Infosys Science Foundation Series in Mathematical Sciences

Aref Jeribi

Problems in
Finite Element
Methods
Aubin Nitsche’s Duality Process, Nodal
Methods and Friedrichs Systems

@seismicisolation
@seismicisolation
Infosys Science Foundation Series

Infosys Science Foundation Series

in Mathematical Sciences

Editors-in-Chief
Gopal Prasad, University of Michigan, Ann Arbor, USA
Irene Fonseca, Carnegie Mellon University, Pittsburgh, PA, USA

Associate Editors
Chandrashekhar Khare, University of California, Los Angeles, USA
Mahan Mj, Tata Institute of Fundamental Research, Mumbai, India
Manindra Agrawal, Indian Institute of Technology Kanpur, Kanpur, India
Ritabrata Munshi, Tata Institute of Fundamental Research, Mumbai, India
S. R. S. Varadhan, New York University, New York, USA
Weinan E, Princeton University, Princeton, USA

@seismicisolation
@seismicisolation
The Infosys Science Foundation Series in Mathematical Sciences, a Scopus-
indexed book series, focuses on high-quality content in the domain of pure and
applied mathematics, biomathematics, financial mathematics, operations research
and theoretical computer science. With this series, Springer hopes to provide readers
with monographs, textbooks, edited volume, handbooks and professional books of
the highest academic quality on current topics in relevant disciplines. Literature in
this series will appeal to a wide audience of researchers, students, educators and
professionals across the world.

@seismicisolation
@seismicisolation
Aref Jeribi

Problems in Finite Element

Methods
Aubin Nitsche’s Duality Process, Nodal
Methods and Friedrichs Systems

@seismicisolation
@seismicisolation
Aref Jeribi
Department of Mathematics and Statistics,
College of Science
Imam Mohammad Ibn Saud Islamic
University (IMSIU)
Riyadh, Saudi Arabia

ISSN 2363-6149 ISSN 2363-6157 (electronic)

Infosys Science Foundation Series
ISSN 2364-4036 ISSN 2364-4044 (electronic)
Infosys Science Foundation Series in Mathematical Sciences
ISBN 978-981-97-5709-1 ISBN 978-981-97-5710-7 (eBook)
https://doi.org/10.1007/978-981-97-5710-7

Mathematics Subject Classification: 65L60, 65N30, 76M10, 78M10, 80M10, 82M10

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2024

This work is subject to copyright. All rights are solely and exclusively licensed 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.
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.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore

If disposing of this product, please recycle the paper.

@seismicisolation
@seismicisolation
To
my mother Sania,
my father Ali,
my wife Fadoua,
my children Adam and Rahma,
my brothers Sofien and Mohamed Amin,
my sister Elhem, and
all members of my extended family

@seismicisolation
@seismicisolation
Preface

Finite and boundary element methods are among the most widely used numerical
discretization methods for the approximate resolution of elliptical boundary prob-
lems. Finite element methods (FEM) are based on a variational formulation of the
partial differential equation (PDE) to be solved. The finite element method is a
numerical technique for solving problems that can be formulated as a functional
minimization, which are described by PDE. Partial differential equations appear in
many mathematical models of economic, biological, and physical phenomena, such
as elasticity, electromagnetism, fluid dynamics, quantum mechanics, model forma-
tion, or derivative evaluation. This book is devoted to the mathematical analysis of
the numerical solution of boundary problems arising in elasticity, acoustic, and elec-
tromagnetic scattering. Numerical simulation is of course the process that makes it
possible to calculate the solutions of these models on computer, and thus to simulate
the physical reality. This book is intended for applied mathematicians, engineers,
scientists, and graduate students wishing to learn and familiarize themselves with
finite element theory. This book includes a course and a series of exercises with very
detailed solutions that I taught for several years to licenses and masters in pure and
applied mathematics, and to engineering schools. The introduction to finite elements
is treated in two parts: finite elements in dimension 1 and finite elements in dimension
2. The finite element method is one of the tools of applied mathematics. In numerical
analysis, the finite element method is used to numerically solve partial differential
equations. When the space of finite elements is a subspace of the space of solutions,
the method is called conforme. In mathematics, it is a question of replacing a compli-
cated problem for which until now we do not know a solution, by a simpler problem
that we know how to solve. This method then appears as a particular Galerkin method.
Aubin Nitsche’s duality process has been considered in some cases conforming
and in others non-conforming. A description of the nodal method for squares or
rectangles and for triangles is given. Of course, in each part, an increase in the
error between exact solution and approximate solution is given. An approximation
of positive symmetric first-order systems in the Friedrichs sense by FEM is given.
Continuous and discontinuous approximation methods adapted to the structure of

vii
@seismicisolation
@seismicisolation
viii Preface

the transport equation, leading to linear systems of quasi-explicit resolution, and,

therefore commonly used in practice.
This book is inspired by a course taught in master’s in mathematics and mechanical
engineering at the University of Sfax (Tunisia) and at the University of Besançon
(France). It is inspired by many more complete works such as [1–8], as well as
various documents from university colleagues.
We would like to thank Hanen Ellouz for the diagrams that have been drawn in
this book.

Sfax, Tunisia Aref Jeribi

April 2024

References

1. M. BenHamadou, A. Jeribi, Analyse numérique matricielle—Méthodes et algorithmes, exer-

cices et problèmes corrigés, Collection références sciences (Ellipses, 2020)
2. P. Brézis, Analyse fonctionelle (Dunod, 1999)
3. P. G. Ciarlet, Introduction à l’analyse numérique matricielle et à l’optimisation (Masson, 1990)
4. A. Jeribi, Denseness, bases and frames in Banach spaces and applications (de Gruyter, Berlin,
2018)
5. S. Kallel, Analyse hilbertienne, Exercices corrigés (Centre de Publication Universitaire,
Collection M/Sciences fondamentales, 2001)
6. P. Lesaint, Sur la résolution des systèmes hyperboliques du premier ordre by des méthodes
d’éléments finis (Thèse, Pierre et Marie Curie, Paris, 1975)
7. P. Lesaint, Introduction aux éléments finis (Cours DEA, Besançon, 1994)
8. P.-A. Raviart, J.-M. Thomas, Introduction à l’analyse numérique des équations aux dériveées
partielles (Masson, 1983)

@seismicisolation
@seismicisolation
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Objective of This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Introduction of Some Elliptical Limit Problems . . . . . . . . . . . . . . . 4
1.4 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Nodal Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 Symmetric First-Order Differential Systems in the Sense
of Friedrichs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6.1 First-Order Systems, Symmetric, and Positive
in the Sense of Friedrichs . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6.2 Continuous Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6.3 Discontinuous Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6.4 Neutron Transport Equation . . . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Outline of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1 Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Bounded Linear Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.3 Convex Sets and Functions . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.4 Isometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5.5 Arithmetic-Geometric Mean Inequality . . . . . . . . . . . . . . 24
2.6 Elementary Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.1 Cauchy-Schwarz Inequality . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.2 Triangular Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

ix
@seismicisolation
@seismicisolation
x Contents

2.6.3 Parallelogram Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.6.4 Polarization Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6.5 Pythagorean Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6.6 Sympson’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7 Review of Lebesgue Integration Theory . . . . . . . . . . . . . . . . . . . . . 27
2.7.1 Lebesgue Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7.2 Space L2 (Ω, R) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7.3 Minkowski’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7.4 Holder’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7.5 Schwarz’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7.6 Support of a Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.8 Weak Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.9 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9.2 Orthogonality Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.10 Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.10.1 Orthonormal Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.10.2 Hilbert Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.11 Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.11.1 Positive Definite Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.11.2 Orthogonal Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.11.3 Monotonic Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.12 Projection on a Closed Convex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.12.1 Projection Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.12.2 A Simple Consequence of the Projection Theorem . . . . . 44
2.13 Contraction Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.14 The Dual of a Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.14.1 Riesz Fréchet’s Representation Theorem . . . . . . . . . . . . . 55
2.14.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.15 Stampacchia Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.15.1 Continuity and Coercivity . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.15.2 Statement of the Stampacchia Theorem . . . . . . . . . . . . . . 60
2.16 Equations and Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.16.1 Solution of Equations Related to Minimization . . . . . . . . 66
2.16.2 General Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.16.3 Quadratic Minimization Problem . . . . . . . . . . . . . . . . . . . . 67
2.16.4 The Variational Equation, Solvability . . . . . . . . . . . . . . . . 68
2.16.5 The Variational Inequality, Solvability . . . . . . . . . . . . . . . 69
2.17 Lax-Milgram Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.17.1 Lax-Milgram Theorem in Hilbert Space . . . . . . . . . . . . . . 73
2.17.2 Lax-Milgram Theorem in Banach Space . . . . . . . . . . . . . 78
2.17.3 Generalization of the Lax-Milgram Theorem . . . . . . . . . 79

@seismicisolation
@seismicisolation
Contents xi

2.18 Reminders on Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

2.18.1 Space D (Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.18.2 Space D , (Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.18.3 Derivation Within the Meaning of Distributions . . . . . . . 83
2.18.4 Regular Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3 Variational Formulation of Boundary Problems . . . . . . . . . . . . . . . . . . 91
3.1 Sobolev Space of Order 1, H 1 (Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.1.2 Separable Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2 Sobolev Space, H01 (Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.2.2 Poincaré Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.2.3 Poincaré-Wirtinger Inequality . . . . . . . . . . . . . . . . . . . . . . 96
3.2.4 Some Properties of the Space H01 (Ω) . . . . . . . . . . . . . . . . 97
3.3 Trace Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3.1 In Dimension n = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3.2 In Dimension n ≥ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.3.3 Application of the Trace Theorem . . . . . . . . . . . . . . . . . . . 102
3.3.4 Green’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.4 Sobolev Spaces of Order m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.4.1 Sobolev Spaces of Order m, H m (Ω) m ≥ 1 . . . . . . . . . . . 104
3.4.2 Separable Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.4.3 Trace Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.4.4 The Laplacian Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.4.5 The Canonical Injection . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.4.6 Sobolev Spaces of Order m, W m,p (Ω) 1 ≤ p ≤ ∞
and m ∈ N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.5 Dirichlet Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.5.1 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.5.2 Equivalence Between Variational Formulation
and Strong Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.5.3 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.5.4 Dirichlet’s Principle for the Laplacian Operator . . . . . . . 122
3.5.5 First Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.5.6 Second Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.6 Neumann Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.6.1 Homogeneous Neumann Boundary Condition . . . . . . . . . 130
3.6.2 Mixed Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
3.6.3 Neumann Boundary Condition . . . . . . . . . . . . . . . . . . . . . . 135
3.6.4 First Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.6.5 Second Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
3.6.6 Third Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

@seismicisolation
@seismicisolation
xii Contents

3.7 Transmission Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

3.8 General Second-Order Elliptic Problems . . . . . . . . . . . . . . . . . . . . . 155
3.8.1 Divergence Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
3.8.2 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
3.8.3 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . 157
3.9 Weak Solution of Elliptic of the Partial Differential
Equations (PDEs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
3.9.1 Dirichlet Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.9.2 Neumann Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.9.3 Robin Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.10 A New Kind of the Fourier Condition . . . . . . . . . . . . . . . . . . . . . . . 164
3.10.1 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
3.10.2 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . 166
3.11 The Convection-Diffusion Problem . . . . . . . . . . . . . . . . . . . . . . . . . 168
3.11.1 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
3.11.2 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . 169
3.12 A Slight Variant of the Plate Problem with Homogeneous
Dirichlet Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3.12.1 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
3.12.2 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . 172
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4 Introduction to Finite Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.1 Finite Element in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.1.1 Lagrange Finite Element . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.1.2 Regular Triangulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.1.3 P-Interpolated Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
4.1.4 Element of Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
4.1.5 Change of Variable Results . . . . . . . . . . . . . . . . . . . . . . . . . 180
4.1.6 Polyhedral Open Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.1.7 Construction of the Approximate Space Vh . . . . . . . . . . . 182
4.1.8 Error Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
4.2 Galerkin Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
4.2.1 Description of the Galerkin Method . . . . . . . . . . . . . . . . . 186
4.2.2 Error Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.3 Examples in Dimension 1 and Introduction to Finite
Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.3.1 Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4.3.2 Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4.3.3 Example 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
4.3.4 Example 4 (General Case) . . . . . . . . . . . . . . . . . . . . . . . . . 194
4.4 Estimates for the Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
4.4.1 We Interpolate by a Polynomial of Degree 1 . . . . . . . . . . 200
4.4.2 We Interpolate by a Polynomial of Degree 2 . . . . . . . . . . 208
4.4.3 We Interpolate by a Polynomial of Degree k . . . . . . . . . . 210

@seismicisolation
@seismicisolation
Contents xiii

4.5 Embedded Beam Bending Problem . . . . . . . . . . . . . . . . . . . . . . . . . 211

4.5.1 Position of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
4.5.2 Existence and Uniqueness of Solution . . . . . . . . . . . . . . . 212
4.5.3 Problem Approached . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.5.4 Error Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
4.5.5 Aubin-Nitsche’s Duality Process . . . . . . . . . . . . . . . . . . . . 219
4.5.6 Inverse Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
4.6 Introduction to Finite Elements in Dimension 2
for Polygonal Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
4.6.1 Triangulation of the Domain . . . . . . . . . . . . . . . . . . . . . . . . 224
4.6.2 Values at the Top . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
4.6.3 Barycentric Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
4.6.4 Approximation of the Solution . . . . . . . . . . . . . . . . . . . . . . 227
4.6.5 Triangulation Numbering . . . . . . . . . . . . . . . . . . . . . . . . . . 232
4.6.6 Examples of Poor Quality Triangles . . . . . . . . . . . . . . . . . 234
4.6.7 Conditions of Not Having Bad Quality Triangles . . . . . . 234
4.6.8 Transformation Between Any Triangle
and the Reference Triangle . . . . . . . . . . . . . . . . . . . . . . . . . 234
4.6.9 Estimates for the Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
4.6.10 Go Up in Degrees P2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
4.7 Conforming Case: Aubin-Nitsche’s Duality Process . . . . . . . . . . . 248
4.8 Introduction to Finite Elements in Dimension 2
for Non-polygonal Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
4.8.1 Domain and Approximate Domain . . . . . . . . . . . . . . . . . . 251
4.8.2 Error Between Exact Solution and Approximate
Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
4.8.3 Curved Border Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
4.9 Rectangular Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
4.9.1 Transformation Between Quadrilateral
and the Reference Square . . . . . . . . . . . . . . . . . . . . . . . . . . 265
4.9.2 Q1 Polynomial for Rectangular Element . . . . . . . . . . . . . . 266
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
5 Non-conforming Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
5.1 Triangulation of the Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
5.1.1 Numbering of the Domain in the Middle
of the Sides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
5.1.2 Values in the Middle of the Sides . . . . . . . . . . . . . . . . . . . . 272
5.2 Approximation of the Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
5.2.1 Approximate Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
5.2.2 Approximate Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
5.2.3 Existence and Uniqueness of Approximate
Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
5.2.4 Estimates for the Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282

@seismicisolation
@seismicisolation
xiv Contents

5.3 Non-conforming Case: Aubin Nitsche’s Duality Process . . . . . . . 288

5.4 Wilson Brick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
5.4.1 Rectangular Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
5.4.2 Approximate Problem, Patch Test for Rectangular
Meshes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
5.4.3 General Estimates for the Errors . . . . . . . . . . . . . . . . . . . . 308
5.5 Finite Element of the Serendipity Family . . . . . . . . . . . . . . . . . . . . 322
Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
6 Nodal Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
6.1 Nodal Method for Squares or Rectangles . . . . . . . . . . . . . . . . . . . . 336
6.1.1 Position of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
6.1.2 Reference Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
6.1.3 Basis Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
6.1.4 Description of the Nodal Method . . . . . . . . . . . . . . . . . . . . 340
6.1.5 Relationship Between Nodal Method
and Non-conforming Method . . . . . . . . . . . . . . . . . . . . . . . 342
6.1.6 Existence and Uniqueness of Solution . . . . . . . . . . . . . . . 347
6.1.7 Error Estimates for the Non-conforming Method . . . . . . 348
6.2 Nodal Method on Triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
6.2.1 Basis Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
6.2.2 Description of the Nodal Method on Triangles . . . . . . . . 367
6.2.3 Barycentric Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
7 Positive Symmetric First-Order Systems Within the Meaning
of Friedrichs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
7.1 First Order of Friedrichs Symmetric Systems . . . . . . . . . . . . . . . . . 379
7.1.1 Strong Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
7.1.2 Green’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
7.1.3 Direct Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
7.2 Existence of a Weak Solution for the Problem
(7.1.2)–(7.1.3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
7.3 Strong Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
7.4 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
7.5 Examples: Choice of the Matrix M . . . . . . . . . . . . . . . . . . . . . . . . . 395
7.6 Some Examples in Physical Problems . . . . . . . . . . . . . . . . . . . . . . . 410
7.6.1 Wave Equation in Dimension 1 . . . . . . . . . . . . . . . . . . . . . 410
7.6.2 Laplacian Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
7.6.3 Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
7.6.4 Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
7.6.5 Tricomi Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
7.6.6 Boundary Layer Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 420

@seismicisolation
@seismicisolation
Contents xv

7.7 Approximation of Friedrichs Systems by Finite Element

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
7.7.1 Rayleigh Ritz Galerkin Method . . . . . . . . . . . . . . . . . . . . . 422
7.7.2 Particular Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
7.7.3 Approximate Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
7.7.4 Convergence of the Approximate Solution
Towards the Exact Solution . . . . . . . . . . . . . . . . . . . . . . . . 433
7.7.5 Application to Finite Elements . . . . . . . . . . . . . . . . . . . . . . 437
7.7.6 Error Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
7.8 Continuous Finite Element Methods . . . . . . . . . . . . . . . . . . . . . . . . 446
7.9 Discontinuous Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
7.9.1 Discontinuous Finite Element Method . . . . . . . . . . . . . . . 454
7.9.2 Error Estimates for Discontinuous Methods . . . . . . . . . . . 463
7.10 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
7.10.1 In Dimension 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
7.10.2 In Dimension 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
7.10.3 The Penalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
7.11 One-Dimensional Transport Equations in Plane Symmetry . . . . . 476
7.12 Change to First-Order System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
7.12.1 Example in Dimension 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
7.12.2 Examples in Dimension 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 507
7.13 Special Case of the Transport Equation . . . . . . . . . . . . . . . . . . . . . . 515
7.14 Petrov Galerkin-Type Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522
8 Approximation of the Transport Equation in Plane
Two-Dimensional Geometry by Continuous and Discontinuous
Finite Element Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
8.1 Transport Equation in Plane Geometry . . . . . . . . . . . . . . . . . . . . . . 525
8.1.1 Position of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
8.1.2 Spherical Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527
8.1.3 Discrete Ordinate Method . . . . . . . . . . . . . . . . . . . . . . . . . . 529
8.1.4 Existence and Uniqueness Result . . . . . . . . . . . . . . . . . . . 531
8.1.5 Approximate Solution Over the Entire Domain . . . . . . . . 532
8.1.6 Approximate Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535
8.2 Continuous Finite Element Methods . . . . . . . . . . . . . . . . . . . . . . . . 539
8.2.1 Approximate Way Using the Quadrature Formula . . . . . 539
8.2.2 Quadrilateral Finite Elements . . . . . . . . . . . . . . . . . . . . . . . 541
8.2.3 Approximate Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542
8.2.4 Construction of the Quadrature Formula . . . . . . . . . . . . . 543
8.2.5 Discrete Norms and Semi-norms . . . . . . . . . . . . . . . . . . . . 544
8.2.6 Interpolate Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544
8.2.7 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
8.2.8 Estimate Between the Function and Its Interpolate . . . . . 547

@seismicisolation
@seismicisolation
xvi Contents

8.3 Scheme of Continuous Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550

8.3.1 Scheme 1 (Generalized DSN Scheme) . . . . . . . . . . . . . . . 550
8.3.2 Scheme 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555
8.3.3 Scheme 3 (Generalized S.N.G. Scheme) . . . . . . . . . . . . . 559
8.4 Discontinuous Finite Element Methods . . . . . . . . . . . . . . . . . . . . . . 563
8.4.1 Existence and Uniqueness Solution . . . . . . . . . . . . . . . . . . 564
8.4.2 Neutron Conservation Property . . . . . . . . . . . . . . . . . . . . . 564
8.4.3 General Bound of the Error . . . . . . . . . . . . . . . . . . . . . . . . 565
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575
9 Exercises with Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
9.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
9.2 Corrected . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
10 Revision Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 709
10.1 Exercises and Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 709
10.2 Problems and Exams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 745

@seismicisolation
@seismicisolation
Symbols

R, C The fields of real and complex numbers,

respectively
N The set of natural numbers
n and p Two non-zero natural numbers
Rn R × · · · × Rn times
Ω Polygonal domain of Rn
mes(Ω) Measure of Ω
area (K) Area of K
P1 Space of polynomials of degree ≤ 1, with
coefficients in R
Pk Space of polynomials of degree ≤ k, with
coefficients in R e
Q1 Space of polynomials of the form i,j≤1 aij xi xj ,
with aij ∈ R e
Qk Space of polynomials of the form i,j≤k aij xi xj ,
with aij ∈ R
A = (aij ) aij denotes the element of A located on the i-th row
and j-th column of the matrix A ∈ Mn,p (R)). i is
the index of the row, and j is the index of the
column
|| · || Norm on X
u⊥ Orthogonal of a vector u
M⊥ Orthogonal of a subspace M
C 0 ([a, b], R) The space of continuous functions over [a, b] with
values in R
C n (Ω, R) The set of functions n times differentiable from Ω
into R whose n-th derivative is continuous
dist(x, y) Distance between x and y
X, Topological dual of X
supp(f ) f function support
PK f The orthogonal projection of f on K

xvii
@seismicisolation
@seismicisolation
xviii Symbols

δnm The Kronecker symbol being worth 1 if n = m, 0

otherwise
Ker(ϕ) The kernel or the null space of the linear map ϕ
Im(ϕ) The image or the range of the linear map ϕ
t
C Transpose of the matrix C
D(Ω) The vector space of indefinitely differentiable
functions over Ω with compact support included in
Ω
D , (Ω) The distribution space on Ω (Schwarz distribution
space)
B(a, r) The open ball with center a and radius r
Tf The regular distribution associated with f
H (x) The Heaviside function at the point x
det(A) Determinant of the square matrix A
<·, ·> The canonical inner product of Rn
dim(E) The dimension of the vector space E
(u1 , u2 , . . . , un ) A family of Rn
vect (or vect(u1 , u2 , . . . , un )) Designates the sub vector space of X n generated by
the vectors u1 , u2 , . . . , un
L2 (Ω, R) The space of functions with square summable over
Ω
<·, ·>0,Ω The inner product of L2 (Ω, R)
|| · ||0,Ω The norm of L2 (Ω, R)
H 1 (Ω) The Sobolev space of order 1 over Ω
<·, ·>1,Ω The inner product of H 1 (Ω)
|| · ||1,Ω The norm of H 1 (Ω)
H01 (Ω) The closure of D(Ω) in H 1 (Ω)
| · |1,Ω The semi-norm of H 1 (Ω) or the norm of H01 (Ω)
H m (Ω) The Sobolev space of order m over Ω
<·, ·>m,Ω The inner product of H m (Ω)
|| · ||m,Ω The norm of H m (Ω)
rank(A) The rank of the matrix A
γ0 The trace mapping
∇u The gradient of u
/\u The Laplacian of u
hK Element diameters K
ρK Sup{diameters of spheres contained in K}
ZK The distances between the midpoints of the
diagonals of K
(τh )h A family of triangulations
|| · ||∞ The infinit norm
u Exact solution of the strong problem
uh Approximate solution
||u − uh || The error between exact solution and approximate
solution

@seismicisolation
@seismicisolation
Symbols xix

ϕj Hats function equal to 1 in xj and zero in all xi i /= j

rh v Vh -interpolated function of v
λi (x, y) Barycentric coordinates of a point M (x, y) with
respect to the points Ai (xi , yi ) 1 ≤ i ≤ 3, not
aligned
T Triangle
h(T ) The larger side of T
ρ(T )
/\
The diameter of the circle inscribed in T
T
/\
Reference triangle
R Reference rectangle
PDEs Partial differential equations
FEM Finite element methods

@seismicisolation
@seismicisolation
Chapter 1
Introduction

The finite element method was born in the aeronautical industry, in the 1950s, more
precisely around the research and development teams of Boeing. Only a few large
companies in the sector have been able to acquire the digital means essential to the
application of the method. The term “finite elements” is then widely developed. It has
grown widely since the 1950s to pervade most areas of applied science, engineering,
and financial markets. Nowadays, the finite element method is implemented in many
academic or commercial software. It remains until now a blind simulation tool, there-
fore an object of study, and still knows many more specific developments (mobile or
three-dimensional problems, applications on machines or massively parallel graphics
cards, deformable meshes, etc.).

1.1 Objective of This Book

The objective of this book is to introduce the reader to numerical simulation and
the world of mathematical modeling, which, in recent decades, have taken on con-
siderable importance in all fields of science and industrial applications (or financial
markets and engineering). Partial differential equations (PDE) appear in many math-
ematical models of economic, biological, and physical phenomena, such as elasticity,
electromagnetism, fluid dynamics, quantum mechanics, model formation, or deriva-
tive evaluation. Mathematical modeling is the art (or science, depending on the point
of view) of representing (or transforming) a physical reality into abstract models
accessible for analysis and calculation. Numerical simulation is of course the pro-
cess that makes it possible to calculate the solutions of these models on computer and
thus to simulate the physical reality. However, closed-form or analytic solutions of
these equations are only available in very specific cases (e.g., for simple geometries
or constant coefficients), and, therefore one must resort to numerical approximations
of these solutions.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 1
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_1
@seismicisolation
@seismicisolation
2 1 Introduction

The finite element method is the basis introduced in this book. Although intro-
duced in dimension .1, we will resume the basic concepts and largely extend the
fields of application in higher dimension. We will see, in particular, how to propose
numerical methods allowing to solve limit problems in higher dimension. Finally,
we will make the link with the notion of variational formulation, which also makes
it possible to better understand the underlying fundamental mathematical principles.
The objectives of this book are multiple. Indeed, it is first of all a question of
understanding how the point of view of the variational formulation makes it possi-
ble, under an unusual approach, to approach problems of partial differential equa-
tions. This point of view turns out to be rich and powerful. It allows in particular to
introduce the theoretical elements, which will lead to the resolution of the problem
(to demonstrate the existence and the unicity of the solution by using the theorem
of Lax-Milgram in an adequate framework), then to build the method of the finite
elements in consider a finite-dimensional subspace, which relies on theoretical con-
siderations to naturally provide a means of approximating the solution (which, in
general, is not explicitly computable).

1.2 Finite Element Method

The finite element method has enjoyed great success in many fields of science and
technology since it was first suggested in elasticity in the fifth decade of the twen-
tieth century. Today, it has become a powerful tool for solving partial differential
equations [1–3].
The key problem of the finite element method is to use a discrete solution on the
finite element space, usually consisting of piecewise polynomials, to approximate
the exact solution on the given space according to some kind of variational principle.
The finite element method (FEM) is a numerical technique for solving problems
that can be formulated as a functional minimization, which are described by partial
differential equations (PDE). A domain of interest is represented as an assembly of
finite elements. Finite element approximation functions are determined in terms of
nodal values of a searched physical field. A continuous physical problem is trans-
formed into a discretized finite element problem with unknown nodal values. For a
linear problem, a system of linear algebraic equations must be solved. Values inside
finite elements can be retrieved using nodal values.
Other steps will be necessary for the formalization of the finite element method
(FEM): measure theory is fundamental to formalize the so-called Sobolev spaces
such as . L 2 (Ω, R) , . H 1 (Ω), and . H01 (Ω) on a fairly regular domain .Ω and estab-
lish that they are Hilbert spaces on which the theorem applies of Lax-Milgram as
well as the concept of distribution to set up the good framework to treat the weak
formulations. Finally, chapters on approximation and interpolation theories define
discrete finite element approximation spaces. The . P1 finite element method is based
on the discrete space of globally continuous and affine functions on each cell.

@seismicisolation
@seismicisolation
1.2 Finite Element Method 3

When the space of finite elements is a subspace of the space of solutions, the
method is called conforme. It is known that, in this case, the finite element solution
converges to the true solution provided that the finite element space approaches the
given space in some sense [1].
In general, for an elliptical boundary value problem of order .2m, the space of
conforming finite elements is a subspace .C m−1 . This means that the shape function
in this conformal finite element space is continuous with its derivatives of order
.m − 1. That is, for a second-order problem, the shape function is continuous and for
a fourth-order problem, the shape function and its first derivatives are continuous.
This is a fairly strong restriction imposed on shape functions in the latter case.
It is proven that to construct a finite element space conforming with .C 1 continuity
for a two-dimensional problem of the fourth order, like the problem of bending of the
plates in elasticity, one needs at least an .18-parameter quintic polynomial for a trian-
gular element [1], and a .16-parameter bicubic polynomial for a rectangular element.
This leads to computational difficulties because the dimension of the associated finite
element space is quite large and its structure is rather complicated.
Therefore, it is desirable to use lower order polynomials with few parameters
while keeping the required continuity property. A genius approach is to subdivide a
given element into many small sub-elements and then use lower order polynomials
on each sub-element to achieve .C 1 continuity across the whole element. It is called
the macro-elements method [1]. An abundant literature has appeared in this direction.
However, due to the complexity of formulating the relevant finite element spaces,
this method does not seem so popular in finite element method calculations.
Another approach consists of directly relaxing the continuity .C m−1 of the finite
element space. This is the so-called non-conforming finite element method which
has had and still has a great impact on the development of finite element methods [1].
However, it has been observed for a short time that some non-conforming elements
converge and others do not. Convergence behavior sometimes depends on mesh
configuration.
Therefore, it is important to have some criterion to verify which non-conforming
element is convergent and which is not. B. M. Irons and A. Razzaque proposed the
Patch Test [4] based on some mechanical consideration and computational experi-
ences. The idea of the patch test is that each element should solve the problem accu-
rately for any constant strain field. The test is very simple and easy to implement in
engineering applications [4].
Wilson’s non-conforming finite element is commonly used to solve problems in
linear elasticity. The Patch Test is satisfied when this element is a rectangle or a
parallelogram and the error made on the constraints is of the order of .h, where .h is
the largest of the element diameters. The Patch Test is not satisfied when the element
is any quadrilateral.
The ambition of this book is to provide the basics that will allow future research
and development or design office engineers to create new numerical algorithms or
new models for more complicated problems not covered in this book. However, even
those who are not destined for such a career have an interest in fully understanding
the challenges of digital simulation. In fact, many political or industrial decisions

@seismicisolation
@seismicisolation
4 1 Introduction

are now taken on the basis of calculations or numerical simulations. It is, therefore,
important that decision-makers have the ability to judge the quality and reliability
of the calculations presented to them. This book will allow them to know the first
criteria that guarantee the relevance and validity of numerical simulations.
The introduction to finite elements is treated in two parts: finite elements in dimen-
sion 1 and finite elements in dimension 2. The finite element method is one of the tools
of applied mathematics. In numerical analysis, the finite element method is used to
solve numerically some partial differential equations. In mathematics, it is a question
of replacing a complicated problem for which until now we do not know a solution, by
a simpler problem that we know how to solve. The finite element method then appears
as a particular Galerkin method. The finite-dimensional (numerical) approximation
of such problems was first studied systematically by Boris Grigorievich Galerkin.
The Galerkin method, which he first published in 1915, is based on a sequence of
finite-dimensional subspaces.(Vn )n ⊂ X ,.Vn+1 ⊂ Vn , that fill the space. X in the limit.
In each finite-dimensional space .Vn , the problem is solved exactly. It can be shown
that under suitable assumptions the sequence of the approximate solutions .(u n )n ,
.u n ∈ Vn , converges to the exact solution .u of the problem. Aubin-Nitsche’s duality
process has been considered in some cases conforming and in others not conform-
ing. We also give a description of the nodal method for squares or rectangles and
for triangles. Of course, in each part, we give an increase of the error between exact
solution and approximate solution. We give an approximation of first-order systems
with positive symmetry in the sense of Friedrichs by finite element methods.
Unfortunately, it has been found in mathematics that patch testing is neither nec-
essary nor sufficient [2, 5–11]. A precise convergence condition, namely the gener-
alized patch test, was suggested by Stummel [12] from a rigorous mathematical point
of view. Many non-conforming finite elements in applications can be checked with
this test [13]. A simplified version of the generalized patch test has been proposed by
the author [14]. However, either the patch test or the generalized patch test is only an
analysis tool for an evaluation of the convergence of the non-conforming elements.
How to build a good element is another, if not more important, problem to solve a
real problem.

1.3 Introduction of Some Elliptical Limit Problems

We consider an elastic membrane .Ω in equilibrium in the plane .x, . y with boundary

∂Ω. We suppose that this membrane is attached along its border .∂Ω, and we pull
.

on this membrane with a force of density . f directed along the .z axis. We seek the
displacement .u(x, y) of the points of this membrane along the axis of .z.

@seismicisolation
@seismicisolation
1.3 Introduction of Some Elliptical Limit Problems 5

f
y

g
∂g
x

Let . E(v) be the total energy of the system for a state of displacement .v. The
displacement .u of the membrane under the action of the force . f is a function that
minimizes the energy . E(v), for all admissible displacements .v (i.e., zero on .∂Ω, and
such that the energy . E(v) makes sense). The internal energy (stress energy) of the
system is assumed to be proportional to the variation of the surface of the membrane
and is equal to ⎡( ⎤
{ ( )2 ( )2 ) 21
⎣ 1+ ∂v ∂v
. + − 1⎦ d xd y.
Ω ∂x ∂y

The potential energy corresponding to the displacement .v is given by

{
. − f v d xdy.
Ω

Except for an additive constant (characterizing the energy of the membrane at rest),
the energy . E(v) is, therefore, written
{ ( ) {
) )1
. E(v) = 1 + |grad v|2 2 − 1 d xd y − f v d xdy.
Ω Ω

If, we assume small displacements, the energy . E(v) is replaced by

{ {
1
. J (v) = |grad v|2 d xd y − f v d xdy.
2 Ω Ω

@seismicisolation
@seismicisolation
6 1 Introduction

We are, therefore, looking for a displacement .u such that

. J (u) = inf{J (v), v ∈ V = allowable displacement space}.

The space .V of admissible displacements can be characterized by

{ }
∂v ∂v
. V = v, v = 0 on ∂Ω, v, , ∈ L (Ω, R) .
2
∂x ∂y

1.4 Existence and Uniqueness

The Lax-Milgram theorem is one of the key ingredients used to build the finite ele-
ment method (FEM). It is a way to establish the existence and uniqueness of the
weak solution of the variational formulation and its discrete approximation. It is
valid for coercive linear operators defined on Hilbert spaces. A corollary known as
Cea’s lemma provides a quantification of the error between the computed approxi-
mation and the unknown exact solution. In particular, the Lax-Milgram theorem is
sufficient to prove the existence and uniqueness of the solution to the (variational
formulation) of the standard Poisson problem defined as follows. We now begin the
theoretical study of elliptic partial differential or derivative equations and boundary
value problems. We hinge on an approach, called the weak approach. There are other
ways to solve or to find the solution of elliptical problems. The weak approach is
quite simple and well-suited to a whole class of approximation methods, as we will
see later. Let .Ω be an open bounded domain of .Rn with piecewise smooth boundary
.Γ = ∂Ω. Given a function . f ∈ L (Ω, R), we search a function .u defined on .Ω
2

checking

. − Δu = f in Ω (1.4.1)
u = 0 on Γ = ∂Ω,
. (1.4.2)
en ∂ 2
where .Δ = i=1 ∂ xi2
is the Laplacian operator. Equations (1.4.1)–(1.4.2) are the
strong formulation of Laplacian’s problem. We replace the problem (1.4.1) and
(1.4.2) by the following problem:
Find .u ∈ H01 (Ω) such that for all .v ∈ H01 (Ω), we have

n {
E {
∂u ∂v
. dx = f v d x. (1.4.3)
i=1 Ω ∂ xi ∂ xi Ω

Its weak formulation (1.4.3) is the connection with the Lax-Milgram theorem given.
We do not intend to limit ourselves to this particular problem, but we emphasize that
our work covers this standard problem which is the basis for the study of many other
physical and mechanical problems.

@seismicisolation
@seismicisolation
1.5 Nodal Method 7

Other mathematical tools can be used to show or prove the existence and unique-
ness of the weak solution to variational problems. For example, one cites the Banach-
Necas-Babuska theorem for Banach spaces, from which one can deduce the Lax-
Milgram theorem, e.g., see [15], or the theory of mixed and saddle point problems,
which is used for example for some fluid problems, e.g., see [16, 17]. However, our
choice is mainly guided by our limited numbers and by the intuitionist logic of the
interactive result demonstrator that we intend to use: we try to select an elemen-
tary and constructive way of proof. This recommends to work initially with Hilbert
spaces rather than with Banach spaces and to try to avoid the use of the Hahn-Banach
theorem whose proof is based on Zorn’s lemma.

1.5 Nodal Method

Let .Ω be a bounded open set of .R2 , with piecewise smooth boundary .∂Ω, either a
triangulation .τh of .Ω by subdividing .Ω into rectangles or triangles . R. In this book,
we define the Nodal method and we show that the Nodal method is identical to
the non-conforming method, in proving that the linear system associated with the
Nodal method is identical with the linear system associated with the non-conforming
method. The error between the exact solution and the approximate solution is given
(indicators and estimators of error, a priori, and a posteriori). In fact, in nuclear
physics, we consider the problem

. − div(D grad u) + σ u = f in Ω
D and σ = cte in relation to each piece
u = 0 on Γ = ∂Ω,

with . D : diffusion, .σ : effective, . f : source, and .u : neutron flux. The respected

balance sheet is
{
. (−DΔu + σ u − f ) d xd y = 0 for all R ⊂ Ω.
R

We define the approximation space

. Vh = {vh ∈ L 2 (Ω, R) such that vh |R is defined by the mean values at the interfaces
. on each rectangle R and mean on R}.
The equation bilan is written
{
. (−DΔu h + σ u h − f ) d xd y = 0 for all R ⊂ Ω.
R

@seismicisolation
@seismicisolation
8 1 Introduction

Either the non-conforming method: Find .u h ∈ Vh such that

E{ E{
. D∇u h · ∇vh + σ u h vh d xd y = f vh d xd y
R⊂Ω R R⊂Ω R

for all .vh ∈ Vh with . D and .σ are equal to constants with respect to each rectangle . R,
v is the average of .vh on . R, i.e.,
. h |R

{
1
v
. h |R = vh d xd y
mes(R) R

and . f |R is the average of . f on . R, i.e.,

{
1
. f |R = f d xdy.
mes(R) R

Then, the Nodal method is identical to the non-conforming method.

1.6 Symmetric First-Order Differential Systems

in the Sense of Friedrichs

Let .Ω be a bounded open set of .Rn , with piecewise smooth boundary .∂Ω and let
.ν = (ν1 , · · · , νn ) be the following directed unit vector the exterior normal to .∂Ω.

The formulation exposed in this book, introduced by Friedrichs [18] in 1958, allows
to take into account partial differential equations of different types
• elliptical:
.

. − Δu = f.

• parabolic:
.
∂u
. − Δu = f.
∂t
• hyperbolic:
.
∂ 2u
. − Δu = f.
∂t 2
or changing type:
• Tricomi’s equation:
.
∂ 2u ∂ 2u
. y − 2 = f
∂x 2 ∂y
U
with .Ω {y = 0} /= ∅, or even of different character depending on the variables:

@seismicisolation
@seismicisolation
1.6 Symmetric First-Order Differential Systems in the Sense of Friedrichs 9

• boundary layer problem:

.

∂q ∂q ∂q ∂ 2q
u
. +v +w − ν 2 = f,
∂x ∂y ∂z ∂z

∂u ∂v ∂w
where .u > 0, . + + = 0, and .ν = cte > 0.
∂x ∂y ∂z
This formalism also makes it possible to define approximation methods by finite
elements using discontinuous functions [19], which is often well suited to problems
with strong gradients.

1.6.1 First-Order Systems, Symmetric, and Positive

in the Sense of Friedrichs

The main object of this work is to approach the solution of the following first-
order systems, symmetric, and positive in the sense of Friedrichs: Find a function
.u : Ω −→ R such that
p

E
n
∂u
. Au(x) := Ai (x) (x) + A0 (x)u(x) = F(x) for x ∈ Ω (1.6.1)
i=1
∂ xi

(B − M)(x)u = 0 for x ∈ ∂Ω,

. (1.6.2)

where .x ∈ Rn , .u(x) ∈ R p , . Ai is a . p × p symmetric matrix .1 ≤ i ≤ n with

Lipchitizian coefficients, . A0 is a . p × p matrix with coefficients in . L ∞ , satisfying

E
n
∂ Ai
. A0 +t A0 + ≥ c0 I, c0 > 0,
i=1
∂ xi

i.e., . A is positive, . F ∈ (L p (Ω, R)) p , . M is a . p × p matrix and is defined for .x ∈

Γ := ∂Ω,
En
.B = νi Ai ,
i=1

with .νi .1 ≤ i ≤ n are the components of the normal vector outside the domain .∂Ω.
We denote by .t M the adjoint matrix of . M and, we assume that

M(x). is continuous in x for x ∈ ∂Ω

M(x) +t M(x) ≥ 0
K er (B − M)(x) + K er (B + M)(x) = R p , for all x ∈ ∂Ω,

@seismicisolation
@seismicisolation
10 1 Introduction

where . K er (B) is the null space of . B. The formal adjoint operator . A∗ of . A is given by

En
∂
. A∗ u(x) = − (Ai (x)u(x)) + A∗0 (x)u(x) for x ∈ Ω.
i=1
∂ x i

We say that .u satisfies the adjoint boundary conditions if

(B +t M)(x)u = 0 for x ∈ ∂Ω.

.

When certain regularity properties on the dependence on .x are satisfied, the problem
(1.6.1)–(1.6.2) admits a unique solution .u ∈ (H 1 (Ω)) p ([2, 4, 20, 21]). Let .Wh
(resp. .Vh ) be
U a finite dimensional space not necessarily included (resp. included)
in .(H 1 (Ω) C 0 (Ω, R)) p . In the first part, one defines and one studies various
methods of approximation of the system (1.6.1)–(1.6.2): continuous methods when
the approximate solution .u h is sought in the space .Vh and discontinuous methods
when the solution .u h is sought in .Wh . Then, we give several examples, where we
study continuous and discontinuous approximation methods, adapted to the structure
of the equations, leading to linear systems of quasi-explicit resolution, and, therefore
commonly used in practice.

1.6.2 Continuous Method

We consider the approximate problem (1.6.1)–(1.6.2) by finite element type methods.

To define the continuous method, we introduce the following formulation of the
problem (1.6.1)–(1.6.2) valid for any solution .u ∈ (H 1 (Ω)) p :
{ { { {
1 1 1
. Au · v d x + u · A∗ v d x + Mu · vds = F · v d x,
2 Ω 2 Ω 2 ∂Ω Ω

for all .v ∈ (H 1 (Ω)) p , where . A∗ is the formal adjoint operator of . A. Let .Vh be a finite
dimensional space included in .(H 1 (Ω)) p , we seek .u h ∈ Vh such that for all .vh ∈ Vh ,
we have
{ { { {
1 1 1
. Au h · vh d x + u h · A ∗ vh d x + Mu h · vh ds = F · vh d x.
2 Ω 2 Ω 2 ∂Ω Ω

We show in particular when the space .Vh is constructed from triangular or quadrilat-
eral finite elements . K , of diameters .≤ h, and if we use polynomials of degree .k in
each element . K , and if the solution .u is regular enough (belonging to .(H k+1 (Ω)) p ),
we have the error markup

||u − u h ||(L 2 (Ω,R)) p ≤ ch k ||u||(H k+1 (Ω)) p .

. (1.6.3)

@seismicisolation
@seismicisolation
1.6 Symmetric First-Order Differential Systems in the Sense of Friedrichs 11

1.6.3 Discontinuous Method

To define the discontinuous method, one introduces a subspace.Wh of finite dimension

of .(L 2 (Ω, R)) p :

. Wh = {vh ∈ (L 2 (Ω, R)) p such that vh |K ∈ (PK ) p , for all K },

where . PK designates the space of shape functions on the element . K . The functions
of .Wh are regular on each element . K , and in general discontinuous at the interfaces
of these elements. We, then, seek .u h ∈ Wh such that for all .vh ∈ Wh
E [{ { (
BT − M T
) ]
. (Au h − F) · vh d x − (u int
h − u ext
h ) · v int
h ds = 0,
K ⊂Ω K ∂K 2
(1.6.4)
with
E
n
. BT = νiK Ai ,
i=1

.ν T exterior normal vector,

U . MT matrix defined on.∂ T , and with the convention.vh = 0
ext

if . S ⊂ ∂Ω. On .∂ T ∂Ω, . MT = M. We show that for a suitable choice of the

transmission matrices . M K , and if . PK contains the space of polynomials of degree
.≤ k the error bound (1.6.3) is satisfied as soon as .u ∈ (H (Ω)) .
1 p

To calculate the integrals involved in the formulas (1.6.4), we use approximate

quadrature. We examine, using techniques developed by Ciarlet and Raviart [22] for
elliptical problems, the influence of these formulas on the precision of the methods.
It is shown that by making reasonable assumptions on the accuracy of the quadrature
formulas, the error introduced by the latter is of the same order as the error introduced
by the finite element method. By using techniques of Dupont [23], some results of
super convergence is given in certain particular cases.

1.6.4 Neutron Transport Equation

Consider the continuous problem of the neutron transport equation

∂u ∂u
. Au := μ +ν + σ u = f for (x, y) ∈ Ω ⊂ R2 ,
∂x ∂y

u(x, y) = 0 for (x, y) ∈ ∂− Ω,

.

where .μ and .ν are two scalars such that .μ2 + ν 2 ≤ 1, .σ > 0, .Ω is a bounded domain
of .R2 , with border .∂Ω,

@seismicisolation
@seismicisolation
12 1 Introduction

∂ Ω = {(x, y) ∈ ∂Ω such that μn x + νn y < 0},

. −

and, .n x and .n y are the components of the exterior normal to .∂Ω. The approxima-
tion of the two-dimensional transport equation in plane symmetry is given. We first
recall certain properties specific to the hyperbolic character of this equation in plane
symmetry. We consider more particularly the neutron transport equation:
(i) In two-dimensional plane geometry .(x, y)
.

∂ϕ ∂ϕ
μ
. +ν + σ ϕ = F in Ω ⊂ R2 ,
∂x ∂y

.ϕ(x, y) = 0 for (x, y) ∈ ∂− Ω,

(see, for instance, any of Refs. [24–57], and [58]).

(ii) In one-dimensional spherical geometry .(r, μ):
.

∂ 2 ∂
μ
. (r ϕ) + r ((1 − μ2 )ϕ) + σ r 2 ϕ = r 2 S on Ω =]0, R[×] − 1, 1[
∂r ∂μ

ϕ(R, μ) = 0, for μ < 0,

.

with .μ ∈ [−1, 1] angular variable, .r spatial variable .0 < r ≤ R.

For these two problems, we study continuous and discontinuous approximation
methods, adapted to the structure of the equations, leading to linear systems of
quasi-explicit resolution, and, therefore commonly used in practice. This then makes
it possible to define, with Reed et al. [59], numerical methods for which the res-
olution is carried out by following the characteristic direction .μ, .ν. We define the
continuous methods in the general case ([59–61]) and we, then, consider particularly
the case, where the finite elements are convex quadrilaterals. If the finite elements
are rectangles and if the space . PK of the shape functions is the space . P1 (resp. . Q 1 )
we find the scheme DSN [62] (resp. SNG [63]). We first introduce the continuous
methods: We seek .u h ∈ Vh such that for all . K
{
. (Au h − f )v d xdy = 0, for all v ∈ PK, ,
K

where the space . PK, of the test functions .v is to be chosen at best. This generalizes
the DSN and SNG ([62, 63]) schemes commonly used in practice. For a suitable
choice of . PK, , we show that the method is stable when the finite elements . K are
convex quadrilaterals having two sides illuminated by the characteristic direction
.(μ, ν). We, then, show error bounds in certain norms, discrete analogues of the norm
.|| · ||0,Ω . We, then, consider a discontinuous method: we seek .u h ∈ Wh such that for
any . K and any function .vh ∈ PK , we have

@seismicisolation
@seismicisolation
References 13
{ {
. (Au h − f )vh d xd y − (μn x + νn y )(u h − th )vh ds = 0,
K ∂− K

where { U
0 on ∂− K ∂− Ω U
t =
. h
external trace of u h on ∂− K \(∂− K ∂− Ω),

and
∂ K = {(x, y) ∈ ∂ K such that μn x + νn y < 0}.
. −

This method is a particular case of discontinuous methods. We show the unconditional

stability of this method, as well as results of error bounds.

1.7 Outline of Contents

Our book consists of 10 chapters.

In Chap. 2, we recall some definitions, notations, and basic information on pro-
jection operators. We also introduce the concept of an abstract result concerning
the Lax-Milgram theorem for Hilbert and Banach space. In Chap. 3, we develop the
notion of Sobolev space and establish that this space is a separable Hilbert space. The
obtained results are used to describe the existence and uniqueness of Dirichlet and
Neumann problems. Chapter 4 is devoted to study the finite element. We introduce
the Galerkin method and, we give an estimate for the errors between the exact solu-
tion and the approached solution. Chapter 5 deals with the non-conforming methods
for a triangulation of the domain by triangle or rectangle. Also, Wilson brick is given.
In Chap. 6, we focus on the study of Nodal method for squares or rectangles. The
next part contains some results related to the relationship between nodal method and
non-conforming method. Chapter 7 is devoted to the analysis of the positive sym-
metric first-order systems within the meaning of Friedrichs. Chapter 8 is devoted to
the approximation of the two-dimensional transport equation in plane symmetry. In
Chap. 9, some exercises with solutions are given. In Chap. 10, some revision issues
are given.

References

1. P.G. Ciarlet, The Finite Element Method for Elliptic Problems (North-Holland, Amsterdam,
NY, 1978)
2. A. Jeribi, Solving various elliptical problems using finite element methods. Lessons, exer-
cises, and corrected problems. (Résolution de divers problèmes elliptiques by des méthodes
d’éléments finis. Cours, exercices, et problèmes corrigés.) (French) Références Sciences. Paris:
Ellipses (2021), 456 p. ISBN 978-2-340-06117-0/pbk
3. O.C. Zienkiewicz, The Finite Element Method, 3rd edn. (Mcgraw-Hill, New York, 1977)

@seismicisolation
@seismicisolation
14 1 Introduction

4. B.M. Irons, A. Razzaque, Experience with the patch test for convergence of finite elements. The
Mathematical Foundations of the Finite Element Method with Applications to Partial Differen-
tial Equations (Proceedings Symposium, Univ. Maryland, Baltimore, Md., 1972) (Academic,
New York, 1972), pp. 557–587
5. A. Jeribi, Spectral Theory and Applications of Linear Operators and Block Operator Matrices
(Springer, New-York, 2015)
6. A. Jeribi, Théorie des points fixes pour la topologie faibleâŁ¯: exercices et problèmes corrigés,
édition Cépaduès (Toulouse, France, 2023)
7. Z.-C. Shi, An explicit analysis of Stummel’s patch test examples. Int. J. Numer. Methods Eng.
20, 1233–1246 (1984)
8. Z.-C. Shi, A convergence condition for the quadrilateral Wilson element. Numer. Math. 44,
349–361 (1984)
9. Z.-C. Shi, On the convergence properties of quadrilateral elements of Sander and Beckers.
Math. Comput. 42, 493–504 (1984)
10. Z.-C. Shi, On the nine degree quasi-conforming plate element. Math. Numer. Sinica 10, 100–
106 (1988)
11. F. Stummel, The limitation of the patch test. Int. J. Numer. Methods Eng. 15, 177–188 (1980)
12. F. Stummel, The generalized patch test SIAM. J. Numer. Anal. 16, 449–471 (1979)
13. Z.-C. Shi, The generalized patch test for Zienkiewicz triangles. J. Comput. Math. 2, 279–286
(1984)
14. Z.-C. Shi Convergence of the TRUNC plate element Comput. Methods Appl. Eng., 62 (1987),
pp. 71-88
15. A. Ern, J.-L. Guermond, Theory and Practice of Finite Elements, Applied Mathematical
Sciences, vol. 159. (Springer, New York, 2004)
16. F. Brezzi, On the existence, uniqueness and approximation of saddle-point problems arising
from Lagrangian multipliers. Rev. FranàSaise Automat. Informat. Recherche Opérationnelle
Sér. Rouge 8(R-2), 129–151 (1974)
17. V. Girault, P.-A. Raviart, Finite Element Methods for Navier-Stokes Equations. Springer Series
in Computational Mathematics, vol. 5 (Springer, Berlin, 1986). Theory and algorithms
18. K.O. Friedrichs, Symmetric positive linear differential equations. Commun. Pure Appl. Math.
11, 333–418 (1958)
19. P. Lesaint, Sur la résolution des systèmes hyperboliques du premier ordre by des méthodes
d’éléments finis (Thèse, Pierre et Marie Curie, Paris, 1975)
20. P.D. Lax, R.S. Phillips, Local boundary conditions for dissipative symmetric linear differential
operators. Commun. Pure Appl. Math. 13, 427–455 (1960)
21. R.S. Phillips, L. Sarason, Singular symmetric positive first order differential operators. J. Math.
Mech. 15, 235–271 (1966)
22. P. G. Ciarlet, P.-A. Raviart, The combined effect of curved boundaries and numerical integration
in isoparametric finite element methods. The Mathematical Foundations of the Finite Element
Method with Applications to Partial Differential Equations (Proc. Sympos., Univ. Maryland,
Baltimore, Md., 1972) (Academic, New York, 1972), pp. 409–474
23. T. Dupont, Galerkin methods for first order hyperbolics: an example. SIAM J. Numer. Anal.
10, 890–899 (1973)
24. B. Abdelmoumen, A. Dehici, A. Jeribi, M. Mnif, Some new properties in Fredholm theory,
Schechter essential spectrum, and application to transport theory. J. Inequal. Appl. Art. ID
852676, 14 pp (2008)
25. B. Abdelmoumen, O. Jedidi, A. Jeribi, Time asymptotic description of an abstract Cauchy
problem solution and application to transport equation. Appl. Math. 59(1), 53–67 (2014)
26. B. Abdelmoumen, A. Jeribi, M. Mnif, Time asymptotic description of the solution to an abstract
Cauchy problem and application to transport equation. Math. Z. 268(3–4), 837–869 (2011)
27. B. Abdelmoumen, A. Jeribi, M. Mnif, Measure of weak noncompactness, some new properties
in Fredholm theory, characterization of the Schechter essential spectrum and application to
transport operators. Ric. Mat. 61(2), 321–340 (2012)

@seismicisolation
@seismicisolation
References 15

28. B. Abdelmoumen, A. Jeribi, M. Mnif, On graph measures in Banach spaces and description
of essential spectra of multidimensional transport equation. Acta Math. Sci. Ser. B (Engl. Ed.)
32(5), 2050–2064 (2012)
29. F. Abdmouleh, A. Ammar, A. Jeribi, A characterization of the pseudo-Browder essential spectra
of linear operators and application to a transport equation. J. Comput. Theor. Transp. 44(3),
141–153 (2015)
30. F. Abdmouleh, A. Jeribi, Symmetric family of Fredholm operators of indices zero, stability of
essential spectra and application to transport operators. J. Math. Anal. Appl. 364(2), 414–423
(2010)
31. F. Abdmouleh, A. Jeribi, Gustafson, Weidman, Kato, Wolf, Schechter, Browder, Rakocevic
and Schmoeger essential spectra of the sum of two bounded operators and application to a
transport operator. Math. Nachr. 284(2–3), 166–176 (2011)
32. A. Ammar, A. Jeribi, A characterization of the essential pseudospectra and application to a
transport equation. Extracta Math. 28(1), 95–112 (2013)
33. A. Ammar, A. Jeribi, K. Mahfoudhi, Global bifurcation from the real leading eigenvalue of the
transport operator. J. Comput. Theor. Transp. 46(4), 229–241 (2017)
34. G.I. Bell, S. Glasstone, Nuclear Reactor Theory (van Nostrand Reinhold Company, 1970)
35. N. Ben Ali, A. Jeribi, N. Moalla, On some subsets of Schechter’s essential spectrum of a matrix
operator and application to transport operator. Math. Nachr. 283(9), 1245–1256 (2010)
36. A. Ben Amar, I. Feki, A. Jeribi, Leray-Schauder and Furi-Pera types fixed point theorems for
the sum of two weakly sequentially continuous mappings and application to transport equation.
Afr. Mat. 25(3), 707–722 (2014)
37. A. Ben Amar, A. Jeribi, B. Krichen, Essential spectra of a .3 × 3 operator ma
38. A. Ben Amar, A. Jeribi, M. Mnif, Some applications of the regularity and irreducibility on
transport theory. Acta Appl. Math. 110(1), 431–448 (2010)
39. S. Charfi, A. Elleuch, A. Jeribi, On the time asymptotic behavior of a transport operator with
bounce-back boundary condition. Applied Mathematics in Tunisia, Springer Proceedings in
Mathematics and Statistics, vol. 131 (Springer, Cham, 2015 21), pp. 329–358
40. S. Charfi, A. Intissar, A. Jeribi, Expansion of solution in terms of generalized eigenvectors for
a rectilinear transport equation. Transp. Theory Statist. Phys. 38(6), 330–345 (2009). 22
41. S. Charfi, A. Jeribi, On a characterization of the essential spectra of some matrix operators and
application to two-group transport operators. Math. Z. 262(4), 775–794 (2009)
42. S. Charfi, A. Jeribi, N. Moalla, Time asymptotic behavior of the solution of an abstract Cauchy
problem given by a one-velocity transport operator with Maxwell boundary condition. Collect.
Math. 64(1), 97–109 (2013)
43. A. Dehici, A. Jeribi, K. Latrach, Spectral analysis of a transport operator arising in growing
cell populations. Acta Appl. Math. 92(1), 37–62 (2006)
44. A. Jeribi, Quelques remarques sur les opérateurs de Fredholm et application à l’équation de
transport. (French) [Some remarks on Fredholm operators and application to the transport
equation] C. R. Acad. Sci. Paris Sér. I Math. 325(1), 43–48 (1997)
45. A. Jeribi, Some remarks on the Schechter essential spectrum and applications to transport
equations. J. Math. Anal. Appl. 275(1), 222–237 (2002)
46. A. Jeribi, K. Latrach, Quelques remarques sur le spectre essentiel et application à l’équation de
transport. (French) [Some remarks on the essential spectrum and application to the transport
equation] C. R. Acad. Sci. Paris Sér. I Math. 323(5), 469–474 (1996)
47. A. Jeribi, H. Megdiche, N. Moalla, On a transport operator arising in growing cell populations.
II. Cauchy problem. Math. Methods Appl. Sci. 28(2), 121–145 (2005)
48. A. Jeribi, M. Mnif, Fredholm operators, essential spectra and application to transport equations.
Acta Appl. Math. 89(1–3), 155-176 (2005) (2006)
49. A. Jeribi, N. Moalla, A characterization of some subsets of Schechter’s essential spectrum and
application to singular transport equation. J. Math. Anal. Appl. 358(2), 434–444 (2009)
50. A. Jeribi, N. Moalla, I. Walha, Spectra of some block operator matrices and application to
transport operators. J. Math. Anal. Appl. 351(1), 315–325 (2009)

@seismicisolation
@seismicisolation
16 1 Introduction

51. A. Jeribi, N. Moalla, S. Yengui, . S-essential spectra and application to an example of transport
operators. Math. Methods Appl. Sci. 37(16), 2341–2353 (2014)
52. A. Jeribi, N. Moalla, S. Yengui, Perturbation theory, . M-essential spectra of .2times2 operator
matrices and application to transport operators. J. Math. Appl. 44, 35–55 (2021)
53. A. Jeribi, S.A. Ould Ahmed Mahmoud, R. Sfaxi, Time asymptotic behaviour for a one-velocity
transport operator with Maxwell boundary condition. Acta Appl. Math. 98(3), 163–179 (2007)
54. A. Jeribi, I. Walha, Gustafson, Weidmann, Kato, Wolf, Schechter and Browder essential spectra
of some matrix operator and application to two-group transport equation. Math. Nachr. 284(1),
67–86 (2011)
55. K. Latrach, A. Jeribi, On the essential spectrum of transport operators on L1-spaces. J. Math.
Phys. 37(12), 6486–6494 (1996)
56. K. Latrach, A. Jeribi, Aref Sur une équation de transport intervenant en dynamique des popu-
lations. (French) [On a transport equation arising in growing cell populations] C. R. Acad. Sci.
Paris Sér. I Math. 325(10), 1087–1090 (1997)
57. K. Latrach, A. Jeribi, H. Megdiche, Time asymptotic behavior of the solution to a Cauchy
problem governed by a transport operator. J. Integral Equ. Appl. 17(2), 121–139 (2005)
58. N. Moalla, M. Damak, A. Jeribi, Essential spectra of some matrix operators and application to
two-group transport operators with general boundary conditions. J. Math. Anal. Appl. 323(2),
1071–1090 (2006)
59. W.H. Reed, T.R. Hill, F.W. Brinkley, K.D. Lathrop, A two dimensional multigroup triangular
mesh planar geometry explicit transport code, LA 5428 MS. Los Alamos Scientific Laboratory
of the University of California (1974)
60. P. Lesaint, Continuous and discontinuous finite element methods for solving the transport
equation. The Mathematics of Finite Elements and Applications, II (Proc. Second Brunel Univ.
Conf. Inst. Math. Appl., Uxbridge, 1975) (Academic, London, 1976), pp. 151–161
61. P. Lesaint, J. Gérin-Roze, Isoparametric finite element methods for the neutron transport
equation. Int. J. Num. Meth. Eng
62. K.D. Lathrop, B. Carlson, Numerical solution of the Boltzmann transport equation. J. Comput.
Phys. 1, 173–197 (1967)
63. K.D. Lathrop, Spacial differencing of the transport equation, Positivity vs. accuracy. J. Comput.
Phys. 4, 475–498 (1969)

@seismicisolation
@seismicisolation
Chapter 2
Fundamentals

We begin by introducing here the mathematical notations and the basic results that
will be used throughout this book.

2.1 Vector Space

A (real) vector space is a set . X , whose elements are called vectors, and in which two
operations, addition and scalar multiplication, are defined as follows:
(i) To every pair of vectors .x and . y corresponds a vector .x + y in such a way that
.

. x + y = y + x and x + (y + z) = (x + y) + z.

. X contains a unique vector .0 (the zero vector or origin of . X ) such that

. x +0= x

for every .x ∈ X , and to each .x ∈ X corresponds a unique vector .−x such that

. x + (−x) = 0.

(ii) To every pair .(α, x), with .α ∈ R and .x ∈ X , corresponds a vector .αx in such a
.
way that
.1x = x, α(βx) = (αβ)x

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 17
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_2
@seismicisolation
@seismicisolation
18 2 Fundamentals

and such that the two distributive laws

.α(x + y) = αx + αy, (α + β)x = αx + βx

hold.
A non-empty subset . M of a vector space . X is called a subspace of . X if .αx + βy ∈
M for all .x, . y ∈ M and all .α, .β ∈ R. The elements .x1 , . . . , xn are said to be linearly
independent if .α1 x1 + · · · + αn xn = 0 implies that .αi = 0 for each .i; otherwise,
they are said to be linearly dependent. An arbitrary collection of vectors is said to be
linearly independent if every finite subset of distinct elements is linearly independent.
The dimension of a vector space . X , denoted by .dim X , is either .0, a positive integer
or .∞. If . X = {0}, then .dim X = 0; if there exist linearly independent .{u 1 , . . . , u n }
such that each .x ∈ X has a (unique) representation of the form

. x = α1 u 1 + · · · + αn u n with αi ∈ R,

then .dim X = n and .{u 1 , . . . , u n } is a basis for . X ; in all other cases .dim X = ∞.

2.2 Norm

Definition 2.2.1 Let . X be a vector space on the real body .R. The mapping

|| · || : X −→ R+
.

is called a norm, if it checks the following properties:

.(i) .||x|| = 0 implies x = 0,
.(ii) .||αx|| = |α|||x||, for all α ∈ R, for all x ∈ X , and

.(iii) .||x + y|| ≤ ||x|| + ||y||, for all x, y ∈ X . ♦

A (real) vector space . X is said to be a normed space if, to every .x ∈ X , there is
associated a non-negative real number .||x||, satisfying Definition 2.2.1. A sequence
.(x n )n , in a normed space . X , is called a Cauchy sequence if, for each .ε > 0, there
exists an integer . N such that .||xm − xn || < ε for all .m, .n ≥ N . We say .xn → x in . X
if .limn→∞ ||xn − x|| = 0 and, in this case, .x is called the limit of .(xn )n . . X is called
complete if every Cauchy sequence in . X converges to a limit in . X .
Example 2.2.1 For . X = Rn and .x = (x1 , . . . , xn ) ∈ Rn , we define the norms

E
n
||x||1 =
. |xi |
i=1

( n ) 21
E
||x||2 =
. |xi |2
i=1

@seismicisolation
@seismicisolation
2.3 Bounded Linear Operators 19

and
. ||x||∞ = sup |xi |.
1≤i≤n

If . X is a vector space and .|| · || is a norm on . X , then the couple .(X, || · ||) is called
a normed vector space. In the following, we use the notation . X is a normed vector
space. If . X is a normed vector space and .|| · || is a norm on . X , then . X remains a
metric space, if we define a distance .dist(·, ·) by

dist(x, y) := ||x − y||,

.

for all .x, . y ∈ X .

Definition 2.2.2 A metric space is said to be separable if it contains a dense countable
part. ♦

2.3 Bounded Linear Operators

Let . X , .Y be real vector spaces. A map .T : X −→ Y is said to be a linear operator

from . X into .Y if
. T (αx + βy) = αT x + βT y

for all .x, . y ∈ D(T ) (domain of .T ) and all .α, .β ∈ R. Let . X , .Y be normed spaces.
A linear operator .T from . X into .Y is said to be bounded if there exists a constant
.m > 0 such that

.||T x|| ≤ m||x|| for all x ∈ X.

We define the operator norm .||T || of .T by

||T || = sup ||T x||

.
||x||=1

= sup ||T x||.

||x||≤1

For . X , .Y are two Banach spaces (i.e., complete normed vector space), let .L (X, Y )
be the Banach space of all bounded linear operators from . X into .Y . A bounded linear
operator . K from . X into .Y is said to be compact if it transforms every bounded set
of . X in a relatively compact set of .Y . In a similar way, . K is said to be compact if for
any bounded sequence .(xn )n of . X , the sequence .(K xn )n has a convergent sequence
in .Y .

Theorem 2.3.1 (Banach-Steinhaus) Let . X be a Banach space and .Y be a normed

space. If . A ⊂ L (X, Y ) such that . sup ||T x|| < ∞ for each fixed .x ∈ X , then
T ∈A

@seismicisolation
@seismicisolation
20 2 Fundamentals

. sup ||T || < ∞. ♦

T ∈A

We recall the following result, the proof of which can also be found in an article by
Brezzi [1].
Lemma 2.3.2 Let . X be a Banach space and, . X 1 and . X 2 be two normed spaces. We
consider two linear operators . Ai ∈ L (X, X i ) .1 ≤ i ≤ 2 such that
(i) .||A1 v|| X 1 + ||A2 v|| X 2 defines on . X a norm equivalent to the norm .||v|| X .
.

(ii) The operator . A1 is compact.

.

Then, the map .v −→ ||A2 v|| X 2 is a norm on the quotient space . X /P equivalent to the
quotient norm, where . P is the kernel of the operator . A2 . ♦
When . X is a (real) normed space, the Banach space .L (X, R) will be called the
(normed) dual or the topological dual space of. X and will be denoted by. X , . Elements
of . X , are called bounded linear functionals or continuous linear functionals on . X .
Frequently, we shall use the notation . f (x) to denote the value of . f ∈ X , at .x ∈ X .
Using this notation, we note that

.| f (x)| ≤ || f |||x||

for all . f ∈ X , , .x ∈ X .

Theorem 2.3.3 (Hahn-Banach) Let . X be a normed space and .Y be a subspace of

X . If . f ∈ Y , , then there exists a bounded linear functional . -
. f ∈ X , such that
. -
f (y) = f (y) for all y ∈ Y, || -
f || X , = || f ||Y , . ♦

Corollary 2.3.4 Let. X be a normed space and.x0 /= 0 in. X . Then, there exists. f ∈ X ,
such that
. || f || = 1, f (x0 ) = ||x0 ||. ♦

The dual space . X ,, of . X , is called the second dual space of . X and is again a Banach
space. Note that to each .x ∈ X , we can associate a unique element . F x ∈ X ,, defined
by . Fx ( f ) = f (x) for all . f ∈ X , . From Corollary 2.3.4, one can also show that
,,
.||Fx || = ||x||. Thus, the (canonical) mapping . J : X −→ X , given by . J x = Fx , is a
linear isometry of . X onto the subspace . J (X ) of . X ,, . Since . J is one-to-one, we can
identify . X with . J (X ). A Banach space . X is called reflexive if its canonical map . J
is onto . X ,, .

2.4 Inner Product

A bilinear form, .a(·, ·), on a linear space . X is a mapping .a : X × X −→ R such

that each of the maps .v −→ a(v, w) and .w −→ a(v, w) is a linear form on . X . It is

@seismicisolation
@seismicisolation
2.4 Inner Product 21

symmetric if .a(v, w) = a(w, v) for all .v, .w ∈ X . A (real) inner product, denoted by
(·, ·), is a symmetric bilinear form on a linear space . X that satisfies
.

(i) .(v, v) ≥ 0 for all .v ∈ X and,

.
(ii) .(v, v) = 0 if, and only if, .v = 0.
.

In fact,

Definition 2.4.1 Let . X be a real vector space. An inner product .(·, ·) on . X is an

mapping

(·, ·) : X × X −→ R
.

(u, v) −→ (u, v)

checking the following properties

.(i) symmetry .(u, v) = (v, u) for all .u, .v ∈ X ,
.(ii) linearity .(u 1 + u 2 , v) = (u 1 , v) + (u 2 , v) for all .u 1 , .u 2 , and .v ∈ X ,
.(λu, v) = λ(u, v) for all .u, .v ∈ X and .λ ∈ R,

.(iii) positivity .(u, u) ≥ 0 for all .u ∈ X ,

.(u, u) = 0 if, and only if, .u = 0. ♦

Definition 2.4.2 A real vector space . X provided with an inner product .(·, ·) named
prehilbertian space. If the space . X is of finite dimension, then . X is called euclidean
space. ♦

Example 2.4.1 .(i) .Rk (.k ∈ N∗ ) provided with inner product

E
k
(x, y) =
. xi yi (2.4.1)
i=1

is a prehilbertian space or Euclidean space.

(ii) Let .l 2 be the space defined by
.

{ ∞
}
E
.l = (xn )n real sequence such that
2
xi2 <∞ .
i=0

l 2 provided with inner product

.

∞
E
(x, y) =
. xi yi (2.4.2)
i=0

is a prehilbertian space.

@seismicisolation
@seismicisolation
22 2 Fundamentals

(iii) The space .C 0 ([a, b], R), of continuous functions on .[a, b] with values in .R,
.

provided with inner product

{ b
( f, g) =
. f (x)g(x) d x (2.4.3)
a

is a prehilbertian space.
(iv) Let . L 2 (]a, b[, R) be the space defined by
.

{ { b ]
. L 2 (]a, b[, R) = f :]a, b[−→ R measurable such that | f (x)|2 d x < ∞ .
a

. L 2 (]a, b[, R) provided with inner product

{ b
.( f, g)0,]a,b[ = f (x)g(x) d x (2.4.4)
a

is a prehilbertian space.

2.5 Hilbert Space

2.5.1 Definition

Definition 2.5.1 An Hilbert space . X is a vector space with an inner product .(·, ·)
and which is complete for the norm
√
. ||u|| = (u, u). ♦

2.5.2 Example

Example 2.5.1 .(i) .Rk (.k ∈ N∗ ) provided with the inner product (2.4.1) is a Hilbert
space.
(ii) .l 2 provided with the inner product (2.4.2) is a Hilbert space.
.

(iii) The space .C 0 ([a, b], R) provided with the inner product (2.4.3) is not a Hilbert
.
space (see Exercise 9.9).
(iv) . L 2 (]a, b[, R) provided with the inner product (2.4.4) is a Hilbert space.
.

@seismicisolation
@seismicisolation
2.5 Hilbert Space 23

2.5.3 Convex Sets and Functions

In what follows, we place ourselves in a Hilbert space . X provided with an inner

product .(·, ·). In what follows, it is assumed that . f : K ⊂ X −→ R is continuous,
. K denoting any part of . X .

Definition 2.5.2 .(i) We say that a set . K ⊂ X is convex if for all .(x1 , x2 ) ∈ K 2 and
.t ∈ [0, 1], .t x 1 + (1 − t)x 2 ∈ K .

.(ii) Let . K be a convex included in . X . The mapping . f : K −→ R is said to be convex

if for all .(x1 , x2 ) ∈ K 2 , .t ∈ [0, 1],

. f (t x1 + (1 − t)x2 ) ≤ t f (x1 ) + (1 − t) f (x2 ). (2.5.1)

(iii) We say that . f is strictly convex if the inequality (2.5.1) is strict for .x /= y,
.

t ∈ [0, 1].
. ♦

2.5.4 Isometry

Definition 2.5.3 Let . E and . F be metric spaces with metrics (e.g., distances)
dist E (·, ·) and .dist F (·, ·). A linear map . f : E −→ F is called an isometry or dis-
.
tance preserving map if for any .u, .v ∈ E one has
. dist E (u, v) = dist F ( f (u), f (v)). ♦

Lemma 2.5.4 Let . E be a Banach space, . F be a normed vector space and . A be an

isometry from . E into . F. Then, the set . A(E) is a closed subspace of the space . F. ♦

Proof Let .(vm )m be a sequence of . A(E) converging to .v in . F. Let .m ∈ N, then there

is .wm ∈ E such that
.vm = A(wm ).

Since .(vm )m is a convergent sequence, it is a Cauchy on . F. Thus, since . A is an

isometry, we have

. lim ||v p − vq || F = lim ||A(w p ) − A(wq )|| F

p, q→∞ p, q→∞

= lim ||w p − wq || E
p, q→∞

= 0.

@seismicisolation
@seismicisolation
24 2 Fundamentals

Hence, .(wm )m is Cauchy on . E. However, . E is complete, so .(wm )m is converging to .w

in . E. Since . A is an isometry, . A is continuous. Hence, .(A(wm ))m converges to . A(w),
which proves
.v = A(w)

and so .v ∈ A(E). Thus, . A(E) is closed. Q.E.D.

2.5.5 Arithmetic-Geometric Mean Inequality

We will make frequent use of the arithmetic-geometric mean inequality, which is

nothing more than the simple observation that, for any real numbers .a and .b,

1 2
ab ≤
. (a + b2 )
2

(just observe that .0 ≤ (a − b)2 = −2ab + a 2 + b2 ). A slightly more complicated

version of the inequality comes by writing
( ) ( ( )2 )
b 1 b
.ab = (εa) ≤ (εa)2 + .
ε 2 ε

Writing .δ in place of .ε2 , we find

δ 2 1
. ab ≤ a + b2 (2.5.2)
2 2δ

for any .δ > 0. Let .a ≥ 0, .b ≥ 0, and .1 < p, q < ∞ such that . 1p + 1

q
= 1. Then, for
all .ε > 0, we have
ap bq
.ab ≤ ε + ε1−q .
p q

2.6 Elementary Properties

2.6.1 Cauchy-Schwarz Inequality

Proposition 2.6.1 Let . X be a prehilbertian space. Then, for all .u, .v ∈ X , we have

.|(u, v)|2 ≤ (u, u)(v, v). (2.6.1)

@seismicisolation
@seismicisolation
2.6 Elementary Properties 25

Proof Let .u, .v ∈ X and .λ ∈ R. If .v = 0, then the result (2.6.1) is trivial. If .v /= 0,

consider the quadratic equation in .λ

||v||2 λ2 + 2(u, v)λ + ||u||2 = (u + λv, u + λv) ≥ 0.

. (2.6.2)

Since the equation of the second degree (2.6.2) has a constant sign, its discriminant
Δ, := ((u, v))2 − (u, u)(v, v) ≤ 0. This completes the proof.
. Q.E.D.

2.6.2 Triangular Inequality

Proposition 2.6.2 Let . X be a prehilbertian space. Then, for all .u, .v ∈ X , we have
1 1 1
. (u + v, u + v) 2 ≤ (u, u) 2 + (v, v) 2 . ♦

Proof Let .u, .v ∈ X . We have

(u + v, u + v) = (u, u) + 2(u, v) + (v, v).

. (2.6.3)

In view of Cauchy-Schwarz inequality (2.6.1), Eq. (2.6.3) gives

1 1
(u + v, u + v) ≤ (u, u) + 2(u, u) 2 (v, v) 2 + (v, v)
.
( 1
)
1 2
≤ (u, u) 2 + (v, v) 2 .

This completes the proof. Q.E.D.

Remark 2.6.3 As a consequence of the Proposition 2.6.2 is that the mapping

|| · || : X −→ R+
.
/
u −→ ||u|| = (u, u)

is a norm on . X . ♦

2.6.3 Parallelogram Identity

Proposition 2.6.4 Let . X be a prehilbertian space. Then, for all .u, .v ∈ X , we have
. ||u + v||2 + ||u − v||2 = 2||u||2 + 2||v||2 . ♦

Theorem 2.6.5 Every non-empty closed convex subset . S of a Hilbert space . X con-
tains a unique element of minimal norm. ♦

@seismicisolation
@seismicisolation
26 2 Fundamentals

Proof Choose .xn ∈ S so that .||xn || → d = inf{||x|| : x ∈ S}. Since . 21 (xn + xm ) ∈

S, we have .||xn + xm ||2 ≥ 4d 2 . Using the parallelogram identity, we see that

||xn − xm ||2 ≤ 2(||xn ||2 − d 2 ) + 2(||xm ||2 − d 2 )

. (2.6.4)

and, therefore .(xn )n is a Cauchy sequence in . X . Since . S is closed, .(xn )n converges

to some .x ∈ S and .||x|| = d. If . y ∈ S and .||y|| = d, then the parallelogram identity
implies, as in (2.6.4), that .x = y. Q.E.D.

2.6.4 Polarization Identity

Proposition 2.6.6 Let . X be a prehilbertian space. Then, for all .u, .v ∈ X , we have
1 1
. (u, v) = ||u + v||2 − ||u − v||2 . ♦
4 4

2.6.5 Pythagorean Relation

Proposition 2.6.7 Let . X be a prehilbertian space. Then, for all .u, .v ∈ X , such that
(u, v) = 0, we have
.

. ||u + v||2 = ||u||2 + ||v||2 . ♦

2.6.6 Sympson’s Formula

Proposition 2.6.8 .(i) If .g ∈ P1 , then

{ b
g(b) + g(a)
. g(x) d x = (b − a) ,
a 2

where . P1 denotes the space of polynomials of degree .≤ 1, with coefficients in .R.

(ii) If .g ∈ P2 , then
.

{ ( ( ) )
b
b−a b+a
. g(x) d x = g(b) + 4g + g(a) ,
a 6 2

where . P2 denotes the space of polynomials of degree .≤ 2, with coefficients in .R.

@seismicisolation
@seismicisolation
2.7 Review of Lebesgue Integration Theory 27

(iii) If two polynomials . p and .q of . P3 coincide in two points and in their derivatives,
.

then they are equal . p = q, where . P3 denotes the space of polynomials of degree .≤ 3,
with coefficients in .R. ♦

2.7 Review of Lebesgue Integration Theory

2.7.1 Lebesgue Spaces

By “domain” we mean a Lebesgue-measurable (usually either open or closed) subset

of .Rn with non-empty interior. We restrict our attention for simplicity to real-valued
functions, . f , on a given domain .Ω, that are Lebesgue measurable; by
{
. f (x) d x
Ω

we denote the Lebesgue integral of . f (.d x denotes Lebesgue measure). For .1 ≤ p <
∞, let
({ ) 1p
.|| f || L p (Ω,R) = | f (x)| d x ,
p
Ω

and for the case . p = ∞ set

|| f || L ∞ (Ω,R) := ess- sup{| f (x)| : x ∈ Ω}.

.

In either case, we define the Lebesgue spaces

. L p (Ω, R) := { f measurable such that || f || L p (Ω,R) < ∞}.

All . L p (Ω, R) spaces with .1 < p < ∞ are reflexive.

Theorem 2.7.1 .{ f function such that f ∈ C p ([a, b], R)} =

. . L p+1 (]a, b[, R),
where .a, .b ∈ R with .a < b. ♦

2.7.2 Space . L 2 (Ω, R)

Let .Ω be an open of .Rn and . L 2 (Ω, R) be the space of functions, defined on .Ω, with
values in .R, of square integrable relatively to the Lebesgue measure .d x1 . . . d xn on
.R . We provide . L (Ω, R) of the following inner product
n 2

{
( f, g)0,Ω =
. f (x)g(x) d x.
Ω

@seismicisolation
@seismicisolation
28 2 Fundamentals

We notice
1
|| f ||0,Ω := ( f, f )0,Ω
.
2

({ ) 21
:= f (x) d x
2
Ω

the corresponding norm. .(L 2 (Ω, R), || · ||0,Ω ) is a Hilbert space.

2.7.3 Minkowski’s Inequality

For .1 ≤ p ≤ ∞ and . f , .g ∈ L p (Ω, R), we have

|| f + g|| L p (Ω,R) ≤ || f || L p (Ω,R) + ||g|| L p (Ω,R) .

.

In view of Minkowski’s inequality and the definitions of .|| · || L p (Ω,R) , the space
L p (Ω, R) is closed under linear combinations, i.e., it is a linear (or vector) space.
.
Moreover, the functionals .|| · || L p (Ω,R) have properties that classify them as norms.

Theorem 2.7.2 For .1 ≤ p ≤ ∞, . L p (Ω, R) is a Banach space. ♦

2.7.4 Holder’s Inequality

For .1 ≤ p, q ≤ ∞ such that .1 = 1

p
+ q1 , if . f ∈ L p (Ω, R) and .g ∈ L q (Ω, R), then
. f g ∈ L (Ω, R) and
1

|| f g|| L 1 (Ω,R) ≤ || f || L p (Ω,R) ||g|| L q (Ω,R) .

.

2.7.5 Schwarz’s Inequality

This is simply Holder’s inequality in the special case. p = q = 2. If. f ,.g ∈ L 2 (Ω, R),
then . f g ∈ L 1 (Ω, R) and
{
. | f (x)g(x)| d x ≤ || f ||0,Ω ||g||0,Ω ,
Ω

1
where .|| f ||0,Ω = ( f, f )0,Ω
2
.

@seismicisolation
@seismicisolation
2.8 Weak Convergence 29

2.7.6 Support of a Function

Let .Ω be an open of .Rn and . f be a defined function on .Ω with real values. The
support of a function . f (x), .supp( f ), is the smallest closed set of values of .x ∈ Ω
outside of which the function . f (x) is identically zero. It is, therefore, the closure of
the set of points .x ∈ Ω such that . f (x) /= 0, i.e.,

supp( f ) := {x ∈ Ω such that f (x) /= 0}.

.

2.8 Weak Convergence

Let . X be a normed space. A sequence .(xn )n ∈ X is said to be weakly convergent to

an element .x ∈ X , written .xn - x, if . f (xn ) → f (x) for all . f ∈ X , .

Theorem 2.8.1 Let .(xn )n be a sequence in . X . Then,

.(i) Weak limits are unique.
.(ii) If .xn → x, then .xn - x.
.(iii) If . x n - x, then .(x n )n is bounded and .||x|| ≤ lim inf ||x n ||. ♦
Proof .(i) Suppose that .x and . y are both weak limits of the sequence .(xn )n and set
z = x − y. Then, . f (z) = 0 for every . f ∈ X , and by Corollary 2.3.4, .z = 0.
.

(ii) Let . f ∈ X , and note that .xn → x implies . f (xn ) → f (x) since . f is continuous.
.

.(iii) Assume .xn - x and consider the sequence .(J xn )n of elements of . X ,, , where
,, ,,
. J : X −→ X is the canonical injection from . X into . X . It is easy to see that . J is a
,
bounded operator. For each . f ∈ X ,

. sup |J xn ( f )| = sup | f (xn )| < ∞

(since . f (xn ) converges). By the Banach-Steinhaus theorem, there is a constant .c

such that .||xn || = ||J xn || ≤ c which implies .(xn )n is bounded. Finally, for . f ∈ X ,

| f (x)| = lim | f (xn )|

.

≤ lim inf || f ||||xn ||

= || f || lim inf ||xn ||

which implies the desired inequality since .||x|| = sup | f (x)|. Q.E.D.
|| f ||=1

We need a result of functional analysis.

Lemma 2.8.2 If .( f k )k is a bounded sequence in . L p (Ω, R), .1 < p < ∞, then there
exists a subsequence .( f km )m which weakly converges to some . f ∈ L p (Ω, R). ♦

@seismicisolation
@seismicisolation
30 2 Fundamentals

2.9 Orthogonality

2.9.1 Definitions

Definition 2.9.1 We say that two vectors .u and .v of . X are orthogonal if

(u, v) = 0
.

and, we write .u⊥v. ♦

Definition 2.9.2 Let . X be a real or complex linear space, .v1 , .v2 , . . . , vk be elements
of . X , and .a1 , .a2 , . . . , ak be real or complex coefficients. The element

E
k
v=
. ai vi
i=1

of . X is said to be a linear combination of the elements .v1 , .v2 , . . . , vk with the coef-
ficients .a1 , .a2 , . . . , ak . ♦

Definition 2.9.3 Let .V be a subset of a linear space . X (not necessarily a subspace of

X ). The linear span of.V , usually denoted by.span(V ), is defined to be the intersection
.
of all subspaces of the space . X that contain the set .V . ♦

U the sum . M + N
Definition 2.9.4 Let . M and . N be subspaces of a linear space . X . By
we mean the linear span of the union of . M and . N , i.e., .span(M o N ). We say that
. X is a direct sum of its subspaces . M, and . N (written as . X = M N ) if
.(i) . X =nM + N ,
.(ii) . M N = {0}.
o
If . X = Mo N , then . N is direct complement of . M and vice versa. We also say that
.X = M N is direct decomposition of . X into subspaces . M, . N . ♦

Definition 2.9.5 A family . F = {v1 , . . . , vn } of a vector space . X on the real body

R is said linearly independent, and its vectors are said to be linearly independent,
.

when for all .λ1 , . . . , λn ∈ R, such that .λ1 v1 + · · · + λn vn = 0, we have .λ1 = · · · =

λn = 0. ♦

Definition 2.9.6 A family .(xi )i∈I is called orthogonal if the vectors which compose
it are two by two orthogonal. ♦

An orthogonal family consisting of non-zero vectors is a linearly independent family.

If .(x1 , . . . , x p ) is an orthogonal family, then the Pythagorean relation is the following
|| p ||2
||E || E
p
|| ||
. || xk || = ||xk ||2 .
|| ||
k=1 k=1

@seismicisolation
@seismicisolation
2.9 Orthogonality 31

Definition 2.9.7 Let .u ∈ X . We call orthogonal of .u, the set

. u ⊥ = {v ∈ X such that (u, v) = 0} . ♦

The mapping

.(u, ·) : X −→ R
v −→ (u, v)

is a continuous linear form. Therefore, .u ⊥ is a closed subspace of . X .

Definition 2.9.8 Let . M be a vector subspace of . X . We call orthogonal of . M, the set

. M ⊥ = {v ∈ X such that (u, v) = 0 for all u ∈ M} . ♦

Remark 2.9.9 Let . M be a vector subspace of . X . Then,

n
.(i) . M
⊥
= u ⊥ is a closed subspace of . X .
n u∈M
(ii) . M
. M ⊥ = {0}. ♦

2.9.2 Orthogonality Results

Lemma 2.9.10 Let .(X, (·, ·)) be an inner product space. Let . M be a subspace of . X .
Then, . M ⊥ is a subspace of . X . ♦

Proof Let .v ∈ M. Then, .(0, v) = 0. Hence, .0 belongs to . M ⊥ . Let .λ, .λ, ∈ R. Let .u,
, ⊥
.u ∈ M . Let .v ∈ M. Then,

(λu + λ, u , , v) = λ(u, v) + λ, (u , , v) = λ0 + λ, 0 = 0.
.

Thus, .λu + λ, u , belongs to . M ⊥ . Hence, . M ⊥ is closed under linear combination.

Therefore, . M ⊥ is a subspace of . X . Q.E.D.

Lemma 2.9.11 A vector subspace . M of . X is dense in . X if, and only if,

M ⊥ = {0}.
. ♦

Proof For any vector subspace . M of . X , it is easy to see that . M ⊥ = (M)⊥ . If . M is a

⊥
dense subspace of . X , i.e., . M = X , then of course,o
.M = X ⊥ = {0}. Reciprocally, if
⊥ ⊥ ⊥
.M = {0}, then .(M) = {0}. Since . X = (M) M, then it follows that . M = X ,
and so . M is dense in . X. Q.E.D.

Lemma 2.9.12 A finite dimensional subspace is closed. ♦

@seismicisolation
@seismicisolation
32 2 Fundamentals

Lemma 2.9.13 Let . X be a Hilbert space. Then,

.(i) For any subsets . M, . N ⊂ X , we have if . M ⊂nN , then . N ⊥ ⊂ M ⊥ .
.(ii) For any subset . M of . X containing zero, . M M ⊥ = {0}.
⊥
.(iii) .{0} = X .
⊥
.(iv) . X = {0}. ♦
n ⊥
Proof .(ii) Let .x ∈ M M . Then, .x ∈ M and .x ∈ M ⊥ . Hence, . M ⊥ ⊂ x ⊥ and so
⊥ ⊥
. x ∈ x . Thus, . x ∈ x if, and only if, .(x, x) = 0 if, and only if, .x = 0.
⊥
n
.(iv) Since . X ⊂ X , then .(ii) implies that . X ⊥ = X X ⊥ = {0}.
Parts .(i) and .(iii) are left to the reader in exercise. Q.E.D.

Lemma 2.9.14 Let . A and . B be two subspaces of a Hilbert space . X . Then,

n
. (A + B)⊥ = A⊥ B ⊥ . ♦
n
Proof Let.x ∈ A⊥ B ⊥ . Then,.(x, a) = 0 for all.a ∈ A and.(x, b) = 0 for all.b ∈ B.
Let . y ∈ A + B, then . y = a + b with .a ∈ A and .b ∈ B. Hence,

(x, a + b) = (x, a) + (x, b) = 0.

.

n
So, . A⊥ B ⊥ ⊂ (A + B)⊥ .
Conversely, let .x ∈ (A + B)⊥ , then .(x, a + 0) = 0 for all .a ∈ A and
n hence .x ∈
A and .(x, 0 + b) = 0 for all .b ∈ B and hence .x ∈ B ⊥ . So, .x ∈ A⊥ B ⊥ , which
⊥

completes the proof. Q.E.D.

Lemma 2.9.15 Let .U be a continuous linear operator on a Hilbert space . X . Then,

. I m(U ∗ ) = (K er (U ))⊥ ,

where . I m(U ) and . K er (U ) denote the range and the null space of .U ,
respectively. ♦

Proof Let .x ∈ I m(U ∗ ), then there is .z ∈ X such that .x = U ∗ (z). Let . y ∈ K er (U ),

we have
∗
.(x, y) = (U (z), y) = (z, U (y)) = 0

for all . y ∈ K er (U ). Hence,

. I m(U ∗ ) ⊂ (K er (U ))⊥ .

Thus,
. I m(U ∗ ) ⊂ (K er (U ))⊥ = (K er (U ))⊥ .

@seismicisolation
@seismicisolation
2.10 Basis 33

Reciprocally, let .x ∈ (I m(U ∗ ))⊥ , then

(x, U ∗ (z)) = 0
.

for all .z ∈ X . This implies that

(U (x), z) = 0
.

for all .z ∈ X . Hence,

U (x) = 0.
.

So, .x ∈ K er (U ). Thus,
(I m(U ∗ ))⊥ ⊂ K er (U ).
.

which give
(K er (U ))⊥ ⊂ ((I m(U ∗ ))⊥ )⊥ .
.

Furthermore,
((I m(U ∗ ))⊥ )⊥ = I m(U ∗ ).
.

So,
. (K er (U ))⊥ ⊂ I m(U ∗ ). Q.E.D.

2.10 Basis

2.10.1 Orthonormal Basis

2.10.1.1 Definition

Definition 2.10.1 .(i) A vector .x ∈ X is called unitary or normed if .||x|| = 1.

(ii) A family .(xi )i is an orthonormal family if it is an orthogonal family whose
.

vectors are normalized. ♦

We assume, in this part, that . X is an Euclidean space with dimension .n. An orthonor-
mal basis of . X is an orthogonal family .(e1 , . . . , en ) whose vectors are unitary. It is in
particular a basis of. X . So, . X has orthonormal basis. If .(e1 , . . . , en ) is an orthonormal
basis of . X , then for all .x ∈ X , .x is uniquely written as

E
n
. x= (x, ei )ei .
i=1

The real .(x, ei ) is called coordinate of .x compared to .ei in the basis .(e1 , . . . , en ).

@seismicisolation
@seismicisolation
34 2 Fundamentals

• If .(e1 , . . . , en ) is an orthonormal basis of . X and if

.

E
n
. x= xi ei
i=1

and
E
n
. y= yi ei ,
i=1

then we can calculate the inner product and the norm by the following formulas:

E
n
. (x, y) = xi yi
i=1

( n ) 21
E
||x|| =
. xi2 .
i=1

Now, given an orthonormal basis of an Euclidean space, we have the following

identities (to be verified in exercise, see [2–7]):

Proposition 2.10.2 Let . X be an Euclidean space and .(e1 , . . . , en ) be an orthonor-

mal basis of . X . Then, for all .x, . y of . X , we have

E
n
. x= (x, ei )ei ,
i=1

E
n
||x||2 =
. |(x, ei )|2 ,
i=1

and
E
n
. (x, y) = (x, ei )(y, ei ). ♦
i=1

2.10.1.2 Orthogonal Projection

An operator . P, on a linear space . F, is a projection if . P 2 = P, i.e., . P z = z for all .z

in the range of . P.
We assume, in this part, that . F is a subspace of . X of finite dimension.

@seismicisolation
@seismicisolation
2.10 Basis 35

• F ⊥ is an additional subspace of . F called additional orthogonal of . F. If . X is himself

. .

of finite dimension, we have in particular

. dim(F ⊥ ) = dim(X ) − dim(F).

o ⊥
.• Let .x ∈ X , which is written only .x = y + z in the direct sum . F F , . y ∈ F and
⊥
. z ∈ F . Then, . y is called orthogonal projection of . x on . F, and is noted . p F (x).

.• If . M is a dense subset of . X and if .x ∈ M ⊥ , then in fact, .x ∈ X ⊥ which implies

. x = 0 (see Lemma 2.9.11).

• If .(e1 , . . . , e p ) is an orthonormal basis of . F, then

.

E
p
. p F (x) = (x, ei )ei .
i=1

• For all .x ∈ X and for all . f ∈ F, we have

.

||x − f || ≥ ||x − p F (x)||

.

with equality if, and only if, . f = p F (x). The amount

.dist(x, F) = ||x − p F (x)||

= inf ||x − f ||
f ∈F

is called distance between .x and . F.

• Since the projection, . p F (x), of an element .x is orthogonal to .x − p F (x) and, using
.

the Pythagorean relation, we have

/
dist(x, F) =
. ||x||2 − || p F (x)||2 .

• Note that the projection, . PF (x), from a point .x on a vector subspace of finite
.
dimension . F, is easily calculated using an orthonormal basis of . F (evidence left in
exercise):
Proposition 2.10.3 Let . X be a prehilbertian space and . F be a vector subspace of
finite dimension with .(e1 , . . . , en ) is an orthonormal basis of . F. Then, for all .x ∈ X ,
we have
En
. PF (x) = (x, ei )ei . ♦
i=1

@seismicisolation
@seismicisolation
36 2 Fundamentals

2.10.2 Hilbert Basis

2.10.2.1 Definition

Definition 2.10.4 Let . X be a Hilbert space. We call Hilbert basis of . X all sequence
(en )n of . X checking
.

(i) .||en || = 1 for all .n ∈ N and

.

(en , em ) = δnm ,
.

with .δnm is the symbol of Kronecker.

.(ii) The vector subspace of . X generated by .(en )n , denoted . F, is dense in . X , i.e.,
.F = X . ♦

Remark 2.10.5 The Hilbert basis is not, in general, an algebraic basis. ♦

2.10.2.2 Gram-Schmidt Process

• If .(x1 , . . . , x p ) is a linearly independent family of . X , then there exists a unique

.
orthonormal family .(e1 , . . . , e p ) satisfying the following two conditions:
(i) for all .k ∈ {1, . . . , p}, .vect (e1 , . . . , ek ) = vect (x1 , . . . , xk );
.
(ii) for all .k ∈ {1, . . . , p}, .(ek , xk ) > 0.
.

The passage of .(x1 , . . . , x p ) to .(e1 , . . . , e p ) named Gram-Schmidt orthonormaliza-

tion method. Recall that the Gram-Schmidt method allows to calculate an orthonor-
mal basis of an Euclidean space from a given basis:
Proposition 2.10.6 (Gram-Schmidt process) Let . X be an Euclidean space and
.(e1 , . . . , en ) be a basis of . X . For each .0 < i < n, note . Fi the vector subspace
, ,
.vect(e1 , . . . , ei ) generated by .e1 , . . . , ei . Then, the family .(e1 , . . . , en ) defined as
follows:

e1,
e, =
. 1
||e1, ||
E
i−1
ei − (ei , ek, )ek,
ei − p Fi−1 (x)
ei, =
k=1
= || || for 1 < i ≤ n,
||ei − p Fi−1 (x)|| || E
i−1 ||
|| , , ||
||ei − (ei , ek )ek ||
|| ||
k=1

is an orthonormal basis of . X . ♦

@seismicisolation
@seismicisolation
2.10 Basis 37

2.10.2.3 Characterization of a Hilbert Basis

Proposition 2.10.7 Let . X be a Hilbert space. An orthogonal sequence .(en )n of . X is

an Hilbert basis if, and only if, .(en )n is maximal, i.e., if there is no normal vector of
. X orthogonal to all vectors of the sequence .(en )n . ♦

Proof Let .(en )n be an Hilbert basis of . X . Suppose that .(en )n is not maximal. Hence,
there is a vector .u 0 /= 0 such that .u 0 is orthogonal to all vectors of the sequence
⊥
.(en )n . We have .u 0 is a closed vector subspace of . X . Thus,

. F ⊂ u⊥
0.

which give
. F ⊂ u⊥
0.

It follows that
. X ⊂ u⊥
0.

Furthermore, .u 0 ∈ X so
. 0 u ⊂ u⊥
0

which is absurd.
Reciprocally, suppose that . F /= X . Since
o ⊥
. X=F F ,

then
⊥
. F /= {0}.
⊥
So, there is .u 0 /= 0 and .u 0 ∈ F such that for all .v ∈ F, we have

(u 0 , v) = 0.
.

which give for all .v ∈ F,

(u 0 , v) = 0.
.

Thus, for all .n ∈ N

(u 0 , en ) = 0.
.

This proves that .(en )n is not maximal, which is absurd. Q.E.D.

Theorem 2.10.8 Any separable Hilbert space . X , of infinite dimension, has a count-
able Hilbert basis. ♦

@seismicisolation
@seismicisolation
38 2 Fundamentals

Proof Let .{gn , n ∈ N} be a dense sequence in . X . We can extract, by induction, a

subsequence .(gni )i such that
.(i) for all .i ∈ N, .{gn j , 0 ≤ j ≤ i} is a linearly independent family,
.(ii) for all .iU∈ N, .vect{gk , 0 ≤ k ≤ n i } = vect{gn j , 0 ≤ j ≤ i} =: E i ,
.(iii) . E := i∈N E i ⊃ {gn , ∈ N}.

By orthogonalizing.(gni )i , by the process of Gram-Schmidt, we construct an orthonor-

mal sequence of .( f ni )i such that for all .i ∈ N, .( f n j )0≤ j≤i is an orthonormal basis of
the finite dimensional space . E i . To show that the sequence .( f ni )i is a Hilbertian basis
of . X , it remains to verify that it is total in . X . Let . f be arbitrary in . X and .ε > 0.
Using .(iii) and the density of .{gn , n ∈ N} in . X , there is .g ∈ E such that

|| f − g|| < ε.
.

Since . E is the meeting of . E i , there is an index .i ∈ N such that .g ∈ E i . However,

the .( f n j )0≤ j≤i form a basis of space of finite dimension . E i so .g decomposes, in this
basis, in the form
E i
.g = a j fn j .
j=0

Of course, .g, .i, and the .a j depend on . f and on .ε, but the reasoning is valid for all . f
and all .ε. The sequence .( f ni )i is, therefore, total in . X . Q.E.D.
Corollary 2.10.9 . X is a separable Hilbert space if, and only if, . X admits a Hilbert
basis. ♦
Example 2.10.1 Let .l 2 be the space defined by
{ ∞
}
E
l 2 = (xn )n real sequence such that
. xi2 < ∞ .
i=1

We provide .l 2 of the following inner product

∞
E
(x, y) =
. xi yi .
i=1

Let .n ∈ N∗ . We pose
e = (0, . . . , 0, 1, 0, . . .).
. n
, ,, ,
(n−1)

We can easily verify that .(en )n is orthonormal. In fact,

∞
E
.||en ||2 = (en )i2 = 1
i=1

@seismicisolation
@seismicisolation
2.11 Matrix 39

and
(en , em ) = δnm .
.

Let .v ∈ l 2 such that .v = (vn )n and

(v, en ) = 0
.

for all .n ∈ N∗ . Then,

∞
E
(v, en ) =
. (en )i vi = vn = 0
i=1

for all .n ∈ N∗ , which proves

.v = 0.

Thus, .(en )n is maximal, which proves, using the Proposition 2.10.7, .(en )n is a Hilbert
basis.

2.11 Matrix

We denote by .Mn (R) the set of square size matrices .n with coefficients in .R.

2.11.1 Positive Definite Matrix

Definition 2.11.1 .(i) We say that the symmetric matrix. A ∈ Mn (R) is positive semi-
definite if
.(Ax, x) ≥ 0

for all .x ∈ Rn .
.(ii) They say she is positive definite if, in addition,

(Ax, x) = 0 implies x = 0.
.

This, therefore, implies that a symmetric matrix is positive definite if, and only if,
(Ax, x) > 0, for all .x /= 0.
. ♦

Lemma 2.11.2 A matrix . A ∈ Mn (R) is positive definite if, and only if, .tr (A) > 0
and .det A > 0, where .tr (A) is the trace of the matrix . A and .det A is the determinant
of the matrix . A. ♦

@seismicisolation
@seismicisolation
40 2 Fundamentals

2.11.2 Orthogonal Matrix

Definition 2.11.3 We call orthogonal matrix of size (or order) .n, any matrix . O ∈
Mn (R) checking .t O O = In (we, then, have . O t O = In ). ♦
• A matrix is orthogonal if, and only if, its columns form an orthonormal basis of
.
Rn , if, and only if, its lines form an orthonormal basis of .Rn .
.

• The set of orthogonal matrices of size .n is a multiplicative group. Moreover, he

.
forms a group called orthogonal group and noted .On (R).

2.11.3 Monotonic Matrix

Definition 2.11.4 A square matrix . A of order .n is said to be monotonic if it is

invertible and its inverse is positive. ♦
Lemma 2.11.5 A square matrix . A of order .n is monotonic if, and only if,
. {v ∈ Rn such that Av ≥ 0} ⊂ {v ∈ Rn such that v ≥ 0}. ♦

2.12 Projection on a Closed Convex

2.12.1 Projection Theorem

Theorem 2.12.1 (Projection) Assume . M is a closed subspace of a Hilbert space . X .

Then, for each .x ∈ X , there exist an unique . y ∈ M and .z ∈ M ⊥ such that .x = y + z.
The element . y is called the projection of .x onto . M. ♦
Proof Let . S = {x − y : y ∈ M}. It is easy to see that . S is convex and closed.
Theorem 2.6.5 implies that there exists an . y ∈ M such that .||x − y|| ≤ ||x − w|| for
all .w ∈ M. Let .z = x − y. For an arbitrary .w ∈ M, .w /= 0, let .α = (z,w)
||w||2
and note
that

||z||2 ≤ ||z − αw||2

.

|(z, w)|2
= ||z||2 − ,
||w||2
⊥ , , , ,
which implies .(z, w) = 0. Therefore,
n .z ∈ M . If .x = y + z for some . y ∈ M, .z ∈
M ⊥ , then . y , − y = z − z , ∈ M M ⊥ = {0}, which implies uniqueness. Q.E.D.

Remark 2.12.2 In particular, if . M is a proper closed subspace of the space . X , then

there exists a non-zero element in . M ⊥ . In fact, for .x ∈ X \M, let . y be the projection
of .x on . M. Then, .z = x − y is a non-zero element of . M ⊥ . ♦

@seismicisolation
@seismicisolation
2.12 Projection on a Closed Convex 41

Theorem 2.12.3 Let . X be a Hilbert space and . K be a non-empty, closed, convex

set, of . X . Then, for all . f ∈ X , there is .u ∈ K unique such that

|| f − u|| = inf || f − v||.

. (2.12.1)
v∈K

In addition, .u is characterized by the property

{
u∈K
. (2.12.2)
( f − u, v − u) ≤ 0 for all v ∈ K .

Proof Let . f ∈ X .
Existence: The set .{|| f − v||, v ∈ K } is a part of .R which is under estimated. So, it
admits a lower bound. Let .δ be this lower bound, i.e.,

δ = inf || f − v||.
.
v∈K

So, there is .(vn )n ∈ K such that .|| f − vn || → δ when .n → ∞. We pose

δ := || f − vn ||. Hence,
. n

.δn → δ when n → ∞. (2.12.3)

Let us show that .(vn )n is a Cauchy sequence. Indeed, according to the parallelogram
identity (Proposition 2.6.4) for . f − vn and . f − vm , we have
|| ||2 || ||
|| vn + vm || || vn − vm ||2
|| || || || = 1 || f − vn ||2 + 1 || f − vm ||2 .
.
|| f − 2 ||
+ ||
2 || 2 2

Hence, || || || ||2
|| vn − vm ||2 || ||
|| || = 1 δ 2 + 1 δ 2 − || f − vn + vm || . (2.12.4)
.
|| 2 || 2 n
2 m || 2 ||

The fact that . K is convex gives . vn +v

2
m
∈ K . Hence, using Eq. (2.12.4), we have

. ||vn − vm ||2 ≤ 2(δn2 + δm2 ) − 4δ 2 . (2.12.5)

Let .ε > 0. Using Eq. (2.12.3), we deduce the existence of a certain . N > 0 such that
for all .n, .m > N , we have
.2(δn + δm ) − 4δ < ε.
2 2 2
(2.12.6)

Using the Eqs. (2.12.6) and (2.12.5), we obtain

||vn − vm || < ε for all n, m > N .

.

@seismicisolation
@seismicisolation
42 2 Fundamentals

So, .(vn )n is a Cauchy sequence on . X . As . X is complete, so .(vn )n is a convergent

sequence in . X . Let
.u = lim vn .
n→∞

As .(vn )n ∈ K and . K is closed, we have .u ∈ K . Hence, .δn = || f − vn || converges to

δ when .n → ∞. It follows that
.

|| f − u|| = inf || f − v||.

.
v∈K

Equivalence between (2.12.1) and (2.12.2).

Let .u ∈ X be such that .u checks (2.12.1). Let .v ∈ K . We pose .w = (1 − t)u +
tv ∈ K for .t ∈]0, 1]. Then,

.|| f − u||2 ≤ || f − w||2

= ( f − w, f − w)
= (( f − u) − t (v − u), ( f − u) − t (v − u))
≤ || f − u||2 − 2t( f − u, v − u) + t 2 ||v − u||2 . (2.12.7)

Equation (2.12.7) gives

t
( f − u, v − u) ≤
. ||v − u||2 . (2.12.8)
2

Now, Eq. (2.12.2) is obtained by making .t tend towards .0+ in Eq. (2.12.8).
Conversely: Let .u ∈ X such that .u checks (2.12.2). Let .v ∈ K , we have

|| f − u||2 − || f − v||2 = ( f − u, f − u) − ( f − v, f − v).

.

Then,

.|| f − u||2 − || f − v||2 = ( f − u, f − u) − ( f − v, f − v)

= ( f − u, v − u) + ( f − u, f − v) − ( f − v, f − v)
= ( f − u, v − u) + ( f − v, v − u)
= 2( f − u, v − u) + (u − v, v − u)
= 2( f − u, v − u) − ||v − u||2 ≤ 0.

This implies that, for all .v ∈ K ,

|| f − u|| ≤ || f − v||.
.

So,
|| f − u|| = inf || f − v||.
.
v∈K

@seismicisolation
@seismicisolation
2.12 Projection on a Closed Convex 43

Uniqueness: We assume that there are .u 1 , .u 2 ∈ K such that

. ( f − u 1 , v − u 1 ) ≤ 0 for all v ∈ K (2.12.9)

and
.( f − u 2 , v − u 2 ) ≤ 0 for all v ∈ K . (2.12.10)

Taking .v = u 2 in (2.12.9) and .v = u 1 in (2.12.10), we obtain

. ( f − u1, u2 − u1) ≤ 0 (2.12.11)

and
( f − u 2 , u 1 − u 2 ) ≤ 0.
. (2.12.12)

By making the sum of (2.12.11) and (2.12.12), we obtain

.||u 2 − u 1 ||2 ≤ 0

and hence .u 2 = u 1 . This completes the proof of the theorem. Q.E.D.

Remark 2.12.4 Let . X be a Hilbert space and . K be a non-empty, closed, convex

set, of . X and let . f ∈ X . Let’s pose .u the orthogonal projection, . PK f , of . f on . K ,
i.e., .u = PK f . Since . K is convex, we can easily see, in the figure below, that the
angle between the vector . f − u and the vector .v − u is always greater than . π2 . Which
justifies that the term .( f − u, v − u) is negative.

f
f −u

u
K

−u
v

v−u

Which only justifies the inequality (2.12.2). ♦

@seismicisolation
@seismicisolation
44 2 Fundamentals

2.12.2 A Simple Consequence of the Projection Theorem

Corollary 2.12.5 Let . X be a Hilbert space and . M ⊂ X be a closed vector subspace

of . X . Then, for all . f ∈ X , there is .u ∈ M unique such that

|| f − u|| = inf || f − v||.

. (2.12.13)
v∈M

In addition, .u is characterized by the property

{
u∈M
. (2.12.14)
( f − u, v) = 0 for all v ∈ M.

♦
Proof Since. M ⊂ X is a closed vector subspace of. X , then. M is a non-empty, closed,
convex set, of . X . For . f ∈ X , according to the closed convex projection theorem
(Theorem 2.12.3), there is .u ∈ M unique such that

|| f − u|| = inf || f − v||.

.
v∈M

In addition, .u is characterized by the property

{
u∈M
. (2.12.15)
( f − u, w − u) ≤ 0 for all w ∈ M.

Let us show the equivalence between (2.12.13) and (2.12.14).

Let .u ∈ X such that .u checks (2.12.13). Hence, .u checks (2.12.15). Let .v ∈ M.
We pose .w = u + tv ∈ M, .t ∈ R∗ . So,

( f − u, w − u) = ( f − u, tv) ≤ 0.
.

Thus,
t( f − u, v) ≤ 0
.

for all .t ∈ R∗ . If .t > 0, then

. ( f − u, v) ≤ 0. (2.12.16)

If .t < 0, then
. ( f − u, v) ≥ 0. (2.12.17)

Equations (2.12.16) and (2.12.17) give

( f − u, v) = 0
.

for all .v ∈ M.

@seismicisolation
@seismicisolation
2.12 Projection on a Closed Convex 45

Conversely: Let .u ∈ M such that

( f − u, v) = 0
.

for all .v ∈ M. Let .v ∈ M, we have

.|| f − u||2 − || f − v||2 = ( f − u, f − u) − ( f − v, f − v)

= 2( f − u, v − u) − ||v − u||2
= −||v − u||2 ≤ 0.

This completes the proof. Q.E.D.

Corollary 2.12.6 Let . X be a Hilbert space and . M ⊂ X be a closed vector subspace

of . X . Then, for all . f ∈ X , there is .u ∈ M unique such that

. || f − u|| = inf || f − v||.

v∈M

In addition, .u is characterized by the property .u ∈ M and . f − u ∈ M ⊥ . ♦

Proposition 2.12.7 The projection . PM on a closed vector subspace . M of a Hilbert

space . X is linear. ♦

Proof Let . f 1 , . f 2 ∈ X , and .λ1 , .λ2 ∈ R. We pose .u = PM (λ1 f 1 + λ2 f 2 ), .u 1 = PM f 1

and .u 2 = PM f 2 . Then, according to the Corollary 2.12.5, we have

. ( f 1 − u 1 , v) = 0 for all v ∈ M (2.12.18)

.( f 2 − u 2 , v) = 0 for all v ∈ M (2.12.19)

and
.(λ1 f 1 + λ2 f 2 − u, v) = 0 for all v ∈ M. (2.12.20)

Equation (2.12.18) gives

. (λ1 u 1 − λ1 f 1 , v) = 0 (2.12.21)

for all .v ∈ M. Equation (2.12.19) gives

.(λ2 u 2 − λ2 f 2 , v) = 0 (2.12.22)

for all .v ∈ M. By summing (2.12.20), (2.12.21), and (2.12.22), we obtain

(λ1 u 1 + λ2 u 2 − u, v) = 0
. (2.12.23)

@seismicisolation
@seismicisolation
46 2 Fundamentals

for all .v ∈ M. In particular, for .v = λ1 u 1 + λ2 u 2 − u in Eq. (2.12.23), we deduce

.||λ1 u 1 + λ2 u 2 − u||2 = 0.

Hence,
. u = λ1 u 1 + λ2 u 2 .

Thus, . PM is linear. Q.E.D.

Proposition 2.12.8 Let . X o be a Hilbert space and . M ⊂ X be a closed vector sub-

space of . X . Then, . X = M M ⊥. ♦
n ⊥
Proof Let .u ∈ M M . Then, .u ∈ M and

(u, v) = 0
.

for all .v ∈ M. In particular, for .v = u, we find

. ||u||2 = 0.

Thus, .u = 0. Hence, n
. M M ⊥ = {0}.

We still have to prove that . X = M + M ⊥ . Let . f ∈ X , then

. PM f ∈ M.

According to the Corollary 2.12.5,

.( f − PM f, v) = 0

for all .v ∈ M. Hence,

. f − PM f ∈ M ⊥ .

It follows that
. f = PM f + f − PM f ∈ M + M ⊥ .

Therefore,
. X ⊂ M + M ⊥.

Furthermore, . M + M ⊥ ⊂ X . Thus,
. X = M + M ⊥. Q.E.D.

@seismicisolation
@seismicisolation
2.12 Projection on a Closed Convex 47

Proposition 2.12.9 Let . X be a Hilbert space and . M ⊂ X be a closed vector sub-

space of . X . Then, the mapping

. Q : X −→ X
f −→ Q( f ) = f − PM f

is the projection operator on . M ⊥ , i.e., . Q = PM ⊥ . ♦

Proof We only show that

. ( f − Q( f ), v) = 0

for all .v ∈ M ⊥ . In fact, let .v ∈ M ⊥ , we have

( f − Q( f ), v) = (PM f, v) = 0.
.

On the other hand, according to the Corollary 2.12.5,

( f − PM ⊥ f, w) = 0
.

for all .w ∈ M ⊥ . According to the uniqueness

. PM ⊥ f = Q( f ).

Thus, . Q is then the operator of projection on . M ⊥ . Q.E.D.

Proposition 2.12.10 Let. X be a Hilbert space and. K be a non-empty, closed, convex

set, of . X . Then,
.||PK f 1 − PK f 2 || ≤ || f 1 − f 2 || (2.12.24)

for all . f 1 , . f 2 ∈ X. ♦

Proof Let . f 1 , . f 2 ∈ X .

f2 − f1
f2 f1

f2 − u 2 f1 − u 1

u2 u2 − u1 u1

u 1 = PK f 1
u 2 = PK f 2

@seismicisolation
@seismicisolation
48 2 Fundamentals

According to the closed convex projection theorem (Theorem 2.12.3), we have

. ( f 1 − PK f 1 , v − PK f 1 ) ≤ 0 (2.12.25)

for all .v ∈ K and

( f 2 − PK f 2 , v − PK f 2 ) ≤ 0
. (2.12.26)

for all .v ∈ K . Taking .v = PK f 2 in (2.12.25) and .v = PK f 1 in (2.12.26), we get

. ( f 1 − PK f 1 , PK f 2 − PK f 1 ) ≤ 0 (2.12.27)

and
.( f 2 − PK f 2 , PK f 1 − PK f 2 ) ≤ 0. (2.12.28)

Either by summing (2.12.27) and (2.12.28), we obtain

(PK f 1 − PK f 2 , −( f 1 − f 2 ) + (PK f 1 − PK f 2 )) ≤ 0.
.

Thus,
||PK f 1 − PK f 2 ||2 − (PK f 1 − PK f 2 , f 1 − f 2 ) ≤ 0.
.

Cauchy-Schwarz inequality gives

||PK f 1 − PK f 2 ||2 ≤ ||PK f 1 − PK f 2 |||| f 1 − f 2 ||.

. (2.12.29)

Equation (2.12.24) is obvious if . PK f 1 = PK f 2 . Otherwise, Eq. (2.12.24) follows

from Eq. (2.12.29). Q.E.D.
Remark 2.12.11 Let . X be a Hilbert space and . M ⊂ X be a closed vector subspace
of . X . Then,
.(i) . PM + PM ⊥ = I d X .
.(ii) . I m(PM ) = M, . I m(PM ⊥ ) = M ⊥ , . K er (PM ) = M ⊥ , and . K er (PM ⊥ ) = M.
.(iii) .||PM || = ||PM ⊥ || = 1. In fact, . PM PM = PM . Hence,

||PM || ≥ 1.
.

On the other hand, according to Proposition 2.12.10, we have

.||PM f || ≤ || f ||.

So,
||PM || ≤ 1.
.

Thus,
. ||PM || = 1. ♦

@seismicisolation
@seismicisolation
2.12 Projection on a Closed Convex 49

Lemma 2.12.12 Let . X be a separable Hilbert space and . M be a closed subspace

of . X . Then, . M is separable. ♦

Proof We have . X is separable so it admits a dense sequence in . X , let .(vm )m this

sequence. We pose
.wm = PM vm

for all .m ∈ N. Let us show that .(wm )m is dense in . M. In fact, let . O be an open
non-empty of . M. Then, there is .U open non-empty of . X such that
n
. O =U M.

Hence, for all .v ∈ O, there is .ε > 0 such that

n
. B(v, ε) M⊂O

with . B(v, ε) is the open ball with center .v and radius .ε. Hence, there is .vn 0 ∈ B(v, ε)
such that
.vn 0 − wn 0 ⊥ v − wn 0 .

Thus,
.||vn 0 − wn 0 ||2 + ||v − wn 0 ||2 = ||v − vn 0 ||2 ,

which prove .wn 0 ∈ B(v, ε) and .wn 0 ∈ M. It follows that .wn 0 ∈ O and so .(wm )m is
dense in . M. Q.E.D.

Theorem 2.12.13 (Bessel inequality) Let . X be a Hilbert space and .(ei )i∈N∗ be an
orthonormal family. Then, for all .x ∈ X,
∞
E
. |(x, ei )|2 ≤ ||x||2 . ♦
i=1

Proof Let .nE∈ N∗ . The projection of .x on the vector space generated

En by .e1 , . . . , en is
n
equal to . i=1 (x, ei )ei and it is orthogonal to .x − i=1 (x, ei )ei (see
Corollary 2.12.6). This orthogonality implies that:
|| n ||2 || ||2
||E || || En ||
|| || || ||
.||x|| = || (x, ei )ei || + ||x − (x, ei )ei ||
2
|| || || ||
i=1 i=1
|| ||2
E n || En ||
|| ||
= |(x, ei )| + ||x −
2
(x, ei )ei ||
|| ||
i=1 i=1
E
n
≥ |(x, ei )|2 .
i=1

@seismicisolation
@seismicisolation
50 2 Fundamentals

The last inequality is true for all .n ∈ N∗ . So, the series

∞
E
. |(x, ei )|2
i=1

is convergent, which completes the proof. Q.E.D.

Theorem 2.12.14 Let . X be a Hilbert space having a countable Hilbert basis

(ei )i∈N∗ . Then, for all .x, . y of . X , we have
.

∞
E
.x = (x, ei )ei ,
i=1

∞
E
||x||2 =
. |(x, ei )|2 ,
i=1

and
∞
E
. (x, y) = (x, ei )(y, ei ). ♦
i=1

Proof Note . F the vector subspace generated by the family .(ei )i∈N∗ and for any .n,
note . Fn the vector subspace generated by .e1 , . . . , en . Since . F is dense in . X , there is
a sequence .(xn )n elements of . F converging to .x. Even, if it means adding enough
zero terms to the start of the sequence and Erepeating terms, we can assume that for
n
each .n, the term .xn belongs to . Fn . Since . i=1 (x, ei )ei is the projection of .x on . Fn ,
we thus obtain, by using Theorem 2.12.3, the inequality
|| ||
|| E
n ||
|| ||
. || x − (x, ei )ei || ≤ ||x − xn ||.
|| ||
i=1

En
which imply that . i=1 (x, ei )ei converges to .x. Passing to the limit in equality
|| n ||2 || ||2
||E || || En ||
|| || || ||
.||x|| = || (x, ei )ei || + ||x − (x, ei )ei ||
2
|| || || ||
i=1 i=1
|| ||
E n || En ||2
|| ||
= |(x, ei )|2 + ||x − (x, ei )ei || ,
|| ||
i=1 i=1

we obtain the second identity of the theorem. Note that by bilinearity with the fol-
lowing inner product, we have

@seismicisolation
@seismicisolation
2.12 Projection on a Closed Convex 51
/ n \
E E
n E
n
. (x, ei )ei , (y, ei )ei = (x, ei )(y, ei ).
i=1 i=1 i=1

By continuity of the inner product, the left term converges to .(x, y), which proves
the last identity. Q.E.D.

