the references. Relating and sim-
ilar problems are addressed in [7, 11, 13, 14, 55, 56]. The basic results on convex
and variational analysis used in this text may found in [13, 24, 25, 40, 47]. Finally,
an extensive study on Sobolev spaces may be found in [1].
Remark 21.1.2. At some points of our analysis we refer to the problems in ques-
tion after discretization. In such a case we refer to their approximations in a finite
element or finite differences context.
21.2 Summary of Results for the Hard Uniaxial Case
We consider first the case of a uniaxial material with no exchange energy. That
is, α = 0 and ϕ (m) = β (1 − |m · e|).
Observe that
ϕ (m) = min{β (1 + m · e), β (1 − m · e)}
where β > 0 and e ∈ R3 is a unit vector.
In the next lines we present the primal formulation and related duality principle.
Define J : Y1 × Y2 × B → R by
J(m, f ,t) = G1 (m, f ,t) + G2 ( f ),

G1 (m, f ,t) = (tg1 (m) + (1 − t)g2(m)) dx
Ω
+Ind0(m) + Ind1( f ) + Ind2(m, f )

− H(x) · m dx, (21.5)
Ω
21.2 Summary of Results for the Hard Uniaxial Case 495
and
1
G2 ( f ) = | f (x)|22 dx.
2 R3
Also,
g1 (m) = β (1 + m · e),
g2 (m) = β (1 − m · e),

0, if |m(x)|2 = 1 a.e. in Ω ,
Ind0 (m) =
+∞, otherwise,

0, if div(− f + mχΩ ) = 0 a.e. in R3 ,
Ind1 (m, f ) =
+∞, otherwise,
and
0, if Curl( f ) = θ , a.e. in R3 ,
Ind2( f ) =
+∞, otherwise.
Observe that as abovementioned,

ϕ (m) dx = β (1 − |m · e|) dx
Ω Ω

= min (tg1 (m) + (1 − t)g2(m)) dx . (21.6)
t∈B Ω
Under additional assumptions to be specified, we have
inf {J(m, f ,t)} = sup {−J ∗ (λ )} (21.7)

(m, f ,t)∈Y1 ×Y2 ×B λ ∈Ŷ ∗
where
J ∗ (λ ) = G̃∗1 (λ ) + G∗2 (λ ) − β dx,
Ω
⎧ ⎫
⎨ 2 1/2 ⎬
3
∂ λ2
G̃∗1 (λ ) = sup
t∈B ⎩ Ω
∑ ∂ xi
+ Hi + β (1 − 2t)ei dx
⎭
i=1
= sup{Ĝ∗1 (λ ,t)}, (21.8)

t∈B
where
2 1/2
3
∂ λ2
Ĝ∗1 (λ ,t) =
Ω
∑ ∂ xi
+ Hi + β (1 − 2t)ei dx,
i=1
and
1
G∗2 (λ ) = |Curl ∗ λ1 − ∇λ2 |22 dx.
2 R3
Finally,
B = {t measurable | t(x) ∈ [0, 1], a.e. in Ω },
and
Ŷ ∗ = {λ = (λ1 , λ2 ) ∈ W 1,2 (R3 ; R3 ) × W 1,2 (R3 ) | λ2 = 0 on ∂ Ω }.
21.3 The Duality Principle for the Hard Case
In the next lines we present one of our main results, which is summarized by the
following theorem.
Theorem 21.3.1. Define J : Y1 × Y2 × B → R by
J(m, f ,t) = G1 (m, f ,t) + G2 ( f ),

G1 (m, f ,t) = (tg1 (m) + (1 − t)g2(m)) dx
Ω
+Ind0(m) + Ind1( f ) + Ind2(m, f )

− H(x) · m dx, (21.9)
Ω
and
1
G2 ( f ) = | f (x)|22 dx.
2 R3
Also,
g1 (m) = β (1 + m · e),
g2 (m) = β (1 − m · e),

0, if |m(x)|2 = 1 a.e. in Ω ,
Ind0 (m) =
+∞, otherwise,

0, if div(− f + mχΩ ) = 0 a.e. in R3 ,
Ind1 (m, f ) =
+∞, otherwise,
and
0, if Curl( f ) = θ , a.e. in R3 ,
Ind2 ( f ) =
+∞, otherwise.
This case refers to a uniaxial material with no exchange energy, that is, α = 0.
Observe that

ϕ (m) dx = β (1 − |m · e|) dx
Ω Ω

= min (tg1 (m) + (1 − t)g2(m)) dx . (21.10)
t∈B Ω

21.3 The Duality Principle for the Hard Case 497
inf {J(m, f ,t)} ≥ sup {−J ∗ (λ )} (21.11)

(m, f ,t)∈Y1 ×Y2 ×B λ ∈Ŷ ∗
where
J ∗ (λ ) = G̃∗1 (λ ) + G∗2 (λ ) − β dx,
Ω
⎧ ⎫
⎨ 2 1/2 ⎬
3
∂ λ2
G̃∗1 (λ ) = sup
t∈B ⎩ Ω
∑ ∂ xi
+ Hi + β (1 − 2t)ei dx
⎭
i=1
= sup{Ĝ∗1 (λ ,t)}, (21.12)

t∈B
where
2 1/2
3
∂ λ2
Ĝ∗1 (λ ,t) =
Ω
∑ ∂ xi
+ Hi + β (1 − 2t)ei dx,
i=1

1
G∗2 (λ ) = |Curl ∗ λ1 − ∇λ2 |22 dx,
2 R3
B = {t measurable | t(x) ∈ [0, 1], a.e. in Ω },

and
Ŷ ∗ = {λ = (λ1 , λ2 ) ∈ W 1,2 (R3 ; R3 ) × W 1,2 (R3 ) | λ2 = 0 on ∂ Ω }.
Furthermore, under these last assumptions, there exists λ0 ∈ Ŷ ∗ such that
−J ∗ (λ0 ) = max {−J ∗ (λ )}.

λ ∈Ŷ ∗
Moreover, after discretization, suppose that t0 ∈ B such that
G̃∗1 (λ0 ) = Ĝ∗1 (λ0 ,t0 ),
is also such that Ĝ∗1 (λ ,t) is locally Lipschitz continuous in a neighborhood of

(λ0 ,t0 ).
Also assume (λ0 ,t0 ) is such that the hypotheses of Corollary 11.1 are satisfied.
Under such hypotheses, defining
∂ (λ 0 )2
∂ xi + Hi + β (1 − 2t0)ei
(m0 )i = 7 2 , ∀i ∈ {1, 2, 3}
∂ (λ 0 )2
∑3i=1 ∂ xi + Hi + β (1 − 2t0)ei
and
f0 = Curl ∗ (λ0 )1 − ∇(λ0 )2 ,
we have that
J(m0 , f0 ,t0 ) = min {J(m, f ,t)}

(m, f ,t)∈Ỹ
= max{−J ∗ (λ )}
λ ∈Ŷ ∗
∗
= −J (λ0 ). (21.13)
Proof. Observe that denoting G0 : Y1 × B → R by

G0 (m,t) = (tg1 (m) + (1 − t)g2(m))dx − H · m dx,
Ω Ω
we have that
J(m, f ,t) = G1 (m, f ,t) + G2 ( f )

≥ G0 (m,t) + G2( f )

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1L2 (R3 ,R3 )
i=1
+div(− f + mχΩ ), λ2 L2 (R3 )
≥ inf {G0 (m,t) + G2 ( f )
(m, f )∈Y1 ×Y2

λ3 3
+
Ω 2 ∑ m2i − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
+div(− f + mχΩ ), λ2 L2 (R3 ) }

= inf { (tg1 (m) + (1 − t)g2(m)) dx
(m, f )∈Y1 ×Y2 Ω

1
− H(x) · m dx + | f (x)|22 dx
Ω 2 R3

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
+div(− f + mχΩ ), λ2 L2 (R3 ) }. (21.14)
This last infimum indicated is attained for functions satisfying the equations
∂ λ2
Hi + β (1 − 2t)ei − λ3 mi + = 0,
∂ xi
if λ2 = 0 on ∂ Ω .
That is,
Hi + β (1 − 2t)ei + ∂∂λx2i
mi =
λ3
and thus from the constraint
3
∑ m2i − 1 = 0
i=1
we obtain 1/2
3
∂ λ2 2
λ3 = ∑ Hi + β (1 − 2t)ei + ∂ xi .
i=1
Also, the infimum in f is attained for functions satisfying
− f + Curl ∗ λ1 − ∇λ2 = θ .
Through such results we get
inf {G0 (m,t) + G2 ( f )

(m, f )∈Y1 ×Y2

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
+div(− f + mχΩ ), λ2 L2 (R3 ) }

2 1/2
3
∂ λ2
=−
Ω
∑ Hi + β (1 − 2t)ei +
∂ xi
dx
i=1

1
− |Curl ∗ λ1 − ∇λ2 |22 dx + β dx
2 R3
Ω
= −Ĝ∗ (λ ,t) − G∗2(λ ) + β dx. (21.15)

Ω

J(m, f ,t) ≥ −Ĝ∗ (λ ,t) − G∗2(λ ) + β dx
Ω

≥ inf {−Ĝ∗ (λ ,t)} − G∗2(λ ) + β dx
t∈B Ω

= −G̃∗1 (λ ) − G∗2(λ ) + β dx
Ω
= −J ∗ (λ ), (21.16)
∀(m, f ,t) ∈ Ỹ = Y1 × Y2 × B, λ ∈ Ŷ ∗ .
Therefore,
inf {J(m, f ,t)} ≥ sup {−J ∗ (λ )} (21.17)

(m, f ,t)∈Y1 ×Y2 ×B λ ∈Ŷ ∗
Finally, from the concavity, continuity, and coerciveness of −J ∗ : Ŷ ∗ → R, by

an application of the direct method of calculus of variations (since it is a standard
procedure we do not give more details here), we have that there exists λ0 ∈ Ŷ ∗ such
that
−J ∗ (λ0 ) = max {−J ∗ (λ )}.
λ ∈Ŷ ∗
Observe that after discretization, we have
− G̃∗1(λ0 ) = inf {−Ĝ∗1(λ0 ,t)}

t∈B
= −Ĝ∗1 (λ0 ,t0 ). (21.18)
Also after discretization, from the hypotheses and Corollary 11.1, we have
∂ Ĝ∗1 (λ0 ,t0 )

δ {G̃∗1 (λ0 )} = . (21.19)
∂λ
Thus, the extremal equation
δ {−J ∗(λ0 )} = θ ,
stands for
∂ Ĝ∗1 (λ0 ,t0 ) ∂ G∗2 (λ0 )
− − = θ,
∂ λ2 ∂ λ2
and
∂ G∗2 (λ0 )
− = θ,
∂ λ1
that is,
⎛ ⎞
∂ (λ 0 )2
3
∂ ⎜ ∂ xi + Hi + β (1 − 2t0)ei ⎟
∑ ∂ xi ⎜
⎝7
⎟
2 χΩ ⎠
∂ (λ 0 )2
i=1
∑3i=1 ∂ xi + Hi + β (1 − 2t0)ei
−div(Curl ∗ (λ0 )1 − ∇(λ0 )2 ) = 0, (21.20)
a.e. in R3 , and
Curl(Curl ∗ (λ0 )1 − ∇(λ0 )2 ) = θ , a.e. in R3 .
Hence
div(m0 χΩ − f0 ) = 0, a.e. in R3 ,
and
Curl( f0 ) = θ , a.e. in R3 .
Now observe that from the definition of m0 we get
∂ Ĝ∗1 (λ0 ,t0 )

(m0 )i = , ∀i ∈ {1, 2, 3}
∂ vi
∂ λ2
where vi = ∂ xi , so that, from a well-known property of Legendre transform, we
obtain

Ĝ∗1 (λ0 ,t0 ) − β dx
Ω
) *
∂ (λ0 )2
= (m0 )i , − G0 (m0 ,t0 )
∂ xi L2 (Ω )

(λ0 )3 3
−
Ω 2 ∑ (m0 )i − 1 dx.
2
(21.21)
i=1
On the other hand, from the definition of f0 , we get
∂ G∗2 (λ0 )
f0 = ,
∂ v1
where
v1 = Curl ∗ λ1 − ∇λ2 ,
so that

G∗2 (λ0 ) = f0 ,Curl ∗ (λ0 )1 − ∇(λ0 )2 L2 (Ω ;R3 ) − | f0 (x)|2 dx
R3
= f0 ,Curl ∗ (λ0 )1 − ∇(λ0 )2 L2 (Ω ;R3 ) − G2 ( f0 ). (21.22)
From (21.21) and (21.22) we obtain
J(m0 , f0 ,t0 ) = G1 (m0 , f0 ,t0 ) + G2 ( f0 )

= G0 (m0 ,t0 ) + G2( f0 )

(λ0 )3 3
+
Ω 2 ∑ (m0 )i − 1 dx
2
i=1
+Curl( f0 ), (λ0 )1 L2 (R3 ,R3 )
+div(− f0 + m0 χΩ ), (λ0 )2 L2 (R3 )

= −Ĝ∗1 (λ0 ,t0 ) − G∗2(λ0 ) + β dx
Ω

= −G̃∗1 (λ0 ) − G∗2(λ0 ) + β dx
Ω
= −J ∗ (λ0 ). (21.23)
From (21.17) we have
inf {J(m, f ,t)} ≥ sup {−J ∗(λ )}.