Theorem 2.12.15 Let . X be a Hilbert space on .R, .(en )n≥1 be a Hilbert basis of . X
and let . E be the vector subspace generated by .a and .(en )n≥2 , with
∞
E en
a=
. .
n=1
n

Then, . E = X . ♦

Proof Suppose that . E /= X , hence there is .x ∈ X and .x ∈

/ E. We have . E is a closed
convex subspace of . X , and . X is a Hilbert space so according to the closed convex
projection theorem (Theorem 2.12.3), there is .u ∈ X unique such that

(x − u, v) = 0
.

for all .v ∈ E. In particular,

(x − u, v) = 0
.

for all .v ∈ E. Hence,

. (x − u, en ) = 0

for all .n ∈ N∗ \{1} and

(x − u, a) = 0.
.

Furthermore,
/ ∞
\
E en
(x − u, a) = x − u,
.
n=1
n
E∞
1
= (x − u, en )
n=1
n
= (x − u, e1 )
= 0.

Hence, for all .n ∈ N∗ ,

(x − u, en ) = 0.
.

@seismicisolation
@seismicisolation
52 2 Fundamentals

Since .(en )n≥1 is a Hilbert basis of . X , we have

∞
E
. x −u = (x − u, en )en = 0.
n=1

Thus, .x − u = 0 and so .x = u ∈ E, which is absurd.

Other method: We have
E∞
en
.a − = e1 .
n=2
n

Consider the sequence

E
n
ek
u =a−
. n .
k=2
k

We have .u n ∈ E and
. lim u n = e1 .
n→∞

Hence,
. 1e ∈ E.

However, . E ⊂ E, so
(en )n≥1 ∈ E.
.

Thus,
. E = X.

This completes the proof. Q.E.D.

Theorem 2.12.16 Let . X be a Hilbert space on .R, and .(en )n≥1 be a Hilbert basis of
X . We pose for .n ≥ 1
.
.u n = en + 2en+1

and
. F = ((u n )n≥1 )

the closed vector subspace generated by .u n . Then,

. F ⊥ = (e),

where
E∞ ( )
1 n−1
.e = − en .
n=1
2

@seismicisolation
@seismicisolation
2.12 Projection on a Closed Convex 53

Furthermore, the expression of the orthogonal projection . PF (x) of an element .x of

X on . F is given by
.

3
. PF (x) = x − (x, e)e. ♦
4
Proof We have

. F ⊥ = {x ∈ X such that (x, u n ) = 0 for all n ∈ N∗ }.

Let .x ∈ X . Since .(en )n is a Hilbert basis, we can write .x under the following form
∞
E
. x= xn en
n=1

with the condition

∞
E
. |xn |2 < ∞.
n=1

If .(x, u n ) = 0, then
∞
E
. xi (ei , en + 2en+1 ) = 0.
i=1

Hence,
. n x + 2xn+1 = 0,

which proves the equalities

1
x
. n+1 = − xn .
2
Thus, ( )
1 n−1
x = −
. n x1 .
2

Let
E∞ ( )
1 n−1
.e= − en .
n=1
2

However,
E∞ ( )
1 2(n−1)
. −
n=1
2

is a convergent series. Hence, .e ∈ F ⊥ . Thus,

. F ⊥ = (e).

@seismicisolation
@seismicisolation
54 2 Fundamentals

According to the closed convex projection theorem (Theorem 2.12.3), we have for
all .x ∈ F ⊥ ,
. x = PF (x) + PF ⊥ (x).

Furthermore,

(x, e)
. PF ⊥ (x) = e
||e||2
3
= (x, e)e.
4
Thus,
3
. PF (x) = x − (x, e)e.
4
This completes the proof. Q.E.D.

2.13 Contraction Mapping Theorem

Let . X be a normed space. A map .T : X −→ X is called a contraction if there exists

a number .k < 1 such that

.||T x − T y|| ≤ k||x − y|| for all x, y ∈ X.

Theorem 2.13.1 (Contraction mapping) Let .T : S ⊂ X −→ S be a contraction on

the closed non-empty subset . S of the Banach space . X . Then, .T has a unique fixed
point, i.e., there exists a unique solution .x ∈ S of the equation .T x = x. Moreover,
. x = lim T x 0 for any choice of . x 0 ∈ S. ♦
n
n→∞

Proof To prove uniqueness, suppose .T x = x, .T y = y. Since .k < 1, we get .x = y

from

||x − y|| = ||T x − T y||

.

≤ k||x − y||.

To show that .T has a fixed point we set up an iteration procedure. For any .x0 ∈ S, set
x
. n+1 = T xn , .n = 0, 1, . . .. Note that .xn+1 ∈ S and .xn+1 = T n+1 x0 . We now claim
that .(xn )n is a Cauchy sequence. Indeed, for any integers .n, . p

@seismicisolation
@seismicisolation
2.14 The Dual of a Hilbert Space 55

||xn+ p − xn || = ||T n+ p x0 − T n x0 ||
.

E
n+ p−1
≤ ||T j+1 x0 − T j x0 ||
j=n

E
n+ p−1
≤ k j ||T x0 − x0 ||
j=n
n
k
≤ ||T x0 − x0 ||.
1−k

Thus, as .n → ∞, .||xn+ p − xn || → 0 independently of . p, so that .(xn )n is a Cauchy

sequence with limit .x ∈ S. Since .T is continuous, we have

. T x = lim T xn
n→∞
= lim xn+1
n→∞
=x

and thus .x is the unique fixed point. Note that the fixed point .x is independent of .x0
since .x is a fixed point and fixed points are unique. Q.E.D.

2.14 The Dual of a Hilbert Space

2.14.1 Riesz Fréchet’s Representation Theorem

Riesz’s theorem provides a canonical way to identify a Hilbert space and its dual.
Theorem 2.14.1 Let. X be a Hilbert space and. X , be its topological dual. Let.ϕ ∈ X , ,
then there is . f ∈ X unique such that

ϕ(v) = ( f, v)
. (2.14.1)

for all .v ∈ X. In addition,

. ||ϕ|| X , = || f || X . (2.14.2)

♦
−1 ,
Proof We pose. M = ϕ ({0}). The fact that.ϕ ∈ X implies that. M is a closed vector
subspace of . X .

Case 1: If . M = X , then .ϕ = 0. We only take . f = 0.

.

@seismicisolation
@seismicisolation
56 2 Fundamentals

Case 2: If . M c X , then we will show that there is .g ∈ X such that

.

⎧
⎨g ∈ /M
. ||g|| X = 1 (2.14.3)
⎩
(g, v) = 0 for all v ∈ M.

In fact, since . M c X , there is .g0 ∈ X and .g0 ∈ / M. We pose .g1 = PM g0 . Consider

the element
g0 − g1
. g := ∈ M ⊥.
||g0 − g1 || X

This, .g checks (2.14.3). Let .v ∈ X , then there is .λ ∈ R and .w ∈ M such that

v = λg + w.
.

Hence,
.ϕ(v) = λϕ(g) + ϕ(w).

Since .ϕ(M) = {0}, then

. ϕ(w) = 0.

So,
ϕ(v) = λϕ(g)
.

and thus
ϕ(v)
λ=
. .
ϕ(g)

By using (2.14.3), we have

0 = (g, w)
.

= (g, v − λg)
= (g, v) − λ.

which give .λ = (g, v). Hence,

ϕ(v) = ϕ(λg + w)
.

= λϕ(g)
= (g, v)ϕ(g)
= (ϕ(g)g, v)
= ( f, v)

with . f = ϕ(g)g ∈ M ⊥ .

@seismicisolation
@seismicisolation
2.14 The Dual of a Hilbert Space 57

Uniqueness: Suppose there are . f 1 , . f 2 ∈ X such that

. ϕ(v) = ( f 1 , v) (2.14.4)

for all .v ∈ X and

.ϕ(v) = ( f 2 , v) (2.14.5)

for all .v ∈ X. By substituting (2.14.4) and (2.14.5), we find

. ( f 1 − f 2 , v) = 0 (2.14.6)

for all .v ∈ X. By taking .v = f 1 − f 2 in (2.14.6), we can deduce

|| f 1 − f 2 ||2X = 0.
.

So,
f = f2 .
. 1

Thus, the uniqueness, which proves (2.14.1).

It remains to show (2.14.2). We have

|ϕ(v)|
.||ϕ|| X , = sup
v∈X, v/=0 ||v|| X
|( f, v)|
= sup
v∈X, v/=0 ||v|| X

≤ || f || X .

On the other hand,

|( f, f )|
. = || f || X .
|| f || X

Thus,
. ||ϕ|| X , = || f || X . Q.E.D.

Remark 2.14.2 Riesz’s theorem proves that the dual of a Hilbert space is also a
Hilbert space. ♦

Remark 2.14.3 .(i) The mapping

Ψ : X , −→ X
.

ϕ −→ Ψ (ϕ) = f

@seismicisolation
@seismicisolation
58 2 Fundamentals

is an isometric isomorphism with . f is the unique function of . X satisfying (2.14.1).

The isomorphism .Ψ identifies . X and . X , .
.(ii) .||Ψ (ϕ)|| = || f || X = ||ϕ|| X , (isometry). ♦

2.14.2 Example

Example 2.14.1 Consider the linear form

ϕ : Rn −→ R
.

x = (x1 , . . . , xn ) −→ ϕ(x) = x1 .

Let .x ∈ Rn . Then,

|ϕ(x)| = |x1 |
.
/
≤ x12 + · · · + xn2
= ||x||.

Hence, .ϕ is continuous. So, .ϕ ∈ (Rn ), . According to Riesz Fréchet’s representation

theorem (Theorem 2.14.1), there is . f ∈ Rn unique such that

ϕ(x) = ( f, x)
. (2.14.7)

for all .x ∈ Rn . Let .e1 = (1, 0, . . . , 0), .e2 = (0, 1, 0, . . . , 0), .. . ., .en = (0, . . . , 0, 1)
be the canonical basis of .Rn . We pose . f = ( f 1 , . . . , f n ). For .x = e1 in (2.14.7), we
have
.( f, e1 ) = ϕ(e1 ) = 1.

which gives
f = 1.
. 1

For .x = e2 in (2.14.7), we have

. ( f, e2 ) = ϕ(e2 ) = 0.

which proves
f = 0.
. 2

Hence, for .x = en in (2.14.7), we have

( f, en ) = ϕ(en ) = 0.
.

@seismicisolation
@seismicisolation
2.15 Stampacchia Theorem 59

which gives
f = 0.
. n

So, . f = (1, 0, . . . , 0).

2.15 Stampacchia Theorem

2.15.1 Continuity and Coercivity

Definition 2.15.1 Let . X be a normed vector space and .a(·, ·) be a bilinear form
on . X
.a : X × X −→ R.

(i) We say that .a(·, ·) is continuous on . X × X if there is a constant .c such that

.

|a(u, v)| ≤ c||u||||v||

.

for all .u, .v ∈ X .

(ii) We say that .a(·, ·) is coercive (or . X -elliptic) if there is a constant .α > 0 such
.
that
.a(v, v) ≥ α||v||
2

for all .v ∈ X . .α is called coercivity constant of .a(·, ·). ♦

Clearly, every . X -elliptic bilinear form .a(·, ·) induces a norm via

/
||v||a :=
. a(v, v). (2.15.1)

This is equivalent to the norm of the Hilbert space . X . The norm (2.15.1) is called
the energy norm.

Lemma 2.15.2 Let . X be a normed vector space and .a(·, ·) be a bilinear form on
.X . Assume that .a(·, ·) is continuous with constant .c ≥ 0, and coercive with constant
.α > 0. Then, .α ≤ c. ♦

Proof Let .u ∈ X be a vector. Assume that .u /= 0. Then,

α||u||2 ≤ a(u, u) ≤ c||u||2 .

.

Hence, .α ≤ c. Q.E.D.

@seismicisolation
@seismicisolation
60 2 Fundamentals

Lemma 2.15.3 Let . X be a Hilbert space, and suppose .a(·, ·) is a symmetric√bilinear

form that is continuous on . X and coercive on a subspace .V of . X . Then, .(V, a(·, ·))
is a Hilbert space. ♦

Proof An immediate consequence of the coercivity of .a(·, ·) is that if .v ∈ V and

.
√ v) = 0, then .v = 0. Hence, .a(·, ·) is an inner product on .V . Now, let .||v||a =
a(v,
a(v, v), and suppose that .(vn )n is a Cauchy sequence in .(V, || · ||a ). By coercivity,
.(vn )n is also Cauchy in .(X, || · || X ). Since . X is complete, there is .v ∈ X such that
.vn → v in the .|| · || X norm. Since . V is closed in . X , then .v ∈ V . Now, .||v − vn ||a ≤
√
c||v − vn || X since .a(·, ·) is continuous. Hence, .vn → v in the .|| · ||a norm, so .(V, || ·
||a ) is complete. Q.E.D.

2.15.2 Statement of the Stampacchia Theorem

Theorem 2.15.4 Let . X be a Hilbert space and .a(·, ·) be a bilinear form on . X ,

continuous and coercive and let . K be a non-empty, closed, convex set, of . X . Then,
for all .ϕ ∈ X , , there is .u ∈ K unique such that

.a(u, v − u) − ϕ(v − u) ≥ 0 (2.15.2)

for all .v ∈ K . Moreover, if .a(·, ·) is symmetric, then the element .u is characterized

by the property
⎧
⎨u ∈ K { ]
. 1 1 (2.15.3)
⎩ a(u, u) − ϕ(u) = inf a(v, v) − ϕ(v) .
2 v∈K 2

Proof Let .ϕ ∈ X , . Then, according to Riesz Fréchet’s representation theorem

(Theorem 2.14.1), there is . f ∈ X unique such that

ϕ(v) = ( f, v)
.

for all .v ∈ X . Let .u ∈ X , the mapping

a(u, ·) : X −→ R
.

v −→ a(u, v)

is continuous. Hence,
a(u, ·) ∈ X , .
.

@seismicisolation
@seismicisolation
2.15 Stampacchia Theorem 61

So, according to Riesz Fréchet’s representation theorem (Theorem 2.14.1), there is

Au ∈ X unique such that
.
.a(u, v) = (A u , v)

for all .v ∈ X . We, thus, define an operator

. A : X −→ X
u −→ Au .

. A is linear (exercise). The operator . A is continuous. In fact, we have

||Au || X = ||a(u, ·)|| X ,

.

|a(u, v)|
= sup
v∈X, v/=0 ||v||
c||u||||v||
≤ sup .
v∈X, v/=0 ||v||

It follows that
||Au || X ≤ c||u||.
.

Hence, . A is continuous. On the other hand,

(Au, u) ≥ α||u||2 .
. (2.15.4)

Let .ρ > 0. We pose

S : X −→ X
. ρ

v −→ Sρ (v) = PK (v + ρ( f − Av )).

Let’s show that there is.ρ ∗ > 0 such that for all.ρ ∈]0, ρ ∗ [, the operator. Sρ is a contrac-
tion. In fact, according to Proposition 2.12.10 (Eq. (2.12.24)) and, using inequality
(2.15.4), we have

||Sρ (v) − Sρ (w)||2 = ||PK (v + ρ( f − Av )) − PK (w + ρ( f − Aw ))||2

.

≤ ||(v − w) − ρ A(v − w)||2

≤ ((v − w) − ρ A(v − w), (v − w) − ρ A(v − w))
≤ ||v − w||2 − 2ρ(v − w, A(v − w)) + ρ 2 ||A(v − w)||2
≤ ||v − w||2 − 2ρα||v − w||2 + ρ 2 ||A(v − w)||2
≤ (1 − 2αρ + ρ 2 c2 )||v − w||2 . (2.15.5)

We pose
. h(ρ) = 1 − 2αρ + ρ 2 c2 .

@seismicisolation
@seismicisolation
62 2 Fundamentals

We have .h(ρ) = 1 if, and only if, .−2αρ + ρ 2 c2 = 0 if, and only if, .ρ = 2α
c2
. The
function .h(·) is differentiable on .[0, 2α
c2
] and

. h , (ρ) = −2α + 2ρc2 .

α
On the other hand, .h , (ρ) = 0 if, and only if, .ρ = c2
. The function .h(·) is decreasing
on .[0, cα2 ], increasing on .[ cα2 , 2α
c2
],

([ α ]) [ α2
]
. h 0, 2 = 1 − 2 , 1
c c

and ([ ]) [ ]
α 2α α2
. h , = 1 − 2 ,1 .
c2 c2 c

On the other hand, since

α||u||2 ≤ a(u, u) ≤ c||u||2 ,

.

we have (we can see also Lemma 2.15.2)

α≤c
.

and
(α) 2α 2 α2
. h =1− +
c2 c2 c2
α2
= 1 − 2 ≥ 0.
c

α 2α
0 c2 c2

h, − 0 +

1 1

h
α2
1−
c2

@seismicisolation
@seismicisolation
2.15 Stampacchia Theorem 63

α2
1−
c2

0 α 2α
c2 c2

We pose .ρ ∗ = 2α
c2
. Hence, for all .ρ ∈]0, ρ ∗ [, we have

0 ≤ h(ρ) < 1.
.

Equation (2.15.5) gives

/
||Sρ (v) − Sρ (w)|| ≤
. h(ρ)||v − w||.

So, . Sρ is contraction for all .ρ ∈]0, ρ ∗ [. Hence, by using Theorem 2.13.1, . Sρ admits
a fixed point and only one .u ∈ X . Thus, .u checks

u = Sρ (u) = PK (u + ρ( f − Au)).
.

which give .u ∈ K and

u = PK (u + ρ( f − Au)).
.

The closed convex projection theorem (Theorem 2.12.3) gives

{
u∈K
.
(u + ρ( f − Au) − u, v − u) ≤ 0 for all v ∈ K .

This is equivalent to saying

{
u∈K
.
ρ( f − Au, v − u) ≤ 0 for all v ∈ K .

This is equivalent to saying

{
u∈K
.
( f − Au, v − u) ≤ 0 for all v ∈ K .

@seismicisolation
@seismicisolation
64 2 Fundamentals

This is equivalent to saying

{
u∈K
.
( f, v − u) − (Au, v − u) ≤ 0 for all v ∈ K .

This is equivalent to saying

{
u∈K
.
ϕ(v − u) − a(u, v − u) ≤ 0 for all v ∈ K .

Uniqueness: We assume that there are .u 1 , .u 2 ∈ K such that

. ( f − Au 1 , v − u 1 ) ≤ 0 (2.15.6)

for all .v ∈ K and

( f − Au 2 , v − u 2 ) ≤ 0
. (2.15.7)

for all .v ∈ K . Taking .v = u 2 in (2.15.6) and .v = u 1 in (2.15.7), and by taking the

sum of these two equations, we obtain

(u 1 − u 2 , f − Au 2 − f + Au 1 ) ≤ 0.
.

Hence,
.a(u 1 − u 2 , u 1 − u 2 ) = (A(u 1 − u 2 ), u 1 − u 2 ) ≤ 0.

Furthermore, .a(·, ·) is coercive, so

. α||u 1 − u 2 ||2 ≤ a(u 1 − u 2 , u 1 − u 2 ) ≤ 0.

Hence, .u 1 = u 2 . Thus, the uniqueness.

Equivalence between (2.15.2) and (2.15.3).
Let .u ∈ X checking (2.15.2). Then,
{
u∈K
.
a(u, v − u) − ϕ(v − u) ≥ 0 for all v ∈ K .

Let’s pose
1
. J (v) = a(v, v) − ϕ(v).
2
Let .v ∈ K , we have

@seismicisolation
@seismicisolation
2.15 Stampacchia Theorem 65

. J (v) = J (u + (v − u))
1
= a(u + (v − u), u + (v − u)) − ϕ(u + (v − u))
2
1 1
= a(u, u) + a(u, v − u) + a(v − u, v − u) − ϕ(u) − ϕ(v − u))
2 2
1
= J (u) + [a(u, v − u) − ϕ(v − u)] + a(v − u, v − u).
2
We can deduce
. J (v) ≥ J (u)

for all .v ∈ K . This is equivalent to

. J (u) = inf J (v).

v∈K

Reciprocally, If .u checks (2.15.3), i.e., .u ∈ K and

. J (u) = inf J (v),

v∈K

then for all .v ∈ K and for all .t ∈]0, 1], we have

w = (1 − t)u + tv = u + t (v − u) ∈ K .
.

Hence,
. J (w) ≥ J (u)

for all .t ∈]0, 1]. By replacing .w by its expression in the last equation, we find

t2
. J (u) + t[a(u, v − u) − ϕ(v − u)] + a(v − u, v − u) ≥ J (u)
2
for all .t ∈]0, 1]. Thus,

t
a(u, v − u) − ϕ(v − u) ≥ − a(v − u, v − u).
.
2

Passing to the limit when .t → 0+ in the last inequality, we find (2.15.2). This com-
pletes the proof. Q.E.D.

@seismicisolation
@seismicisolation
66 2 Fundamentals

2.16 Equations and Minimization

2.16.1 Solution of Equations Related to Minimization

Let

. F : Rn −→ Rn
v −→ F(v).

Consider the problem: Find .u ∈ Rn such that

. F(u) = 0. (2.16.1)

The solution of Eq. (2.16.1) is related to minimization. In fact, if we, can, find a
function . J from .Rn into .R such that . F = ∇ J , and if we can show that . J has a
minimizer .u, i.e.,
. J (u) = min J (v),
n v∈R

then we know that . F(u) = ∇ J (u) = 0. Conversely, suppose we want to find a mini-
mizer .u of the functional . J . If we, can, find an element .u ∈ Rn such that .∇ J (u) = 0,
then the solution .u is a candidate for a minimizer of the functional . J . If, in addition,
the functional . J is convex, all such candidates are minimizers; otherwise, second-
order conditions may be used.

2.16.2 General Case

More general, this approach relates functionals of the form

{
. J (v) = L(x, v(x), ∇v(x)) d x (2.16.2)
Ω

to differential equations. In the case .n = 1, .Ω ⊂ R, (resp., .n > 1, .Ω ⊂ Rn ) these

are ordinary differential equations (resp., partial differential equations). The basic
problem of the calculus of variations is to minimize functionals of the form (2.16.2).
The Dirichlet principle yields equivalence, but not existence. Since the domain of
the functional . J is a subset of a function space, which has infinite dimension, then
the existence problem is hard in general. Therefore, functional analysis comes into
play. Using its methods to show the existence of a minimizer is called the direct
method of the calculus of variations. Such a minimizer solves the Euler equation is
then proved as above by setting to zero the derivatives of . J in all directions. Proving
existence of a minimizer of the functional . J in (2.16.2) by proving first the existence
of a solution to the Euler equation is called the indirect method. If the functional . J is

@seismicisolation
@seismicisolation
2.16 Equations and Minimization 67

not convex, however, solutions of the Euler equation for the functional . J in general
need to have additional properties in order to be minimizers of . J (as is the case in
finite dimensions, where second derivatives of . J play a role).

2.16.3 Quadratic Minimization Problem

Theorem 2.16.1 Let . X be a vector space, . K ⊂ X be convex, let .a : X × X −→ R

be a bilinear form and . F : X −→ R be a linear form. We consider the problem

. min J (v),
v∈K

where
1
. J (v) = a(v, v) − F(v).
2
If .a(·, ·) is symmetric and positive definite, then . J is strictly convex, and for .u ∈ K ,
we have

. J (u) = min J (v) if, and only if, a(u, v − u) ≥ F(v − u) for all v ∈ K . (2.16.3)
v∈K

Furthermore, there exists at most one .u ∈ K with this property. ♦

Proof Let .u, .h ∈ X be arbitrary with .h /= 0. Define

J
. u,h : R −→ R, Ju,h (λ) = J (u + λh).

Then,

1
J
. u,h (λ) = a(u + λh, u + λh) − F(u + λh)
2
λ2
= J (u) + λ(a(u, h) − F(h)) + a(h, h). (2.16.4)
2
Since .a(h, h) > 0, then the quadratic function . Ju,h is strictly convex, and so is . J .
Hence, . J has at most one minimizer.
“.⇒” Let .v ∈ K . By using both (2.16.3) and (2.16.4), it follows that

. J (v) − J (u) = J (u + v − u) − J (u)

1
= (a(u, v − u) − F(v − u)) + a(v − u, v − u) ≥ 0.
2
“.⇐” Let .v ∈ K . Since . K is convex, then .u + λ(v − u) ∈ K for .0 ≤ λ ≤ 1. Hence,

@seismicisolation
@seismicisolation
68 2 Fundamentals

λ2
0 ≤ Ju,v−u (λ) − Ju,v−u (0) = λ(a(u, v − u) − F(v − u)) +
. a(v − u, v − u).
2
(2.16.5)
Dividing by .λ and passing to the limit .λ → 0 yields

0 ≤ a(u, v − u) − F(v − u).

.

This completes the proof. Q.E.D.

The inequality system (one inequality for each .v ∈ K )

u ∈ K , a(u, v − u) ≥ F(v − u) for all v ∈ K ,

.

is said a variational inequality. As (2.16.5) shows, the variational inequality says that
the directional derivative of . J in a minimizer is non-negative for those directions
.v − u which point into the “admissible set” . K .

Corollary 2.16.2 In the situation of Theorem 2.16.1 assume that . K is an affine

subspace of . X , that is, . K = v0 + U with .v0 ∈ X , .U subspace of . X . Then,
. J (u) = min J (v) if, and only if, a(u, w) = F(w) for all w ∈ U. ♦
v∈K
Proof Let .u ∈ K and setting .U = {v − u : v ∈ K }. Then, .a(u, v − u) ≥ F(v − u)
for all .v ∈ K if, and only if, .a(u, w) ≥ F(w) for all .w ∈ U if, and only if, .a(u, w) =
F(w) for all .w ∈ U , (since .w ∈ U implies .−w ∈ U ). Q.E.D.
The system of equations

u ∈ v0 + U, a(u, v) = F(v) for all v ∈ U,

. (2.16.6)

is called the variational equation.

2.16.4 The Variational Equation, Solvability

Theorem 2.16.3 Let .(X, || · ||) be a Banach space, let .a : X × X −→ R be a sym-

metric, continuous and . X -elliptic bilinear form, let . F : X −→ R be linear and con-
tinuous. Then, there is a unique solution .u ∈ X of the variational equation

a(u, v) = F(v) for all v ∈ X.

. (2.16.7)

The solution .u of the variational Eq. (2.16.7) satisfies

1
||u|| ≤
. ||F||, (2.16.8)
α
where .α is the constant of coercivity of .a(·, ·). ♦

@seismicisolation
@seismicisolation
2.16 Equations and Minimization 69

Proof Let .v ∈ X . Then,

α||v||2 ≤ a(v, v) ≤ c||v||2 ,
.

where .α is the constant of coercivity of .a(·, ·) and .c is the constant of continuity of

a(·, ·). Since .α > 0, the quantities
.

/
. (u, v)a = a(u, v), ||v||a = a(v, v),

defines a scalar product whose associated norm .|| · ||a is equivalent to the original
norm .|| · ||. Therefore, .(X, (·, ·)a ) is a complete space and thus a Hilbert space (see
Lemma 2.15.3). Furthermore, . F : (X, (·, ·)a ) −→ R is a continuous function. By
using the Riesz Fréchet’s representation theorem (Theorem 2.14.1), there is a unique
.u ∈ X such that

.(u, v)a = F(v).

Inserting .u for .v in (2.16.7) yields

α||u||2 ≤ a(u, u)
.

= F(u)
≤ ||F||||u||.

This proves (2.16.8) and completes the proof. Q.E.D.

Theorem 2.16.3 state that, under the given assumptions, the solution of the variational
equation is a well-posed problem in the sense that its solution exists, is unique and
depends continuously upon the data (the right-hand side specified by . F). The latter
property has to interpret as follows: the solution operator . S : X , −→ X that maps
,
. F ∈ X (the dual space of . X ) to the solution .u of (2.16.7) is continuous. Indeed, it is
linear (this follows immediately from (2.16.7)), and, according to (2.16.8), we have

||F|| 1
||S(F)|| ≤
. , therefore ||S|| ≤ .
α α

2.16.5 The Variational Inequality, Solvability

In Theorem 2.16.3, we have assumed that the bilinear form is symmetric. It turns
out that this assumption is superfluous. We present this extension in an even more
general context, namely, for the variational inequality.
Theorem 2.16.4 Let . X be a Banach space, let . K ⊂ X be closed, convex and non-
empty. Let .a : X × X −→ R be a continuous, . X -elliptic, and bilinear form with

.a(v, v) ≥ α||v||2 for all v ∈ X, (2.16.9)

@seismicisolation
@seismicisolation
70 2 Fundamentals

let . F : V −→ R be linear and continuous. Then, the variational inequality

a(u, v − u) ≥ F(v − u) for all v ∈ K ,

. (2.16.10)

has a unique solution .u ∈ K . Moreover, if .-

u ∈ K is the solution of (2.16.10) corre-
- : X −→ R in place of . F, then
sponding to a linear and continuous . F

1 -
||u − -
. u || ≤ ||F − F||. (2.16.11)
α
♦
Proof We start the proof by showing (2.16.11). This also implies the uniqueness.
Let .u and .- - respectively. Then,
u be two solutions corresponding to . F, . F,

a(u, -
. u − u) ≥ F(-
u − u),

. u, u − -
a(- - −-
u ) ≥ F(u u ).

Adding these inequalities yields

. a(u − -
u, - - u − u).
u − u) ≥ (F − F)(-

On the other hand,

0 ≤ α||u − -
. u ||2 ≤ a(u − -
u, u − -u)
-
≤ ||F − F||||u − -
u ||.

This proves that (2.16.11) holds. To show existence, we first consider the particular
case, where .a(·, ·) is symmetric. By using Theorem 2.16.1, it suffices to prove that
the associated quadratic functional has a minimizer on the set . K . In order to show
this, we can remark that
1
. J (v) = a(v, v) − F(v)
2
has a minimizer on . K . In order to show this, we can also see that

1
. J (v) ≥ α||v||2 − ||F||||v||
2
(/ / )2
α 1 1
= ||v|| − ||F|| − ||F||2
2 2α 2α

holds for all .v ∈ X . Hence, . J is bounded from below. Define now

.d = inf J (v).
v∈K

@seismicisolation
@seismicisolation
2.16 Equations and Minimization 71

Let .(u n )n be a sequence in . K such that

1
d ≤ J (u n ) ≤ d +
. .
n

The sequence .(u n )n is a Cauchy sequence. In fact, since

( )
1 1
.a(u n − u m , u n − u m ) = 2a(u n , u n ) + 2a(u m , u m ) − 4a (u n + u m ), (u n + u m )
2 2
( )
1
= 4J (u n ) + 4J (u m ) − 8J (u n + u m ) ,
2

we have

α||u n − u m ||2 ≤ a(u n − u m , u n − u m )

.
( ) ( )
1 1
≤4 d+ +4 d + − 8d
n m
( )
1 1
≤4 + .
n m

Since . X is a Banach space, then there exists a .u ∈ X such that the sequence .(u n )n
converges to .u. Since . K is closed, then .u ∈ K . Since . J is continuous, then . J (u) = d.
This concludes the proof in the case, where .a(·, ·) is symmetric. Now, let .a(·, ·) be
arbitrary. We consider the following family of bilinear forms

a (u, v) = a0 (u, v) + tb(u, v), t ∈ [0, 1],

. t

where
1
a (u, v) =
. 0 (a(u, v) + a(v, u)),
2
and
1
b(u, v) =
. (a(u, v) − a(v, u)).
2
Then, .a0 (·, ·) is symmetric and .a1 (·, ·) = a(·, ·). The bilinear forms .at (·, ·) are con-
tinuous and . X -elliptic, and (2.16.9) holds for .at (·, ·) with the same constant .α as for
.at (·, ·), because .at (u, u) = a(u, u) holds for all .u ∈ X . Since .a0 (·, ·) is symmetric,
we claim that: if the variational inequality

a (u, v − u) ≥ G(v − u), v ∈ K ,

. τ

@seismicisolation
@seismicisolation
72 2 Fundamentals

has a unique solution .u ∈ K for every .G ∈ X , , then the same is true if we replace .τ
by .t with
α
.τ ≤ t ≤ τ + , (2.16.12)
2c

where .c is the constant of continuity of .a(·, ·). We prove this claim. Let .t be given,
satisfying (2.16.12). We only have to show the existence of a solution. Let.G : X −→
R be linear and continuous. We look for a solution .u t ∈ K satisfying

a (u t , v − u t ) ≥ G(v − u t ), ∀v ∈ K .
. t

For this purpose, we consider, for arbitrary given .w ∈ X , the variational inequality
with
.aτ (u, v − u) ≥ Fw (v − u), ∀v ∈ K , (2.16.13)

where
. Fw (v) = G(v) − (t − τ )b(w, v).

Let .T : X −→ R be the mapping which associates to each .w ∈ X the solution .u ∈ K

of (2.16.13). It is easy to see that .T is a contraction mapping. Indeed,

1
||T w1 − T w2 || ≤
. ||Fw1 − Fw2 ||
α
1
≤ |t − τ | sup |b(w1 − w2 , v)|
α ||v||=1
1
≤ |t − τ |c||w1 − w2 ||
α
1
≤ ||w1 − w2 ||.
2
In view of Banach’s fixed point theorem (Theorem 2.13.1),.T has a unique fixed point
u . Then, .u t ∈ K , and
. t

a (u t , v − u t ) = aτ (u t , v − u t ) + (t − τ )b(u t , v − u t )
. t

≥ Fu t (v − u t ) + (t − τ )b(u t , v − u t )
= G(v − u t )

for all .v ∈ K . This completes the proof. Q.E.D.

@seismicisolation
@seismicisolation
2.17 Lax-Milgram Theorem 73

2.17 Lax-Milgram Theorem

2.17.1 Lax-Milgram Theorem in Hilbert Space

The Lax-Milgram theorem, which is a generalization of the Riesz representation

theorem.
Theorem 2.17.1 Let . X be a Hilbert space and .a(·, ·) be a continuous bilinear form
and coercive on . X . Given .ϕ ∈ X , , there is .u ∈ X unique such that

. a(u, v) = ϕ(v) (2.17.1)

for all .v ∈ X. The .u satisfy the following condition

1
||u|| ≤
. ||ϕ||, (2.17.2)
α
where .α is the constant of coercivity of .a(·, ·). In addition, if .a(·, ·) is symmetric,
then the element .u is characterized by the property
{ ]
1 1
. a(u, u) − ϕ(u) = inf a(v, v) − ϕ(v) . ♦
2 v∈X 2

Proof Let .u ∈ X be an arbitrary fixed element. Then, the mapping .v −→ a(u, v) is

a linear functional on . X , which is continuous, i.e.,

.|a(u, v)| ≤ c||u||||v||

for all .u, .v ∈ X , and so is .ϕ, i.e.,

|ϕ(v)| ≤ M||v||
.

for all.v ∈ X . In view of the Riesz Fréchet’s representation theorem (Theorem 2.14.1),
there exist unique two elements .φu and .φϕ ∈ X such that

a(u, v) = (φu , v),

. ϕ(v) = (φϕ , v)

for all .v ∈ X . We recall that .w −→ φw is a continuous linear mapping from . X , into

X with operator norm .1. Thus, a solution .u ∈ X satisfies
.

0 = a(u, v) − ϕ(v) = (φu − φϕ , v)

.

for all .v ∈ X , which holds if, and only if, .φu = φϕ in . X . We now wish to solve this
equation using the Banach fixed point theorem (Theorem 2.13.1). For.δ > 0, consider
the mapping .Tδ : X −→ X , by

@seismicisolation
@seismicisolation
74 2 Fundamentals

T (v) = v − δ(φv − φϕ ).
. δ

If .Tδ is a contraction, then by Theorem 2.13.1, there exists a unique fixed point .u
such that .Tδ (u) = u and hence .φu − φϕ = 0. It remains to show that there exists a
constant .δ > 0 such that .Tδ is a contraction mapping, i.e., there exists .0 < L < 1
with
.||Tδ v1 − Tδ v2 || ≤ L||v1 − v2 ||.

Let .v1 , .v2 ∈ X be arbitrary. Then,

||Tδ v1 − Tδ v2 ||2 = ||v1 − v2 − δ(φv1 − φv2 )||2

.

= ||v1 − v2 − δφv1 −v2 ||2

= ||v1 − v2 ||2 − 2δ(v1 − v2 , φv1 −v2 ) + δ 2 (φv1 −v2 , φv1 −v2 ).

Thus,

||Tδ v1 − Tδ v2 ||2 = ||v1 − v2 ||2 − 2δ a(v1 − v2 , v1 − v2 ) + δ 2 a(v1 − v2 , φv1 −v2 )

.

≤ ||v1 − v2 ||2 − 2δα||v1 − v2 ||2 + δ 2 c||v1 − v2 ||||φv1 −v2 ||

≤ (1 − 2δα + δ 2 c)||v1 − v2 ||2 ,

where .α is the constant of coercivity of .a(·, ·). We can, thus, choose .0 < δ < 2 αc
such that . L 2 := (1 − 2δα + δ 2 c) < 1, and the Banach fixed point theorem (The-
orem 2.13.1) yields existence and uniqueness of the solution .u ∈ X . To show the
estimate (2.17.2), assume .u /= 0 (otherwise the inequality holds trivially). Note that
,
.ϕ is a bounded linear functional, hence .ϕ ∈ X . We can, then, apply the coercivity
of .a(·, ·) and divide by .||u|| /= 0 to obtain the following

a(u, u)
. α||u|| ≤
||u||
a(u, v)
≤ sup
v∈X ||v||
ϕ(v)
≤ sup
v∈X ||v||
= ||ϕ||.

Let
1
. J (u) = a(u, u) − ϕ(u).
2
For any .u, .v ∈ X and .t ∈ R, and due to the symmetry of .a(·, ·), we have

t2
. J (u + tv) = J (u) + t (a(u, v) − ϕ(v)) + a(v, v).
2

@seismicisolation
@seismicisolation
2.17 Lax-Milgram Theorem 75

Suppose that .u satisfies the equation .a(u, v) − ϕ(v) = 0 for all .v ∈ X . Then, for
t = 1, we infer that for all .v /= 0,
.

1
. J (u + v) = J (u) + a(v, v) > J (u)
2
holds. So, .u is the unique minimizer of the functional . J (·). Conversely, if .u is the
(unique) minimizer of the functional . J (·), then every directional derivative of the
functional . J (·) at .u must vanish, which implies

d
.0= J (u + tv)|t=0 = a(u, v) − ϕ(v)
dt
for all .v ∈ X . This completes the proof. Q.E.D.

Remark 2.17.2 .(i) If the bilinear form .a(·, ·) is not symmetric, then we can still
define the functional . J (·) in the same way as before and try to minimize it. It is clear
from the proof above that the minimizing element.u does not solve or does not find the
solution of the variational problem associated with .a(·, ·) but the variational problem
associated with the symmetric part of .a(·, ·). Of course, when the two variational
problems are translated into the partial differential equations PDEs, we get entirely
different equations.
(ii) If the bilinear form .a(·, ·) is symmetric, then the unique solution .u ∈ X of
.
Eq. (2.17.1) is nothing else than the unique representing of the linear form.l ∈ X , with
respect to the energetic inner product .(·, ·)a = a(·, ·). In this sense the Lax-Milgram
theorem (Theorem 2.17.1) is a special case of the Riesz Fréchet’s representation
theorem (Theorem 2.14.1). ♦

Example 2.17.1 Let. X = l 2 be the space of infinite sequences.(v1 , v2 , . . .), equipped

with the norm ( ∞ ) 21
E
.||v|| = vn2 .
n=1

The form
∞
E
a(u, v) :=
. 2−n u n vn
n=1

is positive and continuous but not coercive, and

∞
E
. L(v) := 2−n vn
n=1

@seismicisolation
@seismicisolation
76 2 Fundamentals

defines a continuous linear functional. However,

. J (v) = a(v, v) − L(v)

does not reach a minimum of.l 2 . Indeed, a necessary condition for aE minimal solution,
in this case, is that .vn = 1 for .n = 1, 2, . . . , and this contradicts . ∞
n=1 vn < ∞.
2

Proposition 2.17.3 The mapping . X , −→ X , .ϕ −→ u defined by the Lax-Milgram

theorem is linear and continuous. ♦

Proof Let .ϕ1 , .ϕ2 be two linear forms and .u 1 , .u 2 be the corresponding solutions to
the variational problem. For all .λ ∈ R, we have for all .v ∈ X ,

a(u 1 + λu 2 , v) = a(u 1 , v) + λa(u 2 , v)

.

= ϕ1 (v) + λϕ2 (v)

= (ϕ1 + λϕ2 )(v).

Thus, the linearity by uniqueness of the solution. Hence, by using the continuity, we
have

.α||u||2 ≤ a(u, u)
= ϕ(u)
≤ ||ϕ||||u||.

So,
1
||u|| ≤
. ||ϕ||.
α

Thus, we have the continuity, with continuity constant . α1 . Q.E.D.

Example 2.17.2 We provide .Rn by the following inner product

E
n
. (x, y) = xi yi .
i=1

Let. A be a square matrix of order.n symmetric and positive definite. Then, the bilinear
form

a : Rn × Rn −→ R
.

(u, v) −→ a(u, v) = (Au, v)

is continuous and coercive.

@seismicisolation
@seismicisolation
2.17 Lax-Milgram Theorem 77

Continuity of a(·, ·): Let .u, .v ∈ Rn . Then,

.

|a(u, v)| = |(Au, v)|

.

≤ ||Au||2 ||v||2
≤ ||A||2 ||u||2 ||v||2

with ( ) 21
E
n
||u||2 =
. xi2
i=1

and
||Av||2
||A||2 =
. sup
v∈Rn , v/=0 ||v||2

= sup ||Av||2 .
||v||2 =1

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Since the matrix . A is symmetric, then there is . O ∈ On (R), an
.
orthogonal matrix (. O t O =t O O = I ) such that

.
t
O AO = D

is a diagonal matrix, with .t O is the transpose of the matrix . O. Let .u ∈ Rn . Then,

a(u, u) = (Au, u)
.

= (O D t Ou, u)
= (D t Ou,t Ou)
E
n
= λi (t Ou)i2
i=1
( n )
E
≥ min λi ( t
Ou)i2
1≤i≤n
i=1
≥ min λi (t Ou,t Ou)
1≤i≤n

≥ min λi ||u||22 ,
1≤i≤n

with the .λi , .i = 1, . . . , n, are the eigenvalues of the matrix . A. Since . A is positive
definite, all eigenvalues .λi are all strictly positive. Hence,

@seismicisolation
@seismicisolation
78 2 Fundamentals

. min λi > 0.
1≤i≤n

Thus, .a(·, ·) is coercive.

Given .ϕ ∈ (Rn ), and, using the Lax-Milgram theorem (Theorem 2.17.1), there is
.u ∈ R unique such that
n

.a(u, v) = ϕ(v)

for all .v ∈ Rn . This is equivalent to saying

(Au, v) = ϕ(v)
.

for all .v ∈ Rn . On the other hand, .ϕ ∈ (Rn ), , hence according to Riesz Fréchet’s
representation theorem (Theorem 2.14.1), there is . f ∈ Rn unique such that

ϕ(v) = ( f, v)
.

for all .v ∈ Rn . Thus,

. (Au, v) = ( f, v)

for all .v ∈ Rn . This is equivalent to saying

.(Au − f, v) = 0

for all .v ∈ Rn . This is equivalent to saying

. Au = f.

Since . A is symmetric, it implies that the bilinear form .a(·, ·) is symmetric. Thus, .u
checks { ]
1 1
. (Au, u) − ϕ(u) = inf (Av, v) − ϕ(v) .
2 v∈Rn 2

This .u checks . Au = f.

2.17.2 Lax-Milgram Theorem in Banach Space

Corollary 2.17.4 Let . X be a Banach space, let .a : X × X −→ R be a continuous,

X -elliptic, and bilinear form, let . F : X −→ R be linear and continuous. Then, the
.
variational equation
.a(u, v) = F(v) ∀v ∈ X

has a unique solution .u ∈ X . ♦

@seismicisolation
@seismicisolation
2.17 Lax-Milgram Theorem 79

Proof For the proof we apply Theorem 2.16.3 with . K = X . As in Corollary 2.16.2,
the assertion follows from the equivalences .a(u, v − u) ≥ F(v − u) for all .v ∈ X
if, and only if, .a(u, v) ≥ F(v) for all .v ∈ X if, and only if, .a(u, v) = F(v) for all
.v ∈ X . Q.E.D.

2.17.3 Generalization of the Lax-Milgram Theorem

Let . X and .Y be two Banach spaces. For a bounded linear functional . A between . X
and.Y , the range. A(X ) of. A is closed in.Y if, and only if,. A(X ) = (K er (A∗ ))0 , where
∗ , ,
. A : Y −→ X is the adjoint of . A, . K er (A) := {x ∈ X : Ax = 0} is the null space

of an operator . A : X −→ Y , and for .V ⊂ X ,

. V 0 := {x ∈ X , : (x, v) X , ,X = 0 for all v ∈ V }

is the polar of .V . The following generalization of the Lax-Milgram theorem gives

sufficient (and, as can be shown, necessary) conditions for the well-posedness of
(2.17.1).

Theorem 2.17.5 (Banach-Necas-Babuska) Let . X and .Y be Banach spaces and .Y

be reflexive. Let .a : X × Y −→ R be a bilinear form and .ϕ : Y −→ R be a linear
functional satisfying the following assumptions:
(i) Inf-sup-condition: There exists a .c1 > 0 such that
.

a(u, v)
. inf sup ≥ c1 .
u∈X v∈Y ||u||||v||

(ii) Continuity: There exist .c2 , .c3 such that

.

.|a(u, v)| ≤ c2 ||u||||v||,

.|ϕ(v)| ≤ c3 ||v||

for all .u ∈ X , .v ∈ Y .
(iii) Injectivity: For any .v ∈ Y ,
.

.a(u, v) = 0 for all u ∈ X implies v = 0.

Then, there exists a one and only one solution .u ∈ X to (2.17.1) satisfying
1
. ||u|| ≤ ||ϕ||. ♦
c1

@seismicisolation
@seismicisolation
80 2 Fundamentals

Proof The proof is essentially an application of the closed range theorem. We apply
this theorem to the following operator . A : X −→ Y , defined by

(Au, v)Y , ,Y = a(u, v) for all v ∈ Y

.

to show that . A is an isomorphism (i.e., . A is bijective and . A and . A−1 are continuous),
which is equivalent to the claim since (2.17.1) can be expressed as . Au = f . Con-
tinuity of . A easily follows from continuity of .a(·, ·) and the definition of the norm
on .Y , . We next show injectivity of . A. Let .u 1 , .u 2 ∈ X be given with . Au 1 = Au 2 . By
definition, this implies .a(u 1 , v) = a(u 2 , v). Hence, .a(u 1 − u 2 , v) = 0 for all .v ∈ Y .
Hence, the inf-sup-condition implies that

a(u 1 − u 2 , v)
c ||u 1 − u 2 || ≤ sup
. 1 = 0.
v∈Y ||v||

Therefore .u 1 = u 2 . Due to the injectivity of . A, for any .v∗ ∈ A(X ) ⊂ Y , , we have a

unique .u =: A−1 v∗ ∈ X , and the inf-sup-condition yields

a(u, v)
c ||u|| ≤ sup
. 1
v∈Y ||v||
(Au, v)Y , ,Y
= sup
v∈Y ||v||
(v∗ , v)Y , ,Y
= sup
v∈Y ||v||
= ||v∗ ||. (2.17.3)

Therefore, . A−1 is continuous on . A(X ). We next prove that . A(X ) is closed. Let
{vn∗ }n∈N ⊂ A(X ) ⊂ Y , be a sequence converging to a.v∗ ∈ Y , , i.e., there exists.u n ∈ X
.
such that .vn∗ = Au n . The .{vn∗ }n form a Cauchy sequence. From (2.17.3), we infer for
all .n, .m ∈ N that

1
||u n − u m || ≤
. ||A(u n − u m )||
c1
1
= ||vn∗ − vm∗ ||,
c1

which implies that.{u n }n∈N is a Cauchy sequence as well and thus converges to.u ∈ X .
The continuity of . A, then yields

@seismicisolation
@seismicisolation
2.18 Reminders on Distributions 81

.v∗ = lim vn∗

n→∞
= lim Au n
n→∞
= Au.

Thus, we obtain .v∗ ∈ A(X ). We can, therefore, apply the closed range theorem. By
the reflexivity of .Y , we have . A∗ : Y −→ X , and

. K er (A∗ ) = {v ∈ Y : A∗ v = 0}
= {v ∈ Y : (A∗ v, u) X , ,X = 0 for all u ∈ X }
= {v ∈ Y : (Au, v)Y , ,Y = 0 for all u ∈ X }
= {v ∈ Y : a(u, v) = 0 for all u ∈ X }
= {0}

due to the injectivity condition .(iii). Hence, the closed range theorem and the reflex-
ivity of .Y yields

. A(X ) = ({0})0 = {v∗ ∈ Y , : (v∗ , 0)Y , ,Y = 0} = Y , ,

and, therefore surjectivity of . A. Thus, . A is an isomorphism and the claimed estimate

follows from (2.17.3) applied to .v∗ = ϕ ∈ Y , . Q.E.D.

Remark 2.17.6 The term “injectivity condition” is due to the fact that it implies
injectivity of the adjoint operator . A∗ and hence (due to the closed range of . A)
surjectivity of . A. Note that in the symmetric case . X = Y , coercivity of .a(·, ·) implies
both the inf-sup-condition and (via contraposition) the injectivity condition, and we
recover the Lax-Milgram theorem. ♦

2.18 Reminders on Distributions

Distributions are to functions what irrational numbers are to rational numbers. Dis-
tributions are, in fact, a generalization of the notion of function. We will present it
briefly, limiting ourselves to essential concepts such as the derivation of a distribu-
tion. We refer the reader to Girault and Raviart [8], for a simple and modern treatment
of distribution theory. The more informed reader can consult the book of Schwartz
[9], for a more complete and more traditional treatment.

@seismicisolation
@seismicisolation
82 2 Fundamentals

2.18.1 Space .D(Ω)

2.18.1.1 Definition

Definition 2.18.1 Let .Ω be an open of .Rn . All the functions defined on .Ω, that we
consider, are with real values. We denote by .D(Ω) the vector space of indefinitely
differentiable functions on .Ω with compact support in .Ω. ♦

D(Ω) is not reduced to null function. In fact, let .a ∈ Ω and let .r > 0 be such that
.

. B(a, r ) ⊂ Ω,

with . B(a, r ) designates the open ball of center .a and of radius .r . Then, the function
ϕ defined by
.

{ 1
e ||x−a||2 −r 2 if ||x − a|| < r
.ϕ(x) :=
0 if not

belongs to .D(Ω) with

( n ) 21
E
.||x − a|| = (xi − ai )2 ,
i=1

. x = (x1 , . . . , xn ) and .a = (a1 , . . . , an ). Note the interesting result.

Theorem 2.18.2 .D(Ω) is dense in . L 2 (Ω, R). ♦

2.18.1.2 Convergent Sequences in .D(Ω)

We define, in .D(Ω), the pseudo-topological, i.e., we define convergent sequence on

D(Ω). For this, if .ϕ ∈ D(Ω) and if .α = (α1 , . . . , αn ) ∈ Nn , then we define .∂ α ϕ by
.

( )α1 ( )αn
α ∂ ∂
∂ ϕ=
. ··· ϕ
∂ x1 ∂ xn
∂ |α|
= ϕ
∂ x1α1 · · · ∂ xnαn

with .|α| = α1 + · · · + αn . We will say that the sequence .(ϕm )m converges to .ϕ on

D(Ω) if
.

(i) for all .m ∈ N, the support of .ϕm − ϕ stays in a fixed compact . K of .Ω.
.
(ii) for all .α ∈ Nn , .∂ α ϕm converges uniformly to .∂ α ϕ.
.

@seismicisolation
@seismicisolation
2.18 Reminders on Distributions 83

2.18.2 Space .D , (Ω)

2.18.2.1 Definition

We, then, introduce the space .D , (Ω) of continuous linear forms on .D(Ω), in the
following direction:
Let .T : D(Ω) −→ R be linear. .T is continuous if for all convergent sequence .(ϕm )m
to .ϕ on .D(Ω)
. lim (T, ϕm ) = (T, ϕ).
m→∞

D , (Ω) is called the space of distributions on .Ω (Schwarz distribution space).

.

Example 2.18.1 Let .a ∈ Ω. The mapping

δ : D(Ω) −→ R
. a

ϕ −→ (δa , ϕ) = ϕ(a)

is a distribution called mass of Dirac at point .a ∈ Ω.

2.18.2.2 Convergent Sequences in .D , (Ω)

We provide .D , (Ω) of pseudo-topological, i.e., we define convergent sequences in

, ,
.D (Ω). Let .(Tm )m be a sequence of .D (Ω). We will say that the sequence .(Tm )m
,
converges to .T on .D (Ω), if for all .ϕ ∈ D(Ω), we have

. lim (Tm , ϕ) = (T, ϕ).

m→∞

2.18.3 Derivation Within the Meaning of Distributions

Definition 2.18.3 Let .T ∈ D , (Ω). The mapping

. T , : D(Ω) −→ R

defined by: for all .ϕ ∈ D(Ω)

.(T , , ϕ) = −(T, ϕ , )

defined a distribution on .Ω, called derivative of .T . ♦

@seismicisolation
@seismicisolation
84 2 Fundamentals

Proposition 2.18.4 Any distribution is indefinitely differentiable within the meaning

of distributions and, we have for all .α = (α1 , . . . , αn ) ∈ Nn and for all .ϕ ∈ D(Ω),
we have
α |α| α
.(∂ T, ϕ) = (−1) (T, ∂ ϕ).

with .|α| = α1 + · · · + αn . ♦

Proposition 2.18.5 Let .α ∈ Nn be a multi-index. The mapping

.∂ α : D , (Ω) −→ D , (Ω)

is .‘‘continuous”, .i.e., if .(Tm )m is a sequence of .D , (Ω) converging to .T in .D , (Ω),

then .∂ α Tm tends to .∂ α T in .D , (Ω). ♦

Proof For all .ϕ ∈ D(Ω), we have .(Tm , ϕ) converges to .(T, ϕ) when .m tends to .∞.
However,
α |α| α
.(∂ Tm , ϕ) = (−1) (Tm , ∂ ϕ)

and passing to the limit when .m → ∞, we deduce

. (∂ α T, ϕ) = (−1)|α| (T, ∂ α ϕ). Q.E.D.

2.18.4 Regular Distribution

Let .Ω be an open of .Rn and . f ∈ L 2 (Ω, R). Consider the mapping .T f by

. T f : D(Ω) −→ R
{
ϕ −→ (T f , ϕ) = f (x)ϕ(x) d x. (2.18.1)
Ω

We have .T f is linear. Let us show that .T f is continuous. In fact, let .(ϕm )m be a

sequence of .D(Ω) converging to .ϕ in .D(Ω). Hence, the support of .ϕm − ϕ stays in
a fixed compact . K of .Ω. Then, for all .ε > 0, there is . N > 0 such that for .m > N ,
we have ε
.||ϕm − ϕ||∞ < 1 .
|| f ||0,Ω mes(K ) 2

Let .m > N , we have

@seismicisolation
@seismicisolation
2.18 Reminders on Distributions 85
{
|(T f , ϕm ) − (T f , ϕ)| ≤
. | f (x)||ϕm (x) − ϕ(x)| d x
Ω
{
≤ sup |ϕm (x) − ϕ(x)| | f (x)| d x
x∈K K
1
≤ ||ϕm − ϕ||∞ || f ||0,Ω mes(K ) 2
< ε,

which proves .T f is continuous. Thus, .T f ∈ D , (Ω). So, we have the following:

Theorem 2.18.6 Let .Ω be an open of .Rn and . f ∈ L 2 (Ω, R). Consider the mapping
. T f , given in (2.18.1). Then, . T f is a distribution called regular distribution. ♦

Consider the mapping

.ψ : L 2 (Ω, R) −→ D , (Ω)
f −→ ψ( f ) = T f . (2.18.2)

ψ is linear. We show that .ψ is injective. In fact, consider the function . f ∈ L 2 (Ω, R)

.
such that .ψ( f ) = 0. Hence .T f = 0. Thus, for all .ϕ ∈ D(Ω)
{
. f (x)ϕ(x) d x = 0. (2.18.3)
Ω

However, according to the Theorem 2.18.2, .D(Ω) is dense in . L 2 (Ω, R), we will
show that Eq. (2.18.3) remains valid for all .ϕ ∈ L 2 (Ω, R). In fact, let .ϕ ∈ L 2 (Ω, R)
according to the Theorem 2.18.2, there is a sequence .(ϕm )m of .D(Ω) such that .(ϕm )m
converges to .ϕ in . L 2 (Ω, R). Using (2.18.3), we have
{
. f (x)ϕm (x) d x = 0.
Ω

Hence,
|{ { |
| |
| f (x)ϕm (x) d x − f (x)ϕ(x) d x || ≤ || f ||0,Ω ||ϕm − ϕ||0,Ω .
.
|
Ω Ω

Passing to the limit when .m → ∞ in the last equation, we obtain

{
. f (x)ϕ(x) d x = 0 (2.18.4)
Ω

for all .ϕ ∈ L 2 (Ω, R). In particular, for .ϕ = f in (2.18.4), we find

@seismicisolation
@seismicisolation
86 2 Fundamentals
{
. f 2 (x) d x = 0.
Ω

Thus, . f = 0 in . L 2 (Ω, R). Thus, .ψ is injective. So, we have the following:

Proposition 2.18.7 The mapping .ψ, given in (2.18.2), is linear injective. ♦

Remark 2.18.8 If . f ∈ L 2 (Ω, R), then we may identify . f with .T f . Which amounts
to identifying . L 2 (Ω, R) with .ψ(L 2 (Ω, R)) ⊂ D , (Ω). We will systematically do
this identification in all the following and, we write

. L 2 (Ω, R) ⊂ D , (Ω). (2.18.5)

Let . j : L 2 (Ω, R) −→ D , (Ω) be the canonical injection from. L 2 (Ω, R) into.D , (Ω).

Proposition 2.18.9 The injection

. j : L 2 (Ω, R) −→ D , (Ω)

is .‘‘continuous”, .i.e., if .( f m )m is a sequence of . L 2 (Ω, R) which tends to a function

,
. f ∈ L (Ω, R), then . j ( f m ) = T f m converges to . j ( f ) = T f in .D (Ω). ♦
2

Proof Let .ϕ ∈ D(Ω). Then,

|(T fm , ϕ) − (T f , ϕ)| ≤ || f m − f ||0,Ω ||ϕ||0,Ω .

.

Thus,
. lim T fm = T f . Q.E.D.
m→∞

Proposition 2.18.10 The distribution of Dirac .δa is not regular. ♦

Proof We must show that it does not exist of locally integrable function . f (x) such
that
{
.(δa , ϕ) = f (x)ϕ(x) d x
Ω
= ϕ(a)

for all .ϕ ∈ D(Ω). If such a function . f (x) existed, by setting

. ϕ(x) = ϕn (x),

@seismicisolation
@seismicisolation
2.18 Reminders on Distributions 87

where .ϕn (x) is the function defined by

⎧
⎨
1
n2
||x−a||2 − 12
ϕ (x) :=
. n e n if ||x − a|| < 1
⎩ n
0 if not,

we would have {
. f (x)ϕ(x) d x = ϕn (a) = e−1 .
Ω

If . B(a, n) designates the open ball with center .a and radius . n1 , we would also have

.e−1 = |ϕn (a)|

{
≤ | f (x)||ϕn (x)| d x
B(a,n)
{
≤ e−1 | f (x)| d x
B(a,n)

since the functions .ϕn (x) are all bounded by .e−1 . The last inequality constitutes a
contradiction since the term of right tends to .0 when .n tends to infinity. Q.E.D.

Let . f ∈ C 1 (Ω, R). Then, . ∂∂xfi exists and, we have

∂f
. ∈ C 0 (Ω, R).
∂ xi
∂T
Let’s compare .T ∂ f and . ∂ xif . In fact,
∂ xi

. T f : D(Ω) −→ R
{
ϕ −→ (T f , ϕ) = f (x)ϕ(x) d x
Ω

and

. T ∂ f : D(Ω) −→ R
∂ xi
/ \ { ∂f
ϕ −→ T ∂ f , ϕ = (x)ϕ(x) d x.
∂ xi
Ω ∂ xi

@seismicisolation
@seismicisolation
88 2 Fundamentals

Let .ϕ ∈ D(Ω). We have

/ \ / \
∂Tf ∂ϕ
. , ϕ = − Tf ,
∂ xi ∂ xi
{
∂ϕ
=− f (x) (x) d x
Ω ∂ xi
{
∂f
= (x)ϕ(x) d x.
Ω ∂ xi

Thus,
∂Tf
. T ∂f = .
∂ xi ∂ xi
Example 2.18.2 .(a) For .Ω = R, we consider the Heaviside function defined by
{
1 if x > 0
. H (x) :=
0 if x < 0.

Let .TH be a regular distribution defined by

T : D(Ω) −→ R
. H
{
ϕ −→ (TH , ϕ) = H (x)ϕ(x) d x.
R

Let .ϕ ∈ D(Ω). We have

/ \ / \
∂ TH ∂ϕ
. , ϕ = − TH ,
∂x ∂x
{
= − H (x)ϕ , (x) d x
{R∞
=− ϕ , (x) d x
0
= ϕ(0)
= (δ, ϕ),

with .δ is the mass of Dirac at point .0. Thus,

. H , = δ.

(b) For .Ω = R, we consider a continuous and differentiable function . f (x) except

.
at points .xi , .i = 1, 2, . . . , k with .x1 < x2 < · · · < xk , where it has discontinuities of
the first kind (i.e., the left and right limits of . f (x) exist and are finite).

@seismicisolation
@seismicisolation
2.18 Reminders on Distributions 89

graph ( f )

x1 x2 xk

Let .T f be the regular distribution associated with . f . Then, for all .ϕ ∈ D(Ω), we
have
/ \ / \
∂T f ∂ϕ
. ,ϕ = − Tf ,
∂x ∂x
{
=− f (x)ϕ , (x) d x
R
{ x1 { x2 { ∞
=− f (x)ϕ , (x) d x − f (x)ϕ , (x) d x − · · · − f (x)ϕ , (x) d x
−∞ x1 xk
{ x1 { x2
= f , (x)ϕ(x) d x − [ f (x)ϕ(x)]−∞
x1
+ f , (x)ϕ(x) d x
−∞ x1
{ ∞
− [ f (x)ϕ(x)]xx21 + · · · + f , (x)ϕ(x) d x − [ f (x)ϕ(x)]∞
xk
xk
{ x1 { x2
= f , (x)ϕ(x) d x − f (x1 − 0)ϕ(x1 ) + f , (x)ϕ(x) d x
−∞ x1
{ ∞
− f (x2 − 0)ϕ(x2 ) + f (x1 + 0)ϕ(x1 ) + · · · + f , (x)ϕ(x) d x
xk
+ f (xk + 0)ϕ(xk )
{ E
k
= f , (x)ϕ(x) d x + ϕ(x j )( f (x j + 0) − f (x j − 0))
R j=1
/ \
E
k
= T∂f + ( f (x j + 0) − f (x j − 0))δx j , ϕ ,
∂x
j=1

@seismicisolation
@seismicisolation
90 2 Fundamentals

with .δx j is the mass of Dirac at point .x j . Thus,

( ), E
k
. Tf = Tf, + s j δx j ,
j=1

with .s j = f (x j + 0) − f (x j − 0).

References

1. F. Brezzi, On the existence, uniqueness and approximation of saddle-point problems arising

from Lagrangian multipliers. Rev. FranàSaise Automat. Informat. Recherche Opérationnelle
Sér. Rouge 8(R-2), 129–151 (1974)
2. A. Jeribi, Spectral Theory and Applications of Linear Operators and Block Operator Matrices
(Springer, New-York, 2015)
3. A. Jeribi, Denseness, Bases and Frames in Banach Spaces and Applications (de Gruyter, Berlin,
2018)
4. A. Jeribi, Linear Operators and Their Essential Pseudospectra (CRC Press, Boca Raton, 2018)
5. A. Jeribi, Perturbation Theory for Linear Operators: Denseness and Bases with Applications
(Springer, Singapore, 2021). ISBN 978-981-16-2527-5
6. A. Jeribi, M.A. Hammami, A. Masmoudi (ed.), Applied mathematics in Tunisia, in International
Conference on Advances in Applied Mathematics (ICAAM), Hammamet, Tunisia, December
16–19 (2013). Springer Proceedings in Mathematics and Statistics 131 (Springer, Cham, 2015).
ISBN 978-3-319-18040-3/hbk; 978-3-319-18041-0/ebook). xix, 397 p
7. C. Wyss, Riesz bases for . p-subordinate perturbations of normal operator. J. Func. Anal. 258,
208–240 (2010)
8. V. Girault, P.-A. Raviart, Finite element methods for Navier-Stokes equations. Springer Series
in Computational Mathematics, vol. 5 (Springer, Berlin, 1986). Theory and algorithms
9. L. Schwartz, Théorie des distributions (Dunod, Paris, 1965)

@seismicisolation
@seismicisolation
Chapter 3
Variational Formulation of Boundary
Problems

In this chapter, we will focus on the notions of Sobolev spaces of order .1 and of order
.m ≥ 1. Sobolev spaces are functional spaces. More precisely, a Sobolev space is a
vector space of functions provided with the norm obtained by the combination of the
2
. L norm of the function itself and of its derivatives up to a certain order. Derivatives
are understood in a weak sense, within the meaning of distributions in order to make
the space complete. After that, we introduce the notion of problem with elliptical
limits. A physical or mechanical or biological problem is generally described by
the data of differential equations or more certainly by the data of partial differential
equations. Such a formulation is called a strong formulation of the problem. We will
see that it is possible to express these differential equations or partial differential
equations in a “less restrictive” way for the sought solutions. Such a formulation is
called a weak formulation, and its solutions are called a weak solution. Obviously, a
strong solution of the original problem is also a solution of the weak formulation and
vice versa under some conditions of regularity of the solution on open bounded of .Rn
with piecewise smooth boundary. We collect some necessary results from functional
analysis and the weak theory of (elliptic) partial differential equations. This method
is based on the variational formulation of these problems and thus appears as a
particular Galerkin method. We will focus on this aspect in this chapter.

3.1 Sobolev Space of Order .1, . H 1 (Ω)

Let .Ω be an open set of .Rn . Consider a function .v of . L 2 (Ω, R), it is identified with
a distribution of .D , (Ω) still noted .v and we can still define its derivatives

∂v
. , 1 ≤ i ≤ n,
∂ xi

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 91
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_3
@seismicisolation
@seismicisolation
92 3 Variational Formulation of Boundary Problems

as a distribution of .D , (Ω). In general,

∂v
. ∈
/ L 2 (Ω, R).
∂ xi

3.1.1 Definition

Definition 3.1.1 We call Sobolev space of order .1 on .Ω, the space

{ }
∂v
. H 1 (Ω) = v ∈ L 2 (Ω, R) such that ∈ L 2 (Ω, R) for 1 ≤ i ≤ n .
∂ xi
♦
We provide . H 1 (Ω) of the following inner product:
{ ( E n
)
∂u ∂v
(u, v)1,Ω
. = uv + dx
Ω i=1
∂ xi ∂ xi
{ En {
∂u ∂v
= uv d x + dx
Ω i=1 Ω
∂ xi ∂ xi
{ {
= uv d x + (grad u, grad v)d x
{Ω {Ω
= uv d x + ∇u · ∇v d x,
Ω Ω

where .∇u · ∇v is the Euclidean scalar product in .Rn of .∇u and .∇v. We notice
1
. ||v||1,Ω = (v, v)1,Ω
2

the corresponding norm.

Example 3.1.1 1. The function
{ 3
x4 if 0 < x < 1
. f (x) := 3
(−x) 4 if − 1 < x ≤ 0

is in . H 1 (] − 1, 1[) .(but it is not derivable in .x = 0 .).

2. If .Ω =]a, b[ is a bounded open of .R, then any continuous and continuously
differentiable function by pieces on .[a, b] is a function of . H 1 (]a, b[) because

∂Tf
. = T∂f .
∂x ∂x

@seismicisolation
@seismicisolation
3.1 Sobolev Space of Order 1, H 1 (Ω) 93

3. (Hadamard [1]) Let .r and .ϕ be the polar coordinates in the unit disk . B(0, 1) =
{x ∈ R2 such that ||x|| < 1}. The function
∞
E
u(r, ϕ) :=
. k −2 r k! sin(k! ϕ)
k=1

is harmonic in . B(0, 1). If we identify .R2 with .C, then

∞
E
u(z) = Im
. k −2 z k! .
k=1

This shows that {

. |∇u|2 d x = ∞,
B(0,1)

and thus .u ∈
/ H 1 (B(0, 1)). There does not exist any function in . H 1 (B(0, 1)) with the
same boundary value as .u, since for a given boundary value, the harmonic function
is always the one with the smallest value of the . H 1 -semi-norm.

3.1.2 Separable Hilbert Space

Theorem 3.1.2 . H 1 (Ω) is a separable Hilbert space for the norm .|| · ||1,Ω . ♦
Proof Let .(vm )m be a Cauchy sequence in . H (Ω), so .(vm )m is a Cauchy sequence
1

in . L 2 (Ω, R) and ( )
∂vm
.
∂ xi m

is a Cauchy sequence in . L 2 (Ω, R) .1 ≤ i ≤ n. Since . L 2 (Ω, R) is complete, .(vm )m

converges to .v in . L 2 (Ω, R) and ( )
∂vm
.
∂ xi m

converges to .vi in . L 2 (Ω, R) .1 ≤ i ≤ n. Let us show that

∂v
v =
. i
∂ xi

within the meaning of .D , (Ω). In fact, since the canonical injection from . L 2 (Ω, R)
into .D , (Ω) is .‘‘continuous” .(see Proposition 2.18.9), we have .(vm )m converges to .v
in .D , (Ω) and ( )
∂vm
.
∂ xi m

@seismicisolation
@seismicisolation
94 3 Variational Formulation of Boundary Problems

converges to .vi in .D , (Ω) .1 ≤ i ≤ n. Since the mapping

∂
.
∂ xi
( )
is continuous from .D , (Ω) into .D , (Ω) (see Proposition 2.18.5), we have . ∂v m
∂ xi m
converges to . ∂∂vxi in .D , (Ω). Thus,
∂v
v =
. i
∂ xi

in .D , (Ω). We deduce that

∂v
. ∈ L 2 (Ω, R)
∂ xi

and that .(vm )m converges to .v in . H 1 (Ω). Thus, . H 1 (Ω) is a Hilbert space. We still
have to show that the space . H 1 (Ω) is separable. For this, consider

. I : H 1 (Ω) −→(L 2 (Ω, R))n+1

( )
∂v ∂v
v −→ v, ,..., .
∂ x1 ∂ xn

We provide .(L 2 (Ω, R))n+1 with the following inner product:

E
n+1
(u, v) =
. (u i , vi )0,Ω .
i=1

We notice for .u ∈ (L 2 (Ω, R))n+1 ,

( n+1 ) 21
E
.||u||(L 2 (Ω,R))n+1 = ||u i ||20,Ω
i=1

the corresponding norm. The product space .(L 2 (Ω, R))n+1 provided with this inner
product .|| · || is a Hilbert space. Let .v ∈ H 1 (Ω). We can easily verify that

||I (v)||(L 2 (Ω,R))n+1 = ||v||1,Ω .

.

Hence, . I is an isometry. Firstly, according to the Lemma 2.5.4, . I (H 1 (Ω)) is a closed

subspace of .(L 2 (Ω, R))n+1 . Secondly, according to the Lemma 2.12.12, . I (H 1 (Ω))
is separable. Thus,

@seismicisolation
@seismicisolation
3.2 Sobolev Space, H01 (Ω) 95

. -
I : H 1 (Ω) −→I (H 1 (Ω))
v −→-
I (v) = I (v)

is an isometric isomorphism. Hence,

. H 1 (Ω) ∼ I (H 1 (Ω)).

Thus, . H 1 (Ω) is separable. This finishes the proof. Q.E.D.

Lemma 3.1.3 We assume that the open set .Ω is bounded. Then, there exists a con-
stant .C depending on .Ω such that for all .u ∈ H 1 (Ω), we have

( )1
||u||0,∂Ω ≤ C ||u||0,Ω ||u||1,Ω 2 .
.

3.2 Sobolev Space, . H01 (Ω)

3.2.1 Definition

Definition 3.2.1 We denote by . H01 (Ω) the closure of .D(Ω) in . H 1 (Ω), i.e.,

H 1 (Ω)
. H01 (Ω) = D(Ω) .

3.2.2 Poincaré Inequality

Theorem 3.2.2 If .Ω is bounded, then there is a constant .C = C(Ω) > 0 such that

. ||v||0,Ω ≤ C(Ω)|v|1,Ω

for all .v ∈ H01 (Ω) with

( n { | | ) 21
E | ∂v |2
|v|1,Ω = | |
.
|∂x | dx .
i=1 Ω i

Proof We take again here the demonstration of B. Lucquin [2]. Let .v ∈ D(Ω) and
let .-
v be the extension of .v by zero outside of .Ω. Since .Ω is bounded, we can

@seismicisolation
@seismicisolation
96 3 Variational Formulation of Boundary Problems

suppose that .Ω is contained in the band .a ≤ xn ≤ b. Let’s pose .x = (x , , xn ) with

,
. x = (x 1 , . . . , x n−1 ). Then,

{ xn
∂-v ,
v(x , , xn ) =
-
. (x , t) dt.
a ∂ xn

Using Cauchy-Schwarz inequality, we get

|{ |2
| ∂-
xn
v , |
. v(x , , xn )|2 = ||
|- (x , t) dt ||
1
a ∂ xn
{ xn | |
| ∂-v , ||2
≤ (xn − a) |
| ∂ x (x , t)| dt
a n
{ +∞ | |2
| ∂- |
≤ (xn − a) | v (x , , t)| dt.
|∂x |
−∞ n

By integrating the last inequality on .Rn−1 , we deduce

{ { | |2
| ∂- |
|- , ,
v(x , xn )| d x ≤ (xn − a)
2 | v (x , , t)| d x.
.
|
n ∂x
|
Rn−1 R n

By integrating the last inequality on .[a, b], we deduce

{ { b ({ )
, , , ,
. |-
v(x , xn )| d x d xn =2
|-
v(x , xn )| d x 2
d xn
Rn a Rn−1
{ | |2
1 | ∂- |
≤ (b − a)2 | v (x , , t)| d x.
2 |
n ∂x
|
R n

We infer that for all .v ∈ D(Ω), we have

|| ||
1 || ∂v ||2
||v||20,Ω ≤
. (b − a)2 || ||
|| ∂ x || . (3.2.1)
2 n 0,Ω

By density of .D(Ω) in . H01 (Ω), the inequality (3.2.1) stays true for all .v ∈ H01 (Ω),
which completes the proof. Q.E.D.

3.2.3 Poincaré-Wirtinger Inequality

Theorem 3.2.3 Let .Ω be a Lipschitz open subset of .Rn . Then, there exists a constant
.C depending on .Ω such that, for all .v ∈ H (Ω),
1

@seismicisolation
@seismicisolation
3.2 Sobolev Space, H01 (Ω) 97
|| { ||
|| 1 ||
||v − v d x || ≤ C||∇v||0,Ω .
.
|| mes(Ω) Ω ||
0,Ω

Proof We admit the Poincaré-Wirtinger inequality. Q.E.D.

Remark 3.2.4 Even though there is some similarity with Poincaré’s inequality, there
are major differences. In fact, the Poincaré-Wirtinger inequality fails for open sets
that are not regular enough (Lipschitz is sufficient for it to hold), while no reg-
ularity is needed for Poincaré. Instead of proving it, let us simply note that the
Poincaré-Wirtinger is at least reasonable, since the two sides vanish for the constant
functions .v. ♦

3.2.4 Some Properties of the Space . H01 (Ω)

Corollary 3.2.5 Let .Ω be an open bounded set of .Rn . Then,

.(i) The space . H0 (Ω) is a proper subspace of . H (Ω).
1 1

.(ii) The semi-norm

( n || ||2 ) 21
E || ∂v ||
v ∈ H01 (Ω), |v|1,Ω = || ||
.
|| ∂ x ||
i=1 i 0,Ω

is a norm on . H 1 (Ω) equivalent to .|| · ||1,Ω .

(iii) The space . H01 (Ω) is a Hilbert space for the inner product
.

n /
E \
∂u ∂v
.(u, v) := ,
i=1
∂ xi ∂ xi
n {
E ∂u ∂v
= d x.
i=1 Ω ∂ xi ∂ xi

The associated norm is

( n || ||2 ) 21
E || ∂v ||
|v|1,Ω
1
= (v, v) = || || .
|| ∂ x ||
. 2

i=1 i 0,Ω

Proof .(i) The function.v defined for all.x ∈ Ω,.v(x) = 1 is a function of. H 1 (Ω). But
.v ∈
/ H01 (Ω) because otherwise according to the Poincaré inequality (Theorem 3.2.2),

@seismicisolation
@seismicisolation
98 3 Variational Formulation of Boundary Problems

||v||0,Ω ≤ C(Ω)|v|1,Ω = 0,
.

which proves
.mes(Ω) ≤ 0.

This is absurd because, in general, .mes(Ω) > 0.

(ii) Let .v ∈ H01 (Ω). Since .Ω is bounded, according to the Poincaré inequality
.

(Theorem 3.2.2),
.||v||0,Ω ≤ C(Ω)|v|1,Ω .

Hence,

. |v|21,Ω ≤ ||v||21,Ω
:= ||v||20,Ω + |v|21,Ω
≤ (1 + C(Ω)2 )|v|21,Ω .

Thus,

|v|1,Ω ≤ ||v||1,Ω
.
/
≤ 1 + C(Ω)2 |v|1,Ω .

(iii) A consequence of .(ii).

. Q.E.D.

Theorem 3.2.6 .D(Rn ) is dense in . H 1 (Rn ), i.e.,

. H01 (Rn ) = D(Rn ) = H 1 (Rn ).

3.3 Trace Theorem

3.3.1 In Dimension .n = 1

We assume that .Ω is bounded throughout the rest of this paragraph. We place our-
selves in the case .n = 1, i.e., .Ω is an open bounded of .R. We take .Ω =]a, b[. We
will show that we can define the values at the edges of a function .ϕ ∈ H 1 (]a, b[),
i.e., the existence of .ϕ(a) and .ϕ(b).

Theorem 3.3.1 . H 1 (]a, b[) is a subspace of.C 0 ([a, b], R). In addition, the canonical
injection of . H 1 (]a, b[) into .C 0 ([a, b], R) is continuous. ♦

@seismicisolation
@seismicisolation
3.3 Trace Theorem 99

Proof Let .v ∈ C 1 ([a, b], R). According to the first formula for the mean, there is
.ξ ∈ [a, b] such that
{ b
1
.v(ξ ) = v(x) d x.
b−a a

Let .x ∈ [a, b], we obtain

{ x { b
1
.v(x) = v, (t) dt + v(x) d x.
ξ b−a a

Using Cauchy-Schwarz inequality, and for all .x ∈ [a, b], we get

|{ |1 ({ b ) 21
/ | x |2 1
|v(x)| ≤
. |x − ξ | ||
|v, (t)|2 dt || + √ |v(x)|2 d x
ξ b−a a
({ b ) 21 ({ b ) 21
√ , 1
≤ b−a |v (t)| dt
2
+√ |v(x)| d x
2
a b−a a
( )
√ 1
≤ max b − a, √ (||v||0,Ω + ||v, ||0,Ω )
b−a
( )
√ 1
≤ 2 max b − a, √ ||v||1,Ω .
b−a

Thus,
||v||∞ ≤ k||v||1,Ω ,
. (3.3.1)

with ( )
√ 1
.k = 2 max b − a, √ .
b−a

Let .ϕ ∈ H 1 (]a, b[). Since .C 1 ([a, b], R) is dense in . H 1 (]a, b[) (see Theorem 2.7.1),
there is a sequence .(vm )m of .C 1 ([a, b], R) such that .(vm )m converges to .ϕ in
. H (]a, b[). Using the Eq. (3.3.1), we have
1

.||v p − vq ||∞ ≤ k||v p − vq ||1,Ω .

Hence,.(vm )m is a Cauchy sequence in.(C 0 ([a, b], R), || · ||∞ ). Thus,.(vm )m converges
to .v in .C 0 ([a, b], R), which shows that .v is a continuous function. On the other hand,
{ b
.||vm − v||20,Ω = |vm − v|2 d x
a
≤ ||vm − v||2∞ (b − a).

Hence, .(vm )m converges to .v in . L 2 (]a, b[, R) and .(vm )m converges to .ϕ in the space
L 2 (]a, b[, R). Thus, the uniqueness of the limit .v = ϕ in . L 2 (]a, b[, R), which proves
.

@seismicisolation
@seismicisolation
100 3 Variational Formulation of Boundary Problems

that .v ∈ _
ϕ (class of .ϕ). Hence, .v is a continuous representative of ._ϕ . We define the
values at the edges of .ϕ as follows: we pose .ϕ(a) = v(a) and .ϕ(b) = v(b) it exists
because .v is continuous on .[a, b]. We identify .ϕ and .v which amounts to identifying
. H (]a, b[) to a subspace of .C ([a, b], R). Let us show that the canonical injection
1 0

from . H (]a, b[) into .C ([a, b], R) is continuous. In fact, according to (3.3.1), we
1 0

have .||vm ||∞ ≤ k||vm ||1,Ω . Passing to the limit there comes .||v||∞ ≤ k||ϕ||1,Ω . Now,
we have identified .ϕ with .v so .||v||∞ ≤ k||v||1,Ω . Thus, the canonical injection from
. H (]a, b[) into .C ([a, b], R) is continuous.
1 0
Q.E.D.
Remark 3.3.2 The example of .u(x) = x sin 1
x
for .x ∈]0, 1] and .u(0) = 0 shows
that the converse of Theorem 3.3.1 is false. ♦

3.3.2 In Dimension .n ≥ 2

We place ourselves in the case, where .n ≥ 2, i.e.,

.Ω ⊂ Rn , n ≥ 2.

Let .v ∈ H 1 (Ω), we want to define its value at the edge .v|Γ , where .Γ = ∂Ω. This is
not always possible in a classical way because for .n ≥ 2 the functions of . H 1 (Ω) are
not in general continuous. Indeed, the function .ϕ defined by

ϕ(x) = | log ||x||2 |k , x ∈ Ω \ {0},

.

/
with .Ω = B(0, R) ⊂ R2 and .||x||2 = x12 + x22 , belongs to . H 1 (Ω) but it does not
admit any continuous representative .v ∈ C 0 (Ω, R) if .0 < k < 21 . In fact, suppose

@seismicisolation
@seismicisolation
3.3 Trace Theorem 101

that .ϕ ∈ H 1 (Ω) and .v ∈ C 0 (Ω, R) such that

. v=ϕ

almost everywhere. Since .Ω is compact, for .x ∈ Ω, we have

|v(x)| ≤ ||v||∞
.

:= sup |v(x)|.
x∈Ω

On the other hand,

. lim ϕ(x) = ∞.
||x||→0

Hence, there is .r ∈]0, R[ such that

.ϕ(x) > ||v||∞

for all .x ∈ B(0, r ), which proves

ϕ /= v
.

on . B(0, r ) and .mes(B(0, r )) /= 0 which is absurd.

So, the class of functions of . H 1 (Ω), for .n ≥ 2, does not admit a continuous
representative on .Ω. However, we have the following result.
Theorem 3.3.3 Let .Ω be an open bounded domain of .Rn with piecewise smooth
boundary .Γ = ∂Ω. Then, the space .D(Ω) is dense in . H 1 (Ω). Furthermore, the
mapping
.γ0 : v −→ γ0 (v) = v|Γ

of .D(Ω) into . L 2 (Γ, R) is extended by continuity in one map from . H 1 (Ω) into
. L (Γ, R), noted again .γ0 , i.e.,
2

γ : H 1 (Ω) −→L 2 (Γ, R)

. 0

v −→γ0 (v) = v|Γ

is continuous. .γ0 named trace mapping. ♦

If .v ∈ H 1 (Ω), then .γ0 (v) named trace of .v on .Γ . Note that .γ0 (H 1 (Ω)) is a clean
subspace of . L 2 (Γ, R). At most, .γ0 is continuous if, and only if,

||γ0 (v)||0,Γ ≤ M||v||1,Ω

.

with . M is a constant.

@seismicisolation
@seismicisolation
102 3 Variational Formulation of Boundary Problems

3.3.3 Application of the Trace Theorem

The first result is a characterization of the space . H01 (Ω).

Theorem 3.3.4 If .Ω is an open bounded with piecewise smooth boundary .Γ = ∂Ω,
then the space . H01 (Ω) is the kernel of .γ0 , with .γ0 is the trace mapping from . H 1 (Ω)
into . L 2 (Γ, R), i.e.,

{ }
. H01 (Ω) = v ∈ H 1 (Ω) such that γ0 (v) = v|Γ = 0 .

♦
Proof Let us show that
{ }
. H01 (Ω) ⊂ v ∈ H 1 (Ω) such that γ0 (v) = v|Γ = 0 .

Let.v ∈ H01 (Ω). Hence, there is.(vm )m a sequence of.D(Ω) such that.(vm )m converges
to .v in . H01 (Ω). Since .γ0 is continuous, .(γ0 (vm ))m converges to .γ0 (v) in . L 2 (Γ, R).
Moreover,
.γ0 (vm ) = 0

because .vm ∈ D(Ω) with compact support included in the opening .Ω, which proves

γ (v) = 0.
. 0

Thus, .v|Γ = 0.
Reciprocally, (exercise). Q.E.D.

3.3.4 Green’s Formula

The following result concerns Green’s formula. We pose for all .i ∈ {1, . . . , n}, .νi
the .ith directing cosine of the normal .ν at .Γ = ∂Ω directed outwards.

@seismicisolation
@seismicisolation
3.3 Trace Theorem 103
⎛ ⎞
ν1
⎜ . ⎟
.ν = ⎝ . ⎠ .
.
νn

Theorem 3.3.5 We assume that .Ω is an open bounded of .Rn with piecewise smooth
boundary .Γ = ∂Ω. Let .u, .v ∈ H 1 (Ω), then
{ { {
∂u ∂v
. v dx = − u dx + uvνi dσ, 1 ≤ i ≤ n,
Ω ∂ xi Ω ∂ xi Γ

where .dσ is a surface element. ♦

Proof The result is true for all .u, .v ∈ D(Ω). The result follows from the density
of .D(Ω) in . H 1 (Ω) (see Theorem 3.3.3) and the continuity of the trace mapping .γ0 ,
which finishes the proof. Q.E.D.

We have the following lemma (see Ciarlet [3]):

Lemma 3.3.6 Let .Ω be a bounded open set with boundary .Γ , piecewise .C 1 . We

| |
L
assume that .Ω = Ω i , where
i=1
.(i) .Ωi is an open set of .Rn contents in .Ω, whose boundary .Γi is piecewise .C 1 ,
.1 ≤ i ≤ L.
N
.(ii) .Ωi Ω j = ∅ for .i /= j.
Finally, let.v ∈ C 0 (Ω, R) be a function such that for each.i = 1, . . . , L the restric-
tion .v|Ωi of .v to .Ωi belongs to . H 1 (Ωi ). Then, .v ∈ H 1 (Ω). ♦

Proof For .1 ≤ j ≤ n, we define .v j ∈ L 2 (Ω, R) by

∂ ( )
v
. j
|Ωi = v|Ωi .
∂x j

∂v
Let us show that .v j = ∂x j
in the sense of .D , (Ω). Indeed, let .ϕ ∈ D(Ω), we have
{
.(v j , ϕ) = vjϕ dx
Ω
L {
E ∂ ( )
= v|Ωi ϕ d x
i=1 Ωi ∂x j

and using Green’s formula (Theorem 3.3.5), we obtain

L {
E E L {
∂ϕ
(v j , ϕ) = −
. v dx + vϕn j ds.
i=1 Ωi ∂x j i=1 Γi

@seismicisolation
@seismicisolation
104 3 Variational Formulation of Boundary Problems

Since the function .v is continuous and .ϕ vanishes on .Γ , we have

L {
E
. vϕn j ds = 0.
i=1 Γi

So,
{
∂ϕ
.(v j , ϕ) = − v dx
Ω ∂ xj
/ \
∂ϕ
= − v,
∂x j
/ \
∂v
= ,ϕ .
∂x j

We deduce that .v ∈ H 1 (Ω). Q.E.D.

3.4 Sobolev Spaces of Order .m

3.4.1 Sobolev Spaces of Order .m, . H m (Ω) .m ≥ 1

Definition 3.4.1 Let .m be an integer greater than or equal to .1. We call Sobolev
space of order .m on .Ω, the space

. H m (Ω) = {v ∈ L 2 (Ω, R) such that ∂ α v ∈ L 2 (Ω, R), |α| ≤ m},

with .α = (α1 , . . . , αn ) ∈ Nn is an multi-index and .|α| = α1 + · · · + αn . ♦

We provide . H m (Ω) with the following inner product:

E {
(u, v)m,Ω =
. ∂ α u ∂ α v d x.
|α|≤m Ω

We notice 1
||v||m,Ω = (v, v)m,Ω
.
2

the corresponding norm.

Example 3.4.1 .(a) If .n = 1, i.e., .Ω ⊂ R, then

. H 2 (Ω) = {v ∈ L 2 (Ω, R) such that v, ∈ L 2 (Ω, R) and v,, ∈ L 2 (Ω, R)}.

We have

@seismicisolation
@seismicisolation
3.4 Sobolev Spaces of Order m 105

. H 2 (Ω) ⊂ H 1 (Ω)

and { { {
, ,
.(u, v)2,Ω = uv d x + u v dx + u ,, v,, d x.
Ω Ω Ω

(b) If .n = 2, i.e., .Ω ⊂ R2 , then

.

{
2 ∂v ∂v ∂ 2v
. H (Ω) = v ∈ L 2 (Ω, R) such that ∈ L 2 (Ω, R), ∈ L 2 (Ω, R), ∈ L 2 (Ω, R),
∂x ∂y ∂x2
}
∂ 2v 2 ∂ 2v 2 ∂ 2v 2
∈ L (Ω, R), ∈ L (Ω, R), and ∈ L (Ω, R) .
∂ y2 ∂ x∂ y ∂ y∂ x

We have
. H 2 (Ω) ⊂ H 1 (Ω)

and
E{
. (u, v)2,Ω = ∂αu ∂αv d x
|α|≤2 Ω
{ { {
∂u ∂v ∂u ∂v
= uv d xdy + d xd y + d xd y
Ω Ω ∂ x ∂ x Ω ∂y ∂y
{ {
∂ 2u ∂ 2v ∂ 2u ∂ 2v
+ d xd y + d xd y
Ω ∂x ∂x Ω ∂y ∂y
2 2 2 2
{ {
∂ 2u ∂ 2v ∂ 2u ∂ 2v
+ d xd y + d xd y.
Ω ∂ x∂ y ∂ x∂ y Ω ∂ y∂ x ∂ y∂ x

3.4.2 Separable Hilbert Space

Theorem 3.4.2 The space . H m (Ω) is a separable Hilbert space for the following
inner product:
E {
.(u, v)m,Ω = ∂ α u ∂ α v d x.
|α|≤m Ω
♦
Example 3.4.2 Let .Ω be an open bounded domain of .R with piecewise smooth n

boundary .Γ = ∂Ω. Consider . H 2 (Ω). We have

. H 2 (Ω) ⊂ H 1 (Ω).

Let .v ∈ H 2 (Ω), then .v ∈ H 1 (Ω) and so we can define the trace of .v on .Γ = ∂Ω by

@seismicisolation
@seismicisolation
106 3 Variational Formulation of Boundary Problems

γ (v) = v|Γ ∈ L 2 (Γ, R).

. 0

Furthermore, .v ∈ H 2 (Ω), so for .i ∈ [1, . . . , n],

∂v
. ∈ H 1 (Ω).
∂ xi

Hence, we can define the trace of . ∂∂vxi on .Γ = ∂Ω by

( )
∂v ∂v
γ
. 0 = ∈ L 2 (Γ, R)
∂ xi ∂ xi |Γ

so that the normal derivative outside to the border .Γ = ∂Ω

∂v E ∂vn
γ (v) :=
. 1 = νi |Γ
∂ν |Γ i=1
∂ xi

makes sense as an element of . L 2 (Γ, R) .(νi |Γ ∈ L ∞ (Γ, R)).

The mapping

γ : H 2 (Ω) −→L 2 (Γ, R) × L 2 (Γ, R)

.

v −→γ (v) = (γ0 (v), γ1 (v))

is linear and continuous.

3.4.3 Trace Theorem

Theorem 3.4.3 Let .Ω be an open bounded domain of .Rn with piecewise smooth
boundary .Γ = ∂Ω. Then, the mapping

@seismicisolation
@seismicisolation
3.4 Sobolev Spaces of Order m 107

γ : D(Ω) −→ L 2 (Γ, R) × L 2 (Γ, R)

.
( )
∂v
v −→ γ (v) = (γ0 (v), γ1 (v)) := v|Γ ,
∂ν |Γ

extends into a continuous linear map also noted .γ from the space . H 2 (Ω) into
. L (Γ, R) × L (Γ, R). ♦
2 2

3.4.4 The Laplacian Operator

Let .u ∈ H 2 (Ω). We notice .Δu by

.Δu := div(grad u)
:= div(∇u)
En
∂ 2u
:= .
i=1
∂ xi2

Δu ∈ L 2 (Ω, R) since .u ∈ H 2 (Ω). .Δu named the Laplacian of .u.

.

Theorem 3.4.4 Let .Ω be an open bounded domain of .Rn with piecewise smooth
boundary .Γ = ∂Ω. Then, for all function .u ∈ H 2 (Ω) and for all function .v ∈
H 1 (Ω), we have
{ n {
E {
∂u ∂v ∂u
. − Δu v d x = dx − v dσ,
Ω i=1 Ω ∂ xi ∂ xi Γ ∂ν

with
∂u E ∂u
n
. = νi
∂ν i=1
∂ xi |Γ

and .dσ is a surface element. ♦

@seismicisolation
@seismicisolation
108 3 Variational Formulation of Boundary Problems

3.4.5 The Canonical Injection

Definition 3.4.5 We denote by . H02 (Ω) the closure of .D(Ω) in . H 2 (Ω), i.e.,

H 2 (Ω)
. H02 (Ω) = D(Ω) .

♦
H 2 (Ω)
Theorem 3.4.6 Let . H02 (Ω) = D(Ω) . Then,
{ }
∂v
. H0 (Ω) = K er (γ ) = v ∈ H (Ω) such that v|Γ = =0
2 2
∂ν |Γ

with .γ (·) is the trace mapping. ♦

Let.∇ 2 v denote the collection of all.n 2 second-order partial derivatives of the function
.v. We give the following lemma.

Lemma 3.4.7 On the space . H02 (Ω), the semi-norm .||∇ 2 v||0,Ω is a norm. It is equiv-
alent to the . H 2 norm. ♦

Proof It is sufficient to establish a bound from below. Let .v ∈ H02 (Ω). Then, . ∂∂vxi ∈
H01 (Ω) for all .i. Therefore, by using Poincaré inequality (Theorem 3.2.2), we have
|| ( )||2 || ||
|| ∂v || || ∂v ||2
. ||∇ || ||
≥ C || || .
2
|| ∂x || ∂ xi ||1,Ω
i 0,Ω

On the other hand,

|| || || ( )||2 || ||2
|| ∂v ||2 || || || ||
|| || = ||∇ ∂v || + || ∂v ||
.
|| ∂ x || || ||
∂ xi 0,Ω || ∂ xi ||0,Ω
i 1,Ω

and by summing in .i, we get

n || (
E )||2
|| ||
||∇ 2
v||20,Ω = ||∇ ∂v ||
.
|| ∂x ||
i=1 i 0,Ω

≥ C (||∇ v||20,Ω + |v|21,Ω )

2 2

≥ C 2 ||∇ 2 v||20,Ω + C 4 ||v||21,Ω

≥ C 4 ||v||22,Ω

since .v ∈ H01 (Ω) and .C ≤ 1. Q.E.D.

@seismicisolation
@seismicisolation
3.4 Sobolev Spaces of Order m 109

Theorem 3.4.8 Let .Ω be an open bounded domain of .Rn with piecewise smooth
boundary .Γ = ∂Ω. If .m > n2 , then the space

. H m (Ω) ⊂ C 0 (Ω, R)

and the canonical injection from . H m (Ω) into .C 0 (Ω, R) is continuous. ♦

Example 3.4.3 .(a) Let .Ω =]a, b[⊂ R. If .v ∈ H 1 (]a, b[), then .v ∈ C 0 ([a, b], R).
If .v ∈ H 2 (]a, b[), then .v, ∈ H 1 (]a, b[) and so .v, ∈ C 0 ([a, b], R). Thus, .v is of class
.C on .[a, b].
1

.(b) Let .Ω ⊂ R2 . If .v ∈ H 2 (Ω), then according to Theorem 3.4.8, since .2 > 22 ,

. H (Ω) ⊂ C (Ω, R). Thus, .v ∈ C (Ω, R).
2 0 0

Theorem 3.4.9 Let .Ω be an open bounded domain of .Rn with piecewise smooth
boundary .Γ = ∂Ω. Then, . H m (Ω) continuously injects itself into .C s (Ω, R) for all
.m > + s, with .s ∈ N, where .C s (Ω, R) is the set of functions .s times differentiable
n
2
from .Ω into .R whose .nth derivative is continuous. ♦

Let . Pk be the space of polynomials, in .Rn , of the form

E
. ai1 i2 ...in x1i1 x2i2 . . . xnin ,
i 1 +i 2 +···+i n ≤k

with .ai1 i2 ...in ∈ R and let . Q k be the space of polynomials, in .Rn , of the form
E
. ai1 i2 ...in x1i1 x2i2 . . . xnin ,
i 1 ,i 2 ,...,i n ≤k

with .ai1 i2 ...in ∈ R.

We end this section with the following theorems:
Theorem 3.4.10 Let . O ⊂ Rn be an open bounded with piecewise smooth boundary
.∂ O and let . L : H (O) −→ R be a continuous linear form checking
k+1

(i) .|L(v)| ≤ c||L||∗k+1,O ||v||k+1,O for all .v ∈ H k+1 (O), where .|| · ||∗k+1,O designates
.

the norm on the space dual of . H k+1 (O), given by

|L(v)|
||L||∗k+1,O =
. sup .
v∈H k+1 (O), v/=0 ||v||k+1,O

(ii) . L(v) = 0 for all .v ∈ Pk , with . Pk denotes the space of polynomials of degree
.
≤ k, with coefficients in .R.
.

Then,
∗
.|L(v)| ≤ c||L||k+1,O |v|k+1,O

with

@seismicisolation
@seismicisolation
110 3 Variational Formulation of Boundary Problems

⎛ ⎞ 21
E {
|v|k+1,O := ⎝
. (∂ α v)2 d x ⎠ ,
|α|=k+1 O

α = (α1 , . . . , αn ) ∈ Nn the multi-index,

.

( ) ( )
α ∂ α1 ∂ αn
.∂ v = ... v
∂ x1 ∂ xn
∂ |α|
= v
∂ x1α1 . . . ∂ xnαn

and .|α| = α1 + · · · + αn . ♦

Theorem 3.4.11 Let . O ⊂ Rn be an open bounded domain of .Rn with piecewise

smooth boundary .∂ O. Then,
⎛ ⎛ ⎞ ⎞ 21
{ E
v −→ |v|k+1,O := ⎝
. ⎝ |∂ α v|2 ⎠ d x ⎠
O |α|=k+1

is a norm on the quotient space . H k+1 (O)/Pk and there are .α and .β such that

α|v|k+1,O ≤ inf ||v + p||k+1,O ≤ β|v|k+1,O .

.
p∈Pk

Theorem 3.4.12 Let . O ⊂ Rn be an open bounded domain of .Rn with piecewise

smooth boundary .∂ O and let .π ∈ L (H k+1 (O), H m (O)), .m ≤ k + 1 be such that

πp = p
.

for all . p ∈ Pk . Then,

||v − π v||m,O ≤ β||I d − π ||L (H k+1 (O),H m (O)) |v|k+1,O .

.

Proof For all . p ∈ Pk , we have

v − π v = v + p − π(v + p).
.

Hence
||v − π v||m,O ≤ ||I d − π ||L (H k+1 (O),H m (O)) ||v + p||k+1,O .
.

@seismicisolation
@seismicisolation
3.4 Sobolev Spaces of Order m 111

So,
||v − π v||m,O ≤ ||I d − π ||L (H k+1 (O),H m (O)) inf ||v + p||k+1,O .
.
p∈Pk

We deduce the result of which the constant .β being that of Theorem 3.4.11.
Q.E.D.

3.4.6 Sobolev Spaces of Order .m, . W m, p (Ω) .1 ≤ p ≤ ∞

and .m ∈ N

Let .Ω be an open of .Rn . For .1 ≤ p ≤ ∞ and .m ∈ N, we define the Sobolev space

W m, p (Ω) by
.

. W m, p (Ω) = {u ∈ L p (Ω, R) such that ∂ α u ∈ L p (Ω, R) for all α ∈ Nn and |α| ≤ m}

with .α = (α1 , . . . , αn ) ∈ Nn the multi-index,

( ) ( )
α ∂ α1 ∂ αn
.∂ u = ... u
∂ x1 ∂ xn
∂ |α|
= u
∂ x1α1 . . . ∂ xnαn

and .|α| = α1 + · · · + αn . Note that for .m = 0, the space .W 0, p (Ω) is the Lebesgue
space . L p (Ω, R). Note that .W m,2 (Ω) = H m (Ω). A natural norm on .W m, p (Ω) is
defined by
⎧⎛ ⎞ 1p
⎪
⎪ E
⎪
⎨ ⎝ ||∂ α u|| L p (Ω,R) ⎠ if 1 ≤ p < ∞
p
||u||m, p,Ω
. :=
⎪
⎪ 0≤|α|≤m
⎪
⎩ max ||∂ α u|| L ∞ (Ω,R) if p = ∞,
0≤|α|≤m

where .|| · || L p (Ω,R) designates the norm in . L p (Ω, R) and .|| · || L ∞ (Ω,R) designates the
norm in . L ∞ (Ω, R). Equipped with this norm the space .W m, p (Ω) is a Banach space
(i.e., a complete normed vector space). We can show that the norm
⎧ E
⎪
⎪ ||∂ α u|| L p (Ω,R) if 1 ≤ p < ∞
⎨
||u||m, p,Ω
. E
:= 0≤|α|≤m
⎪
⎪ ||∂ α u|| L ∞ (Ω,R) if p = ∞,
⎩
0≤|α|≤m

is an equivalent norm to the previous. The space .W m, p (Ω), therefore, has the same
properties regardless of the used norm. These norms are noted indifferently.|| · ||m, p,Ω .
Consider on the space .W m, p (Ω), the semi-norm

@seismicisolation
@seismicisolation
112 3 Variational Formulation of Boundary Problems
⎧⎛ ⎞ 1p
⎪
⎪ E
⎪
⎨⎝ ||∂ α u|| L p (Ω,R) ⎠ if 1 ≤ p < ∞
p
.|u|m, p,Ω :=
⎪
⎪ |α|=m
⎪
⎩ max ||∂ α u|| L ∞ (Ω,R) if p = ∞
|α|=m

and ⎧( ) 1p
⎪ En || m || p
⎪
⎪ || ∂ u ||
⎪
⎨ || || if 1 ≤ p < ∞
|| ∂ x m || p
. [u]m, p,Ω := i=1 || i || L (Ω,R)
⎪
⎪ || ∂ m u ||
⎪
⎪ || || if p = ∞.
⎩ 1≤i≤n || ∂ x m ||
max
i ∞ L (Ω,R)

Remark 3.4.13 In dimension .1 of space (.n = 1), with .a, .b ∈ R, any element of
W 1, p (]a, b[) (which is therefore a class of functions) can be assimilated to a contin-
.
uous function, at meaning that there is a representative of the class which is contin-
uous. This is because in dimension .1, any function of .W 1, p (]a, b[) can be written
as the integral of its derivative. The function .u ∈ W 1, p (]a, b[) if, and only if, there
exists .-
u ∈ C 0 ([a, b], R) and .v ∈ L p ((a, b), R) such that

u =-
. u

almost everywhere and { x

-
u (x) = -
. u (a) + v(s) ds.
a

In dimension strictly greater than .1, this is false. ♦

Theorem 3.4.14 In the case, where . p is finished, it is also a separable space, i.e., a
normed vector space which contains a dense countable part. ♦

The Sobolev spaces .W m, p (Ω) are reflexive except if . p = 1 or .∞.

Let.(W m, p (Ω)), be the dual space of.W m, p (Ω). We note.( f, u) the duality between
an element .u of .W m, p (Ω) and an element . f ∈ (W m, p (Ω)), and we define the dual
norm by
∗ |( f, v)|
.|| f ||m, p,Ω = sup .
v∈W (Ω), v/=0
m, p ||v|| m, p,Ω

From Lemma 2.3.2, we can show the following result, very useful later.

Lemma 3.4.15 (Lesaint [4]) We consider a bounded open set .Ω, with continuous
boundary (Necas [5, Chap. II]). Let .q be such that .1 ≤ q ≤ ∞, .k and .r are two
positive integers and let . f ∈ (W k+r +1,q (Ω)), be such that

. ( f, u) = 0 for all u ∈ Pk (resp., Q k ). (3.4.1)

@seismicisolation
@seismicisolation
3.4 Sobolev Spaces of Order m 113

Then, there exists a constant .C = C(n, k, r, q, Ω) such that

E
k+r +1
|( f, u)| ≤ C||
. f ||∗k+r +1,q,Ω |u|l,q,Ω , (3.4.2)
l=k+1

E
k+r +1
. resp., |( f, u)| ≤ C|| f ||∗k+r +1,q,Ω [u]l,q,Ω , (3.4.3)
l=k+1

for all .u ∈ W k+r +1,q (Ω). ♦

k+r +1,q
Proof We apply Lemma Er 2.3.2 with . X = W (Ω), . X 1 = W k,q (Ω), . X 2 =
(L (Ω, R)) , . N = l=0 Nl , where . Nl denotes the number of possible derivatives
q N

of order .k + l + 1. The operator . A1 is the identity operator and the operator . A2

matches any function .v ∈ W k+r +1,q (Ω), the set of all its derivatives of order .|α 0 |,
.|α |, . . . , |α |, where .α is a multi-index such as .|α | = k + l + 1, .0 ≤ l ≤ r . The
1 r l l

hypotheses .(i) and .(ii) of Lemma 2.3.2 are, therefore, satisfied and Ethe kernel of . A2
k+r +1
is the space . Pk (. K er (A2 ) = Pk ). We deduce that the semi-norm . l=k+1 | · |l,q,Ω is
a norm on the quotient space .W k+r +1,q (Ω)/Pk , equivalent to the quotient norm

||u||W k+r +1,q (Ω)/Pk = inf ||u + v||k+r +1,q,Ω ,

.
v∈Pk

where .u denotes the equivalence class of any of its .u elements. There thus exists a
constant .C depending only on .n, .k, .r , .q, and .Ω such that

E
k+r +1
||u||W k+r +1,q (Ω)/Pk ≤ C
. |u|l,q,Ω ,
l=k+1

for all .u ∈ W k+r +1,q (Ω). Let . f ∈ (W k+r +1,q (Ω)), satisfy the equality (3.4.1), we
have .( f, u) = ( f, u + v) for all .u ∈ W k+r +1,q (Ω) and all .v ∈ Pk such that

.|( f, u)| ≤ || f ||∗k+r +1,q,Ω inf ||u + v||k+r +1,q,Ω ,

v∈Pk

E
k+r +1
|( f, u)| ≤ C||
. f ||∗k+r +1,q,Ω |u|l,q,Ω ,
l=k+1

which ends the proof of the inequality (3.4.2). To show the inequality (3.4.3), we apply
the Lemma 2.3.2, with . X = W k+r +1,q (Ω), . X 1 = W k,q (Ω), . X 2 = (L q (Ω, R))n(r +1) .
The operator . A1 is the identity, and the operator . A2 is the one that has any function of
k+r +1,q
.W (Ω) matches its .n(r + 1) derivatives of order, respectively, equal to .k + 1,
.k + 2, . . . , k + r + 1, in the directions. x 1 , . . . , x n . A result due to K. T. Smith, proved
in [6], implies that the hypothesis .(i) of Lemma 2.3.2 is satisfied. The hypothesis

@seismicisolation
@seismicisolation
114 3 Variational Formulation of Boundary Problems

(ii) of Lemma 2.3.2 is satisfied thanks to the lemma of Rellich and Konndrachoff
.

demonstrated in [7, Chap. II] and in [5, pp. 17 and 106]. Ek+rOn the other hand, we
+1
have . Q k = K er (A2 ). We deduce that the semi-norm . l=k+1 [·]l,q,Ω is a norm on
the quotient space .W k+r +1,q (Ω)/Q k equivalent to the norm

||-
. u ||W k+r +1,q (Ω)/Qk = inf ||u + v||k+r +1,q,Ω ,
v∈Q k

where .-
u denotes the equivalence class of any of its .u elements. There thus exists a
constant .C > 0 depending only on .n, .k, .q, and .Ω such that

E
k+r +1
||-
. u || W k+r +1,q (Ω)/Q k ≤C [u]l,q,Ω , (3.4.4)
l=k+1

for all .u ∈ W k+r +1,q (Ω). The inequality (3.4.3) is then shown, from the inequality
(3.4.4), in the same way as inequality (3.4.2). Q.E.D.

The following result is a generalization of Lemma 3.4.15 which allows to specify the
value of the constants .C intervening in the inequalities (3.4.2) and (3.4.3), which can
be useful for some applications.
Lemma 3.4.16 (Lesaint [4]) We consider a bounded open set .Ω, with continuous
boundary. Let .q be a number such that .1 ≤ q ≤ ∞ and two integers .k and .r , with
k+r +1,q
.k ≥ 0, .r ≥ 1 and let . f be a continuous linear form on . W (Ω) such that

E
k+r
|( f, u)| ≤ ||
. f ||∗k+r +1,q,Ω (ηl |u|l,q,Ω + μl [u]l,q,Ω ), (3.4.5)
l=k+1

for all .u ∈ Pk+r (resp., . Q k+r ). Then, there exists a constant .C = C(n, k, r, q, Ω)
such that

E
k+r +1
|( f, u)| ≤ C|| f ||∗k+r +1,q,Ω
. (αl |u|l,q,Ω + βl [u]l,q,Ω ), (3.4.6)
l=k+1

for
Ek+rall.u ∈ W k+r +1,q (Ω), with.αl = ηl ,.βl = μl , forE
Ek+r .k + 1 ≤ l ≤ k + r .αk+r +1 = 1 +
Ek+r
k+r
l=k+1 lη , .βk+r +1 = l=k+1 lμ (resp., .αk+r +1 = l=k+1 ηl , .βk+r +1 = 1 + l=k+1
μl ). ♦

Proof We first consider the case, where the equality (3.4.5) is satisfied for .u ∈ Pk+r .
The following property is satisfied (according to the proof of Lemma 3.4.15). The
semi-norm.| · |k+r +1,q,Ω is a norm on the quotient space.W k+r +1,q (Ω)/Pk+r , equivalent
to the quotient norm

.||u||W k+r +1,q (Ω)/Pk+r = inf ||u + v||k+r +1,q,Ω , (3.4.7)

v∈Pk+r

@seismicisolation
@seismicisolation
3.4 Sobolev Spaces of Order m 115

where .u denotes the equivalence class of any of its .u elements. There thus exists a
constant .C = C(n, k, r, q, Ω), and a polynomial .v0 of . Pk+r such that

. ||u + v0 ||k+r +1,q,Ω ≤ C|u|k+r +1,q,Ω . (3.4.8)

Let . f ∈ (W k+r +1,q (Ω)), satisfy the equality (3.4.5), we have

( f, u) = ( f, u + v0 ) − ( f, v0 ).
.

From where we deduce

|( f, u)| ≤ || f ||∗k+r +1,q,Ω ||u + v0 ||k+r +1,q,Ω + |( f, v0 )|.

. (3.4.9)

By using the equality (3.4.5), we have

E
k+r
.|( f, v0 )| ≤ || f ||∗k+r +1,q,Ω (ηl |v0 |l,q,Ω + μl [v0 ]l,q,Ω ). (3.4.10)
l=k+1

According to inequality (3.4.8), we have

. |v0 |l,q,Ω ≤ |u|l,q,Ω + C|u|k+r +1,q,Ω , (3.4.11)

. [v0 ]l,q,Ω ≤ [u]l,q,Ω + C|u|k+r +1,q,Ω , (3.4.12)

for all .l ≤ k + r . By combining the relations (3.4.8), (3.4.9), (3.4.10), (3.4.11), and
(3.4.12), we obtain the inequality (3.4.7). In the case, where the inequality (3.4.5) is
satisfied for .u ∈ Q k+r , there exists, as before, a constant .C = C(n, k, r, q, Ω) and a
polynomial .v0 of . Q k+r such that

||u + v0 ||k+r +1,q,Ω ≤ C[u]k+r +1,q,Ω .

.

We deduce the inequalities

. |v0 |l,q,Ω ≤ |u|l,q,Ω + C[u]k+r +1,q,Ω ,

[v0 ]l,q,Ω ≤ [u]l,q,Ω + C[u]k+r +1,q,Ω ,

.

for any integer .l ≤ k + r . By combining the relations (3.4.9) and (3.4.10), with the
last three relations, we obtain the inequality (3.4.6). Q.E.D.

In the special case, where .Ω is a bounded open set of .R2 , we have the following
result.

@seismicisolation
@seismicisolation
116 3 Variational Formulation of Boundary Problems

Lemma 3.4.17 (Lesaint [4]) We consider a bounded open set .Ω of .R2 , with contin-
uous boundary. Let .q be a number such that .1 ≤ q ≤ ∞ and an integer .k ≥ 0 and
let . f be a continuous linear form on .W k+2,q (Ω) such that
| |
|( f, u)| ≤ ε|| f ||∗k+2,q,Ω [u]k+1,q,Ω for all u ∈ Q k
. Pk+1 . (3.4.13)

Then, there exists a constant .C = C(k, q, Ω) such that

( ( [ ] ) )
∂ 2u
.|( f, u)| ≤ C|| f ||∗k+2,q,Ω (1 + ε) [u]k+2,q,Ω + + ε[u]k+1,q,Ω ,
∂ x∂ y k,q,Ω
(3.4.14)
for all .u ∈ W k+2,q (Ω). ♦

Proof We proceed as to prove Lemmas 3.4.15 and 3.4.16. We apply Lemma 2.3.2,
with . X = W k+2,q (Ω), . X 1 = L q (Ω, R), . X 2 = (L q (Ω, R))4 . The operator . A1 is the
identity operator and the operator . A2 maps any function .v ∈ W k+2,q (Ω) to the
quadruplet ( k+2 )
∂ v ∂ k+2 v ∂ k+2 v ∂ k+2 v
. , , , .
∂ x k+2 ∂ y k+2 ∂ x k+1 ∂ y ∂ x∂ y k+1

Using a result of Smith [6] and the lemma of Rellich and Kondrachoff [5], we show
that the hypotheses .(i) andu .(ii) of Lemma 2.3.2 are satisfied. On the other hand,
we have . K er (A2 ) [= Q k] Pk+1 . From Lemma 3.4.15, we deduce that the semi-
norm.[u]k+2,q,Ω + ∂∂x∂uy is a norm on the quotient space.W k+2,q (Ω)/(Q k u Pk+1 ) ,
2

k,q,Ω
equivalent to the norm

||-
. u ||W k+2,q (Ω)/(Qk u Pk+1 ) = inf
u ||u + v||k+2,q,Ω ,
v∈Q k Pk+1

where .-
u denotes the equivalence class of any of its .uu
elements. There thus exists a
constant .C = C(k, q, Ω) and a polynomial . p0 ∈ Q k Pk+1 such that
( [ ] )
∂ 2u
||u + p0 ||k+2,q,Ω ≤ C [u]k+2,q,Ω
. + . (3.4.15)
∂ x∂ y k,q,Ω

Let . f be a continuous linear form on .W k+2,q (Ω) and satisfy the inequality (3.4.13).
Then,
.( f, u) = ( f, u + p0 ) − ( f, p0 ).

Hence, ( )
|( f, u)| ≤ || f ||∗k+2,q,Ω ||u + p0 ||k+2,q,Ω + ε[ p0 ]k+1,q,Ω .
.

@seismicisolation
@seismicisolation
3.5 Dirichlet Problem 117

According to the inequality (3.4.15), we can write that

( [ ] )
∂ 2u
[ p0 ]k+1,q,Ω ≤ [u]k+1,q,Ω + C [u]k+2,q,Ω
. + .
∂ x∂ y k,q,Ω

By combining the last four relations, we have the inequality (3.4.14). Q.E.D.

3.5 Dirichlet Problem

Let .Ω be an open bounded domain of .Rn with piecewise smooth boundary .Γ = ∂Ω.
Given a function . f ∈ L 2 (Ω, R), we search a function .u defined on .Ω checking

. − Δu = f in Ω (3.5.1)
.u = 0 on Γ = ∂Ω. (3.5.2)

The problem (3.5.1)–(3.5.2) is called strong problem. The boundary condition is

“.u = 0 on .Γ = ∂Ω”, the so-called homogeneous Dirichlet boundary condition.

3.5.1 Variational Formulation

Suppose that the problems (3.5.1) and (3.5.2) admit a fairly regular solution, for
example, .u ∈ H 2 (Ω).
Question: How we are going to choose the appropriate Hilbert space .V to find a
system equivalent to the boundary problem (3.5.1)–(3.5.2).
The general idea is to choose a space.V included in. H 1 (Ω) by adding all the boundary
conditions on .u but not on the derivatives of .u, given in the strong problem (3.5.1)–
(3.5.2). Hence,

. V := {v ∈ H 1 (Ω) such that v = 0 on Γ = ∂Ω} := H01 (Ω).

Let .v ∈ V := H01 (Ω) be a test function. Multiply Eq. (3.5.1) by .v,

. − Δu v = f v

and integrate over .Ω, { {

. −Δu v d x = f v d x.
Ω Ω

@seismicisolation
@seismicisolation
118 3 Variational Formulation of Boundary Problems

Using Theorem 3.4.4, we deduce

n {
E { {
∂u ∂v ∂u
. dx − v dσ = f v dx
i=1 Ω ∂ xi ∂ xi Γ ∂ν Ω

with
∂u E ∂u n
. = νi .
∂ν i=1
∂ xi |Γ

Moreover, .v ∈ H01 (Ω) so .v|Γ = 0. Thus,

{
∂u
. v dσ = 0.
Γ ∂ν

We replace the problem (3.5.1) and (3.5.2) by the following problem:

Find .u ∈ H01 (Ω) such that for all .v ∈ H01 (Ω), we have

n {
E {
∂u ∂v
. dx = f v d x. (3.5.3)
i=1 Ω ∂ xi ∂ xi Ω

Definition 3.5.1 1. The problem (3.5.3) is called the variational formulation (or
weak formulation) of the problem (3.5.1)–(3.5.2).
2. The solution of the problem (3.5.3) is called weak solution of the boundary problem
(3.5.1)–(3.5.2). ♦
Any regular solution (i.e., at least in. H 2 (Ω)) of the boundary problem (3.5.1)–(3.5.2)
is solution of the variational problem (3.5.3). Now, what about the reverse implica-
tion? Further, does a solution of the variational problem solve the boundary value
problem? Indeed, the answer of these questions is basically yes, the two problems
are equivalent. In fact.

3.5.2 Equivalence Between Variational Formulation

and Strong Problem

Let .u ∈ H01 (Ω) be a solution of the problem (3.5.3). We have .D(Ω) ⊂ H01 (Ω).
Then, we can take .v = ϕ ∈ D(Ω) as test function in Eq. (3.5.3). Let’s look at each
term separately. Then, for all .ϕ ∈ D(Ω), we have
n {
E {
∂u ∂ϕ
. dx = f ϕ d x.
i=1 Ω ∂ xi ∂ xi Ω

@seismicisolation
@seismicisolation
3.5 Dirichlet Problem 119

Thus,
n /
E \
∂u ∂ϕ
. , = ( f, ϕ).
i=1
∂ xi ∂ xi

It follows that
n / 2
E \
∂ u
. − , ϕ = ( f, ϕ).
i=1
∂ xi2

Therefore, / \
E
n
∂ 2u
. − , ϕ = ( f, ϕ).
i=1
∂ xi2

Hence, / n \
E ∂ 2u
. + f, ϕ = 0.
i=1
∂ xi2

Again,
. (Δu + f, ϕ) = 0 for all ϕ ∈ D(Ω).

We deduce that
. − Δu = f

within the meaning of distributions on .Ω (i.e., on .D , (Ω)). Since . f ∈ L 2 (Ω, R), the
relation
. − Δu = f

is also true in . L 2 (Ω, R) (see (2.18.5)). Hence,

. − Δu = f

almost everywhere in .Ω. So, (3.5.1). On the other hand, the fact that .u ∈ H01 (Ω)
gives
.u |Γ = γ0 (u) = 0

on . L 2 (Γ, R). Thus, .u = 0 almost everywhere on .Γ . From where (3.5.2).

Remark 3.5.2 If .n ≤ 3 and if .u solution of the problem (3.5.2) belongs to . H 2 (Ω),

then .u = 0 on .Γ . In fact, according to Theorem 3.4.8 and since .2 > 23 , we have

. H 2 (Ω) ⊂ C 0 (Ω, R).

The fact that .u ∈ H 2 (Ω) gives .u is continuous. Since .u = 0 almost everywhere on

.Γ gives .u = 0 everywhere on .Γ . ♦

@seismicisolation
@seismicisolation
120 3 Variational Formulation of Boundary Problems

3.5.3 Existence and Uniqueness

To show the existence and the uniqueness of the solution of (3.5.1)–(3.5.2), it suffices
to show the existence and the uniqueness of the problem (3.5.3). In fact, we consider
the following symmetrical bilinear form:

.a : H01 (Ω) × H01 (Ω) −→ R

n {
E ∂u ∂v
(u, v) −→ a(u, v) = dx
i=1 Ω ∂ xi ∂ xi

and the following linear form

. L : H01 (Ω) −→ R
{
v −→ L(v) = f v d x.
Ω

3.5.3.1 Method 1: Norm of . H 1 (Ω)

We provide . H01 (Ω) with the following inner product:

n {
E ∂u ∂v
(u, v)1,Ω := (u, v)0,Ω +
. d x.
i=1 Ω ∂ xi ∂ xi

Continuity of a(·, ·): Let .u, .v ∈ H01 (Ω). Then,

.

|a(u, v)| ≤ |u|1,Ω |v|1,Ω

.

≤ ||u||1,Ω ||v||1,Ω .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ H01 (Ω). Then,
.

. a(u, u) = |u|21,Ω .

Since .| · |1,Ω and .|| · ||1,Ω are two equivalent norms in . H01 (Ω) (see Corollary 3.2.5
.(ii)), there is .α > 0 such that

|u|1,Ω ≥ α||u||1,Ω
.

for all .u ∈ H01 (Ω). Hence,

a(u, u) ≥ α 2 ||u||21,Ω .
.

@seismicisolation
@seismicisolation
3.5 Dirichlet Problem 121

Thus, .a(·, ·) is coercive.

Continuity of L(·): Let .v ∈ H01 (Ω). Then,
.

{
. |L(v)| ≤ | f | |ϕ| d x
Ω
≤ || f ||0,Ω ||v||0,Ω
≤ || f ||0,Ω ||v||1,Ω .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.5.3)
admits one solution and only one .u ∈ H01 (Ω).

3.5.3.2 Method 2: Semi-norm of . H 1 (Ω)

We provide . H01 (Ω) with the following inner product

n {
E ∂u ∂v
(u, v) :=
. d x.
i=1 Ω ∂ xi ∂ xi

Continuity of a(·, ·): Let .u, .v ∈ H01 (Ω). Then,

.

. |a(u, v)| = |(u, v)|

≤ |u|1,Ω |v|1,Ω .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ H01 (Ω). Then,
.

a(u, u) = (u, u)
.

= |u|21,Ω
1
≥ |u|21,Ω .
2
Hence, .a(·, ·) is coercive.
Continuity of L(·): Let .v ∈ H01 (Ω). Then,
.

{
|L(v)| ≤
. | f | |v| d x
Ω
≤ || f ||0,Ω ||v||0,Ω .

Since .v ∈ H01 (Ω), according to the Poincaré inequality (Theorem 3.2.2), we have

@seismicisolation
@seismicisolation
122 3 Variational Formulation of Boundary Problems

||v||0,Ω ≤ C(Ω)|v|1,Ω .
.

Hence,
|L(v)| ≤ C(Ω)|| f ||0,Ω |v|1,Ω .
.

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.5.3)
admits one solution and only one .u ∈ H01 (Ω).
Remark 3.5.3 The inhomogeneous Dirichlet problem is another problem of interest

. − Δu + cu = f in Ω (3.5.4)
.u = g on Γ = ∂Ω, (3.5.5)
1
with.g ∈ H 2 (∂Ω). The problem (3.5.4)–(3.5.5) is reduced to the homogeneous prob-
lem by taking a function .G ∈ H 1 (Ω) such that .G |∂Ω = g. Let .U = u − G. Then, it
is clear to see that .U ∈ H01 (Ω) and

. − ΔU + cU = −Δu + cu + ΔG − cG
= f + ΔG − cG.

So, we have just written the variational formulation of the homogeneous problem
for .U with right-hand side . F = f + ΔG − cG. ♦

3.5.4 Dirichlet’s Principle for the Laplacian Operator

Consider the Dirichlet problem for the Poisson equation in an open bounded set
Ω ⊂ Rn ,
.

. − Δu = f in Ω (3.5.6)
.u = g on Γ = ∂Ω. (3.5.7)

We also consider the functional

{ {
1
. J (v) = |∇v|2 d x − f v d x, (3.5.8)
2 Ω Ω

where .| · | denotes the Euclidean norm in .Rn . The functional . J and the boundary
value problem (3.5.6)–(3.5.7) are related as follows. Assume that . f : Ω −→ R and
. g : ∂Ω −→ R are continuous, let

. K = {v ∈ C 2 (Ω, R), v(x) = g(x) for all x ∈ ∂Ω}. (3.5.9)

@seismicisolation
@seismicisolation
3.5 Dirichlet Problem 123

Let .u be a minimizer of the functional . J on . K , that is,

u ∈ K , J (u) = min J (v).

. (3.5.10)
v∈K

Let .ϕ ∈ D(Ω) be given. We have .u + λϕ ∈ K for all .λ ∈ R. Consider

. J- : R −→ R, J-(λ) = J (u + λϕ).

Then,
{ {
1
. J-(λ) = |∇(u + λϕ)|2 d x − f (u + λϕ) d x
2 Ω Ω
{ { {
λ2
= J (u) + λ ∇u · ∇ϕ d x − λ f ϕ dx + |∇ϕ|2 d x.
Ω Ω 2 Ω

Since .0 is a minimizer of the differentiable function . J-, then

{ {
-,
.0 = J (0) = ∇u · ∇ϕ d x − f ϕ dx
{Ω Ω

= (−Δu − f )ϕ d x,
Ω

(here we have used Green formula and the fact that .ϕ = 0 on .∂Ω). Since .ϕ ∈ D(Ω)
can be chosen arbitrarily, it follows that (we will prove this later)

. − Δu(x) = f (x) for all x ∈ Ω.

Thus, every minimizer of the functional . J on . K solves the problem (3.5.6)–(3.5.7).

Conversely, let .u ∈ C 2 (Ω, R) be a solution of (3.5.6)–(3.5.7). Then, .u ∈ K . Let
.v ∈ K be arbitrary. Then,

{
. (−Δu − f )(u − v) d x = 0.
Ω

Since .u − v = 0 on .∂Ω, Green formula gives

{ {
. ∇u · (∇u − ∇v) d x − f (u − v) d x = 0,
Ω Ω

where .∇u · ∇v is the Euclidean scalar product in .Rn of .∇u and .∇v. It follows that
{ { { {
.|∇u| dx − f u dx = ∇u · ∇v d x −
2
f v dx
Ω Ω Ω Ω
{ { {
1 1
≤ |∇u|2 d x + |∇v|2 d x − f v d x. (3.5.11)
2 Ω 2 Ω Ω

@seismicisolation
@seismicisolation
124 3 Variational Formulation of Boundary Problems
{
Subtracting . 21 Ω |∇u|2 d x on both sides of (3.5.11), we arrive at

. J (u) ≤ J (v).

Hence, .u is a minimizer of the functional . J on . K . The equivalence .u solves the

problem (3.5.6)–(3.5.7) if, and only if, .u is a minimizer of the functional . J on . K is
called Dirichlet’s principle. In this context, the equation

. − Δu = f

is called the Euler equation or Euler-Lagrange equation corresponding to the func-

tional . J in (3.5.8).
If we set { {
a(u, v) =
. ∇u · ∇v d x, F(v) = f v d x,
Ω Ω

the situation (3.5.6), (3.5.7), (3.5.8), (3.5.9), and (3.5.10), which we considered at the
beginning, becomes a special case of (2.16.6). The variational equation then reads
{ {
. ∇u · ∇v d x = f v d x, for all v in U. (3.5.12)
Ω Ω

So far we did not specify the space . X and the subspace .U . If one starts from the
Poisson equation
. − Δu = f in Ω, (3.5.13)

one may arrive at the variational formulation (3.5.12), immediately, by multiplying

(3.5.13 ) on both sides with functions .v which are zero on .∂Ω, and performing Green
formula. In this manner, one bypasses the functional . J .

3.5.5 First Example

We consider the problem

d 2u
. − + u = f if x ∈]0, 1[ (3.5.14)
dx2
.u(0) = 0 and u(1) = 0 (3.5.15)

with . f ∈ L 2 (]0, 1[, R).

@seismicisolation
@seismicisolation
3.5 Dirichlet Problem 125

3.5.5.1 Variational Formulation

We notice in the boundary conditions of the problem (3.5.14)–(3.5.15), there is two

conditions on .u (.u(0) = 0 and .u(1) = 0) and no condition on the derivative of .u.
Hence, according to the start of this chapter the appropriate Hilbert space is

. V := {v ∈ H 1 (0, 1) such that v(0) = v(1) = 0} = H01 (]0, 1[).

Let .v ∈ H01 (]0, 1[) be a test function. Multiply Eq. (3.5.14) by .v,

d 2u
. − v + uv = f v
dx2
and integrate over .]0, 1[,
{ 1 { 1 { 1
d 2u
. − v dx + uv d x = f v d x.
0 dx2 0 0

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv
. d x − [u , v]10 + uv d x = f v d x,
0 dx dx 0 0

which proves
{ 1 { 1 { 1
du dv
. d x − u , (1)v(1) + u , (0)v(0) + uv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ H01 (]0, 1[) such that for all .v ∈ H01
(]0, 1[), we have { 1 { 1 { 1
du dv
. dx + uv d x = f v d x. (3.5.16)
0 dx dx 0 0

3.5.5.2 Equivalence Between Variational Formulation and Strong

Problem

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + uϕ d x = f ϕ d x.
0 dx dx 0 0

This implies that

@seismicisolation
@seismicisolation
126 3 Variational Formulation of Boundary Problems
{ 1 { 1 { 1
d 2u
. − ϕ dx + uϕ d x = f ϕ d x.
0 dx2 0 0

Thus, { ( 2 )
1
d u
. − 2 + u − f ϕ d x = 0.
0 dx

So, / \
d 2u
. − 2 + u − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx

It follows that
d 2u
. − +u− f =0
dx2

on .D , (]0, 1[). Moreover, . f ∈ L 2 (]0, 1[, R), so

d 2u
. − +u = f
dx2

on . L 2 (]0, 1[, R) (see (2.18.5)). It thus appears that

d 2u
. − +u = f
dx2

almost everywhere in .]0, 1[. Further, .u ∈ H01 (]0, 1[), so

u(0) = u(1) = 0.
.

Thus, the equivalence between the strong problem and the variational formulation.

3.5.5.3 Existence and Uniqueness

We consider the following symmetrical bilinear form

a : V × V −→ R
.
{ {
1
∂u ∂v 1
(u, v) −→ a(u, v) = dx + uv d x
0 ∂x ∂x 0

and the following linear form

@seismicisolation
@seismicisolation
3.5 Dirichlet Problem 127

. L : V −→ R
{ 1
v −→ L(v) = f v d x.
0

. H01 (]0, 1[) is a Hilbert space.

Continuity of a(·, ·): Let .u, .v ∈ H01 (]0, 1[). Then,
.

. |a(u, v)| = |(u, v)1,]0,1[ |

≤ ||u||1,]0,1[ ||v||1,]0,1[ .

Hence, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ H01 (]0, 1[). Then,
.

a(u, u) = (u, u)1,]0,1[

.

= ||u||21,]0,1[
1
≥ ||u||21,]0,1[ .
2
So, .a(·, ·) is coercive.
Continuity of L(·): Let .v ∈ H01 (]0, 1[). Then,
.

{ 1
.|L(v)| ≤ | f | |ϕ| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.5.16)
admits one solution and only one .u ∈ H01 (]0, 1[).

3.5.6 Second Example

We consider the weak formulation: Find .u ∈ H01 (]0, 1[) such that

a(u, v) = L(v)
. (3.5.17)

for all .v ∈ H01 (]0, 1[) with

@seismicisolation
@seismicisolation
128 3 Variational Formulation of Boundary Problems
{ 1 { 1
, ,
. a(u, v) = u v dx + u , v d x,
0 0

{ 1
. L(v) = f v dx
0

and . f ∈ L 2 (]0, 1[, R).

Proposition 3.5.4 .a(·, ·) is a continuous bilinear form on . H01 (]0, 1[). ♦
Proof We have .a(·, ·) is a bilinear form.
Continuity of a(·, ·): Let.u,.v ∈ H01 (]0, 1[), according to Cauchy-Schwarz inequality
.
and the Poincaré inequality (Theorem 3.2.2), we have
{ 1 { 1
, ,
|a(u, v)| ≤
. |u ||v | d x + |u , ||v| d x
0 0
≤ |u|1,]0,1[ |v|1,]0,1[ + |u|1,]0,1[ ||v||0,]0,1[
≤ (1 + C(]0, 1[))|u|1,]0,1[ |v|1,]0,1[ .

Thus, .a(·, ·) is continuous on . H01 (]0, 1[). Q.E.D.

Proposition 3.5.5 For all .u ∈ H01 (]0, 1[), we have

{ 1
. u , u d x = 0.
0

♦
Proof For .u ∈ H01 (]0, 1[), we have
{ 1 { 1
. u,u d x = − uu , d x + [u 2 ]10 .
0 0

Hence, { 1
. 2 u , u d x = u 2 (1) − u 2 (0) = 0.
0

Thus, { 1
. u , u d x = 0.
0

This completes the proof. Q.E.D.

Proposition 3.5.6 .a(·, ·) is coercive. ♦

@seismicisolation
@seismicisolation
3.5 Dirichlet Problem 129

Proof Let .u ∈ H01 (]0, 1[), we have

{ 1 { 1
. a(u, u) = u ,2 d x + u,u d x
0 0
= |u|21,]0,1[
1
≥ |u|21,]0,1[ .
2
So, .a(·, ·) is coercive. Q.E.D.

Proposition 3.5.7 The problem (3.5.17) admits a unique solution .u ∈ H01 (]0, 1[).♦

Proof . L(·) is a linear form.

Continuity of L(·): Let.u ∈ H01 (]0, 1[), according to Cauchy-Schwarz inequality and
.

the Poincaré inequality (Theorem 3.2.2), we have

{ 1
|L(v)| ≤
. | f | |ϕ| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ C(]0, 1[)|v|1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.5.17)
admits one solution and only one .u ∈ H01 (]0, 1[). Q.E.D.

Theorem 3.5.8 .u verify the following boundary problem:

⎧
⎨ d 2u
− 2
+ u , = f if x ∈]0, 1[
. d x
⎩ u(0) = u(1) = 0.

Proof Let .ϕ ∈ D(]0, 1[). We have

{ 1 { 1 { 1
. u,ϕ,d x + u,ϕ d x = f ϕ d x.
0 0 0

Hence, { { {
1 1 1
,, ,
. − u ϕ dx + u ϕ dx = f ϕ d x.
0 0 0

@seismicisolation
@seismicisolation
130 3 Variational Formulation of Boundary Problems

This proves
{ 1
. (−u ,, + u , − f )ϕ d x = 0
0

for all .ϕ ∈ D(]0, 1[). So,

( )
. −u ,, + u , − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).

Thus,
. − u ,, + u , − f = 0

on .D , (]0, 1[). Furthermore, . f ∈ L 2 (]0, 1[, R), so .−u ,, + u , = f on . L 2 (]0, 1[, R).
It thus appears that
,, ,
.−u +u = f

almost everywhere in .]0, 1[. We have .u ∈ H01 (]0, 1[), so

u(0) = u(1) = 0.
.

Hence, .u verify the boundary problem

⎧
⎨ d 2u
− + u , = f if x ∈]0, 1[
. dx2
⎩ u(0) = u(1) = 0.

This finishes the proof. Q.E.D.

3.6 Neumann Problem

3.6.1 Homogeneous Neumann Boundary Condition

Let .Ω be an open bounded domain of .Rn with piecewise smooth boundary .Γ = ∂Ω.
Given . f ∈ L 2 (Ω, R), we find a solution .u defined on .Ω to the problem

. − Δu + u = f in Ω (3.6.1)
∂u
. = 0 on Γ = ∂Ω, (3.6.2)
∂ν |Γ

with
∂u E ∂u
n
. = νi
∂ν |Γ i=1
∂ xi |Γ

and .νi ∈ L ∞ (Γ, R), .1 ≤ i ≤ n.

@seismicisolation
@seismicisolation
3.6 Neumann Problem 131

3.6.1.1 Variational Formulation

Suppose that the solution of the problem (3.6.1) and (3.6.2) is sufficiently regular, for
example, .u ∈ H 2 (Ω). Note that in the boundary conditions of the problem (3.6.1)–
(3.6.2), there are no conditions on .u, there is only one condition normal derivative
of .u. Hence, the appropriate Hilbert space is .V := H 1 (Ω). Let .v ∈ V := H 1 (Ω) be
a test function. Multiply Eq. (3.6.1) by .v,

. − Δu v + uv = f v

and integrate over .Ω,

{ {
. (−Δu + u)v d x = f v d x.
Ω Ω

Thus, { { {
. − Δu v d x + uv d x = f v d x.
Ω Ω Ω

According to Green’s formula, we have

n {
E { { {
∂u ∂v ∂u
. dx − v dσ + uv d x = f v d x.
i=1 Ω ∂ xi ∂ xi Γ ∂ν Ω Ω

We replace the problem (3.6.1)–(3.6.2) by the following variational problem:

Find .u ∈ H 1 (Ω) such that for all .v ∈ H 1 (Ω), we have
n {
E { {
∂u ∂v
. dx + uv d x = f v d x. (3.6.3)
i=1 Ω ∂ xi ∂ xi Ω Ω

3.6.1.2 Equivalence Between Variational Formulation and Strong

Problem

Let .u ∈ H 1 (Ω) be the solution of the problem (3.6.3). Since .D(Ω) ⊂ H 1 (Ω), then
we can take .v = ϕ ∈ D(Ω) as test function in Eq. (3.6.3). Let’s look at each term
separately. Then, for all .ϕ ∈ D(Ω), we have
n {
E { {
∂u ∂ϕ
. dx + uϕ d x = f ϕ d x.
i=1 Ω ∂ xi ∂ xi Ω Ω

@seismicisolation
@seismicisolation
132 3 Variational Formulation of Boundary Problems

According to Theorem 3.3.5, we deduce

En { { {
∂ 2u
. − ϕ d x + uϕ d x = f ϕ d x,
i=1 Ω
∂ xi2 Ω Ω

which proves {
. (−Δu + u − f )ϕ d x = 0
Ω

for all .ϕ ∈ D(Ω). So,

. (−Δ + u − f, ϕ) = 0 for all ϕ ∈ D(Ω).

Thus,
. − Δu + u = f

on .D , (Ω). Since . f , .u ∈ L 2 (Ω, R), and, using (2.18.5), we have

. − Δu + u = f

on . L 2 (Ω, R). This implies

. − Δu + u = f

almost everywhere in .Ω. Thus, (3.6.1).

Let .u ∈ H 1 (Ω) be the solution of the problem (3.6.3). Suppose that .u ∈ H 2 (Ω).
We have
En { { {
∂u ∂v
. dx + uv d x = f v d x.
i=1 Ω
∂ xi ∂ xi Ω Ω

Using Theorem 3.4.4, we deduce for .v ∈ H 1 (Ω),

{ { { {
∂u
. −Δu v d x + v dσ + uv d x = f v d x.
Ω Γ ∂ν Ω Ω

Further,
. − Δu + u = f

on . L 2 (Ω, R). Hence, for all .v ∈ H 1 (Ω), we have

{ {
∂u ∂u
. v dσ = γ0 (v) dσ = 0.
Γ ∂ν Γ ∂ν

Consider

@seismicisolation
@seismicisolation
3.6 Neumann Problem 133

γ : H 1 (Ω) −→ L 2 (Γ, R)
. 0

v −→ γ0 (v) = v|Γ .
1
It is known that . I m(γ0 ) = H 2 (Γ ) c L 2 (Γ, R) (see [8, 9]) and . I m(γ0 ) is dense in
. L (Γ, R). The density of . I m(γ0 ) on . L (Γ, R) entails
2 2

{
∂u
. w dσ = 0 (3.6.4)
Γ ∂ν
∂u
for all .w ∈ L 2 (Γ, R). Taking .w = ∂ν
in (3.6.4), we obtain
{ ( )2
∂u
. dσ = 0,
Γ ∂ν

which implies
∂u
. =0
∂ν

on . L 2 (Γ, R). So,

∂u
. =0
∂ν
almost everywhere in .Γ . Thus, (3.6.2).

3.6.1.3 Existence and Uniqueness

Theorem 3.6.1 The weak problem (3.6.3) admits one solution and only one .u on
H 1 (Ω).
. ♦

Proof We consider the following symmetrical bilinear form:

a : H 1 (Ω) × H 1 (Ω) −→ R
.
n {
E {
∂u ∂v
(u, v) −→ a(u, v) = dx + uv d x
i=1 Ω ∂ xi ∂ xi Ω

and the following linear form:

. L : H 1 (Ω) −→ R
{
v −→ L(v) = f v d x.
Ω

@seismicisolation
@seismicisolation
134 3 Variational Formulation of Boundary Problems

Continuity of a(·, ·): Let .u, .v ∈ H 1 (Ω). Then,

.

|a(u, v)| = |(u, v)1,Ω |

.

≤ ||u||1,Ω ||v||1,Ω .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ H 1 (Ω). Then,
.

a(u, u) = (u, u)1,Ω

.

= ||u||21,Ω
1
≥ ||u||21,Ω .
2
Thus, .a(·, ·) is coercive.
Continuity of L(·): Let .v ∈ H 1 (Ω). Then,
.

{
|L(v)| ≤
. | f | |v| d x
Ω
≤ || f ||0,Ω ||v||0,Ω
≤ || f ||0,Ω ||v||1,Ω .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.6.3)
admits one solution and only one .u ∈ H 1 (Ω). Q.E.D.

3.6.2 Mixed Problem

We can mixed together the Dirichlet and Neumann conditions, so-called mixed prob-
lem, but not at the same place uMore precisely, let .Γ1 and .Γ2 be two
N on the boundary.
subsets of .∂Ω such that .Γ1 Γ2 = ∅ and .Γ1 Γ2 = ∂Ω. Consider the problem

. − Δu + cu = f in Ω (3.6.5)
.u = g1 on Γ1 , (3.6.6)
∂u
. = g2 on Γ2 . (3.6.7)
∂ν |Γ

The variational formulation for the mixed problem (3.6.5)–(3.6.7) (in the case .g1 =
0 for brevity, if not follow the above route) is to: Find .u in the space .V := {v ∈
H 1 (Ω) such that v|Γ1 = 0 on Γ1 } such that

@seismicisolation
@seismicisolation
3.6 Neumann Problem 135
{ { {
. (∇u · ∇v d x + cuv) d x = f v dx + g2 v dσ,
Ω Ω Γ2

for all .v ∈ V .

3.6.3 Neumann Boundary Condition

Let .Ω be an open bounded domain of .Rn with piecewise smooth boundary .Γ = ∂Ω.
Given . f ∈ L 2 (Ω, R), we find a solution .u defined on .Ω to the problem

. − Δu + cu = f in Ω (3.6.8)
∂u
. = g on Γ = ∂Ω, (3.6.9)
∂ν |Γ

with
∂u E ∂u n
. = νi
∂ν |Γ i=1
∂ xi |Γ

and .νi ∈ L ∞ (Γ, R), .1 ≤ i ≤ n. This is an example of a Neumann boundary condi-

tion. When g = 0, we naturally called a homogeneous Neumann boundary condition.
In modeling terms, the Neumann condition is a flow condition. For example, in the
interpretation of thermal equilibrium, the condition corresponds to an imposed heat
flux across the boundary, as opposed to the Dirichlet condition which imposes a tem-
perature on the boundary. The case .g = 0 corresponds to perfect thermal insulation.

3.6.3.1 Variational Formulation

Let us derive the variational formulation informally. Assume first that .u ∈ H 2 (Ω),
take .v ∈ H 1 (Ω), multiply, integrate and use Green’s formula to obtain

. − Δu v + cuv = f v

and integrate over .Ω,

{ {
. (−Δu + cu)v d x = f v d x.
Ω Ω

Thus, { { {
. − Δu v d x + cuv d x = f v d x.
Ω Ω Ω

According to Green’s formula, we have

@seismicisolation
@seismicisolation
136 3 Variational Formulation of Boundary Problems
{ { { {
∂u
. ∇u · ∇v d x − v dσ + cuv d x = f v d x.
Ω Γ ∂ν Ω Ω

We replace the problem (3.6.8)–(3.6.9) by the following variational problem:

Find .u ∈ H 1 (Ω) such that for all .v ∈ H 1 (Ω), we have
{ { { {
. ∇u · ∇v d x + cuv d x = f v dx + gv|∂Ω dσ. (3.6.10)
Ω Ω Ω ∂Ω

3.6.3.2 Equivalence Between Variational Formulation and Strong

Problem

The converse is more interesting. Let .u ∈ H 2 (Ω) be a solution of the above varia-
tional problem (3.6.10). Taking first .v = ϕ ∈ D(Ω), we obtain

. − Δu + cu = f in the sense of D , (Ω)

exactly as in the Dirichlet case. Of course, a test function with compact support
does not see what happens on the boundary, and no information on the Neumann
condition is recovered. Thus, in a second step, we take .v arbitrary in . H 1 (Ω). By
Green’s formula again, we have
{ { {
∂u
. ∇u · ∇v d x = − Δuv d x + v dσ.
Ω Ω Γ ∂ν

Since .u is a solution of the variational problem, it follows that

{ { { {
∂u
. (−Δu + cu)v d x + v dσ = f v dx + gv|∂Ω dσ.
Ω Γ ∂ν Ω ∂Ω

But we already know that

{ {
. (−Δu + cu)v d x = f v d x,
Ω Ω

hence we are left with { {

∂u
. v dσ = gv|∂Ω dσ,
Γ ∂ν ∂Ω

for all .v ∈ H 1 (Ω). For simplicity, we assume here the following: the function .g ∈
1
H 2 (∂Ω), the range of the trace mapping .γ0 and the domain .Ω is smooth. Since
∂u 1
.u ∈ H (Ω), then it follows that . ∈ H 2 (∂Ω). Hence, there is .v ∈ H 1 (Ω) such
2
∂ν
that .v|∂Ω = ∂u
∂ν
− g. With this choice of the function .v, we obtain the following:

@seismicisolation
@seismicisolation
3.6 Neumann Problem 137

{ ( )2
∂u
. −g dσ = 0.
∂Ω ∂ν

Thus,
∂u
. =g
∂ν
which is the Neumann condition.

Remark 3.6.2 Consider the Neumann problem

. − Δu = f in Ω (3.6.11)
∂u
. = g on Γ = ∂Ω, (3.6.12)
∂ν |Γ

in a Lipschitz open set .Ω in .Rn .

(i) We see right away that things are going to be different since we don’t have
.
uniqueness here. In fact, if .u is a solution of the Neumann problem (3.6.11)–(3.6.12),
then .u + s is again a solution for any constant .s. Assume first that .u ∈ H 2 (Ω) and
take .v ∈ H 1 (Ω). By multiplying, integrating and using Green’s formula, we obtain

. − Δu v = f v.

By integrating over .Ω, we obtain

{ {
. −Δuv d x = f v d x.
Ω Ω

According to Green’s formula, we have

{ { {
∂u
. ∇u · ∇v d x − v dσ = f v d x.
Ω Γ ∂ν Ω

Thus, { { {
. ∇u · ∇v d x − gv dσ = f v d x.
Ω Γ Ω

Furthermore, for .v = 1, then, necessarily

{ {
. f dx + g dσ = 0. (3.6.13)
Ω ∂Ω

There is no solution, if the data . f , .g does not satisfy the compatibility condition
(3.6.13). Non-uniqueness and non-existence are in fact dual to each other. There are
several ways to circumvent the two problems, therefore several variational formula-
tions. We choose the space .V as follows

@seismicisolation
@seismicisolation
138 3 Variational Formulation of Boundary Problems
{ { }
. V = v ∈ H 1 (Ω) such that v dx = 0 .
Ω

Since .Ω is a bounded set, then this is well defined, and we thus have the inclusions
. H 1 (Ω) ⊂ L 2 (Ω, R) ⊂ L 1 (Ω, R). It is easy to see that the space.V is a hyperplane of
1 2
. H that is . L -orthogonal to the one-dimensional space of constant functions, which

are the cause of non-uniqueness of solution.

(ii) The space .V is closed. Indeed, let .(vn )n be a sequence in .V such that .vn → v in
.
H 1 (Ω). Then, .vn → v in . L 2 (Ω, R). In view of Cauchy-Schwarz, we have .vn → v
.

in . L 1 (Ω, R). Hence, { {

.0= vn d x → v dx
Ω Ω

and .v ∈ V . Thus, since .V is closed, then .V is a Hilbert space for the scalar product
of the Sobolev space . H 1 (Ω).
(iii) Assume that . f ∈ L 2 (Ω, R), .g ∈ L 2 (∂Ω) satisfy the compatibility condition
.
(3.6.13). Then, the triple
{ { {
. V, a(u, v) = ∇u · ∇v d x, L(v) = f v dx + gv|∂Ω dσ
Ω Ω ∂Ω

defines a variational formulation for the Neumann problem (3.6.11)–(3.6.12). Indeed,

by multiplying the partial differential equations PDE by .v ∈ V , by integrating over
.Ω, and by using Green’s formula, we easily see that if .u solves problem (3.6.11)–

(3.6.12), then .a(u, v) = L(v) for all .v ∈ V . Conversely, let .u ∈ V be a function such
that for all .v ∈ V , .a(u, v) = L(v). We would like to proceed as before. By taking
.v ∈ D(Ω), we deduce the partial differential equation PDE. It doesn’t work here

because .D(Ω) /⊂ V . Let .ϕ ∈ D(Ω), we set

{
1
.ψ = ϕ − ϕ(x) d x.
mes(Ω) Ω

Then, .ψ ∈ V . Thus, we can use .ψ as a test function. Since both .ϕ and .ψ differ by a
constant {
1
.k = ϕ(x) d x,
mes(Ω) Ω

then
∇ψ = ∇ϕ.
.

So, since . f , .g satisfy condition (3.6.13) and .ϕ vanishes on .∂Ω, then

@seismicisolation
@seismicisolation
3.6 Neumann Problem 139
{ { {
. ∇u · ∇ϕ d x = f ψ dx + gψ dσ
Ω
{Ω { ∂Ω

= f (ϕ + k) d x + g(ϕ + k) dσ
{Ω ({ ∂Ω { )
= f ϕ dx + k f dx + g dσ
{Ω Ω ∂Ω

= f ϕ d x.
Ω

Thus, we immediately deduce that .−Δu = f in the sense of distributions, and since
f ∈ L 2 (Ω, R) in the sense of . L 2 (Ω, R) as well. We then choose an arbitrary .v ∈ V
.
and apply Green’s formula again. This gives
{ { { {
∂u
. f v dx + gv dσ = − Δuv d x + v dσ.
Ω ∂Ω Ω ∂Ω ∂ν

Therefore, taking into account that .−Δu = f , we get

{ ( )
∂u
. g− v|∂Ω dσ = 0
∂Ω ∂ν
1
for all .v ∈ V . Now, it is not difficult to see that .γ0 (V ) = H 2 (∂Ω), where .γ0 is the
trace mapping. In fact, choose one .θ ∈ D(Ω) such that
{
. θ d x = 1.
Ω

Then, for all .w ∈ H 1 (Ω),

{
. v=w− w dx θ ∈ V
Ω

and
v
. |∂Ω = w|∂Ω .
1
So, since . H 2 (∂Ω) is dense in . L 2 (∂Ω) (see [8, 9]), then there are enough test func-
tions in .V to ensures that
∂u
. = g.
∂ν
(iv) The problem: Find .u ∈ V such that
.

. a(u, v) = L(v), ∀v ∈ V (3.6.14)

@seismicisolation
@seismicisolation
140 3 Variational Formulation of Boundary Problems

has one and only one solution. We check the hypotheses of the Lax-Milgram theorem
(see Theorem 2.17.1). In fact, we have already proven that .V is a Hilbert space for
the . H 1 scalar product.
Continuity of a(·, ·): Let .u, .v ∈ V . Then,
.

|{ |
| |
.|a(u, v)| = | |
| ∇u · ∇v d x |
Ω
{
≤ |∇u||∇v| d x
Ω
≤ |u|1,Ω |v|1,Ω
≤ ||u||1,Ω ||v||1,Ω .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): For all .v ∈ V , we get
.

{
. v d x = 0.
Ω

Thus, by using the Poincaré-Wirtinger inequality (see Theorem 3.2.3), we have

||v||21,Ω = ||v||20,Ω + ||∇v||20,Ω

.

≤ (C 2 + 1)||∇v||20,Ω .

Therefore,
1
a(u, u) = ||∇u||20,Ω ≥
. ||u||21,Ω .
C2 + 1

Thus, .a(·, ·) is coercive.

Continuity of L(·): Let .v ∈ H 1 (Ω). Then,
.

{ {
|L(v)| ≤
. | f | |v| d x + |g||v|∂Ω | dσ
Ω ∂Ω
≤ || f ||0,Ω ||v||0,Ω + ||g||0,∂Ω ||v|∂Ω ||0,∂Ω
≤ || f ||0,Ω ||v||1,Ω + C||g||0,∂Ω ||v||1,Ω
≤ (|| f ||0,Ω + C||g||0,∂Ω )||v||1,Ω .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.6.14)
has one solution and only one .u ∈ V .

@seismicisolation
@seismicisolation
3.6 Neumann Problem 141

(v) We have not insisted on the regularity necessary to apply Green’s formula or to
.

define. ∂u
1

∂ν
as an element of. H 2 (∂Ω), because it is possible to write down slightly more
complicated arguments that completely do away with such artificial hypotheses.
(vi) The compatibility condition, given in (3.6.13), plays no role in the application
.
of the Lax-Milgram theorem (see Theorem 2.17.1).
.(vii) Since the space .V is . L 2 -orthogonal to the one-dimensional space of constant
functions, then the general solution of the Neumann problem (3.6.11)–(3.6.12) is
of the form .u + s, where .u ∈ V is the unique solution of the problem (3.6.14) and
.s ∈ R is arbitrary. ♦

3.6.4 First Example

We consider the problem

d 2u
. − + u = f if x ∈]0, 1[ (3.6.15)
dx2
,
.u (0) = u(1) = 0 (3.6.16)

with . f ∈ L 2 (]0, 1[, R). This problem corresponds to a metal bar heated by means of
an amount of heat. f . The function.u representing temperature. We fix the temperature
,
.u = 0 in . x = 0 and we assume that the heat flow .u = 0 in . x = 1.

3.6.4.1 Variational Formulation

We notice in the boundary conditions of the problem (3.6.15)–(3.6.16), there is a

condition on .u (.u(1) = 0) and a condition on the derivative of .u (.u , (0) = 0). Hence,
according to the start of this chapter the appropriate Hilbert space is

. V := {v ∈ H 1 (0, 1) such that v(1) = 0}.

Let .v ∈ V be a test function. Multiply Eq. (3.6.15) by .v,

d 2u
. − v + uv = f v
dx2
and integrate over .]0, 1[,
{ 1 { 1 { 1
d 2u
. − v dx + uv d x = f v d x.
0 dx2 0 0

Using Theorem 3.4.4, we deduce

@seismicisolation
@seismicisolation
142 3 Variational Formulation of Boundary Problems
{ 1 { 1 { 1
du dv
. d x − [u , v]10 + uv d x = f v d x,
0 dx dx 0 0

which proves
{ 1 { 1 { 1
du dv
. d x − u , (1)v(1) + u , (0)v(0) + uv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. dx + uv d x = f v d x. (3.6.17)
0 dx dx 0 0

3.6.4.2 Equivalence Between Variational Formulation and Strong

Problem

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + uϕ d x = f ϕ d x.
0 dx dx 0 0

This implies that

{ 1 { 1 { 1
d 2u
. − ϕ dx + uϕ d x = f ϕ d x.
0 dx2 0 0

Thus, { ( 2 )
1
d u
. − 2 + u − f ϕ d x = 0.
0 dx

So, / \
d 2u
. − 2 + u − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx

It follows that
d 2u
. − +u− f =0
dx2

on .D , (]0, 1[). Moreover, . f ∈ L 2 (]0, 1[, R), so

d 2u
. − +u = f
dx2

on . L 2 (]0, 1[, R) (see (2.18.5)). It thus appears that

@seismicisolation
@seismicisolation
3.6 Neumann Problem 143

d 2u
. − +u = f
dx2
almost everywhere in .]0, 1[. Further, .u ∈ V , so

. u(1) = 0.

It remains to show that .u , (0) = 0. In fact, let .v ∈ V , using Green’s formula to the
variational problem, we find
{ 1 { 1 { 1
d 2u
. − v d x + [u , v]10 + uv d x = f v d x.
0 dx2 0 0

Furthermore,
d 2u
. − + u = f,
dx2
hence
u , (0)v(0) = 0
.

for all .v ∈ V . We choose a .v ∈ V such that .v(0) /= 0. Thus,

u , (0) = 0.
.

Thus, the equivalence between the strong problem and the variational formulation.

3.6.4.3 Existence and Uniqueness

We consider the following symmetrical bilinear form:

a : V × V −→ R
.
{ {
1
∂u ∂v 1
(u, v) −→ a(u, v) = dx + uv d x
0 ∂x ∂x 0

and the following linear form:

. L : V −→ R
{ 1
v −→ L(v) = f v d x.
0

. V is a Hilbert space. In fact, consider the mapping

@seismicisolation
@seismicisolation
144 3 Variational Formulation of Boundary Problems

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(1).

ϕ is linear.
.

Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,

.

.|ϕ(v)| = |v(1)|
≤ ||v||∞ .

Further, . H 1 (]0, 1[) is continuously injected into .C 0 ([0, 1], R), so

. ||v||∞ ≤ c||v||1,]0,1[ .

Thus,
|ϕ(v)| ≤ c||v||1,]0,1[ .
.

Hence, .ϕ is continuous. Since .V = ϕ −1 ({0}) and .{0} is closed of .R, the reciprocal
range of closed by a continuous map is closed. Thus, .V is a closed subspace of
. H (]0, 1[). Thus, . V is a Hilbert space.
1

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

|a(u, v)| = |(u, v)1,]0,1[ |

.

≤ ||u||1,]0,1[ ||v||1,]0,1[ .

Hence, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

a(u, u) = (u, u)1,]0,1[

.

= ||u||21,]0,1[
1
≥ ||u||21,]0,1[ .
2
So, .a(·, ·) is coercive.
Continuity of L(·): Let .v ∈ V . Then,
.

{ 1
|L(v)| ≤
. | f | |ϕ| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

@seismicisolation
@seismicisolation
3.6 Neumann Problem 145

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.6.17)
admits one solution and only one .u ∈ V .

3.6.5 Second Example

Consider the problem

d 2u
. − + u = f in ]0, 1[ (3.6.18)
dx2
, ,
.u(0) = u(1), u 0) = u (1), (3.6.19)

with . f ∈ L 2 (]0, 1[, R).

3.6.5.1 Variational Formulation

Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(0) = v(1)} be a test function. Multiply

Eq. (3.6.18) by .v,
d 2u
.− v + uv = f v
dx2
and integrate over .]0, 1[,
{ 1 { 1 { 1
d 2u
. − v dx + uv d x = f v d x.
0 dx2 0 0

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv
. d x − [u , v]10 + uv d x = f v d x,
0 dx dx 0 0

which proves
{ 1 { 1 { 1
du dv
. d x − u , (1)v(1) + u , (0)v(0) + uv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. dx + uv d x = f v d x. (3.6.20)
0 dx dx 0 0

@seismicisolation
@seismicisolation
146 3 Variational Formulation of Boundary Problems

3.6.5.2 Equivalence Between Variational Formulation and Strong

Problem

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + uϕ d x = f ϕ d x.
0 dx dx 0 0

Then, { { {
1 1 1
d 2u
. − ϕ dx + uϕ d x = f ϕ d x.
0 dx2 0 0

Thus, { ( 2 )
1
d u
. − 2 + u − f ϕ d x = 0.
0 dx

So, / \
d 2u
. − 2 + u − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx

It follows that
d 2u
. − +u− f =0
dx2

on .D , (]0, 1[). Moreover, . f ∈ L 2 (]0, 1[, R), so

d 2u
. − +u = f
dx2

on . L 2 (]0, 1[, R). It thus appears that

d 2u
. − +u = f
dx2
almost everywhere in .]0, 1[. Hence, (3.6.18). Moreover, .u ∈ V , so

u(0) = u(1).
.

It remains to show that .u , (0) = u , (1). In fact, let .v ∈ V , using Green’s formula to
the variational problem, we find
{ 1 { 1 { 1
d 2u
. − v d x + [u , v]10 + uv d x = f v d x.
0 dx2 0 0

Moreover,

@seismicisolation
@seismicisolation
3.6 Neumann Problem 147

d 2u
. − + u = f,
dx2
hence
u , (1)v(1) − u , (0)v(0) = 0
.

for all .v ∈ V . Thus,

v(0)(u , (1) − u , (0)) = 0.
.

If we choose a .v ∈ V such that

v(0) /= 0,
.

then
u , (1) = u , (0).
.

Thus, the equivalence between the strong problem (3.6.18)–(3.6.19) and the varia-
tional formulation (3.6.20).

3.6.5.3 Existence and Uniqueness

We consider the following symmetrical bilinear form:

a : V × V −→ R
.
{ {
1
∂u ∂v 1
(u, v) −→ a(u, v) = dx + uv d x
0 ∂x ∂x 0

and the following linear form:

. L : V −→ R
{ 1
v −→ L(v) = f ϕ d x.
0

. V is a Hilbert space. In fact, consider the mapping

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(0) − v(1).

ϕ is linear.
.

Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,

.

|ϕ(v)| = |v(0) − v(1)|

.

≤ 2||v||∞ .

@seismicisolation
@seismicisolation
148 3 Variational Formulation of Boundary Problems

Further, . H 1 (]0, 1[) is continuously injected into .C 0 ([0, 1], R), so

||v||∞ ≤ c||v||1,]0,1[ .
.

Thus,
|ϕ(v)| ≤ 2c||v||1,]0,1[ .
.

Hence, .ϕ is continuous. Further, .V = ϕ −1 ({0}). .{0} is closed of .R, the recipro-

cal range of closed by a continuous map is closed. So, .V is a closed subspace of
. H (]0, 1[). Thus, . V is a Hilbert space.
1

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

|a(u, v)| = |(u, v)1,]0,1[ |

.

≤ ||u||1,]0,1[ ||v||1,]0,1[ .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

a(u, u) = (u, u)1,]0,1[

.

= ||u||21,]0,1[
1
≥ ||u||21,]0,1[ .
2
Thus, .a(·, ·) is coercive.
Continuity of L(·): Let .v ∈ V . Then,
.

{ 1
|L(v)| ≤
. | f | |v| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.6.20)
admits one solution and only one .u ∈ V .
When . f ∈ H 1 (]0, 1[), then the solution .u obtained is solution in the sense usual. In
fact, since
d 2u
.− + u = f,
dx2
then
d 2u
. − = f − u.
dx2

@seismicisolation
@seismicisolation
3.6 Neumann Problem 149

Moreover, .u, . f ∈ H 1 (]0, 1[), so

d 2u
. − ∈ H 1 (]0, 1[).
dx2

This proves .u ∈ H 3 (]0, 1[). Further, .3 > 21 + 2, so according to the Theorem 3.4.9,
. H (]0, 1[) is continuously injected into .C ([0, 1], R). Hence, .u ∈ C ([0, 1], R).
3 2 2

Thus, the obtained solution .u is solution in the usual sense.

3.6.6 Third Example

We consider the problem at the limits: Find .u such that

d 2u
. − + μu = f in ]0, 1[ (3.6.21)
dx2
du
.u(0) = 0, (1) = 1, (3.6.22)
dx

where . f ∈ L 2 (]0, 1[, R), .μ ∈ L ∞ (0, 1) with

μ(x) ≥ β > 0,
.

for all .x ∈]0, 1[. This problem corresponds to a metal bar heated by means of a
quantity of heat. f , the function.u representing the temperature. We fix the temperature
.u = 0 in . x = 0 and we assume that the heat flow

du
. (1) = 1 in x = 1.
dx

3.6.6.1 Variational Formulation

Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(0) = 0} be a test function. Multiply

Eq. (3.6.21) by .v,
d 2u
.− v + μuv = f v
dx2
and integrate over .]0, 1[,
{ 1 { 1 { 1
d 2u
. − v dx + μuv d x = f v d x.
0 dx2 0 0

@seismicisolation
@seismicisolation
150 3 Variational Formulation of Boundary Problems

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv
. d x − [u , v]10 + μuv d x = f v d x.
0 dx dx 0 0

This proves
{ 1 { 1 { 1
du dv
. d x − u , (1)v(1) + u , (0)v(0) + μuv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. dx + μuv d x = f v d x + v(1). (3.6.23)
0 dx dx 0 0

3.6.6.2 Existence and Uniqueness

We consider the following symmetrical bilinear form:

.a : V × V −→ R
{ 1 { 1
du dv
(u, v) −→ a(u, v) = dx + μuv d x
0 dx dx 0

and the following linear form:

. L : V −→ R
{ 1
v −→ L(v) = f ϕ d x + v(1).
0

. V is a Hilbert space. In fact, consider the mapping

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(0).

ϕ is linear.
.

Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,

.

.|ϕ(v)| = |v(0)|
≤ ||v||∞ .

On the other hand, . H 1 (]0, 1[) is injected continuously into .C 0 ([0, 1], R), so

@seismicisolation
@seismicisolation
3.6 Neumann Problem 151

. ||v||∞ ≤ c||v||1,]0,1[ .

Thus,
.|ϕ(v)| ≤ c||v||1,]0,1[ .

Hence, .ϕ is continuous. Furthermore, .V = ϕ −1 ({0}). .{0} is closed of .R, the recip-

rocal range of closed by a continuous map is closed. So, .V is a closed subspace of
. H (]0, 1[). Thus, . V is a Hilbert space.
1

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

.|a(u, v)| ≤ |u|1,]0,1[ |v|1,]0,1[ + ||μ||∞ ||u||0,]0,1[ ||v||0,]0,1[

≤ (1 + ||μ||∞ )||u||1,]0,1[ ||v||1,]0,1[ .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

{ 1 ( )2 { 1
du
a(u, u) =
. dx + μu 2 d x
0 dx 0
≥ |u|21,]0,1[ + β||u||20,]0,1[
≥ min(1, β)||u||21,]0,1[ .

Since .min(1, β) > 0, then .a(·, ·) is coercive.

Continuity of L(·): Let .v ∈ V . Then,
.

{ 1
.|L(v)| ≤ | f | |v| d x + |v(1)|
0
≤ || f ||0,]0,1[ ||v||0,]0,1[ + ||v||∞ .

Furthermore, . H 1 (]0, 1[) is continuously injected into .C 0 ([0, 1], R), so

. ||v||∞ ≤ c||v||1,]0,1[ .

Hence,

. |L(v)| ≤ || f ||0,]0,1[ ||v||1,]0,1[ + c||v||1,]0,1[

= (|| f ||0,]0,1[ + c)||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.6.23)
admits one solution and only one .u ∈ V .

@seismicisolation
@seismicisolation
152 3 Variational Formulation of Boundary Problems

3.6.6.3 Equivalence Between Variational Formulation and Strong

Problem

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + μuϕ d x − ϕ(1) = f ϕ d x.
0 dx dx 0 0

Further, .ϕ(1) = 0. This proves

{ 1 { 1 { 1
d 2u
. − ϕ dx + μuϕ d x = f ϕ d x.
0 dx2 0 0

Thus, { ( )
1
d 2u
. − + μu − f ϕ d x = 0.
0 dx2

So, / \
d 2u
. − 2 + μu − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx

It follows that
d 2u
. − + μu − f = 0
dx2

on .D , (]0, 1[). On the other hand, . f ∈ L 2 (]0, 1[, R), so

d 2u
. − + μu = f
dx2

on . L 2 (]0, 1[, R) (see (2.18.5)). It thus appears that

d 2u
. − + μu = f
dx2

almost everywhere in the domain.]0, 1[. Hence, (3.6.21). Further,.u ∈ V , so.u(0) = 0.

It remains to show that .u , (1) = 1. In fact, let .v ∈ V , using Green’s formula to the
variational problem, we find
{ 1 { 1 { 1
d 2u
. − v d x + [u , v]10 + μuv d x − v(1) = f v d x.
0 dx2 0 0

Hence, { ( 2 )
1
d u
. − 2 + μu − f v d x + (u , (1) − 1)v(1) = 0.
0 dx

@seismicisolation
@seismicisolation
3.7 Transmission Problem 153

On the other hand,

d 2u
. − + μu = f,
dx2

so .(u , (1) − 1)v(1) = 0 for all .v ∈ V . Further, if we choose a .v ∈ V such that .v(1) /=
0, then.u , (1) = 1. Hence, (3.6.22). Thus, the equivalence between the strong problem
and the variational formulation.

3.7 Transmission Problem

Let .Ω be a bounded
uopenuset with fairly regular boundary .∂Ω. We decompose .Ω in
the form .Ω = Ω1 Ω2 E (as in the following figure):

Ωk being an open of .Rn with boundary .Γk , .k = 1, 2. We assume that the .E interface
.

is fairly regular (.C 1 per piece) so that for .v = (v1 , v2 ) ∈ H 1 (Ω1 ) × H 1 (Ω2 ), we can
define the traces .v1|E and .v2|E . We consider the following space .V

. V = {v = (v1 , v2 ) ∈ H 1 (Ω1 ) × H 1 (Ω2 ) such that b1 v1 = b2 v2 on E},

where .b1 and .b2 belong to . L 2 (E, R). The space .V endowed with the norm
( )1
.||v||V = ||v1 ||21,Ω1 + ||v2 ||21,Ω2 2

is a closed subspace of . H 1 (Ω1 ) × H 1 (Ω2 ), so it’s a Hilbert space. On the other hand,
let the functions .aikj ∈ L ∞ (Ωk , R), .k = 1, 2, .1 ≤ i, j ≤ n, and .a0k ∈ L ∞ (Ωk , R),
.k = 1, 2, such that for all . x ∈ Ωk and for all .ξ ∈ R ,
n

E
n E
n
. aikj (x)ξ j ξi ≥ α |ξi |2 ,
i, j=1 i=1

where .α > 0 and

a k (x) ≥ α0 > 0 for all x ∈ Ωk .
. 0

We define the bilinear form

@seismicisolation
@seismicisolation
154 3 Variational Formulation of Boundary Problems
⎛ ⎞
2 {
E E
n
⎝ ∂u ∂v
+ a0k u k vk ⎠ d x
k k
a(u, v) =
. aikj
k=1 Ωk i, j=1
∂ x i ∂ x j

for .u, .v ∈ H 1 (Ω1 ) × H 1 (Ω2 ). We verify that the bilinear form .a(·, ·) is continuous
and .V -elliptic. We finally introduce the linear form
{ {
. L(v) = f 1 v1 d x + f 2 v2 d x,
Ω1 Ω2

where . f k ∈ L 2 (Ωk , R), .k = 1, 2. This form is continuous on .V . Hence, there exists

a pair of functions .u = (u 1 , u 2 ) ∈ V such that

a(u, v) = L(v) for all v ∈ V.

.

Let’s interpret the solved problem. We first choose .v = (v1 , 0), with .v1 ∈ D(Ω1 ),
then comes ( )
E n
∂ ∂u 1
.− ai1j + a01 u 1 = f 1 on Ω1 . (3.7.1)
i, j=1
∂ x j ∂ x i

Similarly by choosing .v = (0, v2 ) with .v2 ∈ D(Ω2 ), we get

E n ( )
∂ 2 ∂u 2
.− ai j + a02 u 2 = f 2 on Ω2 . (3.7.2)
i, j=1
∂ x j ∂ x i

We multiply (3.7.1) by .v1 (resp. (3.7.2) by .v2 ) and we integrate on .Ω1 (resp. on .Ω2 )
and we add member to member. Using Green’s formula on each .Ωk , it comes
⎛ ⎞ ⎛ ⎞
{ E
n { E
n
⎝ ∂u 1 (1) ⎠ ⎝ ∂u 2 (2) ⎠
. ai1j ν v1 ds + ai2j ν v2 ds = 0
∂Ω1 i, j=1
∂ xi j ∂Ω2 i, j=1
∂ xi j

for all .v ∈ V , where .ν (k)

j is the . jth component of the exterior normal to .Ωk . We
deduce (by choosing .v such that .v2 = 0 on .Ω2 , .v1 = 0 on .E)

E
n
∂u 1 (1)
. ai1j ν = 0 on ∂Ω1 \E. (3.7.3)
i, j=1
∂ xi j

Likewise, we have
E
n
∂u 2 (2)
. ai2j ν = 0 on ∂Ω2 \E. (3.7.4)
i, j=1
∂ xi j

@seismicisolation
@seismicisolation
3.8 General Second-Order Elliptic Problems 155

Finally, choosing .v such that .v1 = 0 on .∂Ω1 \E, .v2 = 0 on .∂Ω2 \E, .v1 = b2 ω and
v = b1 ω on .E, we obtain
. 2

E
n
∂u 1 (1) E n
∂u 2 (2)
b
. 2 ai1j ν + b1 ai2j ν = 0 on E. (3.7.5)
i, j=1
∂ xi j i, j=1
∂ xi j

So, we solved the problem (3.7.1), (3.7.2), (3.7.3), (3.7.4), and (3.7.5) with the
condition .b1 u 1 = b2 u 2 on .E.

3.8 General Second-Order Elliptic Problems

3.8.1 Divergence Operator

Consider more general second-order elliptic operators in an open Lipschitz subset

Ω of .Rn . Let .ai j ∈ C 1 (Ω, R) .1 ≤ i, j ≤ n and . A(x) = (ai j (x)) be a .n × n matrix-
.

valued function. Let .u ∈ C 2 (Ω, R) (this regularity can be considerably reduced),

then . A∇u is a vector field whose components are given by

E
n
∂u
.(A∇u)i = ai j .
j=1
∂x j

The divergence of . A∇u is formally given by

E
n
∂(A∇u)i
div(A∇u) =
.
i=1
∂ xi
( n )
E n
∂ 2u En E ∂ai j ∂u
= ai j + .
i, j=1
∂ xi x j j=1 i=1
∂ xi ∂ x j

E
The principal part of this operator . i,n j=1 ai j ∂∂xi x j is of the second order. Let .c, .b, . f ,
2

. g, and .h be functions, .Γ0 , .Γ1 be a partition of .∂Ω as in the mixed problem and

⎞
⎛
ν1
⎜ . ⎟
.ν = ⎝ . ⎠ ,
.
νn

where .νi (.i ∈ {1, . . . , n}) is the .ith directing cosine of the normal .ν at .Γ = ∂Ω
directed outwards. Let us consider the boundary value problem

@seismicisolation
@seismicisolation
156 3 Variational Formulation of Boundary Problems

. − div(A∇u) + cu = f in Ω (3.8.1)
u = h on Γ0 ,
. (3.8.2)
.bu + ν · A∇u = g on Γ1 . (3.8.3)
∂u
When . A = I , we recognize .−div(A∇u) = −Δu and .ν · A∇u = ∂ν
so that we gen-
eralize all the model problems seen so far.

3.8.2 Variational Formulation

First, we reduce the study to the case .h = 0 by subtracting a function with the appro-
priate trace, as before. Suppose that . f ∈ L 2 (Ω, R), .g ∈ L 2 (Γ1 , R), .c ∈ L ∞ (Ω, R)
and .b ∈ L ∞ (Γ1 , R). Then, the variational formulation for problem (3.8.1)–(3.8.3) is
given by: Find .u ∈ V := {v ∈ H 1 (Ω) such that v|Γ1 = 0 on Γ1 } such that

a(u, v) = L(v)
.

for all .v ∈ V , where

{ {
.a(u, v) = (A∇u · ∇v + cuv) d x + bu |Γ1 v|Γ1 dσ
Ω Γ1

and { {
. L(v) = f v dx + gv|Γ1 dσ.
Ω Γ1

In fact, multiply the partial differential equations (PDEs) (3.8.1) by .v ∈ V , integrate

over .Ω,
{ (E n
) { {
∂(A∇u)i
.− v dx + cuv d x = f v d x,
Ω i=1
∂ xi Ω Ω

and, using Green formula, we get

{ E
n { (E
n
) { {
∂v
. (A∇u)i dx − (A∇u)i νi v|Γ1 dσ + cuv d x = f v d x.
Ω i=1
∂ xi Γ1 i=1 Ω Ω

Finally, we have to find .u ∈ V such that

{ { { {
. (A∇u · ∇v + cuv) d x + bu |Γ1 v|Γ1 dσ = f v dx + gv|Γ1 dσ
Ω Γ1 Ω Γ1

for all .v ∈ V . We leave the converse to the reader.

@seismicisolation
@seismicisolation
3.8 General Second-Order Elliptic Problems 157

3.8.3 Existence and Uniqueness

It is easy to see that .V is a Hilbert space and that it is continuous are already known
facts. By the boundedness of the coefficients of the matrix .ai j (x), we leave the proof
of the continuity of the bilinear form
{ {
.a(u, v) = (A∇u · ∇v + cuv) d x + bu |Γ1 v|Γ1 dσ.
Ω Γ1

Since
{ {
a(u, u) =
. (A∇u · ∇u + cu 2 ) d x + bu 2|Γ1 dσ
Ω Γ1
{ {
≥α |∇u| d x + η
2 2
u dx
Ω Ω
≥ min(α, η)||u||21,Ω ,

the .V -ellipticity is also quite obvious. The linear form

{ {
. L(v) = f v dx + gv|Γ1 dσ
Ω Γ1

is continuous. In fact, let .v ∈ V . Then, by using the trace theorem, we have

{ {
|L(v)| ≤
. | f ||v| d x + |g||v|Γ1 | dσ
Ω Γ1
≤ || f ||0,Ω ||v||0,Ω + ||g||0,Γ1 ||v|Γ1 ||0,Γ1
≤ || f ||0,Ω ||v||1,Ω + C||g||0,Γ1 ||v||1,Ω
≤ (|| f ||0,Ω + C||g||0,Γ1 )||v||1,Ω .

Thus, . L(·) is continuous. Let us give a first existence and uniqueness result.
Theorem 3.8.1 Let .Ω be a Lipschitz open subset of .Rn , . f ∈ L 2 (Ω, R), .g ∈
L 2 (Γ1 , R), .c ∈ L ∞ (Ω, R), and .b ∈ L ∞ (Γ1 , R). We suppose that the matrix . A is
uniformly elliptic, i.e., there exists a constant .α > 0 such that

E
n
. ai j (x)ξi ξ j ≥ α|ξ |2
i, j=1

for all .x ∈ Ω and all .ξ ∈ Rn , where .| · | denotes the Euclidean norm in .Rn . We
suppose, in addition, that .c ≥ η > 0 for some constant .η and that .b ≥ 0. Then, the
problem: Find .u ∈ V = {v ∈ H 1 (Ω) such that v|Γ1 = 0 on Γ1 } such that

@seismicisolation
@seismicisolation
158 3 Variational Formulation of Boundary Problems
{ { { {
. (A∇u · ∇v + cuv) d x + bu |Γ1 v|Γ1 dσ = f v dx + gv|Γ1 dσ
Ω Γ1 Ω Γ1

for all .v ∈ V , has one and only one solution. ♦

Proof .V is a Hilbert space, .a(·, ·) is bilinear form, continuous and coercive, and
L(·) is linear continuous. So, the existence, uniqueness, and continuous dependence
.
of the solution on the data by Lax-Milgram theorem (see Theorem 2.17.1). Q.E.D.

Remark 3.8.2 When the matrix . A is not symmetric, the bilinear form .a(·, ·) is not
either, even if the principal part of the operator is symmetric since

E
n
∂2 E n
ai j + a ji ∂ 2
. ai j =
i, j=1
∂ xi x j i, j=1
2 ∂ xi x j

due to the fact that . ∂ ∂xi x j = ∂ ∂x j xi . When . A is symmetric, then so is for the bilinear
2 2

form and we have an equivalent minimization problem with

[{ { ] { {
1
. J (v) = (A∇v · ∇v + cv2 ) d x + 2
bv|Γ dσ − f v dx − gv|Γ1 dσ,
2 Ω Γ1
1
Ω Γ1

minimized over .V .
It is quite clear that we can reduce the regularity from . A to . L ∞ without losing the
existence and uniqueness of the variational problem. The interpretation in terms of
the partial differential equations (PDEs) stops at the divergence form

. − div(A∇u) + cu = f

since we cannot expand the divergence using Leibniz’s formula in this case. Such a
lack of regularity is useful for modeling heterogeneous media. ♦

3.9 Weak Solution of Elliptic of the Partial Differential

Equations (PDEs)

We consider the boundary value problems of the form

E n ( ) E n
∂ ∂u ∂u
.− a jk (x) + b j (x) + c(x)u = f (3.9.1)
j,k=1
∂x j ∂ xk j=1
∂x j

on a bounded open set .Ω ⊂ Rn , where .a jk , .b j , .c ∈ L ∞ (Ω, R), and . f ∈ L 2 (Ω, R).

We do not fix boundary conditions at this time. The problem (3.9.1) is called elliptic
if there is a constant .α > 0 such that

@seismicisolation
@seismicisolation
3.9 Weak Solution of Elliptic of the Partial Differential Equations (PDEs) 159

E
n E
n
. a jk (x)ξ j ξk ≥ α ξ 2j , for all ξ ∈ Rn , x ∈ Ω. (3.9.2)
j,k=1 j=1

Assuming all functions and the domain are sufficiently smooth, we can multiply by
a smooth function .v, integrate over .x ∈ Ω
{ E
n ( ) { En { {
∂ ∂u ∂u
. − a jk (x) v dx + b j (x) v dx + c(x)uv d x = f v d x.
∂x j
Ω j,k=1 ∂ xk Ω ∂x j Ω Ω
j=1

By using Green’s formula, we have

n {
E E n { { n {
E
∂u ∂v ∂u ∂u
. a jk dx + bj v dx + cuv d x − a jk vν j dσ =
∂ x j ∂ xk ∂x j ∂ xk
j,k=1 Ω j=1 Ω Ω j,k=1 ∂Ω
{
. f v d x, (3.9.3)
Ω

where .ν := (ν1 , . . . , νn )T is the outward unit normal on the boundary .∂Ω. We then
search for .u ∈ V “for a suitably chosen function space .V ” satisfying (3.9.3) for all
.v ∈ V including boundary conditions which we will discuss next. In the following,
we will consider the three boundary conditions.

3.9.1 Dirichlet Conditions

For given .g ∈ L 2 (∂Ω, R), we require, in the sense of traces, .u = g on .∂Ω. For the
homogeneous Dirichlet condition, i.e.,.g = 0, we take.V = H01 (Ω), and the boundary
integrals in (3.9.3) vanish since .v = 0 on .∂Ω. Hence, the weak formulation is then
given by: Find .u ∈ H01 (Ω) satisfying

n {
E E n { { {
∂u ∂v ∂u
. a jk dx + bj v dx + cuv d x = f v dx
j,k=1 Ω
∂ x j ∂ xk j=1 Ω ∂x j Ω Ω

for all .v ∈ H01 (Ω).

If .g /= 0, and .g ∈ H 1 (∂Ω) with .∂Ω of class .C 1 (e.g., .g and .∂Ω are sufficiently
smooth), we can find a function .u g ∈ H 1 (Ω) such that .T u g = g. We then set .u =
u + u g , where .-
- u ∈ H01 (Ω) satisfies
{
. u , v) =
a(- f v d x − a(u g , v)
Ω

for all .v ∈ H01 (Ω), where

@seismicisolation
@seismicisolation
160 3 Variational Formulation of Boundary Problems

n {
E E n { {
∂u ∂v ∂u
a(u, v) :=
. a jk dx + bj v dx + cuv d x.
j,k=1 Ω
∂ x j ∂ xk j=1 Ω ∂x j Ω

We give now sufficient conditions on the coefficients .a jk , .b j , .c, and .d such that the
boundary value problems, defined in the previous, have one and only one solution.
Theorem 3.9.1 (Well-posedness) Let .a jk ∈ L ∞ (Ω, R) satisfy the ellipticity con-
dition (3.9.2) with constant .α > 0, let .b j , .c ∈EL ∞ (Ω, R) and . f ∈ L 2 (Ω, R) and
−1 n
. g ∈ L (∂Ω, R) be given, and set .β = α j=1 ||b j || L ∞ (Ω,R) . Then, the prob-
2 2

lem (3.9.1) with homogeneous Dirichlet condition has one and only one solution
.u ∈ H0 (Ω) if we have
1

β
c(x) −
. ≥ 0 for almost all x ∈ Ω.
2
In this case, there is a .C > 0 such that

||u||1,Ω ≤ C|| f ||0,Ω .

.

Furthermore, the non-homogeneous Dirichlet problem, for .g ∈ H 1 (∂Ω), has one

and only one solution satisfying

||u||1,Ω ≤ C(|| f ||0,Ω + ||g||1,∂Ω ).

.

Proof We check the hypotheses of the Lax-Milgram theorem (see Theorem 2.17.1).
By the Holder inequality and the boundedness of the coefficients, we have the con-
tinuity of the bilinear form .a(·, ·) and the linear form . F(·). In fact,
Continuity of a(·, ·): Let .u, .v ∈ H01 (Ω). Then,
.

| |
|E { En { { |
| n ∂u ∂v ∂u |
.|a(u, v)| = || a jk dx + bj v dx + cuv d x ||
| j,k=1 Ω ∂ x j ∂ xk Ω ∂x j Ω |
j=1

E { | | | | E { | | {
n
| ∂u | | ∂v | n
| ∂u |
≤ |a jk | || || |
| | ∂x | dx + |b j | || | |v| d x + |c||u||v| d x
Ω ∂ x j k Ω ∂x j | Ω
j,k=1 j=1

E { | || | E { | |
n
| ∂u | | ∂v | n
| ∂u |
≤ ||a jk || L ∞ (Ω,R) | || | dx + ||b j || L ∞ (Ω,R) | | |v| d x+
| ∂x | | ∂x | | ∂x |
j,k=1 Ω j k j=1 Ω j
{
||c|| L ∞ (Ω,R) |u||v| d x
Ω
E
n E
n
≤ ||a jk || L ∞ (Ω,R) |u|1,Ω |v|1,Ω + ||b j || L ∞ (Ω,R) |u|1,Ω ||v||0,Ω +
j,k=1 j=1

||c|| L ∞ (Ω,R) ||u||0,Ω ||v||0,Ω .

@seismicisolation
@seismicisolation
3.9 Weak Solution of Elliptic of the Partial Differential Equations (PDEs) 161

By using Poincaré inequality (Theorem 3.2.2), we have

⎛ ⎞
E
n E
n
.|a(u, v)| ≤⎝ ||a jk || L ∞ (Ω,R) + C(Ω) ||b j || L ∞ (Ω,R) + C(Ω) ||c|| L ∞ (Ω,R) ⎠ |u|1,Ω |v|1,Ω ,
2

j,k=1 j=1

where .C(Ω) is the constant from Poincaré’s inequality. Thus, .a(·, ·) is continuous.
Continuity of F(·): Let .v ∈ H01 (Ω). Then,
.

|{ |
| |
.|F(v)| = | f v d x ||
|
{Ω
≤ | f ||v| d x
Ω
≤ || f ||0,Ω ||v||0,Ω .

By using Poincaré inequality (Theorem 3.2.2), we have

|F(v)| ≤ C(Ω)|| f ||0,Ω |v|1,Ω ,

.

where .C(Ω) is the constant from Poincaré’s inequality. Thus, . F(·) is continuous.
Coercivity of a(·, ·): It remains for us to show the coercivity of .a(·, ·). Let .u ∈
.

H01 (Ω). Then, the ellipticity of .a jk gives that

{ E
n { E n ( )
∂u ∂u ∂u 2
. a jk dx ≥ α dx
Ω j,k=1
∂ x j ∂ xk Ω j=1 ∂ x j

En || ||
|| ∂u ||2
=α || ||
|| ∂ x ||
j=1 j 0,Ω

= α|u|21,Ω .

By using Young’s inequality.ab ≤ αa 2 + 2α b for.a = |u|1,Ω ,.b = ||u||0,Ω and.α > 0

1 2

as well as repeated application of Holder’s inequality that

⎛ ⎞ 21
En {
a(u, u) ≥ α|u|21,Ω
. −⎝ ||b j ||2L ∞ (Ω,R) ⎠ |u|1,Ω ||u||0,Ω + cu 2 d x
j=1 Ω
⎛ ⎞
{ En
α 2 ⎝c − 1
≥ |u| + ||b j ||2L ∞ (Ω,R) ⎠ u 2 d x.
2 1,Ω Ω 2α j=1

β
The second term of the last inequality is non-negative (since .c − 2
≥ 0), and we
infer by using Poincaré’s inequality (Theorem 3.2.2) that

@seismicisolation
@seismicisolation
162 3 Variational Formulation of Boundary Problems

α 2
a(u, u) ≥
. |u|
2 1,Ω
α 2 α
≥ |u|1,Ω + ||u||20,Ω
4 4C(Ω)2
≥ C|u|21,Ω
α
holds for .C := 4+4C(Ω)2 , where .C(Ω) is the constant from Poincaré’s inequality.

Now, according to the Lax-Milgram theorem (Theorem 2.17.1), the problem (3.9.1)
with homogeneous Dirichlet condition admits one solution and only one.u ∈ H01 (Ω).
Q.E.D.

Naturally, if the data has higher regularity, we can expect more regularity of the
solution as well. The corresponding theory is quite involved, and we give only two
results which will be relevant in the following.
Theorem 3.9.2 Let .Ω ⊂ Rn be a bounded domain with .C k+1 boundary, .k ≥ 0,
.a jk ∈ C (Ω, R), and .b j , .c ∈ W (Ω). Then, for any . f ∈ H k (Ω), the N
k k,∞
solution of
the problem (3.9.1) with homogeneous Dirichlet condition is in . H k+2 (Ω) H01 (Ω),
and, there is a .C > 0 such that

||u||k+2,Ω ≤ C(|| f ||k,Ω + ||u||1,Ω ).

.

♦
Theorem 3.9.3 Let .Ω be a convex polygon in .R2 or a parallelepiped in .R3 , .a jk ∈
C 1 (Ω, R) and .b j , .c ∈ C 0 (Ω, R). Then, the solution of the problem (3.9.1) with
homogeneous Dirichlet condition is in . H 2 (Ω), and there is a .C > 0 such that

.||u||2,Ω ≤ C|| f ||0,Ω .

♦
Remark 3.9.4 For non-convex polygons, .u ∈ H (Ω) is not possible. This is due to
2

the presence of so-called corner singularities at re-entrant corners, which severely

limits the accuracy of finite element approximations. This requires special treatment,
and is a topic of extensive current research. ♦

3.9.2 Neumann Conditions

For given .g ∈ L 2 (∂Ω, R), we require

E
n
∂u
. a jk ν j = g on ∂Ω. (3.9.4)
j,k=1
∂ xk

@seismicisolation
@seismicisolation
3.9 Weak Solution of Elliptic of the Partial Differential Equations (PDEs) 163

In this case, we can substitute this equality in the boundary integral in (3.9.3) and
take .V = H 1 (Ω). The weak formulation is then: Find .u ∈ H 1 (Ω) satisfying

a(u, v) = F(v)
.

for all .v ∈ H 1 (Ω), where

{ {
. F(v) = f v dx + gv dσ.
Ω ∂Ω

Theorem 3.9.5 (Well-posedness) Let .a jk ∈ L ∞ (Ω, R) satisfy the ellipticity con-

∞
dition (3.9.2) with constant .α > 0, letE
.b j , .c ∈ L (Ω, R), . f ∈ L (Ω, R) and . g ∈
2
−1 n
L (∂Ω, R) be given, and set .β = α
2
j=1 ||b j || L ∞ (Ω,R) . Then, the problem (3.9.1)
2

with Neumann condition (3.9.4), for .g ∈ L 2 (∂Ω, R), has one and only one solution
.u ∈ H (Ω) if we have the following:
1

β
c(x) −
. ≥ γ > 0 for almost all x ∈ Ω.
2
In this case, there is a .C > 0 such that

||u||1,Ω ≤ C(|| f ||0,Ω + ||g||0,∂Ω ).

.

3.9.3 Robin Conditions

For given .g ∈ L 2 (∂Ω, R) and .d ∈ L ∞ (∂Ω, R), we require

E
n
∂u
du +
. a jk ν j = g on ∂Ω. (3.9.5)
j,k=1
∂ xk

Again, we can substitute this in the boundary integral and take .V = H 1 (Ω). The
weak formulation is then: Find .u ∈ H 1 (Ω) satisfying
{
a(u, v) +
. duv dσ = F(v)
∂Ω

for all .v ∈ H 1 (Ω).

Theorem 3.9.6 (Well-posedness) Let .a jk ∈ L ∞ (Ω, R) satisfy the ellipticity con-

dition (3.9.2) with constant .α > 0, let .b j , .c ∈ L ∞ (Ω, R), . f ∈ L 2 (Ω, R), and

@seismicisolation
@seismicisolation
164 3 Variational Formulation of Boundary Problems

E
. g ∈ L 2 (∂Ω, R) be given, and set .β = α −1 nj=1 ||b j ||2L ∞ (Ω,R) . Then, the problem
(3.9.1) with Robin condition (3.9.5), for .g ∈ L 2 (∂Ω, R) and .d ∈ L ∞ (∂Ω, R), has
one and only one solution if we have

β
.c(x) − ≥ γ ≥ 0 for almost all x ∈ Ω
2

d(x) ≥ δ ≥ 0 for almost all x ∈ ∂Ω

.

and at least one inequality is strict. In this case, there is a .C > 0 such that

||u||1,Ω ≤ C(|| f ||0,Ω + ||g||0,∂Ω ).

.

3.10 A New Kind of the Fourier Condition

We now introduce a new type of boundary condition, the Fourier condition (also
called Robin’s condition or third boundary condition). Let .b and .c be two functions,
and consider the boundary value problem

. − Δu + cu = f in Ω (3.10.1)
∂u
.bu + = g on Γ = ∂Ω. (3.10.2)
∂ν |Γ

It is easy to see that when.b = 0, we recognize the Neumann problem (and, in a sense,
when .b = +∞ the Dirichlet problem). This condition bears the name of Fourier
who introduced it within the framework of the heat equation. In the interpretation
of heat, . ∂u
∂ν |Γ
represents the flow of heat across the boundary. Let us assume we
are modeling a situation in which the boundary is actually a very thin wall that
isolates .Ω from the outside, where the temperature is .0o . If .g = 0, then the Fourier
condition states that . ∂u
∂ν |Γ
= −bu, i.e., the flux passing heat through the wall is
proportional to the temperature difference between the interior and the outside. For
this interpretation to be physically reasonable, it is obviously necessary that .b ≥
0, that is to say that the heat flows inwards when the exterior is hotter than the
interior and vice versa. It is therefore to be expected that the sign of .b plays a
role. We follow the same pattern as before: Find a variational formulation for the
boundary value problem (3.10.1)–(3.10.2), then apply the Lax-Milgram theorem (see
Theorem 2.17.1) to prove existence and uniqueness.

@seismicisolation
@seismicisolation
3.10 A New Kind of the Fourier Condition 165

3.10.1 Variational Formulation

Suppose that . f ∈ L 2 (Ω, R), .g ∈ L 2 (∂Ω, R), .c ∈ L ∞ (Ω, R), and .b ∈ L ∞ (∂Ω, R).
Then, the variational formulation for the Fourier problem (3.10.1)–(3.10.2) is given
by: Find .u ∈ V = H 1 (Ω) such that

a(u, v) = L(v)
.

for all .v ∈ V , where

V = H 1 (Ω)
.
{ {
a(u, v) = (∇u · ∇v + cuv) d x + bu |∂Ω v|∂Ω dσ
{Ω { ∂Ω

L(v) = f v dx + gv|∂Ω dσ.

Ω ∂Ω

In fact, as always, we multiply the partial differential equations (PDEs) by .v ∈ V ,

integrate over .Ω, and use Green’s formula,
{ { {
∂u
. (∇u · ∇v + cuv) d x = f v dx + v dσ
Ω ∂ν |Γ
{Ω {∂Ω
= f v dx + (g − bu |∂Ω )v|∂Ω dσ,
Ω ∂Ω

hence
{ { { {
. (∇u · ∇v + cuv) d x + bu |∂Ω v|∂Ω dσ = f v dx + gv|∂Ω dσ,
Ω ∂Ω Ω ∂Ω
(3.10.3)
for all .v ∈ V .
Conversely, let us give a solution.u of the variational problem (3.10.3). Taking first
.v = ϕ ∈ D(Ω), all boundary integrals vanish and we get .−Δu + cu = f exactly as
before. Then taking .v arbitrarily, using Green’s formula and the partial differential
equations (PDEs) just obtained, we get
{ { {
∂u
. v|∂Ω dσ + bu |∂Ω v|∂Ω ds = gv|∂Ω dσ,
∂Ω ∂ν |Γ ∂Ω ∂Ω

so that { ( )
∂u
. + bu |∂Ω − g v|∂Ω dσ = 0,
∂Ω ∂ν |Γ

for all .v ∈ V = H 1 (Ω), hence the Fourier boundary condition.

@seismicisolation
@seismicisolation
166 3 Variational Formulation of Boundary Problems

3.10.2 Existence and Uniqueness

Let us give a first existence and uniqueness result.

Theorem 3.10.1 Let .Ω be a Lipschitz open subset of .Rn , . f ∈ L 2 (Ω, R), .g ∈
L 2 (∂Ω, R), .c ∈ L ∞ (Ω, R), and .b ∈ L ∞ (∂Ω, R). Suppose that .c ≥ η > 0 for some
constant .η and that .||b− || L ∞ (∂Ω,R) < min(1,η)
Cγ20
, where .b− = − min(0, b), and .Cγ0 is
the continuity constant of the trace mapping. Then, the problem: Find .u ∈ V such
that
{ { { {
. (∇u · ∇v + cuv) d x + bu |∂Ω v|∂Ω dσ = f v dx + gv|∂Ω dσ,
Ω ∂Ω Ω ∂Ω

for all .v ∈ V , has one and only one solution. ♦

Proof We verify the assumptions of the Lax-Milgram theorem (see Theorem 2.17.1).
We already know that .V is a Hilbert space. Let us consider the bilinear form
{ {
a(u, v) =
. (∇u · ∇v + cuv) d x + bu |∂Ω v|∂Ω dσ.
Ω ∂Ω

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

|{ | { {
| |
| (∇u · ∇v + cuv) d x |≤ |∇u| |∇v| d x + |c(x)||u||v| d x
.
| |
Ω Ω
{Ω
≤ |u|1,Ω |v|1,Ω + ||c||∞ |u||v| d x
Ω
≤ |u|1,Ω |v|1,Ω + ||c||∞ ||u||0,Ω ||v||0,Ω
≤ (1 + ||c||∞ )||u||1,Ω ||v||1,Ω .

The continuity of the bilinear form .a(·, ·) has also already been checked, except for
the integral terms at the boundaries
|{ |
| |
.| bu |∂Ω v|∂Ω dσ || ≤ ||b|| L ∞ (∂Ω,R) ||u |∂Ω ||0,∂Ω ||v|∂Ω ||0,∂Ω
|
∂Ω
≤ Cγ20 ||b|| L ∞ (∂Ω,R) ||u||1,Ω ||v||1,Ω ,

for all .u and .v in .V . So,

|a(u, v)| ≤ (1 + ||c||∞ + Cγ20 ||b|| L ∞ (∂Ω,R) )||u||1,Ω ||v||1,Ω ,

.

for all .u and .v in .V . Thus, .a(·, ·) is continuous. Let’s check it .V -ellipticity of .a(·, ·).
Coercivity of a(·, ·): Obviously .b ≥ −b− , thus
.

@seismicisolation
@seismicisolation
3.10 A New Kind of the Fourier Condition 167
{ {
. a(u, u) = (||∇u||2 + cu 2 ) d x + bu 2|∂Ω dσ
Ω ∂Ω
≥ min(1, η)||u||21,Ω − ||b− || L ∞ (∂Ω,R) ||u |∂Ω ||20,∂Ω
≥ (min(1, η) − Cγ20 ||b− || L ∞ (∂Ω,R) )||u||21,Ω ,

hence the .V -ellipticity.

Consider the linear form
{ {
. L(v) = f v dx + gv|∂Ω dσ.
Ω ∂Ω

The linear form is also known to be continuous. In fact,

Continuity of L(·): Let .v ∈ V ,
.

{ {
.|L(v)| ≤ | f ||v| d x + |g||v|∂Ω | dσ
Ω ∂Ω
≤ || f ||0,Ω ||v||0,Ω + ||g||0,∂Ω ||v|∂Ω ||0,∂Ω
≤ || f ||0,Ω ||v||1,Ω + Cγ0 ||g||0,∂Ω ||v||1,Ω
≤ (|| f ||0,Ω + Cγ0 ||g||0,∂Ω )||v||1,Ω .

Thus, . L(·) is continuous.

Now, the result follows by using Lax-Milgram theorem (see Theorem 2.17.1).
Q.E.D.

Remark 3.10.2 Under the previous hypotheses (see Theorem 3.10.1), we have exis-
tence and uniqueness via Lax-Milgram theorem (see Theorem 2.17.1) provided that
.b is not too negative in some sense. ♦

All these hypotheses give only sufficient conditions. Let us give another set of such
hypotheses.
Theorem 3.10.3 Same hypotheses of Theorem 3.10.1 except that we assume that
c ≥ 0 and that .b ≥ μ > 0 for some constant .μ. Then, the Fourier problem (3.10.1)–
.

(3.10.2) has one and only one solution. ♦

Proof Consider the bilinear form

{ {
a(u, v) =
. (∇u · ∇v + cuv) d x + bu |∂Ω v|∂Ω dσ.
Ω ∂Ω

The continuity of bilinear and linear forms has also already been proven. The only
point to establish is the .V -ellipticity. In fact, we use a compactness argument by
contradiction. For that we admit that the theorem of Rellich is also true in dimension
.n in an open set of Lipschitz. We have

@seismicisolation
@seismicisolation
168 3 Variational Formulation of Boundary Problems
{ {
a(u, u) =
. (|∇u|2 + cu 2 ) d x + bu 2|∂Ω dσ
Ω ∂Ω
{ {
≥ |∇u| d x + μ
2
u 2|∂Ω dσ.
Ω ∂Ω

Let us assume for contradiction that there is no constant .α > 0 such that
{ {
. |∇u| d x + μ
2
u 2|∂Ω dσ ≥ α||u||21,Ω .
Ω ∂Ω

This implies that for all .n ∈ N∗ , there is .u n ∈ H 1 (Ω) such that

{ {
1
. |∇u n |2 d x + μ u 2n |∂Ω dσ < ||u n ||21,Ω .
Ω ∂Ω n

We can suppose without loss of generality that

.||u n ||21,Ω = 1, (3.10.4)

and we have { {
. |∇u n |2 d x + μ u 2n |∂Ω dσ → 0. (3.10.5)
Ω ∂Ω

Now, by (3.10.4), the sequence.(u n )n is bounded in. H 1 (Ω). So, by Rellich’s theorem,
the sequence .(u n )n is relatively compact in . L 2 (Ω, R). We can extract a subsequence
of .(u n )n , always denoted .(u n )n , and .u ∈ L 2 (Ω, R) such that .u n → u in . L 2 (Ω, R).
In view of (3.10.5), .||∇u n ||0,Ω → 0. Therefore, since .∇u n → ∇u in .D , (Ω), we infer
that .∇u = 0 and .u is constant on each connected component of .Ω. Furthermore,

. ||u n − u||21,Ω = ||∇u n ||20,Ω + ||u n − u||20,Ω → 0 (3.10.6)

so that, by continuity of the trace mapping .u n |∂Ω → u |∂Ω in . L 2 (∂Ω, R). Using
(3.10.5), we also have.||u n |∂Ω ||0,∂Ω → 0 since.μ > 0. Therefore,.u |∂Ω = 0. It follows
that .u being a constant with zero trace disappears in every connected component,
i.e., .u |∂Ω = 0. We now realize that (3.10.4) and (3.10.6) contradict each other, so
our premise that there is no .V -ellipticity constant .α is false. Now the result follows
using the Lax-Milgram theorem (see Theorem 2.17.1). Q.E.D.

3.11 The Convection-Diffusion Problem

Now, consider an example of a non-symmetric problem, the convection-diffusion

problem. Give us a vector field .σ . Let .Ω be a bounded open subset of .Rn and suppose
that . f ∈ L 2 (Ω, R), .σ ∈ C 1 (Ω; Rn ) and .c ∈ L ∞ (Ω, R). The convection-diffusion

@seismicisolation
@seismicisolation
3.11 The Convection-Diffusion Problem 169

problem reads

. − Δu + σ · ∇u + cu = f in Ω (3.11.1)
.u = 0 on Γ = ∂Ω. (3.11.2)

In (3.11.1), we have a diffusion term .−Δu and a transport term .σ · ∇u in the same
equation that compete with each other.

3.11.1 Variational Formulation

The variational formulation for the problem (3.11.1)-(3.11.2) is given by: Find .u ∈
V = H01 (Ω) such that
.a(u, v) = L(v)

for all .v ∈ V , where

{
a(u, v) =
. (∇u · ∇v + (σ · ∇u + cu)v) d x
Ω

and {
. L(v) = f v d x.
Ω

Note that the bilinear form is not symmetric.

3.11.2 Existence and Uniqueness

Let us give a first existence and uniqueness result.

Theorem 3.11.1 Let .Ω be a bounded open subset of .Rn , . f ∈ L 2 (Ω, R), .σ ∈
C 1 (Ω; Rn ), and .c ∈ L ∞ (Ω, R). We suppose that .c − 21 divσ ≥ 0. Then, the prob-
lem: Find .u ∈ V such that
{ {
. (∇u · ∇v + (σ · ∇u + cu)v) d x = f v dx
Ω Ω

for all .v ∈ V has one and only one solution. ♦

Proof We check the hypotheses of the Lax-Milgram theorem (see Theorem 2.17.1).
Consider the bilinear form
{
.a(u, v) = (∇u · ∇v + (σ · ∇u + cu)v) d x.
Ω

@seismicisolation
@seismicisolation
170 3 Variational Formulation of Boundary Problems

It is easy to see the continuity of bilinear form .a(·, ·). We just prove the .V -ellipticity
of .a(·, ·). Indeed,
{
.a(u, u) = (|∇u|2 + (σ · ∇u)u + cu 2 ) d x.
Ω

Integrate the last integral by parts

{ { (E
n
)
∂u
. (σ · ∇u)u d x = σi u dx
Ω Ω i=1
∂ xi
{ (E n
)
∂(σi u)
=− u dx
Ω i=1
∂ xi
{ (E n
) { (En
)
∂σi ∂u
=− u dx −
2
σi u dx
Ω i=1
∂ xi Ω i=1
∂ xi
{ {
=− divσ u 2 d x − (σ · ∇u)u d x,
Ω Ω

since all boundary terms vanish, so that

{ {
1
. (σ · ∇u)u d x = − divσ u 2 d x.
Ω 2 Ω

Therefore,
{ ( ( ) )
1
a(u, u) =
. |∇u|2 + c − divσ u 2 d x
Ω 2
≥ |u|21,Ω .

So, the coercivity by the equivalence of the . H 1 semi-norm and the . H 1 norm on
. H0 (Ω). The linear form
1
{
. L(v) = f v d x,
Ω

is continuous. In fact, let .v ∈ H01 (Ω). Then,

{
.|L(v)| ≤ | f | |v| d x
Ω
≤ || f ||0,Ω ||v||0,Ω
≤ || f ||0,Ω ||v||1,Ω .

Thus, . L(·) is continuous. Now, the result follows by using Lax-Milgram theorem
(see Theorem 2.17.1). Q.E.D.

@seismicisolation
@seismicisolation
3.12 A Slight Variant of the Plate Problem with Homogeneous … 171

Remark 3.11.2 We therefore have existence and uniqueness if .c = 0 and .divσ =

0. The case .divσ = 0 is interesting because if .σ represents the velocity field of
a fluid, such as air or water, the condition without divergence is the expression of
incompressibility of the fluid. Under the usual experimental conditions, the two fluids
are indeed considered as incompressible. ♦

3.12 A Slight Variant of the Plate Problem with

Homogeneous Dirichlet Boundary Conditions

Now let’s give a fourth-order example, even if only second-order problems were
advertised in the title of the section. We consider a slight variant of the plate problem
with homogeneous Dirichlet boundary conditions:

Δ2 u + cu = f in Ω
.

u = 0 on Γ = ∂Ω,
∂u
= 0 on Γ = ∂Ω,
∂ν |Γ

with .Ω is bounded in .Rn (.n = 2 in the real case of a plate).

3.12.1 Variational Formulation

Deriving a variational formulation is again fairly routine, but since this is our first
(and only) fourth-order problem, we give some details. The variational space for
= H02 (Ω) which incorporates the two boundary condi-
this Dirichlet problem is .V N
tions. Suppose .u ∈ H (Ω) H02 (Ω). Then, .Δu ∈ H 2 (Ω) and, we can use Green’s
4

formula
{ {
. (Δ u)v d x =
2
(Δ(Δu))v d x
Ω
{Ω
= Δu Δv d x,
Ω

for all .v ∈ H02 (Ω). So, the variational formulation is given by

{ {
. (Δu Δv + cuv) d x = f v d x, (3.12.1)
Ω Ω

@seismicisolation
@seismicisolation
172 3 Variational Formulation of Boundary Problems

for all .v ∈ V which is easily checked to give rise to a solution of the boundary value
problem.

3.12.2 Existence and Uniqueness

Let’s give a first existence and uniqueness result.

Theorem 3.12.1 Let .Ω be a bounded open subset of .Rn , . f ∈ L 2 (Ω, R), and .c ∈
L ∞ (Ω, R). We suppose that .c ≥ 0. Then, the problem (3.12.1) has one and only one
solution. ♦
Proof We verify the hypotheses of the Lax-Milgram theorem (see Theorem 2.17.1).
Consider the bilinear form
{
.a(u, v) = (Δu Δv + cuv) d x.
Ω

Continuity of a(·, ·): Let .u, .v ∈ H02 (Ω). Then,

.

|{ |
| |
|a(u, v)| = || (Δu Δv + cuv) d x ||
.

{Ω {
≤ |Δu| |Δv| d x + |c(x)||u||v| d x
Ω
{Ω
≤ |u|2,Ω |v|2,Ω + ||c||∞ |u||v| d x
Ω
≤ |u|2,Ω |v|2,Ω + ||c||∞ ||u||0,Ω ||v||0,Ω
≤ (1 + ||c||∞ )||u||2,Ω ||v||2,Ω ,

where
.||c||∞ := ess- sup |c(x)|.
x∈Ω

Thus the continuity of the bilinear form .a(·, ·). We just prove the .V -ellipticity.
Coercivity of a(·, ·): We have
.

{
. a(u, u) ≥ (Δu)2 d x.
Ω

En ∂2ϕ
We argue by density. Let .ϕ ∈ D(Ω), since .Δϕ = i=1 ∂ xi2 , we can write

{ { (E )⎛ n ⎞
n
∂ 2
ϕ E ∂ 2ϕ
. (Δϕ)2 d x = ⎝ ⎠ dx
Ω Ω i=1
∂ xi2 j=1
∂ x 2j

@seismicisolation
@seismicisolation
3.12 A Slight Variant of the Plate Problem with Homogeneous … 173

E n {
∂ 2ϕ ∂ 2ϕ
= dx
i, j=1 Ω
∂ xi2 ∂ x 2j
E n {
∂ϕ ∂ 3 ϕ
=− dx
i, j=1 Ω
∂ xi ∂ xi x 2j
E n {
∂ 2ϕ ∂ 2ϕ
= dx
i, j=1 Ω
∂ xi x j ∂ xi x j

with two successive integrations by parts, the first with respect to .xi and the second
with respect to .x j . Thus, for all .ϕ ∈ D(Ω), we get
{ E n { ( 2 )2
∂ ϕ
. (Δϕ)2 d x = dx
Ω i, j=1 Ω
∂ xi x j

= ||∇ 2 ϕ||20,Ω .

Now, by definition, . H02 (Ω) is the closure of .D(Ω) in . H 2 (Ω), so for all .u ∈ H02 (Ω),
there is a sequence .ϕn ∈ D(Ω) such that .ϕn → u in . H 2 (Ω). Passing to the limit in
the equality above, we get
{
. (Δu)2 d x = ||∇ 2 u||20,Ω .
Ω

Since . ∂∂xiϕxnj → ∂∂xi ux j in . L 2 (Ω, R), then the coercivity follows by the equivalence of
2 2

the semi-norm .||∇ 2 u||0,Ω on . H02 (Ω) and the . H 2 norm (see Lemma 3.4.7). The linear
form {
. L(v) = f v d x,
Ω

is continuous. In fact,
Continuity of L(·): Let .v ∈ H02 (Ω). Then,
.

{
|L(v)| ≤
. | f | |v| d x
Ω
≤ || f ||0,Ω ||v||0,Ω
≤ || f ||0,Ω ||v||2,Ω .

Thus, . L(·) is continuous.

Now, the result follows by using Lax-Milgram theorem (see Theorem 2.17.1).
Q.E.D.

@seismicisolation
@seismicisolation
174 3 Variational Formulation of Boundary Problems

Remark 3.12.2 This is another symmetric problem, so we have an equivalent energy

minimization formulation with
{ {
1
. J (v) = (Δv) d x −
2
f v d x,
2 Ω Ω

to be minimized on . H02 (Ω). ♦

References

1. J. Hadamard, Le problème de Cauchy et les équations aux dérivées partielles linéaires hyper-
boliques (Hermann, Paris, 1932)
2. B. Lucquin, Équations aux dérivées partielles et leurs approximations (Mathématiques à
l’université, Ellipses, Paris, 2004)
3. P.G. Ciarlet, Introduction à l’analyse numérique matricielle et à l’optimisation (Masson, 1990)
4. P. Lesaint, Sur la résolution des systèmes hyperboliques du premier ordre by des méthodes
d’éléments finis. Thèse, Pierre et Marie Curie, Paris (1975)
5. J. Necas, Les méthodes directes en théorie des équations elliptiques. (French) (Masson et Cie,
Éditeurs, Paris; Academia, Éditeurs, Prague, 1967) 351 pp
6. K.T. Smith, Inequalities for formally positive integro-differential forms. Bull. A.M.S. 67, 368–
370 (1961)
7. J.L. Lions, problèmes aux limites dans les équations aux dérivées partielles. Séminaire de math-
ématiques supérieures (été 1962). Les presses de l’université de Monréal (1967)
8. P.-A. Raviart, J.-M. Thomas, Introduction à l’analyse numérique des équations aux dériveées
partielles (Masson, 1983)
9. J.N. Reddy, Applied Functional Analysis and Variational Methods in Engineering (McGraw
Hill, New York, 1986)

@seismicisolation
@seismicisolation
Chapter 4
Introduction to Finite Element

Once we have a weak formulation from a strong formulation, “you just have to” cal-
culate the solution! The finite element method is one of the numerical tools developed
to calculate an approximate solution (see [1]). The finite element method proposes to
set up, on the basis of weak formulations, a discrete algorithm (discretization) mak-
ing it possible to search for an approximate solution of a partial differential problem
on a compact field with conditions at the edges and/or in inside the compact. It is,
therefore, a question of answering the questions of existence and uniqueness of the
solution, of stability and convergence of the numerical methods, as well as of appre-
ciating the error between the exact solution and the approximate solution (indicators
and estimators of error, a priori and a posteriori). The purpose of the error indicators
is to indicate to us if “we are far from the solution”, so that we can modify the model if
necessary. Thus, by successive iterations, one will be able to tend towards a solution
closer and closer to the exact solution.

4.1 Finite Element in .Rn

4.1.1 Lagrange Finite Element

Definition 4.1.1 A Lagrange finite element .(K , Σ, P) corresponds to the following

three data:
.(i) A subset . K of .Rn .
UN
.(ii) A set .Σ = i=1 {ai } of . N distinct points of the set . K , called nodes of the finite
element.
.(iii) A space . P, of dimension . N , of functions (shape functions) defined on the
set . K such that the set .Σ is . P-unisolvent i.e., given . N arbitrary real numbers .αi ,
.1 ≤ i ≤ N , there exists a function . p ∈ P and only one such that

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 175
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_4
@seismicisolation
@seismicisolation
176 4 Introduction to Finite Element

. p(ai ) = αi , 1 ≤ i ≤ N .

There is, therefore, a function . pi ∈ P and only one which check

. pi (a j ) = δi j , 1 ≤ j ≤ N , (4.1.1)

δ being the Kronecker symbol. The . N functions . pi are called the basic functions
. ij

of the finite element and any function . p ∈ P can be written in the form

E
N
. p= p(ai ) pi .
i=1 ♦

We say that a finite element .(K , Σ, P) is quadrilateral if it is equivalent (via a

bijection which is here affine in the case of any space dimension and is an element
-, Σ,
of .(Q 1 )2 when the dimension .n = 2) to a finite element .( K - P)- such that . K
- be
the reference hypercube. The set . K is, then, a parallelotope if the dimension .n is
arbitrary, or an arbitrary quadrilateral in dimension .n = 2 (this quadrilateral will be
a parallelogram if the transformation . F is affine).
Example 4.1.1 Let . K be a triangle, the set .Σ K of the values of the functions at the
vertices of the triangle . K is . P1 -unisolvent.
Example 4.1.2 Let . K be a rectangle, the set .Σ K of the values of the functions at
the vertices of the rectangle . K is . Q 1 -unisolvent, where . Q 1 denotes the space of
polynomials of the form .a + bx + cy + d x y. Indeed,
y

y1

y2

x
x2 x1

let . p(x, y) be a polynomial of . Q 1 zero at the four vertices. We have . p(x1 , y1 ) =

p(x1 , y2 ) = 0, so . p(x1 , y) = 0. Similarly . p(x2 , y) = 0. From which we deduce
. p(x, y) = 0. One thus defines a rectangular finite element with four nodes.

Example 4.1.3 Let . K be a triangle. Let .Σ K be the set of values of the functions at
the points which divide each side in the ratio .α, .1 − α, .0 < α < 1.

@seismicisolation
@seismicisolation
4.1 Finite Element in Rn 177

The set .Σ K is not . P2 -unisolvent, where . P2 denotes the space of polynomials of

degree .≤ 2 in .x, . y (of the form .a + bx + cy + d x y + ex 2 + f y 2 ). Indeed, there is
a conic (curve of order .2) which passes through these six points. So any polynomial
of . P2 which is zero at .5 of these points is necessarily zero at the sixth and therefore
cannot take an arbitrary value at these points.
Example 4.1.4 Let . K be a rectangle. Let .Σ K be the set of values of the shape
functions at the midpoints of the sides. The set .Σ K is not . Q 1 -unisolvent. Indeed,
y

a1
y1

a2 a4

y2
a3
x
x2 x1

looking for the polynomial . p(x, y) such that . p(ai ) = 0, .1 ≤ i ≤ 4, we have, by

making the change of variable

2x − (x1 + x2 )
. ξ= ,
x1 − x2

2y − (y1 + y2 )
η=
. ,
y1 − y2

and setting . - p (ξ, η) = p(x, y), . -

p (0, 1) = -
p (0, −1) = -
p (1, 0) = -
p (−1, 0). How-
ever, . -
p (ξ, η) is of the form .α + βξ + γ η + δξ η, hence
⎧
⎪
⎪ α+γ =0
⎨
α−γ =0
.
⎪
⎪ α+β =0
⎩
α−β = 0.

@seismicisolation
@seismicisolation
178 4 Introduction to Finite Element

If we impose on the values . pi , to satisfy the relation

. p1 + p3 = p2 + p4 ,

we can find a unique . P1 polynomial such that . p(ai ) = pi , and, we have

p1 + p3 p2 − p4 p1 − p3
. p(x, y) = + (2x − (x1 + x2 )) + (2y − (y1 + y2 ))
2 x1 − x2 y1 − y3
or
p1 + p3
. -
p (ξ, η) = + ξ( p2 − p4 ) + η( p1 − p3 ).
2
One thus builds a rectangular finite element with .4 nodes, with constraint.

4.1.2 Regular Triangulations

We say that a finite element .(K , Σ, P) is of simpliciel type if it is equivalent to an

-, Σ,
element.( K - P)- such that. K
- is an.n-simplex. We consider the following geometric
parameters associated with any finite element, simplicial or quadrilateral:

h = element diameters K
. K

and
. K ρ = sup{diameters of spheres contained in K }.

Definition 4.1.2 A family of triangulations.(τh )h is said to be regular if the following

conditions are satisfied:
(i) All the finite elements of the triangulation are equivalent to a single finite element
.
-, Σ,
of reference .( K -
- P),
(ii) The quantity
.
. h = max h K
K ∈τh

tend to zero.
(iii) In the case, where the elements . K ∈ τh are either .n-simplexes or parallelotopes,
.
there exists a constant .c such that for all .h, for all . K ∈ τh , we have

hK
. ≤ c.
ρK

(iv) In the case, where all the elements. K ∈ τh are quadrilaterals in dimension.n = 2,
.

there exist two constants .c and .γ such that for all .h , for all . K ∈ τh ,

@seismicisolation
@seismicisolation
4.1 Finite Element in Rn 179

hK
. ≤ c and max | cos θi | ≤ γ < 1,
ρK 1≤i≤4

where .θi are the angles in the vertices . Ai .1 ≤ i ≤ 4. ♦

4.1.3 . P-Interpolated Function

Definition 4.1.3 Consider a finite element .(K , Σ, P) and a function .v defined on a

set containing the set .Σ. We call . P-interpolated function of .v the only function .π v
of the space which satisfies:

π v(ai ) = v(ai ), 1 ≤ i ≤ N .
. ♦

4.1.4 Element of Reference

Definition 4.1.4 From a Lagrange finite element.( K -, Σ, - called hereafter element

- P)
-
of reference, and from an injective map . F of . K in .R , we define a new Lagrange
n

finite element .(K , Σ, P) by

-),
. K = F( K

| |
N
.Σ= {ai }, ai = F(-
ai ), 1 ≤ i ≤ N ,
i=1

. p ◦ F −1 , ∀-
P = { p : K −→ R such that p = - -
p ∈ P},

where the function . F −1 : K −→ K - being the reciprocal map of the map . F. The
- - -
elements .( K , Σ, P) and .(K , Σ, P) are said to be equivalent. ♦

We constantly use the bijections:

- - −→ x = F(-
x∈K
. x ) ∈ K , x ∈ K −→ - -,
x = F −1 (x) ∈ K

. - - −→ p = -
p∈P p ◦ F −1 ∈ P, p ∈ P −→ - -
p = p ◦ F ∈ P.

Similarly, to any function .- - −→ R (resp. .v : K −→ R), we associate the

v:K
function .v : K −→ R (resp. .- - −→ R) defined by
v:K

. v ◦ F −1 (resp. -
v =- v = v ◦ F). (4.1.2)

@seismicisolation
@seismicisolation
180 4 Introduction to Finite Element

The functions . pi = - pi ◦ F −1 , .1 ≤ i ≤ n, where the . -pi being the basic functions

defined in (4.1.1) constitute a basis of . P. On the other hand, the functions . P-
-
interpolated .π v of .v and . P-interpolated .-
π-
v of .-
v, where the functions .v and .-
v satisfy
the relation (4.1.2), are related by the relation

π
-v = -
. π-
v.

In .R2 , the transformations . F used hereafter is the following: The transformation

.F : K- −→ R2 is the restriction to the set . K
- of an invertible affine map i.e., the . F
map is of the form:
.F : -x∈K - −→ F(- x ) = B-x + b,

where . B is an invertible element of .L (R2 ), and .b is a vector of .R2 .

We have the result.
Lemma 4.1.5 Let .(τh )h be a regular family of triangulations with reference element
-, Σ,
(K
. - P)- and let . JK be the Jacobian of the transformation . FK which sends the ref-
erence element . K- to the element . K , for . K ∈ τh and . JS the Jacobian of the restriction
of . FK to a .(n − 1)-any face . S of . K . Then, the following inequalities are satisfied
( )n ( )n
ρK ρK
. c ≤ JK ≤ C ,
hK hK
( )n−1 ( )n−1
ρK ρK
. ≤ JS ≤ ,
hK hK

where the constants .c and .C such that .0 < c < C are independent of .h. ♦

4.1.5 Change of Variable Results

The following change of variable results are demonstrated in [2, 3] and in [4].
Lemma 4.1.6 Let .(τh )h be a regular triangulation family of reference elements
-, Σ,
(K
. - We assume the transformation . FK , of Jacobian . JK , which sends the
- P).
- on . K is affine. Consider a function .-
element . K -), for an integer .k ≥ 0
u ∈ W k,q ( K
−1
and for .q ≥ 1. Then, the function .u = -u ◦ FK belongs to .W k,q (K ), and, we have

u |k,q, K- ≤ C(JK )− q (h K )k |u|k,q,K ,

1
.|-

1
. |u|k,q,K ≤ C(JK ) q (ρ K )−k |-
u |k,q, K- ,

where .C denotes a constant independent of .h. We have the analogous inequalities

by replacing . K by any .(n − 1)-face . S of . K . ♦

@seismicisolation
@seismicisolation
4.1 Finite Element in Rn 181

In the case of quadrilateral elements in dimension .n = 2, we have

Lemma 4.1.7 Let .(τh )h be a regular family of triangulations of .Ω ⊂ R2 , with ref-
erence element .( K-, Σ, - We suppose that the transformation . FK , of Jacobian
- P).
- on . K is an element of .(Q 1 )2 . Consider a function
. J K , which sends the element . K
k,q -
.-
u ∈ W ( K ), for an integer .k ≥ 0 and for .q ≥ 1. Then, the function .u = - u ◦ FK−1
belongs to .W (K ) and, there exists a constant .C independent of .h such that
k,q

( )− q1
[-
.u ]k,q, K- ≤ C inf JK (-
x) h kK |u|k,q,K , for k ≥ 0,
- -
x ∈K

( )− q1 E
k
.|-
u |k,q, K- ≤ C inf JK (-
x) h 2l−k
K K |u|l,q,K , for k ≥ 1,
Z kl (4.1.3)
- -
x ∈K
l=l0

where .l is the smallest integer greater than or equal to . 2k , . Z K is the distances between
the midpoints of the diagonals of . K , and

( )− q1
|-
u |0,q, K- ≤ C inf JK (-
x) |u|0,q,K .
♦
.
- -
x ∈K

4.1.6 Polyhedral Open Set

Definition 4.1.8 We say that an open set .Ω of .Rn is a polyhedral open set if the set
.Ω is a polyhedron. ♦
We, then, define a triangulation .τh of the polyhedral open set .Ω, by giving a certain
number of finite elements .(K , Σ, P), . K ∈ τh , such that the following conditions are
satisfied:
U
.(i) .Ω = K ∈τh K ,
◦ N ◦
.(ii) . K 1 K 2 = ∅ if . K 1 /= K 2 ,
(iii) All elements of the triangulation are polyhedra,
.

.(iv) Any .(n − 1)-face of a finite element . K 1 is either an .(n − 1)-face of another
element . K 2 , the elements . K 1 and . K 2 being then adjacent i.e., part of the boundary
.∂Ω of the open set .Ω,

(v) Whenever two finite elements . K 1 and . K 2 are adjacent, we have

.

N N
.ΣK1 K 2 = ΣK2 K1.

To a triangulation .τh , we associate the set

@seismicisolation
@seismicisolation
182 4 Introduction to Finite Element
| |
Σh =
. ΣK
K ∈τh

which together are called nodes of the triangulation. We define the function spaces
Wh and .Vh by
.
| |
. Wh = PK ,
K ∈τh

Vh = the subspace of .Wh made up of continuous functions at the nodes of the tri-
.

angulation i.e., for any .v ∈ Vh and for any pair of adjacent elements . K 1 and . K 2 ,
we have N N
.v|K 1 (a) = v|K 2 (a), ∀a ∈ Σ K 1 K 2 = ΣK2 K1.

Definition 4.1.9 Let .v be a function defined on .Σh , the function called .Vh -
interpolated function of the function .v, is the only function of the space .Vh which
check the relations
.∀a ∈ Σh , πh v(a) = v(a),

or equivalently, the relations

.∀K ∈ τh , πh v|K = π K v. ♦

4.1.7 Construction of the Approximate Space . Vh

In general, the functions of the space .Vh are not defined on the set .Ω since they do not
have a unique determination along the faces common to the adjacent finite elements,
except for the nodes (this is, in particular, the case when the space .Vh is built from the
non-conforming quadrilateral finite element). The two following definitions make it
possible to state conditions guaranteeing the inclusion

. Vh ⊂ C 0 (Ω, R).

Definition 4.1.10 A finite element .(K , Σ, P) is of class .C 0 if the following

conditions are satisfied:
.(i) . P ⊂ C 0 (K , R), n ,
,
.(ii) If . K is any .(n − 1)-face of the set . K , the set .Σ|K , = Σ K is . P|K , -unisolvent,
where . P|K , = { p|K , , ∀ p ∈ P}. ♦

@seismicisolation
@seismicisolation
4.1 Finite Element in Rn 183

Definition 4.1.11 A triangulation is of class .C 0 if the following two conditions are

satisfied:
(i) All the finite elements of the triangulation are of class .C 0 ,
.

(ii) For any pair of adjacent finite

.
n elements . K 1 and . K 2 of the triangulation, we have
PK 1 |K , = PK 2 |K , with . K , = K 1 K 2 .
. ♦

Let .τh be a triangulation of class .C 0 of an open polyhedral .Ω, the associated space
. Vh , constructed as above, check the inclusion

. Vh ⊂ C 0 (Ω, R).

If we, further, suppose that the inclusions . PK ⊂ H 1 (K ), . K ∈ τh are satisfied, we,

also, have
. Vh ⊂ H (Ω).
1

A finite element equivalent to a finite element of class .C 0 is also of class .C 0 . This,

therefore, makes it possible to construct the space .Vh from a finite element of ref-
-, Σ,
erence .( K - and from a family of transformation . FK , . K ∈ τh , such as those
- P)
described above.
Let .∂Ω be the boundary of the polyhedral open set .Ω. The following result on traces
is a consequence of [5, Theorem 5.5 p. 99].
Lemma 4.1.12 Consider a triangulation.τh of.Ω. Let.m ≥ 0 be an integer and.q > 1
be a number. Then, for any function .u ∈ W m+1,q (Ω), we have
⎛ ⎞ q1
E E
. ⎝ |u|m,q,S ⎠ ≤ C(|u|m,q,Ω + |u|m+1,q,Ω ),
n
K ∈τh S⊂∂Ω ∂K

where . S denotes any .(n − 1)-face of an element . K ∈ τh and, where the constant .C
does not depend on the triangulation .τh . ♦
- be a non-degenerate .n-simplex of .Rn , with vertices .-
Example 4.1.5 Let . K ai .1 ≤
i ≤ n + 1. Any point .-
x∈K- admits a unique representation of the form

E
n+1
-
x=
. λi (-
x )-
ai ,
i=1

with .0 ≤ λi ≤ 1, .1 ≤ i ≤ n + 1, where the .λi are the barycentric coordinates of

the point .-
x with respect to the points .-ai , .1 ≤ i ≤ n + 1. For any integer .k ≥ 1, the
principal lattice of order .k of the .n-simplex . K- is the set
{ { } }
E
n+1 E
n+1
1 k−1
-k
.Σ = -
x= λi -
ai , λi = 1, λi ∈ 0, , . . . , ,1 , 1 ≤ i ≤ n + 1 .
k k
i=1 i=1

@seismicisolation
@seismicisolation
184 4 Introduction to Finite Element

One can show [6] that the set .Σ -, Σ

-k is . Pk -unisolvent. The triplet .( K -k , Pk ), therefore,
constitutes a Lagrange finite element. The finite elements .(K , Σ, P) equivalent to
-, Σ
the triplet .( K -k , Pk ) are called .n-simplexes of type .k.

Example 4.1.6 We consider finite elements defined on hypercube. Let . Q k be the

space of polynomials, in .Rn , of the form
E
. ai1 i2 ···in x1i1 x2i2 · · · xnin ,
i 1 ,...,i n ≤k

- ⊂ Rn , where reference square if

with .ai1 i2 ···in ∈ R. We call reference hypercube . K
.n = 2, the set
- = [−1, 1]n .
.K

For any integer .k ≥ 1, we define the set

{ { } }
-k 2 2
.H = -x = (-
x1 , . . . , -
x n ) ∈ Rn , -
x j ∈ −1, −1 + , . . . , 1 − , 1 , 1 ≤ j ≤ n . (4.1.4)
k k

We show [7] that the set . H -, H

-k is . Q k -unisolvent. The triplet .( K -k , Q k ), therefore,
defines a Lagrange finite element. We call quadrilateral of type .k any finite element
.(K , Σ, P) equivalent via the transformation mentioned above to the finite element
-, H
.( K -k , Q k ). Quadrilaterals of type .k are elements of class .C 0 . We are particularly
interested hereafter in the case of the dimension .n = 2. We note .ξ and .η (resp. .x and
. y) the coordinates of any point . M - (resp. . M) of the element . K- (resp. . K ). Let .(xi , yi )
be the coordinates of the vertices . Ai , .1 ≤ i ≤ 4, of the quadrilateral . K , range of the
- of vertices . A
square . K -i , .1 ≤ i ≤ 4. The transformation . F will be written

1 1 1 1
.x = (1 + ξ )(1 + η)x1 + (1 − ξ )(1 + η)x2 + (1 − ξ )(1 − η)x3 + (1 + ξ )(1 − η)x4
4 4 4 4

1 1 1 1
.y = (1 + ξ )(1 + η)y1 + (1 − ξ )(1 + η)y2 + (1 − ξ )(1 − η)y3 + (1 + ξ )(1 − η)y4 .
4 4 4 4

The functions . p of the space . P are expressed in a very simple way, in coordinates .ξ
and .η, according to their values at the points of the set .Σ. In the case, where .k = 1,
we have in particular

1 1
. p(x, y) = -
p (ξ, η) = (1 + ξ )(1 + η) p(A1 ) + (1 − ξ )(1 + η) p(A2 )+
4 4
1 1
(1 − ξ )(1 − η) p(A3 ) + (1 + ξ )(1 − η) p(A4 ),
4 4

with .(x, y) = F(ξ, η). In the case, where .k = 2, we have

@seismicisolation
@seismicisolation
4.1 Finite Element in Rn 185

1 1
. p(x, y) = -
p (ξ, η) = ξ η(1 + ξ )(1 + η) p(A1 ) + ξ η(ξ − 1)(1 + η) p(A2 )
4 4
1 1 1
+ ξ η(1 − ξ )(1 − η) p(A3 ) + ξ η(1 + ξ )(η − 1) p(A4 ) + η(1 + η)(1 − ξ 2 ) p(A5 )
4 4 2
1 1 1
+ η(η − 1)(1 − ξ 2 ) p(A7 ) + ξ(ξ − 1)(1 − η2 ) p(A6 ) + ξ(1 + ξ )(1 − η2 ) p(A8 )
2 2 2
+ (1 − ξ 2 )(1 − η2 ) p(A9 ),

with .(x, y) = F(ξ, η).

η y

-2
A -1
A A2 A5
1 A1
-
R R
A6 A9
A8
−1 1 ξ

A3 A7 A4

-3
A −1 -4
A x

4.1.8 Error Estimates

Lemma 4.1.13 Let .Ω be an open bounded domain of .Rn , .C 1 per piece, with piece-
wise smooth boundary .Γ = ∂Ω. Let .(τh )h be a regular family of triangulations of
.Ω. Then, for all element . K ∈ τh , we have

| p|1,q,K ≤ c(h K )−1 | p|0,q,K for all p ∈ Pk ,

.

| p|0,q,∂ K ≤ c(h K )− q | p|0,q,K for all p ∈ Pk ,

1
.

for all .q ≥ 1. ♦
The following approximation results, some of which are proved in [8, 9], are useful
for obtaining the error bounds, in what follows.
Lemma 4.1.14 Let .Ω be an open bounded domain of .Rn , .C 1 per piece, with piece-
wise smooth boundary .Γ = ∂Ω. Let .(τh )h be a regular family of triangulations of .Ω,
in simplicial elements (resp. quadrilaterals in dimension .n = 2). We assume that the
- (resp. . Q k ⊂ P)
inclusion . Pk ⊂ P - is satisfied, for an integer .k ≥ 1. Let .s be an inte-
ger such that .1 ≤ s ≤ k + 1 and a number .q ≥ 1, with . q1 − ns < 0. To any function

@seismicisolation
@seismicisolation
186 4 Introduction to Finite Element

.v ∈ W s,q (Ω), we associate its interpolated .πh v ∈ Vh . Then, there exists a constant
.c > 0 independent of .h such that, for all . K ∈ τh , we have, for all .v ∈ W (Ω)
s,q

.|v − πh v|m,q,K ≤ ch s−m

K |v|s,q,K , for 0 ≤ m ≤ s,

s−m− q1
.|v − πh v|m,q,S ≤ ch K |v|s,q,K , for 0 ≤ m ≤ s − 1,

where . S is any face of element . K . If, further, the elements . K are of class .C 0 , we have
for all . K ∈ τh ,

|v − πh v|m,q,S ≤ ch s−m
. K |v|s,q,S , for 0 ≤ m ≤ s,

where . S is any face of . K , for all . K ∈ τh . ♦

4.2 Galerkin Method

4.2.1 Description of the Galerkin Method

We start this section with Galerkin’s method which provides the abstract framework
for studying the convergence of the approximation of the solution of boundary prob-
lems by the finite element method. The space . X , introduced in the Lax-Milgram
theorem (Theorem 2.17.1), is, generally, of infinite dimension. To solve the problem
numerically: Find .u ∈ X such that

a(u, v) = L(v)
. (4.2.1)

for all .v ∈ X , we are led to look for an approximate solution .u h belonging to a

subspace .Vh ⊂ X of finite dimension, let find .u h ∈ Vh such that

.a(u h , vh ) = L(vh ) (4.2.2)

for all .vh ∈ Vh . The problem (4.2.2) is equivalent to the resolution of a linear system.
In fact, let .{w j }1≤ j≤N be a basis of .Vh (with .dim Vh = N ). We decompose .u h on
this basis
EN
.u h = ξi wi . (4.2.3)
i=1

Using (4.2.3), the problem (4.2.2) consiste to find the vector .(ξ1 , . . . , ξ N ) such that

E
N
. ξi a(wi , w j ) = L(w j ) 1 ≤ j ≤ N . (4.2.4)
i=1

@seismicisolation
@seismicisolation
4.2 Galerkin Method 187

Theorem 4.2.1 We assume that the space . X , the bilinear form .a(·, ·) and the linear
form . L(·), satisfy the hypotheses of the Lax-Milgram theorem (Theorem 2.17.1).
Then, the problem (4.2.2) admits a unique solution .u h ∈ Vh . Moreover,

1
.||u h || ≤ ||L||. ♦
α

Proof The space .Vh being of finite dimension, is a closed subspace of . X , and it
is therefore, a Hilbert space for the norm induced by that of . X . We can, therefore,
apply the Lax-Milgram theorem (Theorem 2.17.1) with . X replaced by .Vh . Which
completes the proof. Q.E.D.

Remark 4.2.2 .(i) If the bilinear form.a(·, ·) is symmetrical, then the problem (4.2.1)
is equivalent to find .u ∈ X such that

. J (u) = inf J (v),

v∈X

with
1
. J (v) = a(v, v) − L(v).
2
The problem (4.2.2) is equivalent to find .u h ∈ Vh such that

. J (u h ) = inf J (vh )
h∈Vh

and, we have
. J (u h ) ≥ J (u).

(ii) The problem (4.2.2) admits a unique solution in .Vh . In fact, let .u (1)
.
(2)
h and .u h be
two solutions of the problem (4.2.2), we have

a(u (1)
. h , vh ) = L(vh )

for all .vh ∈ Vh and

a(u (2)
. h , vh ) = L(vh )

for all .vh ∈ Vh . By making the difference of the last two equalities, we deduce

a(u (1)
.
(2)
h − u h , vh ) = 0

for all .vh ∈ Vh . In particular for .vh = u (1) (2)

h − u h , we obtain

.a(u (1) (2) (1) (2)

h − u h , u h − u h ) = 0.

@seismicisolation
@seismicisolation
188 4 Introduction to Finite Element

Since the bilinear form .a(·, ·) is coercive so, we have

. ||u (1) (2)

h − u h || = 0.

Thus,
u (1) = u (2)
. h h .