(m, f ,t)∈Ỹ λ ∈Ŷ ∗
From this and (21.23) we may infer that
J(m0 , f0 ,t0 ) = min {J(m, f ,t)}

(m, f ,t)∈Ỹ
= max{−J ∗ (λ )}
λ ∈Ŷ ∗
∗
= −J (λ0 ). (21.24)

Remark 21.3.2. The reason we consider the problem just after discretization at some
point refers to the fact that we cannot guarantee the equation supt∈B {Ĝ∗1 (λ0 ,t)} =
Ĝ∗1 (λ0 ,t0 ) is satisfied, for the infinite dimensional original problem, by a measur-
able t0 ∈ B. In fact, the global optimal point for the primal formulation may not be
attained for the infinite dimensional problem, but surely it is attained for its finite
dimensional approximation. If the last theorem hypotheses are satisfied, the solution
for primal finite dimensional problem may be obtained by the corresponding dual
one with no duality gap.
21.4 The Semi-Linear Case
In this section we present another relevant result, which is summarized by the

following theorem.
Theorem 21.4.1. Define J : Y1 × Y2 × B → R by
J(m, f ,t) = G0 (m) + G1 (m, f ,t) + G2 ( f ),

α
G0 (m) = |∇m|22 dx,
2 Ω

G1 (m, f ,t) = (tg1 (m) + (1 − t)g2(m)) dx
Ω
+Ind0(m) + Ind1( f ) + Ind2(m, f )

− H(x) · m dx, (21.25)
Ω
and
1
G2 ( f ) = | f (x)|22 dx.
2 R3
Also,
g1 (m) = β (1 + m · e),
g2 (m) = β (1 − m · e),

0, if |m(x)|2 = 1 a.e. in Ω ,
Ind0 (m) =
+∞, otherwise,

0, if div(− f + mχΩ ) = 0 a.e. in R3 ,
Ind1 (m, f ) =
+∞, otherwise,
and
0, if Curl( f ) = θ , a.e. in R3 ,
Ind2 ( f ) =
+∞, otherwise.
We recall the present case refers to a uniaxial material with exchange energy. That
is, α > 0 and ϕ (m) = β (1 − |m · e|).
21.4 The Semi-Linear Case 503
inf {J(m, f ,t)}

(m, f ,t)∈Y1 ×Y2 ×B
≥ sup {−J ∗(m∗ , λ , m̃)} (21.26)
(m∗ ,λ ,m̃)∈A∗
where
J ∗ (m∗ , λ , m̃) = Ĝ∗0 (m∗ , λ ) + G̃∗1(m∗ , λ )

1
+G∗2 (λ ) + λ3 dx − β dx, (21.27)
2 Ω Ω
1
Ĝ∗0 (m∗ , λ ) = sup {−G0 (m) + m2i , m∗i L2 (Ω ) − mi ni , λ2 L2 (∂ Ω ) },
m∈Y1 2
⎧ ⎛ 2 ⎞ ⎫
⎪
⎨1 ∂ λ2 ⎪
⎬
∂ xi + Hi + β (1 − 2t)ei
3
⎜ ⎟
G̃∗1 (m∗ , λ ) = sup ⎝∑ ⎠ dx
t∈B ⎪
⎩ 2 Ω i=1 m∗i + λ3 ⎪
⎭
= sup{Ĝ∗1 (m∗ , λ ,t)}, (21.28)

t∈B
where ⎛ 2 ⎞
∂ λ2
1 ⎜
3
∂ xi + Hi + β (1 − 2t)ei ⎟
Ĝ∗1 (m∗ , λ ,t) = ⎝∑ ⎠ dx,
2 Ω i=1 m∗i + λ3

1
G∗2 (λ ) = |Curl ∗ λ1 − ∇λ2 |22 dx,
2 R3
and
B = {t measurable | t(x) ∈ [0, 1], a.e. in Ω }.
Also,
A∗ = A1 ∩ A2 ∩ A3 ∩ A4 ,
where
Y3 = L2 (Ω ; R3 ) × W 1,2 (R3 ; R3 ) × W 1,2 (R3 ) × L2 (Ω ),
A1 = {(m∗ , λ ) ∈ Y3 | m∗i + λ3 > 0 in Ω , ∀i ∈ {1, 2, 3}}.
∂ m̃i
A2 = {(m∗ , λ , m̃) ∈ Y4 = Y3 × W 1,2 (Ω ; R3 ) | λ2 ni + =0
∂n
on ∂ Ω , ∀i ∈ {1, 2, 3}}, (21.29)
A3 = {(m∗ , λ ) ∈ Y3 | J(m)
˜ > 0, ∀m ∈ Y1 such that m = θ }.
Here,
λ = (λ1 , λ2 , λ3 )
and
1, -
˜
J(m) = G0 (m) − m2i , m∗i L2 (Ω )
2

α 1
= |∇m|22 dx − m2i , m∗i L2 (Ω ) . (21.30)
2 Ω 2
And also,
A4 = {(m∗ , λ , m̃) ∈ Y4 = Y3 × W 1,2 (Ω ; R3 ) | α ∇2 m̃i + m∗i m̃i = 0, in Ω }.
Suppose there exists (m∗0 , λ0 , m̃0 , m0 ) ∈ A∗ × Y1 such that

∗ ∗ 1, ∗
-
δ −J (m0 , λ0 , m̃0 ) − (m0 )i , (m̃0 )i (m0 )i + α ∇ (m̃0 )i L2 (Ω ) = θ .
2
2
Moreover, considering the problem in question after discretization, assume t0 ∈ B

such that
G̃∗1 (m∗0 , λ0 ) = G̃∗1 (m∗0 , λ0 ,t0 ),
is also such that Ĝ∗1 (m∗ , λ ,t) is locally Lipschitz continuous in a neighborhood of
(m∗0 , λ0 ,t0 ).
Also assume (m∗0 , λ0 ,t0 ) is such that the hypotheses of Corollary 11.1 are satis-
fied. Under such hypotheses, we have
∂ (λ 0 )2
∂ xi + Hi + β (1 − 2t0)ei
(m0 )i = , ∀i ∈ {1, 2, 3}
(m0 )∗i + (λ0 )3
and defining
f0 = Curl ∗ (λ0 )1 − ∇(λ0 )2 ,
we have also that
J(m0 , f0 ,t0 ) = min {J(m, f ,t)}

(m, f ,t)∈Ỹ
= max {−J ∗ (m∗ , λ , m̃)}
(m∗ ,λ ,m̃)∈A∗
= −J ∗ (m∗0 , λ0 , m0 ). (21.31)
Proof. Observe that defining G̃0 : Y1 × Y2 × B → R by

G̃0 (m, f ,t) = (tg1 (m) + (1 − t)g2(m))dx − H · m dx,
Ω Ω
we have that
J(m, f ,t) = G0 (m) + G1(m, f ,t) + G2 ( f )

≥ G0 (m) + G̃0(m, f ,t) + G2 ( f )

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
+div(− f + mχΩ ), λ2 L2 (R3 )

1
= G0 (m) − m2i , m∗i L2 (Ω ) + G̃0 (m, f ,t) + G2 ( f )
2

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
1
+div(− f + mχΩ ), λ2 L2 (R3 ) + m2i , m∗i L2 (Ω ) .
2
Thus,

1
J(m, f ,t) ≥ inf G0 (m) − m2i , m∗i L2 (Ω ) + mi ni , λ2 L2 (∂ Ω )
m∈Y1 2
+ inf {G̃0 (m, f ,t) + G2 ( f )
(m, f )∈Y1 ×Y2

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
1
−(− f + mχΩ ), ∇λ2 L2 (R3 ;R3 ) + m2i , m∗i L2 (Ω ) }
2
= −Ĝ∗0 (m∗ , λ ) + inf { (tg1 (m) + (1 − t)g2(m)) dx
(m, f )∈Y1 ×Y2 Ω

1
− H(x) · m dx + | f (x)|22 dx
Ω 2 R3

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
1
−(− f + mχΩ ), ∇λ2 L2 (R3 ;R3 ) + m2i , m∗i L2 (Ω ) }. (21.32)
2
This last infimum in m indicated is attained for functions satisfying the equations
∂ λ2
Hi + β (1 − 2t)ei − (m∗i + λ3 )mi + = 0.
∂ xi
That is,
Hi + β (1 − 2t)ei + ∂∂λx2i
mi = .
m∗i + λ3
Also, the infimum in f is attained for functions satisfying
f = Curl ∗ λ1 − ∇λ2 .
Through such results we get
G̃0 (m, f ,t) + G2 ( f )

λ3 3 2
+
Ω 2
∑ mi − 1 dx + Curl( f ), λ1 L2 (R3 ,R3 )
i=1
1
−(− f + mχΩ ), ∇λ2 L2 (R3 ;R3 ) + m2i , m∗i L2 (Ω )
⎛ 2
2 ⎞
∂ λ2
1 ⎜
3
∂ xi + H i + β (1 − 2t)e i ⎟
≥− ⎝∑ ∗ ⎠ dx
2 Ω i=1 mi + λ 3

1 1
− |Curl ∗ λ1 − ∇λ2|22 dx − λ3 dx + β dx
2 R3 2 Ω Ω

1
= −Ĝ∗ (m∗ , λ ,t) − G∗2(λ ) − λ3 dx + β dx, (21.33)
2 Ω Ω
if
∂ m̃i
λ2 ni + = 0 on ∂ Ω , ∀i ∈ {1, 2, 3}.
∂n
J(m, f ,t) ≥ −Ĝ∗0 (m∗ , λ ) − Ĝ∗ (m∗ , λ ,t) − G∗2(λ )

1
− λ3 dx + β dx
2 Ω Ω
≥ −Ĝ∗0 (m∗ , λ ) + inf{−Ĝ∗ (m∗ , λ ,t)} − G∗2(λ )
t∈B

1
− λ3 dx + β dx
2 Ω Ω
= −Ĝ∗0 (m∗ , λ ) − G̃∗1 (m∗ , λ ) − G∗2(λ )

1
− λ3 dx + β dx
2 Ω Ω
= −J ∗ (m∗ , λ , m̃), (21.34)
∀(m, f ,t) ∈ Ỹ , (m∗ , λ , m̃) ∈ A∗ .

Therefore,
inf {J(m, f ,t)} ≥ sup {−J ∗(m∗ , λ , m̃)} (21.35)

(m, f ,t)∈Y1 ×Y2 ×B (m∗ ,λ ,m̃)∈A∗
Finally, from now on considering the problem after discretization, we have
− G̃∗1(m∗0 , λ0 ) = inf {−Ĝ∗1(m∗0 , λ0 ,t)}

t∈B
= −Ĝ∗1 (m∗0 , λ0 ,t0 ). (21.36)
From the hypotheses and Corollary 11.1 we have that

∗ ∗ ∂ Ĝ∗1 (m∗0 , λ0 ,t0 ) ∂ Ĝ∗1 (m∗0 , λ0 ,t0 )
δ {G̃1 (m0 , λ0 )} = , . (21.37)
∂ m∗ ∂λ
Thus, considering that the following representation holds,
Ĝ∗0 (m∗0 , λ0 ) = −(m̃0 )i ni , (λ0 )2 L2 (∂ Ω ) /2,
from these last results and hypotheses, the extremal equation

∗ ∗ 1, ∗
-
δ −J (m0 , λ0 , m̃0 ) − (m0 )i , (m̃0 )i (m0 )i + α ∇ (m̃0 )i L2 (Ω ) = θ ,
2
2
stands for
, -
∂ −Ĝ∗1 (m∗0 , λ0 ,t0 ) − 12 (m0 )i , (m̃0 )i (m0 )∗i + α ∇2 (m̃0 )i L2 (Ω )
= θ, (21.38)
∂ m∗i
, -
∂ 1
(m ) , (m̃ ) (m )∗ + α ∇2 (m̃ )
2 0 i 0 i 0 i 0 i L (Ω )
2
− = θ, (21.39)
∂ m̃i
, -
∗
∂ 1
2 (m 0 )i , (m̃ 0 )i (m 0 )i + α ∇2
(m̃ 0 )i L (Ω )
2
= θ, (21.40)
∂ mi
∂ Ĝ∗1 (m∗0 , λ0 ,t0 ) 1

− − = θ, (21.41)
∂ λ3 2
that is,
⎛ ⎞2
∂ (λ 0 )2
1 3 ⎝ ∂ xi + Hi + β (1 − 2t0)ei ⎠ 1
∑
2 i=1 (m∗0 )i + (λ0 )3
− = 0, a.e. in Ω .
2
(21.42)
Also,
∂ Ĝ∗0 (m∗0 , λ0 ) ∂ Ĝ∗1 (m∗0 , λ0 ,t0 ) ∂ G∗2 (λ0 )
− − − = θ, (21.43)
∂ λ2 ∂ λ2 ∂ λ2
and
∂ G∗2 (λ0 )
− = θ. (21.44)
∂ λ1
That is, from (21.38),

⎛ ⎞2
∂ (λ 0 )2
+ Hi + β (1 − 2t0)ei
1
− ⎝
∂ xi ⎠ + (m0 )i (m̃0 )i = 0, a.e. in Ω , (21.45)
2 (m∗0 )i + (λ0)3 2
and by (21.39) and (21.40)
(m0 )i (m0 )∗i + α ∇2 (m0 )i = 0, a.e. in Ω ,
(m̃0 )i (m0 )∗i + α ∇2 (m̃0 )i = 0, a.e. in Ω ,

so that
m0 = m̃0 , a.e. in Ω , (21.46)
and hence,
1
Ĝ∗0 (m∗0 , λ0 ) = −G0 (m0 ) + (m0 )2i , (m0 )∗i L2 (Ω ) − (m0 )i ni , (λ0 )2 L2 (∂ Ω ) . (21.47)
2
Moreover, by (21.43),
⎛ ⎞
∂ (λ 0 )2
3
∂ ∂ xi + Hi + β (1 − 2t0)ei
∑ ∂ xi ⎝ (m∗0 )i + (λ0 )3
χΩ ⎠
i=1
−div(Curl ∗ (λ0 )1 − ∇(λ0 )2 ) = 0, (21.48)
a.e. in R3 , and
⎛ ⎞
∂ (λ 0 )2
∂ xi + Hi + β (1 − 2t0)ei
⎝ ⎠ ni − (m̃0 )i ni = 0, on ∂ Ω .
(m∗0 )i + (λ0 )3
Thus, from this, (21.45), (21.46), and (21.42), we obtain

∂ (λ 0 )2
∂ xi + Hi + β (1 − 2t0)ei
(m0 )i = ,
(m∗0 )i + (λ0 )3
and
3
∑ (m0 )2i = 1, a.e. in Ω .
i=1
From (21.44),
Curl(Curl ∗ (λ0 )1 − ∇(λ0 )2 ) = θ , a.e. in R3 .
Hence
div(m0 χΩ − f0 ) = 0, a.e. in R3 ,
and
Curl( f0 ) = θ , a.e. in R3 .
Now observe that from the expression of m0 we get
∂ Ĝ∗1 (m∗0 , λ0 ,t0 )

(m0 )i = , ∀i ∈ {1, 2, 3}
∂ vi
∂ λ2
where vi = ∂ xi , so that from a well-known property of Legendre transform, we
obtain

1
Ĝ∗1 (m∗0 , λ0 ,t0 ) + (λ0 )3 dx − β dx
2 Ω Ω
) *
∂ (λ0 )2
= (m0 )i , − G̃0 (m0 , f0 ,t0 )
∂ xi L2 (Ω )

(λ0 )3 3
−
Ω 2 ∑ (m0 )i − 1 dx
2
i=1
1
− (m0 )2i , (m0 )∗i L2 (Ω ) . (21.49)
2
On the other hand, from the definition of f0 , we get
∂ G∗2 (λ0 )
f0 = ,
∂ v1
where
v1 = Curl ∗ λ1 − ∇λ2 ,
so that

G∗2 (λ0 ) = f0 ,Curl ∗ (λ0 )1 − ∇(λ0 )2 L2 (Ω ;R3 ) − | f0 (x)|2 dx
R3
= f0 ,Curl ∗ (λ0 )1 − ∇(λ0 )2 L2 (Ω ;R3 ) − G2 ( f0 ). (21.50)

J(m0 , f0 ,t0 ) = G0 (m0 ) + G1 (m0 , f0 ,t0 ) + G2 ( f0 )
1
= G(m0 ) − (m0 )2i , (m∗0 )i L2 (Ω ) + G̃0(m0 , f0 ,t0 ) + G2 ( f0 )
2

(λ0 )3 3
+
Ω 2 ∑ (m0 )i − 1 dx + Curl( f0), (λ0 )1 L2 (R3 ,R3 )
2
i=1
1
+div(− f0 + m0 χΩ ), (λ0 )2 L2 (R3 ) + (m0 )2i , (m0 )∗i L2 (Ω )
2
= −Ĝ∗0 (m∗0 , λ0 ) − Ĝ∗1 (m∗0 , λ0 ,t0 ) − G∗2(λ0 )

1
− (λ0 )3 dx + β dx
2 Ω Ω
= −Ĝ∗0 (m∗0 , λ0 ) − G̃∗1 (m∗0 , λ0 ) − G∗2 (λ0 )

1
− (λ0 )3 dx + β dx
2 Ω Ω
= −J ∗ (m∗0 , λ0 , m0 ). (21.51)
inf {J(m, f ,t)} ≥ sup {−J ∗ (m∗ , λ , m̃)}.