(iii) The resolution of the linear system (4.2.4) is as much simpler as it is fast than
.
the matrix
. A = (a(wi , w j ))1≤i, j≤N

has a “pleasant” structure (band matrix for example). We will see in the continuation
on some simple examples that the structure of this matrix depends on the choice of
the basis .{w j }1≤ j≤N for this subspace of finite dimension. ♦

Theorem 4.2.3 (Galerkin orthogonality) Let .(X, (·, ·)) be a real Hilbert space,
a(·, ·) be a continuous bilinear form on . X , . f ∈ X , be a continuous linear form
.
on . X , .Vh ⊂ X be a subspace of finite dimension of . X , .u ∈ X be a solution of (4.2.1),
and .u h ∈ Vh be a solution of (4.2.2). Then, for all .vh ∈ Vh , we have

a(u − u h , vh ) = 0.
. (4.2.5)

Proof Let .vh ∈ Vh . The fact that .Vh ⊂ X , implies that

a(u − u h , vh ) = a(u, vh ) − a(u h , vh )

.

= L(vh ) − L(vh )
= 0.

This completes the proof. Q.E.D.

Remark 4.2.4 If the bilinear form .a(·, ·) is symmetric, it introduces an energetic

inner product
.(·, ·)a := a(·, ·).

It follows from (4.2.5) that

(u − u h , vh )a = 0 for all vh ∈ Vh ,
.

i.e., that the error of the Galerkin approximatione, .u − u h , is orthogonal to the

Galerkin subspace.Vh , in the energetic inner product. Hence, the approximate solution

@seismicisolation
@seismicisolation
4.2 Galerkin Method 189

u ∈ Vh , is an orthogonal projection of the exact solution .u ∈ X onto the Galerkin

. h

subspace .Vh , in the energetic inner product, and thus it is the nearest element in the
space .Vh , to the exact solution .u in the energy norm

a(u − u h , u − u h ) = inf a(u − vh , u − vh ).

. ♦
vh ∈Vh

In contrast to the conforming setting, the well-posedness of (4.2.2) cannot be deduced

from the well-posedness of (4.2.1), but needs to be proved independently. This is
somewhat simpler due to the finite-dimensionality of the space.

Theorem 4.2.5 Let .a : Vh × Vh −→ R be a bilinear form and . L : Vh −→ R be a

linear functional satisfying the following assumptions:
(i) Inf-sup-condition: There exists a .c1 > 0 such that
.

a(u h , vh )
. inf sup ≥ c1 .
u h ∈Vh vh ∈Vh ||u h ||||vh ||

(ii) Continuity: There exist .c2 , .c3 such that

.

. |a(u h , vh )| ≤ c2 ||u h ||||vh ||,

|L(vh )| ≤ c3 ||vh ||
.

for all .u h ∈ Vh , .vh ∈ Vh .

Then, there exists a unique solution .u h ∈ Vh of the problem (4.2.2) satisfying

1
||u h || ≤
. ||L||. ♦
c1

Proof Consider a basis .{ϕ1 , . . . , ϕn } of .Vh and define the matrix . K = (K i j ) ∈ Rn×n ,
. K i j = a(ϕi , ϕ j ). Then, the claim is equivalent to the invertibility of . K . From the inf-

sup-condition, we obtain injectivity of . K by arguing as in the continuous case. By

the rank theorem, this implies surjectivity of . K and hence invertibility. The estimate
follows again from the inf-sup-condition. Q.E.D.

Remark 4.2.6 Note the difference between Theorem 4.2.5 and the Lax-Milgram
theorem in the discrete case: In the latter, the coercivity condition amounts to the
assumption that the matrix . K is positive definite, while the inf-sup and injectivity
condition only amounts to requiring invertibility. ♦

@seismicisolation
@seismicisolation
190 4 Introduction to Finite Element

4.2.2 Error Estimates

We now give an abstract theorem for the increase of the error between the exact
solution .u of the problem (4.2.1) and the approximate solution .u h of the problem
(4.2.2). This very simple but also very important result is due to Céa [10].
Theorem 4.2.7 Let.u be the solution of the problem (4.2.1) and let.u h be the solution
of the problem (4.2.2). Then, we have the following general estimates for the errors

M
||u − u h || ≤
. inf ||u − vh || (4.2.6)
α vh ∈Vh

with .α is the constant of coercivity of .a(·, ·) and . M is the constant of continuity of

a(·, ·).
. ♦

Proof For all .vh ∈ Vh , we have

.α||u − u h ||2 ≤ a(u − u h , u − u h )

= a(u − u h , u − vh ) + a(u − u h , vh − u h ).

We can write
.a(u, vh − u h ) = L(vh − u h )

because .vh − u h ∈ Vh ⊂ X and

. a(u h , vh − u h ) = L(vh − u h ).

Thus,

α||u − u h ||2 ≤ a(u − u h , u − vh )

.

≤ M||u − u h ||||u − vh ||.

We can deduce
M
||u − u h || ≤
. ||u − vh ||
α
for all .vh ∈ Vh . Which leads to the inequality (4.2.6). Q.E.D.

Remark 4.2.8 .(i) Céa’s theorem (Theorem 4.2.7) shows that .u h is quasi-optimal
in the sense that the error .||u − u h || is proportional to the best it can be using the
subspace .Vh .
(ii) Céa’s theorem (Theorem 4.2.7) states that the approximation error .u − u h ,
.
depends on the choice of the Galerkin subspace .Vh , but it does not depend on the
choice of its basis.

@seismicisolation
@seismicisolation
4.2 Galerkin Method 191

(iii) When the bilinear form .a(·, ·) is symmetrical, we can get a better estimate
.

/
M
||u − u h || ≤
. inf ||u − vh ||. (4.2.7)
α vh ∈Vh

In fact, we have
a(u − u h , wh ) = 0
.

for all .wh ∈ Vh . Hence, .u h is the projection of .u on .Vh , within the meaning with the
following inner product
.(·, ·)a := a(·, ·)

we, then, have

. a(u − u h , u − u h ) = inf a(u − vh , u − vh ).
vh ∈Vh

Thus, we can deduce the inequality (4.2.7). ♦

Often it is more useful to estimate the error in a weaker norm. This requires a
duality argument. Let . H be a Hilbert space with inner product .(·, ·) and . X be a
closed subspace satisfying the conditions of the Lax-Milgram theorem such that
the embedding .i : X −→ H is continuous (e.g., . X = H 1 (Ω) C→ L 2 (Ω, R) = H ).
Then, we have the following estimate.

Lemma 4.2.9 (Aubin-Nitsche lemma) Let .u h be the solution of (4.2.2) for given
Vh ⊂ X and .u be the solution of (4.2.1). Then, there exists a .C > 0 such that
.

( )
1
.||u − u h || H ≤ C||u − u h || X sup inf ||ϕg − vh || X
g∈H ||g|| H vh ∈Vh

holds, where for given .g ∈ H , .ϕg is the unique solution of the adjoint problem

a(w, ϕg ) = (g, w) H for all w ∈ X.

.

Furthermore, since .a(·, ·) is symmetric, then the existence of a unique solution of the
adjoint problem is guaranteed by the Lax-Milgram theorem. ♦

Proof We make use of the dual representation of the norm in any Hilbert space,

(g, w) H
||w|| H = sup
. , (4.2.8)
g∈H ||g|| H

where the supremum is taken over all.g /= 0. Now, inserting.w = u − u h in the adjoint
problem, we obtain for any .vh ∈ Vh , using the Galerkin orthogonality and continuity
of .a(·, ·), that

@seismicisolation
@seismicisolation
192 4 Introduction to Finite Element

(g, u − u h ) H = a(u − u h , ϕg )
.

= a(u − u h , ϕg − vh )
≤ C||u − u h || X ||ϕg − vh || X .

Inserting .w = u − u h into (4.2.8), we thus obtain

(g, u − u h ) H
.||u − u h || H = sup
g∈H ||g|| H
||ϕg − vh || X
≤ C||u − u h || X sup
g∈H ||g|| H

for arbitrary .vh ∈ Vh , and taking the minimum over all .vh yields the desired estimate.
Q.E.D.

Remark 4.2.10 The Aubin-Nitsche lemma also holds for nonsymmetric .a(·, ·),
provided both the original and the adjoint problem satisfy the conditions of the
Lax-Milgram theorem. ♦

4.3 Examples in Dimension .1 and Introduction to Finite

Elements

We consider the following equation

d 2u
. − = f for x ∈]0, 1[ (4.3.1)
dx2
.u(0) = u(1) = 0. (4.3.2)

In variational form, the problem (4.3.1)–(4.3.2) is formulated: Find .u ∈ H01 (]0, 1[)
such that
.a(u, v) = ( f, v)0,(0,1)

for all .v ∈ H01 (]0, 1[) with

{ 1
du dv
. a(u, v) = dx
0 dx dx

and { 1
( f, v)0,(0,1) =
. f v d x.
0

@seismicisolation
@seismicisolation
4.3 Examples in Dimension 1 and Introduction to Finite Elements 193

4.3.1 Example 1

Let .Vh be the space generated by the functions

. wk = x(1 − x)x k , 0 ≤ k ≤ N − 1.

Since .wk (0) = wk (1) = 0, then .wk ∈ H01 (]0, 1[). The elements

a = a(wk , wl )
. kl

of the matrix . A, of the linear system (4.2.4), are given by

{ 1
dwk dwl
a =
. kl dx
0 dx dx
2(l + 1)(k + 1)
= .
(k + l + 1)(k + l + 2)(k + l + 3)

This very simple choice for the space .Vh and for the basis leads to a full matrix.

4.3.2 Example 2

Let .Vh be the space generated by the functions

wk = sin(kπ x), 1 ≤ k ≤ N .
.

The elements .akl of the matrix . A are given by

{ 1
dwk dwl
a =
. kl dx
0 dx dx
{ 1
= klπ 2 cos(kπ x) cos(lπ x) d x.
0

Thus, ⎧
⎨0 if k /= l
.akl := k2π 2
⎩ if k = l.
2

We, thus, obtain a diagonal matrix. This comes in fact from the exceptional choice
of the basis of the space .Vh , the basis functions chosen being the . N first proper
functions of the operator
d2
.−
dx2

associated with the Dirichlet boundary conditions (.u(0) = u(1) = 0).

@seismicisolation
@seismicisolation
194 4 Introduction to Finite Element

4.3.3 Example 3

We now consider the problem with variable coefficients

( )
d du
.− a(x) = f for x ∈]0, 1[
dx dx
u(0) = u(1) = 0

with which we associate the bilinear form

{ 1
du dv
.a(u, v) = a(x) d x.
0 d x dx

Let .Vh be the space generated by the functions, given in Example 2,

wk = sin(kπ x), 1 ≤ k ≤ N .
.

The elements .akl of the matrix . A are given by

{ 1
dwk dwl
a =
. kl a(x) dx
0 dx dx
{ 1
= a(x)klπ 2 cos(kπ x) cos(lπ x) d x.
0

We can see that, in general, the elements .akl are all nonzero. This shows that the
choice of the basis carried out above (Example 2), is not adapted to all the problems.

4.3.4 Example 4 (General Case)

We divide the interval .[0, 1] into . I intervals of length .h. We pose .xi = i h, for .0 ≤
i ≤ I . Let

. Wh = {vh ∈ C 0 ([0, 1], R) such that vh is linear on each [xi , xi+1 ]}.

We pose .vh (xi ) = vi so that on each interval .[xi , xi+1 ], we can write

x − xi
v (x) = vi +
. h (vi+1 − vi ).
h
We can define the space .Wh , in an equivalent way, by

@seismicisolation
@seismicisolation
4.3 Examples in Dimension 1 and Introduction to Finite Elements 195

Wh = {vh ∈ C 0 ([0, 1], R) such that vh |(xi ,xi+1 ) ∈ P1 , vh is fully determined

.
by its values vi = vh (xi ) at points, xi 0 ≤ i ≤ I }.

We pose
. Vh = {vh ∈ Wh such that vh (0) = vh (1) = 0}.

A basis of the space .Vh is composed of functions .ϕ j (x) ∈ Vh equal to .1 at .x j and to

zero in all .xi .i /= j. We obtain the hats functions.

ϕj

0 x j−1 xj x j+1 1

Remark 4.3.1 The n fact that.Wh ⊂ H 1 (]0, 1[) gives the following inclusion of spaces
. Vh ⊂ H0 (]0, 1[) C ([0, 1], R). ♦
1 0

The elements .akl of the matrix . A are equal to zero as soon as the supports of the
functions .ϕk (x) and .ϕl (x) are disjoint, i.e., as soon as

|k − l| ≥ 2.
.

Therefore, to find the matrix . A, it suffices to look for the elements .akk , .ak k+1 and
a .
. k k−1

Let us find .akk for .k = 1, . . . , I − 1: We have

{ xk ( )2 { xk+1 ( )2
dϕk dϕk
a
. kk = dx + d x.
xk−1 dx xk dx

Let’s look for the functions .ϕk (x) for .k = 1, . . . , I − 1. In fact,

ϕk (x)

h h

xk−1 xk xk+1

on .]xk−1 , xk [, .ϕk (x) = ax + b. We have .ϕk (xk−1 ) = 0 so

axk−1 + b = 0.
. (4.3.3)

@seismicisolation
@seismicisolation
196 4 Introduction to Finite Element

We have .ϕk (xk ) = 1 so

axk + b = 1.
. (4.3.4)

By differentiating between (4.3.3) and (4.3.4), we deduce

a(xk − xk−1 ) = 1.
.

Which proves .ah = 1 and hence .a = h1 . Equation (4.3.4) gives

xk
b =1−
.
h
kh
=1−
h
= 1 − k.

Thus,
1
ϕ (x)|]xk−1 ,xk [ =
. k x + 1 − k.
h

On .]xk , xk+1 [, .ϕk (x) = ax + b. We have .ϕk (xk ) = 1 so

axk + b = 1.
. (4.3.5)

We have .ϕk (xk+1 ) = 0 hence

axk+1 + b = 0.
. (4.3.6)

By differentiating between (4.3.5) and (4.3.6), we deduce

a(xk+1 − xk ) = −1.
.

Which proves .ah = −1 and so .a = − h1 . Equation (4.3.6) gives

b = −axk+1
.

1
= (1 + k)h
h
= 1 + k.

Hence,
1
ϕ (x)|]xk ,xk+1 [ = − x + 1 + k.
. k
h
So,

@seismicisolation
@seismicisolation
4.3 Examples in Dimension 1 and Introduction to Finite Elements 197
⎧
⎪
⎪0 if x ∈ [x0 , xk−1 ]
⎪
⎪
⎪ 1
⎨ x +1−k if x ∈ [xk−1 , xk ]
.ϕk (x) := h
⎪
⎪ −
1
x +1+k if x ∈ [xk , xk+1 ]
⎪
⎪
⎪
⎩ h
0 if x ∈ [xk+1 , x I ].

Thus,
dϕk 1
. =
d x |]xk−1 ,xk [ h

and
dϕk 1
. =− .
d x |]xk ,xk+1 [ h

On the other hand,

{ ( )2 { xk+1 ( )
xk
1 1 2
a
. kk = dx + − dx
xk−1 h xk h
2
= .
h
Let’s seek .ak k+1 for .k = 1, . . . , I − 2: We have
{ xk+1
dϕk dϕk+1
a
. k k+1 = dx
xk dx dx
{ xk+1
−1 1
= dx
xk h h
1
=− .
h
Let’s seek .ak k−1 for .k = 2, . . . , I − 1: We have
{ xk
dϕk−1 dϕk
a
. k k−1 = dx
xk−1 dx dx
{ xk
−1 1
= dx
xk−1 h h
1
=− .
h
The functions .{ϕi }1≤i≤I −1 form a basis of .Vh . Thus,

I −1
E
u =
. h u j ϕ j (x)
j=1

@seismicisolation
@seismicisolation
198 4 Introduction to Finite Element

with .u j = u h (x j ). It follows that .u h verify the following linear system:

.a(u h , ϕ j ) = L(ϕ j )

for all . j = 1, . . . , I − 1. This implies that

I −1
E
. u i a(ϕi , ϕ j ) = L(ϕ j )
i=1

for all . j = 1, . . . , I − 1. Hence,

{ x j+1
1 2 1
. − u j−1 + u j − u j+1 = f ϕj dx
h h h x j−1

for all . j = 1, . . . , I − 1. So, we have to solve the linear system

{ x j+1
1 1
. − (u j−1 − 2u j + u j+1 ) = f ϕ j d x, j = 1, . . . , I − 1. (4.3.7)
h2 h x j−1
.u 0 = u I = 0. (4.3.8)

Which proves that the matrix of the system is tridiagonal.

Remark 4.3.2 .(i) If one modifies the classification of the basic functions .ϕ j (x), the
matrix of the linear system (4.3.7)–(4.3.8) loses its tridiagonal structure.
.(ii) If we change the basis

ψ2 ψi

x0 = 0 x1 x2 x3 x4 xi 1
ψ1 common support

(ψi )i is a basis of .Vh . Hence, the matrix of the system is full because .a(ψi , ψ j ) /= 0
.

since all .ψi have a common support .]0, x2 [. ♦

Remark 4.3.3 Consider the expression

{ x j+1
1 1
. E j (u) = − (u(x j−1 ) − 2u(x j ) + u(x j+1 )) − f ϕ j d x,
h2 h x j−1

@seismicisolation
@seismicisolation
4.3 Examples in Dimension 1 and Introduction to Finite Elements 199

where .u is the exact solution of the problem (4.3.1)–(4.3.2). If we consider the Taylor
expansions of .u(x j+1 ) and .u(x j−1 ) around the point .x j , we obtain if the solution
belongs to .C 4 ([0, 1]), then

1 d 2u
. − (u(x j−1 ) − 2u(x j ) + u(x j+1 )) = − (x j )
h2 dx2
( )
h2 d 4u d 4u ,
− (x i + θ h) + (x i − θ h) ,
24 d x 4 dx4

with .0 < θ < 1 and .0 < θ , < 1. On the other hand, if we expand the function . f (x)
around the point .x j , we obtain
{ x j+1 { x j+1
1 1 d2 f
. f ϕ j d x = f (x j ) + (x − x j )2 ϕ j (x) (ξ ) d x,
h x j−1 2h x j−1 dx2

where .ξ is a point belonging to .]x j , x[. If . f ,, ∈ C 0 ([0, 1], R), we have

| | | 2 |
| 1 { x j+1 | h2 |d f |
| |
.| f ϕ j d x − f (x j )| ≤ max || 2 || .
| h x j−1 | 12 x∈[0,1] d x

2 4 d2 f
We therefore have, since .− dd xu2 = f (therefore .− dd xu4 = dx2
)
| 4 |
h2 |d u |
.|E j (u)| ≤ max || 4 || .
6 x∈[0,1] dx

However, we can write

( { ) ( )
1 1 x j+1 d2u
. − 2 (u(x j−1 ) − 2u(x j ) + u(x j+1 )) − f ϕ j d x − − 2 (x j ) − f (x j )
h h x j−1 dx
= E j (u).

The relation (4.3.7) is therefore an approximation of the Eq. (4.3.1) up to a term of

order .h 2 (. O(h 2 )) near. The quantity . E j (u) (in finite difference terminology) is called
the truncation error at point .x j . ♦
Remark 4.3.4 We pose .e j = u h (x j ) − u(x j ). We can then show that the error
between the exact solution .u of the problem (4.3.1)–(4.3.2) and the solution .u h
defined in Example 4.3.4, measured by discrete norm
⎛ ⎞ 21
I −1
E I −1 (
E )
e j+1 − e j 2
.||e||h = ⎝h e2j + h ⎠ ,
j=1 j=0
h

with .e = (e0 , . . . , e I ) directly depends on the truncation error i.e.,

@seismicisolation
@seismicisolation
200 4 Introduction to Finite Element

||e||h ≤ c max |E j (u)|

.
1≤ j≤I −1

= O(h 2 ),

if the solution .u ∈ C 4 ([0, 1], R). ♦

4.4 Estimates for the Errors

We consider the following problem

( )
d du
.− a(x) = f for x ∈]0, 1[ (4.4.1)
dx dx
.u(0) = u(1) = 0 (4.4.2)

with . f ∈ L 2 (]0, 1[, R), .a(·) ∈ L ∞ (0, 1) and

a(x) ≥ α > 0
.

for all .x ∈ (0, 1). The problem (4.4.1)–(4.4.2) is equivalent to the following
variational problem: Find .u ∈ H01 (0, 1) such that for all .v ∈ H01 (0, 1), we have
{ 1 { 1
du dv
. a(x) dx = f v d x.
0 dx dx 0

We consider the following symmetrical bilinear form

a : H01 (0, 1) × H01 (0, 1) −→ R

.
{ 1
du dv
(u, v) −→ a(u, v) = a(x) dx
0 dx dx

and, the following linear form

. L : H01 (0, 1) −→ R
{ 1
v −→ L(v) = f v d x.
0

4.4.1 We Interpolate by a Polynomial of Degree .1

We divide the interval .[0, 1] into . I intervals of length .h. We pose .xi = i h, for .0 ≤
i ≤ I . Let .Wh be the space

@seismicisolation
@seismicisolation
4.4 Estimates for the Errors 201

Wh = {vh ∈ C 0 ([0, 1], R) such that vh |(xi ,xi+1 ) ∈ P1 , vh is fully determined

.
by its values vi = vh (xi ) at points, xi 0 ≤ i ≤ I }.

We pose
. Vh = {vh ∈ Wh such that vh (0) = vh (1) = 0}.

A basis of the space .Vh is composed of functions .ϕ j (x) ∈ Vh equal to .1 at .x j and to

zero in all .xi .i /= j.

ϕj

0 x j−1 xj x j+1 1

We obtain the hats functions. Thus,

. Vh = {ϕ j , 1 ≤ j ≤ I − 1}

where .ϕ j are the hats functions. We consider the approximate problem: Find .u h ∈ Vh
solution of
.a(u h , vh ) = L(vh ) (4.4.3)

for all .vh ∈ Vh . This is equivalent to find .u h ∈ Vh solution of

a(u h , ϕ j ) = L(ϕ j )
.

for all . j = 1, . . . , I − 1. According to general estimates for the errors (Theorem

4.2.7)
M
.||u − u h ||1,[0,1] ≤ inf ||u − vh ||1,[0,1]
α vh ∈Vh

with . M = ||a(x)||∞ .
Definition 4.4.1 Let.v be a function of.C 0 ([0, 1], R). We call.rh v the.Vh -interpolated
function of .v, i.e., the unique function of .Vh such that

. h r v(xi ) = v(xi ) 0 ≤ i ≤ I
♦
.
The fact that the solution.u ∈ H01 (0, 1) ⊂ C 0 ([0, 1], R), we can define then.rh u ∈ Vh ,
and, we have
M
.||u − u h ||1,[0,1] ≤ ||u − rh u||1,[0,1] . (4.4.4)
α

@seismicisolation
@seismicisolation
202 4 Introduction to Finite Element

Proposition 4.4.2 If we interpolate by the polynomials of degree .1, then the error
between the function .u with its interpolated .rh u is given by
.(i) .||u − rh u||0,[xi ,xi+1 ] ≤ ch 2 |u|2,[xi ,xi+1 ]
.(ii) .|u − r h u|1,[xi ,xi+1 ] ≤ ch|u|2,[xi ,xi+1 ] . ♦
Proof For the proof of Proposition 4.4.2, we can give the following two methods.
Method 1: We apply the Taylor formula with integral remainder
{ b
du d 2u
.u(b) = u(a) + (b − a) (a) + (b − t) dt.
dx a dx2

Method 2: We will use the Theorems 3.4.11 and 3.4.12. In fact,

(i) We take . O :=]xi , xi+1 [ and .π = rh is linear and continuous from . H 2 (]xi , xi+1 [)
.

into . H m (]xi , xi+1 [) with .m ≤ 2. So, according to the Theorem 3.4.12, we have for
all .m ≤ 2,

. ||u − rh u||m,[xi ,xi+1 ] ≤ C([xi , xi+1 ])||I d − rh ||L (H k+1 (]xi ,xi+1 [),H m (]xi ,xi+1 [)) |u|2,[xi ,xi+1 ] .

Reference segment .[−1, 1]:

We make the following change of variable

1+ξ 1−ξ
. x= xi+1 + xi , ξ ∈ [−1, 1].
2 2
This implies
2( xi+1 xi )
. ξ= x− − , x ∈ [xi , xi+1 ].
h 2 2
Thus, { {
xi+1 1
h
. (u − rh u)2 d x = (u(x(ξ )) − rh (x(ξ )))2 dξ.
xi −1 2

We pose .-
u (ξ ) = u(x(ξ )) and .r-
h u(ξ ) = r h u(x(ξ )). Consider the mapping

1+ξ 1−ξ
-
u −→ -
. r-
u= -
u (1) + -
u (−1).
2 2
-
r-
.u is a polynomial of degree .1 at .ξ . The following formula is nothing more than a
result of change of variable in an integral
{ ( )2 ( )2l { 1 ( l )2
xi+1
dl u h 2 d-u
. dx = dξ. (4.4.5)
xi d xl 2 h −1 dξ l

Let us show that .-

r-
u = r-h u. For that, since they are two polynomials of degree 1, it
suffices to show equality in two points. In fact,

@seismicisolation
@seismicisolation
4.4 Estimates for the Errors 203

r-u(1) = rh u(x(1))
. h

= rh u(xi+1 )
= u(xi+1 )
= u(x(1))
=-
u (1)
=-
r-
u (1)

and

. h r-u(−1) = rh u(x(−1))
= rh u(xi )
= u(xi )
= u(x(−1))
=-
u (−1)
=-
r-
u (−1).

Which proves that

-
r-
.u = r-
hu

and .r-h u is a polynomial of degree .1 at .ξ . We will take in the Theorem 3.4.12, . O =

] − 1, 1[, .-r = π : H 2 (] − 1, 1[) −→ H m (] − 1, 1[) with .m ≤ 2. It must be shown
that .-
r ∈ L (H 2 (] − 1, 1[), H m (] − 1, 1[)) and

-
r-
.p=-
p

for all . -
p ∈ P1 . We take .m = 0 ≤ 2. We will show that

-
r ∈ L (H 2 (] − 1, 1[), L 2 (] − 1, 1[, R))
.

and
-
r-
.p=-
p

for all . -
p ∈ P1 . In fact, .-
r is linear.
Continuity of .-
r : For .ξ ∈ [−1, 1], we have

|-
.r-
u (ξ )| ≤ |-
u (1)| + |-
u (−1)|
≤ 2 max |-
u (ξ )|.
ξ ∈[−1,1]

Further, . H 1 (] − 1, 1[) is continuously injects into .C 0 ([−1, 1], R), so

|-
.r-
u (ξ )| ≤ 2||-
u ||∞

@seismicisolation
@seismicisolation
204 4 Introduction to Finite Element

≤ 2c||-
u ||1,[−1,1]
≤ 2c||-
u ||2,[−1,1] .

By integrating the last inequality between .−1 and .1, we find

{ 1
. (-
r-
u (ξ ))2 dξ ≤ 4c2 2||-
u ||22,[−1,1] .
−1

Which implies
. ||-
r-
u ||0,[−1,1] ≤ c||-
u ||2,[−1,1] .

Thus, prove the continuity of .-

r.
On the other hand,
.-
r-
p=-
p

for all . -
p ∈ P1 because .-
r-
p = r-hp = - p (a polynomial of degree .1 is interpolated by
itself). Thus, .-
r ∈ L (H 2 (] − 1, 1[), L 2 (] − 1, 1[, R)) and

.-
r-
p=-
p

for all . -
p ∈ P1 . According to the Theorem 3.4.12,

||-
. u −-
r-
u ||0,[−1,1] ≤ β||I d − -
r ||L (H 2 (]−1,1[),L 2 (]−1,1[,R)) |-
u |2,[−1,1] .

Furthermore,
{ xi+1 {
h 1
. (u − rh u) d x =
2
u − r-
(- h u) (ξ ) dξ
2
xi 2 −1
h
≤ β 2 ||I d − -
r ||2L (H 2 (]−1,1[),L 2 (]−1,1[,R)) |-
u |22,[−1,1] .
2
On the other hand,
-
u (ξ ) = u(x(ξ )).
.

Hence,

d-
u du d x
. =
dξ d x dξ
h du
= .
2 dx
This proves (see Eq. (4.4.5))

@seismicisolation
@seismicisolation
4.4 Estimates for the Errors 205

{ ( )2 ( )4 { 1 ( 2 )2
xi+1
d 2u h 2 d -
u
. dx = dξ.
xi dx2 2 h −1 dξ
2

So,
{ xi+1 ( )4 { xi+1 ( 2 )2
h 2 2 h d u
. (u − rh u)2 d x ≤ β ||I d − -
r ||2L (H 2 (]−1,1[),L 2 (]−1,1[,R)) d x.
xi 2 h 2 xi d x2

Thus,

h2
. ||u − rh u||0,[xi ,xi+1 ] ≤ r ||L (H 2 (]−1,1[),L 2 (]−1,1[,R)) |u|2,[xi ,xi+1 ] .
β||I d − -
4
If we interpolate by a polynomial of degree .1, we have

||u − rh u||0,[xi ,xi+1 ] ≤ ch 2 |u|2,[xi ,xi+1 ] .

.

(ii) We have
.

{ xi+1 ( )2 ( )2 { 1 ( )2
d h 2 d
. (u − rh u) dx = (-
u −-
r-
u ) dξ.
xi dx 2 h −1 dξ

Furthermore,
d -
u (1) − -
u (−1)
. -
r-
u= .
dξ 2

Since . H 1 (] − 1, 1[) is continuously injects into .C 0 ([−1, 1], R), we have

| |
|d |
| u || ≤ max |-
.
| dξ -
r-
ξ ∈[−1,1]
u (ξ )|

≤ α||-u ||1,[−1,1] .

On the other hand, since .-

r is linear, we have

d r-
d(-u ) d-
u
. (-
r-
u) =
dξ d-u dξ
d-
u
=-r .
dξ

Hence, | |
| d-u ||
|
. -
|r dξ | ≤ α||-
u ||1,[−1,1] .

Integrating between .−1 and .1, we obtain

@seismicisolation
@seismicisolation
206 4 Introduction to Finite Element

{ | |
| d-u ||2
1
|-
|r dξ | dξ ≤ 2α ||-
u ||21,[−1,1] .
2
.
−1

This proves || ||
u ||
|| d- √
||-
r || ≤ u ||1,[−1,1] .
2α||-
.
|| dξ ||
0,[−1,1]

Hence, √
. |-
r-
u |1,[−1,1] ≤ u ||1,[−1,1] .
2α||-

It follows that √
. |-
r-
u |1,[−1,1] ≤ u ||2,[−1,1] .
2α||-

On the other hand,

||-
. r-
u ||0,[−1,1] ≤ c||-
u ||2,[−1,1] .

Thus,
. ||-
r-
u ||1,[−1,1] ≤ c||-
u ||2,[−1,1] .

It thus appears that

-
r ∈ L (H 2 (] − 1, 1[), H 1 (] − 1, 1[)).
.

According to the Theorem 3.4.12,

.|-
u −-
r-
u |1,[−1,1] ≤ ||-
u −-r-
u ||1,[−1,1]
≤ β||I d − -r ||L (H 2 (]−1,1[),H 1 (]−1,1[)) |-
u |2,[−1,1] .

Moreover,
( ( ) ) 21
2 h 4
.|-
u |2,[−1,1] = |u|2,[xi ,xi+1 ] .
h 2

Thus,

{ xi+1 ( )2
d
. (u − rh u) dx
xi dx

( )2
2h
. = |-
u − r- 2
h u|1,[−1,1]
h2
( )
h 2 2 2
≤ β ||I d − -r ||2L (H 2 (]−1,1[),H 1 (]−1,1[)) |-
u |22,[−1,1]
2 h

@seismicisolation
@seismicisolation
4.4 Estimates for the Errors 207
( )2 ( )
h 2 2 h 4 2
≤ β ||I d − -
2
r ||L (H 2 (]−1,1[),H 1 (]−1,1[))
2
|u|2,[xi ,xi+1 ]
2 h h 2
≤ ch 2 |u|22,[xi ,xi+1 ] .

If we interpolate by a polynomial of degree .1, we have

.|u − rh u|1,[xi ,xi+1 ] ≤ ch|u|2,[xi ,xi+1 ] .

Which finishes the proof of the Proposition 4.4.2. Q.E.D.

Theorem 4.4.3 Let .u be the solution of the problem (4.4.1)-(4.4.2). If we interpolate

by the polynomials of degree .1, then

||u − rh u||1,[0,1] ≤ ch|u|2,[0,1] .

. ♦

Proof We have
{ 1 { 1)2 (
d
||u − rh u||21,[0,1] =
. (u − rh u)2 d x + (u − rh u) d x
0 0 dx
−1 {
E xi+1
I { xi+1 ( )2
d
= (u − rh u)2 d x + (u − rh u) d x
i=0 xi xi dx
I −1
E
= ||u − rh u||20,[xi ,xi+1 ] + |u − rh u|21,[xi ,xi+1 ] .
i=0

By using Proposition 4.4.2, we have

I −1
E
||u − rh u||21,[0,1] ≤
. c2 h 4 |u|22,[xi ,xi+1 ] + c2 h 2 |u|22,[xi ,xi+1 ]
i=0
≤ c2 h 2 |u|22,[0,1] .

Thus,
||u − rh u||1,[0,1] ≤ ch|u|2,[0,1] .
.

This completes the proof. Q.E.D.

The following result is a consequence of Theorem 4.4.3 and Eq. (4.4.4).

@seismicisolation
@seismicisolation
208 4 Introduction to Finite Element

Theorem 4.4.4 Let .u be the solution of the problem (4.4.1)–(4.4.2) and .u h be the
solution of the problem (4.4.3). If we interpolate by the polynomials of degree .1, then

||u − u h ||1,[0,1] ≤ ch|u|2,[0,1] .

. ♦

4.4.2 We Interpolate by a Polynomial of Degree .2

Theorem 4.4.5 If we interpolate by the polynomials of degree .2, then

||u − rh u||0,[xi ,xi+1 ] ≤ ch 3 |u|3,[xi ,xi+1 ] .

. ♦

Proof If we interpolate by a polynomial of degree .2, i.e.,

-
r-
.p=-
p for all -
p ∈ P2 .

Let
ξ(1 + ξ ) ξ(ξ − 1)
-
r-
.u (ξ ) = -
u (1) + -
u (−1) + (1 − ξ 2 )-
u (0).
2 2
Let .ξ ∈ [−1, 1]. We have

|-
.r-
u (ξ )| ≤ 3 max |-
u (ξ )|.
ξ ∈[−1,1]

Since the space . H 3 (] − 1, 1[) is continuously injects into .C 0 ([−1, 1], R) (because
m = 3 > n2 = 21 ), then
.
. max |- u (ξ )| ≤ α||-
u ||3,[−1,1] .
ξ ∈[−1,1]

Thus,
|-
.r-
u (ξ )| ≤ 3α||-
u ||3,[−1,1] .

Integrating the last equation, we obtain

{ 1
. |-
r-
u (ξ )|2 dξ ≤ 18α 2 ||-
u ||23,[−1,1] .
−1

Hence, √
. ||-
r-
u ||0,[−1,1] ≤ 3 2α||-
u ||3,[−1,1] .

@seismicisolation
@seismicisolation
4.4 Estimates for the Errors 209

Thus, .-
r is continuous from . H 3 (] − 1, 1[) into . L 2 (] − 1, 1[, R). According to the
Theorem 3.4.12, we have

||-
. u −-
r-
u ||0,[−1,1] ≤ β||I d − -
r ||L (H 3 (]−1,1[),L 2 (]−1,1[,R)) |-
u |3,[−1,1] .

Moreover,
({ ( )2 ) 21 ( ( ) ) 21
1
d 3-
u 2 h 6
. dξ = |u|3,[xi ,xi+1 ] .
−1 dξ 3 h 2

Thus,
5
h2
|-
.u |3,[−1,1] = 5
|u|3,[xi ,xi+1 ] .
22
On the other hand,
{ xi+1 { 1
h
. (u − rh u)2 d x = (- h u) (ξ ) dξ.
u − r- 2
xi 2 −1

We have
h
||u − rh u||20,[xi ,xi+1 ] =
. ||-
u − r- 2
h u||0,[−1,1]
2
h
≤ β 2 ||I d − -r ||2L (H 3 (]−1,1[),L 2 (]−1,1[,R)) |-
u |23,[−1,1]
2
h h5
≤ β 2 ||I d − -r ||2L (H 3 (]−1,1[),L 2 (]−1,1[,R)) 5 |u|23,[xi ,xi+1 ] .
2 2
So,
||u − rh u||20,[xi ,xi+1 ] ≤ ch 6 |u|23,[xi ,xi+1 ] .
.

Thus,
||u − rh u||0,[xi ,xi+1 ] ≤ ch 3 |u|3,[xi ,xi+1 ] .
.

It thus appears that if we interpolate by the polynomials of degree .2, we find the error

||u − rh u||0,[xi ,xi+1 ] ≤ ch 3 |u|3,[xi ,xi+1 ] .

.

This finishes the proof. Q.E.D.

Proposition 4.4.6 For all .m ≤ 3, we have

|| ||
|| d m ||
|| (u − r u) || ≤ ch 3−m |u|3,[xi ,xi+1 ] .
.
|| d x m h ||
♦
0,[xi ,xi+1 ]

@seismicisolation
@seismicisolation
210 4 Introduction to Finite Element

Proof We have
| m |
|d |
| u || ≤ 3 max |-
.
| dξ m -
r-
ξ ∈[−1,1]
u (ξ )|

≤ α||-
u ||3,[−1,1] .

Thus,
-
r ∈ L (H 3 (] − 1, 1[), H m (] − 1, 1[))
.

for all .m ≤ 3. According to the Theorem 3.4.12,

. ||-
u −-
r-
u ||m,[−1,1] ≤ β||I d − -
r ||L (H 3 (]−1,1[),H m (]−1,1[)) |-
u |3,[−1,1] .

On the other hand,

{ xi+1 ( )2
dm
. (u − rh u) dx
xi dxm

( )2 m { 1 ( m )2
h 2 d
. = (-
u −- r- u ) dξ
2 h m
−1 dξ
( )2 m
h 2
≤ β 2 ||I d − -
r ||2L (H 3 (]−1,1[),H m (]−1,1[)) |-
u |23,[−1,1]
2 h
( )2 m ( )
h 2 2 h 6 2
≤ β ||I d − -
2
r ||L (H 3 (]−1,1[),H m (]−1,1[))
2
|u|3,[xi ,xi+1 ]
2 h h 2
≤ ch 2(3−m) |u|23,[xi ,xi+1 ] .

If we interpolate by the polynomials of degree .2, we find the error

|| ||
|| d m ||
|| ||
.
|| d x m (u − rh u)|| ≤ ch 3−m |u|3,[xi ,xi+1 ] .
0,[xi ,xi+1 ]

Q.E.D.

4.4.3 We Interpolate by a Polynomial of Degree . k

Theorem 4.4.7 If we interpolate by the polynomials of degree .k, then for all .m ≤
k + 1,

||u − rh u||m,[0,1] ≤ ch k+1−m |u|k+1,[0,1] .

. ♦

@seismicisolation
@seismicisolation
4.5 Embedded Beam Bending Problem 211

Example 4.4.1 We consider the problem

d 2u
. − = f for x ∈]0, 1[
dx2
u(0) = u(1) = 0.

(i) If we interpolate by the polynomials of degree .1, we find the error

.

||u − u h ||1,[0,1] ≤ ch|u|2,[0,1] .

.

(ii) If we interpolate by the polynomials of degree .2, we find the error

.

||u − u h ||1,[0,1] ≤ ch 2 |u|3,[0,1] .

.

(iii) If we interpolate by the polynomials of degree .k, we find the error

.

||u − u h ||1,[0,1] ≤ ch k |u|k+1,[0,1] .

.

4.5 Embedded Beam Bending Problem

4.5.1 Position of the Problem

Solve numerically ⎧
⎨ d 4u
= f if x ∈]0, 1[
. dx4
⎩ u(0) = u(1) = u , (0) = u , (1) = 0,

with . f ∈ L 2 (]0, 1[, R). It is a problem of bending of embedded beam.

v
Recessed

v : tangent to the curve

Let .v be a virtual trip. Then, the radius of courbure is given by

(1 + y ,2 ) 2
3

. .
y ,,

Energy interne is given by

@seismicisolation
@seismicisolation
212 4 Introduction to Finite Element

{ ( )2
1
v,,
. A 3 dx
0 (1 + v,2 ) 2

and, potential energy is given by

{ 1
. f v d x.
0

Hence, energy is given by

{ ( )2 {
1
v,, 1
. E(v) = A 3 dx − f v d x.
0 (1 + v,2 ) 2 0

If we neglected small displacements .v,2 << 1, we obtain

{ 1 { 1
,,2
. J (v) = A v dx − f v d x.
0 0

Find .u ∈ V such that

. J (u) = inf J (v),
v∈V

with

. V = {v ∈ H 2 (0, 1) such that v(0) = v(1) = 0 and v, (0) = v, (1) = 0} = H02 (0, 1).

4.5.2 Existence and Uniqueness of Solution

. V is a closed subspace of . H 2 (0, 1) because .V = K er (g) with

. g : H 2 (0, 1) −→ R4
v −→ (v(0), v(1), v, (0), v, (1)).

Lemma 4.5.1 . J is strictly convex. ♦

Proof We have { {
1 1
, ,, ,,
(J (v), w) = 2 A
. v w dx − f w d x.
0 0

Hence, { 1
(J ,, (v)w, w) = 2 A
. w,,2 d x.
0

@seismicisolation
@seismicisolation
4.5 Embedded Beam Bending Problem 213

Thus,
(J ,, (v)w, w) ≥ 0, for all w ∈ V.
.

On the other hand, .(J ,, (v)w, w) = 0 gives

{ 1
. 2A w,,2 d x = 0.
0

So, .w,,2 = 0. Thus, .w, = cte. Further, .w, (0) = 0, so .w, = 0. Thus, .w = cte. Gold,
.w(0) = 0, so .w = 0. Hence, . J is strictly convex. Q.E.D.
Lemma 4.5.2 . J is strongly semicontinuous. ♦
Lemma 4.5.3 . J is coercive. ♦
Proof Let .v ∈ V . Hence, .v, ∈ H01 (0, 1), so according to the Poincaré inequality
(Theorem 3.2.2),
, ,,
.||v ||0,(0,1) ≤ c||v ||0,(0,1) .

Further, .v ∈ H01 (0, 1), so according to the Poincaré inequality (Theorem 3.2.2),

.||v||0,(0,1) ≤ c||v, ||0,(0,1) .

Hence, for .||v||0,(0,1) /= 0, we have

. J (v) ≥ c||v||20,(0,1) − || f ||0,(0,1) ||v||0,(0,1)

( )
|| f ||0,(0,1)
≥ c− ||v||20,(0,1) .
||v||0,(0,1)

When .||v||2,(0,1) → ∞, we have the following conclusions .||v,, ||0,(0,1) → ∞ or

,
.||v ||0,(0,1) → ∞ or .||v||0,(0,1) → ∞. Hence, . J (v) → ∞ when .||v||2,(0,1) → ∞. Thus,

. J is coercive. Q.E.D.

V provided of the norm .|| · ||2,(0,1) is a Hilbert space. Hence, the minimum is reached
.
in .u
. J (u) = inf J (v).
v∈V

. J is differentiable
{ 1 { 1
, ,, ,,
(J (v), w) = 2 A
. v w dx − f w d x.
0 0

Hence, .(J , (u), w) = 0 for all .w ∈ V , gives

{ 1 { 1
,, ,,
2A
. u w dx = f w dx
0 0

@seismicisolation
@seismicisolation
214 4 Introduction to Finite Element

for all.w ∈ V (this is the variational formulation). On the other hand,.D(0, 1) ⊂ V , so

. 2 A(u ,, , ϕ ,, ) = ( f, ϕ)

for all .ϕ ∈ D(0, 1). Hence,

2 A(u (4) , ϕ) = ( f, ϕ)
.

for all .ϕ ∈ D(0, 1). Thus,

2 Au (4) = f
.

on .D , (0, 1). Further, since . f ∈ L 2 (]0, 1[, R), we have .u (4) ∈ L 2 (]0, 1[, R). Since
,, ,,,
.u ∈ L (]0, 1[, R), so .u ∈ L 2 (]0, 1[, R). Hence, .u ∈ H 4 (0, 1). Thus,
2

.2 Au (4) = f on ]0, 1[.

We take .2 A = 1. We consider the variational problem: Find .u ∈ V such that

a(u, v) = L(v)
. (4.5.1)

for all .v ∈ V , with { 1

. a(u, v) = u ,, v,, d x
0

and { 1
. L(v) = f v d x.
0

The Lax-Milgram theorem gives us existence and the uniqueness of .u ∈ V .

4.5.3 Problem Approached

Introduce then the space .Vh ⊂ V ⊂ C 1 ([0, 1], R). We divide the interval .[0, 1] into
. I intervals of length .h. We pose . x i = i h, .0 ≤ i ≤ I . Let

. Wh = {vh ∈ L 2 (]0, 1[, R) such that vh |(xi ,xi+1 ) ∈ P3 , vh is entirely determined by

. vh (xi ) and vh, (xi ) for all i}.

We pose

. Vh = {vh ∈ Wh such that vh (0) = vh (1) = 0 and vh, (0) = vh, (1) = 0}.

We consider the problem: Find .u h ∈ Vh such that

@seismicisolation
@seismicisolation
4.5 Embedded Beam Bending Problem 215

a(u h , vh ) = L(vh )
. (4.5.2)

for all .vh ∈ Vh . A basis of the space .Vh is formed by the following two types of
functions .ϕ j and .ψ j :
−
−
−
−
−
−
−
−
−
−
−
−
−
−
−
0 ϕj ψj 1

. Vh is a closed subspace of .V , therefore .Vh is a Hilbert space. Thus, the Lax-Milgram

theorem gives us existence and uniqueness of .u h ∈ Vh . Let’s find the stiffness matrix
and the error .u − u h . We have

I −1
E I −1
E
v (x) =
. h vh (x j )ϕ j (x) + vh, (x j )ψ j (x)
j=1 j=1

and
I −1
E I −1
E
u (x) =
. h u h (x j )ϕ j (x) + u ,h (x j )ψ j (x).
j=1 j=1

So, we come across the following system

. a(u h , ϕ j ) = L(ϕ j )

a(u h , ψ j ) = L(ψ j )
.

for all . j = 1, . . . , I − 1. We can write it in matrix form as follows

⎛ ⎞ ⎛ ⎞
u h (x1 ) L(ϕ1 )
⎛ ⎞
× × 0 0 ··· 0 × × 0 0 ··· 0 ⎜⎜
⎟ ⎜
⎟ ⎜
⎟
⎟
⎜× × × 0 ··· 0 × × × 0 ··· 0⎟ ⎜ .. ⎟ ⎜ .. ⎟
⎜ ⎟⎜ . ⎟ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ .. ⎟ ⎜ .. ⎟
⎜ ⎟⎜ . ⎟ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ u h (x I −1 ) ⎟ ⎜ L(ϕ I −1 ) ⎟
.⎜ ⎟⎜ , ⎟=⎜ ⎟
⎜ ⎟ ⎜ u (x1 ) ⎟ ⎜ L(ψ1 ) ⎟ .
⎜ ⎟⎜ h ⎟ ⎜ ⎟
⎜ ⎟⎜ .. ⎟ ⎜ .. ⎟
⎜ ⎟⎜ . ⎟ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ .. ⎟ ⎜ ⎟
⎜ ⎟⎜ . ⎟ ⎜ ⎟
⎜ ⎟⎜ ⎟ ⎜ .. ⎟
⎝ ⎠⎜ ⎟ ⎜ . ⎟
⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠
u ,h (x I −1 ) L(ψ I −1 )

@seismicisolation
@seismicisolation
216 4 Introduction to Finite Element

We obtain a strip matrix of length .2l + 1 with .l = I . Hence, the shape of this matrix
is not good. So, we arrange the basis so that we find a nicer matrix, that is, one that
contains ordered zeros. So, either
⎛ ⎞ ⎛ ⎞
u h (x1 ) L(ϕ1 )
⎛ ⎞
× × ×× 0 0 0 0 0 0 0 0 ⎜ ⎜
⎟ ⎜
u ,h (x1 )
⎟ ⎜
⎟
⎟ L(ψ1 )
⎜× × ×× 0 0 0 0 0 0 0 0⎟ ⎜ ..
⎟ ⎜ ⎟ ..
⎜ ⎟⎜ ⎟ ⎜. ⎟ .
⎜× × ×××× 0 0 0 0 0 0⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜× × ×××× 0 0 0 0 0 0⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ .. ⎟ ⎜ .. ⎟
⎜0 0 ××××× ×0 0 0 0⎟ ⎜ . ⎟ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜0 0 ××××× ×0 0 0 ⎟ ⎜ ⎜
0 ⎟ ⎜ u h (xi ) ⎟ ⎜ L(ϕi ) ⎟
⎟
.⎜ ⎟
⎜ ⎟ ⎜ u , (xi ) ⎟ = ⎜ L(ψi ) ⎟ .
⎜ ⎟⎜ h ⎟ ⎜ ⎟
⎜ ⎟⎜ .. ⎟ ⎜ .. ⎟
⎜ ⎟⎜ . ⎟ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ ⎟⎜ . ⎟ ⎜ . ⎟
⎝ ⎠ ⎜ .
. ⎟ ⎜ .
. ⎟
⎜ ⎟ ⎜ ⎟
⎝ u h (x I −1 ) ⎠ ⎝ L(ϕ I −1 ) ⎠
u ,h (x I −1 ) L(ψ I −1 )

We obtain a strip matrix of length .2l + 1 with .l = 3.

4.5.4 Error Estimates

We have { 1
a(vh , vh ) =
. vh,,2 d x = |vh |22,(0,1) .
0

We can apply the Poincaré inequality twice, we find

.a(vh , vh ) ≥ α||vh ||20,(0,1) .

On the other hand,

{ 1
a(u h , vh ) =
. u ,,h vh,, d x
0
≤ |u h |2,(0,1) |vh |2,(0,1)
≤ ||u h ||2,(0,1) ||vh ||2,(0,1) .

Thus,

||u − u h ||2,(0,1) ≤ c inf ||u − vh ||2,(0,1)

.
vh ∈Vh

@seismicisolation
@seismicisolation
4.5 Embedded Beam Bending Problem 217

≤ c||u − rh u||2,(0,1) ,

with .rh u ∈ Vh ,
. h r u(xi ) = u(xi )

and
. (rh u), (xi ) = u , (xi )

for all .i = 1, . . . , I − 1.
Theorem 4.5.4 Let .u ∈ H 4 (0, 1). Then,

|u − ri u|m,(xi ,xi+1 ) ≤ ch 4−m |u|4,(xi ,xi+1 ) , 0 ≤ m ≤ 4,

.

with .ri u ∈ P3 such that .ri u = u, .(ri u), = u , in .xi and .xi+1 . ♦
Proof Let
1+ξ 1−ξ
.x = xi+1 + xi .
2 2
We set the change of variable
-
v(ξ ) = v(x(ξ )).
.

We have .- r-
v ∈ P3 , .-
r-
v =-v and .(-v), = -
r- v, in .−1 and in .1. On the other hand, .ri v = v
in .xi and .xi+1 , so .r-iv =-v in .−1 and .1. Furthermore, .(ri v), = v, in .xi and .xi+1 ,
, ,
so .(-
ri v) = - v in .−1 and .1. Further, according to Proposition 2.6.8 .(iii), if two
polynomials . p and .q of . P3 coincide at two points and their derivatives, then they are
equal . p = q. Thus,
-
.r iv = - r-
v.

On the other hand,

{ ( )2
∂m xi+1
|u − ri u|2m,(xi ,xi+1 ) =
. (u − r i u) dx
xi ∂xm
{ 1 ( )2 m ( m )2
2 ∂ h
= (-u − -r -
u ) dξ
−1 h ∂ξ m 2
( )−2 m+1 { 1 ( m )2
h ∂
= (-
u −- r-
u ) dξ.
−1 ∂ξ
2 m

Furthermore,
-
r : H 4 (−1, 1) −→ H m (−1, 1)
.

is linear, continuous and

-
r-
.p=-
p

for all . -
p ∈ P3 . Then,

@seismicisolation
@seismicisolation
218 4 Introduction to Finite Element

||-
. u −-
r-
u ||m,(−1,1) ≤ c|-
u |4,(−1,1) .

On the other hand,

{ (
)2
∂ 4-
1
u
.|-
u |24,(−1,1) = dξ
−1 ∂ξ
4
( )
2 h 8 2
= |u|4,(xi ,xi+1 ) .
h 2

Hence,
|u − ri u|m,(xi ,xi+1 ) ≤ ch 4−m |u|4,(xi ,xi+1 ) .
.

For .m = 4 and since .ri u ∈ P3 , so .(ri u)(4) = 0. So, we have the result already stated.
Hence, it remains for us to show the continuity of .- r : H 4 (−1, 1) −→ H m (−1, 1)
.0 ≤ m ≤ 3. We have

-
r-
.v(ξ ) = - ϕ−1 (ξ ) +-
v(−1)- ϕ1 (ξ ) +-
v(1)- -−1 (ξ ) +-
v, (−1)ψ -1 (ξ ),
v, (1)ψ

with .- -−1 , and .ψ

ϕ1 , .ψ
ϕ−1 , .- -1 are the basis functions. Furthermore,
| |
| ∂m |
| v|| ≤ c(max |- v, |).
.
| ∂ξ m -
r- v| + max |-

Further, . H 2 (−1, 1) ⊂ C 0 ([−1, 1], R), so

| |
| ∂m |
| v|| ≤ c||-
.
| ∂ξ m -
r- v||2,[−1,1]

≤ c||-
v||4,[−1,1] 0 ≤ m ≤ 3.

Hence,
||-
. r-
v||m,[−1,1] ≤ c||-
v||4,[−1,1] 0 ≤ m ≤ 3.

Thus, the continuity. Q.E.D.

Theorem 4.5.5 Let .u be the solution of the problem (4.5.1) and .u h be the solution
of the problem (4.5.2). Then,

||u − u h ||2,(0,1) ≤ ch 2 |u|4,(0,1) .

. ♦

@seismicisolation
@seismicisolation
4.5 Embedded Beam Bending Problem 219

Proof For .m = 2, we have

I −1
E
||u
. − rh u||22,(0,1) = ||u − ri u||22,[xi ,xi+1 ]
i=0
I −1
E
≤ ch 2(4−2) |u|24,[xi ,xi+1 ] .
i=0

Hence,
.||u − rh u||2,(0,1) ≤ ch 2 |u|4,(0,1) .

Thus,
||u − u h ||2,(0,1) ≤ ch 2 |u|4,(0,1) .
.

This completes the proof. Q.E.D.

Question: Can we improve this error.

In fact, we will use Aubin-Nitsche’s duality process.

4.5.5 Aubin-Nitsche’s Duality Process

Theorem 4.5.6 Let .u be the solution of the problem (4.5.1) and .u h be the solution
of the problem (4.5.2). Then,

. ||u − u h ||0,(0,1) ≤ ch 4 |u|4,(0,1) . ♦

Proof We have

|(u − u h , g)0,(0,1) |
||u − u h ||0,(0,1) =
. sup .
g∈L 2 ((0,1),R), ||g||0,(0,1) /=0 ||g||0,(0,1)

At .g ∈ L 2 ((0, 1), R) fixed, let .ϕ be the solution of

d 4ϕ
. = g in (0, 1)
dx4
ϕ(0) = ϕ(1) = ϕ , (0) = ϕ , (1) = 0.

@seismicisolation
@seismicisolation
220 4 Introduction to Finite Element

We have
{ 1
(u − u h , g)0,(0,1) =
. (u − u h )g d x
0
{ 1
d 4ϕ
= (u − u h ) d x.
0 dx4

Using the Green’s formula twice, we obtain

{ [ 3 ]1
d 3ϕ d 1
d ϕ
.(u − u h , g)0,(0,1) =− 3
(u − u h ) d x + (u − u h )
0 dx dx dx3 0
{ 1 2 [ 2 ]1
d ϕ d 2
d ϕ d
= 2 2
(u − u h ) d x − (u − u h ) .
0 dx dx dx2 dx 0

Hence, { 1
d 2ϕ d 2
(u − u h , g)0,(0,1) =
. (u − u h ) d x.
0 dx2 dx2

Further, { {
1 1
. u ,, ϕh,, d x = f ϕh d x
0 0

for all .ϕh ∈ Vh and { {

1 1
. u ,,h ϕh,, d x = f ϕh d x
0 0

for all .ϕh ∈ Vh . Making the difference between the last two equalities, we obtain
{ 1
. (u − u h ),, ϕh,, d x = 0.
0

Hence, { 1
(u − u h , g)0,(0,1) =
. (u − u h ),, (ϕ − ϕh ),, d x.
0

For .ϕh = rh ϕ, we have

{ 1
(u − u h , g)0,(0,1) =
. (u − u h ),, (ϕ − rh ϕ),, d x
0
≤ |u − u h |2,(0,1) |ϕ − rh ϕ|2,(0,1) .

On the other hand,

|ϕ − rh ϕ|2,(0,1) ≤ ch 2 |ϕ|4,(0,1)
.

@seismicisolation
@seismicisolation
4.5 Embedded Beam Bending Problem 221

and
|u − u h |2,(0,1) ≤ ch 2 |u|4,(0,1) .
.

Hence,
(u − u h , g)0,(0,1) ≤ ch 4 |ϕ|4,(0,1) |u|4,(0,1) .
.

Furthermore,
{ ( )2
1
d 4ϕ
|ϕ|24,(0,1) =
. dx
0 dx4
{ 1
= |g|2 d x.
0

So,
|ϕ|4,(0,1) = ||g||0,(0,1) .
.

Thus,
.(u − u h , g)0,(0,1) ≤ ch 4 |u|4,(0,1) ||g||0,(0,1) .

This implies, for .g such that .||g||0,(0,1) /= 0, we have

|(u − u h , g)0,(0,1) |
. ≤ ch 4 |u|4,(0,1) .
||g||0,(0,1)

By switching to sup, we obtain

||u − u h ||0,(0,1) ≤ ch 4 |u|4,(0,1) .

.

This finishes the proof. Q.E.D.

Note that when we applied Aubin-Nitsche’s duality method, we found an error

smaller than the usual method.

4.5.6 Inverse Inequality

Lemma 4.5.7 For all .vh ∈ Vh , we have

|vh |1,(xi ,xi+1 ) ≤ ch −1 ||vh ||0,(xi ,xi+1 ) .

. ♦

@seismicisolation
@seismicisolation
222 4 Introduction to Finite Element

Proof We have
{ xi+1 ( )2 { 1
2 h
. vh,2 d x = vh ),2 dξ.
(-
xi h 2 −1

On the other hand,

-
v (ξ ) = -
. h vh (−1)-
ϕ−1 (ξ ) +-
vh (1)-
ϕ1 (ξ ) +- -−1 (ξ ) +-
vh, (−1)ψ -1 (ξ ).
vh, (1)ψ

Hence,

v, (ξ ) = -
-
. h vh (−1)-,
ϕ−1 (ξ ) +- ϕ1, (ξ ) +-
vh (1)- -−1
vh, (−1)ψ , -1, (ξ ).
vh, (1)ψ
(ξ ) +-

So,
vh, (ξ )| ≤ c(|-
|-
. vh (−1)| + |- vh, (−1)| + |-
vh (1)| + |- vh, (1)|).

Which proves

vh, (ξ )|2 ≤ c(|-

|-
. vh (−1)|2 + |- vh, (−1)|2 + |-
vh (1)|2 + |- vh, (1)|2 ).

Thus,

. vh, (ξ )||20,[−1,1] ≤ c(|-

||- vh (−1)|2 + |- vh, (−1)|2 + |-
vh (1)|2 + |- vh, (1)|2 ).

We pose /
. vh ) =
L(- |-
vh (−1)|2 + |- vh, (−1)|2 + |-
vh (1)|2 + |- vh, (1)|2 .

. L(-vh ) and .||-

vh ||0,[−1,1] are two norms on . P3 and .dim P3 < ∞. Hence, since in a
finite-dimensional space all the norms are equivalent, we have

|-
.vh |21,[−1,1] ≤ cL(-
v h )2
≤ β||-
vh ||20,[−1,1]
2
= β ||vh ||20,(xi ,xi+1 ) .
h
Thus, ( )2
h 2 , 2
. ||-
v (ξ )||20,[−1,1] ≤ β ||vh ||20,(xi ,xi+1 ) .
2 h h h

It follows that

.|vh |1,(xi ,xi+1 ) ≤ ch −1 ||vh ||0,(xi ,xi+1 ) .

Q.E.D.

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 223

Theorem 4.5.8 Let .u be the solution of the problem (4.5.1) and .u h be the solution
of the problem (4.5.2). Then,

||u − u h ||1,(0,1) ≤ ch 3 (|u|4,(0,1) + h|u|4,(0,1) ).

. ♦

Proof We have

|u h − rh u|1,(0,1) ≤ ch −1 ||u h − rh u||0,(0,1)

.
[ ]
≤ ch −1 ||u h − u||0,(0,1) + ||u − rh u||0,(0,1) .

Hence,
.|u h − rh u|1,(0,1) ≤ ch 3 |u|4,(0,1) .

Thus,

|u h − u|1,(0,1) ≤ |u h − rh u|1,(0,1) + |rh u − u|1,(0,1)

.

≤ ch 3 |u|4,(0,1) .

Which proves

||u − u h ||21,(0,1) = ||u − u h ||20,(0,1) + |u − u h |21,(0,1)

.

≤ ch 8 |u|24,(0,1) + ch 6 |u|24,(0,1) .

It follows that
||u − u h ||1,(0,1) ≤ ch 3 (|u|4,(0,1) + h|u|4,(0,1) ).
.

This completes the proof. Q.E.D.

4.6 Introduction to Finite Elements in Dimension .2

for Polygonal Domain

Our presentation of the finite element method in dimension 1 was general enough to
pass easily to the multidimensional case (dimension 2). Of course, some additional
difficulties will arise, but nothing really new. For example, increased attention will
have to be paid especially to the numbering of the degrees of freedom to have a
“pleasant” structure (tape matrix for example) and to avoid matrices requiring too
much memory space.

@seismicisolation
@seismicisolation
224 4 Introduction to Finite Element

4.6.1 Triangulation of the Domain

Ti

Let .Ω be a polygonal domain of .R2 with border .Γ = ∂Ω. We will discretize the
domain.Ω then introduce a finite element space, called linear Lagrange finite element
or . P1 . We, then, approach the solution of our equation by looking for it in a finite-
dimensional approximation space. Let .τh be a triangulation of .Ω by subdividing .Ω
into triangles .(Ti )i=1,...,m which do not overlap and such that

| | | |
m
.Ω= Ti = Ti .
Ti ∈τh i=1

Two triangles .Ti and .T j .i /= j either have a common side, or a common vertex, or
are disjoint.

4.6.2 Values at the Top

Lemma 4.6.1 Let .T ⊂ Ω be a triangle whose vertices are the points . Ai of

coordinates .(xi , yi ) .1 ≤ i ≤ 3.

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 225

A3

A1 A2

Then, there is a polynomial . p(x, y) ∈ P1 (of the form .a + bx + cy) unique taking
arbitrary given values . pi .1 ≤ i ≤ 3, at the vertices . Ai . ♦

Proof Let . p(x, y) = a + bx + cy. We have

p3
A3 (x3 , y3 )

p1 p2
A1 (x1 , y1 ) A2 (x2 , y2 )

at point . A1
. p1 = a + bx1 + cy1 ,

at point . A2
. p2 = a + bx2 + cy2 ,

and at point . A3
. p3 = a + bx3 + cy3 .

Which gives the following system

⎛ ⎞⎛ ⎞ ⎛ ⎞
1 x1 y1 a p1
. ⎝ 1 x 2 y2 ⎠ ⎝ b ⎠ = ⎝ p2 ⎠ .
1 x3 y3 c p3

Since the points . A1 , . A2 and . A3 are not aligned, we have

@seismicisolation
@seismicisolation
226 4 Introduction to Finite Element
| |
| 1 x1 y1 |
| |
.det | 1 x 2 y2 | / = 0.
| |
| 1 x3 y3 |
⎛ ⎞
a
Thus, .⎝ b ⎠ are uniquely determined. Q.E.D.
c

4.6.3 Barycentric Coordinates

Definition 4.6.2 We call barycentric coordinates of a point . M(x, y) with respect to

the points . Ai (xi , yi ) .1 ≤ i ≤ 3, not aligned, the scalars .λi (x, y) (or .λi (M)) such that
(i)
.
E
3
. λi (x, y) = 1
i=1

(ii)
.
E
3
. λi (x, y)xi = x
i=1

(iii)
.
E
3
. λi (x, y)yi = y.
i=1 ♦

Remark 4.6.3 We can always find the scalars .λi (x, y) .1 ≤ i ≤ 3 satisfying the
relations .(i), .(ii) and .(iii) of the Definition 4.6.2. ♦

The polynomial . p of Lemma 4.6.1 is expressed in a simple way in function of

barycentric coordinates. In fact,

. p(x, y) = a + bx + cy
= a(λ1 + λ2 + λ3 ) + b(λ1 x1 + λ2 x2 + λ3 x3 ) + c(λ1 y1 + λ2 y2 + λ3 y3 )
= λ1 (a + bx1 + cy1 ) + λ2 (a + bx2 + cy2 ) + λ3 (a + bx3 + cy3 )
E
3
= λi (x, y) pi .
i=1

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 227

Thus, we have the following relationships:

{
0 if i /= j
λ (A j ) =
. i
1 if i = j.

λ (M) = 0 if . M belongs to the side . A j Ak opposite to the vertex . Ai .

. i

4.6.4 Approximation of the Solution

4.6.4.1 Approximate Space

Let .Wh be the space

Wh = {vh ∈ C 0 (Ω, R) such that vh |Ti ∈ P1 , vh is fully determined

.
by its values at the vertices}.

Hence
. dim Wh = number of vertices.

Thus,

Vh = {vh ∈ Wh such that vh = 0 to the peaks that belong to the border ∂Ω}.
.

n
Proposition 4.6.4 .Wh ⊂ C 0 (Ω) H 1 (Ω). ♦

Proof Let . K 1 = A1 A2 A3 and . K 4 = A2 A3 A4 be two adjacent triangles along side

A2 A3 , . K 1 and . K 4 ⊂ Ω.
.

A3

A1 K1 K4 A4

A2

@seismicisolation
@seismicisolation
228 4 Introduction to Finite Element

Let .vh be a function of .Wh and let .v(1) and .v(4) be the restrictions of .vh to triangles . K 1
and . K 4 , respectively. The restriction of .v(1) to side . A2 A3 only depends on the values
(4)
.v2 and .v3 at vertices . A 2 and . A 3 , and the same is true for .v , and we can write for
any point . M of . A2 A3

v(1) (M) = v(4) (M) = λ2 (M)v2 + λ3 (M)v3 ,

.

where .λ2 (M) and .λ3 (M) are the barycentric coordinates of point . M with respect to
points . A2 and . A3 i.e.,
−−→ −−−→
M A2 · A3 A2
λ (M) = 1 − λ2 (M) =
. 3
−−−→ .
| A 3 A 2 |2
U
The function .vh is therefore continuous on . K 1 K 4 . We deduce that the functions
of .Wh are continuous on .Ω. On the other hand, the restrictions of the functions from
. Wh to each triangle . K are polynomials, therefore belong to . H (K ). Lemma 3.3.6
1

leads to inclusion .Wh ⊂ H (Ω).

1
Q.E.D.

Remark 4.6.5 It is possible to mix the triangular finite elements with 3 nodes
and then
quadrilateral finite elements with 4 nodes. We thus obtain a subspace of
.C (Ω) H 1 (Ω). ♦
0

Remark 4.6.6 In practice, to calculate the coefficients .a(wi , w j ) of the stiffness

matrix, where .{wi } denotes the set of basic functions, the calculations are made
element by element (in the case of the Poisson equation), we therefore have
{ ( )
∂wi ∂w j ∂wi ∂w j
.a(wi , w j ) = + d xd y
Ω ∂x ∂x ∂y ∂y
E { ( ∂wi ∂w j ∂wi ∂w j
)
= + d xd y.
K ∂x ∂x ∂y ∂y
K ⊂Ω

To calculate the integrals on each element . K , we use the change of variables

( ) ( )
x ξ
. = FK ,
y η

with .d xd y = JK (ξ, η) dξ dη, where

| ∂x ∂ x ||
|
| |
| ∂ξ ∂η ||
. J K = det |
| ∂y ∂y |
| |
∂ξ ∂η

and, we have the formulas

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 229

-
∂v ∂-
v ∂y ∂-
v ∂y
J
. K = − ,
∂x ∂ξ ∂η ∂η ∂ξ

-
∂v ∂-
v ∂x ∂-
v ∂x
J
. K =− + ,
∂y ∂ξ ∂η ∂η ∂ξ

where .-
v(ξ, η) = v(x, y) = v ◦ FK . ♦

According to the Lemma 3.3.6, .Wh ⊂ H 1 (Ω). Thus,

. Vh ⊂ H01 (Ω).

Let .Vh be the subspace of .Wh of the zero functions .vh at the vertices located on the
boundary .∂Ω. These functions are in fact zero on the whole boundary and, we have
the inclusion N
. Vh ⊂ C (Ω, R) H01 (Ω).
0

The dimension of .Vh is equal to the number . N of vertices not located on .∂Ω.
Let . A1 , . . . , A N be these vertices (the nodes of the triangulation). We consider the
functions .w j of .Vh such that

.w j (Al ) = δ jl , 1 ≤ j, l ≤ N .

The support of the function .w j is closed from the union of the triangles admitting
A j as support. The set of functions .{w j } Nj=1 is a basis of .Vh and, we have for any
.
function .vh ∈ Vh
EN
.vh = vh (Ai )wi .
i=1

Let .u h ∈ Vh be the solution of the problem (4.2.1) (case of the Poisson equation).
We pose
.u i = u h (Ai ).

The .u i are given by the resolution of the linear system

E
N
. a(wi , w j )u i = L(w j ), 1 ≤ j ≤ N .
i=1

The coefficients.a(wi , w j ) are different from zero only if the supports of the functions
wi and .w j have a non-empty intersection, i.e., if the points . Ai and . A j are the vertices
.
of the same triangle.

@seismicisolation
@seismicisolation
230 4 Introduction to Finite Element

4.6.4.2 Basis of . Vh

The dimension of .Vh is equal to the number . N of vertices of the triangles not located
on .∂Ω. Let . A1 , . . . , A N be these vertices (the nodes of triangulation). A basis of .Vh
are the functions .ϕ1 , . . . , ϕ N defined by

ϕ j (Ai ) = δi j .
.

The support of the function .ϕ j is formed by the union of the triangles admitting . A j
as vertex.

ϕj

Aj

A two-dimensional hat function ϕ j on a general triangle mesh.

4.6.4.3 Approximate Solution

Let .vh ∈ Vh , we have

E
N
v (M) =
. h αi ϕi (M).
i=1

For . M = A j ,

E
N
v (A j ) =
. h αi ϕi (A j )
i=1
= α j δi j
= αj.

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 231

Hence,
v (A j ) = α j .
. h

Thus,
E
N
v (M) =
. h vh (Ai )ϕi (M).
i=1

Let .u h be the solution of the problem: Find .u h ∈ Vh such that

a(u h , vh ) = L(vh )
.

for all .vh ∈ Vh . Which implies

E
N
. a(ϕi , ϕ j )u h (A j ) = L(ϕ j )
j=1

for all .i = 1, . . . , N . We pose

.u j = u h (A j ),
⎛⎞
L(ϕ1 )
⎜ .. ⎟
.f = ⎝
. ⎠
L(ϕ N )

and ⎞⎛
u1
⎜ . ⎟
.u = ⎝ . ⎠ .
.
uN

We are faced with a matrix problem

. Au = f,

where
. A = (ai j )1≤i, j≤N

with .ai j = a(ϕi , ϕ j ).

Remark 4.6.7 .(i) .a(ϕi , ϕ j ) = 0 if . Ai A j is not a side.
.(ii) The diagonal elements of the matrix . A are strictly positive because the bilinear
form .a(·, ·) is coercive.
.(iii) Let .max j | j − i| such that . Ai A j is a side. The matrix . A is a strip matrix of
length .2l + 1 with
.l = max max | j − i|,
i j

@seismicisolation
@seismicisolation
232 4 Introduction to Finite Element

i
⎛ ⎞
⎜ ⎟
2l ⎜
+1 ⎟
⎜ ⎟
⎜ ⎟
⎜ ⎟.
i ⎜ ⎟
⎜ ⎟
⎜ ⎟
⎝ ⎠

4.6.5 Triangulation Numbering

• Let .Ω be the following domain numbered as follows:

.

6 7 8 9 10

1 2 3 4 5

Then, the matrix

. A = (ai j )1≤i, j≤10

is of the form ⎛ ⎞
×× 0 0 0 ×× 0 0 0
⎜× × × 0 0 0 ×× 0 0⎟
⎜ ⎟
⎜0 ××× 0 0 0 ×× 0⎟
⎜ ⎟
⎜0 0 ×× × 0 0 0 × ×⎟
⎜ ⎟
⎜0 0 0 × × 0 0 0 0 ×⎟
.⎜ ⎟.
⎜× 0 0 0 0 ×× 0 0 0⎟
⎜ ⎟
⎜× × 0 0 0 ××× 0 0⎟
⎜ ⎟
⎜0 ×× 0 0 0 ××× 0⎟
⎜ ⎟
⎝0 0 ×× 0 0 0 × × ×⎠
0 0 0 × × 0 0 0 ××

Thus,
. l = 6.

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 233

• If we number .Ω as follows:
.

2 4 6 8 10

1 3 5 7 9

Then, the matrix

. A = (ai j )1≤i, j≤10

is of the form ⎛ ⎞
× ××× 0 0 0 0 0 0
⎜× × 0 × 0 0 0 0 0 0⎟
⎜ ⎟
⎜× 0 ×××× 0 0 0 0⎟
⎜ ⎟
⎜× ××× 0 × 0 0 0 0⎟
⎜ ⎟
⎜0 0 × 0 ×××× 0 0⎟
.⎜ ⎟.
⎜0 0 ×××× 0 × 0 0⎟
⎜ ⎟
⎜0 0 0 0 × 0 × × × ×⎟
⎜ ⎟
⎜0 0 0 0 × × × × 0 ×⎟
⎜ ⎟
⎝0 0 0 0 0 0 × 0 × ×⎠
0 0 0 0 0 0 ××××

Thus,
. l = 3.

This property, very interesting during the resolution of the linear system, depends
closely on the numbering of the vertices. So, the best numbering one is that has a
minumum .l. Hence, there is a numbering algorithm.
The matrix . A is called the system stiffness matrix. The points . Al to which the .u l
values of the solution .u h are attached are called the nodes of the triangulation. The
sides of the triangles are called the edges of the triangulation. So, to have a minimum
.l, when you number a node, you have to number all the neighboring nodes and you

turn counterclockwise.

@seismicisolation
@seismicisolation
234 4 Introduction to Finite Element

4.6.6 Examples of Poor Quality Triangles

The mesh should not have flattened triangles like the two examples of poor quality
triangles:

4.6.7 Conditions of Not Having Bad Quality Triangles

In order not to have triangles of bad quality, one imposes on the triangles.T a condition
of this type
h(T )
. ≤ α,
ρ(T )

with .h(T ) is the largest side of .T , .ρ(T ) is the diameter of the circle inscribed in
T , and .α > 0. This is why often to refine the meshes, one operates as follows: one
.
divides each triangle into .4 by joining the midpoints of its edges:

4.6.8 Transformation Between Any Triangle

and the Reference Triangle

- (resp. . M) of the
We note .ξ and .η (resp. .x and . y) the coordinates of any point . M
- (resp. .T ).
element (triangle) .T

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 235

A2 (x2 , y2 )
η
A3 (x3 , y3 )
-2
1 A

-
T

-3
A -1
A
0 1 ξ A1 (x1 , y1 )

Take care of .(xi , yi ) the coordinates of the vertices . Ai , .1 ≤ i ≤ 3, of the triangle

.T , range of the triangle .T- of vertices . A
-i , .1 ≤ i ≤ 3. We define a triangulation .τh
of the polyhedral open set .Ω ⊂ R , by giving a certain number of finite elements
2

.(T, Σ, P), . T ∈ τh .

Lemma 4.6.8 Let .T ∈ τh be a triangle. Then, there is a unique transformation . FT

- on .T .
who sends the triangle .T ♦
Proof The transformation . FT is given by
{
x = ξ x1 + ηx2 + (1 − ξ − η)x3
. FT :
y = ξ y1 + ηy2 + (1 − ξ − η)y3 .

We have
. -i ) = Ai
FT ( A

and
. -i -
FT ( A A j ) = Ai A j .

To any function .v, defined on .T , we associate the function .- -, by

v, defined on .T

- - = v(M),
v( M)
.

- The transformation . FT can be written

where . M = FT ( M).

. z = FT (-
z) = B-
z+b
( ) ( ) ( )
x ξ x3
where .z = , .-
z= , .b = and
y η y3
( )
x1 − x3 x2 − x3
. B= . (4.6.1)
y1 − y3 y2 − y3

This finishes the proof. Q.E.D.

@seismicisolation
@seismicisolation
236 4 Introduction to Finite Element

4.6.9 Estimates for the Errors

Let’s seek
. ||u − rh u||1,Ω .

Question: What meaning can we give to .rh u.

We have .u ∈ H 2 (Ω) as soon as . f ∈ L 2 (Ω, R). However, . H 2 (Ω) ⊂ C 0 (Ω, R). So,
.u ∈ C (Ω, R). Hence, .r h u exists. Let .τh be a triangulation of .Ω by subdividing .Ω
0

into triangles .T1 , . . . , Tm

| |
m
.Ω = Ti .
i=1

E{ (
Then,
)
||u − rh u||21,Ω =
. (u − rh u)2 + |∇(u − rh u)|2 d xd y.
T ∈τh T

4.6.9.1 -
Reference Triangle . T

The following result is nothing more than a result of change of variable in an integral.

Lemma 4.6.9 Let.T ∈ τh be a triangle and.T - be the reference triangle. Consider the
unique transformation . FT who sends the triangle .T - on .T , given in Lemma 4.6.8 i.e.,
-
. FT : T −→ T , . FT - z = b + B- z with a nonsingular matrix . B, given in (4.6.1). Then,
the mapping .u −→ - -) and for all .v ∈ H s (T ) and
u is bijective from . H s (T ) into . H s (T
.m ≤ s, we have

−1 m 1 √
.(i) .|u|m,T ≤ c||B ||2 (det B) 2 |- u |m,T- with .||B||2 := ρ(B ∗ B), and .ρ(B ∗ B) is the
greatest eigenvalue in module of the matrix . B ∗ B,
− 21
.(ii) .|-
u |m,T- ≤ c||B||m
2 (det B) |u|m,T . ♦

Proof We use the Fréchet derivatives of a function .v. If .v is .l times differentiable

at the point .z = (x, y) ∈ O, we denote . Dl v(z) its .l-th derivative at the point .z,
which is a form .l linear symmetric on .R2 . If .wi , .1 ≤ i ≤ l are .l vectors of .R2 we
note . Dl v(z) · (w1 , . . . , wl ) the value of the .l linear form . Dl v(z) on .w1 , . . . , wl . In
particular, if .(e1 , e2 ) is the canonical basis of .R2 and if .α = (α1 , α2 ) ∈ N2 , we have

.∂ α v(z) = D |α| v(z) · (e1 , . . . , e1 , e2 , . . . , e2 ).

, ,, , , ,, ,
α1 α2

We pose
|Dl v(z) · (w1 , . . . , wl )|
||D α v(z)|| =
. sup ,
w1 ,...,wl ∈R2 , w1 ,...,wl /=0 ||w1 || · · · ||wl ||

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 237

where .||wi || denotes the Euclidean norm of the vector .wi ∈ R2 . There are two
constants .c1 := c1 (l) and .c2 := c2 (l) > 0 such that
({ ) 21
c |v|l,O ≤
. 1 ||Dl v(z)||2 dz ≤ c2 |v|l,O
O

for all .v ∈ H l (O). Using the derivation of composite functions, we have

. Dl- z) · (-
v(- w1 , -
w2 ) = Dl v(z) · (B-
w1 , B-
w2 ).

We can deduce
||Dl-
. z)|| ≤ ||B||l ||Dl v(z)||.
v(-

Hence, { {
. ||Dl-
v(- z ≤ ||B||2l
z)||2 d- ||Dl v(FT (-
z))||2 d-
z.
-
T -
T

So, { {
. ||Dl- z ≤ ||B||2l (det B)−1
z)||2 d-
v(- ||Dl v(z)||2 dz.
-
T T

The existence of the constants .c1 and .c2 defined above leads to the second inequality
of the Lemma 4.6.9. The first inequality is shown in a similar way. Q.E.D.

Lemma 4.6.10 If we interpolate by the polynomials of degree .1, then

||u − r T u||0,T ≤ c||B||22 |u|2,T .

. (4.6.2)
♦

Proof We have
{
.||u − r T u||20,T = (u − r T u)2 d xd y
{T
= (-
u −-
r-
u )2 dξ dη det B.
-
T

Since
. (-
r- u (-
u )(ξ, η) = ξ- u (-
A1 ) + η- u (-
A2 ) + (1 − ξ − η)- A3 ),

then
.|(-
r-
u )(ξ, η)| ≤ 3 max |-
u (ξ, η)|.

-) is continuously injects into .C 0 (T

Gold, . H 2 (T -, R), so

. max |-
u (ξ, η)| ≤ c||-
u ||2,T-.

@seismicisolation
@seismicisolation
238 4 Introduction to Finite Element

Thus,
.|(-
r-
u )(ξ, η)| ≤ 3c||-
u ||2,T-.

-, we obtain
By integrating the last inequality on .T

||-
. r-
u ||0,T- ≤ c||-
u ||2,T-.

Thus, the continuity of .- -) into . L 2 (T

r from . H 2 (T -, R). On the other hand,

-
r-
.p=-
p

for all . -
p ∈ P1 . Hence, according to the Theorem 3.4.12,

||-
. u −-
r-
u ||0,T- ≤ β||I d − -
r ||L (H 2 (T-),L 2 (T-,R)) |-
u |2,T-.

According to the Lemma 4.6.9, we have

u |2,T- ≤ c||B||22 (det B)− 2 |u|2,T .

1
|-
.

Hence,

r ||L (H 2 (T-),L 2 (T-,R)) c||B||22 (det B)− 2 |u|2,T .

1
||-
. u −-
r-
u ||0,T- ≤ β||I d − -

Thus, if we interpolate by the polynomials of degree .1, we have

||u − r T u||0,T ≤ c||B||22 |u|2,T .

.

This completes the proof. Q.E.D.

Lemma 4.6.11 If we interpolate by the polynomials of degree .1, then

.|u − r T u|1,T ≤ c||B −1 ||2 ||B||22 |u|2,T . (4.6.3)

♦

Proof We have
| |
|∂ |
| u || = |-
u (- u (-
.
| ∂ξ -
r- A1 ) − - A3 )|

≤ 2 max |- u (ξ, η)|.

Using the Lemma 4.6.9, we obtain

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 239

|u − r T u|1,T ≤ c||B −1 ||2 (det B) 2 |-

1
. u −-
r-
u |1,T-
≤ c||B −1 ||2 (det B) 2 -
1
c|-
u |2,T-
≤ c||B −1 ||2 ||B||22 |u|2,T .

Thus, if we interpolate by the polynomials of degree .1, we have

.|u − r T u|1,T ≤ c||B −1 ||2 ||B||22 |u|2,T .

This finishes the proof. Q.E.D.

Lemma 4.6.12 We pose .h(T ) the biggest side of .T and .ρ(T ) the diameter of the
circle inscribed in .T . Then,
h(T )
.||B||2 ≤
-)
ρ(T

and

-)
h(T
.||B −1 ||2 ≤ . ♦
ρ(T )

Proof Let .-
w ∈ R2 such that
||-
. -).
w||2 = ρ(T

- such that
We can always find two points .z-1 and .z-2 ∈ T

z- − z-1 = -
. 2 w.

Hence,

. w = B z-2 − B z-1
B-
= (B z-2 + b) − (B z-1 + b)
= z2 − z1

with .z 1 and .z 2 ∈ T . Thus,

||B-
. w||2 = ||z 2 − z 1 ||2
≤ h(T ).

Moreover,
||Bx||2
||B||2 =
. sup .
||x||2 ≤1, x/=0 ||x||2

@seismicisolation
@seismicisolation
240 4 Introduction to Finite Element

Hence,
||B-w||2 h(T )
. ≤ .
||-
w||2 -)
ρ(T

Which proves
h(T )
. ||B||2 ≤ .
-)
ρ(T

Hence, the first inequality. The same goes for the second inequality. Q.E.D.

According to (4.6.2) and (4.6.3) and, using the Lemma 4.6.12, we obtain

h(T )2
.||u − r T u||0,T ≤ c |u|
-)2 2,T
ρ(T

and
-)
h(T )2 h(T
|u − r T u|1,T ≤ c
. |u| .
ρ(T ) ρ(T-)2 2,T

Hence,
E
. ||u − rh u||21,Ω = ||u − rh u||21,T
T ⊂Ω
E h(T )4
≤c |u|22,T .
T ⊂Ω
ρ(T ) 2

If we, also, have the hypothesis

h(T )
. ≤α
ρ(T )

for all .T ⊂ Ω. This assumption eliminates the triangles of poor quality (see
Sect. 4.6.6). Thus, E
.||u − r h u||1,Ω ≤ cα h(T )2 |u|22,T .
2 2

T ⊂Ω

We pose
. h = max h(T ).
T ⊂Ω

So, we have
||u − rh u||21,Ω ≤ ch 2 |u|22,Ω .
.

It follows that
||u − rh u||1,Ω ≤ ch|u|2,Ω .
.

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 241

Theorem 4.6.13 If we interpolate by the polynomials of degree 1, then

||u − rh u||1,Ω ≤ ch|u|2,Ω .

. ♦

4.6.10 Go Up in Degrees . P2

Aj Ai j Ai

Find . p ∈ P2 unique such that

. p(Ai ) = qi

and
. p(Ai j ) = qi j .

4.6.10.1 Uniqueness of the Polynomial for a Triangle with 6 Nodes

- be the reference triangle

Let .T
η

-
A2

-
A23 -
A12

-
A3 -
A13 -
A1 ξ

. -
A1 = (1, 0), .-
A2 = (0, 1), .- A12 = ( 21 , 21 ), . -
A3 = (0, 0), . - A13 = ( 21 , 0), . -
A23 = (0, 21 ).

@seismicisolation
@seismicisolation
242 4 Introduction to Finite Element

Lemma 4.6.14 Let.T - be the reference triangle. Then, there is a polynomial. - p (ξ, η) ∈
P2 (of the form .a + bξ + cη + dξ 2 + eη2 + f ξ η) unique taking arbitrary given
values .q-i .1 ≤ i ≤ 3, .q- - -
i j .1 ≤ i < j ≤ 3, at the vertices . Ai .1 ≤ i ≤ 3, . Ai j .1 ≤ i <
j ≤ 3, respectively. ♦

Proof One of the methods is as follows:

ϕ (ξ, η) is such that .ϕ1 (-
. 1 A1 ) = 1 and .ϕ1 = 0 at the other 5 points. Hence,

ϕ (ξ, η) = ξ(a + bξ + cη)

. 1

because .ϕ1 (-
A2 ) = 0 and .ϕ1 (-
A3 ) = 0 so .ϕ1 = 0 on the side .-
A2 -
A3 . We have,
⎧
⎨ ϕ1 (- A1 ) = 1
. ϕ1 ( -
A12 ) = 0
⎩
ϕ1 ( -
A13 ) = 0.

Hence, ⎧
⎨ a1 + b =b 1
⎪
. 2
(a + 2 ) = 0
⎪
⎩1
2
(a + b2 + 2c ) = 0.

Thus, ⎧
⎨ a = −1
. b=2
⎩
c = 0.

It follows that
. 1ϕ (ξ, η) = ξ(2ξ − 1).

By symmetry
. 2ϕ (ξ, η) = η(2η − 1).

ϕ (ξ, η) is such that .ϕ3 (-

. 3 A3 ) = 1 and .ϕ3 = 0 at the other 5 points. Hence,

ϕ (ξ, η) = (ξ + η − 1)(a + bξ + cη)

. 3

because .ϕ3 (-
A1 ) = 0 and .ϕ3 (-
A2 ) = 0. So, .ϕ3 = 0 on the side .-
A1 -
A2 . We have
⎧
⎨ ϕ3 (- A3 ) = 1
. ϕ3 ( -
A13 ) = 0
⎩
ϕ1 ( -
A23 ) = 0.

Hence,

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 243
⎧
⎨ a = −1
. ( 1 − 1)(−1 + b2 ) = 0
⎩ 21
( 2 − 1)(−1 + 2c ) = 0.

Thus, ⎧
⎨ a = −1
. b=2
⎩
c = 2.

It follows that
ϕ (ξ, η) = (ξ + η − 1)(2ξ + 2η − 1).
. 3

ϕ (ξ, η) is such that .ϕ12 ( -

. 12 A12 ) = 1 and .ϕ12 = 0 at the other 5 points. Hence,

ϕ (ξ, η) = aξ η
. 12

because .ϕ12 (- A1 ) = 0 and .ϕ12 (- A3 ) = 0 hence .ϕ12 = 0 on the side .- A1 -A3 and
.ϕ12 ( A 2 ) = 0 and .ϕ12 ( A 3 ) = 0 so .ϕ12 = 0 on the side . A 2 A 3 . We have .ϕ12 ( -
- - - - A12 ) = 1
so .a = 4. It follows that
.ϕ12 (ξ, η) = 4ξ η.

In the same way, we find

ϕ (ξ, η) = 4η(1 − ξ − η)
. 23

ϕ (ξ, η) = 4ξ(1 − ξ − η).

. 13

We have .q-i given and .q-

i j given. We pose

E
3 E
. -
p (ξ, η) = ϕ j (ξ, η)-
qj + ϕi j (ξ, η)-
qi j .
j=1 1≤i< j≤3

. -
p ∈ P2 , . - -i ) = q-i and . -
p( A -i j ) = q-
p( A i j . Let’s show the uniqueness of . -
p . For that,
consider the mapping

Φ : P2 −→ R6
.

-
p −→ (- -i ) 1 ≤ i ≤ 3, -
p( A -i j ) 1 ≤ i < j ≤ 3).
p( A

Φ is linear and surjective. .dim P2 = 6 and .dim R6 = 6 so .Φ is injective. Thus, the

.
uniqueness of . -
p. Q.E.D.

@seismicisolation
@seismicisolation
244 4 Introduction to Finite Element

4.6.10.2 Barycentric Coordinates

• For . P2 :
.

A2

A23 A12

A3 A13 A1 x

We have
λ (A j ) = δ1 j
. 1

. j = 1, 2, 3 with .λ1 is a polynomial of degree .1. Hence, the function of basis at point
. A1 is
.ψ1 (M) = λ1 (M)(2λ1 (M) − 1).

Likewise, for .ψ2 (M) and .ψ3 (M). Hence,

. ψ j (M) = λ j (M)(2λ j (M) − 1) 1 ≤ j ≤ 3.

Let’s seek .λ12 (M). We have

λ (M) = 4λ1 (M)λ2 (M).

. 12

So,
ψi j (M) = 4λi (M)λ j (M) 1 ≤ i < j ≤ 3.
.

Thus,
E
3 E
. p(M) = qi ψi (M) + qi j ψi j (M).
i=1 1≤i< j≤3

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 245

• For . P3 :
.

6 5

7 10 4

3 8 9 1

Find . p ∈ P3 unique such that

. p(Ai ) = qi , i = 1, . . . , 10.

We have .λ1 (A j ) = δ1 j . j = 1, . . . , 10 with .λ1 is a polynomial of degree .1. Hence,

the function of basis at point . A1 is

1
ψ1 (M) =
. λ1 (M)(3λ1 (M) − 1)(3λ1 (M) − 2).
2
Likewise, for .ψ2 (M) and .ψ3 (M). The function of basis .ψ4 is given by

9
. ψ4 (M) = λ1 (M)λ2 (M)(3λ1 (M) − 1).
2
Likewise, for .ψ5 (M), .ψ6 (M), .ψ7 (M), .ψ8 (M), and .ψ9 (M). The function of basis
ψ10 is given by
.
.ψ10 = 27 λ1 (M)λ2 (M)λ3 (M).

Thus,
E
10
. p(M) = qi ψi (M).
i=1

@seismicisolation
@seismicisolation
246 4 Introduction to Finite Element

4.6.10.3 Errors with the 6-Node Triangles

Proposition 4.6.15 Let .T be a triangle with 6 nodes. Then,

. ||u − r T u||0,T ≤ c(h(T ))3 |u|3,T (4.6.4)

and
(h(T ))3
.|u − r T u|1,T ≤ c |u|3,T . (4.6.5)
ρ(T )
♦

Proof By using Lemma 4.6.9, we have

1
||u − r T u||0,T = (det B) 2 ||-
. u −-
r-
u ||0,T-,

with
- -) −→ L 2 (T
r : H 3 (T
. -, R)

and
E
3 E
-
r-
.u= u(-
ϕ j (ξ, η)- Aj) + ϕi j (ξ, η)- -i j ).
u( A
j=1 1≤i< j≤3

We can easily verify that .-

r is continuous. Hence, using Lemma 4.6.9, we have
1
.||u − r T u||0,T = (det B) 2 ||-
u −-
r-
u ||0,T-
1
-)|-
≤ (det B) 2 c(T u |3,T-
-)(det B)− 2 ||B||32 |u|3,T .
1
≤ (det B) 2 c(T
1

Moreover,
h(T )
.||B||2 ≤
-)
ρ(T
so

-) h(T )3
.||u − r T u||0,T ≤ c(T |u|
ρ(T -)3 3,T
≤ c(h(T ))3 |u|3,T .

Thus, (4.6.4). In the same way, we can prove Eq. (4.6.5). Q.E.D.

If we, also, have the hypothesis

h(T )
. ≤α
ρ(T )

@seismicisolation
@seismicisolation
4.6 Introduction to Finite Elements in Dimension 2 for Polygonal Domain 247

for all .T ⊂ Ω. Then,

||u − u h ||1,Ω ≤ c||u − rh u||1,Ω

.
( ) 21
E
≤c ||u − rh u||1,T
2

T ⊂Ω
( ) 21
E
≤c ||u − rh u||20,T + |u − rh u|21,T
T ⊂Ω
( ) 21
E[ (h(T ))6
]
≤c (h(T )) +
6
|u|3,T
2
.
T ⊂Ω
ρ(T )2

Thus,
( ) 21
E[ ]
.||u − u h ||1,Ω ≤ c (h(T ))6 + α 2 (h(T ))4 |u|23,T .
T ⊂Ω

We pose
. h = max h(T ).
T ⊂Ω

We have
.||u − u h ||1,Ω ≤ c(αh 2 + h 3 )|u|3,Ω .

Furthermore, .h 3 ≤ h 2 , so

||u − u h ||1,Ω ≤ ch 2 |u|3,Ω .

. ♦

Theorem 4.6.16 If we interpolate by the polynomials of degree .2, we have

||u − u h ||1,Ω ≤ ch 2 |u|3,Ω .

. ♦

Usually, we have
Theorem 4.6.17 If we interpolate by the polynomials of degree .k, we have

||u − u h ||m,Ω ≤ ch k+1−m |u|k+1,Ω

.

for all .m ≤ k + 1. ♦

@seismicisolation
@seismicisolation
248 4 Introduction to Finite Element

4.7 Conforming Case: Aubin-Nitsche’s Duality Process

Let .Ω be a polygonal domain of .R2 of border .Γ = ∂Ω. We have

|(u − u h , g)0,Ω |
||u − u h ||0,Ω =
. sup .
g∈L 2 (Ω,R), ||g||0,Ω /=0 ||g||0,Ω

At .g ∈ L 2 (Ω, R) fixed, let .ϕ be the solution of

. − Δϕ = g in Ω
ϕ = 0 on Γ = ∂Ω.

Lemma 4.7.1 We have

.||ϕ||2,Ω ≤ c||g||0,Ω . (4.7.1)
♦

Proof Since the border of .Ω is regular and .g ∈ L 2 (Ω, R), then .ϕ ∈ H 2 (Ω). Hence,
we have a regularity result

.||ϕ||2,Ω ≤ c||g||0,Ω .

Q.E.D.

The variational formulation is given by

.a(ϕ, ψ) = L(ψ)

for all .ψ ∈ H01 (Ω), with

{
a(ϕ, ψ) =
. ∇ϕ∇ψ d xd y
Ω

and {
. L(ψ) = gψ d xd y.
Ω

Theorem 4.7.2 .(i) If we interpolate by . P1 , then

||u − u h ||0,Ω ≤ ch 2 |u|2,Ω .

.

@seismicisolation
@seismicisolation
4.7 Conforming Case: Aubin-Nitsche’s Duality Process 249

(ii) If we interpolate by . P2 , then

.

||u − u h ||0,Ω ≤ ch 3 |u|3,Ω .

. ♦

Proof According to Green’s formula, we have

{
(u − u h , g)0,Ω =
. (u − u h )g d xdy
{Ω
= (u − u h )(−Δϕ) d xd y
Ω
{ {
∂ϕ
= ∇(u − u h ) · ∇ϕ d xd y − (u − u h ) dσ.
Ω ∂Ω ∂ν

Thus, {
(u − u h , g)0,Ω =
. ∇(u − u h ) · ∇ϕ d xd y.
Ω

On the other hand,

. a(u, ψh ) = L(ψh )

for all .ψh ∈ Vh ⊂ H01 (Ω) and

. a(u h , ψh ) = L(ψh )

for all .ψh ∈ Vh ⊂ H01 (Ω). Hence,

a(u − u h , ψh ) = 0
.

for all .ψh ∈ Vh . Let .ψh = rh ϕ, we have

a(u − u h , rh ϕ) = 0.
.

Thus,

(u − u h , g)0,Ω = a(u − u h , ϕ)
.

= a(u − u h , ϕ − rh ϕ).

It follows that
|(u − u h , g)0,Ω | ≤ ||u − u h ||1,Ω ||ϕ − rh ϕ||1,Ω .
.

@seismicisolation
@seismicisolation
250 4 Introduction to Finite Element

(i) If we interpolate by . P1 , then, we obtain

.

|(u − u h , g)0,Ω | ≤ ch|u|2,Ω h|ϕ|2,Ω

.

≤ ch 2 |u|2,Ω |ϕ|2,Ω .

Using (4.7.1), we deduce

.|(u − u h , g)0,Ω | ≤ ch 2 |u|2,Ω ||g||0,Ω .

Thus, for .g such that .||g||0,Ω /= 0,

|(u − u h , g)0,Ω |
. ≤ ch 2 |u|2,Ω .
||g||0,Ω

By switching to sup, we get

.||u − u h ||0,Ω ≤ ch 2 |u|2,Ω .

(ii) If we interpolate by . P2 , then using the Theorem 4.6.17, we obtain

.

.|(u − u h , g)0,Ω | ≤ ||u − u h ||1,Ω ||ϕ − rh ϕ||1,Ω

≤ ch 2 |u|3,Ω h|ϕ|2,Ω .

Using (4.7.1), we deduce

. |(u − u h , g)0,Ω | ≤ ch 3 |u|3,Ω ||g||0,Ω .

Thus, for .g such that .||g||0,Ω /= 0,

|(u − u h , g)0,Ω |
. ≤ ch 3 |u|3,Ω .
||g||0,Ω

By switching to sup, we get

. ||u − u h ||0,Ω ≤ ch 3 |u|3,Ω .

This completes the proof. Q.E.D.

@seismicisolation
@seismicisolation
4.8 Introduction to Finite Elements in Dimension 2 for Non-polygonal Domain 251

4.8 Introduction to Finite Elements in Dimension .2

for Non-polygonal Domain

4.8.1 Domain and Approximate Domain

Let .Ω be a non-polygonal domain in .R2 of border .∂Ω. We consider the following

problem

. − Δu = f in Ω (4.8.1)
.u = 0 on ∂Ω. (4.8.2)

Let’s start by approaching .Ω by a polygonal domain .Ωh that we will triangulate as

previously into triangles .T .

Gh

We define . -
f by {
- f on Ω
. f =
0 elsewhere.

. Vh ⊂ H01 (Ωh ). .Vh is built from triangles with 3 or 6 nodes with .vh = 0 at the nodes
on the border .∂Ωh . The variational formulation is: Find .u h ∈ Vh such that

a (u h , vh ) = -
-
. L(vh ) (4.8.3)

for all .vh ∈ Vh with

@seismicisolation
@seismicisolation
252 4 Introduction to Finite Element
{
-
a (u h , vh ) =
. ∇u h · ∇vh d xd y
Ωh

and {
. -
L(vh ) = -
f vh d xd y.
Ωh

If .Ω is convex, then .Ωh ⊂ Ω. Note that .u h ∈ Ωh . Let’s seek the error .||u − u h ||.
Let’s pose {
∗ u h on Ωh
.u =
0 on Ω\Ωh .

Lemma 4.8.1n There is a constant .c independent of .h such that for all function
u ∈ H 2 (Ω) H01 (Ω), we have
.

h(T )
. inf |u − vh |1,Ωh ≤ ch max |u|2,Ω
vh ∈Vh T ∈τh ρ(T )

with .Vh = {vh ∈ C 0 (Ω, R) such that vh |T ∈ P1 , vh = 0 at the nodes ∈ ∂Ωh }. ♦

Proof We have
. inf |u − vh |1,Ωh ≤ |u − πh u|1,Ωh .
vh ∈Vh

Moreover, E
|u − πh u|21,Ωh =
. |u − πh u|21,T
T ∈τh

and, by using Lemma 4.6.9, we have

|u − πh u|21,T ≤ c det B||B −1 ||22 |-

. u −-
π-
u |21,T-.

According to the Theorem 3.4.12, we have

|-
.u −-
π-
u |1,T- ≤ β||I d − -
π ||L (H 2 (T-),H 1 (T-)) |-
u |2,T-.

Thus,

|u − πh u|21,T ≤ c det B||B −1 ||22 |-

. u |22,T-
≤ c det B||B −1 ||22 ||B||42 (det B)−1 |u|22,T .

Further,
-)
h(T
||B −1 ||2 ≤
.
ρ(T )

and

@seismicisolation
@seismicisolation
4.8 Introduction to Finite Elements in Dimension 2 for Non-polygonal Domain 253

h(T )
||B||2 ≤
. .
-)
ρ(T

Thus,
h(T )
.|u − πh u|1,Ωh ≤ ch max |u|2,Ω .
T ∈τh ρ(T )

Which ends the proof. Q.E.D.

Lemma 4.8.2n There is a constant .c independent of .h such that for all function
u ∈ H 3 (Ω) H01 (Ω), we have
.

3 h(T )
. inf |u − vh |1,Ωh ≤ ch 2 max ||u||3,Ω
vh ∈Vh T ∈τh ρ(T )

with .Vh = {vh ∈ C 0 (Ω, R) such that vh |T ∈ P2 , vh = 0 at the nodes ∈ ∂Ωh }. ♦

n
Proof Let .u ∈ H 3 (Ω) H01 (Ω). .πh u is continuous on .Ωh quadratic by triangle
. T ∈ τh and which interpolates .u at the top points or midpoints of sides of triangle . T
and.πh u ∈
/ Vh because if.[xi , xi+1 ] is side of.∂Ωh such that.πh u(xi ) = πh u(xi+1 ) = 0
this does not imply that .πh u = 0 on the whole side .[xi , xi+1 ] because .πh u ∈ P2 .
Hence, we, then, introduce the function .πh0 u ∈ Vh which interpolates .u at points of
the open .Ωh which are either vertices or midpoints of sides of a triangle .T ∈ τh .
Hence,
. inf |u − vh |1,Ωh ≤ |u − πh u|1,Ωh .
0
vh ∈Vh

On the other hand,

|u − πh0 u|1,Ωh ≤ |u − πh u|1,Ωh + |πh u − πh0 u|1,Ωh .

.

So, by using both Lemma 4.6.9 and Theorem 3.4.12, we have

|u − πh u|1,T ≤ c(det B) 2 ||B −1 ||2 |-

1
. u −-
π-
u |1,T-
≤ c(det B) 2 ||B −1 ||2 β||I d − -
1
π ||L (H 3 (T-),H 1 (T-)) |-
u |3,T-
≤ c(det B) 2 ||B −1 ||2 (det B)− 2 ||B||32 |u|3,T
1 1

-)
h 3 (T ) h(T
≤c 3 |u|
ρ (T -) ρ(T ) 3,T
h 3 (T )
≤c |u|3,T .
ρ(T )

@seismicisolation
@seismicisolation
254 4 Introduction to Finite Element

Thus,
⎛ ⎞ 21
E
|u − πh u|1,Ωh = ⎝
. |u − πh u|21,T ⎠
T ⊂Ωh
⎛ ⎞ 21
E h (T ) 2 ⎠
6
≤⎝ c2 |u| .
T ⊂Ωh
ρ 2 (T ) 3,T

We deduce that
h(T )
.|u − πh u|1,Ωh ≤ ch 2 max |u|3,Ω .
T ⊂Ωh ρ(T )

Note that {
0 if T ⊂ Ω
. |πh u − πh0 u|1,T = n
/= 0 if T ∂Ω /= ∅.

Hence,
⎛ ⎞ 21
E
|πh u − πh0 u|1,Ωh = ⎝
. |πh u − πh0 u|21,T ⎠ .
n
T ⊂Ωh , T ∂Ω/=∅

@seismicisolation
@seismicisolation
4.8 Introduction to Finite Elements in Dimension 2 for Non-polygonal Domain 255

On the other hand,

. (πh u − πh0 u)(Ai ) = 0 for all i = 1, 2, 3

(πh u − πh0 u)(A12 ) = 0

.

. (πh u − πh0 u)(A13 ) = 0

(πh u − πh0 u)(A23 ) = u(A23 )

.

(πh u − πh0 u)|T = u(A23 )ϕ23 ,

.

with .ϕ23 is the function of basis at point . A23 . Since

|πh u − πh0 u|1,T = |u(A23 )||ϕ23 |1,T

.

u is continuous and, using the finite increment theorem, we obtain

.

| |
| ∂u |
|u(A23 ) − u(A4 )| ≤ A23 A4 sup || || .
.
[A23 ,A4 ] ∂ y

Furthermore, . H 2 (]A23 , A4 [) is injects itself continuously into .C 0 ([A23 , A4 ], R), so

there is .α > 0 such that
| | || ||
| ∂u | || ||
. sup | | ≤ α || ∂u || ≤ α||u||3,Ω .
| ∂y | || ∂ y ||
[A23 ,A4 ] 2,Ω

We have the formula

N
.dist(x, Γ ) ≤ c h(T )2 for all x ∈ Γh T. (4.8.4)

Furthermore, since .u(A4 ) = 0 and, using (4.8.4), we have

|u(A23 )| ≤ c h(T )2 ||u||3,Ω

.

and, by using Lemma 4.6.9, we have

|ϕ23 |1,T ≤ c(det B) 2 ||B −1 ||2 |-

1
. ϕ23 |1,T-.

Since
-)
h(T
.||B −1 ||2 ≤
ρ(T )

@seismicisolation
@seismicisolation
256 4 Introduction to Finite Element

and

. det B = 2 mes(T )
1
≤ 2 h 2 (T )
2
= h 2 (T ),

we have
h(T )
|πh u − πh0 u|1,t ≤ c h(T )2
. ||u||3,Ω
ρ(T )
h 3 (T )
≤c ||u||3,Ω .
ρ(T )

Thus,
⎛ ⎞ 21
E h 6
(T )
|πh u − πh0 u|1,Ωh ≤ ⎝
. c2 2 ||u||23,Ω ⎠
n ρ (T )
T ⊂Ωh , T ∂Ω/=∅
⎛ ⎞ 21
E h(T )
≤ c⎝ h 4 (T )⎠ max ||u||3,Ω .
T ⊂∂τh
T ∈τh ρ(T )

Further,
⎛ ⎞ 21 ⎛ ⎞ 21
E E
. ⎝ h 4 (T )⎠ = h ⎝
3
2 h(T )⎠
T ⊂∂τh T ⊂∂τh
3
≤ h 2 mes(Γh )
3
≤ h 2 mes(Γ ).

So,
3 h(T )
|πh u − πh0 u|1,Ωh ≤ ch 2 max
. ||u||3,Ω .
T ∈τh ρ(T )

Thus,

3 h(T )
. inf |u − vh |1,Ωh ≤ ch 2 max ||u||3,Ω .
vh ∈Vh T ∈τh ρ(T ) Q.E.D.

@seismicisolation
@seismicisolation
4.8 Introduction to Finite Elements in Dimension 2 for Non-polygonal Domain 257

Lemma 4.8.3 There is a constant .c independent of .h such that for all function
u ∈ H 2 (Ω), we have
.

. |u|1,Ω\Ωh ≤ ch||u||2,Ω .

♦
Proof We will establish existence of a constant .c such that for all function .v ∈
H (Ω),
1

.||v||0,Ω\Ωh ≤ c(h||v||0,Γ + h |v|1,Ω\Ωh ).

2
(4.8.5)

Indeed, according to density of .D(Ω) in . H 1 (Ω), it suffices to obtain such a bound

for any function in .D(Ω). Let .v ∈ D(Ω), .T ∈ τh and . y = ϕ(x) be the equation from
arc . OT = A-2 A4 A3 of .Γ , then
{ y
∂v
.v(x, y) = v(x, ϕ(x)) + (x, t) dt.
∂
ϕ(x) y

From the Cauchy-Schwarz inequality on . OT , we have

|{ )2 ||
| y ( ∂v
| |
.v (x, y) ≤ 2v (x, ϕ(x)) + 2|ϕ(x) − y| | (x, t) dt |
2 2
| ϕ(x) ∂ y |
|{ ( ) |
| y ∂v 2 |
| |
≤ 2v2 (x, ϕ(x)) + ch 2 (T ) | (x, t) dt | .
| ϕ(x) ∂ y |

By integration on the open . OT , we have

( || ||2 )
|| ∂v ||
||v||20,OT ≤ c h 2 (T )||v||20,Γ n ∂ OT
. + h (T ) ||
4 ||
|| ∂ y || .
0,OT

By summation over all the triangles .T ∈ ∂τh , we obtain

||v||20,Ω\Ωh ≤ c(h 2 ||v||20,Γ + h 4 |v|21,Ω\Ωh ).

.

Thus, (4.8.5). Since the trace mapping.γ0 : H 1 (Ω) −→ L 2 (Γ, R),.v −→ v|Γ is linear
and continuous, we obtain .||v||0,Γ ≤ M||v||1,Ω . Thus, in vertu of (4.8.5), we have

||v||0,Ω\Ωh ≤ ch||v||1,Ω .
.

∂u ∂u
For .v = ∂x
and .v = ∂y
, we deduce
|| || || ||
|| ∂u || || ∂u ||
|| || ≤ ch || ||
.
|| ∂ x || || ∂ x ||
0,Ω\Ωh 1,Ω

@seismicisolation
@seismicisolation
258 4 Introduction to Finite Element

and || || || ||
|| ∂u || || ∂u ||
|| || ≤ ch || ||
.
|| ∂ y || || ∂ y || .
0,Ω\Ωh 1,Ω

Either by summing the last two inequalities, we get

. |u|1,Ω\Ωh ≤ ch||u||2,Ω .

Which ends the proof. Q.E.D.

We introduce the problem:

(H 1): Given a function .g ∈ L 2 (Ω, R), find .ϕ solution in .V of the problem
.

{
a(ϕ, v) =
. gv d xdy
Ω

for all .v ∈ V . This problem admits one solution and only one using the Lax-Milgram
theorem.
Definition 4.8.4 The problem .(H 1) is said to be regular if the mapping .g −→ ϕ
is linear and continuous from . L 2 (Ω, R) into . H 2 (Ω),
n i.e., for all .g ∈ L (Ω, R), the
2

solution .ϕ of the problem .(H 1) belongs to . H (Ω) V and, there is .c > 0 such that
2

||ϕ||2,Ω ≤ c||g||0,Ω .
.

♦
4.8.2 Error Between Exact Solution and Approximate
Solution

Theorem 4.8.5 Let .Ω be an open polyhedral of .Rn .n ≤ 3, and .(τh )h be a regular

family of triangulations of .Ω. It is assumed that there is .k ≥ 1 such that

. - ⊂ H 1 (T
Pk ⊂ P -).

Then, the finite element method is convergent, i.e., the solution .u h of the problem
(4.8.3) converges to the solution .u of (4.8.1)–(4.8.2) in . H 1 (Ω) and

. lim ||u − u h ||1,Ω = 0.

h→0

Furthermore, this method is of order .k, i.e., there is a constant .c independent of .h

such that if .u ∈ H k+1 (Ω), then

.||u − u h ||1,Ω ≤ ch k |u|k+1,Ω .

@seismicisolation
@seismicisolation
4.8 Introduction to Finite Elements in Dimension 2 for Non-polygonal Domain 259

If the problem .(H 1) is regular, then there is a constant .c independent of .h such that
if .u ∈ H k+1 (Ω), then

||u − u h ||0,Ω ≤ ch k+1 |u|k+1,Ω .

. ♦

Theorem 4.8.6 Let .Ω be an open convex of .R2 . Let .u be the solution of the problem:
Find .u ∈ H01 (Ω) such that

E
2 {
∂u ∂v
. ai j d x1 d x2 = f v d x1 d x2
i, j=1
∂ x j ∂ xi Ω

for all .v ∈ H01 (Ω) and .u h be the solution of the problem: Find .u h ∈ Vh such that
{
u h on Ωh
u- =
. h
0 on Ω\Ωh

checking
.a(u-h , v-h ) = L(-
vh )

for all .vh ∈ Vh with

. Vh = {vh ∈ C 0 (Ω, R) such that vh |T ∈ Pk , for all T ∈ τh and vh |Γh = 0}.

We assume that the family .τh is regular. Then, there is a constant .c independent of .h
such that
n 1
.(i) If .u ∈ H (Ω) H0 (Ω) and .k = 1, we have
2

.|u − u-h |1,Ω ≤ ch||u||2,Ω .

n
(ii) If .u ∈ H 3 (Ω)
. H01 (Ω) and .k ≥ 2, we have
3
. |u − u h |1,Ωh ≤ ch 2 ||u||3,Ω .

Furthermore, if the problem .(H 1) is regular, then

||u − u-h ||0,Ω ≤ ch 2 ||u||2,Ω .

. ♦

Proof If .k = 1, we have
( )1
|u − u-h |1,Ω = |u − u-h |21,Ω\Ωh + |u − u-h |21,Ωh 2 .
.

@seismicisolation
@seismicisolation
260 4 Introduction to Finite Element

According to the Lemma 4.8.3, we have

|u − u-h |1,Ω\Ωh ≤ ch||u − u-h ||2,Ω

.

and

|u − u-h |1,Ωh ≤ c inf |u − vh |1,Ωh

.
vh ∈Vh

≤ c|u − πh u|1,Ωh .

Thus, finishes the proof. Q.E.D.

Lemma 4.8.7 Let .τh be a regular family of triangulations for .h ≤ h 0 small enough.
- into .T ∗ with .T ∗ = F ∗ (T
Then, . F ∗ is a bijection from .T -).

-. In fact, . F ∗ − F is a map that

Proof It suffices to show that . F ∗ is injective on .T
checks

(F ∗ − F)(-
. ai ) = 0 for all i = 1, . . . , 5

(F ∗ − F)(-
. a6 ) = a6∗ − a6 .

Furthermore, . F ∗ − F ∈ P22 and . F ∗ = F + (a6∗ − a6 )- p6 , where . -

p6 ∈ P2 is the basis
- relatively to node .-
functions of triangle .T a6 . Let .-
x , .-
y∈T -, then

. F ∗ (-
x ) − F ∗ (-
y) = F(-
x ) − F(-
y) + (-
p6 (-
x) − - y))(a6∗ − a6 ).
p6 (-

@seismicisolation
@seismicisolation
4.8 Introduction to Finite Elements in Dimension 2 for Non-polygonal Domain 261

x ) = B-
Further, . F(- x + b, so

. F ∗ (-
x ) − F ∗ (-
y) = B(-x −-y) + (-
p6 (-
x) − - y))(a6∗ − a6 )
p6 (-
[ ]
= B (-x −-y) + (-
p6 (-
x) − - y))B −1 (a6∗ − a6 ) .
p6 (-

By using (4.8.4), we obtain

|B −1 (a6∗ − a6 )| ≤ ||B −1 ||2 |a6∗ − a6 |

.

h(T-) 2
≤ c h (T )
ρ(T )
h 2 (T )
≤c .
ρ(T )

Since .-
x , .- - and .T
y∈T - is compact, then

|-
.p6 (-
x) − -
p6 (-
y)| ≤ c|-
x −-
y|.

Thus,

|B −1 (F ∗ (-
. x ) − F ∗ (-
y))| ≥ |-
x −-
y| − |-
p6 (-
x) − - y)||B −1 (a6∗ − a6 )|
p6 (-
h 2 (T )
≥ |-
x −-
y| − c |-
x −-y|
ρ(T )
( )
h 2 (T )
≥ 1−c |-
x −- y|.
ρ(T )

If .τh is a regular family of triangulations and for .h ≤ h 0 small enough, we have

h 2 (T )
1−c
. > 0.
ρ(T )

-).
So, . F ∗ is injective. Thus, .T ∗ = F ∗ (T Q.E.D.

@seismicisolation
@seismicisolation
262 4 Introduction to Finite Element

4.8.3 Curved Border Elements

A2
-2
A

A12

-23 -12 A23

A A

A1
A13
-3
A
-13
A -1
A

A3

Consider the transformation

. - −→ E ⊂ P2 × P2
F:T

and .(xi , yi ) the coordinates of . Ai for .1 ≤ i ≤ 3 and .(xi j , yi j ) the coordinates of . Ai j

for .1 ≤ i < j ≤ 3. We pose
.(x, y) = F(ξ, η).

We have
E
3 E
. x= xi -
ϕi (ξ, η) + xi j -
ϕi j (ξ, η),
i=1 1≤i< j≤3

E
3 E
. y= yi -
ϕi (ξ, η) + yi j -
ϕi j (ξ, η),
i=1 1≤i< j≤3

with .-
ϕi .1 ≤ i ≤ 3 and .- -. Thus,
ϕi j .1 ≤ i < j ≤ 3 are the functions of basis of .T

. x = ξ(2ξ − 1)x1 + η(2η − 1)x2 + (1 − ξ − η)(2(1 − ξ − η) − 1)x3 + 4ξ ηx12

+4ξ(1 − ξ − η)x13 + 4η(1 − ξ − η)x23
y = ξ(2ξ − 1)y1 + η(2η − 1)y2 + (1 − ξ − η)(2(1 − ξ − η) − 1)y3 + 4ξ ηy12
+4ξ(1 − ξ − η)y13 + 4η(1 − ξ − η)y23 .

@seismicisolation
@seismicisolation
4.8 Introduction to Finite Elements in Dimension 2 for Non-polygonal Domain 263

-i ) = Ai .1 ≤ i ≤ 3 and . F( A
Hence, . F( A -i j ) = Ai j .1 ≤ i < j ≤ 3. Let us look for
F(T
. -). In fact, for example . A12 ∈ -1 A
/ [A1 , A2 ], then the range of . A -2 is a curve of
degree 2. For example .η = 0, the range of . A -1 A
-3 is given by

. x = ξ(2ξ − 1)x1 + (1 − ξ )(2(1 − ξ ) − 1)x3 + 4ξ(1 − ξ )x13

. y = ξ(2ξ − 1)y1 + (1 − ξ )(2(1 − ξ ) − 1)y3 + 4ξ(1 − ξ )y13 .

Hence,

. x = 2(x1 − x13 + x3 − x13 )ξ 2 + (4x13 − x1 − 3x3 )ξ + x3 (4.8.6)

. y = 2(y1 − y13 + y3 − y13 )ξ 2 + (4y13 − y1 − 3y3 )ξ + y3 . (4.8.7)

We notice that if the term in .ξ 2 is zero, we find the affine transformation that, we
have already studied. To find the equation of the parabola, we take .ξ 2 from (4.8.6)
and replace it in (4.8.7), we find a function in .x, . y and .ξ . We draw .ξ and replace it
in (4.8.6), we then find an equation of a parabola .x 2 as a function of . y 2 .

We assume the following hypothesis

. -i j ||2 = o(h 2 ).
||Ai j − A

We are building
. p ◦ FT−1 , -
PT = { p = - p ∈ P2 in ξ and η}.

Note that if . p ∈ PT , then . p is a function. In addition, there is continuity at the

curvilinear interface. So, we can construct the space .Vh as follows

. Vh = {vh ∈ C 0 (Ω, R) such that vh |T ∈ PT , vh = 0 at the nodes ⊂ ∂Ω}.

@seismicisolation
@seismicisolation
264 4 Introduction to Finite Element

Gh

We are building . -
f by {
- f on Ω
. f =
0 elsewhere.

Find .u h ∈ Vh such that

a (u h , vh ) = -
-
. L(vh )

for all .vh ∈ Vh with {

-
a (u, v) =
. ∇u · ∇v d xdy
Ωh

and {
. -
L(v) = -
f v d xdy.
Ωh

A2

A23 A12
-23
A -12
A
A13

A3 A1
-13
A

Under the assumptions

-i j ||2 = o(h 2 )
||Ai j − A
.

and
h(T )
. ≤ c,
ρ(T )

@seismicisolation
@seismicisolation
4.9 Rectangular Elements 265

we obtain
|u − r T u|m,T ≤ ch 3−m |u|3,T , 0 ≤ m ≤ 3,
.

with .r T u ∈ PT (function of form) and .r T u = u at the nodes. Thus,

||u − u ∗h ||1,Ω = o(h 2 )

.

with {
∗ u h on Ωh
.u h =
0 elsewhere.