(m, f ,t)∈Ỹ (m∗ ,λ ,m̃)∈A∗
From this and (21.51) we may infer that
J(m0 , f0 ,t0 ) = min {J(m, f ,t)}

(m, f ,t)∈Ỹ
= max {−J ∗ (m∗ , λ , m̃)}
(m∗ ,λ ,m̃)∈A∗
= −J ∗ (m∗0 , λ0 , m0 ). (21.52)
21.5 Numerical Examples
In this section we present a numerical two-dimensional example concerning the

hard case. Consider Ω = [0, 1] × [0, 1] ⊂ R2 , the region corresponding to a micro-
magnetic sample. We develop numerical results for the minimization of the simpli-
fied dual functional
⎧ ⎫
⎨ 2 2 1/2 ⎬
∂ λ2
J ∗ (λ2 ) = sup ∑ + Hi + β (1 − 2t)ei dx
i=1 ∂ xi
t∈B ⎩ Ω ⎭

1
+ |∇λ2 |22 dx, (21.53)
2 Ω
with the boundary condition

λ2 = 0 on ∂ Ω .
In such a case we have neglected the external induced magnetic field. Anyway,
observe that
Curl( f0 ) = θ
(in fact it is an appropriate version for the two-dimensional case) stands for the
obviously satisfied equation
Curl(∇λ2 ) = θ .
Finally, units are related to the international system.
21.5.1 First Example
For such an example, for a fixed β > 0, we consider the cases H0 /β =

1.0, H0 /β = 10, and H0 /β = 100, where the magnetic field H is given by
H = H0 i + 0j,
where
i = (1, 0) and j = (0, 1).
√ √
Moreover, e1 = 2/2 and e2 = 2/2, where (e1 , e2 ) is the preferred direction of
magnetization. We have plotted the stream lines for the vector field m = (m1 , m2 )
for these three cases. Please see Figs. 21.1, 21.2, and 21.3. We observe that as H0
increases, the magnetization m direction approaches the magnetic field H one, which
in such an example is given by i = (1, 0).
Remark 21.5.1. It is worth mentioning that as H0 /β is smaller the magnetization m
is closer to (e1 , e2 ). However its direction approaches the H one, as H0 increases.
Such a result is consistent with the concerned problem physics.
21.5.2 Second Example
For such an example, for a fixed β > 0, we consider the cases H0 /β =

0.5, H0 /β = 5.0, and H0 /β = 50, where the magnetic field H is given by
H = H0 Ha (x, y),
where
Ha = x(0.5 − y)i − y(0.5 − x)j.
√ √
Also, again e1 = 2/2 and e2 = 2/2, where (e1 , e2 ) is the preferred direction of
magnetization. For the stream lines of Ha , please see Fig. 21.4. For the magnetiza-
tion m for these three different cases see Figs. 21.5, 21.6, and 21.7. For the parameter
t related to the case H0 /β = 0.5, see Fig. 21.8.
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
Fig. 21.1 First example—stream lines for the magnetization m for H0 /β = 1.0
1.4
1.2
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
Fig. 21.4 Second example—stream lines for external magnetic field Ha

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
Fig. 21.5 Second example—stream lines for the magnetization m for H0 /β = 0.5
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
21.6 Acknowledgments 515
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1
0.8
0.6
0.4
0.2
0
1
1
0.8
0.5 0.6
0.4
0.2
0 0
Fig. 21.8 Second example—parameter t(x) for H0 /β = 0.5
21.6 Acknowledgments
The author is very grateful to Professor Robert C. Rogers for his excellent work
as Ph.D. thesis advisor at Virginia Tech-USA. We thank as well the Mathematics
Department of that institution for the constant financial support received during our
doctoral program.
21.7 Conclusion
In this chapter we develop duality principles for models in ferromagnetism met

in reference [44], for example. The dual variational formulations here presented
are convex (in fact concave) and 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 22
The Generalized Method of Lines Applied
to Fluid Mechanics
22.1 Introduction
In this chapter we develop solutions for the Navier–Stokes system through the
generalized method of lines. The main reference for this chapter is R. Temam [66].
At this point we describe the system in question.
Consider Ω ⊂ R2 an open, bounded, and connected set, whose internal boundary
is denoted by Γ0 and external one is denoted by Γ1 . Denoting by u : Ω → R the field
of velocity in direction x of the Cartesian system (x, y), by v : Ω → R the velocity
field in the direction y, by p : Ω → R the pressure field, so that P = p/ρ , where ρ is
the constant fluid density, ν is the viscosity coefficient, and g is the gravity constant,
the Navier–Stokes PDE system is expressed by
⎧ 2
⎪ ν ∇ u − u∂xu − v∂yu − ∂x P + gx = 0, in Ω ,
⎪
⎨
ν ∇2 v − u∂xv − v∂y v − ∂y P + gy = 0, in Ω , (22.1)
⎪
⎪
⎩
∂x u + ∂y v = 0, in Ω ,
u = v = 0, on Γ0 ,
(22.2)
u = u∞ , v = 0, P = P∞ , on Γ1
In principle we look for solutions (u, v, P) ∈ W 2,2 (Ω ) × W 2,2 (Ω ) × W 1,2 (Ω ) de-

spite the fact that less regular solutions are also possible specially concerning the
weak formulation.
22.2 On the Solution of Steady-State Euler Equation
Through the next result we obtain a linear system whose solution also solves the
steady-state Euler system.
518 22 The Generalized Method of Lines Applied to Fluid Mechanics
Theorem 22.2.1. A solution for the Euler system below indicated, that is,
⎧
⎪
⎪ −u∂x u − v∂yu − ∂x P + gx = 0, in Ω ,
⎪
⎪
⎨
−u∂x v − v∂yv − ∂y P + gy = 0, in Ω , (22.3)
⎪
⎪
⎪
⎪
⎩
∂x u + ∂yv = 0, in Ω ,

"
u · n = 0, on Γ , (22.4)
where u = (u, v), is given by

⎧
⎨ u = ∂x w0 ,
(22.5)
⎩
v = ∂y w0 ,
where w0 is a solution of the equation

⎧ 2
⎨ ∇ w0 = 0 in Ω ,
(22.6)
⎩
∇w0 · n = 0, on Γ .
Proof. For u = ∂x w0 and v = ∂y w0 define
F = −(∂x w0 )2 /2 − (∂yw0 )2 /2 = −u2/2 − v2/2.
The continuity equation in the system (22.3) stands for
∇2 w0 = 0 in Ω . (22.7)
The first two equations in (22.3) correspond to
∂x F − ∂x P + gx = 0, in Ω , (22.8)
and
∂y F − ∂y P + gy = 0, in Ω , (22.9)
which may be solved in P for the particular w0 solution of (22.7) with corresponding
boundary conditions. Since
∂y (∂x F + gx ) = ∂x (∂y F + gy ) = ∂xy P,
it is clear that (22.8) and (22.9) have a solution in P with the proper boundary con-
ditions above described.
22.3 The Generalized Method of Lines for the Navier–Stokes System 519
22.3 The Generalized Method of Lines for the Navier–Stokes

System
In this section we develop the solution for the Navier–Stokes system through the
generalized method of lines. About such a method see Chap. 15 for more details.
Consider
Ω = {(r, θ ) | 1 ≤ r ≤ 2, 0 ≤ θ ≤ 2π },
∂ Ω0 = {(1, θ ) | 0 ≤ θ ≤ 2π },
and
∂ Ω1 = {(2, θ ) | 0 ≤ θ ≤ 2π }.
First for the boundary conditions
u = 0, v = 0, P = P0 (x) on ∂ Ω0 ,
u = u f (x), v = v f (x), P = Pf (x) on ∂ Ω1 ,
we have obtained the following general expressions for the n − th lines (for an ap-
propriate approximate system of equations in polar coordinates):
un (x) = a1 [n] cos(x)P0 (x) + a2[n] cos[x]Pf (x) + a3[n]u f (x)

+a4 [n] cos(x)u f (x)2 + a5[n] sin(x)u f (x)v f (x)
+a6 [n] sin(x)P0 (x) + a7[n] sin(x)Pf (x) + a8[n] sin(x)u f (x)uf (x)
+a9 [n] cos(x)v f (x)uf (x) + a10[n]uf (x)
vn (x) = b1 [n]P0 (x) sin[x] + b2[n]Pf (x) sin(x) + b3 [n]v f (x)

+b4[n] cos[x]u f (x)v f (x) + b5[n] sin(x)v f (x)2
+b6[n] cos(x)P0 (x) + b7[n] cos(x)Pf (x) + b8[n] sin(x)u f (x)vf (x)
+b9[n] cos(x)v f (x)vf (x) + b10 [n]vf (x)
Pn (x) = c1 [n]P0 (x) + c2 [n]Pf (x) + c3 [n] cos(x)2 u f (x)2

+c4 [n] cos(x) sin(x)u f (x)v f (x) + c5 [n] sin(x)2 v f (x)2
+c6 [n] cos(x) sin(x)u f (x)uf (x) + c7[n] cos(x)2 v f (x)uf (x)
+c8 [n] sin(x)2 u f (x)2 + c9 [n] sin(x)2 u f (x)vf (x)
+c10 [n] cos(x) sin(x)v f (x)vf (x) + c11[n] cos(x) sin(x)uf (x)vf (x)
+c12 [n] cos(x)2 vf (x)2 + c13 [n]P0 (x) + c14 [n]Pf (x)
Denoting

2
J(u, v, P) = ν ∇2 u − u∂xu − v∂y u − ∂x P dΩ
Ω

2
+ ν ∇2 v − u∂x v − v∂yv − ∂y P dΩ
Ω
+ (∂x u + ∂yv)2 d Ω , (22.10)
Ω
the coefficients {ai [n]}, {bi [n]}, {ci [n]} may be obtained through the numerical min-
imization of J(u, v, P).
22.3.1 The General Case for Specific Boundary Conditions
For the boundary conditions
u = 0, v = 0, P = P0 (x) on ∂ Ω0 ,
u = u f (x), v = v f (x), P = Pf (x) on ∂ Ω1 ,

where
Ω = {(r, θ ) : | r(θ ) ≤ r ≤ 2r(θ )}
where r(θ ) is a smooth periodic function. We recall that the system in question, in
function of the variables (t, θ ) where t = r/r(θ ), is given by
L(u) − ud1(u) − vd2(u) − d1(P) = 0, (22.11)
L(v) − ud1(v) − vd2(v) − d2 (P) = 0, (22.12)

d1 (u) + d2(v) = 0. (22.13)
where
∂ 2u 1 ∂u
L(u)/ f0 (θ ) = + f2 (θ )
∂t 2 t ∂t
1 ∂ 2u f4 (θ ) ∂ 2 u
+ f3 (θ ) + 2 = 0, (22.14)
t ∂θ∂t t ∂θ2
in Ω . Here f0 (θ ), f2 (θ ), f3 (θ ), and f4 (θ ) are known functions.
More specifically, denoting
−r (θ )
f1 (θ ) = ,
r(θ )
we have
f0 (θ ) = 1 + f1 (θ )2 ,
f1 (θ )
f2 (θ ) = 1 + ,
1 + f1(θ )2
2 f1 (θ )
f3 (θ ) = ,
1 + f1 (θ )2
and
1
f4 (θ ) = .
1 + f1 (θ )2
Also
∂u ∂u
d1 u/ f0 (θ ) = f5 (θ ) + ( f6 (θ )/t) ,
∂t ∂θ
∂u ∂u
d2 u/ f0 (θ ) = f7 (θ ) + ( f8 (θ )/t) ,
∂t ∂θ
where
f5 (θ ) = cos(θ )/r(θ ) + sin(θ )r (θ )/r3 (θ ),
f6 (θ ) = − sin(θ )/r(θ ),
f7 (θ ) = sin(θ )/r(θ ) − cos(θ )r (θ )/r3 (θ ),
f8 (θ ) = cos(θ )/r(θ ).
Observe that t ∈ [1, 2] in Ω .
From (22.11) and (22.12) we may write
d1 (L(u) − ud1(u) − vd2(u) − d1(P))

+d2 (L(v) − ud1(v) − vd2(v) − d2(P)) = 0, (22.15)
d1 [L(u)] + d2[L(v)] = L(d1 (u) + d2(v)) = 0,
and considering that

d1 (d1 (P)) + d2(d2 (P)) = L(P),
from (22.15) we have
L(P) + d1 (u)2 + d2 (v)2 + 2d2(u)d1 (v) = 0, in Ω .
Hence, in fact we solve the approximate system
L(u) − ud1(u) − vd2(u) − d1(P) = 0,
L(v) − ud1(v) − vd2(v) − d2 (P) = 0,

L(P) + d1 (u)2 + d2 (v)2 + 2d2(u)d1 (v) = 0, in Ω .
For the field of velocity u we have obtained the following expressions for the
lines (here x stands for θ ):
Line 1
u1 (x) = 0.1u f (x) + 0.045 f5(x)P0 (x) − 0.045 f5(x)Pf (x)
+0.034 f2(x)u f (x) − 0.0165 f5(x)u f (x)2
−0.0165 f7(x)u f (x)v f (x) − 0.023 f6(x)P0 (x)
−0.011 f6(x)Pf (x) + 0.034 f3(x)uf (x)
−0.005 f6(x)u f (x)uf (x) − 0.005 f8(x)v f (x)uf (x)
+0.008 f4(x)uf (x)
Line 2
u2 (x) = 0.2u f (x) + 0.080 f5(x)P0 (x) − 0.080 f5(x)Pf (x)
+0.058 f2(x)u f (x) − 0.032 f5(x)u f (x)2
−0.032 f7(x)u f (x)v f (x) − 0.037 f6(x)P0 (x)
−0.022 f6(x)Pf (x) + 0.058 f3(x)uf (x)
+0.015 f4(x)uf (x)
Line 3
u3 (x) = 0.3u f (x) + 0.105 f5(x)P0 (x) − 0.105 f5(x)Pf (x)
+0.075 f2(x)u f (x) − 0.045 f5(x)u f (x)2
−0.045 f7(x)u f (x)v f (x) − 0.044 f6(x)P0 (x)
−0.030 f6(x)Pf (x) + 0.075 f3(x)uf (x)
+0.020 f4(x)uf (x)
Line 4
u4 (x) = 0.4u f (x) + 0.120 f5(x)P0 (x) − 0.120 f5(x)Pf (x)
+0.083 f2(x)u f (x) − 0.056 f5(x)u f (x)2
−0.056 f7(x)u f (x)v f (x) − 0.047 f6(x)P0 (x)
−0.037 f6(x)Pf (x) + 0.083 f3(x)uf (x)
+0.024 f4(x)uf (x)
Line 5
u5 (x) = 0.5u f (x) + 0.125 f5(x)P0 (x) − 0.125 f5(x)Pf (x)
+0.085 f2(x)u f (x) − 0.062 f5(x)u f (x)2
−0.062 f7(x)u f (x)v f (x) − 0.045 f6(x)P0 (x)

−0.040 f6(x)Pf (x) + 0.085 f3(x)uf (x)
+0.026 f4(x)uf (x)
Line 6
u6 (x) = 0.6u f (x) + 0.120 f5(x)P0 (x) − 0.120 f5(x)Pf (x)
+0.080 f2(x)u f (x) − 0.064 f5(x)u f (x)2
−0.064 f7(x)u f (x)v f (x) − 0.039 f6(x)P0 (x)
−0.040 f6(x)Pf (x) + 0.080 f3(x)uf (x)
+0.025 f4(x)uf (x)
Line 7
u7 (x) = 0.7u f (x) + 0.105 f5(x)P0 (x) − 0.105 f5(x)Pf (x)
+0.068 f2(x)u f (x) − 0.059 f5(x)u f (x)2
−0.059 f7(x)u f (x)v f (x) − 0.032 f6(x)P0 (x)
−0.037 f6(x)Pf (x) + 0.068 f3(x)uf (x)
+0.023 f4(x)uf (x)
Line 8
u8 (x) = 0.8u f (x) + 0.080 f5(x)P0 (x) − 0.080 f5(x)Pf (x)
+0.051 f2(x)u f (x) − 0.048 f5(x)u f (x)2
−0.048 f7(x)u f (x)v f (x) − 0.022 f6(x)P0 (x)
−0.029 f6(x)Pf (x) + 0.051 f3(x)uf (x)
+0.018 f4(x)uf (x)
Line 9
u9 (x) = 0.9u f (x) + 0.045 f5(x)P0 (x) − 0.045 f5(x)Pf (x)
+0.028 f2(x)u f (x) − 0.059 f5(x)u f (x)2
−0.028 f7(x)u f (x)v f (x) − 0.028 f6(x)P0 (x)
−0.011 f6(x)Pf (x) + 0.017 f3(x)uf (x)

+0.010 f4(x)uf (x)
For the field of velocity v we have obtained for the following expressions for the
lines:
Line 1
v1 (x) = 0.1v f (x) + 0.045 f7(x)P0 (x) − 0.045 f7(x)Pf (x)
+0.034 f2(x)v f (x) − 0.017 f5(x)u f (x)v f (x)
−0.017 f7(x)v f (x)2 − 0.023 f8(x)P0 (x)
−0.011 f8(x)Pf (x) + 0.034 f3(x)vf (x)
−0.005 f6(x)u f (x)vf (x) − 0.005 f8(x)v f (x)vf (x)
+0.008 f4(x)vf (x)
Line 2
v2 (x) = 0.2v f (x) + 0.080 f7(x)P0 (x) − 0.080 f7(x)Pf (x)
+0.058 f2(x)v f (x) − 0.032 f5(x)u f (x)v f (x)
−0.032 f7(x)v f (x)2 − 0.037 f8(x)P0 (x)
−0.022 f8(x)Pf (x) + 0.058 f3(x)vf (x)
+0.015 f4(x)vf (x)
Line 3
v3 (x) = 0.3v f (x) + 0.105 f7(x)P0 (x) − 0.105 f7(x)Pf (x)
+0.075 f2(x)v f (x) − 0.045 f5(x)u f (x)v f (x)
−0.045 f7(x)v f (x)2 − 0.045 f8(x)P0 (x)
−0.030 f8(x)Pf (x) + 0.075 f3(x)vf (x)
+0.020 f4(x)vf (x)
Line 4
v4 (x) = 0.4v f (x) + 0.120 f7(x)P0 (x) − 0.120 f7(x)Pf (x)
+0.083 f2(x)v f (x) − 0.056 f5(x)u f (x)v f (x)
−0.056 f7(x)v f (x)2 − 0.047 f8(x)P0 (x)
−0.037 f8(x)Pf (x) + 0.083 f3(x)vf (x)
+0.024 f4(x)vf (x)
Line 5
v5 (x) = 0.5v f (x) + 0.125 f7(x)P0 (x) − 0.125 f7(x)Pf (x)
+0.085 f2(x)v f (x) − 0.062 f5(x)u f (x)v f (x)
−0.062 f7(x)v f (x)2 − 0.045 f8(x)P0 (x)
−0.040 f8(x)Pf (x) + 0.085 f3(x)vf (x)
+0.026 f4(x)vf (x)
Line 6
v6 (x) = 0.6v f (x) + 0.120 f7(x)P0 (x) − 0.120 f7(x)Pf (x)
+0.068 f2(x)v f (x) − 0.064 f5(x)u f (x)v f (x)
−0.064 f7(x)v f (x)2 − 0.039 f8(x)P0 (x)
−0.040 f8(x)Pf (x) + 0.080 f3(x)vf (x)
+0.026 f4(x)vf (x)
Line 7
v7 (x) = 0.7v f (x) + 0.105 f7(x)P0 (x) − 0.105 f7(x)Pf (x)
+0.068 f2(x)v f (x) − 0.059 f5(x)u f (x)v f (x)
−0.059 f7(x)v f (x)2 − 0.032 f8(x)P0 (x)
−0.037 f8(x)Pf (x) + 0.068 f3(x)vf (x)
+0.023 f4(x)vf (x)
Line 8
v8 (x) = 0.8v f (x) + 0.080 f7(x)P0 (x) − 0.080 f7(x)Pf (x)
+0.051 f2(x)v f (x) − 0.048 f5(x)u f (x)v f (x)
−0.048 f7(x)v f (x)2 − 0.022 f8(x)P0 (x)
−0.029 f8(x)Pf (x) + 0.051 f3(x)vf (x)
+0.018 f4(x)vf (x)
Line 9
v9 (x) = 0.9v f (x) + 0.045 f7(x)P0 (x) − 0.045 f7(x)Pf (x)
+0.028 f2(x)v f (x) − 0.028 f5(x)u f (x)v f (x)
−0.028 f7(x)v f (x)2 − 0.011 f8(x)P0 (x)
−0.017 f8(x)Pf (x) + 0.028 f3(x)vf (x)
+0.010 f4(x)vf (x)
Finally, for the field of pressure P, we have obtained the following lines:
Line 1
P1 (x) = 0.9P0(x) + 0.1Pf (x) − 0.034 f2(x)P1 (x)
+0.034 f2(x)Pf (x) + 0.045 f5(x)2 f0 (x)u f (x)2
+0.090 f5(x) f7 (x) f0 (x)u f (x)v f (x) + 0.045 f7(x)2 f0 (x)v f (x02
−0.034 f3(x)P0 (x) + 0.034 f3(x)Pf (x)
+0.022 f5(x) f6 (x) f0 (x)u f (x)uf (x) + 0.022 f5(x) f8 (x) f0 (x)v f (x)uf (x)
+0.003 f6(x)2 f0 (x)uf (x)2 + 0.022 f6(x) f7 (x) f0 (x)u f (x)vf (x)
+0.022 f7(x) f8 (x) f0 (x)v f (x)vf (x) + 0.007 f6(x) f8 (x) f0 (x)uf (x)vf (x)
+0.003 f8(x)2 f0 (x)vf (x)2 + 0.018 f4(x)P0 (x)
+0.008 f4(x)Pf (x)
Line 2
P2 (x) = 0.8P0(x) + 0.2Pf (x) − 0.058 f2(x)P0 (x)
+0.058 f2(x)Pf (x) + 0.080 f5(x)2 f0 (x)u f (x)2
+0.160 f5(x) f7 (x) f0 (x)u f (x)v f (x) + 0.080 f7(x)2 f0 (x)v f (x)2
−0.058 f3(x)P0 (x) + 0.058 f3(x)Pf (x)
+0.007 f8(x)2 f0 (x)vf (x)2 + 0.028 f4(x)P0 (x)
+0.014 f4(x)Pf (x)
Line 3
P3 (x) = 0.7P0(x) + 0.3Pf (x) − 0.075 f2(x)P0 (x)
+0.075 f2(x)Pf (x) + 0.104 f5(x)2 f0 (x)u f (x)2
−0.075 f3(x)P0 (x) + 0.075 f3(x)Pf (x)
+0.010 f8(x)2 f0 (x)vf (x)2 + 0.034 f4(x)P0 (x)
+0.020 f4(x)Pf (x)
Line 4
P4 (x) = 0.6P0(x) + 0.4Pf (x) − 0.083 f2(x)P0 (x)
+0.083 f2(x)Pf (x) + 0.120 f5(x)2 f0 (x)u f (x)2
−0.083 f3(x)P0 (x) + 0.083 f3(x)Pf (x)
+0.012 f8(x)2 f0 (x)vf (x)2 + 0.035 f4(x)P0 (x)
+0.024 f4(x)Pf (x)
Line 5
P5 (x) = 0.5P0(x) + 0.5Pf (x) − 0.085 f2(x)P0 (x)
+0.085 f2(x)Pf (x) + 0.125 f5(x)2 f0 (x)u f (x)2
−0.085 f3(x)P0 (x) + 0.085 f3(x)Pf (x)
+0.014 f8(x)2 f0 (x)vf (x)2 + 0.033 f4(x)P0 (x)
+0.026 f4(x)Pf (x)
Line 6
P6 (x) = 0.4P0(x) + 0.6Pf (x) − 0.080 f2(x)P0 (x)
+0.080 f2(x)Pf (x) + 0.120 f5(x)2 f0 (x)u f (x)2
−0.080 f3(x)P0 (x) + 0.080 f3(x)Pf (x)

+0.015 f8(x)2 f0 (x)vf (x)2 + 0.028 f4(x)P0 (x)
+0.026 f4(x)Pf (x)
Line 7
P7 (x) = 0.3P0(x) + 0.7Pf (x) − 0.068 f2(x)P0 (x)
+0.068 f2(x)Pf (x) + 0.105 f5(x)2 f0 (x)u f (x)2
−0.068 f3(x)P0 (x) + 0.068 f3(x)Pf (x)
+0.014 f8(x)2 f0 (x)vf (x)2 + 0.022 f4(x)P0 (x)
+0.023 f4(x)Pf (x)
Line 8
P8 (x) = 0.2P0(x) + 0.8Pf (x) − 0.051 f2(x)P0 (x)
+0.051 f2(x)Pf (x) + 0.080 f5(x)2 f0 (x)u f (x)2
−0.051 f3(x)P0 (x) + 0.051 f3(x)Pf (x)
+0.011 f8(x)2 f0 (x)vf (x)2 + 0.015 f4(x)P0 (x)
+0.018 f4(x)Pf (x)
Line 9
P9 (x) = 0.1P0(x) + 0.9Pf (x) − 0.028 f2(x)P0 (x)
+0.028 f2(x)Pf (x) + 0.045 f5(x)2 f0 (x)u f (x)2
−0.028 f3(x)P0 (x) + 0.028 f3(x)Pf (x)
+0.007 f8(x)2 f0 (x)vf (x)2 + 0.008 f4(x)P0 (x)
+0.010 f4(x)Pf (x)
22.3.2 A Numerical Example
We consider for the cases ν = 1 and ν = 0.01,
Ω = {(r, θ ) | 1 ≤ r ≤ 2, 0 ≤ θ ≤ 2π },
∂ Ω0 = {(1, θ ) | 0 ≤ θ ≤ 2π },
and
∂ Ω1 = {(2, θ ) | 0 ≤ θ ≤ 2π }.
For the present example, the boundary conditions are
u = −3.0 sin(θ ), v = 3.0 cos(θ ), on ∂ Ω0 ,

u = v = 0, P = 2.0 on ∂ Ω1 ,
Through the generalized method of lines, truncating the series up to the terms
in d 2 where d = 1/N is the mesh thickness concerning the discretization in r, the
general expression for the velocity and pressure fields on the line n is given by (here
x stands for θ ):
un (x) = a1 [n] cos(x) + a2[n] sin(x) + a3[n] cos(x)3 + a4 [n] cos(x) sin(x)2
vn (x) = b1 [n] cos(x) + b2 [n] sin(x) + b3 [n] sin(x)3 + b4 [n] cos(x)2 sin(x)
Pn (x) = c1 [n] + c2[n] sin(x)4 + c3 [n] cos(x)4 + c4 [n] cos(x)2 sin(x)2 .
We have plotted the field of velocity u, for lines n = 1, n = 5, n = 10, n =

15, and n = 19, for a mesh 20 × 20. Please see Figs. 22.1, 22.2, 22.3, 22.4, and 22.5
for the case ν = 1.0.
For the case ν = 0.01 see Figs. 22.6, 22.7, 22.8, 22.9, and 22.10. For all graphs,
please consider units in x to be multiplied by 2π /20.
Again denoting

2
J= ν ∇2 u − u∂x u − v∂yu − ∂x P dΩ
Ω

2
+ ν ∇2 v − u∂x v − v∂yv − ∂y P dΩ
Ω
+ (∂x u + ∂yv)2 d Ω , (22.16)
Ω
the coefficients {ai [n]}, {bi[n]}, {ci [n]} has been obtained through the numerical
minimization of J, so that for the mesh in question, we have obtained
J ≈ 0.0665 for ν = 1.0,
J ≈ 0.0437 for ν = 0.01.