4.9 Rectangular Elements

4.9.1 Transformation Between Quadrilateral

and the Reference Square

We note .ξ and .η (resp. .x and . y) the coordinates of any point . M - (resp. . M) of the
- (resp.. R). Take care of.(xi , yi ) the coordinates of the vertices. Ai ,.1 ≤ i ≤ 4,
element. R
of the quadrilateral . R, range of the square . R - of vertices . A
-i , .1 ≤ i ≤ 4.
η y

-2
A -1
A A2 a1
1 A1
-
R R
a2 Gk
a4
−1 1 ξ

A3 a3 A4

-3
A −1 -4
A x

Lemma 4.9.1 Let . R be the convex quadrilateral with vertex . Ai (xi , yi ) .1 ≤ i ≤ 4,

-i ) = Ai .1 ≤ i ≤ 4
then there is a transformation . FR ∈ Q 21 unique such that . FR ( A
given by

1 1 1 1
.x = (1 + ξ )(1 + η)x1 + (1 − ξ )(1 + η)x2 + (1 − ξ )(1 − η)x3 + (1 + ξ )(1 − η)x4
4 4 4 4

@seismicisolation
@seismicisolation
266 4 Introduction to Finite Element

1 1 1 1
.y = (1 + ξ )(1 + η)y1 + (1 − ξ )(1 + η)y2 + (1 − ξ )(1 − η)y3 + (1 + ξ )(1 − η)y4 ,
4 4 4 4

where . Q 1 := P1 + {ξ η}. ♦

4.9.2 . Q 1 Polynomial for Rectangular Element

Let . Q 1 := P1 + {ξ η}.
η

-2
A 1 -1
A

-
R

−1 1 ξ

-3
A −1 -4
A

-i ) = qi , .1 ≤ i ≤
Lemma 4.9.2 There is a polynomial . p ∈ Q 1 unique such that . p( A
4. ♦

Proof This polynomial is written as

E
4
. p(ξ, η) = qi -
ϕi (ξ, η),
i=1

with the functions of basis are given by

@seismicisolation
@seismicisolation
4.9 Rectangular Elements 267

1
-
ϕ (ξ, η) =
. 1 (1 + ξ )(1 + η),
4
1
-
ϕ (ξ, η) =
. 2 (1 − ξ )(1 + η),
4
1
-
ϕ (ξ, η) =
. 3 (1 − ξ )(1 − η),
4
and
1
-
ϕ (ξ, η) =
. 4 (1 + ξ )(1 − η).
4
This polynomial . p(ξ, η) is unique since the mapping .Φ defined by

Φ : Q 1 −→ R4
.

-1 ), p( A
p −→ ( p( A -2 ), p( A
-3 ), p( A
-4 ))

is bijective.This finishes the proof. Q.E.D.

Consider the following rectangle

A2 A1

A3 A4

Let . FR be the transformation given in Lemma 4.9.1

1 1 1 1
.x = (1 + ξ )(1 + η)x1 + (1 − ξ )(1 + η)x2 + (1 − ξ )(1 − η)x3 + (1 + ξ )(1 − η)x4
4 4 4 4

1 1 1 1
.y = (1 + ξ )(1 + η)y1 + (1 − ξ )(1 + η)y2 + (1 − ξ )(1 − η)y3 + (1 + ξ )(1 − η)y4 .
4 4 4 4

-i ) = Ai .1 ≤ i ≤ 4 and . FR ( R)
We can easily see that . FR ( A - = R. Furthermore, . FR is
-
bijective from . R into . R. Consider

@seismicisolation
@seismicisolation
268 4 Introduction to Finite Element

. p ◦ FR−1 such that -

PR = { p = - p ∈ Q 1 }.

Let . R1 and . R2 be two rectangles common sides .Γ and let . p1 and . p2 be two polyno-
mials of . Q 1 taking prescribed values at the vertices of . R1 and . R2 . Then, . p1 = p2 on
.Γ . Thus, the continuity between two neighboring rectangles. We pose

. Vh = {vh ∈ C 0 (Ω, R) such that vh |R ∈ PR , vh = 0 at the vertices ⊂ ∂Ω}.

Consider the problem: Find .u h ∈ Vh such that

a(u h , vh ) = L(vh )
.

for all .vh ∈ Vh with {

a(u h , vh ) =
. ∇u h · ∇vh d xd y
Ω

and {
. L(vh ) = f vh d xd y.
Ω

Let .ϕ1 , . . . , ϕ I be a basis of the space .Vh . Then,

{
a(ϕi , ϕ j ) =
. ∇ϕi · ∇ϕ j d xd y
{R {
∂ϕi ∂ϕ j ∂ϕi ∂ϕ j
= d xd y + d xd y
∂x ∂x R ∂y ∂y
{R {
∂-
ϕi ∂-
ϕj ∂-ϕi ∂-ϕj
= J dξ dη + J dξ dη,
-
R ∂ x ∂x -
R ∂ y ∂ y

where
∂x ∂y ∂x ∂y
. J= − .
∂ξ ∂η ∂η ∂ξ

We pose {
ϕi ∂-
∂- ϕj
. A= J dξ dη
-
R ∂ x ∂ x

and {
ϕi ∂-
∂- ϕj
. B= J dξ dη.
-
R ∂y ∂y

Then,
{ ( )( )
1 ∂-
ϕi ∂ y ∂-
ϕi ∂ y ∂-
ϕj ∂y ∂-
ϕj ∂y
. A= − − J dξ dη.
-
R J2 ∂ξ ∂η ∂η ∂ξ ∂ξ ∂η ∂η ∂ξ

@seismicisolation
@seismicisolation
References 269

∂y
It is easy to see that . ∂-
ϕi
∂ξ
is affine in .η and . ∂η is affine in .ξ . So, the terms

ϕi ∂ y
∂- ∂-
ϕi ∂ y ∂-
ϕj ∂y ∂-
ϕj ∂y
. − and −
∂ξ ∂η ∂η ∂ξ ∂ξ ∂η ∂η ∂ξ

are in . P1 but not in . Q 1 . So, we can use for calculating integrals the following
quadrature formula
{ EL
z z
. dξ dη ∼ ωl (-
- bl ),
-
R J l=1
J

ωl ∈ R, .-
where .- - .1 ≤ l ≤ N .
bl ∈ R,

References

1. P. Lesaint, Introduction aux éléments finis (Cours DEA, Besançon, 1994)

2. P.G. Ciarlet, P.-A. Raviart, General Lagrange and Hermite interpolation in.Rn with applications
to finite elements methods. Arch. Rational Mech. Anal. 46, 177–199 (1972)
3. P.G. Ciarlet, P.-A. Raviart, Interpolation theory over curved elements, with applications to finite
element methods. Comput. Methods Appl. Mech. Eng. 1, 217–249 (1972)
4. P.G. Ciarlet, Cours Monréal (1975)
5. J. Necas, Les méthodes directes en théorie des équations elliptiques. (French) Masson et Cie,
Éditeurs, Paris; Academia, Éditeurs, Prague (1967), 351 pp
6. R.A. Nicolaides, On a class of finite elements generated by Lagrange interpolation. SIAM J.
Numer. Anal. 9, 435–445 (1972)
7. P.-A. Raviart, J.-M. Thomas, Introduction à l’analyse numérique des équations aux dériveées
partielles. (Masson, 1983)
8. P.G. Ciarlet, Introduction à l’analyse numérique matricielle et à l’optimisation. (Masson, 1990)
9. P.G. Ciarlet, P.-A. Raviart, The combined effect of curved boundaries and numerical integration
in isoparametric finite element methods. The mathematical foundations of the finite element
method with applications to partial differential equations (Proc. Sympos., Univ. Maryland,
Baltimore, Md., 1972). (Academic, New York, 1972), pp. 409–474
10. J. Céa, Approximation variationnelle des problèmes aux limites. Ann. Inst. Fourier (Grenoble)
14, 345–444 (1964)

@seismicisolation
@seismicisolation
Chapter 5
Non-conforming Methods

Let.g be a polygonal domain of.R2 with piecewise smooth boundary.Γ = ∂g. Given
a function . f ∈ L 2 (g, R), we search a function .u defined in .g checking

. − /\u = f in g (5.0.1)
. u = 0 on Γ = ∂g. (5.0.2)

The variational formulation of the problem (5.0.1)–(5.0.2) is given by: Find .u ∈

H01 (g) such that

a(u, v) = L(v)
. (5.0.3)

for all .v ∈ H01 (g) with

[
a(u, v) =
. ∇u · ∇v d xdy
g

and
[
. L(v) = f v d xdy.
g

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 271
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_5
@seismicisolation
@seismicisolation
272 5 Non-conforming Methods

5.1 Triangulation of the Domain

5.1.1 Numbering of the Domain in the Middle of the Sides

Let .τh be a triangulation of .g in subdividing .g into triangles .(Ti )i=1,··· ,m which

cannot be not cover and such that

{ {
m
.g= Ti = Ti .
Ti ∈τh i=1

We number the midpoints of the sides of the triangles .(Ti )i=1,··· ,m .

Example 5.1.1 Here is an example

1 4

2 3

We want to build an approximation subspace .Vh , of finite dimension. For it, let

. Wh = {vh ∈ L 2 (g, R) such that vh |T ∈ P1 }.

Then,
. Wh ⊂ L 2 (g, R).

5.1.2 Values in the Middle of the Sides

Lemma 5.1.1 There exists a unique polynomial of degree .1 taking values given at
the midpoints of the sides. ♦

@seismicisolation
@seismicisolation
5.1 Triangulation of the Domain 273

1 -2
A -1
A
2

-3
A
0 1 1 ξ
2

Proof Method 1: We can use barycentric coordinates

. p(M) = (1 − 2λ1 (M)) p(A1 ) + (1 − 2λ2 (M)) p(A2 ) + (1 − 2λ3 (M)) p(A3 ).

Method 2: Consider the numbered reference triangle in the middle of the sides: Let’s
find the basic functions .ϕi .1 ≤ i ≤ 3. .ϕ1 (ξ, η) is such that .ϕ1 (-
A1 ) = 1, .ϕ1 (-
A2 ) = 0,
and .ϕ1 (-
A3 ) = 0. Hence, .ϕ1 (ξ, η) is of the form

ϕ (ξ, η) = a + bξ + cη.
. 1

We have ( )
1 1 b c
.ϕ1 ( -
A1 ) = ϕ1 , =a+ + =1
2 2 2 2

and ( )
1 c
ϕ (-
. 1 A2 ) = ϕ1 0, = a + = 0.
2 2

So, .b = 2. On the other hand,

( )
1 b
.ϕ1 ( -
A3 ) = ϕ1 , 0 = a + = 0.
2 2

Hence, .c = 2 and .a = −1. Thus,

ϕ (ξ, η) = −1 + 2ξ + 2η.
. 1

Likewise, .ϕ2 (ξ, η) is such that .ϕ2 (-

A2 ) = 1, .ϕ2 (-
A1 ) = 0, and .ϕ2 (-
A3 ) = 0. Hence,
.ϕ2 (ξ, η) is of the form

.ϕ2 (ξ, η) = a + bξ + cη.

@seismicisolation
@seismicisolation
274 5 Non-conforming Methods

We have ( )
1 c
ϕ (-
. 2A2 ) = ϕ2 0, =a+ =1
2 2

and ( )
1 1 b c
ϕ (-
. 2A1 ) = ϕ2 , =a+ + = 0.
2 2 2 2

Thus, .b = −2. On the other hand,

( )
1 b
.ϕ2 ( -
A3 ) = ϕ2 , 0 = a + = 0.
2 2

So, .c = 0 and .a = 1. Thus,

ϕ (ξ, η) = 1 − 2ξ.
. 2

By symmetry
ϕ (ξ, η) = 1 − 2η.
. 3

Hence, there is a polynomial . -

p of degré .1 unique taking values given to the midpoints
of the sides given by
. - p(-
p (ξ, η) = - p(-
A1 )ϕ1 (ξ, η) + - p(-
A2 )ϕ2 (ξ, η) + - A3 )ϕ3 (ξ, η). Q.E.D.

1 -2 side 2
A -1 side 1
A
2

-3 side 3
A
0 1 1 ξ
2

Remark 5.1.2 Let us denote side .i .1 ≤ i ≤ 3, the sides of the following reference
triangle

. - p(-
p (ξ, η)|side 1 = - p(-
A1 )ϕ1 (ξ, η)|side 1 + - p(-
A2 )ϕ2 (ξ, η)|side 1 + - A3 )ϕ3 (ξ, η)|side 1 .

@seismicisolation
@seismicisolation
5.2 Approximation of the Solution 275

Note that there is no continuity on side 1 except at the point .- A1 , i.e., on side 1,
there is continuity only at .-
A1 . So, for two neighboring triangles, we have continuity
only in the midpoint of the side in common. This justifies the title of this chapter
(non-conforming methods). ♦

5.2 Approximation of the Solution

5.2.1 Approximate Space

Let .Wh be the space

Wh = {vh ∈ L 2 (g, R) such that vh |Ti ∈ P1 , vh is fully determined

.
by its values in the middle of the sides}.

Hence,
. Wh ⊂ L 2 (g, R)

but
. Wh /⊂ H 1 (g).

Let .Vh be the space

. Vh = {vh ∈ Wh such that vh = 0 in the middle of the sides ⊂ ∂g}.

5.2.2 Approximate Solution

E[
We define
a (u h , vh ) =
. h ∇u h · ∇vh d xd y.
T ⊂g T

Remark 5.2.1 We can’t write

[
a (u h , vh ) =
. h ∇u h · ∇vh d xd y
g

because we don’t know if .∇u h ∈ L 2 (g, R). ♦

Let .u h be the solution of the problem: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h (5.2.1)

@seismicisolation
@seismicisolation
276 5 Non-conforming Methods

for all .vh ∈ Vh with [

. L(vh ) = f vh d xd y.
g

Remark 5.2.2 .(i) . L(vh ) has a meaning because . f ∈ L 2 (g, R) and .vh ∈ L 2 (g, R).
.(ii) There are three types of support for the bases function given in the figure opposite
♦
Let us calculate .ah (u h , ϕ5 ) according to .u i = u h (Ai ).
First type of support:

h 3 1
5

4
h

We have [ [
a (u h , ϕ5 ) =
. h ∇u h ∇ϕ5 d xd y + ∇u h ∇ϕ5 d xd y.
K T

@seismicisolation
@seismicisolation
5.2 Approximation of the Solution 277

For the triangle T:

∂u h u h (1) − u h (5) u1 − u5 2
. = = = (u 1 − u 5 )
∂x x1 − x5 h
2
h

∂ϕ5 ϕ5 (1) − ϕ5 (5) 2

. = =−
∂x x1 − x5 h

∂ϕ5 ϕ5 (5) − ϕ5 (4) 2

. = =
∂y y5 − y4 h

∂u h u h (5) − u h (4) 2
. = = (u 5 − u 4 ).
∂y y5 − y4 h

For the triangle K:

∂u h 2
. = (u 5 − u 3 )
∂x h
∂ϕ5 2
. =
∂x h
∂ϕ5 2
. =−
∂y h

∂u h 2
. = (u 2 − u 5 ).
∂y h

Thus,
[ ]
h2 4 4 4 4
.ah (u h , ϕ5 ) = − 2 (u 1 − u 5 ) + 2 (u 5 − u 4 ) + 2 (u 5 − u 3 ) − 2 (u 2 − u 5 ) .
2 h h h h

It follows that
[
a (u h , ϕ5 ) = 2(4u 5 − u 1 − u 2 − u 3 − u 4 ) =
. h f ϕ5 d xd y.
g

@seismicisolation
@seismicisolation
278 5 Non-conforming Methods

Second type of support:

3 5 1

ϕ5 ≡ 0

[ ( ) ]
h2 u1 − u5 2 u5 − u3 2
.ah (u h , ϕ5 ) = − + .
2 h h h h
2 2

Hence, [
a (u h , ϕ5 ) = 2(2u 5 − u 1 − u 3 ) =
. h f ϕ5 d xd y.
g

Third type of support:

1 4

2 3

@seismicisolation
@seismicisolation
5.2 Approximation of the Solution 279

[
a (u h , ϕ5 ) = 2(2u 5 − u 1 − u 3 ) =
. h f ϕ5 d xd y.
g

15 12

16 14 13 11 10

2 8

17 5 9
3 1 7

4 6

18

In 5 [
.2(4u 5 − u 1 − u 2 − u 3 − u 4 ) = f ϕ5 d xd y. (5.2.2)
g

In 1 [
2(2u 1 − u 5 − u 9 ) =
. f ϕ1 d xd y. (5.2.3)
g

In 2 [
2(2u 2 − u 5 − u 14 ) =
. f ϕ2 d xd y. (5.2.4)
g

In 3 [
. 2(2u 3 − u 5 − u 17 ) = f ϕ3 d xd y. (5.2.5)
g

In 4 [
. 2(2u 4 − u 5 − u 18 ) = f ϕ4 d xd y. (5.2.6)
g

We take from the equation (5.2.3), .u 1 =, from the equation (5.2.4), .u 2 =, from the
equation (5.2.5), .u 3 =, and from the equation (5.2.6), .u 4 =, and we replace them in
the equation (5.2.2), we find

@seismicisolation
@seismicisolation
280 5 Non-conforming Methods
[ [ [
1 1
4u 5 − u 9 − u 17 − u 14 − u 18 =
. f ϕ5 d xd y + f ϕ1 d xd y + f ϕ2 d xd y
g 2 [g 2 [g
1 1
. + f ϕ3 d xd y + f ϕ4 d xd y.
2 g 2 g
Thus,
[[ [
1 1 1
. (4u 5 − u 9 − u 17 − u 14 − u 18 ) = f ϕ5 d xd y + f ϕ1 d xd y
h2 [ h2 g [ 2 g [ ]
1 1 1
. + f ϕ2 d xd y + f ϕ3 d xd y + f ϕ4 d xd y .
2 g 2 g 2 g
This is the 5-point Laplacian equation.
We can show that
[ ( )
1 1 1 1 1
. f ϕ5 + ϕ1 + ϕ2 + ϕ3 + ϕ4 d xd y = f (5) + o(h 2 ).
h2 g 2 2 2 2

In fact, according to Taylor formula

∂f ∂f
. f (M) = f (5) + (x − x5 ) (5) + (y − y5 ) (5) + o(h 2 ).
∂x ∂y

Consider the formula of quadrature

[
area(T ) E
3
. g(x, y) d xd y ∼ g(Ai ).
T 3 i=1

This formula is exact for all .g ∈ P2 . We have

[ ( ) [ ]
1 1 1 1 1 h2 4 1 h2
. f (5) ϕ5 + ϕ1 + ϕ2 + ϕ3 + ϕ4 d xd y = f (5) 2 × + ×2×
g 2 2 2 2 3 2 2 3 2
= h 2 f (5).

On the other hand, .(x − x5 )ϕ5 ∈ P2 , so

[ [
∂f ∂f
. (5)(x − x5 )ϕ5 d xd y = (5) (x − x5 )ϕ5 d xd y
g ∂x ∂x g
[ ]
∂f h2
= (5) 2(x5 − x5 ) = 0,
∂x 2×3

[ [ ]
∂f ∂f h2
. (5)(x − x5 )ϕ1 d xd y = (5) 2(x1 − x5 )
g ∂x ∂x 2×3
h3 ∂ f
= (5),
6 ∂x

@seismicisolation
@seismicisolation
5.2 Approximation of the Solution 281

and
[ [ ]
∂f ∂f h2
. (5)(x − x5 )ϕ3 d xd y = (5) 2(x3 − x5 )
g ∂x ∂x 2×3
h ∂f
3
=− (5).
6 ∂x
We notice that the terms cancel each other out two by two. Thus,
[ ( )
1 1 1 1 1
. f ϕ5 + ϕ1 + ϕ2 + ϕ3 + ϕ4 d xd y = f (5) + o(h 2 )
h2 g 2 2 2 2

It follows that
1
. (4u 5 − u 9 − u 17 − u 14 − u 18 ) = f (5) + o(h 2 ).
h2

5.2.3 Existence and Uniqueness of Approximate Solution

Let .u h be the solution of the problem: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh with

E[
a (u h , vh ) =
. h ∇u h · ∇vh d xd y
T ⊂g T

and [
. L(vh ) = f vh d xd y.
g
Lemma 5.2.3 The mapping
1
v −→ ||vh ||h := (ah (vh , vh )) 2
. h

is a norm on .Vh . ♦
Proof Let .vh ∈ Vh such that
a (vh , vh ) = 0.
. h

Let us show that .vh = 0. We have .ah (vh , vh ) = 0, so

E[
. |∇vh |2 d xd y = 0.
T ⊂g T

@seismicisolation
@seismicisolation
282 5 Non-conforming Methods

Thus, for all .T ⊂ g, we have

[
. |∇vh |2 d xd y = 0.
T

It follows that
|∇vh ||T = 0.
.

It thus appears that

v
. h |T = cte.

However, between two neighboring triangles, we have continuity only in the middle
of the side in common. Thus,
.vh = cte

mostly .g. Since .vh = 0 in the middle of the sides .⊂ ∂g, we have .cte = 0. Thus,
v = 0 and so the mapping
. h

1
v −→ ||vh ||h := (ah (vh , vh )) 2
. h

is a norm on .Vh . Q.E.D.

Which proves that:

Theorem 5.2.4 The problem: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh admits a unique solution .u h ∈ Vh . ♦

5.2.4 Estimates for the Errors

Lemma 5.2.5 Let .u be the solution of the problem (5.0.3) and .u h be the solution of
the problem (5.2.1). Then,

||u h − vh ||2h = ah (u − vh , u h − vh ) − E(u, u h − vh )

.

for all .vh ∈ Vh , where

E[ ∂u
. E(u, u h − vh ) = (u h − vh ) ds. ♦
T ⊂g ∂ T
∂n T

@seismicisolation
@seismicisolation
5.2 Approximation of the Solution 283

Proof We have

||u h − vh ||2h := ah (u h − vh , u h − vh )
.

= ah (u h , u h − vh ) − ah (vh , u h − vh )
[
= f (u h − vh ) d xd y − ah (vh , u h − vh )
[g
= (−/\u)(u h − vh ) d xd y − ah (vh , u h − vh )
g
E[
= (−/\u)(u h − vh ) d xd y − ah (vh , u h − vh ).
T ⊂g T

Since .∇(u h − vh ) ∈ H 1 (T ), the Green’s formula gives

E[
||u h − vh ||2h =
. ∇u · ∇(u h − vh ) d xd y
T ⊂g T
E[ ∂u
− (u h − vh ) ds − ah (vh , u h − vh )
T ⊂g ∂ T
∂n T
= ah (u, u h − vh ) − E(u, u h − vh ) − ah (vh , u h − vh ).

Thus,
. ||u h − vh ||2h = ah (u − vh , u h − vh ) − E(u, u h − vh ). Q.E.D.

We choose .vh := rh u ∈ Vh and .rh u coincides with .u in the middle of the sides,
so by using Lemma 5.2.5, we have

||u h − rh u||2h = ah (u − rh u, u h − rh u) − E(u, u h − rh u).

.

Lemma 5.2.6 Let .u be the solution of the problem (5.0.3) and .u h be the solution of
the problem (5.2.1). Then,
. |ah (u − rh u, u h − rh u)| ≤ ch|u|2,g ||u h − rh u||h . ♦

Proof We have
E[
a (u − rh u, u h − rh u) =
. h ∇(u − rh u) · ∇(u h − rh u) d xd y.
T ⊂g T

Hence,

E ([ ) 1 ([
2
)1
2
.|ah (u − rh u, u h − rh u)| ≤ |∇(u − rh u)|2 d xd y |∇(u h − rh u)|2 d xd y .
T ⊂g T T

@seismicisolation
@seismicisolation
284 5 Non-conforming Methods

Since
. |u − rh u|1,T ≤ ch|u|2,T

and
||u − rh u||0,T ≤ ch 2 |u|2,T ,
.

we have
E
|ah (u − rh u, u h − rh u)| ≤ ch
. |u|2,T |u h − rh u|1,T
T ⊂g
( ) 21 ( ) 21
E E
≤ ch |u|22,T |u h − rh u|21,T
T ⊂g T ⊂g
≤ ch|u|2,g ||u h − rh u||h .

This completes the proof. Q.E.D.

T2 S −→n 2

a
T1
→n
− 1

On the other hand, since

∂u 1 ∂u 2
. wh (a) = − w (a)
∂n 1 ∂n 2 h

and the continuity of .wh in .a (.wh1 (a) = wh2 (a)), gives us

[ [ [ [
∂u 1 ∂u 2 ∂u 1 ∂u 2
. w ds + w ds = (w − wh1 (a)) ds + (w − wh2 (a)) ds.
S ∂n 1 h S ∂n 2 h S ∂n 1 h S ∂n 2 h

Further,
∂u ∂u ∂u
. = nx + ny.
∂n 1 ∂x ∂y

Hence, it suffices to study the two following operators

[
∂u
. (wh − wh (a)) ds
S ∂x

and [
∂u
. (wh − wh (a)) ds.
S ∂y

@seismicisolation
@seismicisolation
5.2 Approximation of the Solution 285

-2
A
A2

S
-
a
a
T
-
T
A1

-3
A -1
A A3

By making the change of variable on the reference triangle, it suffices to study

integrals of the form
[ [ ( )
∂u 1
1 ∂-
u ∂y ∂-
u ∂y ||A2 A3 ||
. (wh − wh (a)) ds = − (-
wh − -
wh (-
a )) dη
S ∂y 0 JT ∂ξ ∂η ∂η ∂ξ ||-
A2 -
A3 ||

with | ∂x |
| ∂x |
| |
| ∂ξ ∂η |
. JT := det | |. (5.2.7)
| ∂y ∂y |
| |
∂ξ ∂η
∂y ∂y
We have . JT = o(h 2 ), . ∂ξ = o(h), . ∂η = o(h) and . ||A
|| -
2 A3 ||
A -A ||
= o(h). Hence,
2 3

[ [ [
∂u 1 ∂-
u 1 ∂-
u
. (wh − wh (a)) ds = o(1) wh − -
(- wh (-
a )) dη + o(1) wh − -
(- wh (-
a )) dη.
S ∂y 0 ∂ξ 0 ∂η

We pose
[ 1
∂-
u
. u, -
F(- wh ) = (-
wh − -
wh (-
a )) dη.
0 ∂ξ

Lemma 5.2.7 The mapping .-

u −→ F(-
u, - -) into
wh ) is linear continuous from . H 2 (T
.R of norm

. |F(-
u, -
wh )| ≤ -
c|-
wh |1,T-. ♦

Proof We have
-
r-
.u =-
u

for all .-
u ∈ P1 and . F(-
u, -
wh ) is identically null for all .-
u ∈ P1 . In fact, for .-
u ∈ P1 , we
have
∂-
u
. = cte.
∂ξ

@seismicisolation
@seismicisolation
286 5 Non-conforming Methods

Hence,
[ 1
. (-
wh − -
wh (-
a )) dη = length multiplied by value in the middle = l(-
wh (-
a) − -
wh (-
a )) = 0.
0

On the one hand,

|| ||
u ||
|| ∂-
|F(-
. wh )| ≤ ||
u, - ||
|| ∂ξ || ||-
wh − -
wh (-
a )||0,(0,1) .
0,(0,1)

-) −→ L 2 (∂ T
On the other hand, the trace mapping from. H 1 (T -) is continuous. Hence,
|| || || ||
u ||
|| ∂- u ||
|| ∂-
|| || ≤ c || ||
.
|| ∂ξ || || ∂ξ || - .
0,(0,1) 1,T

Further,

.-
wh (ξ, η) = (1 − 2η)-
w1 + (1 − 2ξ )-
wh (-
a ) + (1 − 2(1 − ξ − η))-
w2 ,

with .-
w1 , .-
w2 and .-
wh (-
a ) are the values in the middle. Hence,

-
wh (0, η) − -
. wh (-
a ) = (1 − 2η)(-
w1 − -
w2 ).

Thus,
[ 1 [ 1
. (-
wh (0, η) − -
wh (-
a ))2 dη = (-
w1 − -
w2 )2 (1 − 2η)2 dη
0 0
1
= (-
w1 − -
w2 )2 .
3
On the other hand,
[ ( )2 ( )2
∂-
wh ∂-
wh
.|-
wh |21,T- = + dξ dη.
-
T ∂ξ ∂η

Further,
∂-
wh
. = −2(-
wh (-
a) − -
w2 )
∂ξ

and
∂-
wh
. = −2(-
w1 − -
w2 ).
∂η

Hence,

@seismicisolation
@seismicisolation
5.2 Approximation of the Solution 287
[
|-
.wh |21,T- = wh (-
4(- a) − -
w2 )2 + 4(-
w1 − -
w2 )2 dξ dη
-
T
[ ]
= 2 (-
wh (-
a) − -
w2 )2 + (- w2 )2 .
w1 − -

So,
([ 1 ) 21
. (-
wh (0, η) − -
wh (-
a )) dη 2
≤ α|-
wh |1,T-.
0

It follows that
.|F(-
u, -
wh )| ≤ c|-
wh |1,T-||-
u ||2,T-.

Thus, the continuity of . F. Q.E.D.

Theorem 5.2.8 Let .u be the solution of the problem (5.0.3) and .u h be the solution
of the problem (5.2.1). Then,
. ||u − u h ||h ≤ ch|u|2,g . ♦
Proof We have
. u, -
F(- wh ) = 0

for all .-
u ∈ P1 . We deduce by using the Theorem 3.4.10,

|F(-
. u, -
wh )| ≤ c|-
u |2,T-|-
wh |1,T-.

Hence,

|F(-
. wh )| ≤ ch 3 h −2 |wh |1,T |u|2,T
u, -
= ch|wh |1,T |u|2,T .

So, for .wh = u h − rh u, we have

|E(u, u h − rh u)| ≤ ch|u|2,g ||u h − rh u||h .

.

Thus,
. ||u − u h ||h ≤ ch|u|2,g .

This completes the proof. Q.E.D.

Variante: There exists a unique polynomial whose mean on each side is given by
[
1
wh (a) =
. wh ds
l(S) S

with .a midpoint of the side . S and .l(S) is the length of the side.

@seismicisolation
@seismicisolation
288 5 Non-conforming Methods

5.3 Non-conforming Case: Aubin Nitsche’s Duality Process

We will consider Aubin Nitsche’s duality process in a non-conforming case, for

example a triangle with three nodes in the middle of the sides.
Let .g be a polygonal domain of .R2 of border .Γ = ∂g. Let .τh be a triangulation of
.g in subdividing .g in triangles .(Ti )i=1,··· ,m and we will use a finite element method
with 3 nodes (degree of freedom in the middle of the sides). Consider
. Vh = {vh ∈ L 2 (g, R); vh |T ∈ P1 , vh entirely determined by values
. in the middle of the sides, vh = 0 at the nodes ⊂ ∂g}.
We have
|(u − u h , g)0,g |
.||u − u h ||0,g = sup .
g∈L 2 (g,R), ||g||0,g /=0 ||g||0,g

1 -2
A -1
A
2

-3
A
0 1 1 ξ
2

At .g ∈ L 2 (g, R) fixed, let .ϕ be the solution of

. − /\ϕ = g in g
ϕ = 0 on Γ = ∂g.

Lemma 5.3.1 We have

||ϕ||2,g ≤ c||g||0,g .
. (5.3.1)

Proof Since the border of .g is regular and .g ∈ L 2 (g, R), then .ϕ ∈ H 2 (g). Hence,
we have a regularity result
. ||ϕ||2,g ≤ c||g||0,g . Q.E.D.

@seismicisolation
@seismicisolation
5.3 Non-conforming Case: Aubin Nitsche’s Duality Process 289

The variational formulation is given by

a(ϕ, ψ) = L(ψ)
.

for all .ψ ∈ H01 (g), with

[
a(ϕ, ψ) =
. ∇ϕ∇ψ d xd y
g

and [
. L(ψ) = gψ d xd y.
g
Lemma 5.3.2 We have [
E E[ ∂ϕ
. (u − u h , g)0,g = ∇(u − u h ) · ∇ϕ d xd y − (u − u h ) dσ. ♦
T ⊂g T T ⊂g ∂ T
∂n T

Proof According to Green’s formula, we have

[
(u − u h , g)0,g =
. (u − u h )g d xdy
[g
= (u − u h )(−/\ϕ) d xd y
g
E[
= (u − u h )(−/\ϕ) d xd y
T ⊂g T
E[ E[ ∂ϕ
= ∇(u − u h ) · ∇ϕ d xd y − (u − u h ) dσ.
T ⊂g T T ⊂g ∂ T
∂n T

This completes the proof. Q.E.D.

Let .ϕh ∈ Vh be that we will choose later. We have
E[ ∂ϕ
(u − u h , g)0,g =ah (u − u h , ϕ − ϕh ) + ah (u − u h , ϕh ) −
. (u − u h ) dσ.
T ⊂g ∂ T
∂n T

Further,
[
a (u h , ϕh ) =
. h f ϕh d xd y
[g
= (−/\u)ϕh d xd y
g
E[
= (−/\u)ϕh d xd y
T ⊂g T
E[ E[ ∂u
= ∇u · ∇ϕh d xd y − ϕh dσ.
T ⊂g T T ⊂g ∂ T
∂n T

@seismicisolation
@seismicisolation
290 5 Non-conforming Methods

Thus,
E[ ∂u E[ ∂ϕ
.(u − u h , g)0,g = ah (u − u h , ϕ − ϕh ) + ϕh dσ − (u − u h ) dσ.
∂n T ∂n
T ⊂g ∂ T ∂T T
T ⊂g

E[
Note that
∂u
. ϕ dσ = 0.
T ⊂g ∂ T
∂n T

In fact, let .T1 and .T2 be two triangles of .g and . S be a side in commun.

T2 S −→n 2

a
T1
→n
− 1

Since .ϕ ∈ H 2 (g), then .ϕ ∈ C 0 (g, R). Hence, .ϕ|S

1
= ϕ|S
2
= ϕ|S and

∂u ∂u
. =− .
∂n 2 ∂n 1

So, [ [ [ ( )
∂u 1 ∂u 2 ∂u ∂u
. ϕ dσ + ϕ dσ = − ϕ dσ = 0.
S ∂n 1 S ∂n 2 S ∂n 1 ∂n 1

Thus,
E[ ∂u E[ ∂u
. (ϕh − ϕ) dσ = ϕh dσ.
T ⊂g ∂ T
∂n T T ⊂g ∂ T
∂n T

For .u ∈ P1 , we have
[ [
∂u ∂u
. (ϕh − ϕ) dσ = (ϕh − ϕ) dσ. (5.3.2)
∂T ∂n T ∂n T ∂T

We do not know if the last integral is zero or not. So, the idea is to choose a .ϕh ∈ Vh
such that the average of .ϕh on each side and on .T is equal to the average of .ϕ on each
side and on .T . Taking .ϕh as above, we find that the equation (5.3.2) is zero. Thus,
[
∂u
. (ϕh − ϕ) dσ = 0
∂T ∂n T

for all .u ∈ P1 . Hence, we have polynomial invariance for . P1 .

@seismicisolation
@seismicisolation
5.3 Non-conforming Case: Aubin Nitsche’s Duality Process 291

1 -2
A -1
A
2

-
S

-3
A
0 1 1 ξ
2

Lemma 5.3.3 The mapping

[ 1
∂-
u
l :-
. u −→ (-
ϕh − -
ϕ )(0, η) dη
0 ∂ξ

-) into .R and identically null for all .-

is linear and continuous from . H 2 (T u ∈ P1 . ♦
Proof According to the Theorem 3.4.10, we have

|l(-
. u )| ≤ c|-
u |2,T-.

We can do better. In fact,

([ ( ) ) 21 ([ ) 21
1
u 2
∂- 1
|l(-
. u )| ≤ dη (- ϕ ) dη .
ϕh − - 2
0 ∂ξ 0

-) −→ L 2 (∂ T
However, .γ : H 1 (T -, R), .v −→ v|∂ T- is linear and continuous. Hence,
|| || || ||
u ||
|| ∂- || ∂- ||
|| || || u ||
.
|| ∂ξ || - ≤ c || ∂ξ || -
0,∂ T 1,T
≤ c||-
u ||2,T-

and
.||-
ϕh − -
ϕ ||0,∂ T- ≤ c||-
ϕh − -
ϕ ||1,T-.

So,
|l(-
. u )| ≤ c||-
ϕh − -
ϕ ||1,T-||-
u ||2,T-. (5.3.3)

This completes the proof. Q.E.D.

@seismicisolation
@seismicisolation
292 5 Non-conforming Methods

Lemma 5.3.4 We have

. |l(-
u )| ≤ c||-
ϕh − -
ϕ ||1,T-|-
u |2,T-. ♦
-) into .R and .l(-
Proof Since .l(·) is linear and continuous from . H 2 (T u ) = 0 for all
.-
u ∈ P1 (see Lemma 5.3.3), we deduce by using Eq. (5.3.3) and the Theorem 3.4.10,

.|l(-
u )| ≤ c||-
ϕh − -
ϕ ||1,T-|-
u |2,T-.

This completes the proof. Q.E.D.

Le.ϕh be which we will choose a priori.rh ϕ, i.e.,.ϕh = rh ϕ. In fact,.ϕh ∈ Vh ,.ϕh |T ∈ P1 ,

r ϕ ∈ P1 and the average of .ϕh on each side is equal to the average of .ϕ on each
. h
side. Hence,
[
1
.ϕh (a) = ϕh
l(S) S
[
1
= ϕ
l(S) S
= ϕ(a),

where .a is the midpoint of the side . S. Hence, .ϕh (a) = ϕ(a). Thus, .ϕh = ϕ at mid-
points of the sides. Further, by definition of .rh ϕ, we have .rh ϕ(ai ) = ϕ(ai ) at mid-
points of the sides. Hence, .rh ϕ|T = ϕh |T at midpoints of the sides of .T . On the other
hand, .rh ϕ|T ∈ P1 and .ϕh |T ∈ P1 , so two polynomials of . P1 which coincides with two
points, so they are equal. So, .ϕh = rh ϕ. Thus,

. ||-
ϕh − -
ϕ ||1,T- ≤ c|-
ϕ |1,T-

as soon as.-
ϕ ∈ P0 . In fact,.-
ϕ = r- hϕ = -r-ϕ . Since .-
r : H 1 (T -) −→ H 1 (T -),.-
ϕ −→ -
r-
ϕ is
linear, continuous and .-r-
ϕ=- ϕ for all .-
ϕ ∈ P0 , then according to the Theorem 3.4.12,
we have
.||-
ϕ−- ϕh ||1,T- ≤ c||I d − - r ||L(H 1 (T-),H 1 (T-)) |-
ϕ |1,T-.

Hence,
||-
. ϕ−-
ϕh ||1,T- ≤ c|-
ϕ |1,T-.

Which proves

.|l(-
u )| ≤ c|-
ϕ |1,T-|-
u |2,T-
≤ c(detB)− 2 ||B||2 |ϕ|1,T (detB)− 2 ||B||22 |u|2,T
1 1

≤ c(detB)−1 ||B||32 |ϕ|1,T |u|2,T .

Further, .detB = o(h 2 ) and .||B||2 = o(h). Hence,

@seismicisolation
@seismicisolation
5.3 Non-conforming Case: Aubin Nitsche’s Duality Process 293

.|l(-
u )| ≤ ch|ϕ|1,T |u|2,T .

We can do better. In fact,

||-
. ϕh − -
ϕ ||1,T- ≤ c|-
ϕ |2,T-

as soon as .-
ϕ ∈ P1 . In fact, since .- -) −→ H 1 (T
r : H 2 (T -), .-
ϕ −→ -
r-
ϕ is linear, contin-
uous and .-
r-ϕ=- ϕ for all .-
ϕ ∈ P1 , then according to the Theorem 3.4.12, we have

||-
. ϕ−-
ϕh ||1,T- ≤ c||I d − -
r ||L(H 2 (T-),H 1 (T-)) |-
ϕ |2,T-.

Hence,
. ||-
ϕ−-
ϕh ||1,T- ≤ c|-
ϕ |2,T-.

Which implies that

|l(-
. u )| ≤ c|-
ϕ |2,T-|-
u |2,T-
≤ c(detB)− 2 ||B||22 |ϕ|2,T (detB)− 2 ||B||22 |u|2,T
1 1

≤ ch 2 |ϕ|2,T |u|2,T .

Thus,
E[ ∂u E
. (ϕh − ϕ) dσ ≤ ch 2 |ϕ|2,T |u|2,T
T ⊂g ∂ T
∂n T T ⊂g
( ) 21 ( ) 21
E E
≤ ch 2
|ϕ|2,T
2
|u|2,T
2

T ⊂g T ⊂g

≤ ch |ϕ|2,g |u|2,g .
2

Let’s seek now a increase of the term

E[ ∂ϕ
. (u − u h ) dσ.
T ⊂g ∂ T
∂n T

We cannot calculate this term because we do not know the error, .u − u h , between .u
and .u h . So, the idea is to insert .πh u. So, we will look for the following two terms
E[ ∂ϕ E [ ∂ϕ
. (u − πh u) dσ + (πh u − u h ) dσ.
T ⊂g ∂ T
∂n T T ⊂g ∂ T
∂n T

We will start first with the term

E[ ∂ϕ
. (u − πh u) dσ.
T ⊂g ∂ T
∂n T

@seismicisolation
@seismicisolation
294 5 Non-conforming Methods

Lemma 5.3.5 We have

E[ ∂ϕ
. (u − πh u) dσ ≤ ch 2 |ϕ|2,g |u|2,g . ♦
T ⊂g ∂ T
∂n T

Proof According to the trace theorem

|[ |
| 1
∂-
ϕ |
.| (-
u −-u )(0, η) dη|| ≤ c||-
r- u −-
r-
u ||1,T-||-
ϕ ||2,T-.
| ∂ξ
0

The mapping
[ 1
∂-
ϕ
. L :-
ϕ −→ (-
u −-
r-
u )(0, η) dη
0 ∂ξ

is linear and continuous from . H 2 (T -) into .R, .|L(-

ϕ )| ≤ c||-
u −-
r-
u ||1,T-||-
ϕ ||2,T- and
ϕ ) = 0 for all .-
. L(- ϕ ∈ P1 . According to the Theorem 3.4.10, we have

|L(-
. ϕ )| ≤ c||-
u −-
r-
u ||1,T-|-
ϕ |2,T-.

On the other hand,.- -) −→ H 1 (T

r : H 2 (T -),.-
u −→ -r-
u is linear, continuous and.-
r-
u =-
u
for all .-
u ∈ P1 , then according to the Theorem 3.4.12, we have

||-
. u −-
r-
u ||1,T- ≤ c||I d − -
r ||L(H 2 (T-),H 1 (T-)) |-
u |2,T-.

Hence,
||-
. u −-
r-
u ||1,T- ≤ c|-
u |2,T-.

Further,

u |2,T- ≤ c(detB)− 2 ||B||22 |u|2,T

1
|-
.

≤ ch|u|2,T .

So,
||-
. u −-
r-
u ||1,T- ≤ ch|u|2,T .

Thus,
.|L(-
ϕ )| ≤ ch 2 |ϕ|2,T |u|2,T .

It follows that

@seismicisolation
@seismicisolation
5.3 Non-conforming Case: Aubin Nitsche’s Duality Process 295

E[ ∂ϕ E
. (u − πh u) dσ ≤ ch 2 |ϕ|2,T |u|2,T
T ⊂g ∂ T
∂n T T ⊂g
( ) 21 ( ) 21
E E
≤ ch 2 |ϕ|22,T |u|22,T
T ⊂g T ⊂g

≤ ch 2 |ϕ|2,g |u|2,g .

This completes the proof. Q.E.D.

Hence, we only have the term

E[ ∂ϕ
. (πh u − u h ) dσ.
T ⊂g ∂ T
∂n T

We pose .wh = πh u − u h ∈ Vh .

T2 S −→n 2

a
T1
→n
− 1

Since
∂ϕ 1 ∂ϕ 2
. w (a) = − w (a),
∂n 1 h ∂n 2 h

we have
[ [ [ [
∂ϕ 1 ∂ϕ 2 ∂ϕ 1 ∂ϕ 2
. w dσ + w dσ = (w − wh1 (a)) dσ + (w − wh2 (a)) dσ.
S ∂n 1 h S ∂n 2 h S ∂n 1 h S ∂n 2 h

To simplify the writing, let’s set

∂ϕ ∂ϕ
. = .
∂n ∂n 1

Let’s seek this term [

∂ϕ
. (wh − wh (a)) dσ.
S ∂n

Since
∂ϕ ∂ϕ ∂ϕ
. = nx + ny,
∂n ∂x ∂y

we only have to look for the integral

@seismicisolation
@seismicisolation
296 5 Non-conforming Methods
[ [ ( )
∂ϕ 1
1 ∂-
ϕ ∂y ∂-
ϕ ∂y ||A2 A3 ||
. (wh − wh (a)) dσ = − (-
wh − -
wh (-
a )) dη
S ∂y 0 JT ∂ξ ∂η ∂η ∂ξ -2 A
|| A -3 ||

with | ∂x |
| ∂x |
| |
| ∂ξ ∂η |
. JT := det | |.
| ∂y ∂y |
| |
∂ξ ∂η

-2
A
A2

S
-
a
a T
-
T
A1

-3
A -1
A A3

∂y ∂y
It is obvious to see. J1T = o(h −2 ),. ∂ξ = o(h),. ∂η = o(h) and. ||A 2 A3 ||
- A
|| A - || = o(h). Hence, 2 3

[ [ [
∂ϕ 1 ∂-
ϕ 1 ∂-
ϕ
. (wh − wh (a)) dσ = o(1) wh − -
(- wh (-
a )) dη + o(1) wh − -
(- wh (-
a )) dη.
S ∂y 0 ∂ξ 0 ∂η

Let’s pose
[ 1
∂-
ϕ
. ϕ, -
F(- wh ) = (-
wh − -
wh (-
a )) dη.
0 ∂ξ
Lemma 5.3.6 The mapping .- ϕ −→ F(- ϕ, - -)
wh ) is linear and continuous from . H 2 (T
into .R of norm .|F(-
ϕ, -
wh )| ≤ c|-
wh |1,T-. ♦
Proof We have
|| ||
ϕ ||
|| ∂-
|F(-
. wh )| ≤ ||
ϕ, - ||
|| ∂ξ || ||-
wh − -
wh (-
a )||0,(0,1) .
0,(0,1)

-) into . L 2 (∂ T
Further, the trace mapping of . H 1 (T -, R) is continuous. Hence,
|| || || ||
ϕ ||
|| ∂- ϕ ||
|| ∂-
|| || ≤ c || ||
.
|| ∂ξ || || ∂ξ || -
0,(0,1) 1,T
≤ c||-
ϕ ||2,T-.

Furthermore,

.-
wh (ξ, η) = (1 − 2η)-
w1 + (1 − 2ξ )-
wh (-
a ) + (1 − 2(1 − ξ − η))-
w2 ,

@seismicisolation
@seismicisolation
5.3 Non-conforming Case: Aubin Nitsche’s Duality Process 297

with .-
w1 , .-
w2 and .-
wh (- -. We
a ) are the values given to the midpoints of the triangle .T
have
.-
wh (0, η) = (1 − 2η)- w1 + -
wh (-a ) + (2η − 1)-
w2 .

Hence,
-
wh (0, η) − -
. wh (-
a ) = (1 − 2η)(-
w1 − -
w2 ).

Which proves
[ 1 [ 1
. (-
wh − -
wh (-
a )) dη = (-
w1 − -
w2 )2 2
(1 − 2η)2 dη.
0 0

Thus, [ 1
1
. (-
wh − -
wh (-
a ))2 dη = (-
w1 − -
w2 )2 .
0 3

On the other hand,

[ ( )2 ( )2
∂-
wh ∂-
wh
.|-
wh |21,T- = + dξ dη.
-
T ∂ξ ∂η

We have
∂-
wh ∂-
wh
. = −2(-
wh (-
a) − -
w2 ) and = −2(-
w1 − -
w2 ).
∂ξ ∂η

Hence,
[
|-
.wh |21,T- = wh (-
4(- a) − -
w2 )2 + 4(-
w1 − -
w2 )2 dξ dη
-
T
[ ]
= 2 (-
wh (-
a) − -
w2 )2 + (- w2 )2 .
w1 − -

So,
||-
. wh − -
wh (-
a )||0,(0,1) ≤ α|-
wh |1,T-.

It follows that
|F(-
. ϕ, -
wh )| ≤ c|-
wh |1,T-||-
ϕ ||2,T-.

Thus, the continuity of . F. This completes the proof. Q.E.D.

The mapping .-
ϕ −→ F(-
ϕ, -
wh ) is identically null for all .-
ϕ ∈ P1 because
[ 1
. (-
wh − -
wh (-
a )) dη = (1 − 0) × values in the middle = -
wh (-
a) − -
wh (-
a ) = 0.
0
Lemma 5.3.7 We have
. |F(-
ϕ, -
wh )| ≤ ch|wh |1,T |ϕ|2,T . ♦

@seismicisolation
@seismicisolation
298 5 Non-conforming Methods

ϕ, -
Proof Since . F(- wh ) = 0 for all .-
ϕ ∈ P1 and, using the Theorem 3.4.10, we have

|F(-
. ϕ, -
wh )| ≤ c|-
wh |1,T-|-
ϕ |2,T-.

Further,

ϕ |2,T- ≤ c(detB)− 2 ||B||2 |wh |1,T (detB)− 2 ||B||22 |ϕ|2,T ,

1 1
|-
.wh |1,T-|-

we deduce
|F(-
. ϕ, -
wh )| ≤ ch|wh |1,T |ϕ|2,T .

This completes the proof. Q.E.D.

Lemma 5.3.8 We have

| |
| E [ ∂ϕ |
| |
. | (πh u − u h ) dσ | ≤ ch 2 |ϕ|2,g |u|2,g . ♦
| ∂n |
T ⊂g ∂ T T

Proof By Lemma 5.3.7, we have

| |
| E [ ∂ϕ | E
| |
.| (πh u − u h ) dσ | ≤ ch |πh u − u h |1,T |ϕ|2,T
| ∂n T |
T ⊂g ∂ T T ⊂g
( ) 21 ( ) 21
E E
≤ ch |πh u − u h |21,T |ϕ|22,T
T ⊂g T ⊂g
≤ ch|ϕ|2,g ||u h − πh u||h .

Furthermore,
.||u h − πh u||h ≤ ||u h − u||h + ||u − πh u||h .

Since
||u h − u||h ≤ ch|u|2,g
.

and
. ||u − πh u||h ≤ ch|u|2,g ,

we have
||u h − πh u||h ≤ ch|u|2,g .
.

@seismicisolation
@seismicisolation
5.3 Non-conforming Case: Aubin Nitsche’s Duality Process 299

Thus, | |
| E [ ∂ϕ |
| |
.| (πh u − u h ) dσ | ≤ ch 2 |ϕ|2,g |u|2,g .
| ∂ T ∂n T |
T ⊂g

This completes the proof. Q.E.D.

Lemma 5.3.9 We have

. ah (u − u h , ϕ − ϕh ) ≤ ch 2 |u|2,g |ϕ|2,g . ♦

Proof We have
E[
a (u − u h , ϕ − ϕh ) =
. h ∇(u − u h ) · ∇(ϕ − ϕh ) d xd y.
T ⊂g T

Hence, [
. ∇(u − u h ) · ∇(ϕ − ϕh ) d xd y ≤ |u − u h |1,T |ϕ − ϕh |1,T .
T

Further,

. |u − u h |1,T ≤ ||u − u h ||1,T

≤ ch|u|2,T

and

|ϕ − ϕh |1,T ≤ c(detB) 2 ||B −1 ||2 |-

1
. ϕ−-
ϕh |1,T-
≤ c|-
ϕ−-
ϕh |1,T-.

Since

.|-
ϕ−-
ϕh |1,T- ≤ ||-
ϕ−-
ϕh ||1,T-
≤ ch|ϕ|2,T ,

then we deduce
a (u − u h , ϕ − ϕh ) ≤ ch 2 |u|2,g |ϕ|2,g .
. h

This completes the proof. Q.E.D.

Theorem 5.3.10 We have

. ||u − u h ||0,g ≤ ch 2 |u|2,g . ♦

@seismicisolation
@seismicisolation
300 5 Non-conforming Methods

Proof By using Lemmas 5.3.9, 5.3.5, and 5.3.8, we have

E[ ∂u E[ ∂ϕ
.(u − u h , g)0,g = ah (u − u h , ϕ − ϕh ) + ϕh dσ − (u − u h ) dσ
∂n T ∂n
T ⊂g ∂ T ∂T
T ⊂g T

≤ ch |u|2,g |ϕ|2,g
2

≤ ch 2 |u|2,g ||ϕ||2,g .

Using (5.3.1), we deduce

.|(u − u h , g)0,g | ≤ ch 2 |u|2,g ||g||0,g .

Hence, for .g such that .||g||0,g /= 0,

|(u − u h , g)0,g |
. ≤ ch 2 |u|2,g .
||g||0,g

By switching to sup, we obtain

.||u − u h ||0,g ≤ ch 2 |u|2,g .

This completes the proof. Q.E.D.

5.4 Wilson Brick

Wilson’s non-conforming element is commonly used to solve linear elasticity prob-

lems (see [1]). We propose to solve some problem numerically using a non-
comforming finite element method, i.e., we will look for an approximate solution
.u h in a finite dimensional space . Vh / ⊂ V . This space . Vh , will be built from the finite
element of Wilson.

5.4.1 Rectangular Elements

The problem (5.0.1)–(5.0.2) takes the following variational form: Find .u ∈ H01 (g)
such that
.a(u, v) = L(v) (5.4.1)

for all .v ∈ H01 (g) with [

. a(u, v) = ∇u · ∇v d x
g

@seismicisolation
@seismicisolation
5.4 Wilson Brick 301

and [
. L(v) = f v d x.
g

We propose to solve this problem numerically by using a non-conforming finite

element method, that is to say we will look for an approximate solution .u h in a finite
dimensional space .Vh /⊂ H01 (g). This space .Vh will be built from the Wilson finite
element.
Let . R - = [−1, 1] × [−1, 1] be the square of reference. We define on . R - the fol-
- - -
lowing finite element .( R, E , P): any function .- -
v ∈ P is uniquely determined by its
data on the set .E - of the degrees of freedom made up of the values .-
vi at the vertices
-i .1 ≤ i ≤ 4 of the square and of the means .-
.A vη of the second derivatives . ∂∂ x-v2
vξ and .-
2

-
and . ∂∂ y-v2 on the square . R.
2

2 1

-
R
∧
∧
>>
ξ

3 4

Let

. - := Q 1 + {1 − ξ 2 } + {1 − η2 }
P
:= P1 + {ξ η} + {1 − ξ 2 } + {1 − η2 }.

We have
1 1 1 1
.-
vh (ξ, η) = − (1 − ξ 2 )-
vξ − (1 − η2 )-
vη + (1 + ξ )(1 + η)-
v1 + (1 − ξ )(1 + η)-
v2
2 2 4 4
1 1
. v3 + (1 + ξ )(1 − η)-
+ (1 − ξ )(1 − η)- v4 ,
4 4
with

@seismicisolation
@seismicisolation
302 5 Non-conforming Methods

∂ 2-
vh
. =-
vξ
∂ξ 2

and
∂ 2-
vh
. =-
vη .
∂η2

Note that
1 1 1
-
v (1, η) =
. h v1 + (1 − η)-
(1 + η)- v4 − (1 − η2 )-
vη .
2 2 2
So, this polynomial depends on the mean value in.η. Therefore, we have no continuity
between two neighboring rectangles. We pose

-
v(ξ, η) = v(x(ξ, η), y(ξ, η)).
. (5.4.2)

Hence, .-
v = v ◦ FR . The space of functions of form . P is given by

. p ◦ FR−1 , -
P = {p = - -
p ∈ P}.

On the quadrilateral, we define the following finite element.(R, E R , P) : any function

p ∈ P is uniquely determined by its data on the set .E R of the degrees of freedom
.
made up of the values. pi at the vertices. Ai .1 ≤ i ≤ 4 and the means second derivatives
[ [
1 ∂ 2v 1 ∂ 2v
. ds and ds,
a2 a4 [a2 a4 ] ∂s 2 a1 a3 [a1 a3 ] ∂s 2

where .s denotes an abscissa along the side .a2 a4 (resp. .a1 a3 ). Let .τh be a triangulation
of the domain .g (supposed to be polyhedral) in a convex quadrilateral . R of diameter
less than or equal to .h. We always assume, thereafter, that the following hypotheses
are uniformly satisfied in .h, there exists a .α > 0 such that

h(R)
. ≤α
ρ(R)

for all . R ∈ τh and, there is .γ ∈]0, 1[ such that

. max | cos(θiR )| < γ

1≤i≤4

for all . R ∈ τh with .h(R) designates (resp. .ρ(R)) the diameter of . R (resp. diameter
of the circle inscribed in . R) and .θiR .1 ≤ i ≤ 4 designates the angles at the vertices
of . R.

Lemma 5.4.1 Let .v and .- v be two functions corresponding by the relation (5.4.2). If
v ∈ H 2 (R), then .-
. - and vice versa and, we have
v ∈ H 2 ( R)

@seismicisolation
@seismicisolation
5.4 Wilson Brick 303

1
.(i) .|v|m,R ≤ c (sup J R ) 2 h(R)−m |-
v|m, R-,
− 21
v|m, R- ≤ c (inf J R ) h(R) |v|m,R ,
.(ii) .|-
m

with .0 ≤ m ≤ 2 and | ∂x
| ∂ x ||
| |
| ∂ξ ∂η || .
. J R := det | (5.4.3)
| ∂y ∂y |
| |
∂ξ ∂η

We pose
. Wh = {vh ∈ L 2 (g, R) such that vh |R ∈ P, vh is entirely determined by values
. at the vertices quadrilaterals and means of second derivatives}.
We define the space .Vh par

. Vh = {vh ∈ Wh such that vh = 0 at the vertices which belongs to ∂g}.

We do not have in general the inclusion

. Vh ⊂ H01 (g).

We consider the problem: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh with

E[
a (u h , vh ) =
. h ∇u h · ∇vh d x
R⊂g R

and [
. L(vh ) = f vh d x.
g

. Vh decomposes in the sum

. Vh = Vh + Vh, ,

where .Vh, ⊂ Vh is the subspace of the zero functions at the vertices of the quadrilat-
erals and .Vh ⊂ Vh is the subspace of the functions for which the degrees of freedom
defined by means on the segments .a2R a4R and .a1R a3R are zero. Furthermore,
n
. Vh ⊂ C 0 (g, R) H01 (g).

@seismicisolation
@seismicisolation
304 5 Non-conforming Methods

Definition 5.4.2 Let .- - We define .-

v ∈ H 2 ( R). r- - . P-interpolated
v ∈ P, - function of .-
v,
by
.-
r- -i ) = -
v( A -i ) 1 ≤ i ≤ 4,
v( A
[ [
∂ 2-
r-
v ∂ 2-
v
. dξ dη = dξ dη
-
R ∂ξ 2 -
R ∂ξ 2

and
[ [
∂ 2-
r-
v ∂ 2-
v
. dξ dη = dξ dη. ♦
- ∂η
R
2 - ∂η2
R

Definition 5.4.3 .(i) Let .v ∈ H 2 (R). We define .r R v ∈ P, . P-interpolated function of

.v, by
-
.r Rv = -
r-
v.

(ii) Let .v ∈ H 2 (g). We define .rh v ∈ Wh , .Wh -interpolated function of .v, by

.

. rh v|R = r R v. ♦

Lemma 5.4.4 .(i) We have the inclusion . P1 ⊂ P. Furthermore, if .v ∈ P1 , then .rh v ∈

P1 .
, , ,
.(ii) Let .vh ∈ Wh . Then, .vh = vh + vh , with .vh ∈ Wh and .vh ∈ Wh , where .vh = vh

at the vertices quadrilaterals.

.(iii) Let .vh ∈ Wh such that .vh |R ∈ P1 , for some . R ∈ τh . Then, .vh |R = vh |R and
,
.vh
|R = 0. ♦

Lemma 5.4.5 Let .vh ∈ Wh . Then,

||vh ||0,g + ||vh, ||0,g ≤ c||vh ||0,g

. (5.4.4)

and
||vh, ||0,g ≤ ch||vh ||h ,
. (5.4.5)

with
⎛ ⎞ 21
E
. ||vh ||h = ⎝ |vh |21,R ⎠ . ♦
R∈τh

Proof We have
[ [
2
. vh d xd y =
2
J R-vh dξ dη
-
R R
[
2
≤ sup J R - vh dξ dη,
-
R

@seismicisolation
@seismicisolation
5.4 Wilson Brick 305

where . J R , given in (5.4.3), is the Jacobian of the transformation . FR . On the other

hand,
[ [
. vh 2 d xd y ≤ -
c sup J R - vh2 dξ dη
R -
R
[ 2
vh
≤-c sup J R d xd y
J
[R R
sup J R
≤-c v2 d xd y.
inf J R R h

Likewise, we can show that

[ [
sup J R
. (vh, )2 d xd y ≤ -
c v2 d xd y.
R inf J R R h

Thus, (5.4.4). On the other hand,

[ [
. (vh, )2 d xd y ≤ -
c sup J R ((- vh ), )2 dξ dη
-
R
[ R

≤-c sup J R (-
vh )2 dξ dη.
-
R

Further, .-
vh −→ |-
vh |1, R- is a norm on . P -/P0 , .-
vh −→ ||- -/P0 and
vh ||0, R- is a norm on . P
( )
-
.dim P/P0 < ∞, then since in a finite dimensional space all the norms are equivalent,

we have
[
. (vh, )2 d xd y ≤ -c sup J R |-
vh |21, R-
R
sup J R
≤-
c |vh |21,R .
inf J R

Thus, (5.4.5). Q.E.D.

Lemma 5.4.6 Let .vh ∈ Wh . Then,

|vh |1,g + ||vh, ||h ≤ c||vh ||h

.

and ⎛ ⎞ 21
E
.||vh, ||h ≤ ch ⎝ |vh |22,R ⎠ . (5.4.6)
R∈τh

@seismicisolation
@seismicisolation
306 5 Non-conforming Methods

Proof Let us show the inequality (5.4.6). We have

[ ( )2 [ () ( , )2
∂vh, ∂vh, 2 ∂vh
. d xd y ≤ c + dξ dη
R ∂x -
R ∂ξ ∂η
[ (( 2 , )2 ( 2 , )2 )
∂ vh ∂ vh
≤c + dξ dη.
-
R ∂ξ 2 ∂η2

We deduce, using Lemma 5.4.1, that

[ ( )2
∂vh,
. d xd y ≤ ch 2 |vh |22,R ,
R ∂x

and the inequality (5.4.6) follows immediately. Q.E.D.

5.4.2 Approximate Problem, Patch Test for Rectangular

Meshes

Let .g be a rectangle cut into rectangles. Consider

E[
a (u h , vh ) =
. h ∇u h · ∇vh d xd y
R⊂g R

and [
. L(vh ) = f vh d xd y,
g

where .u h and .vh ∈ Vh with

{
. Vh := vh ∈ L (g, R) such that vh |R ∈ P2 (R), vh is entirely determined by the
2

. values at the vertices and by the means of derivative seconds

∂2 ∂2 }
. and on each elementR ., vh = 0 at the vertices ⊂ ∂g .
∂x2 ∂ y2
.
Consider the problem: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h (5.4.7)

for all .vh ∈ Vh .

@seismicisolation
@seismicisolation
5.4 Wilson Brick 307

Lemma 5.4.7 The mapping

1
v −→ ||vh ||h := (ah (vh , vh )) 2
. h

is a norm on .Vh and the problem (5.4.7) admits one solution and only one
u ∈ Vh .
. h ♦

Proof Let .vh ∈ Vh such that

. h a (vh , vh ) = 0.

Let us show that .vh = 0. We have

. h a (vh , vh ) = 0,

E[
hence
. |∇vh |2 d xd y = 0.
R⊂g R

Thus, for all . R ⊂ g, we have

[
. |∇vh |2 d xd y = 0.
R

It follows that
|∇vh ||R = 0.
.

It thus appears that

v
. h |R = cte.

However, between two neighboring rectangles, there is continuity at the vertices of

the sides. As a result
.vh = cte

mostly .g. Since .vh = 0 at the vertices sides .⊂ ∂g, we have .cte = 0. Thus, .vh = 0
and so the mapping
1
.vh −→ ||vh ||h := (ah (vh , vh )) 2

is a norm on .Vh . Which proves that the problem: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh admits a unique solution .u h ∈ Vh . Q.E.D.

@seismicisolation
@seismicisolation
308 5 Non-conforming Methods

5.4.3 General Estimates for the Errors

Lemma 5.4.8 Let .u be the solution of the problem (5.4.1) and let .u h be the solution
of the problem (5.4.7). Then,

|E(u, wh )|
. ||u − u h ||h ≤ c inf ||u − vh || + sup , (5.4.8)
vh ∈Vh wh ∈Vh ||wh ||h /=0 ||wh ||h

where
[ E[ ∂u
. E(u, wh ) = (/\u)wh d xd y + ah (u, wh ) = wh ds
R R⊂g ∂ R
∂n R

and . ∂n∂ R is the exterior normal derivative along .∂ R. ♦

Proof We have

||u h − vh ||2h := ah (u h − vh , u h − vh )
.

= ah (u h , u h − vh ) − ah (vh , u h − vh )
[
= f (u h − vh ) d xd y − ah (vh , u h − vh )
g
[
= (−/\u)(u h − vh ) d xd y − ah (vh , u h − vh )
g
E[
= (−/\u)(u h − vh ) d xd y − ah (vh , u h − vh )
R⊂g R
E[
∂u
= ah (u, u h − vh ) − (u h − vh ) ds − ah (vh , u h − vh )
R⊂g ∂ R
∂n R
E[ ∂u
= ah (u − vh , u h − vh ) − (u h − vh ) ds.
R⊂g ∂ R ∂n R

On the other hand,

.|ah (u − vh , u h − vh )| ≤ ||u h − vh ||h ||u − vh ||h .

Hence,
| |
|E [ ∂u |
| |
||u h −
. vh ||2h ≤ ||u h − vh ||h ||u − vh ||h + | (u h − vh ) ds | .
| ∂n |
R⊂g ∂ R R

@seismicisolation
@seismicisolation
5.4 Wilson Brick 309

Thus, for .u h /= vh , we have

| |
|E [ ∂u (u h − vh ) |
| |
.||u h − vh ||h ≤ ||u − vh ||h + | ds | .
| ∂n ||u − v || |
R⊂g ∂ R R h h h

It follows that
|E(u, wh )|
. ||u h − vh ||h ≤ ||u − vh ||h + sup .
wh ∈Vh , ||wh ||h /=0 ||wh ||h

Further,

||u − u h ||h ≤ ||u − vh ||h + ||vh − u h ||h

.

|E(u, wh )|
≤ 2||u − vh ||h + sup .
wh ∈Vh , ||wh ||h /=0 ||wh ||h

In particular, if we go to the inf

|E(u, wh )|
||u − u h ||h ≤ c inf ||u − vh ||h +
. sup .
vh ∈Vh wh ∈Vh , ||wh ||h /=0 ||wh ||h

Thus, (5.4.8). Q.E.D.

The practical convergence criterion called Patch test is written as follows:

. E(u, wh ) = 0 (5.4.9)

for all .wh ∈ Vh and all .u such that . ∂∂ux and . ∂u

∂y
are constant. The following result shows
that this criterion is satisfied in certain cases.
Lemma 5.4.9 We assume that .g is a rectangle with sides parallel to the axes, and
triangulated into rectangles. Then, the criterion (5.4.9) is satisfied. ♦
Proof We introduce the following decomposition

wh = wh + wh, ,
.

where .wh ∈ Vh and .wh = wh at the vertices of triangles. We have

. E(u, wh ) = E(u, wh − wh )
= E(u, wh, )
E[ ∂u ,
= w ds
R∈τ ∂ R
∂n R h
h

@seismicisolation
@seismicisolation
310 5 Non-conforming Methods

because for a square or rectangle

n4
-
A

-
S K S
n1
n2

A n3

[ [ [ [ [
∂u ∂u ∂u ∂u ∂u
. wh ds = wh ds + wh ds + wh ds + wh ds.
∂R ∂n R S ∂n 1 S ∂n 2
- A ∂n 3 - ∂n 4
A

Further,
∂u ∂u ∂u ∂u
. =− and =− .
∂n 1 ∂n 2 ∂n 3 ∂n 4

Hence, [
∂u
. wh ds = 0.
∂R ∂n R

Consider the contribution of a particular item . R to the above sum

η y

-2
A -1
A A2 A1

A3 A4

-3
A -4
A x

[ [ [ [ [
∂u , A1 ∂u , A2 ∂u , A1 ∂u , A4 ∂u ,
. w ds = w dy − w dy + w dx − w d x.
∂R ∂n R h A4 ∂x h A3 ∂x h A2 ∂y h A3 ∂y h

@seismicisolation
@seismicisolation
5.4 Wilson Brick 311

Moreover,
[ [ [ (- ) (- )
A1 ∂u , A2 ∂u , /\y 1 ∂u , ∂u ,
. w dy − w dy = wh (1, η) −
- w (−1, η) dη.
-
A4 ∂x h A3 ∂x h 2 −1 ∂ x ∂x h

If . ∂∂ux = cte, then . ∂∂-ux = cte. Thus,

[ 1
. (wh, (1, η) − wh, (−1, η)) dη = 0
−1

η2 −1
since .wh, (1, η) = wh, (−1, η) = 2
wη . Q.E.D.

Lemma 5.4.10 We assume that the open polyhedral.g is triangulated in any quadri-
lateral. Then, the criterion (5.4.9) is not generally satisfied. ♦

Proof For an element . R, we consider the expression

[
∂u ,
. w ds.
∂R ∂n R h

The contribution of the sides . A4 A1 and . A3 A2 to this expression is written

[ ( ) [ ( )
A1 ∂u y1 − y4 ∂u x1 − x4 A2 ∂u y2 − y3 ∂u x2 − x3
. − wh, ds − − wh, ds.
A4 ∂ x A1 A4 ∂ y A1 A4 A3 ∂ x A2 A3 ∂ y A2 A3

Again,
[ 1 [( - - ) ]
∂u ∂u
. (y1 − y4 ) − (x1 − x4 ) -wh, (1, η) dη
−1 ∂x ∂y
[ 1 [( - - ) ]
∂u ∂u
. − (y2 − y3 ) − wh, (−1, η) dη.
(x2 − x3 ) -
−1 ∂ x ∂ y
∂u ∂u
The latter expression is generally different from zero when .
∂x
and .
∂y
are
constants. Q.E.D.

Remark 5.4.11 To better understand why the criterion (5.4.9) is not satisfied (and
to remedy this drawback), we consider the following equality occurring during the
general increase of the error

a (u h − vh , u h − vh ) =( f, u h − vh )0,g − ah (vh , u h − vh )
. h

=ah (u, u h − vh ) + (−/\u, u ,h − vh, )0,g − ah (vh , u h − vh )

− ah (vh , u ,h − vh, ).

Hence,

@seismicisolation
@seismicisolation
312 5 Non-conforming Methods

a (u h − vh , u h − vh ) = ah (u − vh , u h − vh ) − ah (vh, , u h − vh ) − ah (vh , u ,h − vh, )

. h

+(−/\u, u ,h − vh, )0,g − ah (vh, , u ,h − vh, ). (5.4.10)

The last two terms of the right hand side of (5.4.10) can be grouped into .ah (u −
vh , u h − vh ) term which is easily increased by replacing .vh by .rh u function .Vh -
interpolated by .u and, we have

a (u − rh u, u h − rh u) ≤ ch|u|2,g |u h − rh u|1,g
. h

≤ ch|u|2,g ||u h − rh u||h .

The two right-hand sides of the equality (5.4.10) satisfy the Patch test with .vh = rh u.
Indeed, if . ∂∂ux and . ∂u
∂y
are constants, then

(−/\u, u ,h − vh, )0,g = 0.

.

On the other hand, if . ∂∂ux and . ∂u

∂y
are constants on each element . R, then the restriction
of function .u to this element . R is of the form .a + bx + cy. So, on the reference
square, .-
u ∈ Q 1 . Therefore, .-r- u =- u ∈ Q 1 . We deduce that .(rh u), = 0. Thus,

a ((rh u), , u ,h − (rh u), ) = 0.

. h

Finally, consider the term .ah (vh , wh, ), we have

-h ( ∂ w
E [ ∂v ) - ( -, -, ∂ x
)
, -h ∂ y ∂w-h ∂ y ∂vh ∂w ∂ x ∂w
.ah (vh , wh ) = − + − h + h
dξ dη.
- ∂x
R ∂ξ ∂η ∂η ∂ξ ∂y ∂ξ ∂η ∂η ∂ξ
R⊂g

For each element . R, we have

-, ∂ y -, ∂ y ( )
∂w ∂w ∂y ∂2 y
.
h
− h
= ξ wξ (0, 0) + ξ −
∂ξ ∂η ∂η ∂ξ ∂η ∂ξ ∂η
( )
∂y ∂2 y
. ηwη (0, 0) + η . (5.4.11)
∂ξ ∂ξ ∂η

A sufficient condition so that, .a(rh u, wh, ) = 0 for all .wh ∈ Vh and for all .u such that
∂u ∂y ∂y
.
∂x
and . ∂u
∂y
are constants, is that we replace in the expression (5.4.11), . ∂ξ and . ∂η by
∂y ∂y
.
∂ξ
(0, 0) and . ∂η (0, 0), respectively.
It is, therefore, necessary to modify the bilinear
form .ah (·, ·) in such a way that the
geometry characterizing the elements . R ∈ τh is
treated, in an exact way when considering the degrees of freedom associated with the
vertices of the elements and, in an approximate way when considering the degrees
of freedom defined by the averages on each element. ♦

@seismicisolation
@seismicisolation
5.4 Wilson Brick 313

We choose .vh := rh u ∈ Vh and .rh u coincides with .u at the vertices and the averages
of the second derivatives on each element . R, that is to say,
[ [
1 ∂ 2u 1 ∂ 2rh u
. d xd y = d xd y
mes(R) R ∂x 2 mes(R) R ∂x2

and [ [
1 ∂ 2u 1 ∂ 2rh u
. d xd y = d xd y.
mes(R) R ∂y 2 mes(R) R ∂ y2
Lemma 5.4.12 We have
([ ( )2 [ ( )2 ) 21
∂(u − rh u) ∂(u − rh u)
. d xd y + d xd y ≤ ch l−1 |u|l,R
R ∂x R ∂y

and
|ah (u − rh u, u h − rh u)| ≤ ch l−1 |u|l,g ||u h − rh u||h
.

for .l = 2 or .l = 3. ♦

Proof We have
|[ |
| |
| ∇(u − rh u) · ∇(u h − rh u) d xd y | ≤ |u − rh u|1,R |u h − rh u|1,R .
.
| |
R

Further,
[ ( )2 [ ( )2
∂(u − rh u) ∂(u − rh u)
|u − rh u|21,R =
. d xd y + d xd y.
R ∂x R ∂y

We consider the changement of variable (see Lemma 4.9.1)

1 1 1 1
.x = (1 + ξ )(1 + η)x1 + (1 − ξ )(1 + η)x2 + (1 − ξ )(1 − η)x3 + (1 + ξ )(1 − η)x4
4 4 4 4

1 1 1 1
.y = (1 + ξ )(1 + η)y1 + (1 − ξ )(1 + η)y2 + (1 − ξ )(1 − η)y3 + (1 + ξ )(1 − η)y4 .
4 4 4 4

@seismicisolation
@seismicisolation
314 5 Non-conforming Methods

2
1

3 4

We have .x1 = x4 , .x2 = x3 , . y1 = y2 and . y3 = y4 . Hence,

1+ξ 1−ξ
. x= x1 + x2
2 2
1+η 1−η
. y= y1 + y2 .
2 2
So,
x1 − x2 /\x
. dx = dξ := dξ
2 2
y1 − y2 /\y
dy =
. dη := dη.
2 2
Thus, | ∂x
| ∂ x ||
| |
| ∂ξ
. J R := det |
∂η || = /\x /\y .
| ∂y ∂y | 2 2
| |
∂ξ ∂η

It follows that
[ ( )2 (
)2
∂ 2 /\x/\y
|u
. − rh u|21,R = (-
u −-
r-
u) dξ dη
-
R ∂ξ/\x 4
[ ( ) ( )2
2 2 ∂ /\x/\y
+ (-
u −-
r-
u) dξ dη.
- /\y
R ∂η 4

On the other hand, .- - into . H 1 ( R)

r is linear and continuous from . H l ( R) - and

-
r-
.v =-
v

@seismicisolation
@seismicisolation
5.4 Wilson Brick 315

for all .-
v ∈ Pl−1 with .l = 2 or .3. Since
1 1 1
-
.r- u (ξ, η) = − (1 − ξ 2 )- u ξ − (1 − η2 )- u η + (1 + ξ )(1 + η)- u1
2 2 4
1 1 1
. + (1 − ξ )(1 + η)- u 2 + (1 − ξ )(1 − η)- u 3 + (1 + ξ )(1 − η)- u4,
4 4 4
then
| | | |
|∂ | | 1 1 1 1 |
.| (-
r -
u ) | = |ξ-
u + (1 + η)- u − (1 + η)-
u − (1 − η)-
u + (1 − η)-
u |
| ∂ξ | | ξ
4
1
4
2
4
3
4
4 |
≤ 2 max |-u i | + |-u ξ |.
1≤i≤4

Furthermore, [
1 ∂ 2-
u
-
u =
. ξ dξ dη.
4 - ∂ξ 2
R

Hence,

([ ) 21 ([ ( )2 ) 21
1 ∂ 2-
u
.|-
uξ | ≤ dξ dη dξ dη
4 -
R -
R ∂ξ 2
1
≤ |-
u | -.
2 2, R

- is injects itself continuously into .C 0 ( R,

Since . H 2 ( R) - R), we deduce that

[ | |2 ( )2
|∂ |
| (-
r -
u ) | dξ dη ≤ 2 2 max |-u | + 2|-
u ξ |2
.
|
- ∂ξ
| 1≤i≤4
i
R
≤ c||-
u ||22, R-.

In the same way, we can show that

[ | |2
|∂ |
| (- r -
u ) | dξ dη ≤ c||-
u ||22, R-.
.
-
| ∂η |
R

Thus,
. |-
r-
u |1, R- ≤ c||-
u ||2, R-.

This proves the continuity of

- - −→ H 1 ( R).
r : H 2 ( R)
. -

Since .-
r-
u =-
u for all .-
u ∈ P1 (.l = 2) and, using the Theorem 3.4.12, we obtain

@seismicisolation
@seismicisolation
316 5 Non-conforming Methods

|-
.u −-
r-
u |1, R- ≤ c||I d − -
r ||L(H 2 ( R),H - |-
- 1 ( R)) u |2, R-.

We pose
. h = max(/\x, /\y)

and, we assume that

/\y
.c≤ ≤ C.
/\x
We have
|-
.u −-
r-
u |1, R- ≤ c|-
u |l, R-

and
.|u − rh u|1,R ≤ ch l−1 |u|l,R ,

with .l = 2 or .3. Hence,

([ ( )2 [ ( )2 ) 21
∂(u − rh u) ∂(u − rh u)
. d xd y + d xd y ≤ ch l−1 |u|l,R .
R ∂x R ∂y

Thus,
|ah (u − rh u, u h − rh u)| ≤ ch l−1 |u|l,g ||u h − rh u||h .
.

This completes the proof. Q.E.D.

It remains for us to seek the estimate of

E[ ∂u
. wh ds,
R⊂g ∂ R
∂n R

with .wh = u h − vh .

-
S
R1
S
R2

We have [ [ [
∂u ∂u 1 ∂u 2
. wh ds = w ds + w ds.
S ∂n S S ∂n 1 h S ∂n 2 h

If .u ∈ P0 , then

@seismicisolation
@seismicisolation
5.4 Wilson Brick 317
[
∂u
. (u h − vh ) ds = 0.
∂R ∂n R

If .u ∈ P1 , then [
. (wh1 − wh2 ) ds /= 0.
S

So, we cannot use the Theorem 3.4.10, and so we will change the method. We will
calculate the following term
[ [
∂u 1 ∂u 1
. w ds − w ds.
S ∂n S h -
S ∂n-S h

For .u ∈ P1 , we have
[ [ [
∂u 1 ∂u 1 /\y 1
. w dy − wh dy = c -
wh (1, η) − -
wh (−1, η) dη.
S ∂n S h -
S ∂n -
S 2 −1

Furthermore,
1+η 1−η 1
-
wh (1, η) =
. -
w1 + -
w4 + (η2 − 1)-
wη
2 2 2
1+η 1−η 1
-
wh (−1, η) =
. w2 +
- w3 + (η2 − 1)-
- wη .
2 2 2
Hence,
-
wh (1, η) − -
. wh (−1, η) /= 0.

So, we cannot use the Theorem 3.4.10, and so we will change another times of
method. We are going to work with the polynomial . Q 1 . We consider .sh v |R ∈ Q 1 and

s v=v
. h

at the vertices of rectangles. Hence, .sh v ∈ C 0 (g, R). So, for .u ∈ P1 , find the integral
[ [
∂u 1 ∂u 1
. (w − sh wh1 ) dy − (wh − sh wh1 ) dy
S ∂x h S ∂x
-

come back to find the integral

[ [
. (wh1 − sh wh1 ) dy − (wh1 − sh wh1 ) dy. (5.4.12)
S -
S

Furthermore,
1+η 1−η
-wh |(1,η) =
s-
. w1 +
- -
w4
2 2

@seismicisolation
@seismicisolation
318 5 Non-conforming Methods

and
1+η 1−η
-wh |(−1,η) =
s-
. w2 +
- w3 .
-
2 2

So, the equation (5.4.12) is null if .mes(S) = mes(-

S). Hence, we arrive at the poly-
nomial invariance. Thus,
[ [
∂u 1 ∂u 1
. (wh − sh wh1 ) d y − (wh − sh wh1 ) d y =
S ∂x [[ 1 -
S ∂ x
[ 1 ]
2 /\y ∂-
u ∂-
u
. (1, η)(-wh −- wh )(1, η) dη −
s- wh −-
(−1, η)(- wh )(−1, η) dη .
s-
/\x 2 −1 ∂ξ −1 ∂ξ

We pose
[ [
1 ∂-
u 1 ∂-
u
u, -
.G(- wh ) = wh −-
(1, η)(- wh )(1, η) dη −
s- wh −-
(−1, η)(- wh )(−1, η) dη.
s-
−1 ∂ξ −1 ∂ξ

Lemma 5.4.13 The mapping

.-
u −→ G(-
u, -
wh )

- into .R of norm
is linear continuous from . H 2 ( R)

|G(-
. u, -
wh )| ≤ c|-
wh |1, R-,

identically null for all .-

u ∈ P1 . ♦
Proof Since
1
.(-
wh −-wh )(−1, η) = − (1 − η2 )-
s- wη
2
and
1
. (-
wh −-wh )(1, η) = − (1 − η2 )-
s- wη ,
2
u, -
We have .G(- wh ) = 0 for all .-
u ∈ P1 . On the other hand,
[ || ||
1
∂-
u || ∂-
u ||
. (1, η)(-
wh −-wh )(1, η) dη ≤ ||
s- || (1, η) ||
|| ||(-
wh −-wh )(1, η)||0,(−1,1) .
s-
−1 ∂ξ ∂ξ 0,(−1,1)

Further,
- −→ L 2 (∂ R,
γ : H 1 ( R)
. - R)

is linear and continuous. Hence,

|| || || ||
|| ∂-
u || u ||
|| ∂-
|| || ≤ c || ||
.
|| ∂ξ (1, η)|| || ∂ξ || -
0,(−1,1) 1, R
≤ c||-
u ||2, R-

@seismicisolation
@seismicisolation
5.4 Wilson Brick 319

1
(-
.wh −-wh )(1, η) = − (1 − η2 )-
s- wη
2
and
[ 1
1
.||(-
wh −-wh )(1, η)||20,(−1,1)
s- = (1 − η2 )2 dη -
wη2
4 −1
≤ wη2 .
c-

Furthermore,
[ 2
1 ∂ -
wh
.-
wη = dξ dη
-
mes( R) R- ∂η2
[ 2
1 ∂ -
wh
= dξ dη
4 R- ∂η2
([ ( )2 ) 21
1 ∂ 2-
wh
≤ dξ dη .
2 -
R ∂η2

Hence,
1
|-
.wη | ≤ ||-
wh ||2, R-.
2
Thus,
.||(-
wh −-wh )(1, η)||0,(−1,1) ≤ c||-
s- wh ||2, R-.

On the other hand,

1 1 1
wh (ξ, η) = − (1 − ξ 2 )-
.- wξ − (1 − η2 )- wη + (1 + ξ )(1 + η)- w1
2 2 4
1 1 1
. + (1 − ξ )(1 + η)- w2 + (1 − ξ )(1 − η)- w3 + (1 + ξ )(1 − η)- w4 .
4 4 4
Method 1: We calculate .||-
wh ||2, R-, then we calculate .|-
wh |1, R- and, we easily show that

||-
. wh ||2, R- ≤ c|-
wh |1, R-.

Method 2: .-
v −→ ||-
v||2, R- and .-
v −→ |-
v|1, R- are two norms on . P2/P0 . Since .dim(P2/P0 )
< ∞, then all norms are equivalent. Hence, there is .c > 0 such that

||-
. wh ||2, R- ≤ c|-
wh |1, R-.

Thus,
|G(-
. u, -
wh )| ≤ c|-
wh |1, R-||-
u ||2, R-.

This completes the proof. Q.E.D.

@seismicisolation
@seismicisolation
320 5 Non-conforming Methods

Theorem 5.4.14 We have

. ||u − u h ||h ≤ c(h|u|2,g + h 2 |u|3,g ). ♦
Proof According to the Theorem 3.4.10, we have

|G(-
. u, -
wh )| ≤ c|-
wh |1, R-|-
u |2, R-.

Further, .|-
u |2, R- ≤ ch|u|2,R and .|-
wh |1, R- ≤ c|wh |1,R . Thus,

|G(-
. u, -
wh )| ≤ ch|wh |1,R |u|2,R .

For .wh = u h − vh , we have

E[ ∂u
. (u h − rh u) ds ≤ ch|u|2,g ||u h − rh u||h .
R⊂g ∂ R
∂n R

So,
||u h − rh u||h ≤ ch|u|2,g .
.

Furthermore,
||u − u h ||h ≤ ||u − rh u||h + ||rh u − u h ||h .
.

Since .||u − rh u||h ≤ ch 2 |u|3,g , we have

||u − u h ||h ≤ c(h|u|2,g + h 2 |u|3,g ).

.

This completes the proof. Q.E.D.

Question: Can we improve the upper bound of .|G(- u, -

wh )|, i.e., instead of being
identically zero for .-
u ∈ P1 is it possible that it is identically zero for .-
u ∈ P2 .
In fact, if .-
u = η2 , then
. u, -
G(- wh ) = 0.

If .-
u = ξ η, then
. u, -
G(- wh ) = 0.

In fact,
[ 1 [ 1
. η(-
wh −-wh )(1, η) dη −
s- η(-
wh −- wh )(−1, η) dη
s-
−1
[ 1−1 [
1 1 1
.= − η(1 − η ) dη - wη + η(1 − η2 ) dη -
wη = 0.
2
2 −1 2 −1
If .-
u = ξ 2 , then
. u, -
G(- wh ) /= 0.

@seismicisolation
@seismicisolation
5.4 Wilson Brick 321

In fact,
[ 1 [ 1
. 2(-wh −- wh )(1, η) dη +
s- wh −-
2(- wh )(−1, η) dη
s-
−1 −1
[ 1 [ 1
1 1 8
.= − ×2 wη − × 2
(1 − η ) dη - wη = − -
(1 − η2 ) dη - wη /= 0.
2
2 −1 2 −1 3
So, we have the polynomial invariance only for . P1 . Thus,

.||u − u h ||h = o(h).

Remark 5.4.15 The set of points . A -i .5 ≤ i ≤ 8, is not . Q 1 -unisolvent i.e., there does
not exist a polynomial of degree. Q 1 taking arbitrary values.αi at points. A -i ,.5 ≤ i ≤ 8.
On the other hand, we have the following result: There exists a unique polynomial
-i , .5 ≤ i ≤ 8, from that the
of . P1 , taking arbitrary given values .αi , at the points . A
following relation is satisfied
η

-2 -5
1 A
A -1
A

-
R

-6
A -8
A
−1 -9
A 1 ξ

-3
A -7
−1 A -4
A

α + α7 = α2 + α4 .
. 5

This still allows us to define a quadrilateral finite element, with . K - = [−1, 1] ×

-
[−1, 1], . P = P1 . Any function . - -
p ∈ P is expressed in terms of its values . - -i ) at
p( A
-
points . Ai , .5 ≤ i ≤ 8, as follows

. p(ξ, η) =
1
(- -5 ) + -
p( A -7 )) + ξ (-
p( A -8 ) − -
p( A -6 )) + η (-
p( A -5 ) − -
p( A -7 )),
p( A
2 2 2

@seismicisolation
@seismicisolation
322 5 Non-conforming Methods

and, the following relation is satisfied

. -5 ) + -
p( A -7 )) = p( A
p( A -6 ) + - -8 )
p( A
= 2- -9 ).
p( A

Let.- -i ,.1 ≤ i ≤ 4. The. P-interpolated

v be any function defined at the points. A - function
.-
π-
v of .-
v is the unique satisfying polynomial of . P1 satisfying

-
π-
. -i ) = 1 (-
v( A -i−4 ) +-
v( A -i−3 )),
v( A
2
for .5 ≤ i ≤7. We, then, have
. -
π- -8 ) = 1 (-
v( A -1 ) +-
v( A -4 )).
v( A ♦
2

5.5 Finite Element of the Serendipity Family

- = [−1, 1] × [−1, 1] be the square of reference and let

Let . R

. - = Q 2 \{ξ 2 η2 } + {ξ 3 } + {η3 } = P3 .
P

-2 -12
1A
A -1
A

-
≫

-23
A -14
A
≫
−1 1 ξ

-3
A -34
−1 A -4
A

We pose

@seismicisolation
@seismicisolation
5.5 Finite Element of the Serendipity Family 323
{
. Vh := vh ∈ L 2 (g, R) such that vh |R ∈ P3 (R), vh entirely determined by values at
∂3
. the vertices and midpoints of the sides and by the means of the third derivatives
∂x3
∂3
. and 3 on each elementR., vh = 0 at the vertices and in the middle of the sides ⊂
}∂y
∂g .
- So, this is a non-conforming finite element.
There is no continuity at the edges of . R.
Question: Let’s seek the error .||u − u h ||0,g , by applying Aubin Nitsche’s duality
process.
In fact,
|(u − u h , g)0,g |
.||u − u h ||0,g = sup .
g∈L 2 (g,R), ||g||0,g /=0 ||g||0,g

At .g ∈ L 2 (g, R) fixed, let .ϕ be the solution of

. − /\ϕ = g in g
ϕ = 0 on Γ = ∂g.

Lemma 5.5.1 We have

. ||ϕ||2,g ≤ c||g||0,g . ♦

Proof Since the border of .g is regular and .g ∈ L 2 (g, R), then .ϕ ∈ H 2 (g). Hence,
we have a regularity result
. ||ϕ||2,g ≤ c||g||0,g . Q.E.D.

The variational formulation is given by

.a(ϕ, ψ) = L(ψ)

for all .ψ ∈ H01 (g), with

[
a(ϕ, ψ) =
. ∇ϕ∇ψ d xd y
g

and [
. L(ψ) = gψ d xd y.
g

Lemma 5.5.2 We have

E[ E[ ∂ϕ
. (u − u h , g)0,g = ∇(u − u h ) · ∇ϕ d xd y − (u − u h ) dσ. ♦
R⊂g R R⊂g ∂ R
∂n R

@seismicisolation
@seismicisolation
324 5 Non-conforming Methods

Proof According to Green’s formula, we have

[
. (u − u h , g)0,g = (u − u h )g d xdy
g
[
= (u − u h )(−/\ϕ) d xd y
g
E[
= (u − u h )(−/\ϕ) d xd y
R⊂g R
E[ E[ ∂ϕ
= ∇(u − u h ) · ∇ϕ d xd y − (u − u h ) dσ.
R⊂g R R⊂g ∂ R
∂n R

This completes the proof. Q.E.D.

Let .ϕh ∈ Vh be that we will choose later. We have

E[ ∂ϕ
(u − u h , g)0,g =ah (u − u h , ϕ − ϕh ) + ah (u − u h , ϕh ) −
. (u − u h ) dσ.
R⊂g ∂ R
∂n R

Further,
[
a (u h , ϕh ) =
. h f ϕh d xd y
[g
= (−/\u)ϕh d xd y
g
E[
= (−/\u)ϕh d xd y
R⊂g R
E[ E[ ∂u
= ∇u · ∇ϕh d xd y − ϕh dσ.
R⊂g R R⊂g ∂ R
∂n R

Thus,
E[ ∂u E[ ∂ϕ
.(u − u h , g)0,g = ah (u − u h , ϕ − ϕh ) + ϕh dσ − (u − u h ) dσ.
∂n R ∂n
R⊂g ∂ R ∂R
R⊂g R

We will start with

E[
a (u − u h , ϕ − ϕh ) =
. h ∇(u − u h ) · ∇(ϕ − ϕh ) d xd y.
R⊂g R

Lemma 5.5.3 We have

. ah (u − u h , ϕ − ϕh ) ≤ ch 2 |u|2,g |ϕ|2,g . ♦

@seismicisolation
@seismicisolation
5.5 Finite Element of the Serendipity Family 325

Proof We have
[
. ∇(u − u h ) · ∇(ϕ − ϕh ) d xd y ≤ |u − u h |1,R |ϕ − ϕh |1,R
R
≤ ch|u|2,R ch|ϕ|2,R
≤ ch 2 |u|2,R |ϕ|2,R .

Thus,
a (u − u h , ϕ − ϕh ) ≤ ch 2 |u|2,g |ϕ|2,g .
. h

This completes the proof. Q.E.D.

On the other hand, .ϕ ∈ H 2 (g) ⊂ C 0 (g, R). Hence,

E[ ∂u
. ϕ dσ = 0.
R⊂g ∂ R
∂n R

Thus,
E[ ∂u E[ ∂u
. ϕh dσ = (ϕh − ϕ) dσ.
R⊂g ∂ R
∂n R R⊂g ∂ R
∂n R

For .u ∈ P1 , we have
[
∂u
. A := (ϕh − ϕ) dσ
∂ R ∂n R
[
∂u
= (ϕh − ϕ) dσ.
∂n R ∂ R

We do not know if the last integral is zero or not. So, the idea is to choose .ϕh ∈ Vh
such that the average of each .ϕh on each side is equal to the average of each .ϕ on
each side. So, taking .ϕh as above, we find that the quantity . A is zero. Thus,
[
∂u
. (ϕh − ϕ) dσ = 0
∂R ∂n R

for all .u ∈ P1 . Consider

[ [
1
∂-
u 1
∂-
u
. u) =
l(- (1, η)(-
ϕh − -
ϕ )(1, η) dη − (−1, η)(-
ϕh − -
ϕ )(−1, η) dη.
−1 ∂ξ −1 ∂ξ

@seismicisolation
@seismicisolation
326 5 Non-conforming Methods

-
S -
R

−1

. - into .R and identically null for all .-

l(·) is linear continuous from . H 2 ( R) u ∈ P1 of
norm .≤ c||- ϕh − -ϕ ||1, R-. Hence, according to the Theorem 3.4.10, we have

|l(-
. u )| ≤ c||-
ϕh − -
ϕ ||1, R-|u|2, R-.

Further,
. ||-
ϕh − -
ϕ ||1, R- ≤ c|-
ϕ |1, R-

as soon as .-
ϕ ∈ P0 . Hence,

|l(-
. u )| ≤ ch|ϕ|1,R |u|2,R .

We can do better. In fact,

||-
. ϕh − -
ϕ ||1, R- ≤ c|-
ϕ |2, R-

as soon as .-
ϕ ∈ P1 . Hence,

|l(-
. u )| ≤ ch 2 |ϕ|2,R |u|2,R .

Thus, | |
|E [ ∂u |
| |
.| ϕh dσ | ≤ ch 2 |ϕ|2,g |u|2,g .
| ∂n R |
R⊂g ∂ R

Now, let’s look for an estimate of the term

E [ ∂ϕ
. (u − u h ) dσ.
R⊂g ∂ R
∂n R

We cannot calculate this term because we do not know the error .u − u h . So, the idea
is to insert the function .πh u defined by:

πh u(Ai ) = u(Ai ) 1 ≤ i ≤ 4,
.

@seismicisolation
@seismicisolation
5.5 Finite Element of the Serendipity Family 327

πh u(Ai j ) = u(Ai j ) 1 ≤ i < j ≤ 4,

.

and the average of .πh u of the third derivatives on each side is equal to the average
of .u of the third derivatives on each side, i.e.,
[ [
1 ∂ 3 πh u 1 ∂ 3u
. d xd y = d xd y.
mes(S) S ∂x3 mes(S) S ∂x3

We have
[ [ [
∂ϕ ∂ϕ ∂ϕ
. (u − u h ) dσ = (u − πh u) dσ + (πh u − u h ) dσ.
∂R ∂n R ∂R ∂n R ∂R ∂n R

First, let’s calculate [

∂ϕ
. (u − πh u) dσ.
∂R ∂n R
Lemma 5.5.4 We have
| |
| E [ ∂ϕ |
| |
. | (u − πh u) dσ | ≤ ch 3 ||ϕ||2,g |u|4,g . ♦
| ∂ R ∂n R |
R⊂g

Proof We have
|[ | || ||
| ∂ϕ | || ∂ϕ ||
.| | ||
(u − πh u) dσ | ≤ || || ||u − πh u||0,∂ R .
| ∂n R ∂n R ||0,∂ R
∂R

Further, the trace mapping

γ : H 1 (R) −→ L 2 (∂ R, R)
.

is linear continuous. Hence,

|[ | || ||
| ∂ϕ | || ∂ϕ ||
| | ||
(u − πh u) dσ | ≤ c || || ||u − πh u||1,R
.
| ∂n R ||1,R
∂ R ∂n R

≤ c||ϕ||2,R ||u − πh u||1,R .

However, since . P3 ⊂ P, we have

. ||u − πh u||1,R ≤ ch 3 |u|4,R .

Hence, [
∂ϕ
. (u − πh u) dσ ≤ ch 3 ||ϕ||2,R |u|4,R .
∂R ∂n R

@seismicisolation
@seismicisolation
328 5 Non-conforming Methods

Thus, | |
| E [ ∂ϕ |
| |
.| (u − πh u) dσ | ≤ ch 3 ||ϕ||2,g |u|4,g .
| ∂ R ∂n R |
R⊂g

This completes the proof. Q.E.D.

So, we only have the estimate of the term

E[ ∂ϕ
. (πh u − u h ) dσ.
R⊂g ∂ R
∂n R

Lemma 5.5.5 We have

| |
| E [ ∂ϕ |
| |
. | (πh u − u h ) dσ | ≤ ch 2 (|u|2,g + |u|4,g )|ϕ|2,g . ♦
| ∂n R |
R⊂g ∂ R

2 1

-
S 1 S

3 4

Proof We pose .wh = πh u − u h ∈ Vh .

For [ [
∂ϕ 1 ∂ϕ 1
. wh ds − wh ds
S ∂n R -S ∂n R

and, for .ϕ ∈ P1 , we have

[ [ [
/\y 1
. wh1 dy − wh1 dy = (-
wh (1, η) − -
wh (−1, η)) dη.
S -
S 2 −1

@seismicisolation
@seismicisolation
5.5 Finite Element of the Serendipity Family 329

Furthermore,
ξ η(1 + ξ )(1 + η) ξ η(1 − ξ )(1 + η) ξ η(1 − ξ )(1 − η)
.-
wh (ξ, η) = -
w1 − -
w2 + -
w3
4 4 4
ξ η(1 + ξ )(1 − η) (ξ + 1)(η − 1)(1 + η)
. − -w4 − -
w14
4 2
(ξ + 1)(η + 1)(ξ − 1) (ξ − 1)(η − 1)(1 + η)
. − -
w12 + -
w23
2 2
(ξ + 1)(η − 1)(ξ − 1) (ξ − ξ )
3
(η − η)
3
. + -w34 + -
wξ + -
wη .
2 6 6
So,

η(1 + η) η(1 − η) η3 − η
-
wh (1, η) =
. w1 −
- w4 − (η − 1)(η + 1)-
- w14 + -
wη
2 2 6

η(1 + η) η(1 − η) η3 − η
-
wh (−1, η) =
. -
w2 − -
w3 − (η − 1)(η + 1)-
w23 + -
wη .
2 2 6
We have
.-
wh (1, η) − -
wh (−1, η) /= 0.

Hence, we introduce in our method .sh v|R ∈ P3 \{ξ 3 , η3 },

s v=v
. h

at the vertices and in the middle of the sides. We notice that .sh v ∈ C 0 (g, R). We
E [ ∂ϕ E [ ∂ϕ
have
. wh dσ = (wh − sh wh ) dσ
R⊂g ∂ R
∂n R R⊂g ∂ R
∂n R

since .sh wh ∈ C 0 (g, R) and vanishes two by two. We will calculate the term
[ [
∂ϕ 1 ∂ϕ 1
. (w − sh wh1 ) dy − (wh − sh wh1 ) dy.
S ∂n R h S ∂n R
-

For .ϕ ∈ P1 , we will calculate the term

[ [
. (wh1 − sh wh1 ) dy − (wh1 − sh wh1 ) dy.
S -
S

Further,

η(1 + η) η(1 − η)
-wh (1, η) =
s-
. -
w1 − -
w4 − (η − 1)(η + 1)-
w14
2 2
η(1 + η) η(1 − η)
-wh (−1, η) =
s-
. -
w2 − -
w3 − (η − 1)(η + 1)-
w23 .
2 2

@seismicisolation
@seismicisolation
330 5 Non-conforming Methods

Since .mes(S) = mes(- S), we have

[ [
. (wh1 − sh wh1 ) dy − (wh1 − sh wh1 ) dy = 0.
S -
S

Hence, we arrive at the polynomial invariance. Thus,

[ [
∂ϕ ∂ϕ
. (wh1 − sh wh1 ) d y − (wh1 − sh wh1 ) d y =
∂n R [[ 1 - ∂n ]
S S R [ 1
/\y 2 ∂-ϕ ∂-
ϕ
. wh −-
(1, η)(- s-wh )(1, η) dη − wh −-
(−1, η)(- wh )(−1, η) dη .
s-
2 /\x −1 ∂ξ −1 ∂ξ

We pose
[ [
1 ∂-
ϕ 1 ∂-
ϕ
. H (-
ϕ, -
wh ) = wh −-
(1, η)(- wh )(1, η) dη −
s- wh −-
(−1, η)(- wh )(−1, η) dη.
s-
−1 ∂ξ −1 ∂ξ

The mapping who has

-
ϕ −→ H (-
. ϕ, -
wh )

- into .R of norm .≤ c|-

is linear and continuous from . H 2 ( R) wh |1, R- and identically null
for all .-
ϕ ∈ P1 . Then, according to the Theorem 3.4.10, we have

|H (-
. ϕ, -
wh )| ≤ c|-
ϕ |2, R-|-
wh |1, R-
≤ ch|ϕ|2,R |wh |1,R
≤ ch|ϕ|2,R |πh u − u h |1,R .

Furthermore,
|πh u − u h |1,R ≤ ch|u|2,R + ch 3 |u|4,R .
.

Hence,
|H (-
. ϕ, -
wh )| ≤ ch 2 |u|2,R |ϕ|2,R .

So, | |
| E [ ∂ϕ |
| |
.| (πh u − u h ) dσ | ≤ ch 2 (|u|2,g + |u|4,g )|ϕ|2,g .
| ∂ R ∂n R |
R⊂g

This completes the proof. Q.E.D.

Theorem 5.5.6 We have

. ||u − u h ||0,g ≤ ch 2 (|u|2,g + |u|4,g + h|u|4,g ). ♦

Proof By using Lemmas 5.5.3, 5.5.5 and 5.5.4, we have

|(u − u h , g)0,g | ≤ ch 2 |u|2,g |ϕ|2,g + ch 2 (|u|2,g + |u|4,g )|ϕ|2,g + ch 3 |ϕ|2,g |u|4,g .

.

@seismicisolation
@seismicisolation
5.5 Finite Element of the Serendipity Family 331

This proves

|(u − u h , g)0,g | ≤ ch 2 |u|2,g ||g||0,g + ch 2 ||g||0,g |u|4,g + ch 3 ||g||0,g |u|4,g .

.

Thus, for .g such that .||g||0,g /= 0, we have

|(u − u h , g)0,g |
. ≤ ch 2 (|u|2,g + |u|4,g + h|u|4,g ).
||g||0,g

By switching to sup, we obtain

||u − u h ||0,g ≤ ch 2 (|u|2,g + |u|4,g + h|u|4,g ).

.

This completes the proof. Q.E.D.

We still have to show the following:

Lemma 5.5.7 We have

. |H (-
ϕ, -
wh )| ≤ c||-
ϕ ||2, R-|-
wh |1, R-. ♦

Proof We have

η(1 + η) η(1 − η) η3 − η
. -
wh (1, η) = w1 −
- w4 − (η − 1)(η + 1)-
- w14 + -
wη
2 2 6
η(1 + η) η(1 − η)
-wh (1, η) =
s-
. -
w1 − -
w4 − (η − 1)(η + 1)-
w14
2 2

η(1 + η) η(1 − η) η3 − η
-
wh (−1, η) =
. -
w2 − -
w3 − (η − 1)(η + 1)-
w23 + -
wη
2 2 6
η(1 + η) η(1 − η)
-wh (−1, η) =
s-
. -
w2 − -
w3 − (η − 1)(η + 1)-
w23 .
2 2
Hence,
|| ||
|| ∂-
ϕ ||
.|H (- wh )| ≤ || (1, η)||
ϕ, - ||
|| ||(-
wh −- wh )(1, η)||0,(−1,1)
s-
∂ξ 0,(−1,1) || ||
|| ∂-
ϕ ||
. + ||
|| ∂ξ (−1, η)||
|| ||(-
wh −-wh )(−1, η)||0,(−1,1) .
s-
0,(−1,1)

Further, the mapping

- −→ L 2 (∂ R,
γ : H 1 ( R)
. - R)

@seismicisolation
@seismicisolation
332 5 Non-conforming Methods

is linear and continuous, so

|| ||
|| ∂-
ϕ ||
.|H (- wh )| ≤ c ||
ϕ, - || ∂ξ (1, η) || ||(-
|| - wh −- wh )(1, η)||1, R-
s-
1, R
|| ||
|| ∂-
ϕ ||
+ c || (−1, η)||
||
|| - ||(- wh −- s-wh )(−1, η)||1, R-
∂ξ 1, R
|| 3 ||
|| η − η ||
≤ 2c||ϕ||2, R- ||
|| 6 || - |-
|| wη |.
1, R

Let us show that

||(-
. wh −-wh )(1, η)||1, R- ≤ c|-
s- wh |1, R-.

In fact,
|| 3 ||
|| η − η ||
||(-
. wh −-wh )(1, η)||1, R- = ||
s- ||
|| 6 || - |- wη |
1, R
wη |.
= c|-

Let’s calculate .|-

wh |1, R-. In fact,
ξ η(1 + ξ )(1 + η) ξ η(1 − ξ )(1 + η) ξ η(1 − ξ )(1 − η)
-
wh (ξ, η) =
. w1 −
- w2 +
- -
w3
4 4 4
ξ η(1 + ξ )(1 − η) (ξ + 1)(η − 1)(1 + η)
. − -w4 − -
w14
4 2
(ξ + 1)(η + 1)(ξ − 1) (ξ − 1)(η − 1)(1 + η)
. − -
w12 + -
w23
2 2
(ξ + 1)(η − 1)(ξ − 1) (ξ − ξ )
3
(η − η)
3
. + -w34 + -
wξ + -
wη .
2 6 6
We calculate
([ [ ( )2 [ [ ( )2 ) 21
1 1
∂-
wh 1 1
∂-
wh
|-
.wh |1, R- = dξ dη + dξ dη .
−1 −1 ∂ξ −1 −1 ∂η

Then, we will find .|-

wh |1, R- in terms of .-
wη , .-
w1 , etc. Hence, we can increase

||(-
. wh −-wh )(1, η)||1, R- ≤ c|-
s- wh |1, R-.

@seismicisolation
@seismicisolation
Reference 333

Thus,
|H (-
. ϕ, -
wh )| ≤ c||-
ϕ ||2, R-|-
wh |1, R-.

This completes the proof. Q.E.D.

Reference

1. P. Lesaint, M. Zlamal, Convergence de l’élément de Wilson en élasticité linéaire (Cas des quadri-
latères queçonques, Journés éléments finis, Rennes, 1977)

@seismicisolation
@seismicisolation
Chapter 6
Nodal Methods

Modern nodal methods were developed in the 1970s in the context of reactor cal-
culations to solve the diffusion and transport problems of water reactors, both in
stationary and transient states. The reader may refer to [1] or [2] for a description
of the different types of nodal methods. The basic principle of the nodal methods
consists in breaking down the core of the reactor into cells of relatively large sizes
and of constant composition, where the fluxes are treated in terms of average. These
methods are based on discrete equations obtained by integration on the cells of the
physical balance equations. These balance equations can be completed by using a
transverse integration procedure (partial integration along the transverse direction)
which generates auxiliary equations linking the moments of the flux and those of
the current. The final equations of this formalism are, therefore, derived from the
physical principles underlying the problem to be solved. There are several types of
nodal methods:
1. Polynomial nodal methods: The flow is approximated, in each element . R, by a
polynomial.
2. Analytical nodal methods: These methods use the equations obtained by trans-
verse integrations and approach the second member of these equations by a
polynomial. The resolution is, then, done analytically.
Let .Ω be an open bounded domain of .R2 with piecewise smooth boundary .Γ = ∂Ω.
Given a function . f ∈ L 2 (Ω, R), we search a function .u defined on .Ω checking

. − D/\u + σ u = f in Ω
u = 0 on Γ = ∂Ω

with . D > 0 and .σ > 0.

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 335
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_6
@seismicisolation
@seismicisolation
336 6 Nodal Methods

6.1 Nodal Method for Squares or Rectangles

The nodal methods are interpreted in most cases as non-conforming finite ele-
ment type methods using approximate quadrature formulas. Nodal methods have the
advantage of remaining very close to the physical phenomena they seek to model.
They consist of keeping average or weighted neutron balances for each cell . R. In this
sense, they are similar to methods of the weighted residual type. On the other hand,
these balance equations are supplemented by equations of continuity on average of
the flow on each interface. Nodal methods, therefore, preserve the characteristics
of the neutron systems studied, which makes them attractive to physicists. We are
going to place ourselves within the framework of the study of the polynomial nodal
method.

6.1.1 Position of the Problem

In nuclear physics, we consider the problem

. − div(D grad u) + σ u = f in Ω
D and σ = cte in relation to each piece
u = 0 on Γ = ∂Ω,

with . D : diffusion, .σ : effective, . f : source, .u : neutron flux. The respected balance

sheet is {
. (−D/\u + σ u − f ) d xd y = 0 for all R ⊂ Ω.
R

battalion

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 337

6.1.2 Reference Element

-
We build the reference element . R
η

1 2

-
R

3 1
−1 1 ξ

−1 4

Degree of freedom: the means at the interfaces and the mean over . R. - So, dimension
of the degree of freedom equal to 5. However, we have .dim P1 = 3, .dim P2 = 6 and
- taking .5 basic elements
.dim Q 1 = 4. Notice that .5 is close to .6. So, we construct . P,

of . P2 . We take for example

. - = {1, ξ, η, ξ 2 , η2 }.
P

axis of symmetry

We did not choose .ξ η because .ξ , .η plays the same role and, we have a symmetry of
our domain.

@seismicisolation
@seismicisolation
338 6 Nodal Methods

6.1.3 Basis Functions

- Hence,
Let’s seek the elements of basis. We have .ϕ1 ∈ P.

. 1ϕ (ξ, η) = a + bξ + cη + dξ 2 + eη2 .

We have the average of .ϕ1 on the side .1 is equal to 1 and zero on the other sides and
the average over the domain is zero.
On the side .2, .η = 1 and

ϕ
. 1 |side 2 = a + bξ + c + dξ 2 + e.

Hence, the mean value is

{ 1
1
. (a + bξ + c + dξ 2 + e) dξ = 0.
2 −1

On the side .4, .η = −1 and

ϕ
. 1 |side 4 = a + bξ − c + dξ 2 + e.

Hence, the mean value is

{ 1
1
. (a + bξ − c + dξ 2 + e) dξ = 0.
2 −1

On the side .1, .ξ = 1 and

ϕ
. 1 |side 1 = a + b + cη + d + eη2 .

Hence, the mean value is

{ 1
1
. (a + b + cη + d + eη2 ) dη = 1.
2 −1

On the side .3, .ξ = −1 and

ϕ
. 1 |side 3 = a − b + cη + d + eη2 .

Hence, the mean value is

{ 1
1
. (a − b + cη + d + eη2 ) dη = 0.
2 −1

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 339

- is
On the other hand, the mean value on the domain . R
{ 1 { 1
1
. (a + bξ + cη + dξ 2 + eη2 ) dξ dη = 0.
4 −1 −1

These equations are equivalent to the following system

⎧
⎪ d
⎪
⎪ 2(a + c + + e) = 0
⎪
⎪ 3
⎪
⎪
⎪
⎪ d
⎪
⎪ 2(a − c + + e) = 0
⎪
⎪ 3
⎨
e
. a−b+d + =0
⎪
⎪ 3
⎪
⎪
⎪
⎪ d e
⎪
⎪ a+ + =0
⎪
⎪ 3 3
⎪
⎪
⎪
⎩ a + b + d + e = 1.
3
The solution of this system is given by
⎧
⎪c = 0
⎪
⎪
⎪ e=0
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎨b =
. 2
⎪
⎪ 3
⎪
⎪ d=
⎪
⎪ 4
⎪
⎪
⎪
⎪
⎩a = −1.
4
So, the first function of basis is

1 1 3
. 1 ϕ (ξ, η) = − + ξ + ξ 2 .
4 2 4
Likewise, we find others basis functions. In fact,

1 1 3
ϕ (ξ, η) = − + η + η2
. 2
4 2 4
1 1 3
ϕ (ξ, η) = − − ξ + ξ 2
. 3
4 2 4
1 1 3
. 4 ϕ (ξ, η) = − − η + η2 .
4 2 4

ϕ (ξ, η) depend of .1, .ξ 2 and .η2 and does not depend of .ξ and of .η because the
. 0

averages of .ξ and of .η are zero. Hence,

@seismicisolation
@seismicisolation
340 6 Nodal Methods

ϕ (ξ, η) = a + b(ξ 2 + η2 ).
. 0

For .ξ = 1, we have { 1
1
. (a + b(1 + η2 )) dη = 0.
2 −1

Which proves .a + b + b
3
= 0. Hence,

4
.a = − b.
3
Furthermore, { {
1 1
1
. a + b(ξ 2 + η2 ) dξ dη = 1.
4 −1 −1

Hence, .a + 23 b = 1. So, .a = 2 and .b = − 23 . Thus,

3
ϕ (ξ, η) = 2 − (ξ 2 + η2 ).
. 0
2
We have the following lemma:
Lemma 6.1.1 There is a polynomial . - p∈P - unique taking values given to the means
-
of the interfaces and to the mean of . R, given by
( ) ( ) ( )
1 1 3 1 1 3 1 1 3
.-
p (ξ, η) = − + ξ + ξ 2 q1 + − + η + η 2 q2 + − − ξ + ξ 2 q3
4 2 4 4 2 4 4 2 4
( ) ( )
1 1 3 2 3 2
+ − − η + η q4 + 2 − (ξ + η2 ) q0 ,
4 2 4 2

with .qi .1 ≤ i ≤ 4 mean interfaces of . -

p and .q0 average of . - - i.e.,
p on . R,
{
1
q =
. 0 -
p (ξ, η) dξ dη.
-
mes( R) -
R
♦

6.1.4 Description of the Nodal Method

We define the Nodal method as follows: either a triangulation .τh of .Ω by subdividing

Ω into rectangles .(Ri )i=1,...,m
.

| | | |
m
. Ω= Ri = Ri .
Ri ∈τh i=1

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 341

We define the approximation space

. Vh = {vh ∈ L 2 (Ω, R) such that vh |R is defined by the mean values at the interfaces
on each rectangle R on each rectangle R and mean on R}.

The equation bilan is written

{
. (−D/\u h + σ u h − f ) d xd y = 0 for all R ⊂ Ω.
R

We have the continuity of the average values since

( )
1 3 2
. -
p (ξ, η)|side 1 = q1 + q2 ϕ2 |side 1 + q4 ϕ4 |side 1 + q0 − η
2 2

and { 1
. -
p (ξ, η)|side 1 dη = q1 .
−1

We assume the following hypothesis

{ ( )
∂u 1 ∂u 2
. D1 h + D2 h ds = 0
S ∂n 1 ∂n 2

for all . S side of .Ω and . S /⊂ ∂Ω. The average of .u h is equal to zero on a side . S ⊂ Ω,
i.e., {
1
. u h = 0.
mes(S) S

Vh
J

@seismicisolation
@seismicisolation
342 6 Nodal Methods

The total number of sides is .(I + 1)J + (J + 1)I − 2I − 2J included in .Ω.

Hence, .dim Vh is equal to number of sides included in .Ω + number of squares. Thus,

. dim Vh = (I + 1)J + (J + 1)I − 2I − 2J + I J = 3I J − I − J.

6.1.5 Relationship Between Nodal Method

and Non-conforming Method

Theorem 6.1.2 Either the non-conforming method: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh , where

E{
a (u h , vh ) =
. h D∇u h · ∇vh + σ u h vh d xd y
R⊂Ω R

E{
and
. L(vh ) = f vh d xd y
R⊂Ω R

with . D and .σ are equal to constants with respect to each rectangle . R, .vh |R is the
average of .vh on . R, i.e., {
1
.vh |R = vh d xd y
mes(R) R

and . f |R is the average of . f on . R, i.e.,

{
1
. f |R = f d xdy.
mes(R) R

Then, the Nodal method is identical to the non-conforming method. ♦

Proof Non-conforming method: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh with

E{
a (u h , vh ) =
. h D∇u h · ∇vh + σ u h vh d xd y
R⊂Ω R

and

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 343
{
. L(vh ) = f vh d xd y.
Ω

Nodal method: The .u h checks

{
. (−D/\u h + σ u h − f ) d xd y = 0 for all R ⊂ Ω (6.1.1)
R

and { ( )
∂u 1h ∂u 2
. D1 + D2 h ds = 0 (6.1.2)
S ∂n 1 ∂n 2

∂u h
for all . S ⊂ Ω and . S /⊂ ∂Ω. Note that if . ∂n R |S
= cte, then

∂u 1h ∂u 2
. D1 = −D2 h ,
∂n 1 ∂n 2

i.e., | | | |
| ∂u 1h | | 2|
| D1 | = |−D2 ∂u h | .
.
| ∂n | | ∂n |
1 2

Moreover, we have continuity of the means of .u h along the sides. On . R the basic
functions are noted .ϕ0 , .ϕ1 , .ϕ2 , .ϕ3 , and .ϕ4 .
ϕ2 B

ϕ3 R ϕ1
ϕ0

ϕ4 A

To show that the Nodal method is identical to the non-conforming method, it suffices
to show that the associated linear systems are the same, i.e., we show that the linear
system associated with the Nodal method is identical with the linear system associated
with the non-conforming method. We have

@seismicisolation
@seismicisolation
344 6 Nodal Methods
{ {
. −D/\u h d xd y = −D /\u h d xd y (u h ∈ Vh so /\u h = cte)
R R
= −D/\u h mes(R)
{
= −D/\u h ϕ0 (x, y) d xd y
{ R

= −D/\u h ϕ0 (x, y) d xd y
{ R
{
∂u h
= D∇u h ∇ϕ0 d xd y − D ϕ0 ds.
R ∂R ∂n R
∂u h
Further, .u h ∈ Vh , so . ∂n R
= cte on each side. Hence,
{ {
∂u h ∂u h
. D ϕ0 ds = D ϕ0 ds = 0
∂R ∂n R ∂n R ∂R

because the average of .ϕ0 on each side is zero. Therefore,

{ {
. − D/\u h d xd y = D∇u h ∇ϕ0 d xd y.
R R

On the other hand,

{ {
. σ u h d xd y = σ u h d xd y
R
{R
=σ u h d xd y
R
= σ u h mes(R)
{
= σ uh ϕ0 d xd y
{ R

= σ u h ϕ0 d xd y
R

and
{ {
. f d xdy = f d xdy
R R
= f mes(R)
{
= f ϕ0 d xd y
{ R
= f ϕ0 d xd y.
R

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 345

Hence, the Eq. (6.1.1) is equivalent to

{
. (D∇u h ∇ϕ0 + σ u h ϕ0 − f ϕ0 ) d xd y = 0 for all R ⊂ Ω.
R

By summing on all rectangles . R, we obtain

{
. h a (u h , G0 ) = f G0 d xd y
Ω

with .G0 ∈ Vh and .G0|R = ϕ0 . Consider .G AB ∈ Vh such that the average of .G AB

along . AB is equal to 1 and zero along the others sides.