In any case it seems we have got good qualitative first approximations for the con-
cerned solutions.
22.4 Conclusion
In this chapter we develop solutions for two-dimensional examples of incom-

pressible Navier–Stokes system. Such solutions are obtained through the gener-
alized method of lines. The extension of results to R3 , compressible and time-
dependent cases, is planned for a future work.
−1
−2
−3
0 5 10 15 20
Fig. 22.1 Field of velocity u1 (x)-line n=1, case ν = 1.0

22.4 Conclusion 531
2.5
1.5
0.5
−0.5
−1
−1.5
−2
−2.5
0 5 10 15 20
1.5
0.5
−0.5
−1
−1.5
0 5 10 15 20

0.6
0.4
0.2
−0.2
−0.4
−0.6
−0.8
0 5 10 15 20
0.15
0.1
0.05
−0.05
−0.1
−0.15
−0.2
0 5 10 15 20

22.4 Conclusion 533
−1
−2
−3
0 5 10 15 20
0.8
0.6
0.4
0.2
−0.2
−0.4
−0.6
−0.8
0 5 10 15 20

0.6
0.4
0.2
−0.2
−0.4
−0.6
−0.8
0 5 10 15 20
0.4
0.3
0.2
0.1
−0.1
−0.2
−0.3
−0.4
0 5 10 15 20

22.4 Conclusion 535
0.08
0.06
0.04
0.02
−0.02
−0.04
−0.06
−0.08
0 5 10 15 20

Chapter 23
Duality Applied to the Optimal Control
and Optimal Design of a Beam Model
23.1 Introduction
In this chapter, first we study duality for the optimal control concerning the
energy minimization of a well-known model of beams. The duality principle devel-
oped includes a concave dual variational formulation suitable to obtain numerical
results. For related results in optimization and convex analysis see [13, 14, 40, 47].
For details on the Sobolev spaces involved, see [1, 26]. We emphasize the dual
problem always has a solution through which we may verify the optimality of the
corresponding primal problem one. However in some situations the primal problem
may not have a global minimum, so that in such cases, if there is no duality gap be-
tween the dual and primal problems, the dual formulation global maximum solution
is a weak limit of minimizing sequences for the primal one. At this point we start to
describe the primal problem.
Consider a straight beam represented by the set Ω = [0, l] where l is the beam
length. Consider also the problem of minimizing the beam energy on a fixed in-
terval [0, T ], under the equilibrium equations, that is, the problem of minimizing
J : U → R, where
T l T l
1 1
J(w, u) = EI(wxx )2 dx dt + ρ A(wt )2 dx dt,
2 0 0 2 0 0
subject to
EIwxxxx + ρ Awtt + Cwt + u(x,t)wx − f (x,t) = 0, in Ω × [0, T ],
w(0,t) = wx (0,t) = 0, in [0, T ]
and the initial conditions
w(x, 0) = w1 (x), and wt (x, 0) = w2 (x) in Ω .
538 23 Duality Applied to the Optimal Control and Optimal Design of a Beam Model
It is worth emphasizing that the boundary conditions refer to a clamped beam at

x = 0 and free at x = l.
Here w denotes the field of vertical displacements, E is the Young modulus, I
is a constant which depends on the cross-sectional geometry, ρ is the material den-
sity, A is cross-sectional area, and C > 0 is a constant which also depends on the
type of material. We emphasize to assume E, I, ρ , A,C to be constant on Ω × [0, T ].
Finally, f ∈ L2 ([0, T ]; L2 (Ω )) is an external dynamical load. The objective here is
to obtain the control u(x,t) which minimizes J, so that such a function satisfies the
constraints:
−M0 ≤ u(x,t) ≤ M0 , in Ω × [0, T ],
and l
|u(x,t)| dx ≤ c, in [0, T ],
0
where M0 ∈ R and 0 < c < M0 l.
In this section we develop a dual variational formulation for the optimal control
problem in question. Our main theoretical result is summarized by the next theorem
in which we redefine the functional J without relabeling it.
Theorem 23.2.1. Let ε > 0 be a small constant. Let J : U → R̄ = R ∪ {+∞} be
redefined by
J(w, u) = G(Λ w) + F(Λ w) + Ind(w, u),
where
Λ : U → Y = [L2 ([0, T ]; L2 (Ω ))]3
is given by
Λ w = {Λ1 w, Λ2 w, Λ3 w},
Λ1 w = wxx , Λ2 w = wx , Λ3 w = wt ,
T l T l
Â B̂
G(Λ w) = (wxx )2 dx dt + (wt )2 dx dt
2 0 0 2 0 0
T l
K
− (wx )2 dx dt, (23.1)
2 0 0
T l
K
F(Λ w) = (wx )2 dx dt
2 0 0

ε T l
+ (wxx )2 dx dt
2 0 0

ε T l
+ (wt )2 dx dt (23.2)
2 0 0
where Â = EI − ε , and B̂ = ρ A − ε .
Moreover,
0, if (w, u) ∈ B
Ind(w, u) = (23.3)
+∞, otherwise,
B = {(w, u) ∈ U | EIwxxxx + ρ Awtt + Cwt + uwx − f = 0 in Ω × [0, T ]},
U = Ũ × B̃,
Ũ = U1 ∩U2 ∩U3 ,
B̃ = B1 ∩ B2 ,
U1 = {w ∈ L2 ([0, T ]; H 2 (Ω )) |
w(0,t) = wx (0,t) = 0, in [0, T ]} (23.4)
U2 = {w ∈ L2 ([0, T ]; H 2 (Ω )) ∩ H 1 ([0, T ]; L2 (Ω )) |
w(x, 0) = w1 (x), and wt (x, 0) = w2 (x), in Ω }, (23.5)
U3 = {w ∈ L2 ([0, T ]; H 2 (Ω )) |
wxx (l,t) = wxxx (l,t) = 0, in [0, T ]} (23.6)
B1 = {u ∈ L2 ([0, T ]; L2 (Ω )) |
−M0 ≤ u(x,t) ≤ M0 , in Ω × [0, T ]}, (23.7)
and
B2 = {u ∈ L2 ([0, T ]; L2 (Ω )) |
l
|u(x,t)| dx ≤ c, in [0, T ]}. (23.8)
0
Also,
A∗ = L2 ([0, T ]; L2 (Ω )) × L2 ([0, T ]; H 2 (Ω )), (23.9)
and we assume that K > 0 is the largest constant such that
G(Λ w) ≥ 0, ∀w ∈ Ũ.
Under such hypotheses we have
inf {J(w, u)} ≥ sup {−J ∗(v∗ , λ )}, (23.10)

(w,u)∈U (v∗ ,λ )∈A∗
where
J ∗ (v∗ , λ ) = (G ◦ Λ )∗ (−Λ ∗ v∗ , λ ) + F1∗ (v∗1 − EI λxx )

+F2∗ (v∗2 , λ ) + F3∗ (v∗3 + ρ Aλt − Cλ ) + λ , f L2 , (23.11)
λ ∈ L2 ([0, T ]; H 2 (Ω )) is an appropriate Lagrange multiplier,

ε T l
F1 (Λ1 w) = (wxx )2 dx dt,
2 0 0
T l
K T l
F2 (Λ2 w, λ ) = inf (λ u(x,t)wx ) dx dt + (wx )2 dx dt ,
u∈B̃ 0 0 2 0 0
and T l
ε
F3 (Λ3 w) = (wt )2 dx dt.
2 0 0
Moreover,
F1∗ (v∗1 − EI λxx ) = sup {v1 , v∗1 − EI λxx L2 − F1 (v1 )}

v1 ∈L2
T l
1
= (v∗1 − EI λxx )2 dx dt, (23.12)
2ε 0 0
F2∗ (v∗2 , λ ) = sup {v2 , v∗2 L2 − F2(v2 , λ )}

v2 ∈L2
T l
1
= sup (v∗2 − λ u(x,t))2 dx dt, (23.13)
u∈B̃ 2K 0 0
F3∗ (v∗3 + ρ Aλt − Cλ ) = sup {v3 , v∗3 + ρ Aλt − Cλ L2 − F3(v3 )}

v3 ∈L2
T l
1
= (v∗3 + ρ Aλt − Cλ )2 dx dt.
2ε 0 0
Also,
(G ◦ Λ )∗ (−Λ ∗ v∗ , λ )

= sup Λ w, −v∗ Y − (G ◦ Λ )(w)
w∈U
T
+ λ (0,t)EIwxxx (0,t) dt
0
T
− λx (0,t)EIwxx (0,t) dt
0
l
− λ (x, T )ρ Awt (x, T ) dx
0
l
+ λ (x, 0)ρ Aw2 (x) dx . (23.14)
0
Under such assumptions there exists (v∗0 , λ0 ) ∈ A∗ such that
−J ∗ (v∗0 , λ0 ) = max {−J ∗(v∗ , λ )},

(v∗ ,λ )∈A∗
so that (w0 , u0 ) such that
∂ (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 )
w0 = ,
∂ w∗
where w∗ = −Λ ∗ v∗ and
T l
1
F2∗ (v∗02 , λ0 ) = (v∗02 − λ0 u0 (x,t))2 dx dt
2K 0 0
are also such that
(G ◦ Λ )(w0 ) + F1(Λ1 w0 ) + F2∗∗ (Λ2 w0 , λ0 )

+F3 (Λ3 w0 , λ0 ) + λ0, EIw0xxxx L2
+λ0 , ρ Aw0tt + Cw0t L2 − λ0 , f L2
= min{(G ◦ Λ )(w) + F1(Λ1 w) + F2∗∗ (Λ2 w, λ0 )
w∈Ũ
+F3 (Λ3 w, λ0 ) + λ0, EIwxxxx L2
+λ0 , ρ Awtt + Cwt L2 − λ0 , f L2 }
= max {−J ∗ (v∗ , λ )} = −J ∗ (v∗0 , λ0 ), (23.15)
(v∗ ,λ )∈A∗
where
(G ◦ Λ )(w0 ) + F1(Λ1 w0 ) + F2(Λ2 w0 , λ0 )

+F3 (Λ3 w0 , λ0 ) + λ0, EIw0xxxx L2
+λ0 , ρ Aw0tt + Cw0t L2 − λ0, f L2
= (G ◦ Λ )(w0 ) + F(Λ w0 )
+λ0 , EIw0xxxx + ρ Aw0tt + Cw0t + u0w0x − f L2 .
Furthermore, we emphasize to denote L2 (Ω × [0, T ]) ≡ L2 :
Λ w, v∗ Y = Λ1 w, v∗1 L2 + Λ2 w, v∗2 L2 + Λ3 w, v∗3 L2 ,
where T l
g, hL2 = g(x,t)h(x,t) dx dt
0 0
and
F ∗∗ (Λ2 w, λ ) = sup {Λ2 w, v∗2 L2 − F2∗ (v∗2 , λ )}.
v∗2 ∈A∗
Finally, if K > 0 above specified is such that the optimal inclusion


∂ (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 )
Λ2 ∈ ∂v∗2 F2∗ (v∗02 , λ0 ) (23.16)
∂ w∗
stands for
∂ F̃2∗ (v∗02 , λ0 , u0 )
Λ2 w0 = ,
∂ v∗2
where
∂ (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 )
w0 = ,
∂ w∗
and
T l
1
F̃2∗ (v∗2 , λ , u) = (v∗2 − λ u(x,t))2 dx dt;
2K 0 0
then
J(w0 , u0 ) = min {J(w, u)} = max {−J ∗ (v∗ , λ )} = −J ∗ (v∗0 , λ0 ).

(w,u)∈U (v∗ ,λ )∈A∗
Proof. Observe that
J(w, u) = G(Λ w) + F(Λ w) + Ind(u, w)

≥ G(Λ w) + F(Λ w)
+λ , EIwxxxx + ρ Awtt + Cwt + uwx − f L2 , (23.17)
∀(w, u) ∈ U, (v∗ , λ ) ∈ A∗ , so that
J(w, u) = G(Λ w) + F(Λ w) + Ind(u, w)

≥ Λ w, v∗ Y + G(Λ w)
−Λ1 w, v∗1 L2 + F1 (Λ1 w)
−Λ2 w, v∗2 L2 + F2 (Λ2 w, λ )
−Λ3 w, v∗3 L2 + F3 (Λ3 w)
+λxx , EIwxx L2 − λt , ρ Awt L2
T
+λ ,Cwt L2 − λ (0,t)EIwxxx (0,t) dt
0
T
+ λx (0,t)EIwxx (0,t) dt
0
l
+ λ (x, T )ρ Awt (x, T ) dx
0
l
− λ (x, 0)ρ Aw2 (x) dx
0
−λ , f L2 , (23.18)
∀(w, u) ∈ U, (v∗ , λ ) ∈ A∗ . Thus,


J(w, u) ≥ inf Λ w, v∗ Y + G(Λ w)
w∈Ũ
T
− λ (0,t)EIwxxx (0,t) dt
0
T
+ λx (0,t)EIwxx (0,t) dt
0
l
+ λ (x, T )ρ Awt (x, T ) dx
0
l
− λ (x, 0)ρ Aw2 (x) dx
0
+ inf {−v1 , v∗1 − EI λxx L2 + F1(v1 )}
v1 ∈L2
+ inf {−v2 , v∗2 L2 + F2(v2 , λ )}
v2 ∈L2
+ inf {−v3 , v∗3 + ρ Aλt − Cλ L2 + F3(v3 )}
v3 ∈L2
−λ , f L2 , (23.19)
∀(w, u) ∈ U, (v∗ , λ ) ∈ A∗ .
Therefore,
J(w, u) ≥ −(G ◦ Λ )∗ (−Λ ∗ v∗ , λ ) − F1∗ (v∗1 − EI λxx ) − F2∗ (v∗2 , λ )

−F3∗ (v∗3 + ρ Aλt − Cλ ) − λ , f L2 , (23.20)
∀(w, u) ∈ U, (v∗ , λ ) ∈ A∗ .
Hence,
inf {J(w, u)} ≥ sup {−J ∗ (v∗ , λ )}. (23.21)

(w,u)∈U (v∗ ,λ )∈A∗
Since −J ∗ (v∗ , λ ) is concave, coercive, continuous, and therefore weakly upper

semicontinuous, from an application of the direct method of variations (considering
it is a standard procedure, here we omit more details) there exists (v∗0 , λ0 ) such that
−J(v∗0 , λ0 ) = max {−J ∗ (v∗ , λ )}.