C B F

R1 R2

D A E

On the side . AB, the Eq. (6.1.2) gives

{ {
B
∂u 1h B
∂u 2h
. D dy = D dy.
A ∂x A ∂x

∂u 1
Furthermore, . ∂ xh = cte on . AB, so
{ B
∂u 1h ∂u 1
. D dy = D h × length(AB)
A ∂x ∂ x |AB
{ B
∂u 1
=D h G AB dy
∂ x |AB A

because the average of .G AB is equal to 1. Hence,

{ {
B
∂u 1h ∂u 1h B
. D dy = G AB dy
D
A ∂x A ∂ x |AB
{ {
∂u 1
= D h G AB ds (car G AB = 0)
∂ R1 ∂x ∂ R1 \ AB
{ {
= D∇u 1h ∇G AB d xd y + D/\u 1h G AB d xd y.
R1 R1

@seismicisolation
@seismicisolation
346 6 Nodal Methods

However, ./\u 1h = cte, so

{ {
. D/\u 1h G AB d xd y = D/\u 1h G AB d xd y = 0
R1 R1

because the average of .G AB is equal to zero except on the side . AB is equal to .1.
Hence,
{ {
B
∂u 1h
. D dy = D∇u 1h ∇G AB d xd y
A ∂x R1
{
=− D∇u 2h ∇G AB d xd y.
R2

So, the Eq. (6.1.2) is equivalent to

{ {
. D∇u 1h ∇G AB dxdy + D∇u 2h ∇G AB d xd y = 0.
R1 R2

Further, {
. σ u h 1 G AB d xd y = 0
R1

and {
. f G AB d xd y = 0.
R1

Hence, the Eq. (6.1.2) is equivalent to

{ {
. (D∇u 1h ∇G AB + σ u h G AB − f G AB ) d xd y+
1
(D∇u 2h ∇G AB + σ u h 2 G AB
R1 R2

− f G AB ) d xd y = 0.

Let . R /= R1 and . R2 , then

{
. (D∇u hR ∇G AB + σ u h R G AB − f G AB ) d xd y = 0
R
U
because .G AB = 0 for all . R /= R1 and . R2 because .supp(G AB ) = R1 R2 . Hence,
E{
. (D∇u hR ∇G AB + σ u h R G AB − f G AB ) d xd y = 0.
R⊂Ω R

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 347

Finally, we have the following two equations:

E{
. (D∇u h ∇ϕ0R + σ u h ϕ0R − f ϕ0R ) d xd y = 0
R⊂Ω R

E{
and
. (D∇u h ∇G AB + σ u h G AB − f G AB ) d xd y = 0.
R⊂Ω R

1 ,R2
Let .vh ∈ Vh , .vh be linear combination of .ϕ0R and .G RAB , we have
E{
. (D∇u h · ∇vh + σ u h vh − f vh ) d xd y = 0.
R⊂Ω R

Further, for all . R ⊂ Ω, we have

{ {
. σ u h vh d xd y = σ u h vh d xd y.
R R

E{
Thus,
. (D∇u h · ∇vh + σ u h vh − f vh ) d xd y = 0
R⊂Ω R

if, and only if,

a (u h , vh ) = L(vh ) for all vh ∈ Vh .
. h

Therefore, one ends up with the same linear system and thus the nodal method is
identical to the non-conforming method. Q.E.D.

6.1.6 Existence and Uniqueness of Solution

Lemma 6.1.3 The mapping

1
v −→ ||vh ||h := (ah (vh , vh )) 2
. h

is a norm on .Vh . ♦

Proof If .ah (vh , vh ) = 0, then

. D||∇vh ||0,R = 0

for all . R ⊂ Ω. Hence, .vh |R = cte. Further, .vh |R = 0, so .cte = 0. Thus,

v = 0.
. h Q.E.D.

@seismicisolation
@seismicisolation
348 6 Nodal Methods

Theorem 6.1.4 The problem: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh admits a unique solution .u h ∈ Vh . ♦

6.1.7 Error Estimates for the Non-conforming Method

Either the non-conforming method: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h (6.1.3)

for all .vh ∈ Vh , where

E{
a (u h , vh ) =
. h D∇u h · ∇vh + σ u h vh d xd y
R⊂Ω R

E{
and
. L(vh ) = f vh d xd y.
R⊂Ω R

From Theorem 6.1.4, the problem (6.1.3) admits a unique solution.u h ∈ Vh . We have

||u h − vh ||2h = ah (u h − vh , u h − vh )
.

= ah (u h , u h − vh ) − ah (vh , u h − vh )
= L(u h − vh ) − ah (vh , u h − vh )
{
= f (u h − vh ) d xd y − ah (vh , u h − vh ).
Ω

Furthermore, .u h |R − vh |R = cte by element . R because

{
1
u
. h |R − vh |R = (u h − vh ) d xd y
mes(R) R

and { {
. f d xdy = f d xdy
Ω Ω

because {
1
. f |R = f d xdy.
mes(R) R

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 349

On the other hand,

{
||u h − vh ||2h =
. f (u h − vh ) d xd y − ah (vh , u h − vh )
{Ω {
= f [(u h − vh ) − (u h − vh )]d xd y + f (u h − vh ) d xd y
Ω Ω
−ah (vh , u h − vh ).

We pose .z h = u h − vh . We have
{ {
.||u h − vh ||2h = f (z h − z h ) d xd y + (−div(D grad u) + σ u)z h d xd y
Ω Ω
−a (v , u − vh )
{ h h h E { E {
= f (z h − z h ) d xd y + D∇u · ∇z h d xd y + σ uz h d xd y
Ω R⊂Ω R R⊂Ω R
E { ∂u
− z h ds − ah (vh , z h )
D
∂n
R⊂Ω ∂ R R
{ {
= f (z h − z h ) d xd y + (σ uz h − σ u z h ) d xd y + ah (u − vh , z h )
Ω Ω
E {
∂u
− D z h ds.
∂ R ∂n R
R⊂Ω

We take .vh = rh u ∈ Vh with the average of .rh u is equal to the average of .u along the
sides and on the elements. We will start to mark up the term
E{
. D∇(u − rh u) · ∇z h + σ (u − rh u)z h d xd y.
R⊂Ω R

We have {
. D∇(u − rh u) · ∇z h d xd y ≤ D|u − rh u|1,R |z h |1,R .
R

On the other hand, . P1 ⊂ Pk , so

|u − rh u|1,R ≤ ch|u|2,R .
.

Thus, {
. D∇(u − rh u) · ∇z h d xd y ≤ ch|u|2,R D|z h |1,R .
R

@seismicisolation
@seismicisolation
350 6 Nodal Methods

Since .z h = cte on each element, we have

{ {
. σ (u − rh u)z h d xd y = σ (u − rh u)z h d xd y
R R
≤ σ ||u − rh u||0,R ||z h ||0,R
≤ ch 2 |u|2,R σ ||z h ||0,R .

Hence,
( ) 21 ( ) 21
E E
. h a (u − rh u, z h ) ≤ ch|u|2,Ω D 2 |z h |21,R + ch 2 |u|2,Ω σ 2 ||z h ||20,R .
R⊂Ω R⊂Ω

By applying the formula (2.5.2) with the chosen .δ, we find

1
|ah (u − rr u, z h )| ≤ c(h 2 |u|22,Ω + h 4 |u|22,Ω ) + ||z h ||2h .
.
2
We used the Cauchy-Schwarz inequality to find the error, but we can do better. Indeed,
by applying Green’s formula, we get
{ { {
∂z h
. D∇(u − rh u) · ∇z h d xd y = −D(u − rh u)/\z y d xd y + D (u − rh u) dσ.
R R ∂R ∂n R

Further, .z h ∈ Vh , so .z h ∈ Pk ⊂ P2 . Thus, ./\z h = cte. It follows that

{ {
. −D(u − rh u)/\z y d xd y = −D/\z y (u − rh u) d xd y = 0
R R

because the average of .u is equal to the average of .rh u on each element. On the other
hand, .z h ∈ Vh , so

E
4
-
z =
. h αi ϕi
i=0
= a + bξ + cη + dξ 2 + eη2

and
∂-
zh ∂-
zh ∂-
zh
. = ν1 + ν2 .
∂n R ∂ξ ∂η

On the other hand,

∂-
zh
. (1, η) = cte in η
∂ξ

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 351

∂-
zh
. (−1, η) = cte in η
∂ξ

∂-
zh
. (ξ, 1) = cte in ξ
∂η

and
∂-
zh
. (ξ, −1) = cte in ξ.
∂η

Hence,
∂-
zh
. = cte by side.
∂n R

Since average of .u is equal to the average of .rh u on each side, we have

{ {
∂z h
. D(u − rh u) dσ = cte (u − rh u) dσ = 0.
∂R ∂n R ∂R

Remark 6.1.5 Normal derivatives are constants on each side, i.e., if .vh ∈ Vh , then

∂vh
. = cte
∂n R

on each side. ♦

E{
Thus,
a (u − rh u, z h ) =
. h σ (u − rh u) z h d xd y.
R⊂Ω R

It follows that
|ah (u − rh u, z h )| ≤ ch 2 |u|2,Ω ||z h ||0,Ω .
.

We are now interested in the term

{
. f (z h − z h ) d xd y.
Ω

Lemma 6.1.6 We have

|{ | ( ) 21
| | E
| f (z h − z h ) d xd y || ≤ ch 2 || f ||1,Ω |z h |21,R .
.
|
Ω R⊂Ω

@seismicisolation
@seismicisolation
352 6 Nodal Methods

Proof We have
{ {
. f (z h − z h ) d xd y = ( f − f )(z h − z h ) d xd y
R R

because . f |R = cte and {

. (z h − z h ) d xd y = 0
R

because the average of .z h is equal to the average of .z h . Hence,

|{ |
| |
| f (z h − z h ) d xd y | ≤ || f − f ||0,R ||z h − z h ||0,R .
.
| |
R

Further,
.|| f − f ||0,R = h|| -
f − -
f ||0, R-.

The mapping .π : -
f −→ - - into . L 2 ( R,
f is linear and continuous from . H 1 ( R) - R) and

π-
.f = -
f = -
f

for all . -
f ∈ P0 . Then,
|| -
. f ||0, R- ≤ c|| -
f − - f ||1, R-.

Hence,
|| -
. f − -
f ||0, R- ≤ c|| f ||1,R .

Thus,
. || f − f ||0,R ≤ ch|| f ||1,R .

On the other hand,

||z h − z h ||0,R ≤ ch|z h |1,R .
.

Hence, {
. f (z h − z h ) d xd y ≤ ch 2 || f ||1,R |z h |1,R .
R

So,
|{ | ( ) 21
| | E
| f (z h − z h ) d xd y || ≤ ch 2 || f ||1,Ω |z h |21,R .
.
|
Ω R⊂Ω

This completes the proof. Q.E.D.

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 353

We still have to show the continuity of the linear mapping

π: -
. f −→ -
f.

Lemma 6.1.7 The mapping .π : - f −→ - - into

f is linear and continuous from . H 1 ( R)
- R) (we even have continuity from . L 2 ( R,
L 2 ( R,
. - R) into . L 2 ( R,
- R)). ♦

Proof We have
{ {
. (-
f )2 dξ dη = -
f-f dξ dη
- -
R R
{ {
1 1
≤ (-
f )2 dξ dη + (-
f )2 dξ dη.
2 -
R 2 -
R

Hence,
{ {
. (-
f )2 dξ dη ≤ (-
f )2 dξ dη
-
R -
R
≤ || -
f ||21, R-.

Thus,
. π: -
f −→ -
f

- into . L 2 ( R,
is linear and continuous from . H 1 ( R) - R) (we even have continuity from
2 - 2 -
. L ( R, R) into . L ( R, R)). Q.E.D.

Now, we are going to look for an estimate of the term

{
. (σ uz h − σ u z h ) d xd y.
Ω

Lemma 6.1.8 We have

|{ | ( ) 21
| | E
.| |
| (σ uz h − σ u z h ) d xd y | ≤ ch |u|1,Ω |z h |21,R .
2
Ω R⊂Ω

Proof We have
{ {
. (uz h − u z h ) d xd y = (u − u)(z h − z h ) d xd y
R R

@seismicisolation
@seismicisolation
354 6 Nodal Methods

because .(u − u)(z h − z h ) = uz h − uz h − uz h + u z h and

{ {
. −uz h + u z h d xd y = z h (u − u) d xd y
R R
{
= cte (u − u) d xd y
R
= 0.

Hence,
{
. (σ uz h − σ u z h ) d xd y ≤ ||u − u||0,R ||z h − z h ||0,R
R
≤ ch 2 |u|1,R |z h |1,R .

Thus,
|{ | ( ) 21
| | E
| |
| (σ uz h − σ u z h ) d xd y | ≤ ch |u|1,Ω |z h |1,R .
2 2
.
Ω R⊂Ω

This completes the proof. Q.E.D.

Let’s seek an estimate of the term

E{ ∂u
. D z h ds.
R⊂Ω ∂R ∂n R

Lemma 6.1.9 We have

| | ( ) 21
|E { ∂u | E
| |
.| D z h ds | ≤ ch 2 |u|3,Ω |z h |21,R .
| ∂ R ∂n R |
R⊂Ω R⊂Ω

Proof On each . R, we define .π∂ R z h on .∂ R by

π z
. ∂ R h |S = average de z h on S,

where . S is beside. Note that

E{ ∂u
. D π∂ R z h ds = 0.
R⊂Ω ∂R ∂n R

Indeed, we have by hypothesis on a side . S, conservation of the flow, that is to say,

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 355
{ ( )
∂u ∂u
. D1 + D2 ds = 0
S ∂n 1 ∂n 2

and, we have the continuity of .π∂ R z h , i.e.,

π z
. ∂ R1 h |S = π∂ R2 z h |S

because .π∂ R z h |S is equal to the average of .z h on . S and, according to the definition of

the space .Vh , we have the continuity of the means on each side. Hence, the continuity
of .π∂ R z h . Thus,
{ { {
∂u ∂u 1 ∂u 2
. D π∂ R z h ds = D1 π∂ R1 z h ds + D2 π∂ R2 z h ds
S ∂n R S ∂n 1 S ∂n 2
{ ( )
∂u 1 ∂u 2
= D1 + D2 ds π∂ R1 z h
S ∂n 1 ∂n 2
=0

R1
S R2

D1 D2

because .π∂ R1 z h = cte is continuous and, we have conservation of the flow. Thus,
E{ ∂u E{ ∂u
. D z h ds = D (z h − π∂ R z h ) ds.
R⊂Ω ∂R ∂n R R⊂Ω ∂ R
∂n R

C B

D A

@seismicisolation
@seismicisolation
356 6 Nodal Methods

Further,
{ {
∂u ∂u
. D (z h − π AB z h ) dy − D (z h − πC D z h ) dy =
AB ∂ x CD ∂ x
[{ { 1 ]
/\y 1 ∂-u ∂-
u
D (1, η)(-zh − -
π-
z h )(1, η) dη − (−1, η)(-
zh − -
π-
z h )(−1, η) dη
/\x −1 ∂ξ −1 ∂ξ

with .-
π- - and moreover, the mean of .-
z h is defined on .∂ R z h is equal to the mean of .-
π-
zh
on each side. We pose
{ 1 { 1
∂-
u ∂-
u
u ,-
.G(- zh ) = (1, η)(-
zh − -
π-
z h )(1, η) dη − (−1, η)(-
zh − -
π-
z h )(−1, η) dη.
−1 ∂ξ −1 ∂ξ

Method 1: We have

-
z (ξ, η) = α0 ϕ0 + α1 ϕ1 + α2 ϕ2 + α3 ϕ3 + α4 ϕ4 .
. h

Hence,

-
z (1, η) = α0 ϕ0 (1, η) + α1 ϕ1 (1, η) + α2 ϕ2 (1, η) + α3 ϕ3 (1, η) + α4 ϕ4 (1, η)
. h
( ) ( )
α0 α2 α4 α2 − α4 3 3 3
= + α1 − − + η + − α0 + α 2 + α 4 η 2
2 4 4 2 2 4 4

and

.-
z h (−1, η) = α0 ϕ0 (−1, η) + α1 ϕ1 (−1, η) + α2 ϕ2 (−1, η) + α3 ϕ3 (−1, η) + α4 ϕ4 (−1, η)
( ) ( )
α0 α2 α4 α2 − α4 3 3 3
= + α3 − − + η + − α0 + α2 + α4 η 2 .
2 4 4 2 2 4 4

On the other hand,

{ 1
1
-
π-
.z h (1, η) = -
z h (1, η) dη
2 −1
( ) ( ) ( )
3 1 1 1 3 1 1
= 2− − α0 + − + + α1 + − + α2
2 2 4 2 4 4 4
( ) ( )
1 1 3 1 1
+ − − + α3 + − + α4
4 2 4 4 4
= α1 .

Likewise,
.-
π-
z h (−1, η) = α3 .

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 357

Thus,
( ) ( )
α0 α2 α4 α2 − α4 3 3 3
(-
.zh − -
π-
z h )(1, η) = − − + η + − α0 + α 2 + α 4 η 2
2 4 4 2 2 4 4

and
( ) ( )
α0 α2 α4 α2 − α4 3 3 3
.(-
zh −-
π-
z h )(−1, η) = − − + η + − α0 + α2 + α4 η 2 .
2 4 4 2 2 4 4

Method 2: We write .-
z h under the form

-
z (ξ, η) = a + bξ + cη + dξ 2 + eη2 .
. h

Hence,
-
z (1, η) = a + b + cη + d + eη2
. h

and e
-
π-
.z h (1, η) = a + b + d + .
3
Likewise,
-
z (−1, η) = a − b + cη + d + eη2
. h

and e
-
π-
.z h (−1, η) = a − b + d + .
3
Thus,
e
.(-
zh − -
π-
z h )(1, η) = cη + eη2 −
3
and e
.(-
zh − -
π-
z h )(−1, η) = cη + eη2 − .
3
u ,-
It follows that in the two methods .G(- z h ) = 0 for all .-
u ∈ P1 . So, we get the
polynomial invariance.
Question: Can we improve the polynomial invariance of . P1 .
.

u = η2 , then . ∂-
In fact, if .- u
∂ξ
= 0 and hence .G(-
u ,- u = ξ η, then . ∂-
z h ) = 0. If .- u
∂ξ
=η
and so .G(-u ,-z h ) = 0 because
{ ( { 1 (
1
e) e)
. η cη + eη2 − dη − η cη + eη2 − dη = 0.
−1 3 −1 3

@seismicisolation
@seismicisolation
358 6 Nodal Methods

u = ξ 2 , then . ∂-
If .- u
∂ξ
= 2ξ and so .G(-u ,-
z h ) = 0. Hence, we have polynomial invariance
. P2 . The mapping
.-
u −→ G(- u ,-
zh )

- into .R of norm .≤ c|-

is linear and continuous from . H 3 ( R) z h |1, R- and identically null
for all .-
u ∈ P2 . Hence,

|G(-
. u ,-
z h )| ≤ c|-
u |3, R-|-
z h |1, R-
≤ ch 2 |u|3,R |z h |1,R .

Thus,
| | ( ) 21
|E { ∂u | E
| |
.| D z h ds | ≤ ch |u|3,Ω
2
|z h |1,R .
2
| ∂n R |
R⊂Ω ∂ R R⊂Ω

This completes the proof. Q.E.D.

It thus appears that

.||u h − rh u||h ≤ ch 2 (|u|2,Ω + |u|3,Ω + || f ||1,Ω + |u|1,Ω ).

Hence, we still have to show that

|G(-
. u ,-
z h )| ≤ c|-
z h |1, R-||-
u ||3, R-. (6.1.4)

In fact,
|| ||
|| ∂-
u ||
.|G(-
u ,- ||
z h )| ≤ || (1, η)|||| ||(-
zh − -
π- z h )(1, η)||0,(−1,1)
∂ξ 0,(−1,1)
|| ||
|| ∂-
u ||
+ || ||
|| ∂ξ (−1, η)|| ||(-zh − -π-
z h )(−1, η)||0,(−1,1) .
0,(−1,1)

Further, the mapping

- −→ L 2 (∂ R,
γ : H 1 ( R)
. - R)

is linear and continuous. Hence,

|| ||
|| ∂-
u ||
.|G(- z h )| ≤c1 || (1, η)||
u ,- ||
|| - ||(- zh − -
π-z h )(1, η)||1, R-
∂ξ 1, R
|| ||
|| ∂-
u ||
+ c2 || ||
|| ∂ξ (−1, η)|| - ||(- zh − - π-z h )(−1, η)||1, R-.
1, R

@seismicisolation
@seismicisolation
6.1 Nodal Method for Squares or Rectangles 359

So,

|G(-
. u ,-
z h )| ≤ c||-
u ||3, R-(||(-
zh − -
π-
z h )(1, η)||1, R- + ||(-
zh − -
π-
z h )(−1, η)||1, R-).

Furthermore, .(- zh − -π- z h )(1, η) = cη + eη2 − 3e . A simple calculation of the term

.||(-
zh − -
π-
z h )(1, η)||1, R- leads to

||(-
. zh − -
π-
z h )(1, η)||1, R- = α|c| + β|e|.

A simple calculation of the term .|- z h |1, R- with .-

z h (ξ, η) = a + bξ + cη + dξ 2 + eη2
leads to
.|-
z h |1, R- = α1 |a| + α2 |b| + α3 |c| + α4 |d| + α5 |e|.

Hence, there is a constant .c > 0 such that

||(-
. zh − -
π-
z h )(1, η)||1, R- ≤ c|-
z h |1, R-.

In the same way, we can show that

||(-
. zh − -
π-
z h )(−1, η)||1, R- ≤ c|-
z h |1, R-.

Thus, the Eq. (6.1.4). Since .rh u = u for all .u ∈ P1 , we have

||u − rh u||h ≤ ch|u|2,Ω .

.

Thus,
.||u − u h ||h ≤ ||u − rh u||h + ||rh u − u h ||h = o(h).

Hence, the question is: can we do better. In fact,

E{
||u − rh u||2h =
. D|∇(u − rh u)|2 + σ (u − rh u)2 d xd y.
R⊂Ω R

On the other hand,

{ {
. D|∇(u − rh u)|2 d xd y =D −(u − rh u)/\(u − rh u) d xd y
R R
{
∂
+D (u − rh u) (u − rh u) ds.
∂R ∂n R

Further,
{ ( )2 ( )2
∂ ∂
|u
. − rh u|21,R = (u − rh u) + (u − rh u) d xd y.
R ∂x ∂y

@seismicisolation
@seismicisolation
360 6 Nodal Methods

If .u ∈ P1 , then .rh u = u. If .-
u = ξ η, then .-
r- - and .-
u∈P r-
u = 0. In fact,

E
4
-
r-
.u= αi ϕi
i=1

with .αi is equal to the average of .-

r-
u , is equal to the average of .-
u , and is equal to
zero. Thus, .-
r-
u = 0. Furthermore,
{ ( ) { 1
1
u 2
∂-
. dη = η2 dη /= 0.
−1 ∂ξ −1

Hence, we have no polynomial invariance for . P2 . Hence, we can’t do better. But the
important thing is that, we have
( ) 21
E{
. D|∇(u − u h )|2 d xd y ≤ ch(|u|2,Ω + |u|3,Ω + || f ||1,Ω + |u|1,Ω ),
R⊂Ω R

and
( ) 21
E{
. σ (u − u h ) d xd y 2
≤ ch 2 (|u|2,Ω + |u|3,Ω + || f ||1,Ω + |u|1,Ω ).
R⊂Ω R

In fact, { {
. D|∇(u − u h )| d xd y = D 2
|∇(u − u h )|2 d xd y.
R R

Hence,
.|u − u h |1,R ≤ |u − rh u|1,R + |rh u − u h |1,R .

- Hence,
On the other hand, . P1 ⊂ P.

|u − rh u|1,R ≤ ch|u|2,R .
.

So,
. |u − u h |21,R ≤ 2ch 2 |u|22,R + 2|rh u − u h |21,R .

Thus,
( ) 21
E{
. D|∇(u − u h )| d xd y 2
≤ ch(|u|2,Ω + |u|3,Ω + || f ||1,Ω + |u|1,Ω ).
R⊂Ω R

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 361

On the other hand,

{ {
. σ (u − u h )2 d xd y = σ (u − u h )2 d xd y.
R R

Thus,
. ||u − u h ||20,R ≤ 2||u − rh u||20,R + 2||u h − rh u||20,R .

Since . P1 ⊂ Pk ,
. ||u − rh u||0,R ≤ ch 2 |u|2,Ω ,

then
( ) 21
E{
. σ (u − u h )2 d xd y ≤ ch 2 (|u|2,Ω + |u|3,Ω + || f ||1,Ω + |u|1,Ω ).
R⊂Ω R

6.2 Nodal Method on Triangles

6.2.1 Basis Elements

The approximate solution .u h ∈ Vh , which is piecewise polynomial, check that the

mean of .u h and the mean of the currents . D ∂u
∂n
are continuous at interfaces. So, the
balance equation is written
{
. (−D/\u h + σ u h − f ) d xd y = 0 for all T ⊂ Ω.
T

We assume in the following hypothesis

{ ( )
∂u 1h ∂u 2
. D1 + D2 h ds = 0
S ∂n 1 ∂n 2

for all . S side of .Ω and . S /⊂ ∂Ω.

Degree of freedom: average values on the sides and average values on the elements.
So, we have .4 degrees of freedom. For example we will take . P- = Q1.

@seismicisolation
@seismicisolation
362 6 Nodal Methods

-
T

The question that arises is: is this method well defined, i.e., does there exist a unique
polynomial of Q 1 taking mean values given on each side and on the elements. Indeed,
we suppose that this polynomial exists and is unique. Note ϕ0 , ϕ1 , ϕ2 , and ϕ3 the
basis elements.
η

ϕ3 -
T ϕ2
ϕ0

0 ϕ1 1 ξ

ϕ is written
. 1
ϕ (ξ, η) = a + bξ + cη + eξ η.
. 1

For .η = 0, we have
{ 1
1 b
. (a + bξ ) dξ = a + = 1.
1−0 0 2

For .ξ = 0, we have
{ 1
1 c
. (a + cη) dη = a + = 0.
1−0 0 2

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 363

For .ξ + η = 1, we have
{ 1
1 b c e
. (a + bξ + c(1 − ξ ) + eξ(1 − ξ )) dξ = a + + + = 0.
1−0 0 2 2 6

- gives
The average of .ϕ1 on .T
{
a b c e
. (a + bξ + cη + eξ η) dξ dη = + + + = 0.
-
T 2 6 6 24

Hence, we fall on the following linear system

⎧
⎪ b
⎪
⎪ a+ =1
⎪
⎪ 2
⎪
⎪
⎪
⎪
c
⎨a + = 0
2
.
⎪
⎪ b c e
⎪
⎪ a+ + + =0
⎪
⎪ 2 2 6
⎪
⎪
⎪
⎩ a + b + c + e = 0.
2 6 6 24
The solution of this system is given by
⎧
⎪ a = −1
⎪
⎨
b=4
.
⎪c = 2
⎪
⎩
e = −12.

Thus,
ϕ (ξ, η) = −1 + 4ξ + 2η − 12ξ η.
. 1

By symmetry, we find

ϕ (ξ, η) = −1 + 4η + 2ξ − 12ξ η.
. 3

ϕ is written
. 2
ϕ (ξ, η) = a + bξ + cη + eξ η.
. 2

For .η = 0, we have
{ 1
1 b
. (a + bξ ) dξ = a + = 0.
1−0 0 2

For .ξ = 0, we have

@seismicisolation
@seismicisolation
364 6 Nodal Methods
{ 1
1 c
. (a + cη) dη = a + = 0.
1−0 0 2

For .ξ + η = 1, we have
{ 1
1 b c e
. (a + bξ + c(1 − ξ ) + eξ(1 − ξ )) dξ = a + + + = 1.
1−0 0 2 2 6

- gives
The average of .ϕ2 on .T
{
a b c e
. (a + bξ + cη + eξ η) dξ dη = + + + = 0.
-
T 2 6 6 24

Hence, we fall on the following linear system

⎧
⎪ b
⎪
⎪ a+ =0
⎪
⎪ 2
⎪
⎪
⎪
⎪
c
⎨a + = 0
2
.
⎪
⎪ b c e
⎪
⎪ a+ + + =1
⎪
⎪ 2 2 6
⎪
⎪
⎪
⎩ + b + c + e = 0.
a
2 6 6 24
The solution of this system is given by
⎧
⎪
⎪ a = −3
⎨
b=6
.
⎪
⎪ c=6
⎩
e = −18.

Hence,
ϕ (ξ, η) = −3 + 6ξ + 6η − 18ξ η.
. 2

ϕ is written
. 0

ϕ (ξ, η) = a + bξ + cη + eξ η.
. 0

We find ⎧
⎪
⎪ a = 24
⎨
b = −48
.
⎪
⎪ c = −48
⎩
e = 144.

So,
ϕ (ξ, η) = 24 − 48ξ − 48η + 144ξ η.
. 0

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 365

Thus, there exists a unique . -

p polynomial of . Q 1 taking mean values given on each
side and on the elements. Hence,

E
3
. -
p= αi ϕi
i=0

with .αi .1 ≤ i ≤ 3 is equal to the average of . -

p on the interface (on each side) and
α is equal to the average of . -
. 0 -. So, we can construct an element of . Q 1 .
p over .T
Consequently, the Nodal method is, thus, defined by

. Vh = {vh ∈ L 2 (Ω, R) such that vh |T ∈ Q 1 , vh is entirely determined by values

averages on each side and on the elements}.

We have { ( )
∂u 1h ∂u 2
. D1 + D2 h ds = 0
S ∂n 1 ∂n 2

for all . S ⊂ Ω and . S /⊂ ∂Ω and

{
. (−D/\u h + σ u h − f ) d xd y = 0 for all T ⊂ Ω.
T

The question now is: is the Nodal method identical to the non-conforming method:
Find .u h ∈ Vh such that
.ah (u h , vh ) = L(vh )

for all .vh ∈ Vh with

E{
a (u h , vh ) =
. h D∇u h · ∇vh + σ u h vh d xd y
T ⊂Ω T

E{
and
. L(vh ) = f vh d xd y,
T ⊂Ω T

with .vh |R is the average of .vh on .T , i.e.,

{
1
v
. h |T = vh d xd y
mes(T ) T

and . f |R is the average of . f on .T , i.e.,

{
1
. f |T = f d xdy.
mes(T ) T

@seismicisolation
@seismicisolation
366 6 Nodal Methods

So, we are in front of a linear system

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh . The questions that we ask:

.(i) Is there existence and uniqueness of the solution.
.(ii) What is the order of the error between the exact solution .u and the approximate
solution .u h . Give .||u − u h ||.
.(iii) Is there a non-conforming method equivalent to this Nodal method.

To answer the question .(iii), we are not going to follow the same method that we
used in the Nodal method for squares. The essential tool that has been used in the
Nodal method for squares is that the derivative normal with respect to each side is a
constant. On the other hand, for this example, we do not have this condition. Indeed,
for example
.ϕ0 (ξ, η) = 24(1 − 2ξ − 2η + 6ξ η).

Hence,
∂ϕ0
. (0, η) = 24(−2 + 6η)
∂ξ

and
∂ϕ0
. (ξ, 0) = 24(−2 + 6ξ ).
∂η

So, ( )
∂ϕ0 ∂ϕ0
. + (ξ, 1 − ξ ) = 48.
∂ξ ∂η

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 367

-=
Thus, not all normal derivatives are constants. Hence, instead of working with . P
Q 1 , we work with a method that we will describe below.

6.2.2 Description of the Nodal Method on Triangles

-2
A

-3
A 2 -1
A

We have .4 degrees of freedom, the mean over each side and the mean over the
element. In addition, we want the normal derivative to be a constant by side so that
we can find as for the square, an equivalent non-conforming method. The idea is
to choose .v(ξ, η) ∈ P2 and, to find conditions on the coefficients so that, we have a
normal derivative equal to a constant by side. Let

.v(ξ, η) = a + bξ + cη + dξ η + eξ 2 + f η2 .

Hence,
∂v
. (ξ, η) = b + dη + 2eξ
∂ξ

∂v
. (ξ, η) = c + dξ + 2 f η.
∂η

Thus,
∂v
. (0, η) = b + dη
∂ξ

∂v
. (ξ, 0) = c + dξ.
∂η

@seismicisolation
@seismicisolation
368 6 Nodal Methods

The normal derivative of .v with respect to the side .3 is

( )
∂v ∂v
. + (ξ, 1 − ξ ) = b + c + d + 2eξ + 2 f (1 − ξ ).
∂ξ ∂η

So, for the normal derivatives to be equal to a constant by sides, it suffices to take
d = 0 and .e = f . Hence,
.

v(ξ, η) = a + bξ + cη + f (ξ 2 + η2 ).
.

Therefore, we have .4 degrees of freedom so, we have .4 coefficients to determine.

We have
.v(ξ, 0) = a + bξ + f ξ
2

and
.v(0, η) = a + cη + f η2 .
{ 1
Since . v(ξ, 0) dξ = 0, we have
0

b f
a+
. + = 0.
2 3
{ 1
Since . v(0, η) dη = 1, we have
0

c f
. a+ + = 1.
2 3
{ 1
Since . v(ξ, 1 − ξ ) dξ = 0, we have
0

b c f f
.a+ + c − + + f + − f = 0.
2 2 3 3
The last equation {
. v(ξ, η) dξ dη = 0
-
T

gives
a b c f f
. + + + + = 0.
2 6 6 12 12
Hence, we fall on the following linear system

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 369
⎧
⎪
⎪ b f
⎪
⎪ a+ + =0
⎪
⎪ 2 3
⎪
⎪
⎪
⎪ c f
⎨a + + = 1
.
2 3
⎪
⎪a + b + c + 2 f = 0
⎪
⎪
⎪
⎪ 2 2 3
⎪
⎪
⎪
⎪ a b c f
⎩ + + + = 0.
2 6 6 6

The solution of this system is given by

⎧
⎪a = 1
⎪
⎨
b = −4
.
⎪
⎪ c = −2
⎩
f = 3.

Hence,
v(ξ, η) = 1 − 4ξ − 2η + 3(ξ 2 + η2 ).
.

This polynomial checks that the mean on the side .1 is equal to .1 and that the means
on the other sides are zero. Thus,

ϕ (ξ, η) = 1 − 4ξ − 2η + 3(ξ 2 + η2 )
. 1

ϕ (ξ, η) = 1 − 2ξ − 4η + 3(ξ 2 + η2 )
. 2

ϕ (ξ, η) = −1 + 3(ξ 2 + η2 )
. 3

and
ϕ (ξ, η) = 6ξ + 6η − 9(ξ 2 + η2 ).
. 0

Write the Nodal method

{
. (−D/\u h + σ u h − f ) d xd y = 0 for all T ⊂ Ω.
T

We assume the following hypothesis

{ ( )
∂u 1h ∂u 2
. D1 + D2 h ds = 0
S ∂n 1 ∂n 2

for all . S side of .Ω and . S /⊂ ∂Ω.

The normal derivatives on the reference triangle are constants but are they
constants on any triangles. The answer is unfortunately no. Indeed,

@seismicisolation
@seismicisolation
370 6 Nodal Methods

∂u ∂u ∂u
. = nx + ny.
∂n ∂x ∂y

A2

( )
−
→ y2 − y1
n
x1 − x2

A1
A3

-
∂u -
∂u -
∂u
. = nx + ny.
∂n ∂x ∂y

Hence,
-
∂u ∂-
u ∂y ∂-
u ∂y
J
. T (ξ, η) = −
∂x ∂ξ ∂η ∂η ∂ξ

-
∂u ∂-
u ∂x ∂-
u ∂x
J
. T (ξ, η) = − +
∂y ∂ξ ∂η ∂η ∂ξ

with . JT is given in (5.2.7).

∂-
u ∂-
u
+
∂ξ ∂η

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 371

Thus,

- [( ) ( ) ]
∂u 1 u ∂y
∂- ∂-
u ∂y ∂-
u ∂x ∂-
u ∂x
. (ξ, η) = − nx + − + ny
∂n JT ∂ξ ∂η ∂η ∂ξ ∂ξ ∂η ∂η ∂ξ

with
. x = ξ x1 + ηx2 + (1 − ξ − η)x3 ,

. y = ξ y1 + ηy2 + (1 − ξ − η)y3 ,

(see Lemma 4.6.8), and

y2 − y1 x1 − x2
n =
. x , ny = ,
l l
with .l is the length of side. It follows that

- [ ]
∂u 1 ∂-u ∂-
u
. (ξ, η) = a+ b ,
∂n l JT ∂ξ ∂η

with
a = (y2 − y3 )(y2 − y1 ) − (x2 − x3 )(x1 − x2 )
.

and
b = −(y1 − y3 )(y2 − y1 ) + (x1 − x3 )(x1 − x2 ).
.

The problem is that we only know that

∂-
u ∂-
u
. + = cte
∂ξ ∂η

but we don’t know if

∂-
u ∂-
u
a
. +b = cte
∂ξ ∂η

for .a /= b. But if .a = b, we have

∂-
u
. = cte
∂n
but this condition is not always true. Hence, the idea of introducing barycentric
coordinates.

@seismicisolation
@seismicisolation
372 6 Nodal Methods

6.2.3 Barycentric Coordinates

A2

a1− −a3

A3 a2 A1

We consider polynomials of degree .2

ϕ = 1 − 6λi + 6λi2 1 ≤ i ≤ 3
. i

( )
E
3
ϕ =2 2−3
. 0 λi2 .
i=1

We check that these functions are the basic elements. Indeed,

. 1 ϕ = 1 − 6λ1 + 6λ21 .

On the side . A2 A3 , .λ1 = 0, so .ϕ1 = 1. Thus, the average of .ϕ1 is equal to .1. On the
side . A1 A2 , .λ1 ranges from .0 to .1, so
{ 1
6 6
. (1 − 6λ1 + 6λ21 )dλ1 = 1 − + = 0.
0 2 3

On the side . A1 A3 , .λ1 ranges from .0 to .1, so

{ 1
. (1 − 6λ1 + 6λ21 )dλ1 = 0
0

and { ( ( ) )
mes(T ) 6 6
. ϕ1 = 1+ 1− + 2 = 0.
T 3 2 4

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 373

Likewise, for others. Hence, .{ϕi }0≤i≤3 are the basis elements. We have

∂ϕi
.
∂n

not constant on the sides because . ∂ϕ

∂n
i
∈ P1 . So, we will assume the following hypoth-
esis to show that the Nodal method is equal to the non-conforming method. So, we
will assume { {
∂u 1h ∂u 1
. D1 ds = D1 h ϕ3 ds (6.2.1)
S ∂n 1 ∂T ∂n 1

with . S = A1 A2 .
A2
ϕ3

A3 1
S 2

A1

On . A2 A3 and . A3 A1 , . ∂u
∂n
h
∈ P1 because .ϕi ∈ P2 . For that, let’s seek the roots of the
polynomial .1 − 6λ + 6λ2 . We pose
√
1 ± 3
.λ = ±
2 2×3

the abscissa of Gauss Legendre on .[0, 1]. Hence, .1 − 6λ + 6λ2 along a side is equal
to the Legendre polynomial of degree .2. However, the Legendre polynomial . pi ,
defined on .[0, 1] of degree .i, satisfies
{ 1
. pi (x) p j (x) d x = δi j
0

and { 1
. pi (x)q(x) d x = 0
0

@seismicisolation
@seismicisolation
374 6 Nodal Methods

for any polynomial .q of strict degree less than .i, with .δi j is the Kronecker symbol.
So, from this condition, we have the desired hypothesis (6.2.1), i.e.,
{ {
∂u 1 ∂u 1h
. D1 h ds = D1 ϕ3 ds.
S ∂n 1 ∂T ∂n 1

∂u 1
In fact, . ∂nh1 ∈ P1 on the sides . A1 A3 and . A3 A2 and, .ϕ3 is a Legendre polynomial of
degree .2, so {
∂u 1h
. ϕ3 = 0
A1 A3 ∂n 1

and {
∂u 1h
. ϕ3 = 0.
A2 A3 ∂n 1

Hence, { {
∂u 1h ∂u 1h
. D1 ds = D1 ϕ3 ds.
S ∂n 1 ∂T ∂n 1

We end this chapter with a theorem which shows that the Nodal method is identical
to the non-conforming method.

Theorem 6.2.1 Either the non-conforming method: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h (6.2.2)

for all .vh ∈ Vh , where

E{
a (u h , vh ) =
. h D∇u h · ∇vh + σ u h vh d xd y
T ⊂Ω T

E{
and
. L(vh ) = f vh d xd y
T ⊂Ω T

with . D and .σ are equal to constants with respect to each triangle .T , .vh |T is the
average of .vh on .T , i.e., {
1
.vh |T = vh d xd y
mes(T ) T

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 375

and . f |T is the average of . f on .T , i.e.,

{
1
. f |T = f d xdy.
mes(R) T

Then, the Nodal method is identical to the non-conforming method. ♦

Proof We recall the reasoning made in [3]. We consider the equations of the nodal
method and we will show that they can be written in the variational form (6.2.2). Let
.G A1 A2 ∈ Vh be the zero-mean function everywhere except on . A 1 A 2 . Hence,

{ {
∂u 1 ∂u 2h
. D1 h G A1 A2 ds + D2 G A1 A2 ds = 0
∂ T1 ∂n 1 ∂ T2 ∂n 2

because { {
∂u 1h ∂u 2h
. D1 ds + D2 ds = 0.
S ∂n 1 S ∂n 2

According to Green’s formula, we have

{ { {
. D1 /\u 1h G A1 A2 d xd y + D1 ∇u 1h ∇G A1 A2 d xd y + D2 /\u 2h G A1 A2 d xd y
T1 T1 { 2 T

. + D2 ∇u 2h ∇G A1 A2 d xd y = 0.
T2

Further, ./\u 1h = cte, ./\u 2h = cte,

{
. G A1 A2 d xd y = 0
T1

and {
. G A1 A2 d xd y = 0.
T2

Hence, { {
. D1 ∇u 1h ∇G A1 A2 d xd y + D2 ∇u 2h ∇G A1 A2 d xd y = 0.
T1 T2

Furthermore, {
. σ u h 1 G A1 A2 d xd y = 0
T1

{
. σ u h 2 G A1 A2 d xd y = 0
T2

{
. f G A1 A2 d xd y = 0
T1

@seismicisolation
@seismicisolation
376 6 Nodal Methods

and {
. f G A1 A2 d xd y = 0.
T2

Hence,
{ {
. (D1 ∇u 1h ∇G A1 A2 + σ u h 1 G A1 A2 − f G A1 A2 ) dxdy+ (D2 ∇u 2h ∇G A1 A2 + σ u h 2 G A1 A2
T1 T2
− f G A1 A2 ) dxdy = 0.

Let .G AB ∈ Vh be the zero-mean function everywhere except on . AB and .T /= T1 and

T /= T2 , we have
.

{
. (D1 ∇u hT ∇G AB + σ u h T G AB − f G AB ) d xd y = 0
T
U
because .supp(G AB ) = T1 T2 . Hence,
E{
. (D1 ∇u hT ∇G AB + σ u h T G AB − f G AB ) d xd y = 0.
T ⊂Ω T

On the other hand, {

. (−D/\u h + σ u h − f ) d xd y = 0
T

for all .T ⊂ Ω. Then,

{ { {
∂u h
. −D/\u h d xd y = D∇u h ∇ϕ0 d xd y − D ϕ0 ds
T1 T1 ∂ T1 ∂n T

with
. 0 ϕ = 2(2 − 3(λ21 + λ22 + λ23 )).

On the side . A1 A2 , we have .λ3 = 0 and so

ϕ = 2(2 − 3(λ21 + λ22 )).

. 0

Since
E
3
. λi = 1,
i=1

@seismicisolation
@seismicisolation
6.2 Nodal Method on Triangles 377

so on side . A1 A2 , .λ1 + λ2 = 1, .λ2 = 1 − λ1 and .λ22 = (1 − λ1 )2 . Thus,

ϕ
. 0 |A A
1 2
= 2(2 − 3(λ21 + (1 − λ1 )2 ))
= 2(2 − 3(λ21 + 1 + λ21 − 2λ1 ))
= −2(1 − 6λ1 + 6λ21 ).

It is a Legendre polynomial. Since

∂u h
. ∈ P1 ,
∂n T

we have {
∂u h
. D ϕ0 = 0.
A1 A2 ∂n T

Thus, { {
. −D/\u h d xd y = D∇u h ∇ϕ0 d xd y.
T1 T1

Further, { { {
. σ u h d xd y = σ u h ϕ0 d xd y = σ u h ϕ0 d xd y
T1 T1 T1

and { {
. f d xdy = f ϕ0 d xd y.
T1 T1

Hence, {
. (D∇u h ∇ϕ0 + σ u h ϕ0 − f ϕ0 ) d xd y = 0.
T1

E{
Thus,
. (D∇u h ∇ϕ0T + σ u h ϕ0T − f ϕ0T ) d xd y = 0
T ⊂Ω T

because .supp(ϕ0 ) = T1 . Let .vh ∈ Vh , be the linear combination of .ϕ0T and .G AB , so

E{
. (D∇u h · ∇vh + σ u h vh − f vh ) d xd y = 0.
T ⊂Ω T

Furthermore, { {
. σ u h vh d xd y = σ u h vh d xd y.
T T

@seismicisolation
@seismicisolation
378 6 Nodal Methods

E{
Hence,
. (D∇u h · ∇vh + σ u h vh − f vh ) d xd y = 0.
T ⊂Ω T

The last equation is equivalent to finding .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh . Thus, the Nodal method is identical to the non-conforming method.
This completes the proof. Q.E.D.

Remark 6.2.2 Let

. P = {1 − 6λi + 6λi2 1 ≤ i ≤ 3, 2(2 − 3(λ21 + λ22 + λ23 ))}.

Then, we have the inclusion . P0 ⊂ P but . P1 /⊂ P. In fact, .λ1 ∈ P1 and .λ1 ∈

/ P. ♦

References

1. R.D. Lauwrence, Progress in nodal methods for the solution of the neutron diffusion and transport
equations. Prog. Nucl. Energy 17 (1986)
2. K. Koebke, M.R. Wagner, Progress in nodal reactor analysis. Atomkernnenergie Kerntechnik
43 (1983)
3. C. Fedon-Magnaud, Etude théorique de quelques méthodes nodales de résolution de l’équation
de la diffusion, Thèse de doctorat de l’université de Paris, vol. 6 (1983)

@seismicisolation
@seismicisolation
Chapter 7
Positive Symmetric First-Order Systems
Within the Meaning of Friedrichs

This chapter is devoted to the analysis of the positive symmetric first-order systems
within the meaning of Friedrichs.

7.1 First Order of Friedrichs Symmetric Systems

Let .Ω be a bounded open set of .Rn , whose boundary .∂Ω is continuous [1] and a
time continuously differentiable by pieces (we will say .C 1 by parts). We consider
the Friedrichs symmetric systems

E
n
∂u
. Au(x) = Ai (x) (x) + A0 (x)u(x), (7.1.1)
i=1
∂ xi

where .x ∈ Rn , .u(x) ∈ R p , . Ai is a . p × p symmetric matrix .1 ≤ i ≤ n with Lip-

chitizian coefficients and . A0 is a . p × p matrix, not necessarily symmetrical, with
coefficients in . L ∞ (Ω, R). We want to solve

E
n
∂u
. Ai (x) (x) + A0 (x)u(x) = F(x) for x ∈ Ω (7.1.2)
i=1
∂ xi

(B − M)(x)u = 0 for x ∈ ∂Ω,

. (7.1.3)

where . F ∈ (L p (Ω, R)) p , . M is a . p × p matrix and is defined for .x ∈ Γ := ∂Ω,

E
n
. B= νi Ai
i=1

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 379
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_7
@seismicisolation
@seismicisolation
380 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

with .νi .1 ≤ i ≤ n are the components of the normal vector outside the domain .∂Ω.
We will say that .u is a classical solution if .u ∈ (C 1 (Ω, R)) p . We denote by .t M the
adjoint matrix of . M and, we assume that

M(x). is continuous in x for x ∈ ∂Ω (7.1.4)

M(x). +t M(x) ≥ 0 (7.1.5)
K er (B
. − M)(x) + K er (B + M)(x) = R , for all x ∈ ∂Ω.
p
(7.1.6)

The formal adjoint operator . A∗ of . A is given by

En
∂
∗
. A u(x) = − (Ai (x)u(x)) + A∗0 (x)u(x) for x ∈ Ω.
i=1
∂ x i

We say that .u satisfies the adjoint boundary conditions if

.(B +t M)(x)u = 0 for x ∈ ∂Ω. (7.1.7)

7.1.1 Strong Solution

Definition 7.1.1 We will say that .u is a strong solution of the problem (7.1.2)–
(7.1.3), if there is a sequence .(u j ) j of functions of .(H 1 (Ω)) p satisfying the boundary
conditions (7.1.3) and such that
( )
. lim ||u j − u||(L 2 (Ω,R)) p + ||F − Au j ||(L 2 (Ω,R)) p = 0.
j→∞ ♦
Definition 7.1.2 We say that the operator. A defined on a subspace.W of.(C 1 (Ω, R)) p
is coercive, if there exists a constant .α > 0 such that

(Av, v)(L 2 (Ω,R)) p ≥ α||v||2(L 2 (Ω,R)) p for all v ∈ W.

.
♦

There are different theorems for the existence of a strong solution, depending on
the assumptions made on the boundary .∂Ω, the second member . F, the matrices
. B(x) and . M(x), . x ∈ ∂Ω. We quote one of them [2]. For more details, we refer to

Friedrichs [3], Lax and Phillips [2], Phillips and Sarason [4].

Theorem 7.1.3 Let . A be a positive operator. We assume that the boundary .∂Ω is
of class .C 2 , that the hypotheses (7.1.4)–(7.1.5)–(7.1.6) are satisfied and that the
subspace . K er (B(x) − M(x)) of .R p is generated by vectors varying continuously
with .x for .x ∈ ∂Ω. Then, for all . F ∈ (L 2 (Ω, R)) p , the problem (7.1.2)–(7.1.3) has
a unique strong solution. ♦

@seismicisolation
@seismicisolation
7.1 First Order of Friedrichs Symmetric Systems 381

Question: Can we underestimate the term

{
.(Au, u)(L 2 (Ω,R)) p = (Au, u)2 d x ≥ · · ·
Ω

to have coercivity, with .(·, ·)2 is the euclidean scalar product of .R p , given by

E
p
(u, v)2 =
. u i vi
i=1

and .u i (resp. .vi ) is the .i-th component of the vector .u (resp. .v) and
{
(u, v)(L 2 (Ω,R)) p =
. (u, v)2 d x.
Ω

For that we will introduce Green’s formula.

7.1.2 Green’s Formula

Let .ϕ, .ψ ∈ (D(Ω)) p , then

(Aϕ, ψ)(L 2 (Ω,R)) p = (ϕ, A∗ ψ)(L 2 (Ω,R)) p ,

.

with
En
∂
. A∗ v = − (Ai (x)v) + A∗0 v. (7.1.8)
i=1
∂ x i

In fact,
{
. (Aϕ, ψ)(L 2 (Ω,R)) p = (Aϕ, ψ)2 d x
Ω
{ (E n
)
∂ϕ
= Ai (x) (x) + A0 ϕ(x), ψ dx
Ω i=1
∂ xi
2
{ E n ( ) {
∂ϕ
= (x), Ai (x)ψ d x + (ϕ, A∗0 ψ)2 d x.
Ω i=1 ∂ xi 2 Ω

Now,. Ai with Lipschitzian coefficients, therefore, differentiable and, therefore,. ∂∂xi Ai

has a meaning. So,

@seismicisolation
@seismicisolation
382 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

{ En ( ) {
∂
(Aϕ, ψ)(L 2 (Ω,R)) p
. = − ϕ, (Ai (x)ψ) d x + (ϕ, A∗0 ψ)2 d x.
Ω i=1
∂ xi 2 Ω

Thus,
En
∂
. A∗ ψ = − (Ai (x)ψ) + A∗0 ψ.
i=1
∂ x i

Theorem 7.1.4 We assume that .Ω is an open bounded of .Rn with piecewise smooth
boundary .Γ = ∂Ω. Let .u, .v ∈ (H 1 (Ω)) p , then
{ { {
∗
. (Au, v)2 d x = (u, A v)2 d x + (Bu, v)2 dσ,
Ω Ω Γ

where . A (resp. . A∗ ) is defined in (7.1.1) (resp. (7.1.8)),

E
n
. B= νi Ai
i=1

with .νi .1 ≤ i ≤ n are the components of the normal vector outside the domain .∂Ω
and .dσ is a surface element. ♦

7.1.3 Direct Problems

Let .v ∈ (H 1 (Ω)) p . According to Green’s formula (Theorem 7.1.4), we have

{
(Av, v)(L 2 (Ω,R)) p =
. (Av, v)2 d x
{Ω {
∗
= (v, A v)2 d x + (Bv, v)2 dσ
Ω Γ
{ ( En
) {
∂ ∗
= v, − (Ai (x)v) + A0 v d x + (Bv, v)2 dσ.
Ω i=1
∂ xi Γ
2

On the other hand,

En
∂ En
∂ Ai (x) En
∂v
. (Ai (x)v) = v+ Ai (x)
i=1
∂ x i i=1
∂ x i i=1
∂ xi
E
n
∂ Ai (x)
= v + Av − A0 v.
i=1
∂ xi

@seismicisolation
@seismicisolation
7.1 First Order of Friedrichs Symmetric Systems 383

Hence,
{ ( E
n
) { {
∂ Ai (x)
.(Av, v)(L 2 (Ω,R)) p = v, − v + A0 v + A∗0 v dx − (v, Av)2 d x + (Bv, v)2 dσ.
Ω ∂ x i Ω Γ
i=1 2

Thus,
{ ( ( E n
) ) {
∂ ∗
.2(Av, v)(L 2 (Ω,R)) p = v, − (Ai (x)) + A0 + A0 v d x + (Bv, v)2 dσ.
Ω ∂ xi Γ
i=1 2

So, we have the following.

Lemma 7.1.5 Let .v ∈ (H 1 (Ω)) p . Then,

{ ( ( E n
) ) {
∂ ∗
.2(Av, v)(L 2 (Ω,R)) p = v, − (Ai (x)) + A0 + A0 v d x + (Bv, v)2 dσ.
Ω ∂ xi Γ
i=1 2

Lemma 7.1.6 For all.u (resp..v) in.(H 1 (Ω)) p and satisfying the boundary conditions
(7.1.3) (resp. (7.1.7)), we have

(Au, v)(L 2 (Ω,R)) p = (u, A∗ v)(L 2 (Ω,R)) p .

.

Remark 7.1.7 If the term {

. (Bv, v)2 dσ < 0,
Γ

then we cannot absorb it by the other term to have the coercivity. ♦

Let
En
∂
.C(x) = A0 (x) + A∗0 (x) − (Ai (x)).
i=1
∂ xi

Definition 7.1.8 We say that the system (7.1.2)–(7.1.3) is positive in the sense of
Friedrichs if there exists .α > 0 such that

.
t
XC(x)X ≥ 2α t X X (7.1.9)

for all . X ∈ R p . ♦

@seismicisolation
@seismicisolation
384 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

If the condition (7.1.9) is satisfied, then

{ ( ( E n
) )
∂ ∗
. v, − (Ai (x)) + A0 + A0 v d x ≥ 2α||v||2(L 2 (Ω,R)) p
Ω i=1
∂ x i
2

for all .v ∈ (H 1 (Ω)) p . Hence, we only have the term

{
. (Bv, v)2 dσ
Γ

that we would like it to be positive or zero. For that, a sufficient condition is to define
the subspace .W by

. W = {v ∈ (C 1 (Ω, R)) p such that B(x)v(x) = M(x)v(x) for all x ∈ Γ }, (7.1.10)

where . M is a continuous coefficient matrix on .Γ and . M +t M is positive semi-

definite. We say that the conditions at the limits . Bv = Mv are semi-admissible if
. M + M is positive semi-definite. If .v ∈ W , then
t

{
. (Bv, v)2 dσ ≥ 0.
Γ

In fact, since . M +t M is positive semi-definite, we have

{ { {
. (Bv, v)2 dσ = ((B − M)v, v)2 dσ + (Mv, v)2 dσ
Γ Γ
({ { Γ )
1
= (Mv, v)2 dσ + (v, Mv)2 dσ
t
2 Γ Γ
({ )
1
= ((M + M)v, v)2 dσ ≥ 0.
t
2 Γ

Hence, {
. (Bv, v)2 dσ ≥ 0.
Γ

So, if .v ∈ W , then {
. (Av, v)2 d x ≥ α||v||2(L 2 (Ω,R)) p .
Ω

Thus, we only have coercivity in .(L 2 (Ω, R)) p . Consequently, we have the existence
and uniqueness of the solution only in .(L 2 (Ω, R)) p of the problem: Find .u ∈ R p
such that
.(Au, v)(L 2 (Ω,R)) p = L(v)

for all .v ∈ W , where . L(v) = (F, v)(L 2 (Ω,R)) p .

@seismicisolation
@seismicisolation
7.2 Existence of a Weak Solution for the Problem (7.1.2)–(7.1.3) 385

7.2 Existence of a Weak Solution for the Problem

(7.1.2)–(7.1.3)

To define a weak solution of the problem (7.1.2)–(7.1.3), we multiply the relation

Au = F (which has a formal meaning) by a regular test function .ϕ. We have Green’s
.
formula (valid for .u and .ϕ ∈ (H 1 (Ω)) p ),
{ { {
. (Au, ϕ)2 d x = (Bu, ϕ)2 dσ + (u, A∗ ϕ)2 d x.
Ω Γ Ω

The boundary conditions (7.1.10) allow to write

{ { {
1 1
. (Bu, ϕ)2 dσ = ((B − M)u, ϕ)2 dσ + ((B + M)u, ϕ)2 dσ
Γ 2 Γ 2 Γ
{
1
= (u, (B +t M)ϕ)2 dσ.
2 Γ

Since we are looking for a weak solution .u ∈ (L 2 (Ω, R)) p , a simple (and natural)
way to make sense of the integral
{
. (u, (B +t M)ϕ)2 dσ
Γ

is to force the test function .ϕ to satisfy the boundary conditions

(B +t M)ϕ = 0.
.

We will say that .u is a weak solution of the problem (7.1.2)–(7.1.3) if .u ∈

(L 2 (Ω, R)) p and if we have

(u, A∗ ϕ)(L 2 (Ω,R)) p = (F, ϕ)(L 2 (Ω,R)) p for all ϕ ∈ (C 1 (Ω, R)) p
.

with .(B +t M)ϕ = 0 on .Γ .

Remark 7.2.1 If .u is a classical solution of (7.1.2)–(7.1.3), then .u is also a weak

solution. ♦

Lemma 7.2.2 Every strong solution is a weak solution. ♦

Proof For all .ϕ ∈ (C 1 (Ω, R)) p , with .(B +t M)ϕ = 0 on .Γ , we have

.(u j , A∗ ϕ)(L 2 (Ω,R)) p = (Au j , ϕ)(L 2 (Ω,R)) p .

@seismicisolation
@seismicisolation
386 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Passing to the limit for . j → +∞, we get

.(u, A∗ ϕ)(L 2 (Ω,R)) p = (F, ϕ)(L 2 (Ω,R)) p .

Q.E.D.

Let .z ∈ (H 1 (Ω)) p , according to Green’s formula (Theorem 7.1.4), we have

{ { {
∗
. (Au, z)2 d x = (u, A z)2 d x + (Bu, z)2 dσ.
Ω Ω Γ

We can write
B−M B+M
. B= + .
2 2
Hence,
{ { (( ) ) { (( ) )
B−M B+M
. (Bu, z)2 dσ = u, z dσ + u, z dσ
Γ Γ 2 Γ 2
{ ( ( ) ) 2 2
B +t M
= u, z dσ
Γ 2 2

because . B is symmetric since the . Ai are also symmetrical. Hence, according to the
last equality, we have taken
.(B − M)u = 0

because . M is arbitrary. So, to get rid of the boundary term, we can take the next
space { }
∗
. W = z ∈ (C (Ω, R)) such that (B + M)z |∂Ω = 0 .
1 p t

Consider the weak problem: Find .u ∈ (L 2 (Ω, R)) p such that

{ {
∗
. (u, A z)2 d x = (F, z)2 d x (7.2.1)
Ω Ω

for all .z ∈∗ W .
Question: Is there existence of .u, uniqueness and regularity.
If, we have the existence of .u, then . Au = F in .D , (Ω). In this case can one then
recover the boundary conditions.

Remark 7.2.3 .(i) .(B +t M)z = 0 is an adjoining boundary condition. Indeed, if

we consider . A∗

@seismicisolation
@seismicisolation
7.2 Existence of a Weak Solution for the Problem (7.1.2)–(7.1.3) 387

En
∂
. A∗ z = − (Ai z) + A∗0 z
i=1
∂ x i
( n )
En
∂z E ∂ Ai
=− Ai + − + A∗0 z.
i=1
∂ x i i=1
∂ x i

If we associate to . A∗ the . B A∗ , we have

E
n
. B A∗ =− νi Ai = −B A .
i=1

The choice of
E
n
. B A∗ = − νi Ai
i=1

results from the fact that

( n )
E
n
∂z E ∂ Ai
∗ ∗
. A z=− Ai + − + A0 z.
i=1
∂ xi i=1
∂ xi

The space .W A∗ is given by

{ }
. W A∗ = z ∈ (C 1 (Ω, R)) p such that (B A∗ − M1 )z |∂Ω = 0 .

If we choose . M1 =t M, then .(B A∗ − M1 )z = −B A −t M = 0. Thus, the name con-

ditions to the adjoining limits.
(ii) If we have uniqueness of the problem (7.2.1), then we can hope for the existence
.
for the direct problem. ♦

Lemma 7.2.4 The operator . A∗ with the adjoining boundary conditions satisfies
{
. (A∗ z, z)2 d x ≥ α||z||2(L 2 (Ω,R)) p
Ω

for all .z ∈∗ W . ♦

Proof Let .z ∈∗ W , we have

En
∂
. A∗ z = − (Ai z) + A∗0 z.
i=1
∂ x i

@seismicisolation
@seismicisolation
388 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Hence,
{ { (( E n
) )
∗ ∂ ∗
. (A z, z)2 d x = − (Ai (x)) + A0 + A0 − A z, z d x.
Ω Ω i=1
∂ xi
2

Thus, { {
1
. (A∗ z, z)2 d x = (C(x)z, z)2 d x
Ω 2 Ω

with
En
∂
. C(x) = A0 + A∗0 − (Ai (x)).
i=1
∂ xi

Since the system is positive in the sense of Friedrichs, we have

.
t
XC(x)X ≥ 2α t X X.

Hence, {
. (A∗ z, z)2 d x ≥ α||z||2(L 2 (Ω,R)) p
Ω

for all .z ∈∗ W . This allows us to say that there is uniqueness in .∗ W for the solution
of . Au = F. Q.E.D.

Theorem 7.2.5 Let . F ∈ (L 2 (Ω, R)) p . We assume that . A is positive in the sense
of Friedrichs and that the boundary conditions (7.1.10) are semi-admissible. Then,
there is .u ∈ (L 2 (Ω, R)) p such that
{ {
∗
. (u, A z)2 d x = (F, z)2 d x
Ω Ω

for all .z ∈∗ W . ♦

Proof Let

. V = {v = A∗ z, z ∈∗ W }
= I m(A∗ )
= A∗ (∗ W ).

.V is a vector subspace of .(L 2 (Ω, R)) p . We equip it of the norm of .(L 2 (Ω, R)) p .
Usually is not dense in .(L 2 (Ω, R)) p and .(V, || · ||(L 2 (Ω,R)) p ) is not always complete
and in general is not dense in .(L 2 (Ω, R)) p . On .V , we define the map . L(·), with
∗
.v ∈ V , we associate . z ∈ W in a unique way

@seismicisolation
@seismicisolation
7.2 Existence of a Weak Solution for the Problem (7.1.2)–(7.1.3) 389
{
v −→
. (F, z)2 d x
Ω

which is linear, and continuous on .(L 2 (Ω, R)) p . In fact,

.|L(v)| ≤ ||F||(L 2 (Ω,R)) p ||z||(L 2 (Ω,R)) p

1
≤ ||F||(L 2 (Ω,R)) p ||v||(L 2 (Ω,R)) p
α
because

(v, z)(L 2 (Ω,R)) p =(A∗ z, z)(L 2 (Ω,R)) p

.

≥α||z||2(L 2 (Ω,R)) p

and
1
||z||(L 2 (Ω,R)) p ≤
. ||v||(L 2 (Ω,R)) p .
α

According to the Han Banach theorem, we can extend . L by . L which is linear and
continuous on .(L 2 (Ω, R)) p and, checking

|L(v)|
||L|| = sup
.
||v||(L 2 (Ω,R)) p
1
≤ ||F||(L 2 (Ω,R)) p .
α

According to the Riesz Fréchet theorem, there is .u ∈ (L 2 (Ω, R)) p such that

. L(v) = (u, v)(L 2 (Ω,R)) p

for all .v ∈ (L 2 (Ω, R)) p and

||u||(L 2 (Ω,R)) p = ||L||.

.

We restrict ourselves to the space .V ⊂ (L 2 (Ω, R)) p , so

{
. L(v) = (F, z)2 d x
Ω
= (u, v)(L 2 (Ω,R)) p
= (u, A∗ z)(L 2 (Ω,R)) p

for all .z ∈∗ W . Q.E.D.

@seismicisolation
@seismicisolation
390 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Corollary 7.2.6 We assume that . A is positive in the sense of Friedrichs and that the
boundary conditions (7.1.10) are semi-admissible. Then, for all . F ∈ (L 2 (Ω, R)) p ,
the problem (7.1.2)–(7.1.3) has at least one weak solution that verifies

1
||u||(L 2 (Ω,R)) p ≤
. ||F||(L 2 (Ω,R)) p .
α
♦

7.3 Strong Solution

Lemma 7.3.1 If a strong solution exists, it is unique. ♦

Proof Let .u and .- u be two distinct strong solutions. There are two sequences .(u j ) j
and .(-
u j ) j of functions of .(C 1 (Ω, R)) p satisfying .(B − M)u j = (B − M)-
u j = 0 on
.Γ such that

( )
. lim ||u j − u||(L 2 (Ω,R)) p + ||F − Au||(L 2 (Ω,R)) p = 0,
j→+∞

and ( )
. lim ||-
uj −- u ||(L 2 (Ω,R)) p = 0.
u ||(L 2 (Ω,R)) p + ||F − A-
j→+∞

By using Lemma 7.1.5, we have

{ {
.2(A(u j −-
u j ), u j − -
u j )(L 2 (Ω,R)) p = (C(u j − -
u j ), u j − -
u j )2 d x + (B(u j − -
u j ), u j − -
u j )2 dσ
Ω Γ
{ {
= (C(u j − -
u j ), u j − -
u j )2 d x + (M(u j − -
u j ), u j − -
u j )2 dσ.
Ω Γ

Hence,
.(A(u j − -
u j ), u j − -
u j )(L 2 (Ω,R)) p ≥ α||u j − -
u j ||2(L 2 (Ω,R)) p .

Passing to the limit when . j → +∞, it comes

0 ≥ α||u j − -
. u j ||2(L 2 (Ω,R)) p .

Thus, .u j = -
u j. Q.E.D.

Remark 7.3.2 Using Lemma 7.1.5, we can also show that if a strong solution .u
exists, then

1
||u||(L 2 (Ω,R)) p ≤
. ||F||(L 2 (Ω,R)) p .
α
♦

@seismicisolation
@seismicisolation
7.4 Boundary Conditions 391

Remark 7.3.3 .(i) We took .(V, || · ||(L 2 (Ω,R)) p ) but we can take .(V, || · ||1,Ω ) and, we
use the fact that the canonical injection from . H 1 (Ω) into . L 2 (Ω, R) is continuous.
(ii) There is uniqueness of the solution .u ∈ (L 2 (Ω, R)) p such that
.

. (u, A∗ z)(L 2 (Ω,R)) p = (F, z)(L 2 (Ω,R)) p

{ }
for all .z ∈∗ W = z ∈ (C 1 (Ω, R)) p such that (B +t M)z |∂Ω = 0 only for .u ∈
(H 1 (Ω)) p (to be able to apply Green’s formula).
.(iii) Inhomogeneous boundary conditions can be introduced, for example for the
problem (7.1.2)–(7.1.3) by writing

(B − M)(u − g) = 0 on ∂Ω,
.

where .g is given in . L 2 (∂Ω, R). ♦

7.4 Boundary Conditions

We have . Au = F on .D , (Ω). We assume that .u ∈ (H 1 (Ω)) p (to be able to apply

-∗
Green’s formula). Since .C 1 (Ω, R) = H 1 (Ω), from .∗ W we can define the space .W
by { }
- ∗ = z ∈ (H 1 (Ω)) p such that (B +t M)z |∂Ω = 0 .
.W (7.4.1)

- ∗ and we integrate on .Ω, we obtain

We multiply by .z ∈ W

(Au, z)(L 2 (Ω,R)) p = (F, z)(L 2 (Ω,R)) p

.

- ∗ . According to Green’s formula

for all .z ∈ W
{
(u, A∗ z) +
. (Bu, z)2 dσ = (F, z).
Γ

Hence, {
. (Bu, z)2 dσ = 0.
Γ

We replace . B par
B−M B+M
. B= + ,
2 2
we obtain
{ (( ) ) { ( ( ) )
B−M B +t M
. u, z dσ + u, z dσ = 0.
Γ 2 2 Γ 2 2

@seismicisolation
@seismicisolation
392 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Thus, { (( ) )
B−M
. u, z dσ = 0
Γ 2 2

- ∗ . Suppose that this entails .((B − M)u, z)(x) = 0 for all .x ∈ Γ . We

for all .z ∈ W
would like to conclude that .(B − M)u(x) = 0.
Lemma 7.4.1 If the following condition
n
. I m(B − M) (K er (B +t M))⊥ = {0}

is satisfied, then .(B − M)u(x) = 0. ♦

Proof We have .((B − M)u, z)(x) = 0 for all .z ∈ K er (B + M). Hence, .(B −
t

M)u ∈ (K er (B +t M))⊥ . Gold, .(B − M)u ∈ I m(B − M), so

n
(B − M)u ∈ I m(B − M)
. (K er (B +t M))⊥ = {0}.

Thus, .(B − M)u(x) = 0. Q.E.D.

The proof of the following theorem is based on Lemmas 2.9.14 and 2.9.15.
Theorem 7.4.2 . K er (B − M)(x) + K er (B + M)(x) = R p for all .x ∈ Γ . ♦
Preuve. According to Lemma 2.9.15, we have

(K er (B +t M))⊥ = I m(B + M).

.

n
Hence, since . I m(B − M) (K er (B +t M))⊥ = {0}, then
n
. I m(B − M) I m(B + M) = {0}.

Again, n
. I m(B − M) I m(B + M) = {0}.

Further, according to Lemma 2.9.14, we have

( )⊥
. (I m(B − M))⊥ + (I m(B + M))⊥ = {0}.

Hence,
(I m(B − M))⊥ + (I m(B + M))⊥ = R p .
.

Thus,
. K er (B − M)(x) + K er (B + M)(x) = R p

for all .x ∈ Γ . C.Q.F.D.

@seismicisolation
@seismicisolation
7.4 Boundary Conditions 393

Definition 7.4.3 If . M satisfies

. K er (B −t M)(x) + K er (B +t M)(x) = R p

we, then, say that the boundary conditions are admissible. ♦

Lemma 7.4.4 We assume that . M +t M is positive semi-definite. Then, for .v ∈ R p ,

. Mv = 0 if, and only if, . Mv = 0. ♦
t

Proof Since . M +t M is positive semi-definite, then for all .v ∈ R p , we have

. ||(M +t M)v||22 ≤ ||M +t M||2 t v(M +t M)v

≤ 2||M +t M||2 t vMv.

If. Mv = 0, then.(M +t M)v = 0 and so.t Mv = 0. If.t Mv = 0, then.(M +t M)v = 0

and so . Mv = 0. Q.E.D.

Lemma 7.4.5 Let . B and . M be two matrices satisfying the following

n
. I m(B − M) I m(B + M) = {0}.

Then, if .v ∈ R p satisfies . Bv = 0, we have . Mv = 0. ♦

Proof Let .x = (B − M)v and . y = (B + M)v. Then, .x + y = 0 implies .x = −y

and so .x = 0 and . Mv = 0. Q.E.D.

Lemma 7.4.6 Let . B =t B and . M be a matrix such that . M +t M ≥ 0. So, we have

the equivalence between the two following assertions
n
.(i) . I m(B − M) Inm(B + M) = {0} or . K er (B −t M) + K er (B +t M) = R p .
.(ii) . I m(B − M) I m(B +t M) = {0} or . K er (B − M) + K er (B + M) =
t

Rp. ♦
n
Proof .(i) ⇒ (ii) Let .w ∈ I m(B −t M) I m(B +t M). Let us show that .w = 0.
In fact, we can write
.2B = B + M + B − M
t t

and
2t M = B +t M − (B −t M).
.

There is .u + such that

. w = (B +t M)u +

and, there is .u − such that

w = (B −t M)u − .
.

@seismicisolation
@seismicisolation
394 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Furthermore, we can write.u − = v1 + v2 with.v1 ∈ K er (B +t M) and.v2 ∈ K er (B −t

M), and .u + = -
v1 +-v2 with .-v1 ∈ K er (B +t M) and .- v2 ∈ K er (B −t M). On the
other hand,
.w = (B − M)v1
t

and
.w = (B +t M)-
v2 .

Hence,
(B +t M)v1 = 0
.

and
. (B −t M)-
v2 = 0.

Thus,

2t M(v1 +-
. v2 ) = (B +t M)(v1 +-v2 ) − (B −t M)(v1 +-
v2 )
= (B + M)-
t
v2 − (B − M)v1 = 0.
t

Hence, . M(v1 +-
v2 ) = 0. On the other hand,

2B(v1 −-
. v2 ) = (B +t M)(v1 −-v2 ) + (B −t M)(v1 −-
v2 )
= −(B + M)-
t
v2 + (B −t M)v1 = 0.

Hence,. M(v1 −- v2 ) = 0. Thus,. Mv1 = 0 and. M-v2 = 0. Thus,.t Mv1 = 0 and.t M-v2 =
0. Further, .w = (B +t M)- v2 = B-v2 = 0, so .w = 0.
In the same way, we show the implication .(ii) ⇒ (i). Q.E.D.

Lemma 7.4.7 Let . M be a matrix such that . M +t M ≥ 0 and

. K er (B − M) + K er (B + M) = R p .

.(i) If . B is positive semi-definite, then the only choice possible of . M is . M = B.

.(ii) If . B is negative semi-definite, then the only choice possible of . M is
. M = −B. ♦

Proof .(i) We assume that . B ≥ 0. Let us show that . K er (B + M) ⊂ K er (B − M).

In fact, let .u ∈ K er (B + M), so .(B + M)u = 0. On the other hand, . B is positive
semi-definite so
.0 ≤ (Bu, u) = (−Mu, u) ≤ 0

because
1
(Mu, u) =
. ((M +t M)u, u) ≥ 0.
2

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 395

Hence, .(Bu, u) = 0. Proving . Bu = 0. So, . Mu = 0 and

.(B − M)u = 0.

It follows that

. K er (B − M) + K er (B + M) = K er (B − M) = R p .

Thus, . B = M.
Likewise, for .(ii). Q.E.D.

7.5 Examples: Choice of the Matrix . M

It is always possible to choose a . M matrix satisfying the admissibility conditions.

Indeed, we can write . B = Q D t Q, with . Q being orthogonal and . D being a diagonal
matrix. We choose
. M = Q|D| Q,
t
(7.5.1)

where .|D| is the diagonal matrix whose elements are the absolute values of the
elements of . D. We check that

. K er (B − M) + K er (B + M) = R p .

When . B is semi-definite, either positive or negative, this is the only possible choice
according to Lemma 7.4.7. In other cases, there are other possible choices.

Example 7.5.1 We consider the problem

du
. + σ u = f on ]0, 1[,
dx

with.σ > 0. This system is positive (with.α = σ ). We have. B(0) = −1 and. B(1) = 1,
so . M(0) = M(1). We do not impose a boundary condition for .x = 1, and we impose
one for .x = 0 (for example .u = g).

Example 7.5.2 Let.Ω =]0, 1[. We consider the problem (one-dimensional transport
equation in plane symmetry [5])

du
. + σ u = f on ]0, 1[, (7.5.2)
dx

u(0) = 0,
. (7.5.3)

@seismicisolation
@seismicisolation
396 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

where.σ > 0. Equation (7.5.2) can be considered as a symmetric system of Friedrichs.

We have . B(0) = −1 and . B(1) = 1. By applying Lemma 7.4.7, he comes . M(0) =
1 = M(1). Hence, .(B − M)(0) = −2 and .(B − M)(1) = 0. We must, therefore,
impose a boundary condition for.x = 0 (condition (7.5.3)) and no boundary condition
for .x = 1. We can also replace condition (7.5.3) by .u(0) = g.

Example 7.5.3 Let .Ω =]0, 1[. We consider the problem

. − u ,, + u = f on ]0, 1[.

We pose .v = u , . Then, {
−u , + v = 0
−v, + u = f.
.

Hence, ( ) ( ) ( )( ) ( )
0 −1 ∂ u 10 u f
. + = .
−1 0 ∂x v 01 v 0

We pose ( )
0 −1
. A1 =
−1 0

and ( )
10
. A0 = .
01

On the other hand,

En
∂
.C(x) = A0 (x) + A∗0 (x) − (Ai (x)) = 2I.
i=1
∂ xi

Hence, the system .−u ,, + u = f is positive in the sense of Friedrichs. The matrix . B
is given by
. B = ν A1

with .ν is the components of the external normal vector. Let’s calculate . B(0). The
normal vector .ν in .0 has for component .ν = (−1, 0).
−
→
ν
−

0 1

So,

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 397

. B(0) = (−1)A1
( )
0 −1
= (−1)
−1 0
( )
01
= .
10

Let . M(0) be the matrix ( )

ab
. M(0) = .
cd

We have ( )
−a 1 − b
. (B − M)(0) =
1 − c −d

and ( )
a 1+b
(B + M)(0) =
. .
1+c d

We cannot have . K er (B − M) = {0} because otherwise . K er (B + M) = R2 . So,

. B = −M. However, . M + M positive semi-definite or negative, therefore, . B is not,
t

absurd. So, .dim K er (B − M) ≥ 1. Therefore, .det(B − M) = 0. Similarly, we can

show that .det(B + M) = 0. Thus,
{
ad − (1 − b)(1 − c) = 0
.
ad − (1 + b)(1 + c) = 0.

All calculation done we find .c = −b and .ad − 1 + b2 = 0. Gold,

( )
2a b + c
.M + M =
t
b + c 2d
( )
2a 0
=
0 2d
≥0.

Hence, .a ≥ 0 and .d ≥ 0. Gold, .ad = 1 − b2 ≥ 0. Proving .|b| ≤ 1. Hence, we finally

have relations .a ≥ 0, .d ≥ 0, .b + c = 0, .|b| ≤ 1, and .ad − 1 + b2 = 0. Let .(u, v) ∈
K er (B − M), then ( )( ) ( )
−a 1 − b u 0
. = .
1 − c −d v 0

This is equivalent to {
−au + (1 − b)v = 0
. (7.5.4)
(1 − c)u − dv = 0.

@seismicisolation
@seismicisolation
398 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

(i) Dirichlet boundary condition: .u(0) = 0. The system (7.5.4) becomes

.

{
(1 − b)v = 0
.
dv = 0.

Hence, the choice are .b = 1, .a ≥ 0, .c = −1, .d = 0 correspond to the problem of

Dirichlet.
(ii) Neumann boundary condition: .v = 0. The system (7.5.4) becomes
.

{
−au = 0
.
(1 − c)u = 0.

Hence, the choice are.b = −1,.d ≥ 0,.a = 0,.c = 1 correspond of Neumann problem.
1−b2
(iii) Mixed boundary condition: The choice .−1 < b < 1, .a > 0, .d =
.
a
, .c = −b
correspond to the problem Mixed.
We proceed in the same way for .x = 1.

Remark 7.5.1 1. For the problem of Neumann we only took .b = −1 because .b = 1

is not suitable. In fact, if .b = 1, then .c = −1 and .−2u = 0. Hence, .u = 0 absurd
because .u /= 0.
2. In the two cases considered, Dirichlet boundary condition or Neumann boundary
condition, we can notice that we have . K er (B − M) = K er (B +t M) for .x = 0 and
. x = 1. ♦

Example 7.5.4 Let .Ω =]0, 1[. We consider the problem

d 2u
. − = f on ]0, 1[.
dx2

We pose .v = u , . Then, {
−v, = f
−u , + v = 0.
.

Hence, ( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + = . (7.5.5)
−1 0 ∂x v 01 v 0

We pose ( )
0 −1
. A1 =
−1 0

and ( )
00
. A0 = .
01

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 399

Further,

∂
.C(x) = A0 (x) + A∗0 (x) − (A1 (x))
( ) ∂x
00
= .
02

Hence, .C is not positive in the sense of Friedrichs. To make this system positive in
the sense of Friedrichs, we multiply by a matrix of the form
( )
ab
. .
cd

Thus,
( )( ) ( ) ( )( )( ) ( )( )
ab 0 −1 ∂ u ab 00 u ab f
. + = .
cd −1 0 ∂x v cd 01 v cd 0

Which proves
( ) ( ) ( )( ) ( )
−b −a ∂ u 0b u af
. + = .
−d −c ∂x v 0d v cf

We pose ( )
−b −a
. A1 =
−d −c

and ( )
0b
. A0 = .
0d

We want the matrix . A1 to be symmetric. Hence, .a = d. Thus,

( ) ( )
d 0 b b, a ,
C = A0 + A∗0 − A1 = + .
d , c,
.
dx b 2d

So, ( )
b, b + a,
C= .
b + a c, + 2a
,
.

Hence, .C is positive definite if, and only if, .tr (C) > 0 and .det C > 0 (see Lemma
2.11.2). Again,
, ,
.b + c + 2a > 0 (7.5.6)

and
.b, (c, + 2a) − (b + a , )2 > 0. (7.5.7)

@seismicisolation
@seismicisolation
400 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

(i) Dirichlet boundary condition: .u(0) = u(1) = 0. We have

.

( ) ( )
ba αβ
. B(0) = , M(0) = .
ac γ δ

Hence, ( )
b−α a−β
. (B − M)(0) =
a−γ c−δ

and ( )
b+α a+β
(B + M)(0) =
. .
a+γ c+δ

We want ( )
u(0)
.(B − M)(0) =0
v(0)

if, and only if, .u(0) = 0. Hence, .a = β and .c = δ. So,

( )
b + α 2a
.(B + M)(0) = .
a + γ 2c

If . B(0) is not positive semi-definite, then .dim K er (B − M)(0) ≥ 1. Hence, .det(B −

M)(0) = 0. If. B(0) is not negative semi-definite, then .dim K er (B + M)(0) ≥ 1. So,
.det(B + M)(0) = 0. Thus,

2c(b + α) − 2a(a + γ ) = 0.
. (7.5.8)

We want that . M(0) + M ∗ (0) to be positive semi-definite. In fact,

( )
2α a + γ
. M(0) + M ∗ (0) =
a + γ 2c

is positive semi-definite if
2α + 2c ≥ 0
. (7.5.9)

and
4αc − (a + γ )2 ≥ 0.
. (7.5.10)

Hence, .α + c ≥ 0 and .αc ≥ 0. Proving .α ≥ 0 and .c ≥ 0. We choose for example

( ) ( )
ab 1 ϕ(x)
. = .
cd 0 1

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 401

Equation (7.5.6) gives .ϕ , + 2 > 0. Equation (7.5.7) gives .ϕ , (0 + 2) + ϕ 2 > 0, so

,
.2ϕ − ϕ > 0. Equation (7.5.8) gives .−2(1 + γ ) = 0. Hence, .γ = −1, .β = 1 and
2

.δ = 0. Equation (7.5.9) gives .α ≥ 0. Equation (7.5.10) gives .1 + γ = 0, so .γ = −1.

Thus, ( ) ( ) ( )( ) ( )
−ϕ −1 ∂ u 0ϕ u f
. + = .
−1 0 ∂x v 01 v 0

We have ( )
−ϕ(1) −1
. B(1) =
−1 0

and ( )
α1 β1
. M(1) = .
γ 1 δ1

Hence, ( )
−ϕ(1) − α1 −1 − β1
. (B − M)(1) =
−1 − γ1 −δ1

and ( )
−ϕ(1) + α1 −1 + β1
(B + M)(1) =
. .
−1 + γ1 δ1

On the other hand, .detB(1) = −1, so the eigenvalues are opposed. Hence,
B(1) is neither positive semi-definite ni negative semi-definite. So, .dim K er (B −
.
M)(1) ≥ 1 and .dim K er (B + M)(1) ≥ 1. Hence, .det(B − M)(1) = 0 and .det(B +
M)(1) = 0. Thus,
.δ1 (α1 + ϕ(1)) − (1 + γ1 )(1 + β1 ) = 0 (7.5.11)

and
. 1 δ (α1 − ϕ(1)) − (−1 + γ1 )(−1 + β1 ) = 0. (7.5.12)

Since we have the Dirichlet boundary conditions, the second column of .(B − M)(1)
is zero. So, .−1 − β1 = 0 and .−δ1 = 0. Therefore, .β1 = −1 and .δ1 = 0. Equation
(7.5.11) gives .−(1 + γ1 ) × 0 = 0. Similarly Eq. (7.5.12) gives .−(γ1 − 1)(−2) = 0,
so .γ1 = 1. We want . M(1) + M ∗ (1) to be positive semi-definite. Indeed,
( )
∗ 2α1 γ1 + β1
. M(1) + M (1) =
γ1 + β1 2δ1
( )
2α1 0
=
0 0

is positive semi-definite if .2α1 ≥ 0, so .α1 ≥ 0. Hence,

( ) ( )
α1 −1 −ϕ(1) − α1 0
. M(1) = , (B − M)(1) = ,
1 0 −2 0

@seismicisolation
@seismicisolation
402 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

with .α1 ≥ 0. .ϕ(0) and .ϕ(1) are arbitrary. We take for example

ϕ(0)
. ϕ(x) =
1 − ϕ(0)x

with .0 < ϕ(0) < 1. We have

(ϕ(0))2
. ϕ , (x) =
(1 − ϕ(0)x)2
= (ϕ(x))2 .

Hence,
2ϕ , (x) − ϕ 2 (x) = ϕ 2 (x) > 0.
.

(ii) Neumann boundary condition: .u , (0) = u , (1) = 0. For

.

( ) ( )
ab 1 ϕ(x)
. = .
ca 0 1

Hence,
( ) ( )
ϕ(0) − α 1 − β ϕ(0) + α 1 + β
(B − M)(0) =
. , (B + M)(0) = .
1−γ −δ 1+γ δ

Since we have the Neumann boundary conditions, the first column of .(B − M)(0)
is zero. So, .α = ϕ(0) and .γ = 1. On the other hand, .detB(0) = −1, so . B(0) is
neither positive semi-definite nor negative semi-definite. So, we have .dim K er (B −
M)(0) ≥ 1 and .dim K er (B + M)(0) ≥ 1. Therefore, .det(B − M)(0) = 0 and .det
(B + M)(0) = 0. Thus,

. − δ(ϕ(0) − α) − (1 − β)(1 − γ ) = 0

and
δ(ϕ(0) + α) − (1 + β)(1 + γ ) = 0.
.

Which proves .2δα − 2(1 + β) = 0. Hence, .δα = 1 + β. The positivity of .(M +t

M)(0) entails that .α + δ ≥ 0 and .4αδ − (β + γ )2 ≥ 0. Further,
( )
−ϕ(1) − α1 −1 − β1
(B − M)(1) =
.
−1 − γ1 −δ1

and ( )
−ϕ(1) + α1 −1 + β1
. (B + M)(1) = .
−1 + γ1 δ1

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 403

Since we have the Neumann boundary conditions, the first column of .(B − M)(1)
is zero. Hence, .α1 = −ϕ(1) and .γ1 = −1. So,

. dim K er (B − M)(1) ≥ 1.

Proving.det(B − M)(1) = 0 and so.δ1 (α1 + ϕ(1)) − (1 + γ1 )(1 + β1 ) = 0. We have

detB(1) = −1, so . B(1) is neither positive semi-definite ni negative semi-definite.
.

Hence, .dim K er (B + M)(1) ≥ 1. So, .det(B + M)(1) = 0. Thus,

δ (α1 − ϕ(1)) − (−1 + γ1 )(−1 + β1 ) = 0.

. 1

Which proves .−δ1 ϕ(1) + β1 − 1 = 0. The positivity of .(M +t M)(1) gives us

.α1 + δ1 ≥ 0 and.4α1 δ1 − (β1 + γ1 ) ≥ 0. Hence,.α1 δ1 ≥ 0. Thus,.α1 ≥ 0 and.δ1 ≥ 0.
2

Hence, .ϕ(0) ≥ 0, .ϕ(1) ≤ 0, .ϕ /= 0 and .2ϕ , > ϕ 2 . Hence, .ϕ is strictly increasing, this
is absurd because .ϕ(0) ≥ 0 and .ϕ(1) ≤ 0. So, Neumann is not applicable. In fact,
suppose that the problem
,,
. − u = f on ]0, 1[

and
u , (0) = u , (1) = 0
.

admits a solution .u ∈ H 2 (]0, 1[). Hence,

{ 1
. − u , (1) + u , (0) = f (x) d x = 0.
0

So, this system only admits a solution if

{ 1
. f (x) d x = 0.
0

Example 7.5.5 Let .Ω =]0, 1×]0, T [. We consider the problem

⎧
⎪ ∂u ∂ 2u
⎨ − 2 = f 0 < x < 1, 0 < t < T
∂t ∂x
⎩ u(0, t) = u(1, t) = 0
.
⎪
u(x, 0) = u 0 (x).

We pose
∂u
. = v.
∂x
Again,
∂u
v−
. = 0.
∂x

@seismicisolation
@seismicisolation
404 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Hence,
∂u ∂v
. − = f.
∂t ∂x
Thus,
( ) ( ) ( ) ( ) ( )( ) ( )
10 ∂ u 0 −1 ∂ u 00 u f
. + + = .
00 ∂t v −1 0 ∂x v 01 v 0

This symmetric system does not satisfy the conditions of positivity (7.1.9). However,
if we pose {
u = eμt u 1 μ > 0
v = eμt v1
.

the symmetric system obtained is positive

( ) ( ) ( ) ( ) ( )( ) ( )
10 ∂ u1 0 −1 ∂ u1 μ0 u1 e−μt f
. + + = .
00 ∂t v1 −1 0 ∂x v1 01 v1 0

We have ( )
−1 0
. B(x, 0) = .
0 0

and ( )
10
. B(x, T ) = .
00

By using Lemma 7.4.7, we have

( )
10
. M(x, 0) = M(x, T ) = ,
00

which allows us to find the fact that we must give ourselves an initial condition.

Example 7.5.6 Let .Ω =]0, 1[. We consider the problem

d 2u
. − = f on ]0, 1[,
dx2

u(0) = u , (1) = 0.
.

We pose .v = u , . Then, {
−v, = f
(7.5.13)
−u , + v = 0.
.

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 405

Hence, ( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + = .
−1 0 ∂x v 01 v 0

We pose ( )
0 −1
. A1 =
−1 0

and ( )
00
. A0 = .
01

On the other hand,

∂
C(x) = A0 (x) + A∗0 (x) −
. (A1 (x))
( ) ∂ x
00
= .
02

So, .C is not positive in Friedrichs sense. To make this system positive within the
meaning of Friedrichs we multiply by a matrix of the form
( )
1 ϕ(x)
. .
0 1

Hence, ( ) ( ) ( )( ) ( )
−ϕ −1 ∂ u 0ϕ u f
. + = .
−1 0 ∂x v 01 v 0

We have ( )
ϕ(0) 1
. B(0) =
1 0

and ( )
αβ
. M(0) = .
γ δ

So, ( )
ϕ(0) − α 1 − β
. (B − M)(0) = .
1−γ −δ

We want ( )
u(0)
.(B − M)(0) =0
u , (0)

if, and only if, .u(0) = 0. Hence, .β = 1 and .δ = 0. On the other hand,

@seismicisolation
@seismicisolation
406 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
( )
ϕ(0) + α 2
(B + M)(0) =
. .
1+γ 0

. B(0) is not positive semi-definite so .dim K er (B − M)(0) ≥ 1. So,

det(B − M)(0) = 0.
.

. B(0) is not negative semi-definite so .dim K er (B + M)(0) ≥ 1. Thus,

det(B + M)(0) = 0.
.

Hence,.2(1 + γ ) = 0. Proving.γ = −1. We want. M(0) + M ∗ (0) to be positive semi-

definite. Hence, ( )
2α 0
.
0 0

either positive semi-definite. So, .2α + 0 ≥ 0 and .0 ≥ 0. Thus, .α ≥ 0. Then,

( )
α 1
. M(0) =
−1 0

with .α ≥ 0. On the other hand,

( )
−ϕ(1) −1
. B(1) =
−1 0

and ( )
α1 β1
. M(1) = .
γ 1 δ1

Hence, ( )
−ϕ(1) − α1 −1 − β1
(B − M)(1) =
. .
−1 − γ1 −δ1

We want ( )
u(1)
.(B − M)(1) =0
u , (1)

if, and only if, .u , (1) = 0. Hence, .α1 = −ϕ(1) and .γ1 = −1. On the other hand,
( )
−2ϕ(1) −1 + β1
(B + M)(1) =
. .
−2 δ1

. B(1) is not negative semi-definite, so .dim K er (B + M)(1) ≥ 1. Thus,

. det(B + M)(1) = 0.

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 407

Proving .−2δ1 ϕ(1) + 2(−1 + β1 ) = 0. Thus, .β1 = δ1 ϕ(1) + 1. The positivity of

(M + M ∗ )(1),
.
( )
∗ −2ϕ(1) −1 + β1
.(M + M )(1) =
−1 − β1 2δ1

gives us .−ϕ(1) + δ1 ≥ 0 and .−4ϕ(1)δ1 − (−1 + β1 )2 ≥ 0. Proving .−ϕ(1)δ1 ≥ 0.

Hence, .−ϕ(1) ≥ 0 and .δ1 ≥ 0. Thus, .ϕ(1) ≤ 0 and .δ1 ≥ 0. Hence,
( )
−ϕ(1) δ1 ϕ(1)
. M(1) =
−1 δ1

with .ϕ(1) ≤ 0 and .δ1 ≥ 0. So, .ϕ(0) is any and .ϕ(1) ≤ 0. We take for example

ϕ(0)
ϕ(x) =
.
1 − ϕ(0)x
ϕ(0)
with .ϕ(0) < 0 and .ϕ(1) = 1−ϕ(0)
< 0. We have

(ϕ(0))2
ϕ , (x) =
.
(1 − ϕ(0)x)2
= (ϕ(x))2 .

Hence,
. 2ϕ , (x) − ϕ 2 (x) = ϕ 2 (x) > 0.

So, the system multiplied by a matrix

( )
1 ϕ(x)
.
0 1

is positive within the meaning of Friedrichs. Hence, this system has a unique solution
(u, v).
.

Remark 7.5.2 If we multiply by

( )
1 ϕ(x)
. ,
0 1

the Neumann does not work.

Question: If we multiply by
( )
ab
.
cd

does Neumann work.

@seismicisolation
@seismicisolation
408 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

In fact,
d 2u
. − = f on ]0, 1[,
dx2

u , (0) = u , (1) = 0.
.

We pose .v = u , . Then, {
−v, = f
−u , + v = 0.
.

Hence, ( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + = .
−1 0 ∂x v 01 v 0

We multiply by ( )
ab
.
cd

we obtain ( ) ( ) ( )( ) ( )
−b −a ∂ u 0b u af
. + = .
−d −c ∂x v 0d v cf

We pose ( )
−b −a
. A1 =
−d −c

and ( )
0b
. A0 = .
0d

We want the matrix . A1 to be symmetric, so .a = d. Gold,

∂
C(x) = A0 (x) + A∗0 (x) −
. (A1 (x))
( ∂
) x
b, b + a,
= .
b + a c, + 2a
,

C(x) is positive definite if, and only if, .b, + c, + 2a > 0 and .b, (c, + 2a) − (b +
.
a , )2 > 0 (see Lemma 2.11.2). On the other hand,
( ) ( )
ba αβ
. B(0) = , M(0) = .
ac γ δ

Hence, ( )
b−α a−β
(B − M)(0) =
.
a−γ c−δ

@seismicisolation
@seismicisolation
7.5 Examples: Choice of the Matrix M 409

and ( )
b+α a+β
(B + M)(0) =
. .
a+γ c+δ

We want ( )
u(0)
(B − M)(0) =0
u , (0)
.

if, and only if, .u , (0) = 0. Since we have the Neumann boundary conditions, the
first column of .(B − M)(0) is zero. So, .b = α and .a = γ . . B(0) is not negative
semi-definite. So, .dim K er (B + M)(0) ≥ 1. Hence,

det(B + M)(0) = 0.
.

Again, .(b + α)(c + δ) − (a + β)(a + γ ) = 0 and .2b(c + δ) − 2a(a + β) = 0. On

the other hand, ( ) ( )
−b −a α1 β1
. B(1) = , M(1) = .
−a −c γ 1 δ1

Hence, ( )
−b − α1 −a − β1
. (B − M)(1) =
−a − γ1 −c − δ1

and ( )
−b + α1 −a + β1
(B + M)(1) =
. .
−a + γ1 −c + δ1

Since we have the Neumann boundary conditions, the first column of .(B − M)(1) is
zero. So,.α1 = −b and.γ1 = −a.. B(1) is not negative semi-definite. So,.dim K er (B +
M)(1) ≥ 1. Hence, .det(B + M)(1) = 0. Gold, .−2b(−c + δ1 ) + 2a(−a + β1 ) = 0.
We take ( ) ( )
ab 1b
. = .
cd 11

Hence, ( )
0 −1
. M(0) =
1 δ

with .δ any and ( )

−1 δ1
. M(1) =
−1 δ1

with .δ1 arbitrary. ♦

@seismicisolation
@seismicisolation
410 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

7.6 Some Examples in Physical Problems

7.6.1 Wave Equation in Dimension 1

We consider the problem

∂ 2u ∂ 2u
. − 2 = f. (7.6.1)
∂t 2 ∂x
Let’s pose
∂u
v=
.
∂t
and
∂u
.w= .
∂x
Hence,
∂v ∂w
. − = f.
∂t ∂x
So, (7.6.1) gives ⎧
⎪ ∂v ∂w
⎪
⎪ − = f
⎪
⎨ ∂t ∂x
∂u
. −v =0
⎪ ∂t
⎪
⎪
⎩ − ∂u + w = 0.
⎪
∂x
Thus,
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
01 0 u 0 0 −1 u 0 0 0 u f
∂ ∂
.⎝1 0 0⎠ ⎝ v ⎠ + ⎝ 0 0 0 ⎠ ⎝ v ⎠ + ⎝ 0 −1 0 ⎠ ⎝ v ⎠ = ⎝ 0 ⎠ .
00 0 ∂t w −1 0 0 ∂x w 0 0 1 w 0

So, we are faced with a first-order system with symmetric matrices. We can work
with .2 × 2 matrices instead of .3 × 3 matrices. Indeed,
( ) ( ) ( ) ( ) ( )( ) ( )
10 ∂ v 0 −1 ∂ v 00 v f
. + + =
0 1 ∂t w −1 0 ∂x w 0 0 w 0

if, and only if, ⎧

⎪ ∂v ∂w
⎨ − = f
∂t ∂x
.
⎪ ∂v ∂w
⎩− + = 0.
∂x ∂t

@seismicisolation
@seismicisolation
7.6 Some Examples in Physical Problems 411

Remark 7.6.1 We have chosen the second equation in this way (instead of this form
− ∂w
.
∂t
+ ∂∂vx = 0) to get the second symmetric matrix. ♦

7.6.2 Laplacian Operator

On .Ω ⊂ R2 , we consider the problem

. − Δu = f. (7.6.2)

Let’s pose
∂u
v=
.
∂x
and
∂u
w=
. .
∂y

Hence,
( )
∂ 2u ∂ 2u
. − Δu = − + 2
∂x 2 ∂y
∂v ∂w
=− −
∂x ∂y
= f.

So, (7.6.2) gives ⎧

⎪ ∂u
⎪
⎪ − +v =0
⎪
⎨ ∂∂ux
⎪
. − +w =0
⎪
⎪ ∂y
⎪
⎪ ∂v ∂w
⎪
⎩− − = f.
∂x ∂y

Thus,
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
0 −1 0 u 0 0 −1 u 000 u f
∂ ∂
. ⎝ −1 0 0 ⎠ ⎝ v ⎠+⎝ 0 0 0 ⎠ ⎝ v ⎠ + ⎝0 1 0⎠⎝ v ⎠ = ⎝ 0 ⎠.
∂x ∂y
0 0 0 w −1 0 0 w 001 w 0

So, we are faced with a first-order system with symmetric matrices. We can work
with .2 × 2 matrices instead of .3 × 3 matrices. Indeed,
( ) ( ) ( ) ( ) ( )( ) ( )
−1 0 ∂ v 0 −1 ∂ v 00 v f
. + + =
0 1 ∂x w −1 0 ∂y w 00 w 0

@seismicisolation
@seismicisolation
412 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

if, and only if,

⎧
⎪ − ∂v − ∂w = f
⎨
∂x ∂y
.
⎪ ∂w ∂v
⎩ − = 0.
∂x ∂y

Remark 7.6.2 We have chosen the second equation in this way (instead of this form
− ∂w
.
∂x
+ ∂∂vy = 0) to get the second symmetric matrix. ♦

7.6.3 Heat Equation

Let .Ω =]0, 1×]0, T [. We consider the problem

⎧
⎪ ∂u ∂ 2u
⎨ − a 2 = f 0 < x < 1, 0 < t < T
∂t ∂x
⎩ u(0, t) = u(1, t) = 0
.
⎪
u(x, 0) = u 0 (x).

We pose
∂u
a
. = av.
∂x
Again,
∂u
av − a
. = 0.
∂x
Hence,
∂u ∂v
. −a = f.
∂t ∂x
Thus,
( ) ( ) ( ) ( ) ( )( ) ( )
10 ∂ u 0 −a ∂ u 00 u f
. + + = .
00 ∂t v −a 0 ∂x v 0a v 0

We pose ( )
10
. A1 =
00
( )
0 −a
. A2 =
−a 0

and

@seismicisolation
@seismicisolation
7.6 Some Examples in Physical Problems 413
( )
00
. A0 = .
0a

C(x) is given by
.
( )
0 0
C(x) =
. .
0 2a

C(x) is not defined positive. If .a is not constant, then

.

⎞ ⎛
∂a
⎜ 0 ∂x ⎟
.C(x) = ⎝ ⎠.
∂a
2a
∂x

C(x) is not defined positive. So, to make this system positive, we set the change of
.
function {
u = eμt u 1 μ > 0
v = eμt v1 .
.

Then,
∂u
e−μt a
. = e−μt av.
∂x
Hence,
∂u 1
a
. = av1 .
∂x
Proving ( )
∂u ∂v
e−μt
. −a = f e−μt
∂t ∂x

and
∂u ∂u 1
e−μt
. = μu 1 +
∂t ∂t

because .u = eμt u 1 . Hence,

⎧
⎪ ∂u 1 ∂v1
⎪ −μt
⎨ μu 1 + ∂t − a ∂ x = f e
⎪
∂u
u = eμt u 1 , −a + av1 = 0
.
⎪
⎪
⎪
⎩ ∂ x
v = eμt v1 .

Thus,
( ) ( ) ( ) ( ) ( )( ) ( )
10 ∂ u1 0 −a ∂ u1 μ0 u1 e−μt f
. + + = .
00 ∂t v1 −a 0 ∂x v1 0a v1 0

@seismicisolation
@seismicisolation
414 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

On the other hand, ( )

−1 0
. B(x, 0) = .
0 0

. B(x, 0) is negative semi-definite, so the only choice of . M is

. M(x, 0) = −B(x, 0)
( )
10
= .
00

1 x

Hence, ( )
−2 0
(B − M)(x, 0) =
. .
0 0

So, we cannot impose initial conditions on .u because the first column of .(B −
M)(x, 0) is not zero. Furthermore,
( )
10
. (B − M)(x, T ) =
00

is positive semi-definite. Hence, the only choice of . M is

. M(x, T ) = B(x, T ).

On the other hand, ( )

00
(B − M)(x, T ) =
. .
00

Hence, nothing is imposed on .T because the Heat equation is retrograde. On the

other hand, ( ) ( )
0a γ a
. B(0, t) = , M(0, t) =
a0 −a 0

@seismicisolation
@seismicisolation
7.6 Some Examples in Physical Problems 415

with .γ ≥ 0. Hence,
( ) ( )
−γ 0 γ 2a
(B − M)(0, t) =
. , (B + M)(0, t) = .
2a 0 0 0

We have
. K er (B − M)(0, t) = {(α, β) ∈ R2 such that α = 0}

and
. K er (B + M)(0, t) = {(α, β) ∈ R2 such that γ α + 2aβ = 0}.

Gold,
. K er (B − M)(0, t) + K er (B + M)(0, t) = R2 .

Hence, ( ) ( )
0 −a γ −a
. B(1, t) = , M(1, t) = .
−a 0 a 0

So, ( ) ( )
−γ 0 γ −2a
.(B − M)(1, t) = , (B + M)(1, t) = .
−2a 0 0 0

Gold,
. K er (B − M)(1, t) + K er (B + M)(1, t) = R2 .

Hence, we can impose Dirichlet or Neumann.

∂T ∂T
∂x
=0 ∂x
=0

If . f = 0, then
{ 1
. u(x, t) → cte = u(x, 0) d x when t → ∞.
0

Question: Can we do Friedrichs without a change of function?

@seismicisolation
@seismicisolation
416 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Liquid aluminum

Magnetic field

Coupling
for not moving away from liquid aluminum

Solid

7.6.4 Wave Equation

Let .Ω = ]0, 1[×]0, T [, .T > 0, we consider the problem

∂ 2u ∂ 2u
−. = f 0 < x < 1, 0 < t < T
∂t 2 ∂x2
u(0, t) = u(1, t) = 0
u(x, 0) = u 0 (x) (7.6.3)
∂u
(x,. 0) = u 1 (x). (7.6.4)
∂t
We pose
∂u ∂u
. v= , w= .
∂t ∂x
Hence,
∂v ∂w
. − = f.
∂t ∂x
We will assume that the cross derivatives are equal. We have

∂v ∂w
. = .
∂x ∂t

@seismicisolation
@seismicisolation
7.6 Some Examples in Physical Problems 417

So,
∂v ∂w
. − + = 0.
∂x ∂t
Thus, ( ) ( ) ( ) ( ) ( )
10 ∂ v 0 −1 ∂ v f
. + = . (7.6.5)
01 ∂t w −1 0 ∂x w 0

We pose ( )
10
. A1 =
01

and ( )
0 −1
. A2 = .
−1 0

We have
C(x) = (0).
.

C(x) is not positive. Let’s pose .v = eμt v1 and .w = eμt w1 . Hence,

.

∂v ∂v1
. = μeμt v1 + eμt
∂t ∂t
and
∂w ∂w1
. = μeμt w1 + eμt .
∂t ∂t
Proving
∂v1 ∂w1
. + μv1 − = e−μt f
∂t ∂x
and
∂w1 ∂v1
. + μw1 − = 0.
∂t ∂x
Thus,
( ) ( ) ( ) ( ) ( ) ( ) ( −μt )
10 ∂ v1 0 −1 ∂ v1 μ0 v1 e f
. + + = .
0 1 ∂t w1 −1 0 ∂x w1 0 μ w1 0

We have ( ) ( )
−1 0 10
. B(x, 0) = and B(x, T ) = .
0 −1 01

Then, by using Lemma 7.4.7, we have

@seismicisolation
@seismicisolation
418 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

. M(x, 0) = M(x, T )
( )
10
= .
01

We thus find the initial conditions (7.6.3) and (7.6.4). For .x = 0 and .x = 1, we use
the matrices given in the previous examples. The conditions, for the limit .x = 0, that
can be taken into account in an admissible way are of the form

. − av + (1 − b)w = 0 (7.6.6)

with .a ≥ 0 and .|b| ≤ 1.

Remark 7.6.3 Characteristics method. The system (7.6.5) can still be written (when
f = 0)
.
∂ ∂
. (v + w) − (v + w) = 0
∂t ∂x
∂ ∂
. (v − w) + (v − w) = 0.
∂t ∂x

So, the function .v + w (resp. .v − w) is constant on the characteristic directions . ddtx =

−1 (resp. . ddtx = 1).
t
te
v+

cs
=
w

w
=

v−
cs
te

x =1 x

For a point such as . M the solution is known from the initial data in . P and . Q. For
a point such as . N the solution is known from the values .(v − w)(R) and .(v + w)(S).
However, we know

@seismicisolation
@seismicisolation
7.6 Some Examples in Physical Problems 419

. (v + w)(R) = (v + w)(Q)

and
(v − w)(S) = (v − w)(P).
.

The solution in . N will be known if we impose boundary conditions of the type

v(0, t) + αw(0, t) = 0, α /= 1
. (7.6.7)

v(1, t) + βw(1, t) = 0, β /= −1.

.

It can be seen that the condition (7.6.6) is more restrictive than the condition (7.6.7),
so the boundary conditions of the characteristic type are not all necessarily taken
into account in an admissible way in the Friedrichs framework. ♦

Remark 7.6.4 The general problem of taking into account, in an admissible way,
of any type of boundary conditions is, it seems, always open. ♦

7.6.5 Tricomi Equation

Consider the Tricomi equation

∂ 2u ∂ 2u
. y − = f.
∂x2 ∂ y2

We pose
∂u ∂u
v=
. , w= .
∂x ∂y

We have ( ) ( ) ( ) ( ) ( )
y0 ∂ v 0 −1 ∂ v f
. + = .
01 ∂x w −1 0 ∂y w 0

Since
E
2
∂ Ai
. A0 + A∗0 − = 0,
i=1
∂ xi

the system is not positive in the sense of Friedrichs. To make the system positive, we
premultiply by the matrix ( )
b cy
. .
c b

@seismicisolation
@seismicisolation
420 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

We get the symmetric system

( ) ( ) ( ) ( ) ( )
by cy ∂ϕ v −cy −b ∂ϕ v bf
. + =
cy b ∂x w −b −c ∂ y w cf

and ⎛ ( ) ⎞
∂c ∂b ∂b ∂c
E2
∂ A ⎜ y − +c −y ⎟
∗
. A0 + A0 −
i
=⎜ ∂y ∂x ∂y ∂x ⎟.
∂ x ⎝ ∂b ∂c ∂b ∂c ⎠
i=1 i −y − +
∂y ∂x ∂x ∂y

If we choose for example .b = −b0 − b1 x, .b1 > 0, .c = c0 , with .c0 + b1 y > 0 for all
y such that .(x, y) ∈ Ω, the matrix
.

E
2 ( )
∂ Ai c0 + b1 y 0
. A0 + A∗0 − =
i=1
∂ xi 0 b1

is then positive definite. Depending on the different choices of functions .b(x, y) and
c(x, y), different types of boundary conditions can be taken into account. We can
.
successively consider the elliptical part and the hyperbolic part and we will consider
the case where a part of .∂Ω is characteristic (i.e., .νx d x + ν y dy with .dy = ± √d xy ).

7.6.6 Boundary Layer Problem

Boundary layer problem

∂q ∂q ∂q ∂ 2q
u
. +v +w − ν 2 = f, on Ω×]0, Z [,
∂x ∂y ∂ξ ∂ξ

∂u ∂v ∂w
where . + + = 0, .ν = cte > 0, .u > 0, and
∂x ∂y ∂ξ

q(x, y, 0) = q(x, y, Z ) = 0
.

with .q is given on .∂− Ω×]0, Z [, .Ω =]0, 1[×]0, 1[, and

∂ Ω = {(x, y) ∈ ∂Ω such that u n x + v n y < 0}.

. −

We pose
∂q
ν =ν
. p .
∂ξ

@seismicisolation
@seismicisolation
7.6 Some Examples in Physical Problems 421

We get the system

( ) ( ) ( ) ( ) ( ) ( ) ( )( ) ( )
u0 ∂ p v0 ∂ p w −ν ∂ p 00 p f
. + + + = .
00 ∂x q 0 0 ∂y q −ν 0 ∂ξ q 0ν q 0

The condition
∂u ∂v ∂w
. + + =0
∂x ∂y ∂ξ

implies that this system is not positive. We premultiply with a matrix of the form
( )
ab
.
0a

(for reasons of symmetry). We have

⎛ ( )⎞
∂a ∂a ∂a ∂b ∂a
E
3
∂ Ai ⎜ −u ∂ x − v ∂ y − w ∂ξ + ν ∂ξ ν b+
∂ξ ⎟
. A0 + A∗0 − =⎜
⎝
(
∂a
) ⎟.
⎠
i=1
∂ xi ν b+ 2aν
∂ξ

It is necessary to choose .a and .b so that this matrix is positive definite and also so
that the boundary conditions can be taken into account
( ) ( )
bν − aw aν α1 aν
.• For .ξ = 0, . B = , .M = , with .α1 ≥ 0.
aν 0 −aν 0
( ) ( )
aw − bν −aν α2 −aν
.• For .ξ = Z , . B = ,M = , with .α2 ≥ 0.
−aν 0 aν 0
( )
−au 0
.• For . x = 0, . B = .
0 0
( )
−av 0
.• For . y = 0, . B = .
0 0
( )
au 0
.• For . x = 1, . B = .
0 0
( )
av 0
.• For . y = 1, . B = .
0 0
The choice .a = 1 makes it possible to take into account the conditions .q given on
∂ Ω×]0, Z [. The choice .b = b(ξ ) = − 1+ξ
. −
1
, then guarantees positivity.

@seismicisolation
@seismicisolation
422 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

7.7 Approximation of Friedrichs Systems by Finite

Element Methods

7.7.1 Rayleigh Ritz Galerkin Method

In this paragraph, we define an approximate problem (7.1.2)–(7.1.3) in an abstract

way. For this, let .Ω be a bounded polyhedral open set of .Rn and let .Vh be a finite-
dimensional subspace of .(H 1 (Ω)) p , where .h is a parameter intended to tend to zero,
and let . X h be a subspace of .Vh . We will look for an approximate solution .u h ∈ X h
and show that if certain abstract hypotheses are satisfied, this approximate solution
converges, in a sense defined later, towards the exact solution .u of the problem
(7.1.2)–(7.1.3). The introduction of subspace. X h of.Vh can make it possible to impose,
possibly, on the approximate solution .u h to satisfy certain boundary conditions, for
example
{ { }
. X h = vh ∈ Vh such that ((B − M)vh , wh )2 ds = 0 for all wh ∈ Vh .
∂Ω
(7.7.1)
From the definition of a weak solution of the problem (7.1.2)–(7.1.3), we can naturally
define the following approximate problem: Find .u h ∈ X h such that

(u h , A∗ vh )(L 2 (Ω,R)) p = (F, vh )(L 2 (Ω,R)) p

. (7.7.2)

for all .vh ∈ X h∗ , where

{ { }
. X h∗ = vh ∈ Vh such that ((B +t M)vh , wh )2 ds = 0 for all wh ∈ Vh .
∂Ω
(7.7.3)

Remark 7.7.1 .(i) The problem (7.7.2)–(7.7.3) makes sense because the functions
v and .wh belong to .(H 1 (Ω)) p and, we can define their traces on .∂Ω in subspace
. h
1
.(H 2 (∂Ω)) of .(L (∂Ω)) .
p 2 p

(ii) There is not in general uniqueness of the solution for the problem (7.7.2).
.

(iii) In equality (7.7.2), we cannot necessarily choose .vh = u h , because we do not

.
have in general the inclusion

. X h ⊂ X h∗ .

♦
Problem (7.7.2) generally leads to the solution of a linear system for which the
number of unknowns is not equal to the number of equations. It is then difficult to

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 423

show that problem (7.7.2) is .Vh -elliptic and to prove the existence of an approximate
solution. We are, therefore, going to define an approximate problem for which we
seek the solution .u h in the same space as the functions .vh , which requires a particular
treatment of the boundary conditions.
Question: Is there existence, uniqueness of the solution of the problem (7.7.2) and
what is the error .u − u h .

7.7.2 Particular Case

If . X h = X h∗ , then
(u h , A∗ u h )(L 2 (Ω,R)) p = (F, u h )(L 2 (Ω,R)) p .
.

Hence,
{ {
1 1
. ((A + A∗ )u h , u h )2 d xd y − (Bu h , u h )2 ds = (F, u h )(L 2 (Ω,R)) p .
2 Ω 2 ∂Ω

Furthermore, . X h = X h∗ , so
{ {
1
. (t Mu h , u h )2 ds = ((M +t M)u h , u h )2 ds
∂Ω 2 ∂Ω
{
= (Bu h , u h )2 ds ≥ 0.
∂Ω

As . A is positive, we have

(u h , A∗ u h )(L 2 (Ω,R)) p ≥ α||u h ||2(L 2 (Ω,R)) p .

.

If the solution .u ∈ (H 1 (Ω)) p , then

. Au = F

in .(L 2 (Ω, R)) p and .(B − M)u = 0. Hence,

{
1 1 1
. (Au, v)(L 2 (Ω,R)) p + (u, A∗ v)(L 2 (Ω,R)) p + (Mu, v)2 ds = (F, v)(L 2 (Ω,R)) p
2 2 2 ∂Ω

for all .v ∈ (H 1 (Ω)) p if, and only if,

. Au = F, and (B − M)u = 0.

@seismicisolation
@seismicisolation
424 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

In fact, we multiply the equation . Au = F by .v and by integrating over .Ω, then using
Green’s formula, we find
{
1 1 1
. (Au, v)(L 2 (Ω,R)) p + (u, A∗ v)(L 2 (Ω,R)) p + (Bu, v)2 ds = (F, v)(L 2 (Ω,R)) p .
2 2 2 ∂Ω

Since .(B − M)u = 0, we have

{ {
1 1
. (Bu, v)2 ds = (Mu, v)2 ds.
2 ∂Ω 2 ∂Ω

So, we have the following.

Theorem 7.7.2 If .u ∈ (H 1 (Ω)) p is the solution of the problem (7.1.2)–(7.1.3), then
we can write for every .v ∈ (H 1 (Ω)) p ,
{
1 1 1
. (Au, v)(L 2 (Ω,R)) p + (u, A∗ v)(L 2 (Ω,R)) p + (Mu, v)2 ds = (F, v)(L 2 (Ω,R)) p .
2 2 2 ∂Ω
(7.7.4)
Equivalently, we can write for any .v ∈ (H 1 (Ω)) p ,
{
1
(Au, v)(L 2 (Ω,R)) p
. + ((M − B)u, v)2 ds = (F, v)(L 2 (Ω,R)) p
2 ∂Ω

and

{
1
(u, A∗ v)(L 2 (Ω,R)) p +
. (u, (B +t M)v)2 ds = (F, v)(L 2 (Ω,R)) p .
2 ∂Ω

Remark 7.7.3 For the three equivalent formulations, the functions .u and .v belong
to the same space, but boundary terms appear in the equations. ♦

Let .τh be a triangulation of .Ω, of class .C 0 , in finite elements of diameter less than
.h. We assume, throughout, that the family of triangulations .(τh )h is regular (see
Definition 4.1.2). This implies, in particular, that each finite element .(K , Σ K , PK ),
. K ∈ τh , is the range by an invertible isoparametric transformation . FK (affine when

the element . K is a .n-simplex or a parallelotope, or belonging to .(Q 1 )2 when the

element . K is a convex quadrilateral in dimension .n = 2) of a finite element with
-, Σ,
references .( K - If we, further, suppose that the inclusion
- P).

. P -, R)
- ⊂ C k+1 ( K

is satisfied for an integer .k ≥ 1, we, also, have the inclusion

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 425

. PK ⊂ C k+1 (K , R), for all K ∈ τh .

Finally, we can always assume that the Jacobian . JK of the transformation . FK is

strictly positive. We denote . JS the Jacobian of the restriction of the transformation
. FK to any face . S of the element . K ∈ τh . To the triangulation .τh , we associate . Vh of
.(H (Ω)) and, we consider the approximate problem: Find .u h ∈ Vh such that
1 p

{
1 1 1
. (Au h , vh )(L 2 (Ω,R)) p + (u h , A∗ vh )(L 2 (Ω,R)) p + (Mu h , vh )2 ds = (F, vh )(L 2 (Ω,R)) p
2 2 2 ∂Ω
(7.7.5)
for all .vh ∈ Vh . Now, we apply Green’s formula, we find
{ {
1 1 1 1
. (Au h , vh )(L 2 (Ω,R)) p + (Au h , vh )(L 2 (Ω,R)) p − (Bu h , vh )2 ds + (Mu h , vh )2 ds =
2 2 2 ∂Ω 2 ∂Ω
(F, vh )(L 2 (Ω,R)) p

for all .vh ∈ Vh . Proving

{ (( ) )
B−M
(Au h , vh )(L 2 (Ω,R)) p −
. u h , vh ds = (F, vh )(L 2 (Ω,R)) p .
∂Ω 2 2

If .vh = u h , we have
{
1 1
. ((A + A∗ )u h , u h )(L 2 (Ω,R)) p + (Mu h , u h )2 ds = (F, u h )(L 2 (Ω,R)) p .
2 2 ∂Ω

If . F = 0, then let us show .u h = 0. Indeed,

1
. ((A + A∗ )u h , u h )(L 2 (Ω,R)) p ≥ α||u h ||2(L 2 (Ω,R)) p
2
and {
. (Mu h , u h ) ds ≥ 0.
∂Ω

Hence,
. α||u h ||2(L 2 (Ω,R)) p ≤ ||F||(L 2 (Ω,R)) p ||u h ||(L 2 (Ω,R)) p .