(v∗ ,λ )∈A∗
Such an optimal point is attained through the extremal equations:

∂ (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 ) ∂ F1∗ (v∗01 − EI λ0xx )
Λ1 − = θ, (23.22)
∂ w∗ ∂ v∗1

∂ (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 )
Λ2 ∈ ∂v∗2 F2∗ (v∗02 , λ0 ), (23.23)
∂ w∗

∂ (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 ) ∂ F3∗ (v∗03 + ρ Aλ0t − Cλ0 )
Λ3 − = θ. (23.24)
∂ w∗ ∂ v∗3
Hence, for w0 such that
∂ (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 )
w0 = , (23.25)
∂ w∗
we get
∂ F1∗ (v∗01 − EI λ0xx )
Λ1 w0 − = θ, (23.26)
∂ v∗1
Λ2 w0 ∈ ∂v∗2 F2∗ (v∗02 , λ0 ), (23.27)
and
∂ F3∗ (v∗03 + ρ Aλ0t − Cλ0 )
Λ3 w0 − = θ. (23.28)
∂ v∗3
Thus, from the last three relations, we obtain
F1∗ (v∗01 − EI λ0xx ) = Λ1 w0 , v∗01 − EI λ0xx L2 − F1(Λ1 w0 ),
F2∗ (v∗02 , λ0 ) = Λ2 w0 , v∗02 L2 − F2∗∗ (Λ2 w0 , λ0 ),

and
F3∗ (v∗03 + ρ Aλ0t − Cλ0) = Λ3 w0 , v∗03 + ρ Aλ0t − Cλ0L2 − F3(Λ3 w0 ).
From (23.25), the extremal condition concerning the variation in λ and these last
three equalities, we get
J ∗ (v∗0 , λ0 ) = (G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 ) + F1∗ (v∗01 − EI λ0xx ) + F2∗ (v∗02 , λ0 )

+F3∗ (v∗03 + ρ Aλ0t − Cλ0 ) + λ0, f L2
= −(G ◦ Λ )(w0 ) − F1(Λ1 w0 ) − F2∗∗ (Λ2 w0 , λ0 )
−λ0 , EIw0xxxx L2 − λ0 , ρ Aw0tt + Cw0t L2
−F3 (Λ3 w0 ) + λ0 , f L2 . (23.29)
Similarly as above, we may infer that

inf {G(Λ w) + F1(Λ1 w) + F2∗∗ (Λ2 w, λ0 )
w∈Ũ
+F3 (Λ3 w) + λ0 , EIwxxxx + ρ Awtt + Cwt − f L2 }
≥ −J ∗ (v∗ , λ0 ), ∀v∗ such that (v∗ , λ0 ) ∈ A∗ . (23.30)
min{G(Λ w) + F1 (Λ1 w) + F2∗∗ (Λ2 w, λ0 )

w∈Ũ
+F3 (Λ3 w) + λ0 , EIwxxxx + ρ Awtt + Cwt − f L2 }
= G(Λ w0 ) + F1(Λ1 w0 ) + F2∗∗ (Λ2 w0 , λ0 )
+F3 (Λ3 w0 ) + λ0 , EI(w0 )xxxx + ρ A(w0)tt + C(w0 )t − f L2
= −J ∗ (v∗0 , λ0 ) = max {−J ∗ (v∗ , λ )}. (23.31)
(v∗ ,λ )∈A∗
Finally, if K > 0 above specified is such that
∂ F̃2∗ (v∗02 , λ0 , u0 )
Λ2 w0 = ,
∂ v∗2
where
T l
1
F̃2∗ (v∗2 , λ , u) = (v∗2 − λ u(x,t))2 dx dt,
2K 0 0
denoting
T l T l
K
F̃2 (Λ2 w, λ , u) = (λ u(x,t)wx ) dx dt + (wx )2 dx dt,
0 0 2 0 0
we have
F2∗ (v∗02 , λ0 ) = F̃2∗ (v∗02 , λ0 , u0 ) = Λ2 w0 , v∗02 L2 − F̃2(Λ2 w0 , λ0 , u0 )
= Λ2 w0 , v∗02 L2 − F2(Λ2 w0 , λ0 ). (23.32)
Moreover, the variation in λ in the dual formulation gives us the extremal inclu-
sion:
∗ ∗ ∗ ∗
∂ F1 (v01 − EI λ0xx ) ∂ F3 (v03 + ρ Aλ0t − Cλ0)
EI + ρ A
∂ v∗1 xx ∂ v∗3 t
∂ F3∗ (v∗03 + ρ Aλ0t − Cλ0)
+C − f ∈ [∂v∗2 F2∗ (v∗02 , λ0 )](−u0 ), (23.33)
∂ v∗3
so that
EIw0xxxx + ρ Aw0tt + Cwt + u0w0x − f = 0, in Ω × [0, T ].
Hence, from this last equation, (23.29) and (23.32) we have
J(w0 , u0 ) = G(Λ w0 ) + F(Λ w0 ) + Ind(w0 , u0 )
= G(Λ w0 ) + F1(Λ1 w0 )
+F̃2 (Λ2 w0 , λ0 , u0 ) + F3(Λ3 w0 )
+λ0 , EIw0xxxx + ρ Aw0tt + Cw0t + u0w0x − f L2
= −(G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 ) − F1∗ (v∗01 ) − F̃2∗ (v∗02 , λ0 , u0 )
−F3∗ (v∗03 ) − λ0 , f L2
= −(G ◦ Λ )∗ (−Λ ∗ v∗0 , λ0 ) − F1∗ (v∗01 ) − F2∗ (v∗02 , λ0 )
−F3∗ (v∗03 ) − λ0 , f L2
= −J ∗ (v∗0 , λ0 ). (23.34)

J(w0 , u0 ) = min {J(w, u)} = max {−J ∗ (v∗ , λ )} = −J ∗ (v∗0 , λ0 ).
(w,u)∈U (v∗ ,λ )∈A∗

23.3 Some Closely Related Simpler Examples with Numerical

Results
Consider a straight beam with circular cross-sectional area given by A(x), where
x ∈ [0, l], l being the beam length and [0, l] = Ω its axis. Suppose such a beam is sim-
ply supported, so that w ∈ U, where w : Ω → R is the field of vertical displacements
and
U = {w ∈ W 2,2 (Ω ) : w(0) = w(l) = 0}.
Also, the beam in question is assumed to be under a compressive axial load P
applied at x = l. We shall look for the optimal distribution A(x) which maximizes
the buckling load P, where the following designed constraints must be satisfied:

A(x) dx = V = cAmax l,
Ω
where 0 < c < 1 and

0 < Amin ≤ A(x) ≤ Amax , in Ω .
Hence, our optimization problem translates into minimizing −P, subject to
c0 (A(x)2 w,xx ),xx + Pw,xx = 0, in Ω ,
where c0 > 0 is an appropriate constant to be specified, so that P is such that

c0 P
A(x)2 w2,xx dx − w2,x dx ≥ 0,
2 Ω 2 Ω
∀w ∈ U. Furthermore, as above indicate, we must have
0 < Amin ≤ A(x) ≤ Amax , in Ω ,
and
Ω
where 0 < c < 1.
Observe that from the concerned constraints

c0 P
A(x)2 w2,xx dx = w2,x dx,
2 Ω 2 Ω
so that through the appropriate constraint for the concerned eigenvalue problem, that
is,
w2,x dx = 1,
Ω
we get
c0 P
A(x)2 w2,xx dx = .
2 Ω 2
23.3 Some Closely Related Simpler Examples with Numerical Results 547
Hence, we may define the above optimization problem by the minimization of
J(w, A) + Ind(w, P, A),
where
−c0
J(w, A) = A(x)2 w2,xx dx
2 Ω
and
0, if (w, P, A) ∈ A∗ ,
Ind(w, P, A) = (23.35)
+∞, otherwise,
where
A∗ = A1 ∩ A2 ∩ A3 ,
A1 = {(w, P, A) ∈ U × R+ × L2 (Ω ) such that

c0 (A(x)2 w2,xx ),xx − Pw,xx = 0, in Ω } (23.36)
A2 = {(P, A) ∈ R+ × L2 (Ω ) such that

˜ P, A) ≥ 0, ∀w ∈ U}
J(w, (23.37)
where
˜ P, A) = c0
J(w, A(x)2 w2,xx dx −
P
w2,x dx.
2 Ω 2 Ω
Finally,

A3 = w ∈ U | (w,x )2 dx = 1 .
Ω
At this point, denoting

c0
G(w,xx , A) = A(x)2 (w,xx )2 dx,
2 Ω
we define the extended functional Jλ (w, P, A) by
Jλ (w, P, A) = −G(w,xx , A)
+λ , (c0 A(x)2 w,xx )xx + Pw,xx L2 (23.38)
where λ is an appropriate Lagrange multiplier.

Observe that
Jλ (w, P, A) = w,xx , v∗ L2 − G(w,xx , A)

−w,xx , v∗ L2 + λ , c0 (A(x)2 w,xx )xx + Pw,xx L2 , (23.39)
so that
Jλ (w, P, A) ≤ sup {v, v∗ L2 − G(v, A)}+

v∈L2
sup {−w,xx , v∗ L2 + λ , c0 (A(x)2 w,xx )xx + Pw,xx L2 }, (23.40)
w∈U
and therefore
Jλ (w, P, A) ≤ G∗ (v∗ , A) + Ind1(v∗ , λ , P, A),
where
1 (v∗ )2
G∗ (v∗ , A) = dx,
2c0 Ω A(x)2
and
0, if (v∗ , λ , P, A) ∈ B∗ ,
Ind1 (v∗ , λ , P, A) = (23.41)
+∞, otherwise,
where
B∗ = {(v∗ , λ , P, A) ∈ L2 × L2 × R+ × L2 |
v∗,xx − c0 (A(x)2 λxx ),xx + Pwxx = 0, in Ω ,
and v∗ (0) = v∗ (l) = 0}. (23.42)
Summarizing the partial duality principle obtained, we have
Jλ (w, P, A) ≤ inf {G∗ (v∗ , A) + Ind1(v∗ , λ , P, A)}.

v∗ ∈L2
Having this inequality in mind, we suggest the following algorithm to get critical
points relating the original problem. It is worth emphasizing we have not formally
proven its convergence:
1. Set k = 1 and choose w̃01 ∈ U such that

[(w̃01 ),x ]2 dx = 1.
Ω
2. Set A1 = cAmax .
3. Set n = 1 and w̃1k = w̃0k .
4. Calculate wnk ∈ U by solving the equation
c0 (Ak (x)2 (wnk ),xx ),xx + (w̃nk ),xx = 0, in Ω .
5. Define
w̃kn+1 = wnk /Skn ,
where 7
Skn = [(wnk ),x ]2 dx.
Ω
6. Set n → n + 1 and go to step (4), up to the satisfaction of an appropriate conver-
gence criterion.
7. Define
w̃0k+1 = lim w̃nk ,
n→∞
1
Pk = lim .
n→∞ Skn
8. Define
v∗ = c0 Ak (x)(w̃0k+1 )xx ,
and obtain Ak+1 (x) by
Ak+1 (x) = argminA∈C∗ {G∗ (v∗ , A)},
where
C∗ = C1 ∩C2 ,

C1 = {A ∈ L2 | A(x) dx = V = cAmax l},
Ω
and
C2 = {A ∈ L2 | 0 < Amin ≤ A(x) ≤ Amax , in Ω }.
9. Set k → k + 1 and go to step (3), up to the satisfaction of an appropriate conver-
gence criterion.
23.3.1 Numerical Results
We present numerical results l = 1.0, c0 = 105 , c = 0.7, Amin /α = 0.3, and

Amax /α = 1.0 for an appropriate α > 0. Here units refer to the international system.
We have obtained the buckling load P = 6.0777 · 105, and for the optimal A(x),
see Fig. 23.1. The eigenvalue P1 = 4.8320 · 105 corresponds to A(x) = cAmax . Ob-
serve that P > P1 as expected. Anyway, we have obtained just a critical point; at this
point, we are not able to guarantee global optimality.
23.3.2 A Dynamical Case
In this section we develop analysis for a beam model dynamics, similarly as in

the last section. Specifically, we consider the motion of a beam on an interval [0, T ].
The beam model in question is the same as in the last section, so that the dynamical
equation is given by
c0 (A(x)2 w(x,t),xx )xx + ρ A(x)w(x,t),tt = 0, in Ω = [0, l],

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0 0.2 0.4 0.6 0.8 1
Fig. 23.1 Optimal distribution of area A(x)/α , intending to maximize the buckling load
where ρ denotes the beam density and w(x,t) : Ω × [0, T ] → R denotes the field
of vertical displacements. The motion results from proper initial conditions not yet
specified.
For the last equation we look for a solution of the form
w(x,t) = eiω t u(x),
where ω is the first natural frequency.

Replacing such a solution in the last equation we get

eiω t c0 (A(x)2 u,xx )xx − ω 2ρ A(x)u(x) = 0, in Ω ,
so that
(c0 (A(x)2 u,xx )xx − ω 2ρ A(x)u(x) = 0, in Ω . (23.43)
At this point we consider the problem of finding A(x) which maximizes the fun-
damental frequency ω , subject to (23.43):

u2 dx = 1,
Ω

1 ω2
co A(x)2 (u,xx )2 dx − ρ A(x)u2 dx ≥ 0,
2 Ω 2 Ω
∀u ∈ U.
Moreover, the following design constraints must be satisfied:

Ω
and
0 < Amin ≤ A(x) ≤ Amax , in Ω .
This problem is mathematically similar to the previous one, related to the maxi-
mization of the buckling load.
Thus, similarly as in the last section, we define Jλ (u, A, ω ) by
Jλ (u, ω , A) = −G(u,xx , A)
+λ , (c0 A(x)2 u,xx )xx − ω 2 ρ A(x)uL2 (23.44)
where
c0
G(u,xx , A) = A(x)2 (u,xx )2 dx,
2 Ω
and λ is an appropriate Lagrange multiplier.
Observe that
Jλ (w, ω , A) = u,xx , v∗ L2 − G(u,xx , A)

−u,xx , v∗ L2 + λ , (c0 A(x)2 u,xx )xx − ω 2 ρ uL2 , (23.45)
so that
Jλ (w, ω , A) ≤ sup {v, v∗ L2 − G(v, A)

v∈L2
+ sup {−u,xx , v∗ L2 + λ , c0 (A(x)2 u,xx )xx − ω 2 ρ A(x)uL2 },
w∈U
and therefore
Jλ (w, P, A) ≤ G∗ (v∗ , A) + Ind2(v∗ , λ , P, A),
where
1 (v∗ )2
G∗ (v∗ , A) = dx,
2c0 Ω A(x)2
and
0, if (v∗ , λ , ω , A) ∈ B∗ ,
Ind1 (v∗ , λ , ω , A) = (23.46)
+∞, otherwise,
where
B∗ = {(v∗ , λ , ω , A) ∈ L2 × L2 × R+ × L2 |
v∗,xx − c0 (A(x)2 λxx ),xx − ω 2ρ A(x)λ = 0, in Ω ,
and v∗ (0) = v∗ (l) = 0}. (23.47)
Summarizing the partial duality principle obtained, we have
Jλ (w, ω , A) ≤ inf {G∗ (v∗ , A) + Ind1(v∗ , λ , ω , A)}.