So,
1
.||u h ||(L 2 (Ω,R)) p ≤ ||F||(L 2 (Ω,R)) p .
α
Thus, if . F = 0, then .u h = 0. Hence, we have the following.

Theorem 7.7.4 The approximate problem (7.7.5) has a unique solution .u h ∈ Vh

and, we have the upper bound

@seismicisolation
@seismicisolation
426 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

1
||u h ||(L 2 (Ω,R)) p ≤
. ||F||(L 2 (Ω,R)) p . (7.7.6)
α
♦
We have a formula equivalent to Eq. (7.7.5). Indeed, by applying Green’s formula to
Eq. (7.7.5), we find
{
∗ 1
(u h , A vh )(L 2 (Ω,R)) p
. + ((B + M)u h , vh )2 ds = (F, vh )(L 2 (Ω,R)) p .
2 ∂Ω

If we choose .vh ∈ X h∗ , we have

.(u h , A∗ vh )(L 2 (Ω,R)) p = (F, vh )(L 2 (Ω,R)) p .

In practice, the integrals occurring in (7.7.5) are rarely calculated exactly, either
because the matrices . Ai (x), .0 ≤ i ≤ n and . M(x), where the second member . F do
not have a simple analytical expression, either because these quantities are only
known (via physical measurements) at a certain number of points in the domain .Ω.
To calculate these integrals in an approximate way, we give ourselves:
-
(i) A quadrature formula on the reference finite element . K
.

{ E
N
. -
f (-
x ) d-
x∼ wl -
- f (-
bl ),
-
K l=1

wl ∈ R, .-
where .- -, .1 ≤ l ≤ N .
bl ∈ K
(ii) A quadrature formula on a face .-
. -
S of dimension .n − 1 of . K
{ E
L
. -
g (- s∼
s) d- - g (-
wl, - bl, ),
-
S l=1

where .- bl, ∈ -
wl, ∈ R, .- S, .1 ≤ l ≤ L.
Given . K ∈ τh and a face . S of dimension .n − 1 of . K , we have
{ {
. f (x) d x = - x ) JK (-
f (- x ) d-
x,
K -
K
{ {
. g(x) d x = -
g (-
s) JS (-
s) d-
s.
S -
S

In this way, we have

{ E
N
. f (x) d x ∼ wl,K f (bl,K ), (7.7.7)
K l=1

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 427

with .wl,K = JK (- wl , .bl,K = FK (-

bl )- bl ), .1 ≤ l ≤ N ,

{ E
L
. g(s) ds ∼ wl,S g(bl,S ), (7.7.8)
S l=1

with .wl,S = JS (- wl, , .bl,S = FK (-

bl, )- bl, ), .1 ≤ l ≤ L. In what follows, we will assume
that the weights .- wl (therefore also .wl,K , . K ∈ τh ), .1 ≤ l ≤ N and the weights .- wl,
(thus also .wl,K , . K ∈ τh ), .1 ≤ l ≤ L, are strictly positive.

7.7.3 Approximate Problem

When using the formulas (7.7.7) and (7.7.8) to calculate the integrals, the approximate
problem is formulated as follows: Find .u ∗h ∈ Vh , such that

1 E E
L
1 1
.(Au ∗h , vh )h + (u ∗h , A∗ vh )h + wl,S (Mu ∗h , vh )2 (bl,S ) = (F, vh )h ,
2 2 2 S⊂∂Ω l=1
(7.7.9)
for all .vh ∈ Vh , where

EE
N
(u, v)h =
. wl,K (u, v)2 (bl,K ).
K ∈τh l=1

The bilinear form

1 1
(u h , vh ) −→
. (Au h , vh )h + (u h , A∗ vh )h
2 2

has an obvious meaning when we have the inclusion .Vh ⊂ (C 1 (Ω, R)) p . When this
inclusion is not satisfied, the expression

1 1
. (Au h , vh )h + (u h , A∗ vh )h (7.7.10)
2 2
still makes sense because (7.7.10) represents a sum of integrals over each element
K , calculated using the quadrature formula (7.7.7).
.

The following problems arise: How to choose the quadrature formulas on the
- and on any face .-
finite element of reference . K - so that:
S of dimension .n − 1 of . K
(i) the problem (7.7.9) admits a unique solution .u ∗h ∈ Vh ,
.

(ii) . lim ||u ∗h − u||(L 2 (Ω,R)) p = 0,

.
h→0

@seismicisolation
@seismicisolation
428 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

(iii) the .u ∗h − u error is of the same order as the .u h − u error corresponding to the
.

finite element method (7.7.5), where the integrals are calculated exactly.
By first placing ourselves in an abstract framework, and, using certain general
hypotheses, we will obtain results of convergence and error bounding. We have
the following error bounds.

Theorem 7.7.5 (Lesaint [6]) Suppose there is a constant .α > 0 such that

1 1
. (Avh , vh )h + (vh , A∗ vh )h ≥ α||vh ||2(L 2 (Ω,R)) p (7.7.11)
2 2
and
1 E E
L
. wl,S (Mvh , vh )2 (bl,S ) ≥ 0, for all vh ∈ Vh . (7.7.12)
2 S⊂∂Ω l=1

Then, the solution .u ∗h of the problem (7.7.9) satisfies the inequality

⎧ ⎛ |{ |
⎪ | |
⎪
⎨ | ((B − M)(u − vh ), w)2 ds ||
⎜ |
.||u − u ∗h ||(L 2 (Ω,R)) p ≤ C inf ⎜ ⎝||u − vh ||(H 1 (Ω)) p + sup
∂Ω
+
⎪
⎪ v ∈V ||w||(L 2 (Ω,R)) p
⎩ h h w∈Vh

| |
|1 |
| (Avh , w)(L 2 (Ω,R)) p + 1 (vh , A∗ w)(L 2 (Ω,R)) p − 1 (Avh , w)h − 1 (vh , A∗ w)h |
|2 2 2 2 |
+ sup +
w∈Vh ||w||(L 2 (Ω,R)) p
| { |⎞
|1 1 E E
L |
| |
| (Bvh , w)2 ds − wl,S (Mvh , w)2 (bl,S )| ⎟
| 2 ∂Ω 2 |⎟
S⊂∂Ω l=1 ⎟
+ sup ⎟+
||w||(L 2 (Ω,R)) p ⎟
w∈Vh
⎠

| |⎫
| |⎪
| EE N
|⎪
|(F, w)(L 2 (Ω,R)) p − |⎪
| w (F, w) (b ) |⎪
⎪
|⎪
l,K 2 l,K
| K ∈τh l=1
⎬
+ sup , (7.7.13)
||w||(L 2 (Ω,R)) p ⎪
⎪
w∈Vh ⎪
⎪
⎪
⎪
⎭

where the constant .C only depends on .α, the matrices . Ai , .0 ≤ i ≤ n. If, we further
assume that, there is a constant .β such that

1 E E
L
. wl,S (Mvh , vh )2 (bl,S ) ≥ β||vh ||2(L 2 (∂Ω,R)) p , (7.7.14)
2 S⊂∂Ω l=1

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 429

then we have the increase

{
(
− u ∗h ||(L 2 (Ω,R)) p + ||u − u ∗h ||(L 2 (∂Ω,R)) p ≤ C inf ||u − vh ||(H 1 (Ω)) p + ||u − vh ||(L 2 (∂Ω,R)) p +
.||u
vh ∈Vh
| |
|1 |
| (Avh , w)(L 2 (Ω,R)) p + 1 (vh , A∗ w)(L 2 (Ω,R)) p − 1 (Avh , w)h − 1 (vh , A∗ w)h |
|2 2 2 2 |
+ sup +
w∈Vh ||w|| (L (Ω,R))
2 p
| { |⎞
|1 1 E E
L |
| |
| (Bvh , w)2 ds − wl,S (Mvh , w)2 (bl,S )| ⎟
| 2 ∂Ω 2 |⎟
S⊂∂Ω l=1 ⎟
+ sup ⎟+
||w||(L 2 (∂Ω,R)) p ⎟
w∈Vh
⎠

| |⎫
| EE |⎪
| N
|⎪
|(F, w)(L 2 (Ω,R)) p − |⎪
| w (F, w) (b )|⎪
⎪
|⎪
l,K 2 l,K
| K ∈τh l=1
⎬
+ sup , (7.7.15)
||w||(L 2 (Ω,R)) p ⎪
⎪
w∈Vh ⎪
⎪
⎪
⎪
⎭

where the constant .C only depends on .α, .β, and matrices . Ai , .0 ≤ i ≤ n. ♦

Proof The hypotheses (7.7.11) and (7.7.12) imply that the problem (7.7.9) has a
unique solution .u ∗h ∈ Vh . Now, consider any element .vh ∈ Vh and let .w = u h − vh .
We have by the equalities (7.7.4) and (7.7.9)

1 E E EE
L N
1 1
. (Aw, w)h + (w, A∗ w)h + wl,S (Mw, w)2 (bl,S ) = wl,K (F, w)2 (bl,K )−
2 2 2
S⊂∂Ω l=1 K ∈τh l=1
{
1
− (F, w)(L 2 (Ω,R)) p + (A(u − vh ), w)(L 2 (Ω,R)) p + ((M − B)(u − vh ), w)2 ds+
2 ∂Ω
1 1 1 1
+ (Avh , w)(L 2 (Ω,R)) p + (vh , A∗ w)(L 2 (Ω,R)) p − (Avh , w)h − (vh , A∗ w)h +
2 2 2 2
{
1 E E
L
1
+ (Bvh , w)2 ds − wl,S (Mvh , w)2 (bl,S ). (7.7.16)
2 ∂Ω 2
S⊂∂Ω l=1

By using the hypotheses (7.7.11) and (7.7.12), we can, therefore, write, for all
v ∈ Vh
. h

@seismicisolation
@seismicisolation
430 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
⎛ {
⎜ ((M − B)(u − vh ), w)2 ds
. ||w||(L 2 (Ω,R)) p ≤C⎜
⎝||u − vh ||(H 1 (Ω)) p + sup
∂Ω
+
w∈Vh ||w||(L 2 (Ω,R)) p
| |
|1 |
| (Avh , w)(L 2 (Ω,R)) p + 1 (vh , A∗ w)(L 2 (Ω,R)) p − 1 (Avh , w)h − 1 (vh , A∗ w)h |
|2 2 2 2 |
+ sup +
w∈Vh ||w||(L 2 (Ω,R)) p
| { |
|1 1 E E
L |
| |
| (Bvh , w)2 ds − wl,S (Mvh , w)2 (bl,S )|
| 2 ∂Ω 2 |
S⊂∂Ω l=1
+ sup +
w∈Vh ||w||(L 2 (∂Ω,R)) p
| |⎞
| |
| EE N
|
|(F, w)(L 2 (Ω,R)) p − w (F, w) (b ) |⎟
| l,K 2 l,K |⎟
| K ∈τh l=1 |⎟
+ sup ⎟.
||w||(L 2 (Ω,R)) p ⎟
w∈Vh ⎟
⎠

We easily deduce, thanks to the triangular inequality, the inequality (7.7.14). From
the equality (7.7.16) and the hypotheses (7.7.11) and (7.7.14), we can write, for all
.vh ∈ Vh

(
. ||w||(L 2 (Ω,R)) p + ||w||(L 2 (∂Ω,R)) p ≤ C ||u − vh ||(H 1 (Ω)) p + ||u − vh ||(L 2 (∂Ω,R)) p +
| |
|1 |
| (Avh , w)(L 2 (Ω,R)) p + 1 (vh , A∗ w)(L 2 (Ω,R)) p − 1 (Avh , w)h − 1 (vh , A∗ w)h |
|2 2 2 2 |
+ sup +
w∈Vh ||w|| (L (Ω,R))
2 p
| { |
|1 1 E E
L |
| |
| (Bvh , w)2 ds − wl,S (Mvh , w)2 (bl,S )|
| 2 ∂Ω 2 |
S⊂∂Ω l=1
+ sup +
w∈Vh ||w||(L 2 (∂Ω,R)) p
| |⎞
| |
| EE N
|
|(F, w)(L 2 (Ω,R)) p − w (F, w) (b ) |
2 l,K | ⎟
|
|⎟
l,K
| K ∈τh l=1 ⎟
+ sup ⎟.
||w||(L 2 (Ω,R)) p ⎟
w∈Vh ⎟
⎠

From the triangle inequality, we have

||u − u ∗h ||(L 2 (Ω,R)) p + ||u − u ∗h ||(L 2 (∂Ω,R)) p ≤ ||w||(L 2 (Ω,R)) p + ||w||(L 2 (∂Ω,R)) p +
.

||u − vh ||(L 2 (Ω,R)) p + ||u − vh ||(L 2 (∂Ω,R)) p .

The last two inequalities lead to the upper bound (7.7.15). Q.E.D.

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 431

Remark 7.7.6 The hypothesis (7.7.11) is satisfied if we can verify that, for all .vh ∈
Vh , we have

1 EE
N
. wl,K ((A + A∗ )v, v)2 (bl,K ) ≥ α||v||2(L 2 (Ω,R)) p .
2 K ∈τ l=1
h

For this, it suffices that for all . K ∈ τh and for all . p ∈ (PK ) p , we have

1E
N
. wl,K ((A + A∗ ) p, p)2 (bl,K ) ≥ α|| p||2(L 2 (K ,R)) p .
2 l=1

Similarly, the hypotheses (7.7.12) and (7.7.14) are satisfied if we can verify that for
all .v ∈ Vh , we have, respectively, the inequalities

E
L
. wl,S ((M + M ∗ )v, v)2 (bl,S ) ≥ 0,
l=1

and

1E
L
. wl,S ((M + M ∗ )v, v)2 (bl,S ) ≥ β||v||2(L 2 (S,R)) p .
4 l=1

From the inequality (7.7.13), it is possible to establish a general result of convergence

of the approximate solution .u h ∈ Vh towards the exact solution .u when .h tends to
zero. For this, we introduce the space
{ }
. - = u ∈ (H 1 (Ω)) p such that (B − M)u |∂Ω = 0 .
W

Theorem 7.7.7 (Lesaint [6]) We assume that the hypotheses (7.7.11) and (7.7.12)
are satisfied, for a constant .α > 0, independent of .h, and that the solution .u of the
problem (7.1.2)–(7.1.3) belongs to the space .(H 1 (Ω)) p . Suppose on the other hand
- , dense in .W
that there exists a subspace .H of .W - , and an mapping .rh : H −→ X h
such that

@seismicisolation
@seismicisolation
432 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

∀v ∈ H , lim ||v − rh v||(H 1 (Ω)) p = 0,

. (7.7.17)
h→0

{
|v − rh v||w| ds
∂Ω
.∀v ∈ H , lim sup = 0, (7.7.18)
h→0 w∈Vh ||w||(L 2 (Ω,R)) p
| |
|1 |
| (Arh v, w)(L 2 (Ω,R)) p + 1 (rh v, A∗ w)(L 2 (Ω,R)) p − 1 (Arh v, w)h − 1 (rh v, A∗ w)h |
|2 2 2 2 |
.∀v ∈ H , lim sup = 0,
h→0 w∈Vh ||w||(L 2 (Ω,R)) p
(7.7.19)
| { |
|1 1 E E
L |
| |
| (Brh v, w)2 ds − wl,S (Mrh v, w)2 (bl,S )|
| 2 ∂Ω 2 |
S⊂∂Ω l=1
.∀v ∈H, lim sup = 0,
h→0 w∈Vh ||w||(L 2 (∂Ω,R)) p
(7.7.20)
and
| |
| EE |
| N
|
|(F, w)(L 2 (Ω,R)) p − w (F, w) (b ) |
| l,K 2 l,K |
| K ∈τh l=1 |
. lim sup = 0. (7.7.21)
h→0 w∈Vh ||w||(L 2 (Ω,R)) p

Then, the solution .u ∗h ∈ Vh of the problem (7.7.9) converges to the exact solution
.u ∈ (H (Ω)) of the problem (7.1.2)–(7.1.3), i.e.,
1 p

. lim ||u − u ∗h ||(L 2 (Ω,R)) p = 0.

h→0

Proof Let .ε > 0 be arbitrarily small. We choose .v ∈ H such that

ε
.||u − v||(H 1 (Ω)) p ≤ ,
8C
where .C is the constant that appears in the inequality (7.7.13). We, then, choose
h ≤ h 0 (ε) small enough so that
.

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 433

ε
||v − rh v||(H 1 (Ω)) p ≤ ,
{ 8C
|v − rh v||w| ds
ε
sup ∂Ω ≤ ,
| w∈Vh ||w||(L 2 (Ω,R)) p 4C |
|1 |
| (Arh v, w)(L 2 (Ω,R)) p + 1 (rh v, A∗ w)(L 2 (Ω,R)) p − 1 (Arh v, w)h − 1 (rh v, A∗ w)h |
|2 2 2 2 |
sup +
||w||(L 2 (Ω,R)) p
w∈Vh | { |
. |1 1 E E
L |
| |
| (Brh v, w)2 ds − wl,S (Mrh v, w)2 (bl,S )|
| 2 ∂Ω 2 | ε
S⊂∂Ω l=1
+ sup ≤ ,
w∈Vh | ||w|| (L (∂Ω,R))
2 p
| 4C
| |
| EE N
|
|(F, w)(L 2 (Ω,R)) p − wl,K (F, w)2 (bl,K )||
|
| K ∈τh l=1 | ε
sup ≤ .
w∈Vh ||w||(L 2 (Ω,R)) p 4C

If, in the inequality (7.7.13), we choose .vh = rh v, we have for .h ≤ h 0 (ε)

||u − u ∗h ||(L 2 (Ω,R)) p ≤ ε,

.

which leads to convergence. Q.E.D.

Remark 7.7.8 The positivity conditions (7.7.11) and (7.7.12) play the role of stabil-
ity conditions and the hypotheses (7.7.17), (7.7.18), (7.7.19), (7.7.20), and (7.7.21)
play the role of consistency conditions. ♦

7.7.4 Convergence of the Approximate Solution Towards

the Exact Solution

We, now, consider the problem of the convergence of the approximate solution
- ∗ be the space defined in (7.4.1). We make the
towards the exact solution. Let .W
following assumption on the subspace . X h of .(H 1 (Ω)) p .
- ∗ , dense in .W
(H 2) There is a subspace .W ∗ of .W
. - ∗ , and a mapping .rh : W ∗ −→ X h
such that

lim ||v
. − r h v||(H 1 (Ω)) p = 0
h→0
⎛ { ⎞
⎜ |v − r h v||w h |ds ⎟
lim ⎜
⎝ sup ∂Ω ⎟ = 0.
h→0 wh ∈X h ||wh ||(L 2 (Ω,R)) p ⎠

@seismicisolation
@seismicisolation
434 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Theorem 7.7.9 (Lesaint [6]) We assume that the second member . F belongs to
(L 2 (Ω, R)) p and that hypothesis .(H 2) is satisfied. Let .u h ∈ X h be the solution
.
of the approached problem (7.7.5). Then, from the sequence .u h , we can extract a
subsequence .u h , which converges weakly in .(L 2 (Ω, R)) p to a solution .u of the weak
problem. ♦
Proof By using Theorem 7.1.4, equality (7.7.5) can be written
{
1
.(u h , A∗ vh )(L 2 (Ω,R)) p + (u h , (B +t M)vh )2 ds = (F, vh )(L 2 (Ω,R)) p for all vh ∈ X h .
2 ∂Ω
(7.7.22)

Let .v ∈ W ∗ and .rh v ∈ X h , where .rh is defined in .(H 2). Then,

lim (v. − rh v) = 0 strongly in (L 2 (Ω, R)) p (7.7.23)

h→0
∗ ∗
lim (A
. v − A r h v) = 0 strongly in (L (Ω, R))
2 p
(7.7.24)
h→0
⎛ { ⎞
⎜ |v − r h v||w h |ds ⎟
lim ⎜
⎝. sup
∂Ω ⎟ = 0. (7.7.25)
h→0 wh ∈X h ||wh ||(L 2 (Ω,R)) p ⎠

On the other hand, we can write

|{ | |{ |
| | | |
| | |
(u h , (B + M)rh v)2 ds | = |
t
(u h , (B + M)(rh v − v))2 ds ||
t
.
|
∂Ω ∂Ω
⎛ { ⎞
⎜ |v − r h v||w h |ds ⎟
≤c⎜ sup
⎝w ∈X
∂Ω ⎟ ||u h ||(L 2 (Ω,R)) p ,
⎠
h h
||w ||
h (L (Ω,R))
2 p

(7.7.26)

where .c is a constant independent of .h. By using Theorem 7.7.4, the sequence .(u h ) is
bounded in .(L 2 (Ω, R)) p , uniformly in .h. By using Lemma 2.8.2, we can therefore
extract a subsequence .(u h , ) which converges weakly in .(L 2 (Ω, R)) p towards an
element .u of .(L 2 (Ω, R)) p . When .h , tends to zero, we can write using the relations
(7.7.22), (7.7.23), (7.7.24), (7.7.25), and (7.7.26)

(u, A∗ v)(L 2 (Ω,R)) p = (F, v)(L 2 (Ω,R)) p for all v ∈ W ∗ .

.

- ∗ , then
Since the subspace .W ∗ is dense in .W

- ∗.
(u, A∗ v)(L 2 (Ω,R)) p = (F, v)(L 2 (Ω,R)) p for all v ∈ W
.

The .u element is, therefore, a weak solution of the problem (7.1.2)–(7.1.3). Q.E.D.

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 435

- be the space
Let .W
{ }
. - = v ∈ (H 1 (Ω)) p such that (B − M)v|∂Ω=0 .
W

We make the following assumption on the subspace . X h of .(H 1 (Ω)) p .

- , dense in .W
(H 3) There is a subspace .W of .W
. - , and a mapping .rh : W −→ X h such
that

lim ||v
. − r h v||(H 1 (Ω)) p = 0
h→0
⎛ { ⎞
⎜ |v − r h v||w h |ds ⎟
lim ⎜
⎝ sup ∂Ω ⎟ = 0.
h→0 wh ∈X h ||wh ||(L 2 (Ω,R)) p ⎠

Theorem 7.7.10 (Lesaint [6]) Let.u be the strong solution (in the sense of the Defini-
tion 7.1.1) of the problem (7.1.2)–(7.1.3) and .u h ∈ X h be the solution of the approx-
imate problem (7.7.5). We assume that the hypothesis .(H 3) is satisfied. Then, the
sequence .(u h ) strongly converges to .u in .(L 2 (Ω, R)) p , when .h tends to zero. ♦
Proof We consider the following expression, defined for all .vh ∈ X h
{
1 1
. Eh = ((A + A∗ )(u h − vh ), u h − vh )(L 2 (Ω,R)) p + (M(u h − vh ), u h − vh )2 ds.
2 2 ∂Ω

By using Lemma 7.1.5, Theorem 7.1.4, and the hypotheses (7.1.4)–(7.1.5)–(7.1.6),

we have
{
1 1
. Eh = (C(u h − vh ), u h − vh )(L 2 (Ω,R)) p + (M(u h − vh ), u h − vh )2 ds.
2 2 ∂Ω

Hence,
. E h ≥ α||u h − vh ||2(L 2 (Ω,R)) p . (7.7.27)

According to the definition of the approached problem (7.7.5) and Theorem 7.1.4,
we have
{
1
. Eh = (F, u h − vh )(L 2 (Ω,R)) p − (Avh , u h − vh )(L 2 (Ω,R)) p + ((B − M)vh , u h − vh )2 ds.
2 ∂Ω

- , we can write
For all .v ∈ W
{
1
. E h = (F − Av, u h − vh ) 2 + (A(v − vh ), u h − vh )(L 2 (Ω,R)) p + ((B − M)(vh − v), u h − vh )2 ds.
(L (Ω,R)) p 2 ∂Ω

-
Using the inequality (7.7.27), it comes, for any .v ∈ W

@seismicisolation
@seismicisolation
436 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
⎛ { ⎞
⎜ |v − vh ||wh |ds ⎟
.||u h − v||(L 2 (Ω,R)) p ≤ c⎜
⎝||F − Av||(L 2 (Ω,R)) p + ||v − vh ||(H 1 (Ω)) p + sup
∂Ω ⎟.
⎠
wh ∈X h ||wh || 2
(L (Ω,R)) p

(7.7.28)
We, now, consider the inequality (7.7.28) for .v ∈ W and, we choose .vh = rh v. When
.h tends to zero, we have

. lim ||u h − v||(L 2 (Ω,R)) p ≤ c||F − Av||(L 2 (Ω,R)) p for all v ∈ W ,

h→0

- , we have
and since .W is dense in .W

. -.
lim ||u h − v||(L 2 (Ω,R)) p ≤ c||F − Av||(L 2 (Ω,R)) p for all v ∈ W (7.7.29)
h→0

The equality (7.7.29) is valid for the functions .u j of the Definition 7.1.1 and it comes,
for any index . j

. lim ||u h − u||(L 2 (Ω,R)) p ≤ c(||u − u j ||(L 2 (Ω,R)) p + ||F − Au j ||(L 2 (Ω,R)) p ).
h→0

Taking the limit of the right-hand side as . j tends to infinity, we get

. lim ||u h − u||(L 2 (Ω,R)) p = 0.

h→0

This completes the proof. Q.E.D.

Theorem 7.7.11 (Lesaint [6]) Let .u ∈ (H 1 (Ω)) p be the strong solution of the prob-
lem (7.1.2)–(7.1.3) and .u h ∈ X h be the solution of the approximate problem (7.7.5).
Then,
⎛ { ⎞
⎜ |u − v h ||w h |ds ⎟
.||u − u h ||(L 2 (Ω,R)) p ≤ c inf ⎜|u − vh |1,2,Ω + sup ⎟,
∂Ω
vh ∈X h ⎝ wh ∈X h ||wh ||(L 2 (Ω,R)) p ⎠

where .c is a constant independent of .h. ♦

Proof By using (7.7.28), we have
⎛ { ⎞
⎜ |v − vh ||wh |ds ⎟
.||u h − v||(L 2 (Ω,R)) p ≤ c ⎜
⎝||F − Av||(L 2 (Ω,R)) p + |v − vh |1,2,Ω + sup
∂Ω ⎟,
wh ∈X h ||wh ||(L 2 (Ω,R)) p ⎠

- . For .v = u in the last inequalities, we have

for all .v ∈ W

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 437
⎛ { ⎞
⎜ |u − vh ||wh |ds ⎟
||u − u h ||(L 2 (Ω,R)) p ≤ c ⎜
⎝|u − vh |1,2,Ω + wsup
∂Ω ⎟,
||wh ||(L 2 (Ω,R)) p ⎠
.
∈X h h

for all .vh ∈ X h . This completes the proof. Q.E.D.

7.7.5 Application to Finite Elements

We will assume here that the open set .Ω is polyhedral, but the results given here
can be extended to the case of any open set by using the same techniques (curved
isoparametric finite elements) as in [7–9]. Let .τh be a triangulation of .Ω in finite
elements . K with diameter less than or equal to .h. The methods defined here can be
extended to the case of finite elements of the Hermite type as in [10]. We will assume
that all the elements . K of the triangulation are the ranges by affine transformations
(or belonging to .(Q 1 )2 in the case of the dimension .n = 2) of a finite element of
-. We define the function spaces .Wh and .Vh by
reference . K
| |
. Wh = PK ,
K ∈τh

Vh = the subspace of .Wh made up of continuous functions at the nodes of the tri-
.

angulation i.e., for any .v ∈ Vh and for any pair of adjacent elements . K 1 and . K 2 , we
have n n
.v|K 1 (a) = v|K 2 (a), ∀a ∈ Σ K 1 K 2 = ΣK2 K1.

We have the following inclusion

n
. Vh ⊂ (H 1 (Ω) C 0 (Ω, R)) p .

The space . X h can be defined in one of two ways

. X h1 = Vh

. X h2 = {vh ∈ Vh such that (B − M)vh = 0 at nodes belonging to ∂Ω}.

The functions .vh of the space . X h2 thus discretely satisfy the boundary condition
(7.1.3). On the other hand, it is easy to build a basis of the space. X h2 , locally. Moreover,
when the matrix . B − M has its coefficients constant along the faces included in the
boundary .∂Ω, we have

@seismicisolation
@seismicisolation
438 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{
. ((B − M)vh , wh )2 ds = 0 for all vh ∈ X h2 , wh ∈ Vh ,
∂Ω

and the space . X h2 is included in the space . X h defined in (7.7.1). We consider the
approximate problem (7.7.5) which we write again here for more clarity: Find .u h ∈
X h such that, for all .vh ∈ X h , we have
{
1 1 1
. (Au h , vh )(L 2 (Ω,R)) p + (u h , A∗ vh )(L 2 (Ω,R)) p + (Mu h , vh )2 ds = (F, vh )(L 2 (Ω,R)) p ,
2 2 2 ∂Ω
(7.7.30)
where . X h is one of the spaces . X hi , .i = 1, 2, defined above.

Remark 7.7.12 When . X h = X h1 = Vh , the formulation (7.7.30) of the approximate

problem leads to an implicit processing of the boundary conditions. We actually
use here a discontinuous finite element method along the border .∂Ω the value of the
approximate solution .u h along .∂Ω, inside .Ω, not necessarily satisfying the boundary
conditions (7.1.3). ♦

We have the following theorem.

Theorem 7.7.13 (Lesaint [6]) Let .(τh )h be a regular family of triangulations of .Ω.
We assume that in the case of simplicial elements (resp. quadrilaterals), we have the
- (resp. . Q k ⊂ P)
inclusion . Pk ⊂ P - for an integer .k ≥1. Furthermore, if we suppose
that the second member . F belongs to .(L 2 (Ω) p , then the unique solution .u h ∈ X h of
the problem (7.7.30) satisfies the inequality

1
.||u h ||(L 2 (Ω,R)) p ≤ ||F||(L 2 (Ω,R)) p . (7.7.31)
α
♦

Proof The inequality (7.7.31) is nothing other than the inequality (7.7.6) of Theorem
7.7.4. Q.E.D.

Theorem 7.7.14 (Lesaint [6]) Let .(τh )h be a regular family of triangulations of .Ω.
We assume that in the case of simplicial elements (resp. quadrilaterals), we have the
- (resp. . Q k ⊂ P)
inclusion . Pk ⊂ P - for an integer .k ≥1. We suppose that the second
member . F belongs to .(L 2 (Ω) p . Then, we have the following results:
.(i) When . X h = X h1 , from the sequence .(u h ), we can extract a subsequence .(u h , )
which converges weakly in .(L 2 (Ω, R)) p , when .h , tends to zero, to a weak solution
.u of the problem (7.1.2)–(7.1.3).

.(ii) We now assume that .u is the strong solution of the problem (7.1.2)–(7.1.3). Then,
if . X h = X h1 , or . X h2 , then the sequence .(u h ) strongly converges in .(L 2 (Ω, R)) p , when
.h tends to zero, to the strong solution .u of the problem (7.1.2)–(7.1.3). ♦

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 439

Proof To show the weak and strong convergences, it suffices to verify that the
- ∗ and .W
hypotheses .(H 2) and .(H 3) are satisfied. Let .W - be the spaces defined,
respectively, by
{ }
. - ∗ = v ∈ (H 1 (Ω)) p such that (B +t M)v|∂Ω = 0
W

and { }
. - = v ∈ (H 1 (Ω)) p such that (B − M)v|∂Ω = 0
W

γ ∗ and .-
and, let .- γ be the spaces defined, respectively, by
{ }
-∗ = v ∈ (H 2 (Ω)) p such that (B +t M)v|∂Ω = 0
γ
. (7.7.32)

and { }
. γ = v ∈ (H 2 (Ω)) p such that (B − M)v|∂Ω = 0 .
-

If the boundary .∂Ω is of class .C 2 and if the subspaces . K er (B(x) − M(x)) and
. K er (B(x) + M(x)) included in .R , are generated by vectors whose components,
t n

defined for .x ∈ ∂Ω, are twice continuously differentiable on .∂Ω, we can show by
reasoning with local maps, as in [1, Chapter 1], or in [2, p. 431], that the space .- γ∗
(resp. .- - ∗ -
γ ) is dense in .W (resp. .W ). These density results can extend [4, pp. 237–241]
in case the domain .Ω is polyhedral. We, now, suppose that, we have .n ≤ 3, so that
the inclusion
. H (Ω) ⊂ C (Ω, R)
2 0

is satisfied. Let .πh : (H 2 (Ω)) p −→ X h be the interpolation operator. By applying

Lemma 4.1.14, we can write

.||v − πh v||1,2,Ω ≤ ch|v|2,2,Ω (7.7.33)

for all .v ∈ (H 2 (Ω)) p . By using Lemmas 4.1.13 and 4.1.14, we have

{
. |wh ||v − πh v|ds ≤ ch||wh ||(L 2 (Ω,R)) p |v|2,2,Ω
∂Ω

for all .wh ∈ Vh and all .v ∈ (H 2 (Ω)) p . Hence,

{
|wh ||v − πh v|ds
∂Ω
. sup ≤ ch|v|2,2,Ω , (7.7.34)
wh ∈Vh ||wh ||(L 2 (Ω,R)) p

γ ∗ as (7.7.32)
for all .v ∈ (H 2 (Ω)) p . So, the hypothesis .(H 2) is satisfied by defining .-
and choosing for application .rh the operator .πh . For all .v ∈ - γ , we have

@seismicisolation
@seismicisolation
440 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

. πh v ∈ X h2 ⊂ X h1 = Vh .

Inequalities (7.7.33) and (7.7.34) hold for any .v ∈ - γ and the hypothesis .(H 3) is,
therefore, satisfied in both cases,. X h = Vh and. X h = X h2 by choosing.rh = πh . Q.E.D.

Remark 7.7.15 .(i) When the dimension .n is greater than or equal to 4, the previous
results are still valid, replacing the interpolation operator .πh by a suitable operator
.-
πh , constructed using a process of regularization and suppose that, we have the
inequality, for a constant independent of the triangulation

h
. ≤ c, (7.7.35)
ρ

where
. ρ = inf ρ K
K ∈τh

and
. h = sup h K .
K ∈τh

(ii) The weak convergence result is still valid for . X h = X h2 , when the following
.

equality is satisfied
. K er (B − M) = K er (B + M).
t

γ ∗ belongs to the space

Indeed, in this case, the interpolate .πh v of any function .v of .-
2
.Xh. ♦
Theorem 7.7.16 (Lesaint [6]) We assume that the hypotheses of Theorem 7.7.13 are
satisfied for an integer .k ≥ 1 and, that the solution .u belongs to .(H s+1 (Ω)) p , for an
integer .s such that .1 ≤ s ≤ k, and . n2 − 1 < s. Then, for . X h = X h1 or . X h2 , we have

||u − u h ||(L 2 (Ω,R)) p ≤ ch s |u|s+1,2,Ω ,

. (7.7.36)

where .c is a constant independent of .h. ♦

Proof Let .u ∈ (H (Ω)) and .πh u ∈ Vh be its interpolated. Since the function .u
s+1 p

check the boundary conditions (7.1.3), we have

.πh u ∈ X h2 ⊂ X h1 = Vh .

According to Theorem 7.7.11, we can write

⎛ { ⎞
⎜ |u − πh u||wh |ds ⎟
||u − u h ||(L 2 (Ω,R)) p ≤ c ⎜
.
⎝||u − πh u||1,2,Ω + wsup
∂Ω ⎟.
⎠
∈X h h
||wh ||(L 2 (Ω,R)) p

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 441

According to Lemma 4.1.14, we have

||u − πh u||1,2,Ω ≤ ch s |u|s+1,2,Ω .

.

According to Lemmas 4.1.14 and 4.1.13, we have

|{ |
| |
.| |u − πh u||wh |ds || ≤ ch s |u|s+1,2,Ω ||wh ||(L 2 (Ω,R)) p ,
|
∂Ω

for all .vh ∈ Vh . The last three inequalities lead to Eq. (7.7.36). Q.E.D.

Remark 7.7.17 .(i) When we have the inequality .k ≤ n2 − 1, the bound (7.7.36)
is still valid, as at the end of the proof of Theorem 7.7.13, and assuming that the
inequality (7.7.35) is satisfied. We can also obtain an upper bound analogous to the
inequality (7.7.36), if we assume that the solution .u ∈ (W s+1,q (Ω)) p , for an integer
.s and a number .q such that .1 ≤ s ≤ k + 1 and .n − q(s + 1) < 0. We, then, have

.||u − u h ||(L 2 (Ω,R)) p ≤ ch s |u|s+1,q,Ω .

(ii) The previous theorem shows that when we use polynomials of degree .≤ k, we
.
can obtain precision in .o(h k ). ♦

7.7.6 Error Estimates

We pose
{
1 1
. Eh = ((A + A∗ )(u h − vh ), u h − vh )(L 2 (Ω,R)) p + (M(u h − vh ), u h − vh )2 ds,
2 2 ∂Ω

with .u h ∈ R p and .vh arbitrary. Since . A is positive and . M is positive semi-definite,

then
. E h ≥ α||u h − vh || 2
(L (Ω,R)) p .
2

We have
{
1 ∗ 1 1
. (A (u h − vh ), u h − vh )(L 2 (Ω,R)) p = (A(u h − vh ), u h − vh )(L 2 (Ω,R)) p − (B(u h − vh ), u h − vh )2 ds.
2 2 2 ∂Ω

Hence,
{
1
. E h = (A(u h − vh ), u h − vh )(L 2 (Ω,R)) p − ((B − M)(u h − vh ), u h − vh )2 ds.
2 ∂Ω

Further,

@seismicisolation
@seismicisolation
442 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{
1
.(Au h , u h − vh )(L 2 (Ω,R)) p − ((B − M)u h , u h − vh )2 ds = (F, u h − vh )(L 2 (Ω,R)) p
2 ∂Ω
= (Au, u h − vh )(L 2 (Ω,R)) p .

Thus, since .(B − M)u = 0, we have

{
1
.α||u h − vh ||2(L 2 (Ω,R)) p ≤ (A(u − vh ), u h − vh )(L 2 (Ω,R)) p − ((B − M)(u − vh ), u h − vh )2 ds.
2 ∂Ω

Let’s pose
. Vh = {Pk in pieces k ≥ 1}.

We replace .vh by .rh u, we have

{
1
.α||u h − rh u||2(L 2 (Ω,R)) p ≤ (A(u − rh u), u h − rh u)(L 2 (Ω,R)) p − ((B − M)(u − rh u), u h − rh u)2 ds.
2 ∂Ω

We pose .wh = u h − rh u, then

n (
{ E ) {
∂
.(A(u − r h u), wh )(L 2 (Ω,R)) p = Ai (u − rh u), wh d xd y + (A0 (u − rh u), wh )2 d xd y.
Ω i=1 ∂ xi 2 Ω

We pose .u m the m-th component of vector .u. Then,

. |u m − rh u m |1,Ω ≤ ch k |u m |k+1,Ω .

Hence,
|{ ( ) |
| ∂ |
| Ai (u − rh u), wh d xd y || ≤ ch k |u|k+1,Ω ||wh ||(L 2 (Ω,R)) p .
.
| ∂ xi
Ω 2

We, now, want to increase the following term

{
1
. ((B − M)(u − rh u), wh )2 ds. (7.7.37)
2 ∂Ω

According to the trace theorem

||wh ||0,∂Ω ≤ c||wh ||1,Ω .

.

Hence,
|{ |
| |
| ((B − M)(u − rh u), wh )2 ds || ≤ c||u − rh u||1,Ω ||wh ||1,Ω
.
|
∂Ω
≤ ch k |u|k+1,Ω ||wh ||1,Ω .

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 443

Note that we cannot overload .||wh ||1,Ω because we only have the coercivity in . L 2 .
So, we turn back, that is to say that it is not desirable to apply the trace theorem. So,
we use the inverse inequality (see Lemma 4.5.7). Indeed, according to Lemma 4.5.7,
we have
−1
.||wh ||1,Ω ≤ ch ||wh ||0,Ω .

Hence,
|{ |
| |
| ((B − M)(u − rh u), wh )2 ds || ≤ ch k−1 |u|k+1,Ω ||wh ||0,Ω .
.
|
∂Ω

Thus, if we interpolate by the polynomials of degree .k (. Pk ), then the error is given

by
.||u h − r h u||0,Ω ≤ ch |u|k+1,Ω .
k−1
(7.7.38)

If we interpolate by the polynomials of degree .1 (. P1 ), then the error is given by

. ||u h − rh u||0,Ω = o(1).

This is not satisfactory.

Other methods: We basis ourselves on the Cauchy-Schwarz inequality on (7.7.37)
and, we use the trace theorem only for the term .u − rh u but not for the term .wh . If,
we have
−1
.||wh ||0,∂Ω ≤ ch 2 ||wh ||0,Ω , (7.7.39)

then
1
||u h − rh u||0,Ω ≤ ch k− 2 |u|k+1,Ω .
.

This error estimates is better than the error estimates (7.7.38). Let’s show (7.7.39).
Indeed, let us increase the term
E {
. wh2 ds.
S⊂∂Ω S

We have { {
l(S)
. wh2 ds = -
wh2 ds.
S 1 -
S

Since in a finite-dimensional space all the norms are equivalent, we have

{ {
. wh2 ds ≤ -
c -wh2 dξ dη.
S -
T

For example, triangle with three nodes

@seismicisolation
@seismicisolation
444 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

{ ( ( ) )
1 w1 + -
- w3 2
. -
wh2 dξ dη = w1 + 4
-2
+-2
w3
-
T 6 2
1 2
= (-
w +-
w1-
w3 + -
w32 )
3 1
and

(( )2 ( )2 ( )2 )
11 w1 + -
- w2 w2 + -
- w3 w1 + -
- w3
. + +
23 2 2 2
1 2
= w +-
(- w22 + -
w32 + -
w1 -
w2 + -
w1 -
w3 + -
w2 -
w3 ).
12 1

Hence, { {
. - c1 h −2
wh2 dξ dη ≤ - wh2 d xd y
-
T T

and { {
−1
. -
wh2 ds ≤ ch wh2 d xd y.
-
S T

E { E{
Thus,
−1
. wh2 ds ≤ ch wh2 d xd y.
S⊂∂Ω S T ⊂Ω T

Proving (7.7.39). Now, we will increase the term (7.7.37).

In dimension 1: Do we have this inequality

|α(0)(u − rh u)(0)wh (0)| ≤ ch β |u|k+1,(0,1) ||wh ||0,(0,1) .

.
−

0 h 1 −1 1
Reference segment

In fact, we set the change of variable .x = 1+ξ

2
h + 1−ξ
2
0. Let’s pose

. u, -
E(- wh ) = -
α (−1)(-
u −-
r-
u )(−1)-
wh (−1).

The mapping .-u −→ E(- u, -

wh ) is linear and continuous from . H k+1 (−1, 1) into .R of
norm .≤ -
c||-
wh ||0,(−1,1) and identically null for all . - -k . Then,
p∈P

@seismicisolation
@seismicisolation
7.7 Approximation of Friedrichs Systems by Finite Element Methods 445

.|E(-
u, -
wh )| ≤ c|-
u |k+1,(−1,1) ||-
wh ||0,(−1,1)
≤ ch k+1 |u|k+1,(0,h) h − 2 |wh |0,(0,h) h − 2 .
1 1

Thus,
|E(-
. u, -
wh )| ≤ ch k |u|k+1,(0,h) |wh |0,(0,h) .

We can do even better: instead of working with . H k+1 (−1, 1), we work with
.W
k+1,∞
(−1, 1) and, we find

. |E(-
u, -
wh )| ≤ c||-
u ||k+1,∞,(−1,1) ||-
wh ||0,(−1,1)
≤ ch k+1 ||u||k+1,∞,(0,h) h − 2 |wh |0,(0,h) .
1

Thus,
1
. |E(-
u, -
wh )| ≤ ch k+ 2 ||u||k+1,∞,(0,h) |wh |0,(0,h) .

So, we won . 21 screen. On the other hand, since . H 2 (−1, 1) is continuously injected
into .C 0 ([−1, 1], R), we have

|E(-
. u, -
wh )| ≤ c|-
u −- r-
u |(−1)|-
wh |(−1)
≤ c(|-
u (−1)| + |-
r-
u (−1)|)|-
wh (−1)|
≤ c max |-
u (ξ )| max |-
wh (ξ )|
ξ

≤ c||-
u ||2,(−1,1) ||-
wh ||2,(−1,1) .

Hence,
.|E(-
u, -
wh )| ≤ c||-
u ||k+1,(−1,1) ||-
wh ||0,(−1,1) .

Thus, the continuity of . E(·, ·).

In dimension 2: We have
{ {
. ((B − M)(u − rh u), wh )2 ds = l(S) (( - - u −-
B − M)(- r-
u ), -
wh )2 dη.
S -
S

We pose {
. u, -
E(- wh ) = (( - - u −-
B − M)(- r-
u ), -
wh )2 dη.
-
S

@seismicisolation
@seismicisolation
446 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

-
S S

The mapping.- u −→ E(- u, - -) into.R of norm

wh ) is linear and continuous from. H k+1 (T
.≤ -
c||-
wh ||0,T- and identically null for all . -
p∈P-k , then

|E(-
. u, -
wh )| ≤ c|-
u |k+1,T-||-
wh ||0,T-
≤ ch k+1 |u|k+1,T h −1 h −1 |wh |0,T
≤ ch k−1 |u|k+1,T |wh |0,T .

-) instead of . H k+1 (T
Remark 7.7.18 We can take .W k+1,∞ (T -). ♦

7.8 Continuous Finite Element Methods

Let .Ω ⊂ R2 be a polyhedron and let . Z h be the space

. Z h = {vh ∈ C 0 (Ω, R) such that vh |T ∈ Pk }.

Theorem 7.8.1 We assume that the solution of the problem

. Au = F in Ω

(B − M)u = 0 on ∂Ω
.

belongs to . H k+1 (Ω). If the solution .u h of the problem:

Find .u h ∈ Vh such that

(u h , A∗ vh )(L 2 (Ω,R)) p = (F, vh )(L 2 (Ω,R)) p

.

@seismicisolation
@seismicisolation
7.8 Continuous Finite Element Methods 447

for all .vh ∈ X h∗ , is built from the space . Z h , then for .k ≥ 1 and for .n ≤ 3, we have

||u − u h ||0,Ω = o(h k ).

.

♦
Proof We pose
( ) {
1 1
. Eh = (A + A∗ )(u h − vh ), u h − vh + (M(u h − vh ), u h − vh )2 ds.
2 (L 2 (Ω,R)) p 2 ∂Ω

Then, { {
1
. (Mwh , wh )2 ds = ((M + M ∗ )wh , wh )2 ds.
∂Ω 2 ∂Ω

Since . A is positive and . M +t M is positive semi-definite, we have

. E h ≥ α||u h − vh ||2(L 2 (Ω,R)) p .

On the other hand,

{
1
.(Au h , u h − vh )(L 2 (Ω,R)) p − ((B − M)u h , u h − vh )2 ds = (F, u h − vh )(L 2 (Ω,R)) p
2 ∂Ω
= (Au, u h − vh )(L 2 (Ω,R)) p .

Hence,
{
1
(A(u h − u), u h − vh )(L 2 (Ω,R)) p =
. ((B − M)u h , u h − vh ) ds.
2 ∂Ω

Which proves
{
1
. Eh = (A(u h − u + u − vh ), u h − vh )(L 2 (Ω,R)) p − ((B − M)(u h − vh ), u h − vh )2 ds.
2 ∂Ω

Thus, since .(B − M)u = 0, we have

{
1
. E h = (A(u − vh ), u h − vh )(L 2 (Ω,R)) p − ((B − M)(u − vh ), u h − vh )2 ds.
2 ∂Ω

We take .vh = rh u. We pose .wh = u h − rh u, we have

n (
{ E ) {
∂
.(A(u − rh u), wh )(L 2 (Ω,R)) p = Ai (u − rh u), wh d xd y + (A0 (u − rh u), wh )2 d xd y.
Ω i=1 ∂ xi 2 Ω

@seismicisolation
@seismicisolation
448 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

We pose .u m , .m-th component of the vector .u, then

.|u m − rh u m |1,Ω ≤ ch k |u m |k+1,Ω .

Hence,
|{ ( ) |
| ∂ |
| A (u − r u), w d xd y | ≤ ch k |u|k+1,Ω ||wh ||(L 2 (Ω,R)) p .
.
| i
∂ x
h h |
Ω i 2

Thus,
.|(A(u − rh u), (u h − rh u)| ≤ ch k |u|k+1,Ω ||u h − rh u||(L 2 (Ω,R)) p .

Now, we are interested in the increase of the term

{
. ((B − M)(u − rh u), wh )2 ds. (7.8.1)
∂Ω

1st idea: We use the trace theorem

|{ |
| |
| ((B − M)(u − rh u), wh )2 ds || ≤ c||u − rh u||1,Ω ||wh ||1,Ω
.
|
∂Ω
≤ ch k |u|k+1,Ω ||wh ||1,Ω .

We cannot overload .||wh ||1,Ω because we only have coercivity in . L 2 (Ω, R). So,
we turn back. So, it is not satisfactory to apply the trace theorem. We will use the
following inverse inequality

. ||wh ||1,Ω ≤ ch −1 ||wh ||0,Ω .

Thus,
|{ |
| |
.| ((B − M)(u − rh u), wh )2 ds || ≤ ch k−1 |u|k+1,Ω ||wh ||0,Ω .
|
∂Ω

Hence,

. ||u h − rh u||0,Ω ≤ ch k−1 |u|k+1,Ω .

If we interpolate by . P1 , we find

||u h − rh u||0,Ω = o(1).

.

This is not satisfactory.

@seismicisolation
@seismicisolation
7.8 Continuous Finite Element Methods 449

2nd idea: We basis ourselves on Cauchy-Schwarz inequality on (7.8.1) and, we use

the trace theorem only for the term .u − rh u but no for .wh . If, we have

||wh ||0,∂Ω ≤ ch − 2 ||wh ||0,Ω ,

1
. (7.8.2)

then
|{ |
| |
| ((B − M)(u − rh u), wh )2 ds || ≤ c||u − rh u||0,∂Ω ||wh ||0,∂Ω
.
|
∂Ω

≤ c||u − rh u||1,Ω ch − 2 ||wh ||0,Ω

1

1
≤ ch k− 2 |u|k+1,Ω ||wh ||0,Ω .

Hence,
1
||u h − rh u||0,Ω ≤ ch k− 2 |u|k+1,Ω .
.

So, this last increase is better than the others. So, we still have to show (7.8.2). Indeed,
E{
.||wh ||0,∂Ω =
2
wh2 ds.
S⊂Ω S

Since in a finite-dimensional space all the norms are equivalent, we obtain

{ {
l(S)
. wh2 ds = -
wh2 ds
1 -
S
{ S

c l(S) -
≤- wh2 dξ dη.
-
T

Furthermore, .l(S) = o(h) so

wh ||0,T- ≤ c||B||0 (det B)− 2 ||wh ||0,T

1
.||-
≤ ch −1 ||wh ||0,T .

Thus,
{
. wh2 ds ≤ chh −2 ||wh ||20,T .
S

It follows that

||wh ||20,∂Ω ≤ chh −2 ||wh ||20,Ω .

.

@seismicisolation
@seismicisolation
450 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

It thus appears that

||wh ||0,∂Ω ≤ ch − 2 ||wh ||0,Ω .

1
.

Q.E.D.

Remark 7.8.2 .(i) Theorem 7.8.1 is always valid if one uses isoparametric convex
quadrilateral finite elements by using on the reference square polynomials of degree
.≤ k.
−1
.(ii) The best increase of the error that we found is of the order . O(h 2 ). ♦

Question: Can we do better.

In fact,
In dimension 1: We pose .(B − M)(0) = α(0) and .(B + M)(1) = α(1). We want
show that
.|α(0)(u − r h u)(0)wh (0)| ≤ ch |u|k+1,(0,h) ||wh ||0,(0,h) .
k+1
−

−
0 h 1 −1 1
Reference segment

We pose
. u, -
E(- wh ) = -
α (−1)(-
u −-
r-
u )(−1)-
wh (−1).

The mapping
.-
u −→ E(-
u, -
wh )

is linear and continuous from . H k+1 (−1, 1) into .R of norm .≤ -

c||-
wh ||0,(−1,1) and

. p, -
E(- wh ) = 0

for all . -
p ∈ Pk , then

|E(-
. u, -
wh )| ≤ c|-
u |k+1,(−1,1) ||-
wh ||0,(−1,1)
≤ ch k+1 |u|k+1,(0,h) h − 2 ||wh ||0,(0,h) h − 2
1 1

≤ ch k |u|k+1,(0,h) ||wh ||0,(0,h) .

Thus,
.||u h − rh u||0,(0,1) ≤ ch k |u|k+1,(0,1) .

We can do better, i.e., instead of working with . H k+1 (−1, 1), we work with .W k+1,∞
(−1, 1). We have

@seismicisolation
@seismicisolation
7.8 Continuous Finite Element Methods 451

|E(-
. u, -
wh )| ≤ c||-
u ||k+1,∞,(−1,1) ||-
wh ||0,(−1,1)

because the mapping .-u −→ E(- u, -

wh ) is linear and continuous from .W k+1,∞ (−1, 1)
into .R of norm .≤ -
c||-
wh ||0,(−1,1) and

. p, -
E(- wh ) = 0

for all . -
p ∈ Pk . Since
1
||-
. u ||k+1,∞,(−1,1) ≤ c||B||k+1 (det B) ∞ ||u||k+1,∞,(0,h)
≤ ch k+1 ||u||k+1,∞,(0,h) ,

we have

wh )| ≤ ch k+1 ||u||k+1,∞,(0,h) h − 2 ||wh ||0,(0,h)

1
|E(-
. u, -
1
≤ ch k+ 2 ||u||k+1,∞,(0,h) ||wh ||0,(0,h) .

Thus,
1
||u h − rh u||0,(0,1) ≤ ch k+ 2 ||u||k+1,∞,(0,1) .
.

As a result, we won . 21 screen altogether. It remains the continuity of . E(·, ·). Since
. H (−1, 1) injects itself continuously into .C ([−1, 1], R), we have
2 0

|E(-
. u, -
wh )| ≤ c|-
u −- r-
u |(−1)|wh |(−1)
≤ c(|-
u |(−1) + |-
r-
u |(−1))|-
wh |(−1)
≤ c max |-
u (ξ )| max |-
wh (ξ )|
ξ ξ

≤ c||-
u ||2,(−1,1) ||-
wh ||2,(−1,1) .

However, in a finite-dimensional space all the norms are equivalent, we obtain

. |E(-
u, -
wh )| ≤ c||-
u ||k+1,(−1,1) ||-
wh ||0,(−1,1) .

Thus, the continuity of . E(·, ·).

In dimension 2: We have
{ {
. ((B − M)(u − rh u), wh )2 ds = l(S) (( - - u −-
B − M)(- r-
u ), -
wh )2 dη.
S -
S

Let’s pose
{
. u, -
E(- wh ) = (( - - u −-
B − M)(- r-
u ), -
wh )2 dη.
-
S

@seismicisolation
@seismicisolation
452 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

The mapping
-
u −→ E(-
. u, -
wh )

-) into .R of norm .≤ -
is linear and continuous from . H k+1 (T c||-
wh ||0,T- and

. p, -
E(- wh ) = 0

for all . -
p ∈ Pk . Then,

|E(-
. u, -
wh )| ≤ c|-
u |k+1,T-||-
wh ||0,T-
≤ ch k+1 |u|k+1,T h −1 h −1 ||wh ||0,T
≤ ch k−1 |u|k+1,T ||wh ||0,T .

Furthermore,
( ) 21 ( ) 21
E E E
. |u|k+1,T ||wh ||0,T ≤ |u|2k+1,T ||wh ||20,T .
T ⊂Ω T ⊂Ω T ⊂Ω

Since .l(S) = o(h), we have

.||u h − rh u||0,Ω ≤ ch k |u|k+1,Ω .

-), we work with

We can do better. Indeed, instead of working with . H k+1 (T
k+1,∞ -
.W (T ). We have

|E(-
. u, -
wh )| ≤ c||-
u ||k+1,∞,T-||-
wh ||0,T-
≤ ch k+1 ||u||k+1,∞,T h −1 h −1 ||wh ||0,T
≤ ch k ||u||k+1,∞,T ||wh ||0,T .

Hence, {
. ((B − M)(u − rh u), wh )2 ds ≤ ch k+1 ||u||k+1,∞,T ||wh ||0,T .
S

Thus,

@seismicisolation
@seismicisolation
7.8 Continuous Finite Element Methods 453
{ E
. ((B − M)(u − rh u), wh )2 ds ≤ ch k+1 ||u||k+1,∞,T ||wh ||0,T
∂Ω T ⊂Ω
E
≤ ch k+1 ||u||k+1,∞,Ω 1 × ||wh ||0,T
T ⊂Ω
⎛ ⎞1 ⎛ ⎞1
E 2
E 2

≤ ch k+1 ||u||k+1,∞,Ω ⎝ 12 ⎠ ⎝ ||wh ||20,T ⎠

T ⊂Ω T ⊂Ω
1
≤ ch k+1 ||u||k+1,∞,Ω (h −1 ) 2 ||wh ||0,Ω
1
≤ ch k+ 2 ||u||k+1,∞,Ω ||wh ||0,Ω .

So,
1
||u h − rh u||0,Ω ≤ ch k+ 2 ||u||k+1,∞,Ω .
.

-) injects itself continuously into

It remains the continuity of . E(·, ·). Since . H 2 (T
0 -
.C ( T , R), we have

|E(-
. u, -
wh )| ≤ c max |-
u (ξ )| max |-
wh (ξ )|
ξ ξ

≤ c||-
u ||2,T-||-
wh ||2,T-.

However, in a finite-dimensional space all the norms are equivalent, we obtain

|E(-
. u, -
wh )| ≤ c||-
u ||k+1,T-||-
wh ||0,T-.

Thus, the continuity of . E(·, ·).

Example 7.8.1 We consider the problem

du
. + σ u = f on ]0, 1[
dx

.u(0) = b,

with .σ > 0. We pose .w = u − b, we have

dw
. + σ w = f − σ b on ]0, 1[
dx

. w(0) = 0.

We have. M(0) = 1,. M(1) = 1,. B(0) = −1,. B(1) = 1,.(B − M)(0) = −2, and.(B −
M)(1) = 0. Then,

@seismicisolation
@seismicisolation
454 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{ 1 ( )
dwh 1
. + σ wh − f + σ b vh d x − (−2)wh (0)vh (0) = 0
0 dx 2

and { ( )
1
du h
. + σ uh − f vh d x + (u h (0) − b)vh (0) = 0.
0 dx

ψi
−

−
0 xi−1 xi xi+1 1

{
u j+1 − u j−1 h x j+1
. + σ (u j+1 + 4u j + u j−1 ) − f ψ j d x = 0, 1 ≤ j ≤ I − 1
2 6 x j−1

{
u1 − u0 h h
. + σ (2u 0 + u 1 ) − f ψ0 d x + u 0 − b = 0.
2 6 0

7.9 Discontinuous Method

7.9.1 Discontinuous Finite Element Method

We consider the problem

. Au = F

.(B − M)u = 0.

Let .Vh ⊂ (L 2 (Ω, R)) p be the space defined by

. Vh = {vh ∈ (L 2 (Ω, R)) p such that vh |T ∈ Pk1 × Pk2 × · · · × Pk p }.

We search .u h ∈ Vh such that

E[ { ((
BT − M T
) ) ]
. (Au h − F, vh )(L 2 (T,R)) p − (u int
h − u h ), vh
ext int
ds = 0
T ⊂Ω ∂T 2 2
(7.9.1)
for all .vh ∈ Vh , with
E
n
. BT = νiT Ai
i=1

@seismicisolation
@seismicisolation
7.9 Discontinuous Method 455

ν T exterior normal vector, . MT matrix defined on .∂ T , the functions .u int

.
ext
h (resp. .u h )
designate the values of .u h along .∂ T inside (resp. outside) .T . For a face . S ⊂ ∂Ω,
h = 0 if it is a homogeneous problem. The
we will impose (with the convention) .u ext
matrix . MT is to be chosen for all the faces not included in .∂Ω. For
n the other faces,
we have . MT = M, given by the boundary conditions i.e., on .∂ T ∂Ω, . MT = M.
Question: If . F = 0 can we show that .u h = 0.
We replace .u h by .vh in (7.9.1), we obtain

E[ { ((
BT − M T
) ) ]
. (Au h , u h )(L 2 (T,R)) p − (u int
h − u h ), u h
ext int
ds = 0.
T ⊂Ω ∂T 2 2

Hence,
[
E (1 )
1
{
. (A + A∗ )u h , u h + (BT u int
h , u h ) ds
int
2 (L 2 (T,R)) p 2 ∂T
T ⊂Ω
{ (( ) ) ]
BT − M T
− (u int
h − u ext
h ), u int
h ds = 0.
∂T 2 2

2
1
S

On the other hand,

{ {
1
. (MT (u int
h − u ext
h ), u int
h 2 ) ds = (M1 (u 1h − u 2h ), u 1h )2 ds
2 ∂T S
{
+ (M2 (u 2h − u 1h ), u 2h )2 ds
{ S
= (M S (u 1h − u 2h ), u 1h − u 2h )2 ds,
S

with. M1 = M2 = M S . Hence, the idea of choosing. M S such that. M S +t M S is positive

semi-definite. So, we have a positive construction. So, we have the problem.
We pose

@seismicisolation
@seismicisolation
456 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

E[ { ((
BT − M T
) ) ]
. E(wh , z h ) = (Awh , z h )(L 2 (T,R)) p − (whint − whext ), z int
h ds .
∂T 2 2
T ⊂Ω
(7.9.2)
We have
.α||wh ||20,Ω ≤ E(wh , wh )

and
. E(u h , vh ) = (F, vh )(L 2 (Ω,R)) p

for all .vh ∈ Vh .

Let two elements .T1 and .T2 be adjacent along a face . S. We make the following
assumption

. MT1 ,S = MT2 ,S = M S , with Ms + M S∗ positive semi-definite (7.9.3)

for any face . S inside the domain .Ω.

Theorem 7.9.1 We assume that the hypothesis (7.9.3) is satisfied. Then, the problem
(7.9.1) admits a unique solution .u h ∈ Wh and, we have

α||u h ||(L 2 (Ω,R)) p ≤ ||F||(L 2 (Ω,R)) p .

. (7.9.4)

Proof Consider the expression defined in (7.9.2). We have

{
1 1 ( )
(Au h , u h )(L 2 (T,R)) p =
. ((A + A∗ )u h , u h )(L 2 (T,R)) p + h , uh
BT u int int
2
ds.
2 2 ∂T

E{
On the other hand,
. (BT u ext
h , u h )2 ds = 0
int

T ⊂Ω ∂T

h = 0 for any part of .∂ T ⊂ ∂Ω. Finally,

with the convention .u ext

E{ E{
. (MT (u int
h − u h , u h )2 ds =
ext int
(MT (u int
h − u h ), u h − u h )2 ds,
ext int ext

T ⊂Ω ∂T S⊂S S

where .S denotes the set of faces of the elements .T , with the convention .u ext
h = 0 if
. S ⊂ ∂Ω. So, he comes

E (1 ) E {
. E(u h , u h ) = (A + A∗ )u h , u h + (MT (u int
h − u h ), u h − u h )2 ds.
ext int ext
2 (L 2 (T,R)) p S
T ⊂Ω S⊂S

On the other hand,

@seismicisolation
@seismicisolation
7.9 Discontinuous Method 457
E
. E(u h , u h ) = (F, u h )(L 2 (T,R)) p .
T ⊂Ω

By using the positivity of . A, the hypothesis (7.9.3) and the Cauchy-Schwarz inequal-
ity, we obtain the inequality (7.9.4). Q.E.D.

Example 7.9.1 We consider the problem

du
. + σ u = f on ]0, 1[,
dx

with .σ > 0. We have . M(0) = 1, . M(1) = 1, . B(0) = −1 and . B(1) = 1. Then,

I −1 {
E xi+1 ( )
du h 1 − + −
. + σ uh − f vh d x − (1 − m i+1 )(u i+1 − u i+1 )vi+1
xi dx 2
i=0
1
− (−1 − m i )(u i+ − u i− )vi+ = 0
2

with .m 0 = 1, .m I = 1 and .m i ≥ 0 .1 ≤ i ≤ I − 1. Note

v
. h |(x ,x )
i i+1
= vi+ 21 .

Hence,

{ xi+1 1 1
.hσ u i+ 1 − f dx − (1 − m i+1 )(u i+ 1 − u i+ 3 ) + (1 + m i )(u i+ 1 −u i− 1 ) = 0
2 xi 2 2 2 2 2 2

1 ≤ i ≤ I − 2.

{ h
1
. hσ u 21 − f d x − (1 − m 1 )(u 21 − u 23 ) + u 21 − u(0) = 0
0 2
{ 1
1
. hσ u I − 21 − f d x + (1 + m I −1 )(u I − 21 − u I − 23 ) = 0.
1−h 2

We choose for example .m i = 1 for all .i, we obtain

{ xi+1
. hσ u i+ 21 + u i+ 21 − u i− 21 = f d x, 0 ≤ i ≤ I − 1 (7.9.5)
xi

u
. −1
2
= u(0).

@seismicisolation
@seismicisolation
458 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

It is the retrograde Euler which is precise, of the order of .h. If .σ = 0, Eq. (7.9.5)
gives { xi+1
.u i+ 1 − u i− 1 = f d x, 0 ≤ i ≤ I − 1
2 2
xi

u
. −1
2
= u(0).

Hence, if we know .u(0) we, can find all the others.

Example 7.9.2 We take .m i = 1 for all .i and

x − xi − xi+1 − x +
u
. h |(xi ,xi+1 ) = u i+1 + ui .
h h
We have
I −1 {
E xi+1 ( )
du h
. + σ uh − f vh d x + (u i+ − u i− )vi+ = 0
i=0 xi dx

with
u − = u(0).
. 0

Furthermore, { ( )
h
du h
. + σ u h vh d x + (u + − +
0 − u 0 )v0 = 0
0 dx

and { xi+1 ( )
du h
. + σ u h vh d x + (u i+ − u i− )vi+ = 0.
xi dx

For .vh = 1, we have

u− +
1 − u0 h
. h + σ (u − + u+ + −
0 ) + u 0 − u 0 = 0.
h 2 1
For .vh = x, we have
( )
h2 u− +
1 − u0 h h u+ + u−
. +σ 4 0 1
+ hu −
1 = 0.
2 h 6 2 2

Hence, ( )
h h +
. 1+σ u− −
1 + σ u0 = u0 (7.9.6)
2 2

and ( ) ( )
h − h
. 1 + 2σ u1 − 1 − σ u+
0 = 0. (7.9.7)
3 3

@seismicisolation
@seismicisolation
7.9 Discontinuous Method 459

By multiplying Eq. (7.9.6) by .1 − σ h3 and, Eq. (7.9.7) by .σ h2 , and by summing, we

find ( ) ( )
h h2 h h2 h
. 1+σ − σ2 + σ + σ2 u−
1 = 1 − σ u−
0.
6 6 2 3 3

Proving ( ) ( )
h h2 h
. 1 + 2σ + σ2 u−
1 = 1 − σ u−
0.
3 6 3

Thus,
1 − σ h3
u− =
. 1 2 u−
0.
1 + 2σ h3 + σ 2 h6

Gold,
1−x
.
1 + 23 x + 16 x 2
( )( ( ) ( ) )
1 2 1 2 2 1 2 2 2 1 2 3
. = 1− x 1− x − x + x+ x − x+ x + o(x )
4
3 3 6 3 6 3 6
( )( )
1 2 5 2
= 1− x 1 − x + x 2 − x 3 + o(x 4 )
3 3 18 27
1 1
= 1 − x + x 2 − x 3 + o(x 4 ).
2 6
On the other hand,
u(x) = u(0)e−σ h
.

is the exact solution. Hence, all calculations done

u − − u(h) = o(σ h 4 ).
. 1

So, we have a method of the order of .o(h 3 ) to solve the linear system because on
each cell, the error is of the order .o(σ h 4 ). Hence,

1
. o(σ h 4 ) = o(h 3 ).
h
Example 7.9.3 Either the problem

du
. + σ u = f on ]0, 1[,
dx

u(0) = μ,
.

@seismicisolation
@seismicisolation
460 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

with.σ > 0 and.μ ∈ R. We consider the abscissas.xi ,.0 ≤ i ≤ I , distinct, with.x0 = 0

and .x I = 1. We set .h i = xi+1 − xi . Let

. Vh = {vh ∈ L 2 ((0, 1), R) such that vh |[xi ,xi+1 ] ∈ Pk }.

The discontinuous finite element method is written: Find .u h ∈ Vh such that

I −1 {
E ( ) I −1
xi+1
du h 1E
+ σ u h − f vh d x − (1 − m i )(u i− − u i+ )vi−
. xi d x 2
i=0 E I −1 i=1
− 21 i=1 (−1 − m i )(u i+ − u i− )vi+ + (u + +
0 − μ)v0 = 0,

for all .vh ∈ Vh , where .u i± = vh (xi ± 0), and where the .m i , .1 ≤ i ≤ I − 1, are scalars
.≥ 0 . In particular, if we work with polynomials of degree .0, we obtain the system

h (σ u i+ 21 − f i+ 21 ) − 21 (1 − m i+1 )(u i+ 21 − u i+ 23 )
. i

1
. + (1 + m i )(u i+ 21 − u i− 21 ) = 0, for 1 ≤ i ≤ I − 1, (7.9.8)
2
1
h (σ u 21 − f 21 ) − (1 − m 1 )(u 21 − u 23 ) + (u 21 − μ) = 0,
. 0 (7.9.9)
2
1
. h I −1 (σ u I − 21 − f I − 21 ) + (1 + m I −1 )(u I − 21 − u I − 23 ) = 0, (7.9.10)
2
where .u i+ 21 designates the restriction of .u h to the segment .[xi , xi+1 ] and . f i+ 21
designates the average of . f on this segment. We note that if we choose .m i = 1,
.1 ≤ i ≤ I − 1, the linear system is uncoupled into the set of equations

h (σ u i+ 21 − f i+ 21 ) + u i+ 21 − u i− 21 = 0, 0 ≤ i ≤ I − 1,
. i

with the convention .u − 21 = μ. If we then assign the value .u i+ 21 to the point .xi+1 , we
express the implicit Euler scheme and, we have

u(xi+1 ) − u i+ 21 = O(h), for all i.

.

Remark 7.9.2 In the case where one of the matrices . Ai (for example . A1 ) can be
reduced to a diagonal matrix with coefficients .≥ 0, the value .x1 can be treated as
a time variable. Suppose .Ω =]0, X 1 [×Ω , , where .Ω , is an open set of .Rn−1 . We
consider the abscissas .x1, j ∈ [0, X 1 ], .0 ≤ j ≤ J − 1, .x1,J −1 = X 1 and, we choose

. M(x1, j , x , ) = |B(x1, j , x , )|.

The problem then reduces to a set of subproblems on the domains .]x1, j , x1, j+1 [×Ω , ,
.0 ≤ j ≤ J − 1, processed sequentially from . j = 0 to . j = J − 1.

@seismicisolation
@seismicisolation
7.9 Discontinuous Method 461

z
y

x1,0 x1,1 x1,2 x1,3 x = x1

Remark 7.9.3 In the case of the transport equation in plane geometry, we have an
analogous situation, for example in dimension .2, the equation is written

∂ϕ ∂ϕ
μ
. +ν + σ ϕ = f, (x, y) ∈ Ω,
∂x ∂y

.ϕ = 0 on .∂− Ω = {(x, y) ∈ ∂Ω such as μn x + νn y < 0}. We can verify that we are

in the framework of the Friedrichs systems. If the domain .Ω is a union of . N triangles
and (or) convex quadrilaterals, we can show that we can number these elements .T j ,
.1 ≤ j ≤ N such that for all . j, we have

⎡ ⎤
| | | |
∂ Tj ⊂ ⎣
. − ∂+ Ti ⎦ ∂− Ω.
i< j

We can thus perform the calculations in a decoupled way, element by element, going
from upstream to downstream following the characteristic direction.(μ, ν), the matrix
. M T being equal to .−BT on .∂− T and . BT on .∂+ T , for . T ⊂ Ω.

@seismicisolation
@seismicisolation
462 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

13
14
10 9
11 12
4
3 7 8
2
6
1
5

Item numbering is not necessarily unique. ♦

Remark 7.9.4 We still have a situation of the same type for the transport equation
in one-dimensional spherical geometry

∂ 2 ∂
.μ (r ϕ) + r (1 − μ2 )ϕ + σ r 2 ϕ = r 2 f
∂r ∂μ

or equivalently

∂ϕ ∂ϕ
μr 2
. + r (1 − μ2 ) + σ r 2 ϕ = r 2 f, for 0 < r < R, −1 < μ < 1,
∂r ∂μ

with .ϕ(R, μ) = 0 for .μ < 0. It is checked that this boundary condition is the only
one to impose and that the system is positive if the norm with weight is used
{ R { 1
. r 2 ϕ 2 dr dμ.
0 −1

To solve this problem numerically, we cut the domain.]0, R[×] − 1, 1[ into rectangles

. K i j =]ri , ri+1 [×]μ j , μ j+1 [,

0 ≤ i ≤ I − 1, .−J ≤ j ≤ J − 1, with .0 = r0 < r1 < · · · < r I = R, .−1 = μ−J <

.

μ−J +1 < · · · < μ0 = 0 < μ1 < · · · < μ J = 1. We calculate the approximate solu-
tion successively on the rectangles . K I −1,−J , K I −2,−J , · · · , K I −1,−J +1 , K I −2,−J +1 ,
· · · , etc. following the numbering given in the figure (this numbering is not the only
one possible).

@seismicisolation
@seismicisolation
7.9 Discontinuous Method 463

26 27 28 29 30

21 22 23 24 25

16 17 18 19 20
r
15 14 13 12 11

10 9 8 7 6

5 4 3 2 1

7.9.2 Error Estimates for Discontinuous Methods

We consider the problem

. Au = F

(B − M)u = 0.
.

Let .Vh ⊂ (L 2 (Ω, R)) p be the space defined by

. Vh = {vh ∈ (L 2 (Ω, R)) p such that vh |T ∈ Pk × Pk × · · · × Pk }.

We search .u h ∈ Vh such that

E[ { ((
BT − M T
) ) ]
. (Au h − F, z h )(L 2 (T,R)) p − (u int
h − u h ), z h
ext int
ds = 0
T ⊂Ω ∂T 2 2

for all .z h ∈ Vh , with

E
n
. BT = νiT Ai ,
i=1

@seismicisolation
@seismicisolation
464 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

ν T exterior normal vector, and . MT matrix defined on .∂ T . For .wh , .z h ∈ Vh , we pose

.

E[ { ((
BT − M T
) ) ]
. E(wh , z h ) = (Awh , z h )(L 2 (T,R)) p − (whint − whext ), z int
h ds .
∂T 2 2
T ⊂Ω

On the other hand,

. E(wh , wh ) = (F, wh )(L 2 (Ω,R)) p − E(vh , wh ) (7.9.11)

E[ { ((
BT − M T
) ) ]
= (A(u − vh ), wh )(L 2 (T,R)) p + (whint − whext ), z int
h ds
∂T 2 2
T ⊂Ω

with .vhext = 0 on .∂Ω. If . M S +t M S either positive semi-definite and . A is positive,

then
.α||wh ||0,Ω ≤ E(wh , wh )
2

and
. E(u h , vh ) = (F, vh )(L 2 (Ω,R)) p

for all .vh ∈ Vh with . S is beside. Let

. wh = u h − vh .

Then,
α||u h − vh ||20,Ω
.

. ≤ E(u h − vh , u h − vh )
≤ (F, u h − vh )(L 2 (Ω,R)) p − E(vh , u h − vh )
≤ (Au, u h − vh )(L 2 (Ω,R)) p − E(vh , u h − vh )
E[ { ((
BT − M T
) ) ]
≤ (A(u − vh ), u h − vh )(L 2 (T,R)) p + (vhint − vhext ), u int
h − vh
int
ds .
∂T 2 2
T ⊂Ω

We choose .vh = rh u the interpolated continuous. For . S /⊂ ∂Ω, since we interpolate

by continuous functions, we have

(rh u)int = rh u = (rh u)ext

.

and so
{ (( ) )
BT − M T
. (rh u int − rh u ext ), u int
h − rh u
int
ds = 0.
∂T 2 2

For . S ⊂ ∂Ω, we have

.(B − M)u = 0

@seismicisolation
@seismicisolation
7.9 Discontinuous Method 465

and
{ {
l(S)
. ((B − M)(u − rh u), whint )2 ds = (( - - u −-
B − M)(- r-
u ), -
whint )2 ds.
S 1 -
S

We pose {
. u, -
z(- wh ) = (( - - u −-
B − M)(- r-
u ), -
whint )2 ds.
-
S

The mapping
-
u −→ z(-
. u, -
wh )

-) into .R of norm .≤ -
is linear and continuous from . H k+1 (T c||-
wh ||0,T- and

. p, -
z(- wh ) = 0

for all . -
p ∈ Pk . Hence,

. |z(-
u, -
wh )| ≤ c|-
u |k+1,T-||-
wh ||0,T-
≤ ch k+1 |u|k+1,T h −1 h −1 ||wh ||0,T
≤ ch k−1 |u|k+1,T ||wh ||0,T .

So,
{ (( ) )
BT − M T
. (vhint − vhext ), u int
h − vhint ds ≤ ch k |u|k+1,T ||wh ||0,T .
∂T 2 2

Thus,
E{ (( ) )
BT − M T
. (vhint − vhext ), u int
h − vhint ds
T ⊂Ω ∂T 2 2
E
. ≤ ch k |u|k+1,T ||wh ||0,T
T ⊂Ω
( ) 21 ( ) 21
E E
≤ ch k
|u|2k+1,T ||wh ||20,T
T ⊂Ω T ⊂Ω

≤ ch k |u|k+1,Ω ||wh ||0,Ω .

On the other hand,

@seismicisolation
@seismicisolation
466 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
E E
. (A(u − rh u), u h − rh u)(L 2 (T,R)) p ≤ ||u − rh u||1,T ||u h − rh u||0,T
T ⊂Ω T ⊂Ω
( )1 ( )1
E 2 E 2
≤ ||u − rh u||21,T ||u h − rh u||20,T
T ⊂Ω T ⊂Ω
≤ ||u − rh u||1,Ω ||u h − rh u||0,Ω
≤ ch k |u|k+1,Ω ||u h − rh u||0,Ω .

Thus,
.||u h − rh u||0,Ω ≤ ch k |u|k+1,Ω

for all .k ≥ 1

Theorem 7.9.5 We assume that . M S + M S∗ is positive semi-definite, for all . S ⊂ S ,

where .S is the set of faces. We assume that .Wh is constructed using polynomials of
degree .≤ k (triangular elements, isoparametric quadrilaterals, and generalization
to .n ≥ 3) and that the exact solution .u ∈ (W k+1,∞ (Ω))2 . Then,

.||u − u h ||0,Ω = O(h k ). (7.9.12)

Proof In the relation (7.9.11), we replace .vh by the interpolation .rh u of .u (.rh u = u
at the nodes). Edge terms corresponding to faces . S /⊂ ∂Ω disappear. The boundary
terms corresponding to faces . S ⊂ ∂Ω are bounded as in Theorem 7.9.1. Finally, we
have
.(A(u − r h u), wh )(L 2 (T,R)) p ≤ ch ||u||k+1,T ||wh ||0,T .
k

We deduce the inequality (7.9.12). Q.E.D.

Remark 7.9.6 The markup (7.9.12) is not optimal and does not allow to justify the
numerical results obtained by using polynomials of degree zero within the framework
of the transport equation. ♦

Question: Can we do better.

In fact, the mapping
-
u −→ z(-
. u, -
wh )

-) into .R of norm .≤ -
is linear and continuous from .W k+1,∞ (T c||-
wh ||0,T- and

. p, -
z(- wh ) = 0

@seismicisolation
@seismicisolation
7.9 Discontinuous Method 467

for all . -
p ∈ Pk . Hence,

|z(-
. u, -
wh )| ≤ c|-
u |k+1,T-||-
wh ||0,T-
≤ ch k+1 ||u||k+1,∞,T h −1 ||wh ||0,T
≤ ch k ||u||k+1,∞,T ||wh ||0,T .

So,
{ (( ) )
BT − M T
. (vhint − vhext ), u int
h − vhint ds ≤ ch k+1 ||u||k+1,∞,T ||wh ||0,T .
∂T 2 2

Thus,
|{ |
| |
| |
| ((B − M)(rh u − u), u h − vh )2 ds | ≤ ch |u|k+1,T ||u h − vh ||0,T .
int int k+1
.
S

We have
E{ (( ) ) E
BT − M T
. (vhint − vhext ), u int
h − vh
int
ds ≤ ch k+1 |u|k+1,∞,Ω ||wh ||0,T .
2
T ⊂Ω ∂ T 2 T ⊂Ω

Gold,
E E
. ||wh ||0,T = 1 × ||wh ||0,T
T ⊂Ω T ⊂Ω
( ) 21 ( ) 21
E E
≤ 12 ||wh ||20,T
T ⊂Ω T ⊂Ω
− 21
≤h ||wh ||0,Ω .

So,
E{ (( ) )
BT − M T 1
. (vhint − vhext ), u int
h − vh
int
ds ≤ ch k+ 2 ||u||k+1,∞,Ω ||wh ||0,Ω .
2
T ⊂Ω ∂ T 2

Thus,
| |
| E { (( B − M ) ) |
| T T | 1
.| (vh − vh ), u h − vh
int ext int int
ds | ≤ ch k+ 2 ||u||k+1,∞,Ω ||u h − rh u||0,Ω .
| ∂T 2 2 |
T ⊂Ω

It follows that
1
. ||u h − rh u||0,Ω ≤ c(h k |u|k+1,Ω + h k+ 2 ||u||k+1,∞,Ω ).

@seismicisolation
@seismicisolation
468 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Theorem 7.9.7 We assume . M S + M S∗ is positive definite (uniformly in .h) for all

. S ⊂ S , . S / ⊂ ∂Ω. We assume that . Wh is constructed with simplicial elements and
polynomials of degree .≤ k. Then,

1
. ||u − u h ||0,Ω = O(h k+ 2 ).
♦

7.10 Example

Let .Ω be a polygonal domain of .R2 with piecewise smooth boundary .Γ = ∂Ω.

Given a function . f ∈ L 2 (Ω, R), we search a function .u defined on .Ω checking

. − Δu = f in Ω
u = 0 on Γ = ∂Ω.

The variational formulation is given by: Find .u ∈ V := H01 (Ω) such that

a(u, v) = L(v)
.

for all .v ∈ V with {

.a(u, v) = ∇u · ∇v d xdy
Ω

and {
. L(v) = f v d xdy.
Ω

The approximate problem is given by: Find .u h ∈ Vh such that

.a(u h , vh ) = L(vh )

for all .vh ∈ Vh , with .Vh is a subspace of .V of finite dimension. If .Vh ⊂ H01 (Ω), then
a finite element of class .C 0 , i.e., a conforming finite element. If .Vh /⊂ H01 (Ω), then
a non-conforming finite element.

@seismicisolation
@seismicisolation
7.10 Example 469

−
−

−
If, we have a non-conforming finite element, then
E{
a (u h , vh ) =
. h ∇u h · ∇vh d xd y
T ⊂Ω T

for all .vh ∈ Vh with

. Vh = {vh ∈ L 2 (Ω, R) such that vh |T ∈ PT + the conditions at the limits}.

Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh . This is equivalent to find .u h ∈ Vh such that

{ {
. ∇u h · ∇vh d xd y = f vh d xd y (7.10.1)
T T

for all .T ⊂ Ω.

7.10.1 In Dimension 1

Equation (7.10.1) becomes

{ {
x j+1
∂u h ∂vh x j+1
. dx = f vh d x (7.10.2)
xj ∂x ∂x xj

for all .vh ∈ Vh .

First idea: We pose .Vh the set

. Vh = {vh ∈ L 2 (0, 1) such that vh |(xi ,xi+1 ) = cte}.

@seismicisolation
@seismicisolation
470 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Hence, Eq. (7.10.2) gives { x j+1

. f d x = 0.
xj

This condition is not always true.

Second idea: We pose .Vh the set

. Vh = {vh ∈ L 2 (0, 1) such that vh |(xi ,xi+1 ) ∈ P1 }.

−

−
xj x j+1

The two functions in the figure above are the basic functions of the .Vh space. We
pose
+
.v j = vh (x j + 0), v−j+1 = vh (x j+1 − 0).

Hence,
du h u −j+1 − u +j
. = ∈ P0
dx h
and
dvh v−j+1 − v+j
. = ∈ P0 .
dx h
Thus, Eq. (7.10.2) gives
( )( ) {
u −j+1 − u +j v−j+1 − v+j x j+1
. h = f vh d x (7.10.3)
h h xj

for all .vh ∈ Vh . In particular, for the elements of basis .vh(1) and .vh(2) with

x − xj
v(1) =
. h
h
and
x j+1 − x
. h v(2) = .
h
Furthermore,
v(1)− = vh(1) (x j+1 − 0) = 1, v(2)−
. j+1 j+1 = 0

@seismicisolation
@seismicisolation
7.10 Example 471

and
v(1)+ = vh(1) (x j + 0) = 0, v(2)+
. j j = 1.

Equation (7.10.3) gives

{
1 − x j+1
x − xj
. (u − u +j ) = f (x) dx (7.10.4)
h j+1 xj h

and {
1 − x j+1
x j+1 − x
.− (u − u +j ) = f (x) d x. (7.10.5)
h j+1 xj h

Thus, Eq. (7.10.4) becomes

{ x j+1
−
.u j+1 − u +j = f (x)(x − x j ) d x (7.10.6)
xj

and Eq. (7.10.5) becomes

{ x j+1
. − (u −j+1 − u +j ) = f (x)(x j+1 − x) d x. (7.10.7)
xj

By summing (7.10.6) and (7.10.7), we obtain

{ x j+1
. f (x)(x j+1 − x j ) d x = 0.
xj

Again, { x j+1
. f (x) d x = 0. (7.10.8)
xj

This last condition is not always true. Hence, Eq. (7.10.8) as the necessary condition
for the existence of a solution.

7.10.2 In Dimension 2

Is there a solution to the approximate problem? The answer is no. Indeed, let .vh ∈ Vh
such that
.ah (vh , vh ) = 0.

Can we show that .vh = 0. Indeed, let .vh ∈ Vh such that

a (vh , vh ) = 0.
. h

@seismicisolation
@seismicisolation
472 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

E{
Then,
. |∇vh |2 d xd y = 0.
T ⊂Ω T

Thus, for all .T ⊂ Ω, we have

{
. |∇vh |2 d xd y = 0.
T

It follows that
|∇vh ||T = 0.
.

Thus,
v
. h |T = cte.

Hence, we cannot say that .vh = 0 on .Ω. Hence, we are missing information linking
one triangle to another. For that let’s pose

1
. J (vh ) = ah (vh , vh ) − L(vh ).
2
If .Vh is built as before, we have: Find .u h ∈ Vh such that

. J (u h ) = inf J (vh ).
vh ∈Vh

This is equivalent to: Find .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh . In the general case this problem does not have a single solution.
Hence, the idea of taking

. K = {vh ∈ Vh with some conditions of continuity to be defined}.

We know that the solution, if it exists, is characterized by

(J , (u h ), vh − u h ) ≥ 0
.

for all .vh ∈ K .

In dimension 1: Continuity at points .x j , so we fall on the conforming method. From
where, we have nothing new.
In dimension 2: Let . K be the set

. K = {vh ∈ Vh such that vh is continuous in the middle of the sides}.

@seismicisolation
@seismicisolation
7.10 Example 473

So, we come across the non-conforming method. Hence, there is nothing new. So,
the idea is to introduce another idea «it’s penalization».

7.10.3 The Penalization

Let .ε > 0 and let’s pose

{
1 E
. ε J (vh ) = J (vh ) + (v+ − vh− )2 ds.
ε T ⊂Ω ∂ T h

2
1

vh+ of the triangle 1

−

In dimension 1: .vh (x0 ) = 0 and .vh (x I ) = 0. Hence,

I −1 { ( )2 { I −1
1E x j+1
∂vh 1
1E
. Jε (vh ) = dx − f vh d x + (vh (x +j ) − vh (x −j ))2 .
2 j=0 xj ∂x 0 ε j=1

Consider the problem: Find .u εh ∈ Vh such that

J (u εh ) = inf Jε (vh ),
. ε (7.10.9)
vh ∈Vh

with .Vh = {vh ∈ L 2 (Ω, R) such that vh |(xi ,xi+1 ) ∈ Pk }, .k ∈ N.

If .k = 0, let’s pose .v j+ 21 = vh |(x j ,x j+1 ) . If the solution of the problem (7.10.9) exist,
then its gradient is zero at point .u εh , i.e.,

I −1 {
E { I −1
x j+1 ∂u εh ∂vh 1 2E ε +
. dx − f vh d x + (u h (x j ) − u εh (x − + −
j ))(vh (x j ) − vh (x j )) = 0
∂x ∂x ε
j=0 x j 0 j=1
(7.10.10)
for all .vh ∈ Vh . We choose
{
1 on ]x j , x j+1 [
v =
. h
0 elsewhere.

@seismicisolation
@seismicisolation
474 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Equation (7.10.10) gives

{ x j+1
2[ ε + ]
. − f dx + (u h (x j ) − u εh (x −j )) + (u εh (x +j+1 ) − u εh (x −j+1 ))(−1) = 0.
xj ε

With notations .v j+ 21 = vh (x +j ) and .v j+ 21 = vh (x −j+1 ), we have

{ x j+1
2[ ε ]
. − f dx + (u j+ 1 − u εj− 1 ) − (u εj+ 3 − u εj+ 1 ) = 0.
xj ε 2 2 2 2

Hence, {
2[ ε ] x j+1
. −u j+ 3 + 2u εj+ 1 − u εj− 1 = f d x.
ε 2 2 2
xj

Thus, ( { )
1 [ ε ε ε
] ε 1 x j+1
.− u 3 − 2u 1 + u 1 = f dx .
h 2 j+ 2 j+ 2 j− 2 2h h xj

On the other hand,

. f (x) = f (x j+ 21 ) + (x − x j+ 21 ) f , (x j+ 21 ) + o(h 2 ).

So,
{ x j+1 { x j+1 { x j+1
. f (x) d x = f (x j+ 21 ) d x + (x − x j+ 21 ) f , (x j+ 21 ) d x + h o(h 2 ).
xj xj xj

Further, { x j+1
. (x − x j+ 21 ) f , (x j+ 21 ) d x = 0
xj

hence { x j+1
1
. f (x) d x = f (x j+ 21 ) + o(h 2 ).
h xj

For .ε = 2h, we have

⎛ ⎞⎛u1 ⎞ ⎛
f (x 12 )
⎞ ⎛ ⎞
2 −1 0 0 ··· 0 o(h 2 )
⎟⎜ ⎟ ⎜ f (x 32 ) ⎟
2
⎜ −1 −1 ··· ⎟ ⎜
2 ⎟
⎟⎜ ⎟ ⎜ ⎜ o(h 2 ) ⎟
2 0 0 u 23
⎜
⎜ .. .. .. .. ..⎟⎜ ⎟ ⎜ ⎟ ⎜
f (x 5 ) ⎟ ⎜ o(h ) ⎟
1 ⎜ . . . . .⎟⎜ u5 ⎟ ⎜ ⎟
⎜ 0 ⎟⎜ 2 ⎟ ⎜ 2 ⎟
.. ⎟ + ⎜ .. ⎟
⎜ . .. .. .. .. ⎟ ⎜ ...
⎜ ⎟=⎜
⎟ ⎜ . ⎟ ⎜ . ⎟ .
h ⎜ ..
.
2
⎜ . . . . 0 ⎟⎟⎜ ⎟ ⎜ ⎟ ⎜
⎜ .
⎟
⎟
⎜ . ⎟⎜ . ⎟ ⎜ .. ⎟ ⎜ . ⎟
.. .. .. ⎜ . ⎟ ⎜ . ⎟ .
⎝ .. . . . −1 ⎠ ⎝ . ⎠ ⎝ ⎠ ⎝ . ⎠
.. .. ..
0 ··· ··· 0 −1 2 . .

@seismicisolation
@seismicisolation
7.10 Example 475

. J is strictly convex continuous and

. lim J (v) = +∞.

||v||→∞

Furthermore, . J is differentiable and

I −1
1E
. Φ= (vh (x +j ) − vh (x −j ))2
ε j=1

is convex, continuous, positive, differentiable, and

J (u εh ) = inf Jε (vh ).
. ε
vh ∈Vh

Then, .u εh exist and .u εh weakly converges to .u when .ε → 0. The absence of continuity

implies that the Poincaré inequality does not apply. Therefore, we will introduce the
Lagrengian E
.L (v, q) = J (v) + q j (v+j − v−j ),
j

with .q j Lagrenge multiplier. There is . p j ∈ R such that

E
a (u h , vh ) − L(vh ) +
. h p j (v+j − v−j ) = 0
j

with E
. p j (u +j − u −j ) = 0.
j

The couple .(u, p) is a saddle point. So, we can use Uzawa algorithm:

.q k+1
j = q kj + ρ((u kj )+ − (u kj )− )

and E
a (u kh , vh ) − L(vh ) +
. h q kj (v+j − v−j ) = 0.
j

One thus falls on an elliptical problem which does not have uniqueness of solution
because one does not have the coercivity. If we add the continuity, we will find the
coercivity and, we find the standard problem.

@seismicisolation
@seismicisolation
476 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

7.11 One-Dimensional Transport Equations in Plane

Symmetry

We consider the problem

du
. + σ u = f if x ∈]0, 1[ (7.11.1)
dx
.u(0) = μ. (7.11.2)

We assume that .u h (x j ) known. We want to calculate .u h (x j+1 ). Indeed, we have

{ x j+1 ( )
du h
. + σ uh − f vh d x = 0 (7.11.3)
xj dx

with .vh is a test function. We search .u h ∈ Pk .

Question: How many equations do we write knowing that we know .u h (x j ) (a point)
for any .vh in a space of dimension .k + 1 (for example . Pk ).
k = 1: Equation (7.11.3) gives
.

{
u j+1 − u j u j+1 + u j x j+1
.h + σh = f (x) d x.
h 2 xj

What can be done with the discontinuity? The idea that makes us walk is to write

u
. h |[x j ,x j+1 [ = u h |]x j ,x j+1 [ + H (x − x j )(u h (x +j ) − u h (x −j ))

and
du h du h
. = (interior ) + δ(x−x j ) (u h (x +j ) − u h (x −j )),
dx dx

with . H (x − x j ) is the Heaviside function at point .x − x j and .δ(x−x j ) is the mass of

Dirac at point .x − x j . Let us look for .u h ∈ Pk , knowing .u h (x −j ) known, for all .vh in
a space of dimension .k + 1. Therefore,
{ x j+1 ( )
du h
. + σ uh − f vh d x + (u h (x +j ) − u h (x −j ))vh (x +j ) = 0. (7.11.4)
xj dx

For .k = 0: We replace in (7.11.4) .vh by .1, we find

{ x j+1
u (x +j ) − u h (x −j ) + σ hu h |]x j ,x j+1 [ −
. h f (x) d x = 0.
xj

@seismicisolation
@seismicisolation
7.11 One-Dimensional Transport Equations in Plane Symmetry 477

We notice.u j+ 21 = u h (x +j ) = u h |]x j ,x j+1 [ and.u j− 21 = u h (x −j ). Equation (7.11.4) gives

{ x j+1
1 1
. (u j+ 21 − u j− 21 ) + σ u j+ 21 = f (x) d x.
h h xj

This is Euler’s retrograde method.

For .k = 1: We replace in (7.11.4) .vh by .1, we find
{
u −j+1 − u +j h x j+1
. h + σ (u −j+1 + u +j ) − f (x) d x + u +j − u −j = 0.
h 2 xj

We replace in (7.11.4), .vh by .x − x j and, using Proposition 2.6.8, we get

{ x j+1 { x j+1
du h du h
. (x − x j ) d x = (x − x j ) d x
xj dx dx xj
( )
du h (x j+1 − x j ) + (x j − x j )
= h
dx 2
u −j+1 − u +j h
= h
h 2
and
{ x j+1
σh ( )
. σ u h (x − x j ) d x = u h (x j+1 )(x j+1 − x j ) + 4u h (x j+ 21 )(x j+ 21 − x j )
xj 6
( − +
)
σh u + u h
u −j+1 h + 4
j+1 j
= .
6 2 2

Equation (7.11.4) gives

− +
( ) {
h u j+1 − u j σh u +j + u −j+1 h x j+1
. h+ 4 + u −j+1 h − f (x)(x − x j ) d x = 0.
2 h 6 2 2 xj

So, we are in front of the following system

{
u −j+1 − u +j h x j+1
. h + σ (u −j+1 + u +j ) − f (x) d x + u +j − u −j = 0
h 2 xj

and

h − σ h2 ( + ) { x j+1
. (u j+1 − u +j ) + u j + 2u −j+1 − f (x)(x − x j ) d x = 0
2 6 xj

@seismicisolation
@seismicisolation
478 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

of unknowns .u +j and .u −j+1 as a function of . f and .u −j . For example for . f = 0, we find

⎧
⎪ h
⎨ u −j+1 − u +j + σ (u −j+1 + u +j ) + u +j − u −j = 0
.
2 2
⎪
⎩ h (u − − u + ) + σ h (u + + 2u − ) = 0.
j+1 j
2 6 j j+1

Which proves ⎧( )
⎪
⎪ h h
⎨ 1+σ u −j+1 + σ u +j = u −j
. ( 2 ) (2 )
⎪
⎪ h − h
⎩ 1 + 2σ u j+1 − 1 − σ u +j = 0.
3 3

Thus, ⎧
⎪ +
⎪ 1 + 2σ h3
⎨uj = 2 u −j
. 1+ σ 2 h6 + 2
3
σh
⎪
⎪ 1−σ h3
⎩ u −j+1 = 2 2 h2 u −j .
1+ 3 σ h+ σ 6

It is the approximate solution of the system (7.11.1)–(7.11.2). The exact solution of

the system (7.11.1)–(7.11.2) is

u(h) = μe−σ h
.

and ( )
− 1 − σ h3
.u 1 = σ 2 h2
μ, u −
0 = μ.
1 + 23 σ h + 6

Let’s seek the error .u(h) − u −

1 . We have

u(h) − u −
. 1
( )
−σ h
1 − σ h3
. = e − μ
1 + 23 σ h + σ 6h
2 2

[ ( )( ( ) )]
−σ h h 2 σ 2h2 1 2 1 2 2 2
= e − 1−σ 1 − σh − + σh + σ h + o(h ) μ
3
3 3 6 2 3 6
[ ( )(
1 2 2 1 2 1
= 1 − σ h + σ h + o(h ) − 1 − σ h
3
1 − σ h − σ 2h2
2 3 3 6
( ) )]
1 2 1 2 2 2
+ σh + σ h + o(h 3 ) μ.
2 3 6

All calculation done

u(h) − u −
. 1 = μ o(h ).
3

@seismicisolation
@seismicisolation
7.11 One-Dimensional Transport Equations in Plane Symmetry 479

Example 7.11.1 Let .Ω =]0, 1[. We consider the problem (one-dimensional trans-
port equations in plane symmetry [11])

du
+. σ u = f (7.11.5)
dx
u(0) =
. 0. (7.11.6)

For.σ = 1, Eq. (7.11.5) can be considered as a symmetric Friedrichs system. We have

. B(0) = −1 and . B(1) = 1. Applying Lemma 7.4.7, we get . M(0) = 1 = M(1). Let
.(B − M)(0) = −2 and .(B − M)(1) = 0. We must impose a boundary condition for
. x = 0 (condition (7.11.6)) and no boundary condition for . x = 1. We can also replace
the condition (7.11.6) by .u(0) = g.
For .σ > 0, Eq. (7.11.5) is positive in the sense of Friedrichs with .α = σ . In fact,

C(x) = σ + σ ∗ = 2σ
.

and, for all . X ∈ R, we have

.
t
XC(x)X = 2σ t X X ≥ 2σ t X X.

−
→
ν
−

0 1

. B(0) = (−1) × 1 = −1 and . B(1) = (1) × 1 = 1. Hence, . M(0) = M(1) = 1. In

fact, . M(0) +t M(0) ≥ 0. Suppose that . M(0) = α, then

. K er (B(0) − M(0)) + K er (B(0) + M(0)) = R.

Hence, . K er (−1 − α) + K er (−1 + α) = R. Thus, .α = 1 or .α = −1. Since . B(0) is

defined negative, then . M(0) = −B(0) = 1. Thus, .α = 1 and . M(0) = 1. Likewise,
we can show that . M(1) = 1. We do not impose any condition on the limit for .x = 1
because .(B − M)(1) = 0 and we impose a condition for .x = 0 (for example .u = g)
because .(B − M)(0) /= 0.
We consider the following division of the interval

0 = x0 < x1 < · · · < x I = 1.

.

Let .Vh be the space of continuous functions whose restriction at each interval
[x j , x j+1 ], .0 ≤ j ≤ I − 1, is a polynomial of . Pk . The space . X h2 is the space of
.
functions from .Vh to zero for .x = x0 . We solve the following approximate problems:

@seismicisolation
@seismicisolation
480 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

(i) Find .u h ∈ X h1 = Vh such that

.

{ 1 ( )
du h
. + σ uh − f vh d x + u h (0)vh (0) = 0 (7.11.7)
0 dx

for all .vh ∈ Vh .

(ii) Find .u h ∈ X h2 such that
.

{ 1 ( )
du h
. + σ uh − f vh d x = 0 (7.11.8)
0 dx

for all .vh ∈ X h2 .

In both cases, we have the error bound

||u − u h ||0,(0,1) ≤ ch k |u|k+1,2,Ω ,

.

where
. h= sup (x j+1 − x j ).
0≤ j≤I −1

When .k = 1, the problems (7.11.7) and (7.11.8) are written, respectively,

{ x1
u1 − u0 2 u1 2u 0 2
. + σ u0 + + = f v0 d x (7.11.9)
x1 − x0 3 3 x1 − x0 x1 − x0 0
( )
x j+1 − x j x j − x j−1
.u j+1 − u j−1 +σ (2u j + u j+1 ) + (2u j + u j−1 ) =
3 3
{ x j+1
f v j d x for 1 ≤ j ≤ I − 1, (7.11.10)
x j−1
( ) {
u I − u I −1 2u I u I −1 2 xI
. +σ + = f vI d x (7.11.11)
x I − x I −1 3 3 x I − x I −1 x I −1

and

u =0
. 0
( )
x j+1 − x j−1 x j − x j−1
u j+1 − u j−1 +σ (2u j + u j+1 ) + (2u j + u j−1 ) =
3 3
{ x j+1
f v j d x for 1 ≤ j ≤ I − 1,
x j−1
( ) {
u I − u I −1 2u I u I −1 2 xI
+σ + = f v I d x,
x I − x I −1 3 3 x I − x I −1 x I −1

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 481

where .u j = u h (x j ) and, where .v j designates the function of .Vh equal to .1 at the point
x j and to zero at the points.xi ,.i /= j, for.0 ≤ j ≤ I . In the case, where.x j+1 − x j = h,
.
for .0 ≤ j ≤ I − 1, Eqs. (7.11.9)–(7.11.10)–(7.11.11) become
( ) {
u1 − u0 2 u1 2u 0 2 h
. +σ u0 + + = f v0 dx
h 3 3 h h 0
{ ( j+1)h
h
u j+1 − u j−1 + σ (u j+1 + 4u j + u j−1 ) = f v j d x for 1 ≤ j ≤ I − 1,
3 ( j−1)h
( ) { h
2u I u I −1
u I − u I −1 + σ h + =2 f v I d x.
3 3 1−h

We have the following majoration

||u − u h ||0,(0,1) = o(h 2 ).

.

To obtain the solution .u h at the different nodes, it is necessary to solve a linear system
whose matrix is .(2k + 1)-diagonal.

7.12 Change to First-Order System

Question: Why don’t we change a second-order problem to a first-order problem

and do the same reasoning as above?

7.12.1 Example in Dimension 1

If, we have a second-order system in the form .−u ,, = f + the boundary conditions,
we make the following change of variable: we set .v = u , . Therefore, .−v, = f . So,
we find a system of two equations with two first-order unknowns.
−

discontinuity derivative so mass of Dirac

In fact,

. − u, + v = 0 (7.12.1)
. − v, = f. (7.12.2)

Equation (7.12.1) gives

I −1 {
E xi+1 E
. (−u ,h + vh ) ph d x − (u i+ − u i− ) pi+ = 0
i=0 xi i

@seismicisolation
@seismicisolation
482 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

for all . ph ∈ Vh and Eq. (7.12.2) gives

I −1 {
E xi+1 E { 1
. −vh, qh dx − (vi+ − vi− )qi+ = f qh d x
i=0 xi i 0

for all .qh ∈ Vh with

. Vh = {vh ∈ L 2 ((0, 1), R) such that vh |(xi ,xi+1 ) ∈ P1 }.

E
Remark 7.12.1 The term . i (u i+ − u i− ) pi+ appeared since .u h is discontinuous.
Indeed,
+ −
.u h |[x ,x [ = u h |]x ,x [ + H (x − x j )(u h (x i ) − u h (x i ))
i i+1 i i+1

and
du h du h
= (interior.) + δ(x−xi ) (u h (xi+ ) − u h (xi− )),
.
dx dx
with . H (x − x j ) is the Heaviside function at point .x − x j and .δ(x−x j ) is the mass of
Dirac at point .x − x j . ♦
Example 7.12.1 We consider the problem

d 2u
. − + u = f on ]0, 1[, (7.12.3)
dx2

u(0) = u(1) = 0.
. (7.12.4)

Let’s pose .v = u , , we have

. − v, + u = f

. − u , + v = 0.

Hence, ( ) ( ) ( ) ( )( ) ( )
u 0 −1 ∂ u 10 u f
. A := + = .
v −1 0 ∂x v 01 v 0

We have ( ) ( ) ( )( )
01 ∂ ϕ1 10 ϕ1
. A∗ ϕ = + .
10 ∂x ϕ2 01 ϕ2

We consider the following division of the interval

.0 = x0 < x1 < · · · < x I = 1,

where .x j = j h, .0 ≤ j ≤ I.

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 483

• Polynomial . P1 : Let .Vh be the space of functions .(ϕh , ψh ) whose components are
.

linear on each interval .[xi , xi+1 ], with .xi = i h, .0 ≤ i ≤ I , and such that .ϕh (0) =
ϕh (1) = 0. Consider the space

. Z h = _ψ j _1≤ j≤I −1 ,

−
ψj

−
0 x j−1 xj x j+1

with .ψ j are the hats functions and .Vh is the space

. Vh = Z h × Z h .

On the other hand,

( ) ( )
01 0 −1
. B(0) = , B(1) =
10 −1 0

and ( ) ( )
α0 1 α1 −1
. M(0) = , M(1) =
−1 0 1 0

with .α0 ≥ 0 and .α1 ≥ 0. Let’s pose . X h∗ be the space

. X h∗ = {(vh1 , vh2 ) ∈ Vh such that vh1 (0) = vh1 (1) = 0}.

In fact, ( ) ( )
α0 0 α1 0
(B +t M)(0) =
. , (B +t M)(1) = .
2 0 −2 0

Remark 7.12.2 In the space . X h∗ that we took at the start, we can replace it with

. X h∗ = {vh ∈ Vh such that (B +t M)vh = 0 at the nodes ∈ ∂Ω}.

Let .u h = (u 1h , u 2h ) ∈ Vh checking

(u h , A∗ ϕh )(L 2 ((0,1),R))2 = (F, ϕh )(L 2 ((0,1),R))2

.

for all .ϕh ∈ X h∗ . Hence,

@seismicisolation
@seismicisolation
484 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{ 1 ( ) { 1 ( ) { 1
dϕh2 dϕh1
. u 1h + ϕh1 dx + u 2h + ϕh2 dx = f ϕh1 d x
0 dx 0 dx 0

for all .ϕh ∈ X h∗ . In particular, for .ϕh1 = ψ j and .ϕh2 = 0, we have

{ 1 { 1 { 1
dψ j
. u 1h ψ j d x + u 2h dx = f ψ j d x.
0 0 dx 0
−

−
0 j −1 j j +1 1

Further, .u 1h ψ j ∈ P2 because .u 1h ∈ P1 and .ψ j ∈ P1 , and, using Proposition 2.6.8, we

obtain
{ xj
h[ 1 ]
. u 1h ψ j d x = (u h ψ j )(x j−1 ) + 4(u 1h ψ j )(x j− 21 ) + (u 1h ψ j )(x j )
x j−1 6
[ ( 1 )( ) ]
h u h (x j ) + u 1h (x j−1 ) ψ j (x j ) + ψ j (x j−1 )
= 4 + u h (x j )
1
6 2 2
h
= (u 1j + u 1j−1 + u 1j )
6
h
= (2u 1j + u 1j−1 )
6
and { x j+1
h
. u 1h ψ j d x = (2u 1j + u 1j+1 ).
xj 6

On the other hand,

{ 1 { x j+1
dψ j dψ j
. u 2h dx = u 2h dx
0 dx x j−1 dx
{ xj { x j+1
1 1
= u 2h d x − u 2h d x.
h x j−1 h xj

Further, using Proposition 2.6.8, we obtain

{
1 u j + u j−1 u 2j + u 2j−1
xj 2 2
1
. u 2h d x = h =
h x j−1 h 2 2

and {
1 x j+1 u 2j + u 2j+1
. u 2h d x = .
h xj 2

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 485

Hence, { 1
dψ j 1
. u 2h d x = [u 2j−1 − u 2j+1 ].
0 dx 2

Thus,
{ 1
h 1 1
. (u j−1 + 4u 1j + u 1j+1 ) + [u 2j−1 − u 2j+1 ] = f ψ j d x, 1 ≤ j ≤ I − 1,
6 2 0

ψ0
−

−
x1
0

{ 1
1 h
. − (u 11 + u 10 ) + (2u 20 + u 21 ) = f ψ0 d x,
2 6 0

{ 1
1 1 h
. (u I + u I −1 ) + (2u 2I + u 2I −1 ) = f ψ I d x,
1
2 6 0

and
u 1 = u 1I = 0.
. 0

In particular for .ϕh1 = 0 and .ϕh2 = ψ j , we have

1 h
. − (u 1j+1 − u 1j−1 ) + (u 2j+1 + 4u 2j + u 2j+1 ) = 0, 1 ≤ j ≤ I − 1.
2 6
We have .2I equations with.2(I + 1) unknown. So, it suffices to know two unknowns,
for example, .u 10 and .u 1I to solve this system.
• . Polynomial Pk : The space .Vh is the space of vector functions .(ϕh , ψh ) such that
.
each component belongs to the space of continuous functions whose restriction at
each interval.[x j , x j+1 ],.0 ≤ j ≤ I − 1 is a polynomial of. Pk . The space. X h2 is defined
by
. X h = {(ϕh , ψh ) ∈ Vh such that ϕh (0) = ϕh (1) = 0}.
2

We solve the following approximate problems:

@seismicisolation
@seismicisolation
486 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

(i) Find .(u h , vh ) ∈ X h1 = Vh such that

.

{ 1( dvh
) { 1( du h
) ( a )
. − + uh − f ϕh d x + − + vh ψh d x + u h (0) − ϕh (0) + ψh (0) +
0 dx 0 dx 2
(a )
u h (1) ϕh (1) + ψh (1) = 0 (7.12.5)
2

for all .(ϕh , ψh ) ∈ Vh .

(ii) Find .(u h , vh ) ∈ X h2 such that
.

{ 1 ( ) { 1( )
dvh du h
. − + u h − f ϕh d x + − + vh ψh d x = 0. (7.12.6)
0 dx 0 dx

for all .(ϕh , ψh ) ∈ X h2 .

For problems (7.12.5) and (7.12.6), we have the error bound
|| ||
|| du ||
.||u − u h ||0,(0,1) + || ||
|| d x − vh || ≤ ch k ||u||k+2,(0,1) .
0,(0,1)

For .k = 1, it is easy to prove that

|| ||
|| du ||
.||u − u h ||0,(0,1) + || || 3

|| d x − vh || ≤ ch 2 ||u||4,(0,1) .
0,(0,1)

The problems (7.12.5) and (7.12.6) lead to the resolution of linear systems having,
respectively, .2(I k + 1) and .2I k unknowns. The solution .(u h , vh ) of the problem
(7.12.6) exactly satisfies the boundary conditions

u (0) = u h (1) = 0.
. h

When .k = 1, the problem (7.12.5) is written

1 (v v1 )
0
− (u 0 + u 1 ) + h
. + =0
2 3 6
( ) { h
2u 0 u1
−(v1 − v0 ) + h + =2 f ϕ0 d x
3 3 0
1 (v v I −1 )
I
(u I + u I −1 ) + h + =0
2 3 6
( ) { 1
2u I u I −1
−(v I − v I −1 ) + h + =2 f ϕI d x
3 3 1−h
{ 1
h
−(vi+1 − vi−1 ) + (u i+1 + 4u i + u i−1 ) = 2 f ϕ I d x, 1 ≤ i ≤ I − 1
3 0
h
−(u i+1 − u i−1 ) + (vi+1 + 4vi + vi−1 ) = 0, 1 ≤ i ≤ I − 1,
3

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 487

where .(u i , vi ) = (u h (xi ), vh (xi )) .0 ≤ i ≤ I and, where .ϕi designates the function
equal to.1 for.x = xi and to zero for.x = x j ,. j /= i,.0 ≤ i ≤ I . For.k = 1, the problem
(7.12.6) is equivalent to the last linear system, whose first and third equations are
replaced by
.u 0 = u I = 0.

Consider the following quadrature formula

{ xi+1
h
. f (x) d x = ( f (xi ) + f (xi+1 )), 0 ≤ i ≤ I − 1.
xi 2

This quadrature formula induces the discrete scalar product

I −1
E h
(u, ϕ)h =
. ((uϕ)(xi ) + (uϕ)(xi+1 )).
i=0
2

We pose the following problem: Find .(u h , vh ) ∈ Vh such that

( ) ( )
1 dvh 1 du h
− + u h − f, ϕh + − + vh , ψh +
2 dx ( ) h( 2 dx ) h
.
1 dϕh 1 dψh (7.12.7)
+ vh , + uh , = 0,
2 dx h 2 dx h

for all .(ϕh , ψh ) ∈ Vh , or equivalently

{ 1 ( )
du h dvh
. − ψh − ϕh d x + (u h − f, ϕh )h + (vh , ψh )h = 0.
0 dx dx

If we put .(u i , vi , f i ) = (u h (xi ), vh (xi ), f (xi )), .0 ≤ i ≤ I , the problem (7.12.7), can
be written in the form of the linear system

vi+1 − vi−1
. − + u i = fi , 1 ≤ i ≤ I − 1
2h
u i+1 − u i−1
. − + vi = 0, 1 ≤ i ≤ I − 1
2h

u 1 − u(0)
. − + v0 = 0
h

u(1) − u I −1
. − + v I = 0.
h

The truncation error at the points .xi , .1 ≤ i ≤ I − 1, is in .o(h 2 ). We assume that the
solution .u of the problem (7.12.3)–(7.12.4) belongs to . H 4 (0, 1) and that the second

@seismicisolation
@seismicisolation
488 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

member . f belongs to . H 2 (0.1). Let .(u h , vh ) be the solution to the problem (7.12.7).
Let .A be the operator ( )
0 −1 ∂
.A := .
−1 0 ∂x

We consider

. X h = _A (wh − Φh ), wh − Φh _0,(0,1) + (wh − Φh , wh − Φh )h ,

with .wh = (u h , vh ), .Φh = (ϕh , ψh ) ∈ Vh . We have

. X h ≥ ||wh − Φh ||20,(0,1) . (7.12.8)

On the other hand, if we put ( )

10
. A=A +
01

and .w = (u, v), we have

X h = _A(w − Φh ), wh − Φh _0,(0,1) + ( f, u h − ϕh )h − _ f, u h − ϕh _0,(0,1) +

.
+_Φh , wh − Φh _0,(0,1) − (Φh , wh − Φh )h .
(7.12.9)

The first term of the expression . X h can be bounded by

3
|_A(w − Φh ), wh − Φh _0,(0,1) | ≤ Ch 2 ||u||4,(0,1) ||wh − Φh ||0,(0,1) .
. (7.12.10)

On the other hand, we have

I −1
( { )
E x j+1
. ( f, ϕh )h − _ f, ϕh _0,(0,1) = α j h f (x j ) − f ϕj dx ,
j=1 x j−1

with
I −1
E
ϕ =
. h αjϕj,
j=1

where the .ϕ j , .1 ≤ j ≤ I − 1, are the linear functions in each interval and such that
ϕ j (xi ) = δi j , .0 ≤ i ≤ I . It is easy to see that
.

⎛ ⎞ 21
I −1
E
.⎝ |α j |2 ⎠ ≤ Ch − 2 ||ϕh ||0,(0,1)
1

j=1

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 489

and | { x j+1 |
| |
| | 5
. |h f (x j ) − f ϕ j d x | ≤ Ch 2 || f ||2,(0,1) , 1 ≤ j ≤ I − 1.
| x j−1 |

We, then, deduce

|( f, u h − ϕh )h − _ f, u h − ϕh _0,(0,1) | ≤ Ch 2 || f ||2,(0,1) ||u h − ϕh ||0,(0,1) . (7.12.11)

.

Similarly, it is easy to show, by replacing .rh w by .(rh w − w) + w, that

3
. |_rh w, wh − Φh _0,(0,1) − (rh w, wh − Φh )h | ≤ Ch 2 ||u||4,(0,1) . (7.12.12)

The inequalities (7.12.8) to (7.12.12) lead to the increase

|| ||
|| du ||
||u − u h ||0,(0,1) + || || 3
.
|| d x − v h || ≤ C(h 2 ||u||4,(0,1) + h 2 || f ||2,(0,1) ).
0,(0,1)

Example 7.12.2 Consider the problem

d 2u
. − + αu = f on ]0, 1[, α ≥ 0 (7.12.13)
dx2

u(0) = u(1) = 0.
. (7.12.14)

We pose .v = u , . Then, {
−v, + αu = f
−u , + v = 0.
.

Hence, ( ) ( ) ( )( ) ( )
0 −1 ∂ u α0 u f
. + = .
−1 0 ∂x v 01 v 0

We pose ( )
0 −1
. A1 =
−1 0

and ( )
α0
. A0 = .
01

Further,

@seismicisolation
@seismicisolation
490 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

∂
. C(x) = A0 (x) + A∗0 (x) − (A1 (x))
( ) ∂x
2α 0
= .
0 2

C(x) is not positive within the meaning of Friedrichs if .α = 0. We have

.

( ) ( )
01 a0 1
. B(0) = , M(0) =
10 −1 0

with .a0 ≥ 0 and

( ) ( )
0 −1 a1 −1
. B(1) = , M(1) =
−1 0 1 0

with .a1 ≥ 0. If .α > 0, we have the existence and uniqueness of the solution of the
problem (7.12.13)–(7.12.14). If .α = 0, we have only existence of solution of the
problem (7.12.13)–(7.12.14).
−

−
0 1

Ih = 1
x j = jh

On the other hand,

( ) ( )
−a1 0 −a0 0
(B − M)(1) =
. , (B − M)(0) = .
−2 0 2 0

v
. h |(x j ,x j+1 ) ∈ P1 , .vh (x j ) = v j , we have

{ 1( dvh
) { 1(
du h
)
1
. − + αu h z h1 d x + − + vh z h2 d x − (−a0 u h (0)z h1 (0) + 2u h (0)z h2 (0))
0 dx 0 dx 2
{ 1
1
− (−a1 u h (1)z h1 (1) − 2u h (1)z h2 (1)) = f z h1 d x.
2 0

We take .z h1 = ϕ j and .z h2 = 0, after we take .z h1 = 0 and .z h2 = ϕ j with .ϕ j are the hats

functions, we obtain with the rating

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 491

dvh v j − v j−1
=
d x |(x j−1 ,x j ) h{
v j+1 − v j−1 αh x j+1
− + (u j−1 + 4u j + u j+1 ) = f ϕjdx 1 ≤ j ≤ I − 1
2 6 x j−1
u j+1 − u j−1 h
− + (v j−1 + 4v j + v j+1 ) = 0, 1 ≤ j ≤ I − 1
2 6 { h
.
v1 − v0 αh a0
− + (2u 0 + u 1 ) + u 0 = f ϕ0 d x
2 6 2 0
u1 − u0 h
− + (2v0 + v1 ) − u 0 = 0
2 6 {1
− v I −v2 I −1 + αh 6
(2u I + u I −1 ) + a21 u I = 1−h f ϕ I d x
− u I −u2 I −1 + h6 (2v I + v I −1 ) + u I = 0.

If .α > 0, we have the existence and uniqueness of the solution of the problem
(7.12.13)–(7.12.14). In fact, for .z h1 = u h and .z h2 = vh , we have
{ 1 ( ) { 1
d a0 2 a1 2
. − (u h vh ) + αu h + vh d x + u 0 − u 0 v0 + u 1 + u 1 v1 =
2 2
f u h d x.
0 dx 2 2 0

Hence, { { {
1 1 1
.α u 2h d x + vh2 d x ≤ f uh d x
0 0 0

because .a0 ≥ 0 and .a1 ≥ 0. Cauchy-Schwarz inequality gives

{ 1 { 1
. (u 2h + vh2 ) d x ≤ cte(α) f 2 d x.
0 0

Thus, we have existence and uniqueness of the solution to the problem (7.12.13)–
(7.12.14).

Example 7.12.3 We can sometimes dispense with premultiplying. In fact, consider

the system
d 2u
.− = f on ]0, 1[,
dx2

. u(0) = u(1) = 0.

We pose .v = u , . Then, {
−v, = f
−u , + v = 0.
.