v∗ ∈L2
Having such an inequality in mind, we develop an algorithm to obtain critical points,

similar to that of the previous sections (we do not give the details here).
For the same constraints in A(x) (in particular c = 0.7) as for the previous exam-
ple, again for l = 1, c0 = 105, and ρ = 10, we obtain the optimal ω 2 = 7.3885 · 105.
For the optimal A(x) see Fig. 23.2.
For the case A(x) = cAmax we have obtained ω12 = 6.8070 · 105. Units refer to the
international system.
Observe that the optimal ω > ω1 as naturally expected. Anyway, we emphasize
to have calculated just a critical point. Again at this point we cannot guarantee global
optimality.
0.9
0.8
0.7
0.6
0.5
0.4
0 0.2 0.4 0.6 0.8 1
Fig. 23.2 Optimal distribution of area A(x)/α , intending to maximize ω
23.4 Conclusion
In this chapter we develop a concave dual variational formulation for the opti-
mal control of a well-known beam model. In practice, the results may be applied
to the energy minimization with piezoelectric actuators, through which the beam
vibration may be controlled (see [49] for related results). In a second step, we study
the optimal design for this same beam model, with the objective of maximizing its
buckling load and fundamental frequency, respectively. In both cases the numerical
results obtained are consistent with the problem physics. Finally, we emphasize the
approach here developed may be applied to many other situations, such as for plate
and shell models for example.
References
1. R.A. Adams, Sobolev Spaces (Academic Press, New York, 1975)

2. R.A. Adams, J.F. Fournier, Sobolev Spaces, 2nd edn. (Elsevier, Oxford, UK, 2003)
3. G. Allaire, Shape Optimization by the Homogenization Method (Springer, New York, 2002)
4. J.F. Annet, Superconductivity, Superfluids and Condensates. Oxford Master Series in
Condensed Matter Physics, Oxford University Press, New YorK (2010)
5. H. Attouch, G. Buttazzo, G. Michaille, Variational Analysis in Sobolev and BV Spaces.
MPS-SIAM Series in Optimization (SIAM, Philadelphia, 2006)
6. G. Bachman, L. Narici, Functional Analysis (Dover Publications, New York, 2000)
7. J.M. Ball, R.D. James, Fine mixtures as minimizers of energy. Arch. Ration. Mech. Anal. 100,
15–52 (1987)
8. M.P. Bendsoe, O. Sigmund, Topology Optimization, Theory Methods and Applications
(Springer, Berlin, New York 2003)
9. F. Bethuel, H. Brezis, F. Helein, Ginzburg-Landau Vortices (Birkhäuser, Basel, 1994)
10. F. Botelho, Variational Convex Analysis. Ph.D. Thesis (Virginia Tech, Blacksburg, 2009)
11. F. Botelho, Dual variational formulations for a non-linear model of plates. J. Convex Anal.
17(1), 131–158 (2010)
12. F. Botelho, Existence of solution for the Ginzburg-Landau system, a related optimal control
problem and its computation by the generalized method of lines. Appl. Math. Comput. 218,
11976–11989 (2012)
13. F. Botelho, Topics on Functional Analysis, Calculus of Variations and Duality (Academic
Publications, Sofia, 2011)
14. F. Botelho, On duality principles for scalar and vectorial multi-well variational problems.
Nonlinear Anal. 75, 1904–1918 (2012)
15. F. Botelho On the Lagrange multiplier theorem in Banach spaces, Comput. Appl. Math. 32,
135–144 (2013)
16. H. Brezis, Analyse Fonctionnelle (Masson, Paris, 1987)
17. L.D. Carr, C.W. Clark, W.P. Reinhardt, Stationary solutions for the one-dimensional nonlinear
Schrödinger equation. I- case of repulsive nonlinearity. Phys. Rev. A 62, 063610 (2000)
18. I.V. Chenchiah, K. Bhattacharya, The relaxation of two-well energies with possibly unequal
moduli. Arch. Rational Mech. Anal. 187, 409–479 (2008)
19. M. Chipot, Approximation and Oscillations. Microstructure and Phase Transition, The IMA
Volumes in Mathematics and Applications, Oxford, UK, vol. 54, 1993, pp. 27–38
20. R. Choksi, M.A. Peletier, J.F. Williams, On the phase diagram for microphase separation of
diblock copolymers: an approach via a nonlocal Cahn-Hilliard functional. SIAM J. Appl.
Math. 69(6), 1712–1738 (2009)
21. P. Ciarlet, Mathematical Elasticity, vol. I, Three Dimensional Elasticity (North Holland,
Elsevier, 1988)
22. P. Ciarlet, Mathematical Elasticity, vol. II Theory of Plates (North Holland, Elsevier, 1997)
to Non-Convex Variational Models, DOI 10.1007/978-3-319-06074-3,
554 References
23. P. Ciarlet, Mathematical Elasticity, vol. III – Theory of Shells (North Holland, Elsevier, 2000)
24. B. Dacorogna, Direct Methods in the Calculus of Variations (Springer, New York, 1989)
25. I. Ekeland, R. Temam, Convex Analysis and Variational Problems (Elsevier-North Holland,
Amsterdam 1976).
26. L.C. Evans, Partial Differential Equations, Graduate Studies in Mathematics, vol. 19
(American Mathematical Society, Providence, Rhode Island, 1998)
27. U. Fidalgo and P. Pedregal: A General Lower Bound for the Relaxation of an Optimal Design
Problem with a General Quadratic Cost Functional and a General Linear State Equation,
Journal of Convex Analysis 19, number 1, 281–294 (2012)
28. N.B. Firoozye, R.V. Khon, Geometric Parameters and the Relaxation for Multiwell Ener-
gies. Microstructure and Phase Transition, the IMA volumes in mathematics and applications,
vol. 54, 85–110 (Oxford, UK, 1993)
29. I. Fonseca, G. Leoni, Modern Methods in the Calculus of Variations, L p Spaces (Springer,
New York, 2007)
30. A. Galka, J.J. Telega, Duality and the complementary energy principle for a class of geomet-
rically non-linear structures. Part I. Five parameter shell model Arch. Mech. 47, 677–698;
Part II. Anomalous dual variational principles for compressed elastic beams. Arch. Mech. 47,
699–724 (1995)
31. D.Y. Gao, G. Strang, Geometric nonlinearity: potential energy, complementary energy and the
gap function. Q. J. Appl. Math. 47, 487–504 (1989a)
32. D.Y. Gao, On the extreme variational principles for non-linear elastic plates. Q. Appl. Math.
XLVIII(2), 361–370 (1990)
33. D.Y. Gao, Pure complementary energy principle and triality theory in finite elasticity. Mech.
Res. Comm. 26(1), 31–37 (1999)
34. D.Y. Gao, General analytic solutions and complementary variational principles for large
deformation nonsmooth mechanics. Meccanica 34, 169–198 (1999)
35. D.Y. Gao, Finite deformation beam models and triality theory in dynamical post-buckling
analysis. Int. J. Non Linear Mech. 35, 103–131 (2000)
36. D.Y. Gao, Duality Principles in Nonconvex Systems, Theory, Methods and Applications
(Kluwer, Dordrecht, 2000)
37. M. Giaquinta, S. Hildebrandt, Calculus of Variations I. A Series of Comprehensive Studies
in Mathematics, vol. 310 (Springer, Berlin, 1996)
38. M. Giaquinta, S. Hildebrandt, Calculus of Variations II. A Series of Comprehensive Studies
in Mathematics, vol. 311 (Springer, Berlin, 1996)
39. E. Giusti, Direct Methods in the Calculus of Variations (World Scientific, Singapore, reprint,
2005)
40. K. Ito, K. Kunisch, Lagrange Multiplier Approach to Variational Problems and Applications.
Advances in Design and Control (SIAM, Philadelphia, 2008)
41. A. Izmailov, M. Solodov, Otimização, vol. 2 (IMPA, Rio de Janeiro, 2007)
42. A. Izmailov, M. Solodov, Otimização, vol. 1, 2nd edn. (IMPA, Rio de Janeiro, 2009)
43. R.D. James, D. Kinderlehrer, Frustration in ferromagnetic materials. Continuum Mech.
Thermodyn. 2, 215–239 (1990)
44. H. Kronmuller, M. Fahnle, Micromagnetism and the Microstructure of Ferromagnetic Solids
(Cambridge University Press, Cambridge, UK, 2003)
45. L.D. Landau, E.M. Lifschits, Course of Theoretical Physics, Vol. 5- Statistical Physics, Part 1
(Butterworth-Heinemann, Elsevier, reprint, 2008)
46. E.M. Lifschits, L.P. Pitaevskii, Course of Theoretical Physics, Vol. 9- Statistical Physics,
Part 2 (Butterworth-Heinemann, Elsevier, reprint, 2002)
47. D.G. Luenberger, Optimization by Vector Space Methods (Wiley, New York, 1969)
48. G.W. Milton, Theory of composites. Cambridge Monographs on Applied and Computational
Mathematics (Cambridge University Press, Cambridge 2002)
49. A. Molter, O.A.A. Silveira, J. Fonseca, V. Bottega, Simultaneous Piezoelectric Actuator and
Sensor Placement Optimization and Control Design of Manipulators with Flexible Links
Using SDRE Method, Mathematical Problems in Engineering, Hindawi, 2010, pp. 1–23
References 555
50. P. Pedregal, Parametrized measures and variational principles, Progress in Nonlinear Differ-
ential Equations and Their Applications, vol. 30 (Birkhauser, Basel, 1997)
51. P. Pedregal and B. Yan, On Two Dimensional Ferromagnetism. Proceedings of the Royal So-
ciety of Edinburgh, 139A, 575–594 (2009)
52. M. Reed, B. Simon, Methods of Modern Mathematical Physics, Volume I, Functional Analysis
(Reprint Elsevier, Singapore, 2003)
53. S.M. Robinson, Strongly regular generalized equations. Math. Oper. Res. 5, 43–62 (1980)
54. R.T. Rockafellar, Convex Analysis (Princeton University Press, Princeton, New Jersey, 1970)
55. R.C. Rogers, Nonlocal variational problems in nonlinear electromagneto-elastostatics, SIAM
J. Math. Anal. 19(6), 1329–1347 (1988)
56. R.C. Rogers, A nonlocal model of the exchange energy in ferromagnet materials. J. Integr.
Equat. Appl. 3(1), 85–127 (1991)
57. W. Rudin, Real and Complex Analysis, 3rd edn. (McGraw-Hill, New York, 1987)
58. W. Rudin, Functional Analysis, 2nd edn. (McGraw-Hill, New York, 1991)
59. H. Royden, Real Analysis, 3rd edn. (Prentice Hall India, New Delhi, 2006)
60. O. Sigmund, A 99 line topology optimization code written in Matlab. Struc. Muldisc. Optim.
21, 120–127 (2001)
61. J.J.A. Silva, A. Alvin, M. Vilhena, C.Z. Petersen, B. Bodmann, On a Closed form Solu-
tion of the Point Kinetics Equations with Reactivity Feedback of Temperature. International
Nuclear Atlantic Conference- INAC- ABEN, Belo Horizonte, MG, Brazil, October 24–28,
2011, ISBN: 978 -85-99141-04-05
62. E.M. Stein, R. Shakarchi, Real Analysis, Princeton Lectures in Analysis III, (Princeton
University Press, 2005)
63. J.C. Strikwerda, Finite Difference Schemes and Partial Differential Equations, 2d edn. SIAM
(Philadelphia, 2004)
64. D.R.S. Talbot, J.R. Willis, Bounds for the effective constitutive relation of a nonlinear
composite. Proc. R. Soc. Lond. 460, 2705–2723 (2004)
65. 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 non-convex boundary condition. Minimax approach, C.R. Acad. Sci.
Paris, Serie II 308, 1313–1317 (1989)
66. R. Temam, Navier-Stokes Equations (AMS, Chelsea, reprint 2001)
67. J.F. Toland, A duality principle for non-convex optimisation and the calculus of variations.
Arch. Rath. Mech. Anal. 71(1), 41–61 (1979)
68. J.L. Troutman, Variational Calculus and Optimal Control, 2nd edn. (Springer, New York,
1996)
Index
A Closure, 5, 28, 51, 88, 92, 132, 150, 180, 185

Absolute continuous measures, 109, 113 Compact imbedding, 216–221
Adjoint operator, 58–64, 89, 255, 263, 283, Compactness, 9–10, 13–19, 48–52, 217–219,
428 274
Affine continuous, 253, 254 Compact operator, 64–66
Arzela–Ascoli theorem, 19–21, 194, 217 Compact set, 9, 10, 15–18, 80, 134, 137, 151,
154, 157–159, 171, 182, 191–195, 228,
B 267, 276
Baire category theorem, 26–28 Completeness, 11, 364
Balanced sets, 8 Complex Ginzburg–Landau system, 363,
Banach space, 11, 13, 41, 42, 44, 46–53, 55, 373–392
57, 58, 60, 61, 63, 64, 73, 74, 177, 180, Complex Hilbert space, 36, 77, 94
181, 183, 225, 226, 231, 233, 234, 252, Composites, 218, 447, 451–455
253, 255, 257, 261, 263, 264, 266, 269,
Conductivity, 447, 451–455
282, 283, 287, 288, 292, 293, 297, 299,
Cone, 63, 287, 288, 290, 300, 304
300, 304, 307, 308, 311, 345, 365, 383,
384, 427, 428, 431 Connected set, 225, 248, 321, 322, 331, 337,
343, 345, 375, 377, 394, 403, 407, 426,
Beam model, 537–552
433, 435, 437, 447, 450, 452, 456, 465,
Bipolar functional, 254
477, 484, 517
Borel set, 100, 153, 160, 166
Bounded function, 22 Constrained optimization, 287–317
Bounded operator, 57–58 Continuity, 5–8, 22–24, 44, 45, 49, 84, 169,
Bounded self-adjoint operator, 67–69, 77–86 200, 229, 249, 255–256, 518
Bounded set, 7, 22, 51, 64, 170, 172, 186, 187, Continuous function, 5, 42, 43, 79, 80,
191, 196, 197, 199, 208, 213, 217, 220, 180–181, 186, 197, 200, 248, 253, 257,
363, 445, 493 271, 427
Continuous operator, 23–24, 428, 429
C Control, 377–381, 390–392, 456–459,
Calculus of variations, 225–249, 263, 279, 537–552
283, 364, 449, 459, 499 Convergence, 9, 41, 49, 80, 97, 103–106, 112,
Cauchy–Schwarz inequality, 31, 32, 110 125, 126, 128, 129, 178, 180, 217, 230,
Cauchy sequence, 11, 13, 14, 19, 26, 32, 33, 354, 359, 381, 385, 401, 407, 418, 423,
38, 70, 93, 170, 177, 183, 201, 209, 213, 471, 473, 490, 548
216, 220, 281, 297, 308, 473 Convex analysis, 251–285, 341, 343, 362, 425,
Closed graph theorem, 30–31 446, 516, 537
Closed set, 4–6, 13, 49, 134–135, 137, 138, Convex envelope, 253, 254
144, 160, 168 Convex function, 251–260, 288, 291, 359, 434
to Non-Convex Variational Models, DOI 10.1007/978-3-319-06074-3,
558 Index
Convex set, 7, 8, 48, 208, 251–260, 291, 302, Gram–Schmidt orthonormalization, 39–40
307, 383 Graph, 30, 88, 91, 271, 529
Countably compact set, 10, 16
H
D Hahn–Banach theorem, 41–56, 62, 185, 303
Dense set, 20, 26, 40, 52, 145 Hahn decomposition, 108–109
Distribution, 167–174, 183, 357, 456, 460, Hausdorff space, 7, 9, 48, 149–153
462, 463, 484, 486, 546, 550, 552 Helly lemma, 51
du Bois–Reymond lemma, 243–246 Hilbert space, 31–37, 39, 40, 59, 64, 66, 75,
Duality, 261–264, 280–285, 291–292, 77, 86, 88, 94, 110, 184
321–341, 343–362, 393–424, 437–446, Hölder continuous, 181
451–455, 465–491, 493–516, 537–552 Hyperplane, 44, 46, 48, 256, 287, 289
Duality principle, 321–339, 341, 343,
345–353, 361, 362, 366, 368, 394–400, I
425–464, 466–469, 478–484, 494, Inequality, 31, 32, 37, 47, 76, 105, 110, 136,
496–502, 516, 538–545, 548, 551 152, 176, 177, 179, 183, 184, 190,
Dual space, 42–43, 171, 427 192, 203–205, 207–209, 215, 219, 220,
Dual variational formulation, 343, 365–368, 299–303, 357, 358, 364, 366, 368, 381,
447, 476, 491, 493, 516, 537, 538, 552 409, 410, 482
Inner product, 31, 36, 110, 184
E Integral, 102–103, 106
Egorov theorem, 144–145 Integration, 99–128, 147–166, 199, 245, 248,
Eigenvalue, 73, 76, 465, 470, 474, 546, 549 322, 344, 375, 426, 465, 477, 493
Ekeland variational principle, 280–285 Interior, 5, 8, 12, 25, 26, 27, 28, 30, 73, 74,
Elasticity, 321–341, 343, 441, 447, 477–491 129, 252, 288, 289, 302
ε -net, 15, 17, 18, 191 Inverse mapping theorem, 30–31, 61, 293
Epigraph, 252, 253 Inverse operator, 73
Euler system, 517, 518
J
F Jordan, 109
Fatou’s lemma, 104–106
Finite-dimensional space, 4 K
Fluid mechanics, 517–535 Kakutani theorem, 49, 51
Fréchet derivative, 231, 312, 316
Fréchet differentiability, 231, 239, 292, 300, L
304, 309, 313 Lagrange multiplier, 288, 290, 292–305, 312,
Frequency, 550, 552 443, 466, 480, 482, 540, 547, 551
Fubini theorem, 122–128, 190, 193, 208 Lebesgue decomposition, 113–114
Function, 5, 21, 22, 99, 101–106, 138–146, Lebesgue dominated convergence theorem,
167, 180–182, 237, 240, 251–260, 272, 105–106, 126, 129, 178, 230
309–311, 434, 476, 486 Lebesgue integral, 103, 105, 181
Functional, 34, 41–45, 58, 62, 110, 153, 170, Lebesgue measurable function, 138–146
172, 179, 184, 185, 225, 233, 234, 239, Lebesgue measurable set, 129
251, 254, 257–258, 282, 288, 290–292, Lebesgue monotone convergence theorem,
359, 365, 370, 396, 425, 441, 449, 486, 103–106, 123, 125, 128, 129, 178
490, 494, 510, 547 Lebesgue points, 162–166
Lebesgue space, 175
G Legendre functional, 257
Gâteaux derivative, 226, 255 Legendre–Hadamard condition, 234–236
Gâteaux differentiability, 254–255, 311–317 Legendre transform, 257–259, 325, 500, 509
Gâteaux variation, 226–230, 313 Limit, 10, 12, 16, 38, 47, 64, 65, 70, 72, 86, 93,
Generalized method of lines, 307, 382–389, 163, 178, 189, 209, 226, 233, 235, 242,
392, 400, 404–411, 517–535 371, 372, 537
Ginzburg–Landau system, 374–377, 392–424 Limit point, 10, 12, 16, 93
Global existence, 374–377, 392 Linear function, 22, 23, 422
Index 559
Linear functional, 34, 41, 45, 58, 62, 110, 153, Nonnegative measurable functions, 103, 104,
170, 172, 179, 184, 185 109, 113, 125, 127, 141, 142
Linear mapping, 21–22 Norm, 11, 13, 30, 32, 43, 47, 52, 53, 56, 64,
Linear operator, 24, 25, 28, 30, 57–97, 185, 66, 69, 70, 72, 74, 80, 97, 175, 176, 180,
214, 255, 259, 263, 309, 312, 428 181, 183, 186, 187, 257, 273, 297, 307,
Locally convex space, 7, 170 308, 322, 345, 374, 384, 385, 469, 482
Local minimizer, 237, 299, 312 Normable space, 13
Local minimum, 226, 228–229, 234, 239, 247, Normed space, 11
292, 299, 300, 304–306, 339, 340 Null space, 21–22, 61, 293, 298
Lower semicontinuous, 151, 152, 163, Numerical, 321, 339–341, 353–362, 369–372,
251–260, 263, 267, 270, 280–282, 428, 382, 389–392, 400–411, 418, 419, 423,
432 424, 460–462, 466, 469–476, 484–491,
510–511, 529–530, 537, 546–552
M
O
Matrix version of generalized method of lines,
Open covering, 9
404–411
Open mapping theorem, 28–30
Maximal, 37, 42, 254
Open set, 4, 5, 9, 12, 44–46, 99, 100, 129, 133,
Maximize, 261, 352, 428, 431, 482, 546, 550,
137, 149–152, 155, 157, 159, 161, 163,
552
167, 175, 180, 195, 196, 198, 199, 205,
Maximum, 337
206, 228, 257, 271–273, 276
Measurability, 114–122
Operator, 23–24, 26, 28, 30, 57–97, 171, 180,
Measurable function, 99–101, 103–105, 109,
185, 199, 213, 218, 255, 259, 263, 283,
113, 125, 138–147, 257
307, 309, 312, 344, 365, 383, 385, 425,
Measurable set, 99, 102, 106, 107, 109,
426, 428, 429, 437
115–117, 121, 126, 129, 133–138, 141,
Operator topology, 57, 58
144, 166, 273
Optimal control, 377–381, 392, 456–459,
Measure, 99–166, 175, 242, 271, 322, 374,
537–552
477, 478, 485, 493
Optimal design, 456–459, 477–491, 537–552
Measure space, 101, 109, 122, 129 Optimality conditions, 300, 321, 476
Metric, 11–13, 53, 57, 282 Optimal shape, 486
Metric space, 11–19, 26, 52 Optimization, 261–264, 287–317, 349, 360,
Micro-magnetism, 493–516 452, 466, 477, 484, 486, 490, 491, 537,
Milman Pettis theorem, 55–56 546, 547
Minimizer, 41, 234, 263, 279, 344, 361, 364, Orthogonal complement, 32
370, 374, 435, 436, 516 Orthonormal basis, 36–40, 64
Minimum, 226, 228–229, 234, 239, 247, 267, Orthonormal set, 36–39
292, 299, 300, 304–306, 332, 339, 340, Outer measure, 114–122, 129–134
445, 537
Multi-well problem, 425–446 P
Phase transition, 425, 426, 437–446, 450–451
N Plate model, 321, 343–362, 456–459
Natural boundary conditions, 248–249 Positive functional, 153, 272, 445
Natural frequency, 550 Positive operator, 66–73, 79
Navier–Stokes system, 517, 519–530 Positive set, 106–108
Necessary conditions, 228–229, 239, 299, Projection, 49, 66, 82, 87, 94
304–306, 312, 340, 363
Neighborhood, 5–9, 22, 23, 46, 48, 51, 53–56, Q
150, 170, 252, 253, 255, 258, 263, 271, Quantum mechanics, 465–476
275, 278, 279, 309, 312, 313, 316, 334,
336, 338, 497, 504 R
Nested sequence, 13 Radon–Nikodym theorem, 109–114, 166
Newton’s method, 400, 402–405, 407, 411, Range, 21–22, 76, 96, 101, 161, 298
415–424 Real analysis, 3
560 Index
Real set, 225 309, 311, 322, 345, 365, 383, 384, 427,
Reflexive spaces, 48 428, 431, 494, 537
Relative compactness, 14, 18, 19, 191–195, Spectral theorem, 77–86, 94–97
220 Spectrum, 73–76
Resolvent, 73 Square root of a positive operator, 66–73
Riesz Strong operator topology, 57
lemma, 34, 89 Sufficient conditions, 234, 341, 343, 362, 365,
representation theorem, 110, 153–162, 394, 423
179–180 Symmetric operator, 92
S T
Scalar field, 3 Topological dual space, 42–43, 427
Schrödinger equation, 465, 466, 476 Topological space, 4, 5, 10, 99–101, 151
Self-adjoint operator, 67, 69, 71, 72, 76–86, Topological vector space, 3–40, 46, 168
91–97 Topology, 4, 6, 12, 13, 41–58, 167–169, 252,
Semi-linear case in micro-magnetism, 502–510 253, 257, 275, 282
Separable spaces, 52 Totally bounded set, 15, 17, 18, 194, 195
Sequentially compact, 15–18 Trace theorem, 213–216, 322, 344, 375, 379,
Set, 4–10, 12, 13, 15–19, 22, 25, 36–39, 447, 465, 477, 493
44–46, 48, 49, 51, 52, 64, 80, 99–100,
102, 106–109, 115–117, 120, 121, U
129, 133–138, 141, 144, 145, 149–155, Uniform convergence, 19, 180, 240
157–161, 163, 166–168, 170–172, 175, Uniformly convex space, 55–56
180, 182, 186, 187, 191–199, 205, 206, Unique, 3, 32–34, 89, 101, 109, 113, 153, 179,
208, 213, 217, 220, 225, 228, 248, 184, 257, 267, 307, 309, 310, 312, 355,
251–260, 267, 271–273, 276, 291, 302, 384, 409, 411, 493
307, 321, 322, 331, 337, 343, 345, 363, Upper bound, 37
375, 377, 383, 394, 403, 407, 426, 433, Upper semicontinuous, 79, 81, 82, 151, 152,
435, 437, 445, 447, 450, 452, 456, 465, 266, 267, 269, 431, 449, 459, 483
477, 484, 493, 517
Signed measures, 106–109 W
Simple function, 101–106, 127, 141, 142, 145, Weak convergence, 41, 381
272 Weakly closed, 48, 51, 52, 253
Sobolev imbedding theorem, 175, 186–213, Weakly compact, 51, 267–269
357, 363, 364, 375, 447, 465, 477, 493 Weak operator topology, 58
Sobolev space, 43, 175–220, 322, 494, 537 Weak-star topology, 48, 53
Space, 3–44, 46–53, 55–61, 63–64, 66, 73–75, Weak topology, 41–56, 252
77, 86, 88, 94, 99–101, 109, 110, 122, Weierstrass–Erdmann conditions, 245–247
129, 149–153, 168, 170, 171, 175–220, Weierstrass necessary condition, 239–243
225, 226, 231, 233, 234, 252, 253, 255,
257, 261, 263, 264, 266, 269, 282, 283, Z
287, 288, 292–293, 297–300, 304, 307, Zorn’s lemma, 37, 42

Functional Analysis and Applied Optimization in Banach Spaces

Uploaded by

Copyright:

Available Formats

Functional Analysis and Applied Optimization in Banach Spaces

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Functional Analysis and Applied Optimization in Banach Spaces

Uploaded by

Copyright:

Available Formats

Fabio Botelho

With Contributions by Anderson Ferreira

ISBN 978-3-319-06073-6 ISBN 978-3-319-06074-3 (eBook)

Mathematics Subject Classification: 46N10, 46E15, 46N50, 49J40, 49K20

© Springer International Publishing Switzerland 2014

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Summary of Main Results

The main results of this work are summarized as follows.

Duality Applied to Elasticity

Duality Applied to a Plate Model

Chapter 13 develops dual variational formulations for the two-dimensional equa-

Duality Applied to Ginzburg–Landau-Type Equations

Duality Applied to Multi-well Variational Problems

Duality for a Model in Quantum Mechanics

In Chap. 19 we develop a duality principle and computation for a class of nonlin-

Duality Applied to the Optimal Design in Elasticity

Duality Applied to Micro-magnetism

The main focus of Chap. 21 is the development of dual variational formulations

Duality Applied to Fluid Mechanics

In Chap. 22 we develop approximate solutions for the incompressible Navier–

Duality Applied to the Optimal Control and Optimal Design

This monograph is based on my Ph.D. thesis at Virginia Tech and I am especially

USA, two wonderful institutions in the American continent, forever in my heart

Pelotas, Brazil Fabio Botelho

Part I Basic Functional Analysis

1 Topological Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 The Hahn–Banach Theorems and Weak Topologies . . . . . . . . . . . . . . . . 41

3 Topics on Linear Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.4 The Square Root of a Positive Operator . . . . . . . . . . . . . . . . . . . . . . . . 66

4 Basic Results on Measure and Integration . . . . . . . . . . . . . . . . . . . . . . . . 99

5 The Lebesgue Measure in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

6 Other Topics in Measure and Integration . . . . . . . . . . . . . . . . . . . . . . . . . 147

8 The Lebesgue and Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

Part II Variational Convex Analysis

9 Basic Concepts on the Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . 225

10 Basic Concepts on Convex Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

11 Constrained Variational Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287

11.7 On the Banach Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 307

Part III Applications

12 Duality Applied to Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321

13 Duality Applied to a Plate Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343

14 About Ginzburg–Landau-Type Equations: The Simpler Real

15 The Full Complex Ginzburg–Landau System . . . . . . . . . . . . . . . . . . . . . . 373

16 More on Duality and Computation for the Ginzburg–Landau

16.3 On the Numerical Procedures for Ginzburg–Landau-Type

17 On Duality Principles for Scalar and Vectorial Multi-well

18 More on Duality Principles for Multi-well Problems . . . . . . . . . . . . . . . . 447

19 Duality and Computation for Quantum Mechanics Models . . . . . . . . . 465

20 Duality Applied to the Optimal Design in Elasticity . . . . . . . . . . . . . . . . 477

20.4 Another Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486

21 Duality Applied to Micro-Magnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493

22 The Generalized Method of Lines Applied to Fluid Mechanics . . . . . . 517

23 Duality Applied to the Optimal Control and Optimal Design

N The set of natural numbers

1.2 Vector Spaces

We denote by F a scalar field. In practice this is either R or C, the set of real or

F. Botelho, Functional Analysis and Applied Optimization in Banach Spaces: Applications 3

Definition 1.2.3 (Vector Subspace). Let U be a vector space. A set V ⊂ U is said

Definition 1.2.21 (Base). A collection σ ⊂ σ is said to be a base for σ if every