Hence, ( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + = .
−1 0 ∂x v 01 v 0

@seismicisolation
@seismicisolation
492 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

We pose ( )
0 −1
. A1 =
−1 0

and ( )
00
. A0 = .
01

Gold,

∂
. C(x) = A0 (x) + A∗0 (x) − (A1 (x))
( ) ∂x
00
= .
02

C(x) is not positive within the meaning of Friedrichs. We have

.

( ) ( )
01 a 1
. B(0) = , M(0) =
10 −1 0

with .a ≥ 0 and ( ) ( )
0 −1 a1 −1
. B(1) = , M(1) =
−1 0 1 0

with .a1 ≥ 0. Then,

( ) ( )
−a 0 a2
(B − M)(0) =
. , (B + M)(0) = .
2 0 00

On the other hand, . K er (B − M)(0) + K er (B + M)(0) = R2 and .(M +t M)(0) is

positive semi-definite. Hence,
( ) ( )
a0 a1 0
.(B +t M)(0) = , (B +t M)(1) = .
20 −2 0

Let .W = {ϕ = {(ϕ1 , ϕ2 ) ∈ C 1 ([0, 1], R) × C 1 ([0, 1], R) such that ϕ1 (0) = ϕ1

(1) = 0} and .V = {v = A∗ ϕ, ϕ ∈ W } with
( ) ( ) ( ) ( )( )
u 0 −1 ∂ u 00 u
. A = +
v −1 0 ∂x v 0 1 v

and ( ) ( ) ( ) ( )( )
ϕ1 01 ∂ ϕ1 00 ϕ1
. A∗ = + .
ϕ2 10 ∂x ϕ2 01 ϕ2

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 493

The operator
. A∗ : W −→ V

is bijective (isomorphism). In fact, . A∗ is linear. . A∗ is injective. In fact

( ) ( )
∗ ϕ1 0
. A = ,
ϕ2 0

so
dϕ2
. =0
dx
hence
dϕ1
. + ϕ2 = 0.
dx
Which proves .ϕ2 = cte. The fact that

dϕ1
. = −cte
dx

gives .ϕ1 = −cte x + b. Gold, .ϕ1 (0) = 0, so .b = 0 and .ϕ1 (1) = 0 gives .cte = 0.
Thus, .ϕ1 = 0 and .ϕ2 = 0. Thus, . A∗ is bijective from .W into .V . Hence, to .w ∈ V we
associate of unique way .ϕ ∈ W such that

. A∗ ϕ = w.

Consider the linear form . L

. L : V −→R
{ 1
w −→L(w) = (F, ϕ)(L 2 ((0,1),R))2 = f ϕ1 d x.
0

. L(·) is continuous, i.e.,

. |L(w)| ≤ c|| f ||0,(0,1) ||w||(L 2 ((0,1),R))2 .

In fact,
.|L(w)| ≤ || f ||0,(0,1) ||ϕ1 ||0,(0,1) .

Furthermore, . A∗ ϕ = w, so
dϕ2
. = w1
dx

@seismicisolation
@seismicisolation
494 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

and
dϕ1
. + ϕ2 = w2 . (7.12.15)
dx
Which proves
{ 1 { 1 { 1
dϕ1
. w2 d x = dx + ϕ2 d x
0 0 dx 0
{ 1
= ϕ1 (1) − ϕ1 (0) + ϕ2 d x
0
{ 1
= ϕ2 d x.
0

So, according to the theorem of the mean, there is a .ξ ∈]0, 1[ such that
{ 1 { 1
. w2 d x = ϕ2 d x
0 0
= ϕ2 (ξ ).

Thus,
{ x
dϕ2
ϕ (x) = ϕ2 (ξ ) +
. 2 dt
ξ dx
{ x
= ϕ2 (ξ ) + w1 dt.
ξ

It follows that
{ 1 { x
ϕ (x) =
. 2 w2 d x + w1 dt.
0 ξ

Hence,

||ϕ2 ||0,(0,1) ≤ ||w2 ||0,(0,1) + ||w1 ||0,(0,1) .

.

On the other hand, using (7.12.15), we have

{ x { x { x
dϕ1
. dt + ϕ2 dt = w2 dt.
0 dx 0 0

So, { x
ϕ (x) − ϕ1 (0) =
. 1 (w2 − ϕ2 ) dt.
0

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 495

Thus,

||ϕ1 ||0,(0,1) ≤ ||w2 ||0,(0,1) + ||ϕ2 ||0,(0,1)

.

≤ 2||w2 ||0,(0,1) + ||w1 ||0,(0,1) .

It follows that
|L(w)| ≤ c|| f ||0,(0,1) ||w||(L 2 ((0,1),R))2 .
.

It thus appears that . L(·) is a continuous linear map from .V into .R. Hence, . L(·)
extends according to Han Banach’s theorem into an mapping . L(·)

. L : (L 2 ((0, 1), R))2 −→ R.

According to the Riesz Fréchet theorem, there is .u = (u 1 , u 2 ) ∈ (L 2 ((0, 1), R))2

unique such that

. L(w) = (u, w)(L 2 ((0,1),R))2

for all .w ∈ V . Again, there is .u = (u 1 , u 2 ) ∈ (L 2 ((0, 1), R))2 unique such that

(u, w)(L 2 ((0,1),R))2 = (u, A∗ ϕ)(L 2 ((0,1),R))2

.
{ 1
= f ϕ1 d x
0

for all .ϕ ∈ W . Hence,

{ 1 ( ) { 1
dϕ2 dϕ1
. u1 + u2 + ϕ2 dx = f ϕ1 d x
0 dx dx 0

for all .ϕ ∈ W . Which implies that

/ \
∂ϕ1
. u2, = _ f, ϕ1 _.
∂x

Thus,
du 2
. − = f
dx
and
du 1
. − + u 2 = 0.
dx

On the other hand, the function .u 2 ∈ H 1 (0, 1) because . f ∈ L 2 ((0, 1), R) and .u 2 ∈
L 2 ((0, 1), R) so .u 1 ∈ H 1 (0, 1). Hence, we can apply Green’s formula to recover the
boundary conditions. We have

@seismicisolation
@seismicisolation
496 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{ 1 { 1 ( ) { 1
dϕ2 dϕ1
. u1 dx + u2 + ϕ2 dx = f ϕ1 d x.
0 dx 0 dx 0

Hence,

{ 1 { 1
du 1 du 2
. − ϕ2 d x + u 1 (1)ϕ2 (1) − u 1 (0)ϕ2 (0) − ϕ1 d x + u 2 (1)ϕ1 (1) − u 2 (0)ϕ1 (0)
0 dx 0 dx
{ 1 { 1
+ u 2 ϕ2 d x = f ϕ1 d x.
0 0

Proving
u (1)ϕ2 (1) − u 1 (0)ϕ2 (0) + u 2 (1)ϕ1 (1) − u 2 (0)ϕ1 (0) = 0.
. 1

Gold, .u 1 (0) = u 1 (1) = 0, we have

u (1)ϕ1 (1) − u 2 (0)ϕ1 (0) = 0

. 2

for all .ϕ ∈ W . Thus, .u 2 (1) = u 2 (0) = 0.

Example 7.12.4 Consider the problem

d 2u
. − = f on ]0, 1[,
dx2

.u , (0) = u , (1) = 0
{ 1
. f (x) d x = 0.
0

We pose .v = u , . Then, {
−v, = f
−u , + v = 0.
.

Hence, ( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + = .
−1 0 ∂x v 01 v 0

We pose ( )
0 −1
. A1 =
−1 0

and ( )
00
. A0 = .
01

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 497

Gold,

∂
. C(x) = A0 (x) + A∗0 (x) − (A1 (x))
( ) ∂x
00
= .
02

C(x) is not positive within the meaning of Friedrichs. We have

.

( ) ( )
01 0 −1
. B(0) = , M(0) =
10 1 γ

with .γ ≥ 0 and ( ) ( )
0 −1 0 1
. B(1) = , M(1) =
−1 0 −1 γ1

with .γ1 ≥ 0. Then,

( ) ( )
0 2 00
(B − M)(0) =
. , (B + M)(0) = .
0 −γ 2γ

(B − M)(0) expresses the boundary conditions on .u. On the other hand,

.

. K er (B − M)(0) = {(α, β) ∈ R2 such that β = 0}

and
. K er (B + M)(0) = {(α, β) ∈ R2 such that 2α + γβ = 0}.

Gold,
. K er (B − M)(0) + K er (B + M)(0) = R2

and ( )
0 0
. M+ M= t
,
0 2γ

with .γ ≥ 0. Hence,
( ) ( )
0 −2 0 0
(B − M)(1) =
. , (B + M)(1) = .
0 −γ1 −2 γ1

Further,
. K er (B − M)(1) + K er (B + M)(1) = R2

and .(M +t M)(1) is positive semi-definite. Hence, we can express the boundary
conditions (admissible conditions). So,

@seismicisolation
@seismicisolation
498 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
( ) ( )
02 0 −2
.(B +t M)(0) = , (B +t M)(1) = .
0γ 0 γ1

We pose

.W = {ϕ = (ϕ1 , ϕ2 ) ∈ C 1 ([0, 1], R)/R × C 1 ([0, 1], R) such that ϕ2 (0) = ϕ2 (1) = 0}

and
. V = {v = A∗ ϕ, ϕ ∈ W }.

The mapping .ϕ ∈ W −→ v ∈ V is bijective. We are building then the mapping

. v −→ L(v)
n 2∗
on .V (L × L 2 ((0, 1), R)), with
{ { 1 }
. L 2∗
= v1 ∈ L 2 ((0, 1), R) such that v1 d x = 0 .
0

Remark 7.12.3 We intercepted.V by. L 2∗ × L 2 ((0, 1), R) because subsequently, we

have
dϕ2
. = v1
dx
implies
{ 1 { 1
dϕ2
. dx = v1 d x = 0.
0 dx 0

Hence, we want a Hilbert space such that

{ 1
. v1 d x = 0.
0

♦
We have { 1
. L(v) = f v1 d x.
0

. L(·) is well defined and continuous on . L 2∗ × L 2 ((0, 1), R). On the other hand,
( ) ( ) ( ) ( )( )
v1 01 ∂ v1 00 v1
. A∗ = + .
v2 10 ∂x v2 01 v2

. A∗ is injective. In fact, let .ϕ ∈ W such that . A∗ ϕ = 0. Let us show that .ϕ = 0. We

have

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 499

dϕ2
. =0
dx
and
dϕ1
. + ϕ2 = 0.
dx

Hence, .ϕ2 = cte. Proving .ϕ1 = cte and so .ϕ˚1 = 0.

. L(·) is continuous because using Theorem 3.4.11, we have

|L(v)| ≤ || f ||0,(0,1) ||ϕ1 + c||0,(0,1) ∀c

.

≤ || f ||0,(0,1) inf ||ϕ1 + c||0,(0,1)

c
≤ || f ||0,(0,1) ||ϕ˚1 || L 2/R . (7.12.16)

So, taking . P0 = R in Theorem 3.4.11, we obtain the mapping

. v −→ |ϕ1 |1,(0,1)

is a norm on . H 1 (0, 1)/R . Hence, using (7.12.16), we obtain

|| ||
|| dϕ1 ||
.|L(v)| ≤ c|| f ||0,(0,1) || || .
|| d x ||
0,(0,1)

Let .ϕ ∈ W such that . A∗ ϕ = v, we have

dϕ2
. = v1
dx
and
dϕ1
. + ϕ2 = v2 .
dx
Hence, || ||
|| dϕ1 ||
|| || ≤ ||v2 ||0,(0,1) + ||ϕ2 ||0,(0,1) .
.
|| d x ||
0,(0,1)

Further, { x
ϕ (x) = ϕ2 (0) +
. 2 v1 (t) dt.
0

So,
||ϕ2 ||0,(0,1) ≤ ||v1 ||0,(0,1) .
.

Thus,

@seismicisolation
@seismicisolation
500 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
|| ||
|| dϕ1 ||
|| || ≤ ||v2 ||0,(0,1) + ||v1 ||0,(0,1)
.
|| d x ||
0,(0,1)
√ ( )1
≤ 2 ||v2 ||20,(0,1) + ||v1 ||20,(0,1) 2 .
n
The mapping . L(·) is linear and continuous from .V (L 2∗ × L 2 ((0, 1), R)) into .R.
Hence, . L(·) is prolonged by continuity in . L(·) defined on . L 2∗ × L 2 ((0, 1), R) with

||L|| = ||L||.
.

So, according to the Riesz Fréchet theorem, there is .u = (u 1 , u 2 ) ∈ L 2∗ × L 2

((0, 1), R)) unique such that
.||u|| = ||L||

and
. L(v) = (u, v)

for all .v ∈ L 2∗ × L 2 ((0, 1), R). Hence,

{ 1
∗
(u, A ϕ)(L 2 ((0,1),R))2 =
. f ϕ1 d x
0
≤ (F, ϕ)(L 2 ((0,1),R))2

with ( )
f
. F= .
0

Example 7.12.5 We can sometimes dispense with premultiplying. We consider the

system ( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + =
−1 0 ∂x v 01 v 0

with { 1
. f (x) d x = 0.
0

Question 1: Give the admissible . M(0) and . M(1) matrices to take into account the
Neumann conditions. In fact, we have
( )
01
. B(0) =
10

and ( )
αβ
. M(0) = .
γ δ

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 501

Hence, ( )
−α 1 − β
(B − M)(0) =
. .
1 − γ −δ

Since we have the Neumann boundary conditions, the first column of .(B − M)(0)
is zero. So, .α = 0 and .γ = 1. On the other hand, .det B(0) = −1, so . B(0) is neither
positive semi-definite nor negative semi-definite. So,

. dim K er (B + M)(0) ≥ 1.

Which gives .det(B + M)(0) = 0 with

( )
0 1+β
(B + M)(0) =
. .
2 δ

Thus, .−2(1 + β) = 0 and so .β = −1. On the other hand, . M(0) + M ∗ (0) is positive
semi-definite. Then, ( )
∗ 0 0
. M(0) + M (0) = .
0 2δ

Hence, .δ ≥ 0. Thus, ( )
0 −1
. M(0) =
1 δ

with .δ ≥ 0. Gold,
( ) ( )
0 −1 α1 β1
. B(1) = , M(1) = .
−1 0 γ 1 δ1

Hence, ( )
−α1 −1 − β1
(B − M)(1) =
. .
−1 − γ1 −δ1

Since we have the Neumann boundary conditions, the first column of .(B − M)(1) is
zero. So, .α1 = 0 and .γ1 = −1. Now, .det B(1) = −1, therefore, . B(1) is neither pos-
itive semi-definite nor negative semi-definite. So, .dim K er (B + M)(1) ≥ 1. Hence,
.det(B + M)(1) = 0 with

( )
α1 −1 + β1
(B + M)(1) =
. .
−1 + γ1 δ1

So, .α1 δ1 − (−1 + β1 )(−1 + γ1 ) = 0 and .2(−1 + β1 ) = 0. Thus, .β1 = 1. Since

M(1) + M ∗ (1) is positive semi-definite,
.

( )
0 0
. ≥ 0.
0 2δ1

@seismicisolation
@seismicisolation
502 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Hence, .δ1 ≥ 0. Thus, ( )

0 1
. M(1) =
−1 δ1

with .δ1 ≥ 0.
Question 2: Show that our problem admits a weak solution. We can consider the
space

.W = {ϕ = (ϕ1 , ϕ2 ) ∈ C 1 ([0, 1], R)/R × C 1 ([0, 1], R) such that ϕ2 (0) = ϕ2 (1) = 0}.

In fact, let .V be the space

. V = {v = A∗ ϕ, ϕ ∈ W }.

(i) Let us show that for all .v ∈ V , there is .ϕ ∈ W unique such that . A∗ ϕ = v, with
.

( )
∗ ∂ϕ2 ∂ϕ1
. A (ϕ1 , ϕ2 ) = , + ϕ2 .
∂x ∂x

In fact, only show the injection of . A∗ . Let .ϕ ∈ K er (A∗ ). Let us show that .ϕ = 0.
We have . A∗ ϕ = 0, so
∂ϕ2
. =0
∂x
and
∂ϕ1
. + ϕ2 = 0.
∂x
Proving
∂ϕ1
ϕ =−
. 2 .
∂x
Thus,
∂ϕ2 ∂ 2 ϕ1
. = − 2 = 0.
∂x ∂x

Hence, .ϕ2 = cte. Thus, .ϕ1 = −cte x + b. Gold, .ϕ2 (0) = ϕ2 (1) = 0, so .cte = 0.
Thus, .ϕ1 = b. So, .ϕ˚1 = 0. Hence, . A∗ is injective.
(ii) Let’s show the following inclusion
.

{ { 1 }
. V ⊂ (L 2 )∗ := (v1 , v2 ) ∈ (L 2 ((0, 1), R))2 such that v1 d x = 0 .
0

In fact, let .v ∈ V , so .v = A∗ ϕ, with .ϕ ∈ W . We have

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 503

∂ϕ2
. = v1
∂x
and
∂ϕ1
. + ϕ2 = v2 .
∂x
Hence,
{ {
1
∂ϕ21
. v1 d x = dx
0 0 ∂x
= ϕ2 (1) − ϕ2 (0) = 0.

So, .v ∈ (L 2 )∗ . Thus, .V ⊂ (L 2 )∗ . The subspace .(L 2 )∗ provided of the norm induced

by .(L 2 ((0, 1), R))2 is a Hilbert space.
(iii) Let now .v ∈ V be an arbitrary element, we associate with it .ϕ ∈ W unique
.
solution of . A∗ ϕ = v and we pose
{ 1
. L(v) = f ϕ1 d x,
0

where .ϕ1 is a representative of the class .ϕ˚1 . This linear form is well defined because
{ 1
. f (x) d x = 0, and ϕ˚1 ∈ C 1 ([0, 1], R)/R .
0

Show that this linear form . L(·) is continuous for the norm .(L 2 )∗ . In fact,
Method 1: We have
. |L(v)| ≤ || f ||0,(0,1) ||ϕ1 ||0,(0,1)

for any representative .ϕ1 of the class .ϕ˚1 . Hence,

|L(v)| ≤ || f ||0,(0,1) ||ϕ˚1 || L 2/R

.
|| ||
|| ∂ϕ1 ||
≤ c|| f ||0,(0,1) ||
|| ∂ x ||
|| .
0,(0,1)

Further, || ||
|| ∂ϕ1 ||
|| || ≤ ||v2 ||0,(0,1) + ||ϕ2 ||0,(0,1) .
.
|| ∂ x ||
0,(0,1)

On the other hand,

@seismicisolation
@seismicisolation
504 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{ x
ϕ (x) = ϕ2 (0) +
. 2 v1 dt
{ x 0

= v1 dt.
0

Hence,
.||ϕ2 ||0,(0,1) ≤ ||v1 ||0,(0,1) .

Thus,

. |L(v)| ≤ c|| f ||0,(0,1) (||v1 ||0,(0,1) + ||v2 ||0,(0,1) )

√
≤ c 2|| f ||0,(0,1) ||v||(L 2 ((0,1),R))2

with
( )1
||v||(L 2 ((0,1),R))2 = ||v1 ||20,(0,1) + ||v2 ||20,(0,1) 2 .
.

Method 2: If we define the space .V by

. V = {ϕ = (ϕ1 , ϕ2 ) ∈ C 1 ([0, 1], R) × C 1 ([0, 1], R) such that ϕ1 (0) = ϕ1 (1) = 0},

then

|L(v)| ≤ || f ||0,(0,1) ||ϕ1 ||0,(0,1) .

.

According to formula of the average, there is .ξ ∈]0, 1[ such that

{ 1 { 1 ( )
dϕ1
. v2 d x = + ϕ2 dx
0 0 dx
{ 1
= ϕ1 (1) − ϕ1 (0) + ϕ2 d x
0
{ 1
= ϕ2 d x
0
= ϕ2 (ξ ).

Hence,
{ x
ϕ (x) = ϕ2 (ξ ) +
. 2 v1 (t) dt
ξ
{ 1 { x
= v2 d x + v1 (t) dt.
0 ξ

Thus,
||ϕ2 ||0,(0,1) ≤ ||v2 ||0,(0,1) + ||v1 ||0,(0,1) .
.

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 505

Furthermore,
dϕ1
. + ϕ2 = v2
dx
so
{ x
ϕ (x) = ϕ1 (0) +
. 1 (v2 − ϕ2 ) dt
{ x 0

= (v2 − ϕ2 ) dt.
0

Thus,

||ϕ1 ||0,(0,1) ≤ ||v2 ||0,(0,1) + ||ϕ2 ||0,(0,1)

.

≤ 2||v2 ||0,(0,1) + ||v1 ||0,(0,1) .

Hence,
( )1
||ϕ1 ||0,(0,1) ≤ c ||v2 ||20,(0,1) + ||v1 ||20,(0,1) 2
.

=c||v||(L 2 ((0,1),R))2

So,
|L(v)| ≤ c|| f ||0,(0,1) ||v||(L 2 ((0,1),R))2 .
.

It follows that

. L : V −→ R
v −→ L(v)

extends according to Han Banach’s theorem into an mapping . L(·)

. L : (L 2 )∗ −→ R
√
of norm .≤ c 2|| f ||0,(0,1) . According to the Riesz Fréchet theorem, there is a couple
2 ∗
.u = (u 1 , u 2 ) ∈ (L ) unique such that

. L(v) = (u, v)(L 2 ((0,1),R))2

and

. ||u||(L 2 )∗ = ||L||.

So, we have existence of an element .u ∈ (L 2 )∗ such that

@seismicisolation
@seismicisolation
506 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{ 1
. f ϕ1 d x = L(v)
0
= (u, v)(L 2 ((0,1),R))2
= (u, A∗ ϕ)(L 2 ((0,1),R))2

for all .ϕ ∈ W. Again,

{ { ( ) {
1
∂ϕ2 1
∂ϕ1 1
. u1 dx + u2 + ϕ2 dx = f ϕ1 d x
0 ∂x 0 ∂x 0

for all .ϕ ∈ W. We take .ϕ ∈ W such that .ϕ2 = 0, we obtain

{ {
1
∂ϕ1 1
. u2 dx = f ϕ1 d x.
0 ∂x 0

Hence, / \
∂ϕ1
. u2, = _ f, ϕ1 _.
∂x

Thus,
du 2
. − = f in D , (0, 1). (7.12.17)
dx
On the other hand,
/ \ / \
∂u 1 ∂u 2
. − , ϕ2 + − , ϕ1 + _u 2 , ϕ2 _ = _ f, ϕ1 _.
∂x ∂x

Using Eq. (7.12.17), we obtain

/ \
∂u 1
. − + u 2 , ϕ2 = 0.
∂x

So,
∂u 1
. − + u 2 = 0.
∂x

Hence, . Au = F in .D , (0, 1) × D , (0, 1). Now, . f ∈ L 2 ((0, 1), R). Therefore, .u 2 ∈

H 1 (0, 1) and.u 2 ∈ L 2 ((0, 1), R). Hence,.u 1 ∈ H 1 (0, 1). So,.u = (u 1 , u 2 ) ∈ H 1 (0, 1)
× H 1 (0, 1). Thus, we can apply Green’s formula to recover the boundary conditions.
Indeed, let .ϕ ∈ W , then
{ { ( ) {
1
∂ϕ2 1
∂ϕ1 1
. u1 dx + u2 + ϕ2 dx = f ϕ1 d x.
0 ∂x 0 ∂x 0

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 507

Hence,

{ {
1 ∂u 1 1 ∂u 2
. − ϕ2 d x + u 1 (1)ϕ2 (1) − u 1 (0)ϕ2 (0) − ϕ1 d x + u 2 (1)ϕ1 (1) − u 2 (0)ϕ1 (0)
0 ∂x 0 ∂x
{ 1 { 1
+ u 2 ϕ2 d x = f ϕ1 d x.
0 0

So,
u (1)ϕ2 (1) − u 1 (0)ϕ2 (0) = u 2 (0)ϕ1 (0) − u 2 (1)ϕ1 (1)
. 1

and
u (0)ϕ1 (0) = u 2 (1)ϕ1 (1)
. 2

for all .ϕ ∈ W . Thus, .u 2 (0) = u 2 (1).

Example 7.12.6 Case where we do not know how to recover positivity. Consider
the problem {
div u = f 1
.
Rot u = f 2 ,

with ( )
u1
.u= .
u2

Again ⎧
⎪ ∂u 1 ∂u 2
⎨ + = f1
∂ x 1 ∂ x2
.
⎪ ∂u ∂u
⎩ 1 − 2 = f2 .
∂ x2 ∂ x1

7.12.2 Examples in Dimension 2

Example 7.12.7 We consider the problem

−Δu. + u = f in Ω :=]0, 1[×]0, 1[

u=0 on ∂Ω. (7.12.18)

We pose
∂u ∂u
. = v, = w.
∂x ∂y

@seismicisolation
@seismicisolation
508 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Hence,
∂v ∂w
. − − + u = f.
∂x ∂y

Thus,
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
0 −1 0 u 0 0 −1 u 100 u f
∂ ∂
. ⎝ −1 0 0 ⎠ ⎝v ⎠+⎝ 0 0 0 ⎠ ⎝ v ⎠ + ⎝0 1 0⎠⎝ v ⎠ = ⎝ 0 ⎠.
∂x ∂y
0 0 0 w −1 0 0 w 001 w 0

We have ⎛ ⎞
0 −νx −ν y
. B = ⎝ −ν x 0 0 ⎠,
−ν y 0 0

where .νx and .ν y are the components of the exterior normal .ν to .∂Ω. A possible
choice for the matrix . M, allowing to find the boundary conditions (7.12.18) is the
following ⎛ ⎞
a −νx −ν y
. M = ⎝ νx 0 0 ⎠ , with a > 0.
νy 0 0

We still have in this case

. K er (B − M) = K er (B +t M) for x ∈ ∂Ω.

We cut the domain .Ω = [0, 1] × [0, 1] into rectangles with sides .Δx = 1I and .Δy =
1
J
. Let .xi = iΔx, . y j = jΔy, . Ai j = (xi , y j ), .u h (Ai j ) = u i j , for .0 ≤ i, j ≤ I, J .
Let .Vh be the space of vector functions .(ϕh , ψh , ηh ) such that each component is a
continuous function whose restriction to each rectangle is a polynomial of . Q k in .R2 ,
where . Q k is the space of polynomials of the form
E
. ai j x i x j ,
i, j≤k

with .ai j ∈ R. The space . X k2 will be defined by

. X h2 = {(ϕh , ψh , ηh ) ∈ Vh such that ϕh |∂Ω = 0},

corresponding to the next choice of the matrix . M along .∂Ω

⎛ ⎞
0 10
. M = ⎝ −1 0 0 ⎠ .
0 00

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 509

We solve the following approximate problems:

(i) Find .(u h , vh , wh ) ∈ Vh such that
.

{ ( ) { ( )
∂vh ∂wh ∂u h
. − − + u h − f ϕh d xd y + − + vh ψh d xd y+
Ω ∂x ∂y Ω ∂x
{ 1 { ( )
∂u h
(−(u h ψh )(0, y) + (u h ψh )(1, y)) dy + − + wh ηh d xd y+
0 Ω ∂y
{ 1
(−(u h ψh )(x, 0) + (u h ψh )(x, 1)) d x = 0 (7.12.19)
0

for all .(ϕh , ψh , ηh ) ∈ Vh .

(ii) Find .(u h , vh , wh ) ∈ X h2 such that
.

{ ( ) { ( )
∂vh ∂wh ∂u h
. − − + u h − f ϕh d xd y + − + vh ψh d xd y+
Ω ∂x ∂y Ω ∂x
{ ( )
∂u h
− + wh ηh d xd y = 0 (7.12.20)
Ω ∂y

for all .(ϕh , ψh , ηh ) ∈ X h2 .

For these two problems, we have the following error bound
|| || || ||
|| ∂u || || ∂u ||
.||u − u h ||0,Ω + ||
|| ∂ x − v ||
h || + ||
|| ∂ y − w ||
h || ≤ ch k ||u||k+2,2,Ω .
0,Ω 0,Ω

When .k = 1, the error bound of the problem (7.12.19), is given by

|| || || ||
|| ∂u || || ∂u ||
||u − u h ||0,Ω + || || || || 3
.
|| ∂ x − v h || + || ∂ y − w h || ≤ ch 2 (||u||4,2,Ω + ||u||3,∞,Ω )
0,Ω 0,Ω

and, the error bound of the problem (7.12.20), is given by

|| || || ||
|| ∂u || || ∂u ||
||u − u h ||0,Ω + || || || || 3
.
|| ∂ x − v h || + || ∂ y − w h || ≤ ch 2 ||u||4,2,Ω .
0,Ω 0,Ω

When .k = 1, we are going to write the linear system corresponding to the problem
(7.12.20), in the case, where we calculate the integrals in an approximate way on
each rectangle, by the formula
{
area(K ) E
4
. f d xdy = f (ai ),
K 4 i=1

where the points .ai are the vertices of the rectangle . K , .1 ≤ i ≤ 4. We have

@seismicisolation
@seismicisolation
510 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

vi+1, j −vi−1, j w −w
− 2Δx
− i, j+12Δy i, j−1 + u i j = f i j , for 1 ≤ i, j ≤ I − 1, J − 1
u i+1, j −u i−1, j
− 2Δx + vi j = 0, for 1 ≤ i, j ≤ I − 1, J − 1
u −u
. − i, j+12Δy i, j−1 + wi j = 0, for 1 ≤ i, j ≤ I − 1, J − 1
u 0 j = w0 j = 0, 0 ≤ j ≤ J,
u −u
− 1 jΔx 0 j + v0 j = 0, 0 ≤ j ≤ J,

with analogous equations along the other sides of the rectangle .Ω.

Let .v ∈ X ∗ , with
{ }
. X ∗ = v ∈ (H 1 (Ω)) p such that (B +t M)v|∂Ω = 0 .

We take .vh = rh v the interpolated of .v.

Question: Does .v − rh v tend to .0 on .(H 1 (Ω)) p when .h → 0. Furthermore,

.(u, A∗ v)(L 2 (Ω,R)) p = (F, v)(L 2 (Ω,R)) p

n 2 n
for all .v ∈ X ∗ (H (Ω)) p . Is the space . X ∗ (H 2 (Ω)) p dense in . X ∗ .
We consider the problem
{
Au = F in Ω
.
(B − M)u = 0 on ∂Ω.

We can think for example .(B − M)(u − g) = 0 on .∂Ω, where .g is a given function.
We have .Vh ⊂ (H 1 (Ω)) p . We search .u h ∈ Vh such that
{ (( ) )
B−M
(Au h , z h )(L 2 (Ω,R)) p −
. uh , zh ds = (F, z h )(L 2 (Ω,R)) p
∂Ω 2 2

for all .z h ∈ Vh . We have

1
||u h ||(L 2 (Ω,R)) p ≤
. ||F||(L 2 (Ω,R)) p .
α
The problem is that . A is not always positive. We can make it positive by premulti-
plying, but what can we say about things without multiplying. Indeed,
Example 7.12.8 We consider the problem
{
−Δu = f in Ω
.
u = 0 on ∂Ω.

We pose
∂u ∂u
. = v, = w.
∂x ∂y

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 511

Hence,
∂v ∂w
. − − = f.
∂x ∂y

Thus,
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
0 −1 0 u 0 0 −1 u 000 u f
∂ ∂
. ⎝ −1 0 0 ⎠ ⎝v ⎠+⎝ 0 0 0 ⎠ ⎝ v ⎠ + ⎝0 1 0⎠⎝ v ⎠ = ⎝ 0 ⎠.
∂x ∂y
0 0 0 w −1 0 0 w 001 w 0

We pose
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0 −1 0 0 0 −1 000
. A 1 = ⎝ −1 0 0 ⎠ , A 2 = ⎝ 0 0 0 ⎠ , and A 0 = ⎝ 0 1 0 ⎠ .
0 0 0 −1 0 0 001

C(x) is given by
.
⎛ ⎞
000
.C(x) = ⎝ 0 2 0 ⎠ .
002

C(x) is not positive within the meaning of Friedrichs. We have

.

⎛ ⎞ ⎛ ⎞
010 α0 1 0
. B(0, y) = ⎝ 1 0 0 ⎠ , M(0, y) = ⎝ −1 0 0 ⎠ .
000 0 00

Hence, ⎛ ⎞
−α0 0 0
.(B − M)(0, y) = ⎝ 2 0 0⎠.
0 00

. M +t M is positive semi-definite if .α0 ≥ 0. On the other hand,

⎛ ⎞
α0 2 0
.(B + M)(0, y) = ⎝ 0 0 0 ⎠ .
0 00

Gold,
. K er (B − M)(0, y) = {(z 1 , z 2 , z 3 ) ∈ R3 such that z 1 = 0}

and
. K er (B + M)(0, y) = {(z 1 , z 2 , z 3 ) ∈ R3 such that α0 z 1 + 2z 2 = 0}.

@seismicisolation
@seismicisolation
512 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

We have
. K er (B − M)(0, y) + K er (B + M)(0, y) = R3 .

We pose

. V = {z = A∗ ϕ, ϕ = (ϕ1 , ϕ2 , ϕ3 ) ∈ (C 1 (Ω, R))3 , ϕ1 = 0 on ∂Ω}.

We took the space .V as above because

⎞ ⎛
α0 0 0
.(B + M)(0, y) = ⎝ 2 0 0 ⎠
t

0 00

and

. K er (B +t M)(0, y) = {(ϕ1 , ϕ2 , ϕ3 ) ∈ (C 1 (Ω, R))3 such that ϕ1 = 0}.

We have
∂ϕ ∂ϕ
. A∗ ϕ = −A1 − A2 + A0 ϕ.
∂x ∂y

At .z ∈ V , we associate
{
. L(z) = f ϕ1 d x
Ω
= (F, ϕ).

. A∗ is injectif. In fact, . A∗ ϕ = 0, so

∂ϕ2 ∂ϕ3
. + =0 (7.12.21)
∂x ∂y
∂ϕ1
. + ϕ2 = 0 (7.12.22)
∂x
∂ϕ1
. + ϕ3 = 0. (7.12.23)
∂y

We multiply Eq. (7.12.22) by . ∂ϕ

∂x
1
and Eq. (7.12.23) by . ∂ϕ
∂y
1
, and we integrate on .Ω,
we get
{ ( )2 { ( )2 { {
∂ϕ1 ∂ϕ1 ∂ϕ1 ∂ϕ1
. d xd y + d xd y + ϕ2 d xd y + ϕ3 d xd y = 0.
Ω ∂x Ω ∂y Ω ∂x Ω ∂y

Green’s formula gives

@seismicisolation
@seismicisolation
7.12 Change to First-Order System 513

{ ( )2 ( )2 { {
∂ϕ1 ∂ϕ1 ∂ϕ2 ∂ϕ3
. + d xd y − ϕ1 d xd y − ϕ1 d xd y
Ω ∂x ∂y Ω ∂x Ω ∂y
{
+ ϕ1 (ϕ2 + ϕ3 )ndσ = 0.
∂Ω

Using (7.12.21), we obtain

{ ( )2 ( )2
∂ϕ1 ∂ϕ1
. + d xd y = 0.
Ω ∂x ∂y

Thus, .∇ϕ1 = 0 and so .ϕ1 = cte. Further, .ϕ1 = 0 on .∂Ω so .ϕ1 = 0. Proving .ϕ2 =
ϕ3 = 0. Thus, . A∗ is injective.
. L(·) is continuous. In fact, let . z ∈ V , according to the Poincaré inequality (The-
orem 3.2.2), we have

. |L(z)| ≤ || f ||0,Ω ||ϕ1 ||0,Ω

≤ c(Ω)|| f ||0,Ω ||∇ϕ1 ||0,Ω .

Let .z ∈ V , .z = A∗ ϕ. Hence,

∂ϕ2 ∂ϕ3
. + = z1 (7.12.24)
∂x ∂y
∂ϕ1
. + ϕ2 = z 2 (7.12.25)
∂x
∂ϕ1
. + ϕ3 = z 3 . (7.12.26)
∂y

We multiply Eq. (7.12.25) by . ∂ϕ

∂x
1
and Eq. (7.12.26) by . ∂ϕ
∂y
1
, and we integrate on .Ω,
we get
{ ( )2 ( )2 { {
∂ϕ1 ∂ϕ1 ∂ϕ1 ∂ϕ1
. + d xd y + ϕ2 d xd y + ϕ3 d xd y =
Ω ∂x ∂y Ω ∂x Ω ∂y
{ {
∂ϕ1 ∂ϕ1
z2 d xd y + z3 d xd y.
Ω ∂x Ω ∂y

Green’s formula gives

{ ( )2 ( )2 { {
∂ϕ1 ∂ϕ1 ∂ϕ2 ∂ϕ3
. + d xd y − ϕ1 d xd y − ϕ1 d xd y =
Ω ∂x ∂y Ω ∂x Ω ∂y
{ {
∂ϕ1 ∂ϕ1
z2 d xd y + z3 d xd y.
Ω ∂x Ω ∂y

@seismicisolation
@seismicisolation
514 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Using (7.12.24), we obtain

{ {
1 1 1
. |∇ϕ1 |2 d xd y ≤ ||z 1 ||0,Ω ||ϕ1 ||0,Ω + ||z 2 ||20,Ω + ||z 3 ||20,Ω + |∇ϕ1 |2 d xd y.
Ω 2 2 2 Ω

According to the Poincaré inequality (Theorem 3.2.2), we have

{
1 1 1
. |∇ϕ1 |2 d xd y ≤ c(Ω)||z 1 ||0,Ω ||∇ϕ1 ||0,Ω + ||z 2 ||20,Ω + ||z 3 ||20,Ω .
2 Ω 2 2

Further, using Eq. (2.5.2), we obtain

1
c(Ω)||z 1 ||0,Ω ||∇ϕ1 ||0,Ω ≤
. ||∇ϕ1 ||20,Ω + c(Ω)2 ||z 1 ||20,Ω .
4
Hence,
{ ( )E3
1 1
. |∇ϕ1 |2 d xd y ≤ max , c(Ω)2 ||z i ||20,Ω .
4 Ω 2 i=1

Thus,
. |L(z)| ≤ c|| f ||0,Ω ||z||(L 2 (Ω,R))3 .

So, we extend it to . L(·) defined on .(L 2 (Ω, R))3 in .R which is linear and continuous.
So, according to the Riesz Fréchet theorem, there exists .u ∈ (L 2 (Ω, R))3 such that

(u, z)(L 2 (Ω,R))3 = L(z)

.

for all .z ∈ (L 2 (Ω, R))3 . For .z ∈ V , .z = A∗ ϕ. Hence,

(u, A∗ ϕ)(L 2 (Ω,R))3 = L(z).

.

Again, {
(u, A∗ ϕ)(L 2 (Ω,R))3 =
. f ϕ1 .
Ω

In .(D , (Ω))3 , . Au = F with

⎞ ⎛ ⎛ ⎞
f u1
.F = ⎝ 0 ⎠ , u = ⎝ u2 ⎠ .
0 u3

Hence,

@seismicisolation
@seismicisolation
7.13 Special Case of the Transport Equation 515

∂u 2 ∂u 3
. − − = f
∂x ∂y
∂u 1
− + u2 = 0 (7.12.27)
∂x
∂u 1
. + u 3 = 0. (7.12.28)
∂y

Equations (7.12.27) and (7.12.28) gives .u 1 ∈ H 1 (Ω).

Remark 7.12.4 If, we have

. − Δu = f

on .D , (Ω) and if, furthermore, . f ∈ L 2 (Ω, R), this does not imply that .u 1 ∈ H 2 (Ω)
and this according to the boundary conditions. Indeed, if, we have Neumann or
Dirichlet conditions, then .u 1 ∈ H 2 (Ω). On the other hand, if, we have a coupling
between the boundary conditions of Neumann and Dirichlet (mixed), we cannot say
that .u 1 ∈ H 2 (Ω) because the problems arise at the intersections of the boundary
conditions. ♦

7.13 Special Case of the Transport Equation

In the case of the transport equation in plane geometry, for example in dimension .2,
the equation is written (see Remark 7.9.3)

∂ϕ ∂ϕ
.μ +ν + σ ϕ = f, (x, y) ∈ Ω,
∂x ∂y

ϕ = 0 on .∂− Ω = {(x, y) ∈ ∂Ω such as μn x + νn y < 0}. One uses to solve this

.
problem the discontinuous method exposed previously and one chooses the matrices
of transmission . MT so that the problem can be solved element by element, i.e.,

. MT = BT on ∂+ T, MT = −BT on ∂− T,

where . BT = μn x,T + νn y,T , .(n x,T , n y,K ) exterior normal vector to .∂ T ,

∂ T = {(x, y) ∈ ∂ T such that BT > 0}

. +

and
∂ T = {(x, y) ∈ ∂ T such that BT < 0}.
. −

On each mesh .T , we seek the solution .u h in a space of functions of the form . PT

(for example polynomials of degree .≤ k if .T is a triangle or polynomials of . Q k
isoparametric if .T is any quadrilateral). We pose

@seismicisolation
@seismicisolation
516 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
| |
. Wh = PT .
T ⊂Ω

The problem is, therefore, posed as follows: Find .u h ∈ Wh such that

{
(Au h − f, vh ) L 2 (T,R) −
. BT (u int
h − u h )vh ds = 0,
ext int
(7.13.1)
∂− T

for all .T ⊂ Ω and all .vh ∈ Wh , where .u ext

h = 0 on .∂− Ω.

Remark 7.13.1 .(i) This method was introduced by Reed and Hill [12] for solving
the transport equation in two-dimensional plane geometry with triangular meshes.
.(ii) If, in (7.13.1), we replace .vh by the constant function equal to .1, we obtain the

particle conservation property:

{ { { {
. h ds + σ
BT u int u h d xd y = |BT |u ext
h ds + f d xdy.
∂− T T ∂− T T

♦
To obtain error bounds, we use the following lemma.
Lemma 7.13.2 Let .u h be the solution to the problem (7.13.1). For any .T ⊂ Ω, any
v ∈ π {(PT ), T ⊂ Ω} and for any function .q ∈ L 2 (∂− T, R), we have the relation
. h

{ {
1
B (u int
h − vh ) ds + 2 ∂− T BT{(u h − q) ds
int 2 1 ext 2
{ 2 ∂+ T T
− ∂− T BT ((u h − vh ) − (u h − q)) ds + T σ (u h − vh )2 d xd y =
1 int int ext 2
. {2 { (7.13.2)
∂+ T BT (u − vh )(u{h − vh ) ds + ∂− T BT (u − q)(u h − vh ) ds
int int int int int

+ T (u − vh )A∗ (u h − vh ) d xd y,

where . A∗ ϕ = −μ ∂ϕ
∂x
− ν ∂ϕ
∂y
+ σ ϕ. ♦
Proof We pose .w = u h − vh , .ξ = u ext
h − q on .∂− T . We consider the expression
{ {
y =−
. h BT (wint − ξ )wint ds + (Aw)w d xdy.
∂− T T

Green’s formula implies

{ { {
1
. yh = BT (w ) ds − BT (w − ξ )w ds + σ w2 d xd y.
int 2 int int
2 ∂T ∂− T T

On the other hand,

1 int 2 1 2 1 int
.(wint − ξ )wint = (w ) − ξ + (w − ξ )2 .
2 2 2

@seismicisolation
@seismicisolation
7.13 Special Case of the Transport Equation 517

We obtain
{ { { {
1 1 1
. yh = BT (wint )2 ds + BT ξ 2 ds − BT (wint − ξ )2 ds + σ w2 d xd y.
2 ∂+ T 2 ∂− T 2 ∂− T T

On the other hand, from (7.13.1), we have

{ {
. yh = BT (vh − q)w ds + A(u − vh )w d xdy.
int int
∂− T T

Again using Green’s formula, we have

{ { {
. yh = BT (u − vhint )wint ds + BT (u − q)wint ds + (u − vh )A∗ w d xd y,
∂+ T ∂− T T

which completes the proof. Q.E.D.

Theorem 7.13.3 We assume that the n elements .T ⊂ Ω are rectangles and that the
exact solution .u belongs to . H k+2 (Ω)
U W
k+1,∞
(Ω). Let a numbering of elements
. Ti as in Remark 7.9.3 and let .Ωi = {T j , j ≤ i}. Then,

||u − u h ||0,Ωi = O(h k+1 ),

. (7.13.3)
({ ) 21
. B(u − u h ) ds 2
= O(h k+1 ), (7.13.4)
∂+ Ωi

for any index .i ≤ N . ♦

Proof On the interval .[−1, 1], we consider Gauss Radau’s quadrature formula
{ 1 E
k+1
. g(ξ ) dξ ∼ wi f (ηi ), (7.13.5)
−1 i=1

with.ηk+1 = 1, exact for polynomials of degree.≤ 2k. To any continuous function.- v on

the square - = [−1, 1] × [−1, 1], we associate the polynomial.-
.T r-v ∈ Q (of the form
E k
. -
0≤i, j≤k ai j ηi η j ), equal to.v at points. Ai j with coordinates.(ηi , η j ),.1 ≤ i, j ≤ k. Let
. T = A 1 A 2 A 3 A 4 be a rectangle in.Ω. To any continuous function.v on. T , we associate
.- - by the usual isoparametric transformation. To any function .v, continuous on
v on .T
. T , we associate .r (v) ∈ Q k equal to .v to the points . Ai j , image by . F of the points . A -i j .
In expression (7.13.2), we replace .vh by .r (u |T ) for all .T ⊂ Ω, and .q by .vhext . The
right-hand side of (7.13.2) is the sum of the following terms:

@seismicisolation
@seismicisolation
518 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs
{
.Σ = σ (u − vh )wh d xd y,
T

({ { { )
A1 A2 ∂wh
.X =μ (u − vhint )whint d y − (u − vhext )whint d y − (u − vh ) d xd y ,
A4 A3 T ∂x

({ { { )
A1 A4 ∂wh
.Y =ν (u − vhint )whint d x − (u − vhext )whint d x − (u − vh ) d xd y .
A2 A3 T ∂y

We have
|Σ| = O(h k+1 )||w||0,T .
.

The map .u −→ X is identically zero for all .u ∈ Q k . If .u = x k+1 , .u |A1 A4 is a constant

in . y, so .(u − vh int )|A1 A4 = 0. The same is true for side . A3 A2 . On the other hand,
{ { ({ )
∂wh y2 x4
∂wh
. (u − vh ) d xd y = (u − vh ) dx dy,
T ∂x y3 x3 ∂x

where the .(xi , yi ) are the coordinates of the points . Ai , .1 ≤ i ≤ 4. The polynomial
∂wh
.
∂x
is of degree .≤ k − 1 at .x. At fixed . y, the integral
{ x4
∂wh
. (u − vh ) dx
x3 ∂x

is calculated exactly by the quadrature formula (7.13.5), therefore equal to zero. If

u = y k+1 , .u − vh is independent of .x, so . X = 0. The map .u −→ X is identically
.

zero for all .u ∈ Pk+1 . This polynomial invariance makes it possible, with the usual
techniques of change of variable, to show that

. |X | ≤ ch k+1 ||wh ||0,T .

We have the same increase of .Y . Finally,

||u − rh u||0,Ωi = O(h k+1 ).

.

We deduce the relation (7.13.3). If .u ∈ W k+1,∞ (Ω), we have on the other hand
({ ) 21
. (u − rh u) ds 2
= O(h k+1 ),
∂Ωi

and, we obtain the markup (7.13.4). Q.E.D.

Remark 7.13.4 .(i) In the context of convex quadrilaterals having two illuminated
faces, we have the upper bound

||u − u h ||0,Ωi = O(h k+1 + h k−1 z), for all i,

.

@seismicisolation
@seismicisolation
7.14 Petrov Galerkin-Type Methods 519

where .z = max{Z T , T ⊂ Ω}, and . Z T denotes the distance between the midpoints
of the diagonals of the quadrilateral .T .
A2 A1

ZT

A3

A4

(ii) In the case of convex quadrilaterals with any number of illuminated faces (.1–.3)
.
1
or in the case of triangles, Theorem 7.9.7 gives an error in . O(h k+ 2 ). ♦

7.14 Petrov Galerkin-Type Methods

The idea of Petrov Galerkin-type methods consists in using test functions, linear
combinations of the usual test functions and derivatives thereof. We can for example
define the problem: Find .u h ∈ Vh such that
{
1
(Au h − F, vh + h Avh )(L 2 (Ω,R)) p −
. ((B − M)vh , vh )2 ds = 0, (7.14.1)
2 ∂Ω

for all .vh ∈ Vh , where the parameter .h is equal to the maximum diameter of the finite
elements .T ⊂ Ω.
Lemma 7.14.1 We assume that the operator . A is positive and that the matrix . M is
semi-admissible. Then, the problem (7.14.1) admits a unique solution .u h and

@seismicisolation
@seismicisolation
520 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

/
1 h
||u h ||0,Ω ≤
. + ||F||0,Ω .
α2 α

♦
Proof We set, for all .v and .w ∈ H 1 (Ω),
{
1
. E(w, v) = (Aw, v + h Av) (L 2 (Ω,R)) p − ((B − M)w, v)2 ds.
2 ∂Ω

Green’s formula leads

{
1 1
. E(vh , vh ) = ((A + A∗ )vh , vh )(L 2 (Ω,R)) p + h||Avh ||20,Ω + (Mvh , vh )2 ds.
2 2 ∂Ω

So,
. E(vh , vh ) ≥ α||vh ||20,Ω + h||Avh ||20,Ω .

We deduce the unique existence of the solution .u h . On the other hand,

.α||u h ||20,Ω + h||Au h ||20,Ω ≤ E(u h , u h )

=(F, u h + h Au h )(L 2 (Ω,R)) p
α 1 h h
≤ ||u h ||20,Ω + ||F||20,Ω + ||Au h ||20,Ω + ||F||20,Ω .
2 2α 2 2
So ( )
α h 1 h
. ||u h ||20,Ω + ||Au h ||20,Ω ≤ + ||F||20,Ω .
2 2 2α 2

Which completes the proof of the lemma. Q.E.D.

Theorem 7.14.2 We assume that the operator . A is positive and that the matrix . M
is semi-admissible and that the space .Vh is constructed with polynomials of degree
.≤ k on triangles (or polynomials . Q k isoparametric on quadrilaterals) with usual
generalization for .n ≥ 3. On the other hand, we assume that the matrix . M + M ∗ is
positive semi-definite. Then,
1
||u − u h ||0,Ω = O(h k+ 2 )
.

as soon as the exact solution .u is regular enough. ♦

Proof We set .e = u − u h . We have
{
1 1
. E(e, e) = ((A + A∗ )e, e)(L 2 (Ω,R)) p + h||Ae||20,Ω + (Me, e)2 ds.
2 2 ∂Ω

@seismicisolation
@seismicisolation
7.14 Petrov Galerkin-Type Methods 521

On the other hand, for any .vh ∈ Vh , we have

. E(e, e) = E(e, u − vh ) + E(e, vh − u h ).

Gold,
. E(e, vh − u h ) = 0.

It, therefore, remains to bound . E(e, u − vh ). We have

{
1
. E(e, u − vh ) = (Ae, u − vh ) 2 + h(Ae, A(u − vh ))(L 2 (Ω,R)) p − ((B − M)e, u − vh )2 ds.
(L (Ω,R)) p 2 ∂Ω

We have, by replacing .vh by .rh u, function .Vh interpolated of .u:

h 1
. |(Ae, u − rh u)(L 2 (Ω,R)) p | ≤ ||Ae||20,Ω + ||u − rh u||20,Ω
4 h
h
≤ ||Ae||0,Ω + O(h 2k+1 ),
2
4
and
h h
. h|(Ae, A(u − rh u))(L 2 (Ω,R)) p | ≤ ||Ae||20,Ω + ||A(u − rh u)||20,Ω
2 2
h
≤ ||Ae||20,Ω + O(h 2k+1 ).
2
On the other hand, we can write

{ {
. ((B − M)(u − u h ), u − rh u)2 ds = ((B − M)(u − rh u), u − rh u)2 ds+
∂Ω ∂Ω
{
((B − M)(rh u − u h ), u − rh u)2 ds.
∂Ω

We have, as soon as .u ∈ W k+1,∞ (Ω),

|{ |
| |
.| ((B − M)(u − r u), u − r u) ds | = O(h 2k+2 ).
| h h 2 |
∂Ω

As in Theorem 7.8.1, Remark 7.8.2 .(i), we have

|{ |
| |
.| ((B − M)(r u − u ), u − r u) ds | ≤ c1 h k+ 21 ||u h − rh u||0,Ω .
| h h h 2 |
∂Ω

@seismicisolation
@seismicisolation
522 7 Positive Symmetric First-Order Systems Within the Meaning of Friedrichs

Finally,

||u h − rh u||0,Ω ≤ ||u h − u||0,Ω + ||u − rh u||0,Ω

.

≤ ||e||0,Ω + O(h k+1 ).

Combining all these inequalities, we get

{
1
.α||e||0,Ω + (Me, e)2 ds + h||Ae||20,Ω ≤ E(e, e)
2
2 ∂Ω
3
≤ h||Ae||20,Ω + O(h 2k+2 ) + εc1 ||e||20,Ω ,
4

where the constant .ε > 0 is arbitrary. We finally know that

{
. (Me, e)2 ds ≥ 0.
∂Ω

α
By choosing .ε = 2c1
, we get

α 1
. ||e||20,Ω + h||Ae||20,Ω ≤ O(h 2k+1 ).
2 4
This completes the proof. Q.E.D.

Remark 7.14.3 In the definition of the scheme (7.14.1), we can replace the term
h Avh by .γ h Avh , where .γ is a strictly positive scalar to be chosen, which has the
.
effect of modulating the quantity dissipation associated with this term. ♦

References

1. J. Necas, Les méthodes directes en théorie des équations elliptiques. (French) Masson et Cie,
Éditeurs, Paris; Academia, Éditeurs, Prague (1967), 351 p
2. P.D. Lax, R.S. Phillips, Local boundary conditions for dissipative symmetric linear differential
operators. Comm. Pure Appl. Math. 13, 427–455 (1960)
3. K.O. Friedrichs, Symmetric positive linear differential equations. Comm. Pure Appl. Math. 11,
333–418 (1958)
4. R.S. Phillips, L. Sarason, Singular symmetric positive first order differential operators. J. Math.
Mech. 15, 235–271 (1966)
5. K.D. Lathrop, B. Carlson, Transport theory: the method of discrete ordinates, in Computing
Methods in Reactor Physics, ed. by H. Greenspan, C.N. Kelerb, D. Okrent, (Gordon and Breach,
1968), pp. 165–266
6. P. Lesaint, Sur la résolution des systèmes hyperboliques du premier ordre by des méthodes
d’éléments finis (Thèse, Pierre et Marie Curie, Paris, 1975)
7. P.G. Ciarlet, Cours Monréal (1975)
8. P.G. Ciarlet, P.-A. Raviart, Interpolation theory over curved elements, with applications to finite
element methods. Comput. Methods Appl. Mech. Engrg. 1, 217–249 (1972)

@seismicisolation
@seismicisolation
References 523

9. P.G. Ciarlet, P.-A. Raviart, The combined effect of curved boundaries and numerical integration
in isoparametric finite element methods: the mathematical foundations of the finite element
method with applications to partial differential equations, in Proceedings of the Symposium,
University Maryland, Baltimore, Md., 1972 (Academic Press, New York, 1972), pp. 409–474
10. P.G. Ciarlet, P.-A. Raviart, General Lagrange and Hermite interpolation in. R n with applications
to finite element methods. Arch. Rational Mech. Anal. 46, 177–199 (1972)
11. K.D. Lathrop, B. Carlson, Numerical solution of the Boltzmann transport equation. J. Comput.
Phys. 1, 173–197 (1967)
12. W.H. Reed, T.R. Hill, Triangular mesh methods for the neutron transport equation, in Confer-
ence on Mathematical Models and Computational Techniques for Analysis of Nuclear Systems
(Ann. Arbor., Michigan, April 9–11 1973)

@seismicisolation
@seismicisolation
Chapter 8
Approximation of the Transport
Equation in Plane Two-Dimensional
Geometry by Continuous and
Discontinuous Finite Element Methods

In this chapter, we study continuous and discontinuous approximation methods,

adapted to the structure of the equations, leading to linear systems of quasi-explicit
resolution and, therefore, commonly used in practice. All results of this chapter are
due to P. Lesaint and may be found in [1].

8.1 Transport Equation in Plane Geometry

8.1.1 Position of the Problem

We want to solve the transport equation in plane geometry, which is written in the
form of the following first-order problem:
∂u ∂u
μ
. +ν + σ u = f for (x, y) × (μ, ν) ∈ G × Q (8.1.1)
∂x ∂y

u(x, y, μ, ν) = 0 for (x, y, μ, ν) ∈ (∂G × Q)− ,

. (8.1.2)

with .μ and .ν are two scalars such that .μ2 + ν 2 ≤ 1, .σ > 0, .G is a bounded domain
of .R2 , with border .∂G, . Q is the unit disk .μ2 + ν 2 ≤ 1 and, where we put

(∂G × Q)− = {(x, y) × (μ, ν) ∈ ∂G × Q such that B := μn x + νn y < 0},

.

n and .n y designating the components of the normal outside .∂G (see, for instance,
. x

any of Refs. [2–13]). The function .u(x, y, μ, ν) represents a flux of neutrons at the
point of coordinates.(x, y), in the angular direction indicated by the vector.(μ, ν). The
quantity .σ represents a cross section and the term . f (x, y, μ, ν) takes into account

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 525
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_8
@seismicisolation
@seismicisolation
526 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

scattering, fission, and sources. The boundary conditions (8.1.2) express that the flux
of neutrons entering the system is zero. Let’s pose . A1 = μ, . A2 = ν, and . A0 = σ .
Nuclear center:
.( )
u: neutron flux
μ
. : flow direction.
ν
.σ : cross section
. f : source.

In dimension 1:
∂u
. + σ u = f.
∂x
Hence, { {
a a
.u(a) + σ u(s) ds = u(0) + f (s) ds.
0 0

.u(a):
{ a output stream
.σ 0 u(s) ds: what was absorbed
.u(0): the entrance
{a
.
0 f (s) ds what we brought.

In dimension 2: We pose ( )
μ
.v= .
ν

Hence,
v.∇u + σ u = f.
.

. B|∂G = μn x + νn y ,

where .n x and .n y are the components of the exterior normal to .∂G,

@seismicisolation
@seismicisolation
8.1 Transport Equation in Plane Geometry 527

∂ G = {(x, y) ∈ ∂G such that μn x + νn y < 0},

. −

∂ G = ∂G\∂− G.
. +

Incoming flow given: .u = g on .∂− G and .(B − M)(u − g) = 0.

8.1.2 Spherical Geometry

We consider the transport equation in one-dimensional geometry which is written in

the form of the following first-order problem ([14–17])

∂ 2 ∂
.μ (r u) + r ((1 − μ2 )u) + σ r 2 u = r 2 f (8.1.3)
∂r ∂μ

.u(R, μ) = 0, for μ < 0, (8.1.4)

with .μ ∈ [−1, 1] angular variable, .r spatial variable .0 < r ≤ R.

The function .u(r, μ) represents a flux of neutrons located at the distance .r from the
origin and having a velocity .−
→
v such that

.
−
→
r ·−
→
v = r |−
→
v |μ,

where .−
→r denotes the radius vector. The quantity .σ represents a cross section and
the function . f takes into account fission terms and sources. The boundary condition
(8.1.4) means that no neutron enters the sphere of radius . R. The equation (8.1.3) can,
also, be written

@seismicisolation
@seismicisolation
528 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

∂u ∂u
μr 2
. + 2r μu + r (1 − μ2 ) − 2μr u + σ r 2 u = r 2 f.
∂r ∂μ

Hence,
∂u ∂u
μr 2
. + r (1 − μ2 ) + σ r 2u = r 2 f (8.1.5)
∂r ∂μ

+ boundary conditions. The equation (8.1.5) render in the framework of Friedrichs

symmetric systems [18] and the admissible boundary conditions are nothing other
than the conditions (8.1.4). Indeed, for.r = 0 or for.μ = ±1, we have. B = 0. Accord-
ing to Lemma 7.4.7, we, therefore, have . M = 0 and, therefore, there are no bound-
ary conditions to impose for .r = 0 or for .μ = ±1. For .r = R, we have . B = μR 2 .
According to Lemma 7.4.7, we have . M = |μ|R 2 , and we deduce the boundary
conditions (8.1.4). We pose
{ 1
ml =
. μl u(r, μ) dr dμ, for l = 0, 1, 2, . . . .
−1

By formally integrating the equation (8.1.3) with respect to the variable .μ, we get
{
∂ 2 1
. (r m 1 ) + σ r 2 m 0 = r 2 f dμ. (8.1.6)
∂r −1

This relationship shows that the directional transfer mechanism expressed by the term
∂
.
∂μ
(r (1 − μ2 )u) has no influence on the total number of particles. If we integrate in
.r , the relation (8.1.6) between .a and .b, it comes

{ b { b { 1
. b m 1 (b) − a m 1 (a) +
2 2
σ r m 0 (r ) dr =
2
r 2 f dr dμ,
a a −1

which expresses the following conservation relation:

Outgoing stream+Captures=Incoming stream+Sources.
The equation (8.1.5) divided by .r 2 and the boundary conditions (8.1.4) can still be
written
du
. + σ u = f if x ∈]0, 1[
ds
u(0) = 0,

where . du
ds
denotes the derivative along the characteristic curves of equation

r 2 (1 − μ2 ) = cte.
.

@seismicisolation
@seismicisolation
8.1 Transport Equation in Plane Geometry 529

The solution .u at a point .s0 of a characteristic curve, therefore, only depends on

the values of .u (and the physical constants .σ and . f ) at the points of this character-
istic, upstream of .s0 . We pose . A1 = μr 2 , . A2 = r (1 − μ2 ), . A0 = σ r 2 and . F = r 2 f .
. B(R, μ) = μR gives if .μ ≥ 0, then . B = M and . B − M = 0 nothing to express. If
2

.μ ≤ 0, then . B = −M we express .1 (see figure above).

. B(0, μ) = 0
. B(r, ±1) = 0

dr dμ
. =
r μ
2 r (1 − μ2 )
1
gives .r (1 − μ2 ) 2 = cte. .C(x) is given by

.C(x) = −2μr + 2μr + 2σ r 2 = 2σ r 2 .

8.1.3 Discrete Ordinate Method

In practice, to perform the calculations, we often decouple the space variables .(x, y)
and the angular variables .(μ, ν), using the discrete ordinate method [14, Chap. 5],
which consists of choosing a set of angular directions .(μm , νm ), .1 ≤ m ≤ M and
solving the partial differential equations separately

∂u m ∂u m
μm
. + νm + σm u m = f m for (x, y) ∈ G,
∂x ∂y

@seismicisolation
@seismicisolation
530 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

u (x, y) = 0 for (x, y) ∈ ∂−m G,

. m

where

∂ m G = {(x, y) ∈ ∂G such that μm n x + νm n y < 0}, 1 ≤ m ≤ M

. −

u (x, y) = u(x, y, μm , νm ).
. m

In the following, we are interested in the numerical approximation of .u m (x, y), for
each fixed value .m, using finite element methods. To simplify the writing, we delete
.m and the continuous problem is written

∂u ∂u
. Au := μ +ν + σ u = f for (x, y) ∈ G ⊂ R2 , (8.1.7)
∂x ∂y

u(x, y) = 0 for (x, y) ∈ ∂− G,

. (8.1.8)

where
∂ G = {(x, y) ∈ ∂G such that B := μn x + νn y < 0}.
. − (8.1.9)

For all .u, .v ∈ H 1 (G), we have

{ ( ) { ( )
∂u ∂u ∂v ∂v
. μ +ν + σ u v d xdy = −μ −ν + σ v u d xdy+
G ∂x ∂y G ∂x ∂y
{
. (μn x + νn y )uv ds. (8.1.10)
∂G

By choosing .v = 1 in the equality (8.1.10), we can write

{ { ( )
∂u ∂u
. f d xdy = μ +ν + σ u d xd y
G ∂x ∂y
{G {
= σ u d xdy + (μn x + νn y )u ds,
G ∂G

or in an equivalent way
{ { { {
. (μn x + νn y )u ds + σ u d xdy = − (μn x + νn y )u ds + f d xdy.
∂+ G G ∂− G G

This last relation expresses the neutron conservation. The solution .u at point . M0 with
coordinates .(x0 , y0 ) depends only on the values of .σ , . f and .u along the characteristic
passing through point . M0 (i.e., line . D with equation .μ(y − y0 ) + ν(x − x0 ) = 0)
and upstream of . M0 , i.e., for .(x, y) ∈ D such that, we have

@seismicisolation
@seismicisolation
8.1 Transport Equation in Plane Geometry 531

μ(x − x0 ) < 0.
.

This remark plays a fundamental role during the numerical approximate problem
(8.1.7)–(8.1.8). Indeed, we will try to define numerical schemes such that the con-
servation of neutrons is satisfied in a discrete way and such that the approximate
solution .u h can be calculated by following the characteristic directions (i.e., by track
or following the neutrons).
Furthermore,
{ ( )
∂u ∂u
(Au, u)0,G
. = μ +ν + σ u u d xdy
∂x ∂y
{G {
1
= σ u 2 d xd y + (μn x + νn y )u 2 ds. (8.1.11)
G 2 ∂G

On.∂− G,. B < 0 so. M = −B. Thus,. B − M = 2B < 0. Furthermore,.(B − M)u = 0

if, and only if, .u = 0. Let

∂ G = {(x, y) ∈ ∂G such that B = μn x + νn y > 0}.

. + (8.1.12)

On .∂+ G, . B ≥ 0 so . M = B. Nothing is expressed at the exit.

8.1.4 Existence and Uniqueness Result

The problem (8.1.7)–(8.1.8), therefore, corresponds to a positive Friedreichs sym-

metric system, which reduces to a single equation and, we can cite the existence
result of a solution given in [18, pp. 382–383].
Theorem 8.1.1 We assume that the boundary .∂G is of class .C 2 (or is of class
.C piecewise see [19, p. 241]). Then, for all . f belonging to . H (G), the problem
2 1

(8.1.7)–(8.1.8) has a unique strong solution .u belonging to . H (G).

1
♦

Remark 8.1.2 Theorem 8.1.1 is actually a regularity result for the solution of the
problem (8.1.7)–(8.1.8). In the general case, the solution does not belong to . H 2 (G),
even if the second member . f is very regular, as the following simple example
shows. ♦

Example 8.1.1 Let .G be the square .]0, 1[×]0, 1[. We choose . f = σ = 1 and, we
assume that .μ and .ν are strictly positive. The solution .u(x, y) of (8.1.7)–(8.1.8) is
given by { y
1 − e− ν if νx > μy
.u(x, y) =
1 − e− μ if νx < μy.
x

@seismicisolation
@seismicisolation
532 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

8.1.5 Approximate Solution Over the Entire Domain

We will approach the problem (8.1.7)–(8.1.8) by imposing the following principle:

Let .τh be a triangulation of .G in elements . K . We pose

∂ K = {(x, y) ∈ ∂ K such that μn x,K + νn y,K < 0},

. −

∂ K = {(x, y) ∈ ∂ K such that μn x,K + νn y,K > 0},

. +

where .n x,K and .n y,K are the components of the exterior normal .n K to .∂ K .

⎧
⎨ We calculate the approximate solution u h on an element K only if
. we already know the value of u h on the lightened portion ∂− K (8.1.13)
⎩
border ∂ K .

We will say that the element . K is frontier if it has at least one side included in .∂G,
and that the element . K is semi-frontier if it is not frontier and if at least one of its
vertices belongs to .∂G. The boundary elements . K 1 , . K 2 , . . . , K r corresponding to
.∂− G are numbered clockwise.

@seismicisolation
@seismicisolation
8.1 Transport Equation in Plane Geometry 533

When two successive boundary elements . K i and . K i+1 have no common side, there
is at least one semi-boundary element between . K i and . K i+1 . We say that a side of
i i+1
. K (resp. . K ) is attached to . K i+1 (resp. . K i ) if this side is either common to . K i
i+1
and . K , or if it is a subset of the union of the sides of the semi-boundary elements
located between . K i and . K i+1 (see figure above).
The (8.1.13) principle can solve the problem (8.1.7)–(8.1.8) by approximately fol-
lowing the characteristic direction. We now recall the proof of a lemma [20, pp.
15–19] which shows that it is possible to calculate the approximate solution over the
entire domain .G, by imposing the principle (8.1.13).

Lemma 8.1.3 We can classify the elements . K of .τh in an order . K 1 , . K 2 , . . . , K N

such that for all . j, .1 ≤ j ≤ N , each side of . K j contained in .∂− K j is either a subset
of .∂− G or a subset of .∂+ K i for an index .i < j. ♦

Proof Let us show that there exists at least one boundary element . K such that, we
have .∂− K ⊂ ∂− G. For this, suppose that .∂− K i /⊂ ∂− G for all .i = 1, . . . , r and show
that we arrive at a contradiction. We consider the first boundary element . K 1 and, we
use the notations of the following figure:

@seismicisolation
@seismicisolation
534 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

In the triangular (resp. quadrilateral) case, the side .[A1 , A3 ] (resp. .[A1 , A4 ]) of
.K 1 is a subset of .∂+ K 1 . Otherwise . K 1 would not be the first boundary element of
.∂− G. Then, the side .[A 2 , A 3 ] of . K which is attached to . K is included in .∂− K .
1 2 1

Otherwise, we would have .∂− K = [A1 , A2 ] ⊂ ∂− G, which was excluded above.

1

We deduce that the side of . K 2 which is attached to . K 1 belongs to .∂+ K 2 . More

generally, we have the following property, for all .i = 1, . . . , r − 1: the side of . K i
which is attached to . K i+1 is a subset of .∂− K i and the side of . K i+1 that is attached
to . K i is a subset of .∂+ K i+1 . We now consider the last boundary element . K r and, we
use the notations of the following figure (Fig. 8.1).
In the triangular (resp. quadrilateral) case, the side .[A1 , A3 ] (resp. .[A1 , A4 ]) of
.K
r
is a subset of .∂+ K r . Moreover, the side .[A2 , A3 ] is also a subset of .∂+ K r .
Otherwise . K r would not be the last boundary element of .∂− G. We, therefore, obtain
.∂− K = [A 1 , A 2 ] ⊂ ∂− G, which was excluded above. So, there is a border element. K
r

such as.∂− K ⊂ ∂− G. Now, n let. K 1 be a border element of.∂− G such that.∂− K 1 ⊂ ∂− G.

We define .G1 = G\(G K 1 ). Each face included in .∂− G1 is either a subset of .∂− G
or a subset of.∂+ K 1 . From what was demonstrated earlier, there is a border element. K 2
corresponding to .∂− G1 and such that .∂− K 2 ⊂ ∂− G1 etc. By repeating this process,
we take into account all the elements of .G and, we obtain a classification of these

Fig. 8.1 Triangular and quadrilateral

@seismicisolation
@seismicisolation
8.1 Transport Equation in Plane Geometry 535

elements in the order . K 1 , . K 2 , . . . , K N such that the conclusions of the Lemma 8.1.3
are satisfied. Q.E.D.

Remark 8.1.4 The numbering of the elements .1, . . . , N is not necessarily unique,
as shown in Fig. 8.1. This is also the case as soon as there are at least two boundary
elements . K 1 and . K 2 relative to .∂− G and such that
. ∂− K 1 ⊂ ∂− G and ∂− K 2 ⊂ ∂− G. ♦

8.1.6 Approximate Solution

To approach the problem (8.1.7)–(8.1.8), we consider a triangulation .τh of .G in

triangular or quadrilateral finite elements. K , of diameter.≤ h. We recall the following
notations:

@seismicisolation
@seismicisolation
536 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

. Kh = diameter of K

ρ = sup{diameter of circles contained in K }

. K

θ
. i,K , 1 ≤ i ≤ 4 = angles of K , if K is a quadrilateral,

.Z K = distances between the midpoints of the diagonals of K , if K is a quadrilateral.

Suppose there is a constant .C independent of .h such that

h ≤ Cρ K for all K ∈ τh
. K (8.1.14)

and a constant .γ independent of .h, with .0 < γ < 1, such that

. max | cos(θi,K )| ≤ γ for any quadrilateral K ∈ τh . (8.1.15)

1≤i≤4

• When the element . K is a triangle, we assume that it is the range by invertible

.
-
affine transformation . FK (see the proof of Lemma 4.6.8) of a reference triangle . K
-
(in general, we choose for . K the isosceles right triangle whose sides of the right
angle are of length equal to one).
In the triangle case, to each element . K of .τh , we associate a space of finite dimension
such that
. PK = { p = -p ◦ FK−1 , - p ∈ Pk }. (8.1.16)

• When. K is a non-degenerate convex quadrilateral, it is the range by a transformation

.
FK (see Lemma 4.9.1) belonging to .(Q 1 )2 of the reference square . K
. - = [−1, 1] ×
[−1.1].

@seismicisolation
@seismicisolation
8.1 Transport Equation in Plane Geometry 537

In the quadrilateral case, to each element . K of .τh , we associate a space of finite

dimension such that
p ◦ FK−1 , -
. PK = { p = - p ∈ Q k }. (8.1.17)

In the two cases, we have the inclusion

. PK ⊂ H 1 (K ), for all K ∈ τh .

Setting | |
. Wh = PK ,
K ∈τh

we have the inclusion

. Wh ⊂ L 2 (G).

If the degrees of freedom of the element . K- are chosen as in Example 4.1.5 in the
triangular case, and as in Example 4.1.6, in the quadrilateral case, and if we denote
. Vh the subspace of . Wh of the continuous functions at the nodes of the triangulation,
we, then, have n
. Vh ⊂ H (G) C 0 (G, R).
1

We note . X h the subspace of .Vh defined by

. X h = {vh ∈ Vh such that vh = 0 at nodes belonging to ∂− G}.

It is possible to approach the problem (8.1.7)–(8.1.8) using a Galerkin type method.

For that, let .Vh (we can choose .Vh = X h ) be a finite-dimensional subspace of . H 1 (G).
We seek .u h ∈ Vh such that

@seismicisolation
@seismicisolation
538 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …
{ ( ) {
∂u h ∂u h
.μ +ν + σ u h − f vh d xd y − (μn x + νn y )u h vh ds = 0
G ∂x ∂y ∂− G
(8.1.18)
for all .vh ∈ Vh . The numerical resolution of the problem (8.1.18) requires the inver-
sion of a linear system whose number of unknowns is equal to the number of nodes
of the triangulation. The matrix of the system is not, in general, triangular block and
one cannot solve while following the characteristic direction. We pose
{ ( ) {
∂vh ∂vh
. Y (vh ) = μ +ν + σ vh vh d xd y − (μn x + νn y )vh2 ds.
G ∂x ∂y ∂− G

According to (8.1.11), we have

{ {
1
. Y (vh ) = σ vh2 d xd y + |μn x + νn y |vh2 ds.
G 2 ∂G

This leads to the existence of a unique approximate solution .u h for the problem
(8.1.18). On the other hand, we have
{
.Y (u h ) = f u h d xd y
G
≤ || f ||0,G ||u h ||0,G ,

from which, we deduce the following result:

Theorem 8.1.5 We suppose that .σ (x, y) ≥ α > 0 and that . f belongs to . L 2 (G).
Then, the problem (8.1.18) has a unique solution .u h ∈ Vh and, we have
. ||u h ||0,G ≤ C|| f ||0,G . ♦

Remark 8.1.6 We can always reduce to the case, where we have .σ (x, y) ≥ α > 0,
because, using the change of function
( )
y
λ μ+ν
x
w
u=e
. ,

the equation (8.1.1) becomes

( )
∂w ∂w −λ y
μ+ν
x
. μ +ν + (σ + 2λ)w = f e . ♦
∂x ∂y
When the spaces . PK , . K ∈ τh are chosen as in (8.1.16) or (8.1.17), and the space
. Vh is constructed as above, we have the following result.
Theorem 8.1.7 We assume that the hypotheses (8.1.14) and (8.1.15), as well as the
inclusions
- ⊂ Pk when K
.P - is triangle (8.1.19)

. P - is a square
- ⊂ Q k when K (8.1.20)

@seismicisolation
@seismicisolation
8.2 Continuous Finite Element Methods 539

are satisfied for an integer .k ≥ 1. Let .u h ∈ Vh be the solution of the problem (8.1.18).
Then, if the solution .u of the problem (8.1.7)–(8.1.8) belongs to . H k+1 (G), we have
. ||u − u h ||0,G ≤ Ch k |u|k+1,2,G . ♦
When the inclusions (8.1.19) and (8.1.20) are satisfied, we can replace .vh by .1 in
the equality (8.1.18). Applying the Eq. (8.1.10), we obtain the neutron conservation
relation
{ { { {
. (μn x + νn y )u h ds + σ u h d xd y = − (μn x + νn y )u h ds + f d xdy.
∂+ G G ∂− G G

Theorem 8.1.8 We assume that the domain .G is a rectangle triangulated into equal
- = Q 1 . Let .u h ∈ Vh be
rectangles, that the hypothesis (8.1.14) is satisfied, and that . P
the solution of the problem (8.1.18). Then, if the solution .u of the problem (8.1.7)–
(8.1.8) belongs to . H 3 (G), we have
. ||u − u h ||0,G + ||u − u h ||0,∂G ≤ Ch 2 ||u||3,2,G . ♦
Remark 8.1.9 Theorems 8.1.7 and 8.1.8 and Remark 8.1.2 show that in some prac-
tical cases, it may be unnecessary to use polynomials of degree.≥ 2. A similar remark
is found in [21, p. 10]. ♦

8.2 Continuous Finite Element Methods

Let .τh be a triangulation of .G in finite elements . K and let . PK be a finite-dimensional

space defined on . K for . K ∈ τh . We pose

. M K = number of nodes of K not located on ∂− K .

We give ourselves a set .{v j } M K

j=1 of . M K linearly independent test functions.

Definition 8.2.1 Continuous methods: We seek .u h ∈ PK , such that

{
. (Au h − f )v j d xd y = 0 for 1 ≤ j ≤ M K , (8.2.1)
K

u given to nodes located on ∂− K ,

. h (8.2.2)

where . A is the operator given in (8.1.7). ♦

8.2.1 Approximate Way Using the Quadrature Formula

In practice, we have to calculate the integrals involved in the equality (8.2.1) in an

approximate way using the quadrature formula

@seismicisolation
@seismicisolation
540 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

{ E
NK
. g d xdy ∼ wl,K g(bl,K ), (8.2.3)
K l=1

where the points .bl,K belong to . K and the weights .wl,K are positive, .1 ≤ l ≤ N K .
The equation (8.2.1) is, then, written

E
NK
. wl,K ((Au h − f )v j )(bl,K ) = 0, 1 ≤ j ≤ M K . (8.2.4)
l=1

We assume the following hypotheses:

(H 4) The constant function is a linear combination of the functions .v j , .1 ≤ j ≤ M K .
.

(H 5) The integral
.
{ ( )
∂vh ∂vh
. μ +ν d xd y
K ∂x ∂y

is calculated exactly, for all .vh ∈ PK .

(H 6) For all .vh ∈ PK and for any face included in .∂ K , the integral of .vh along this
.
face is entirely determined by the values of .vh at the nodes located on this side.
According to the hypothesis .(H 4), we can replace .v j by the constant one in the
equality (8.2.1), and, we have

E
NK
. wl,K (Au h − f )(bl,K ) = 0.
l=1

According to the hypothesis .(H 5), it comes

{ ( ) E
NK
∂u h ∂u h
. μ +ν d xd y + wl,K (σ u h − f )(bl,K ) = 0,
K ∂x ∂y l=1

and is again
{ E
NK {
. (μn x + νn y )u h ds + wl,K (σ u h )(bl,K ) = − (μn x + νn y )u h ds+
∂+ K l=1 ∂− K

E
NK
. wl,K f (bl,K ). (8.2.5)
l=1

The hypothesis .(H 6) and the equality (8.2.5) express in a discrete way the
conservation of neutrons.

@seismicisolation
@seismicisolation
8.2 Continuous Finite Element Methods 541

Lemma 8.2.2 A sufficient condition for the scheme (8.2.4) to discretely satisfy a
particle conservation property is that the three hypotheses .(H 4), .(H 5), and .(H 6)
are simultaneously verified. ♦

We assume that the domain.G is triangulated into convex quadrilaterals. K of diameter

≤ h such that four sides come from a vertex belonging to .G.
.

We make the following (stability) hypothesis:

All quadrilaterals K ∈ τh have two and only two illuminated faces.

. (8.2.6)

- = [−1, 1] ×
Let. FK ∈ (Q 1 )2 be the transformation that sends the reference square. K
[−1, 1] to the quadrilateral . K . Let .(x, y) = FK (ξ, η). We have

1 1 1 1
.x = (1 + ξ )(1 + η)x1 + (1 − ξ )(1 + η)x2 + (1 − ξ )(1 − η)x3 + (1 + ξ )(1 − η)x4
4 4 4 4

1 1 1 1
.y = (1 + ξ )(1 + η)y1 + (1 − ξ )(1 + η)y2 + (1 − ξ )(1 − η)y3 + (1 + ξ )(1 − η)y4 ,
4 4 4 4

where the .(xi , yi ) are the coordinates of the vertices . Ai , .1 ≤ i ≤ 4 of the quadrilat-
eral . K .

8.2.2 Quadrilateral Finite Elements

Consider two quadrilateral finite elements:

• The quadrilateral of type .k, characterized by the triple .(K , E K , PK ), where
.

-k ), H
E K = FK ( H
. -k being defined as (4.4), (8.2.7)

@seismicisolation
@seismicisolation
542 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

. p ◦ FK−1 , ∀-
PKk = { p = - p ∈ Q k }. (8.2.8)

• The non-conforming quadrilateral, characterized by the triple .(K , E K, , PK, ), where

.

E K, = {Ai }i=5
.
8
,

. p ◦ FK−1 , ∀-
PK, = { p = - p ∈ P1 },

and, we have the relation

. p(A5 ) + p(A7 ) = p(A6 ) + p(A8 ) = 2 p(A9 ).

8.2.3 Approximate Space

We define the space .Vhk (resp. .Vh, ) of functions whose restriction to each element . K
belongs to . PKk (resp. . PK, ) and which are continuous at the nodes of triangulation. We
have the inclusion n
. Vh ⊂ C (G, R) H 1 (G)
k 0

but, in general, we do not have the inclusion

. Vh, ⊂ C 0 (G, R).

We pose

. X hk = {vh ∈ Vhk such that vh = 0 at nodes located on ∂− G}

and
. X h, = {vh ∈ Vh, such that vh = 0 at nodes located at ∂− G}.

When the finite element used in the continuous method (8.2.1), (8.2.2) is the quadri-
lateral of type .k, the relation (8.2.2) and the hypothesis (8.2.6) imply that we already
know .2k + 1 values of the approximate solution .u h . There remains, therefore, a
number of unknowns equal to .(k + 1)2 − 2k − 1 = k 2 = dim PKk−1 .
We can choose the space . PKk−1 as the space of the test functions.
When using the non-conforming quadrilateral, two values of the approximate
solution are already known, and given the relation (8.2.7), only one unknown remains
to be determined. We choose (to satisfy the hypothesis .(H 4)) the test function equal
to one (Fig. 8.2).

@seismicisolation
@seismicisolation
8.2 Continuous Finite Element Methods 543

A2 A5
A1

A6 A8

A4
A3 A7
)
) (
(

Fig. 8.2 Quadrilateral of type .k (k = 3), 9 unknowns non-conforming quadrilateral, 1 unknown

8.2.4 Construction of the Quadrature Formula

The quadrature formula (8.2.3) can be constructed as follows.

• We give ourselves, on the segment .[−1, 1], the quadrature formula
.

{ 1 E
k
. f (ξ ) dξ ∼ -
wi f (gi ), (8.2.9)
−1 i=1

where the .gi , .1 ≤ i ≤ k, are the abscissas of Gauss-Legendre [22] and the .-
wi are the
(positive) weights such that the formula (8.2.9) is exact for the polynomials of . P2k−1 .
-, we can write the following quadrature formula, exact for the
• On the square . K
.
polynomials of . Q 2k−1
{ 1 { 1 E
k
. f (ξ, η) dξ dη ∼ -
wi -
w j f (gi , g j ), (8.2.10)
−1 −1 i, j=1

and, on the finite element . K , it comes

{ E
k
. f (x, y) d xd y ∼ wiKj f (giKj ), (8.2.11)
K i, j=1

where
.wiKj = JK (gi , g j )-
wi -
w j , giKj = FK (-
gi j ), -
gi j = (gi , g j ).

@seismicisolation
@seismicisolation
544 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

8.2.5 Discrete Norms and Semi-norms

We define the following discrete norms and semi-norms:

We define .| · |h,K ,k and .| · |h,k by
⎛ ⎞ 21
E
k
.|v|h,K ,k = ⎝ wiKj (v(giKj ))2 ⎠ ,
i, j=1

and ⎛ ⎞ 21
E
|v|h,k = ⎝
. |v|2h,K ,k ⎠ ,
K ∈τh

and, we note .(·, ·)h,K ,k and .(·, ·)h,k the corresponding scalar products. We will omit
the index .k whenever there is no ambiguity. Let . S be any face included in the .∂G
boundary. We consider on . S the quadrature formula, exact for the polynomials of
. P2k−1 , using the Gauss-Legendre points . Bi , and the corresponding weights .wi , .1 ≤
i ≤k
{ E k
. f (s) ds ∼ wi f (Bi ).
S i=1

We define .(·)h,S,k and .(·)h,k by

( k ) 21
E
(v)h,S,k =
. wi (v(Bi ))2 ,
i=1

and ( ) 21
E
. (v)h,k = (v)2h,S,k ,
S⊂∂G

and, we note .(·, ·)h,S,k and .(·, ·)h,k the corresponding scalar products.

8.2.6 Interpolate Function

Let .Yhk−1 be the space

Y k−1 = {w ∈ L 2 (G, R) such that w|K ∈ PKk−1 , ∀K ∈ τh }.

. h

@seismicisolation
@seismicisolation
8.2 Continuous Finite Element Methods 545

We first construct functions .rh u ∈ X hk and .rh, u ∈ X h, which interpolate the function
.u at certain points and, we estimate the expressions

|(A(u − rh u), w)0,G |

.|A(u − rh u)|h,k , |A(u − rh, u)|h,1 , sup
w∈Yhk−1 ||w||0,G

that we will use them later. If the interpolate .rh u (resp. .rh, u) is defined as before,
Example 4.1.6 (resp. Remark 5.4.15) and if we brutally use the Lemma 4.1.14, we
get the markups
.|A(u − r h u)|h,k ≤ Ch |u|k+1,2,G ,
k

|A(u − rh, u)|h,1 ≤ Ch|u|2,2,G ,

.

|(A(u − rh u), w)0,G |

. sup ≤ Ch k |u|k+1,2,G .
w∈Yhk−1 ||w||0,G

We will show that we can in fact obtain bounds in .o(h k+1 ), .o(h 2 ) and .o(h k+1 ),
respectively.

Definition 8.2.3 To any function .v defined and continuous on an element . K of .τh ,

of vertices . Ai , .1 ≤ i ≤ 4, we associate its . PK, -interpolated function .r K, v as follows

r, v = -
. K v ◦ FK−1 , -
r ,- v = v ◦ FK ,

r ,-
where the transformation . FK is defined in Lemma 4.9.1, and .- v is the unique
polynomial of . P1 satisfying Remark 5.4.15
1 - -i−3 )), 5 ≤ i ≤ 7.
. r , v(Ai ) =
- (-
v( Ai−4 ) +-
v( A ♦
2
Let.v be a function of.C 0 (G, R). Its. X h, -interpolated function.rh, v is the unique element
of . X h, whose restriction to each element . K ∈ τh is equal to .r K, v. Let . K be an element
of .τh , and let .v ∈ H 3 (G) and .rh, v ∈ X h, be its interpolated. We have
( ) ( )
∂ , ∂y ∂ , ∂y ∂ ,
.area(K ) (v − rh v) (G K ) = (-
v −-r- v) − (-
v −-r- v) (0, 0),
∂x ∂η ∂ξ ∂ξ ∂η
(8.2.12)
where .G K denotes the range of . FK of the point with coordinates .ξ = η = 0.
• On the segment .[−1, 1], we consider the quadrature formula
.

{ 1 E
k+1
. f (ξ ) dξ ∼ -
wil f (li ), (8.2.13)
−1 i=1

@seismicisolation
@seismicisolation
546 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

where the .li , .1 ≤ i ≤ k + 1, .l1 = −1, .lk+1 = 1, are the abscissas of Gauss-Lobatto
[22] and the .- wil are the weights such that the quadrature formula (8.2.13) is exact for
the polynomials of . P2k−1 . Let .- li j be the point of the reference square with coordinates
.(li , l j ), .1 ≤ i, j ≤ k + 1.

Definition 8.2.4 To any function .v ∈ C 0 (K , R), we associate its . PKk -interpolated

function .r K v in the following way

r v =-
. K v ◦ FK−1 , -
r- v = v ◦ FK , (8.2.14)

where .-
r- v at points .-
v is the unique polynomial of . Q k equal to .- li j , .1 ≤ i, j ≤ k + 1. ♦

Let.v be a function of.C 0 (G, R). Its. X hk -interpolated function.rh v is the unique element
of . X hk whose restriction to each element . K ∈ τh is equal to .r K v. Let . K be a rectangle
of .τh , .v be any function of . H k+2 (G), and for all .w ∈ PKk−1 , then we have
/ \ { ( )
∂ ∂y ∂
. (v − rh v), w = (-
v −-
r-
v) -w dξ dη. (8.2.15)
∂x 0,K ∂η -
K ∂ξ

8.2.7 Hypotheses

We consider the following hypotheses:

(H 7) For any element . K ∈ τh , we have the inequality
.

. Z K ≤ ch 2K . (8.2.16)

(H 8) Let . Ai , .1 ≤ i ≤ 4 be the vertices of any element . K of .τh , and, we assume the

.
inequalities:

.|μ(y1 − y2 + y3 − y4 ) − ν(x1 − x2 + x3 − x4 )|
≤ C0 h K |μ(y1 − y4 + y2 − y3 ) − ν(x1 − x4 + x2 − x3 )|, (8.2.17)

.|μ(y1 − y2 + y3 − y4 ) − ν(x1 − x2 + x3 − x4 )|
≤ C0 h K |μ(y1 − y2 + y4 − y3 ) − ν(x1 − x2 + x4 − x3 )|, (8.2.18)

for all . K ∈ τh , where the constant .C0 is independent of .τh .

Remark 8.2.5 Let .h 0 be a real such that .0 < h 0 < C10 . The hypothesis .(H 8) implies
that if, we have .h ≤ h 0 , the two angles formed by the characteristic direction .(μ, ν)

@seismicisolation
@seismicisolation
8.2 Continuous Finite Element Methods 547

−−−→ −−−→ −−−→ −−−→

and two opposite sides . A1 A2 and . A4 A3 (or (. A2 A3 and . A1 A4 ) either have the same
sign, or are both equal to 0. This assumption is not equivalent to the assumption
(8.2.6) and allows to take into account the case, where an element has two sides
parallel to the feature. ♦

8.2.8 Estimate Between the Function and Its Interpolate

Lemma 8.2.6 (Lesaint [1]) Let .v ∈ H 3 (G) and .rh, v ∈ X h, be its interpolated. We
assume that .(τh )h is a regular family of triangulations. Then, for all . K ∈ τh , we have
| | | |
|∂ | |∂ |
| (v − r , v)| + | (v − r , v)| ≤ C(Z K |v|2,2,K + h 2K |v|3,2,K ),
|∂x
. h | |∂y h |
h,K ,1 h,K ,1
(8.2.19)
where .C is a constant .> 0 independent of .h. ♦
Proof Let . K be an element of .τh . The linear mapping

∂
-
v −→
. (- r ,-
v −- v)
∂ξ

-) into .R and is identically zero for all .-

is continuous from . H 3 ( K v ∈ P2 . According
to Lemma 3.4.15, we deduce that
| |
|∂ |
| ,
v)(0, 0)|| ≤ C|-
.
| ∂ξ (-
v −-
r- v|3,2, K- .

Using the Lemma 4.1.7, inequality (4.1.3), it comes

| |
|∂ |
| ,
v)(0, 0)|| ≤ C inf (JK (ξ, η))− 2 (h K Z K |v|2,2,K + h 3K |v|3,2,K ).
1
.
| ∂ξ (-
v −-
r-
−1≤ξ,η≤1
(8.2.20)
∂ ∂
The previous inequality (8.2.20) is also valid when we replace . ∂ξ by . ∂η . On the other
hand, one can show that
| | | |
|∂y | |∂y |
| | | |
.
| ∂η (0, 0)| + | ∂ξ (0, 0)| ≤ Ch K . (8.2.21)

By combining the relations (8.2.12), (8.2.20), and (8.2.21), we have

| |
|∂ |
| , |
.
| ∂ x (v − rh v)| ≤ C(Z K |v|2,2,K + h 2K |v|3,2,K ).
h,K ,1

The previous inequality is also valid with . ∂∂x replaced by . ∂∂y . From which, we deduce
the inequality (8.2.19). Q.E.D.

@seismicisolation
@seismicisolation
548 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

When the elements . K ∈ τh are rectangles, we have

Lemma 8.2.7 (Lesaint [1]) We assume that the domain .G is the square .(]0, 1[)2 ,
triangulated as above. Let .v be any function of . H k+2 (G), where .k is an integer .≥ 1,
and let .rh v be its . X hk -interpolated function. Then, for all . K ∈ τh , we have
|/ \ || ||/ \ ||
| ∂
| | | ∂ |
.| (v − rh v), w |+| (v − rh v), w | ≤ Ch k+1
K |v|k+2,2,K ||w||0,K ,
| ∂x 0,K | | ∂ y 0,K |
(8.2.22)
for all .w ∈ PKk−1 , where .C is a constant .> 0 independent of .h. ♦

Proof Let . K be a rectangle of .τh . The linear mapping

{ ( )
∂
-
v −→
. (-
v −-
r-
v) -w dξ dη
-
K ∂ξ

-) into .R of norm .≤ C||-

is continuous from . H k+2 ( K - v||k+2,2, K- ||-
w||0, K- , and it is iden-
U
tically zero for all .-
v ∈ Q k Pk+1 . Indeed, if .-v ∈ Q k , we have .- r-v =- v; if, we have
.-
v = ηk+1 , then .-
r-
v only depends on .η; finally if .-
v = ξ k+1 , then .-
v −- r-v is polynomial
of degree .k + 1 in .ξ which vanishes at the Gauss-Lobatto abscissa, i.e.,

-
v −-
. r-
v = (ξ − l1 )(ξ − l2 ) · · · (ξ − lk+1 ).

The first derivative in.ξ of this polynomial is a polynomial of degree.k which vanishes
at the abscissa of Gauss-Legendre [22]

∂
. (-
v −-
r-
v) = (k + 1)(ξ − g1 ) · · · (ξ − gk ). (8.2.23)
∂ξ

Since .-
w is a polynomial of . Q k−1 , the integral
{ ( )
∂
. (-
v −-
r-
v) -w dξ dη
-
K ∂ξ

is calculated exactly using the quadrature formula (8.2.10). The relation (8.2.23)
implies that this integral is zero. Applying the Lemma 3.4.17, it comes
|{ ( ) |
| ∂ |
| w dξ dη|| ≤ C|-
- v|k+2,2, K- ||-
.
| - ∂ξ (-
v −-
r-
v) - w||0, K- .
K

The element . K being a rectangle, we have according to the Lemma 4.1.6

v|k+2,2, K- ≤ C(JK )− 2 (h K )k+2 |v|k+2,2,K ,

1
.|-

and

@seismicisolation
@seismicisolation
8.2 Continuous Finite Element Methods 549

w||0, K- ≤ C(JK )− 2 ||w||0,K .

1
||-
.

On the other hand, we have

| |
|∂y |
.| | (JK )−1 ≤ Ch −2
| ∂η | ≤ Ch K , K .

By combining the relation (8.2.15) with the last five inequalities, we obtain the
upper bound |
|/ ∂ \ ||
| |
.| (v − rh v), w | ≤ Ch k+1
K |v|k+2,2,K ||w||0,K .
| ∂x 0,K |

This inequality is also valid when . ∂∂x is replaced by . ∂∂y . From which, we deduce the
inequality (8.2.22). Q.E.D.

Lemma 8.2.8 (Lesaint [1]) Let .v be any function of . H k+2 (G), for an integer .k ≥ 1
and let .rh v be its . X hk -interpolated function, defined in (8.2.14). We assume that .(τh )h
is a regular family of triangulations. Then, for all . K ∈ τh , we have
| | | |
|∂ | |∂ |
| | | |
| ∂ x (v − rh v)| + | ∂ y (v − rh v)| ≤ C Z K h K |v|k+1,2,K + Ch K |v|k+2,2,K ,
k−1 k+1
.
h,k h,k
(8.2.24)
where .C is a constant .> 0 independent of .h, for all . K ∈ τh . ♦

Proof Let . K be any quadrilateral in .τh . We have

( ) ( )
∂ ∂y ∂ ∂y ∂
. J K (-
gi j ) (v − rh v) (gi j ) = (-
v −-
r-
v) − (-
v −-
r-
v) (-
gi j ),
K
∂x ∂η ∂ξ ∂ξ ∂η
(8.2.25)
for .1 ≤ i, j ≤ k. The linear map
( )
∂
-
v −→
. (-
v −-
r-
v) (-
gi j )
∂ξ
U
-) into .R and is identically zero for all .-
is continuous from . H k+2 ( K v ∈ Q k Pk+1 .
Indeed, as in the proof of Lemma 8.2.7, we have .- v −-r-v = 0 for all .-
v ∈ Q k , and .-
r-
v
only depends on the variable .η if .- v = ηk+1 . If .-
v = ξ k+1 , then the equality (8.2.23)
holds, which means that ( )
∂
. (-
v −-
r-
v) (- gi j ) = 0.
∂ξ

Applying the Lemma 3.4.17, it comes

|( ) | ( [ 2 ] )
| ∂ | ∂-v
| | -
.
| ∂ξ (-
v −-
r- gi j )| ≤ C [-
v) (- v]k+2,2, K- +
∂ξ ∂η k,2, K-
. (8.2.26)

@seismicisolation
@seismicisolation
550 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

∂
We have a similar inequality for the term . ∂η (-
v −-
r-
v). On the other hand, according
to Lemma 4.1.7, we have

(JK (ξ, η))− 2 h k+2

1
[-
.v]k+2,2, K- ≤ C inf K |v|k+2,2,K , (8.2.27)
−1≤ξ,η≤1

and
[ ]
∂ 2-
v
(JK (ξ, η))− 2 (h kK Z K |v|k+1,2,K + h k+2
1
. ≤C inf K |v|k+2,2,K ).
∂ξ ∂η -
k,2, K −1≤ξ,η≤1
(8.2.28)
Finally, the Lemma 4.1.5 allows to write

sup JK (ξ, η)
. ≤ C. (8.2.29)
inf JK (ξ, η)

By combining the relations (8.2.25), (8.2.26), (8.2.27), (8.2.28), and (8.2.29), for
| values of|.i and . j, .1 ≤ i, j ≤ k, we obtain the inequality (8.2.24) for
the different
the term .| ∂∂x (v − rh v)|h,K ,k . The proof is identical for upper bounding the term
| |
|∂ |
.|
∂y
(v − r h v) | . Q.E.D.
h,K ,k

8.3 Scheme of Continuous Methods

We can now define the following continuous methods.

8.3.1 Scheme 1 (Generalized DSN Scheme)

Find .u h ∈ X h, such that

area(K ) (Au h − f )(G K ) = 0, for all K ∈ τh ,

.

where .G K denotes the range of . FK of the point with coordinates .ξ = η = 0, or

equivalently
.(Au h − f )(G K ) = 0, for all K ∈ τh .

Scheme 1 can be written explicitly, for .k = 1, by using the following formulas

∂v v ∂y
∂- v ∂y
∂-
. K J = − , (8.3.1)
∂x ∂ξ ∂η ∂η ∂ξ

@seismicisolation
@seismicisolation
8.3 Scheme of Continuous Methods 551

∂v ∂-
v ∂x ∂-
v ∂x
J
. K =− + , (8.3.2)
∂y ∂ξ ∂η ∂η ∂ξ

where .-v(ξ, η) = v(x, y), with

( .(x, y) = )FK (ξ, η). The relations (8.3.1) and (8.3.2)
lead to the expression . JK μ ∂v
∂x
h
+ ν ∂v
∂y
h
is a polynomial of . Q 1 for the variables .ξ
and .η. We deduce that the hypothesis .(H 5) is satisfied, i.e.,
( ) { ( )
∂vh ∂vh ∂vh ∂vh
.area(K ) μ +ν (G K ) = μ +ν d xd y,
∂x ∂y K ∂x ∂y

for all .vh ∈ PK, . On the other hand, along any face of . K , any function .vh of . PK, is
a polynomial of . P1 and the integral of .vh along this face depends only of the value
.vh in the middle of this face, and the hypothesis .(H 6) is satisfied. So, we have the
following.
Lemma 8.3.1 Scheme 1 satisfies the assumptions .(H 4), .(H 5), and .(H 6); the
conservation of neutrons is, therefore, verified in a discrete way. ♦
Scheme 1 can, also, be written, on each quadrilateral . K of vertices . Ai = (xi , yi ),
1 ≤ i ≤ 4, with .u h (Ai ) = u i , .5 ≤ i ≤ 9
.

(u 8 − u 6 )(μ(y1 + y2 − y3 − y4 ) − ν(x1 + x2 − x3 − x4 )) +
.

. + (u 5 − u 7 )(−μ(y1 − y2 − y3 + y4 ) + ν(x1 − x2 − x3 + x4 )) +

. + ((y1 − y3 )(x2 − x4 ) + (y2 − y4 )(x1 − x3 ))(σ (A9 )u 9 − f (A9 )) = 0,

u + u 8 = u 5 + u 7 = 2u 9 .
. 6

In particular, when the quadrilateral . K is a rectangle whose sides . A1 A4 and . A2 A3

(resp. . A2 A1 and . A3 A4 ) are parallel to the axis of . y (resp. of .x) and are of length
equal to ./\y (resp. ./\x), it comes

u8 − u6 u5 − u7
μ
. +ν + σ (A9 )u 9 = f (A9 ),
/\x /\y

. 6u + u 8 = u 5 + u 7 = 2u 9

which is nothing but the D.S.N. [23].

Lemma 8.3.2 (Lesaint [1]) We assume that the hypothesis .(H 8) is satisfied and
that .h is small enough. The vertices . Ai = (xi , yi ), .1 ≤ i ≤ 4 of any element . K of .τh
can always be numbered so that the lit sides are . A2 A3 and . A3 A4 (Fig. 8.2). For all
,
.vh ∈ X h , we can, then, write, by setting .vi = vh (Ai ), .5 ≤ i ≤ 8

@seismicisolation
@seismicisolation
552 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

(( ) )
∂vh ∂vh 1 − 3C0 h K
.2 area(K ) μ +ν vh (G K ) ≥ (μ(y1 − y4 ) − ν(x1 − x4 ))v82 +
∂x ∂y 1 − C0 h K

1 − 3C0 h K
. + (μ(y2 − y1 ) − ν(x2 − x1 ))v52 +
1 − C0 h K

1 + C0 h K
. + (μ(y3 − y2 ) − ν(x3 − x2 ))v62 +
1 − C0 h K

1 + C0 h K
. + (μ(y4 − y3 ) − ν(x4 − x3 ))v72 , (8.3.3)
1 − C0 h K

where .C0 is the constant involved in the hypothesis (8.2.17). Then, Scheme 1 has a
unique solution .u h such that
1
|u h |h,1 + (|μn x + νn y | 2 u h )h,1 ≤ C| f |h,1 ,
. (8.3.4)

where the constant .C is independent of .h. ♦

Proof Using the notations of the following figure (Fig. 8.3)
We can write for all .vh ∈ X h,

A2 A5
A1

A6 GK A8

A3 A7 A4

Fig. 8.3 Quadrilateral

@seismicisolation
@seismicisolation
8.3 Scheme of Continuous Methods 553

(( ) ) E
4
∂vh ∂vh 1
.area(K ) μ +ν vh (G K ) = (μ(yi+1 − yi ) − ν(xi+1 − xi ))vi+4 +
2
∂x ∂y 2
i=1

1
. (μ(y1 − y2 + y3 − y4 ) − ν(x1 − x2 + x3 − x4 ))((v5 − v7 )2 − (v8 − v6 )2 ),
4
(8.3.5)
where we noted .x5 = x1 , . y5 = y1 . When the expression .μ(y1 − y2 + y3 − y4 ) −
ν(x1 − x2 + x3 − x4 ) is positive (resp. negative), we increase the term .(v8 − v6 )2 by
.2v8 + 2v6 (resp. .(v5 − v7 ) by .2v5 + 2v7 ). The inequality (8.3.3) is deduced from the
2 2 2 2 2

relation (8.3.5) using the hypothesis .(H 8), inequality (8.2.17) (resp. (8.2.18)). When
we have .σ (x, y) ≥ α > 0 (and, we can always reduce to this situation according to
Remark 8.1.6), we can write
{
. σ vh2 d xd y ≥ Ch 2K (v(G K ))2 . (8.3.6)
K

By combining the inequalities (8.3.5) and (8.3.6) for all elements . K of .τh , we obtain,
replacing .vh by .u h
1
.|u h |2h,1 + (|μn x + νn y | 2 u h )2h,1 ≤ C|( f, u h )h,1 |,

from which, we deduce the existence of the solution .u h and the increase
(8.3.4). Q.E.D.
Remark 8.3.3 The assumption .(H 7) is not, in general, sufficient to guarantee the
stability of Scheme 1, as shown by the example described in Fig. 8.3. The character-
istic direction is parallel to the side . A2 A3 and the quadrilateral is very adjacent to a
rectangle. We have the equality .μ(y2 − y3 ) − ν(x2 − x3 ) = 0. It is deduced that

(( ) ) ( )
∂vh ∂vh 1 1
.area(K ) μ +ν vh (G K ) ≥ (μ(y1 − y4 ) − ν(x1 − x4 )) v82 − (v8 − v6 )2 +
∂x ∂y 2 2

1 1
. (μ(y2 − y1 ) − ν(x2 − x1 ))v52 − (μ(y3 − y4 ) − ν(x3 − x4 ))v72 .
2 2
The term .(v8 − v6 )2 then prevents us from asserting that Scheme 1 is stable. ♦
Let .(τh )h be a regular family of triangulations. We assume that the hypothesis .(H 8)
is satisfied and that .h is small enough. Let .u h ∈ X h, be the solution of Scheme 1. For
all .vh ∈ X h, , we define the expression . E h, by

. E h, = (A(u h − vh ), u h − vh )h,1 .

Using the Lemma 8.3.2, we can write

E h, ≥ C|u h − vh |2h,1 + (|B| 2 (u h − vh ))2h,1 .

1
.

@seismicisolation
@seismicisolation
554 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

On the other hand, we have by the definition of Scheme 1

. E h, = (A(u − vh , u h − vh )h,1 .

By using the Cauchy Schwarz inequality, then the triangular inequality, we deduce
from the last two relations the inequality
1 1
|u − u h |h,1 + (|B| 2 (u − u h ))h,1 ≤ C inf , (|A(u − vh )|h,1 + (|B| 2 (u − vh ))h,1 ),
.
vh ∈X h

where .C is a constant .> 0 independent of .h. So, we have

Theorem 8.3.4 Let .(τh )h be a regular family of triangulations. We assume that the
hypothesis .(H 8) is satisfied and that .h is small enough. Let .u h ∈ X h, be the solution
of Scheme 1. Then, if the solution .u of the problem (8.1.7)–(8.1.8) belongs to the
space .C 1 (G, R), we have
1 1
|u − u h |h,1 + (|B| 2 (u − u h ))h,1 ≤ C inf , (|A(u − vh )|h,1 + (|B| 2 (u − vh ))h,1 ),
.
vh ∈X h

where .C is a constant .> 0 independent of .h. ♦

Moreover, reasoning as in Lemma 4.1.14, we have

|u − rh, u|h,K ,1 ≤ Ch 2K |u|2,2,K .

.

By summing over all the elements . K of .τh , we obtain

.|u − rh, u|h,1 ≤ Ch 2 |u|2,2,G . (8.3.7)

Let . S be any side of an element . K ∈ τh , with ends . Ai and . A j and midpoint . Ai j .

According to the Definition 8.2.3, we have
| |
−−→ | 1 |
.(u − rh, u)h,S,1 = | Ai A j | ||u(Ai j ) − (u(Ai ) + u(A j ))|| .
2

We deduce, by reasoning as in Lemma 4.1.14

. (u − rh, u)h,S,1 ≤ Ch 2K |u|2,2,S .

If the solution .u of the problem (8.1.7)–(8.1.8) belongs to the space . H 3 (G), then
by summing over all . S faces included in .∂G and, using Lemma 4.1.12, we have

@seismicisolation
@seismicisolation
8.3 Scheme of Continuous Methods 555

⎛ ⎞ 21
E E
(u − rh, u)h,1 ≤ Ch 2 ⎝
. |u|22,2,S ⎠
n
K ∈τh S⊂∂G ∂K

= Ch ||u||3,2,G .
2
(8.3.8)

By combining the inequalities (8.3.7) and (8.3.8) with the Lemma 8.2.6 and the
Theorem 8.3.4, we obtain the upper bound
1
|u − u h |h,1 + (|B| 2 (u − u h ))h,1 ≤ C(Z |u|2,2,G + h 2 |u|3,2,G ),
.

where . Z = max{Z K , K ∈ τh }. Moreover, if we assume the hypothesis .(H 7), we

deduce the inequality
1
. |u − u h |h,1 + (|B| 2 (u − u h ))h,1 ≤ Ch 2 ||u||3,G ,

where .C is a constant .> 0 independent of .h. So, we have the following error bounds:
Theorem 8.3.5 Let .(τh )h be a regular family of triangulations. We assume that the
hypotheses .(H 7) and .(H 8) are satisfied and that .h is small enough. Let .u h ∈ X h,
be the solution of Scheme 1. Then, if the solution .u of the problem (8.1.7)–(8.1.8)
belongs to the space . H 3 (G), we have
1
. |u − u h |h,1 + (|B| 2 (u − u h ))h,1 ≤ Ch 2 ||u||3,G ,

where .C is a constant .> 0 independent of .h. ♦

8.3.2 Scheme 2

Find .u h ∈ X hk such that, for all . K ∈ τh

{
. (Au h − f )vh d xd y = 0, for all vh ∈ PKk−1 . (8.3.9)
K

Lemma 8.3.6 Scheme 2 satisfies the assumptions .(H 4), .(H 5), and .(H 6); the
conservation of neutrons is, therefore, verified in a discrete way. ♦

Proof The hypotheses .(H 4), and .(H 5) are obviously satisfied. On the other hand,
the restriction of a function .vh of . PKk to any face of the element . K is a polynomial
of degree .≤ k entirely determined by its values at the nodes located on this face, and
the hypothesis .(H 6) is thus satisfied. Q.E.D.

We will consider Scheme 2, in the case, where the domain .G is the square .[0, 1]2 ,
triangulated into rectangles . K lm with sides parallel to the axes, with vertices . Alm ,
. Al+1,m , . Al,m+1 , . Al+1,m+1 , where . Alm is the coordinate point . xl = l/\x, . ym = m/\y,

@seismicisolation
@seismicisolation
556 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

0 ≤ l ≤ L, .0 ≤ m ≤ M, with . L/\x = M/\y = 1. Let . K be any element of .τh . To

.

any function .v of . L 2 (K , R), we associate its projection .π Kk v defined by

{
. (v − π Kk v)w d xdy = 0, ∀w ∈ Q k . (8.3.10)
K

To any function .v of . L 2 ((xl , xl+1 ), R), we associate its projection .πxk v defined by
{ xl+1
. (πxk v − v)w d x = 0, ∀w ∈ Pk .
xl

We define .π yk v in the same way, starting from .v ∈ L 2 ((ym , ym+1 ), R). Now, let
.v(x, y) be a continuous function on . K lm , we can define the projections .(π x v)(y) and
k

.(π y v)(x) to . y and . x fixed, respectively, and, we have

k

πxk π yk v = π yk πxk v
.

= π Kk lm v.

Finally, to any function .v of . L 2 (G, R), (resp. of . L 2 ((0, 1), R)), we associate its pro-
jection .πk v (resp. .π k v), function whose restriction to each element . K (resp. interval
.]xl , xl+1 [) is equal to .π K v (resp. .π x v). In the equality (8.3.9), we can replace .vh by
k k

.πk−1 u h , and, we obtain

{
. (Au h − f )πk−1 u h d xd y = 0, for all K ∈ τh .
K

For any function .vh of . Q k , . ∂v

∂x
h
is a polynomial of degree .≤ k − 1 for the variable .x
and of degree .≤ k for the variable . y. So,
{ {
∂vh ∂vh k−1 k−1
. (πk−1 vh ) d xd y = (π π y vh ) d xd y
∂x ∂x x
K
{K
∂vh k−1
= (π y vh ) d xd y.
K ∂x

Hence, { { ( )
∂vh k−1 ∂vh
. (πk−1 vh ) d xd y = πy (π yk−1 vh ) d xd y.
K ∂x K ∂x

On the other hand, we can easily show that for any function .v ∈ C 1 (K , R), we have

∂ ∂v
. (π k−1 v) = π yk−1 .
∂x y ∂x

@seismicisolation
@seismicisolation
8.3 Scheme of Continuous Methods 557

From the last two equalities, we deduce, for all .vh ∈ Q k

{ {
∂vh ∂
. (πk−1 vh ) d xd y = (π k−1 vh )π yk−1 vh d xd y.
K ∂x K ∂x y

So, we have
{ {
∂
. (Au h )(πk−1 u h ) = μ (π k−1 u h )(π yk−1 u h ) d xd y +
K K ∂x y
{ {
∂
ν (πxk−1 u h )(πxk−1 u h ) d xd y + σ (πk−1 u h )2 d xd y
K ∂y K
{
= f (πk−1 u h ) d xd y. (8.3.11)
K

By summing the equalities such as (8.3.11) for all the rectangles of the domain .G
and, by using the Schwarz inequality, we obtain the inequality
1
||πk−1 u h ||0,G + |||μn x + νn y | 2 (π k−1 u h )||0,∂G ≤ C|| f ||0,G ,
.

where the constant .C does not depend on .h. So, we have the following:
Lemma 8.3.7 We assume that the domain .G is the square .(]0, 1[)2 , triangulated as
above. Then, Scheme 2 admits a unique solution .u h such that
1
||πk−1 u h ||0,G + |||μn x + νn y | 2 (π k−1 u h )||0,∂G ≤ C|| f ||0,G ,
.

where the constant .C does not depend on .h. ♦

For all .vh ∈ X hk , we define the expression . E h by

. E h = (A(u h − vh ), πk−1 (u h − vh ))0,G . (8.3.12)

According to Lemma 8.3.7, we have

1
. E h ≥ C||πk−1 (u h − vh )||20,G + |||B| 2 π k−1 (u h − vh )||20,∂G . (8.3.13)

On the other hand, the definition of Scheme 2 implies that

. E h = (A(u − vh ), πk−1 (u h − vh ))0,G . (8.3.14)

Let . K be any rectangle of .τh . To any function . p ∈ PKk , we associate its projection
.π K p ∈ PKk−1 . According to the definition of the projection on the space . PKk−1 ,
k−1

we have
{ {
. pq d xdy = (π Kk−1 p)q d xdy, for all q ∈ PKk−1 . (8.3.15)
K K

@seismicisolation
@seismicisolation
558 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

Lemma 8.3.8 Let . K be any rectangle of .τh . To any function . p ∈ PKk , we associate
its projection .π Kk−1 p ∈ PKk−1 . Then,

. p = π Kk−1 p at points giKj , 1 ≤ i, j ≤ k, (8.3.16)

where the points .giKj are defined at (8.2.11). ♦

Proof The functions . pq and .(π Kk−1 p)q are, respectively, polynomials of . Q 2k−1 and
. Q 2k−2 . So, by using (8.3.15), we have

{
( p, q)h,K ,k =
. pq d xdy
{ K

= (π Kk−1 p)q d xdy

K
= (π Kk−1 p, q)h,K ,k .

The points .giKj , .1 ≤ i, j ≤ k, defined in (8.2.11) form a set . Q k−1 -unisolvent. We

deduce the equality (8.3.16). Q.E.D.

We have the following.

Theorem 8.3.9 (Lesaint [1]) We place ourselves within the framework of the
assumptions of the Lemma 8.3.7. Let .u h ∈ X hk be the solution of Scheme 2. Then, if
the solution .u of the problem (8.1.7)–(8.1.8) belongs to the space . H 1 (G), we have
1
|u − u h |h,k + (|B| 2 (u − u h ))h,k ≤
.

( )
|(A(u − vh ), w)0,G | 1
.C inf sup + |u − vh |h,k + (|B| (u − vh ))h,k ,
2
vh ∈X hk w∈Y k−1 ||w||0,G
h
(8.3.17)
where .C is a constant .> 0 independent of .h. ♦

Proof Consider . E h defined in (8.3.12). From the two relations (8.3.13) and (8.3.14),
we deduce the inequality

1 |(A(u − vh ), w)0,G |
||πk−1 (u h − vh )||0,G + |||B| 2 π k−1 (u h − vh )||0,∂G ≤ C sup
. .
w∈Yhk−1 ||w||0,G

Using the Lemma 8.3.8, it comes

.||πk−1 (u h − vh )||0,G = |πk−1 (u h − vh )|h,k = |u h − vh |h,k .

@seismicisolation
@seismicisolation
8.3 Scheme of Continuous Methods 559

Similarly, one can show that

.||π k−1 (u h − vh )||0,∂G = (π k−1 (u h − vh ))h,k

= (u h − vh )h,k .

By combining the last three relations with the triangle inequality, we obtain the upper
bound (8.3.17). Q.E.D.

Theorem 8.3.10 Let .(τh )h be a regular family of triangulations from rectangle .G

to rectangles . K . Let .v ∈ X hk be the solution of Scheme 2. Then, if the solution .u of
the problem (8.1.7)–(8.1.8) belongs to the space . H k+2 (G), we have
1
. |u − u h |h,k + (|B| 2 (u − u h ))h,k ≤ Ch k+1 ||u||k+2,2,G , (8.3.18)

where .C is a constant .> 0 independent of .h. ♦

Proof According to Lemma 4.1.14, we have

||u − rh u||0,K ≤ Ch k+1

. K |u|k+1,2,K .

On the other hand, one can show that

|u − rh u|h,K ,k ≤ Ch k+1
. K |u|k+1,2,K ,

and
. (u − rh u)h,S,k ≤ Ch k+1
K |u|k+1,2,K .

Summing over all . K elements of .τh and, using the Lemma 4.1.12, we get

. ||u − rh u||0,G ≤ Ch k+1 |u|k+1,2,G , (8.3.19)

|u − rh u|h,k ≤ Ch k+1 |u|k+1,2,G ,

. (8.3.20)

. (u − rh u)h,k ≤ Ch k+1 ||u||k+2,2,G . (8.3.21)

By combining the last three inequalities, (8.3.19)–(8.3.20)-(8.3.21), with the Lemma

8.2.7 and the Theorem 8.3.9, we obtain the upper bound (8.3.18). Q.E.D.

8.3.3 Scheme 3 (Generalized S.N.G. Scheme)

Find .u h ∈ X hk such that, for all . K ∈ τh

E
k
. wiKj ((Au h − f )vh )(giKj ) = 0, for all vh ∈ PKk−1 , (8.3.22)
i, j=1

@seismicisolation
@seismicisolation
560 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

where equivalently
(Au h − f )(giKj ) = 0, 1 ≤ i, j ≤ k.
. (8.3.23)

We multiply the equality (8.3.23) by .u h (giKj )wiKj and we sum for all the indices .i and
. j, .1 ≤ i, j ≤ k, we comes

E
k
. wiKj (Au h − f )u h (giKj ) = 0, (8.3.24)
i, j=1

for all . K ∈ τh .
Remark 8.3.11 (i) The equation (8.3.22) is deduced from the equation (8.3.9)
using the quadrature formula (8.2.11).
(ii) Scheme 3 can be written explicitly for .k = 1 using the formulas (8.3.1) and
(8.3.2).
(iii) Scheme 3, for .k = 1, can be written, with .u h (Ai ) = u i , .1 ≤ i ≤ 4

.(u 1 − u 3 )(μ(y2 − y4 ) − ν(x2 − x4 )) + (u 2 − u 4 )(−μ(y1 − y3 ) + ν(x1 − x3 )) +

( )
u1 + u2 + u3 + u4
. + ((y1 − y3 )(x4 − x2 ) + (y2 − y4 )(x1 − x3 )) σ (A9 ) − f (A9 ) = 0.
4

Lemma 8.3.12 Scheme 3 satisfies the assumptions .(H 4), .(H 5), and .(H 6); the
conservation of neutrons is, therefore, verified in a discrete way. ♦
( )
Proof When.vh ∈ PKk , the expression. JK μ ∂v ∂x
h
+ ν ∂v
∂y
h
is a polynomial of. Q k in the
variables .ξ and .η. The quadrature formula (8.2.10) being exact for the polynomials
of . Q 2k−1 , we deduce that the hypothesis .(H 5) is satisfied, i.e.,

E
k ( ) { ( )
∂vh ∂vh ∂vh ∂vh
. wiKj μ +ν (bi j ) =
K
μ +ν d xd y,
i, j=1
∂x ∂y K ∂x ∂y

for all . K ∈ PKk . The hypothesis .(H 6) is verified as in the case of Scheme 2. Q.E.D.

We, then, consider Scheme 3 in the case, where the domain .G is triangulated into
arbitrary convex quadrilaterals satisfying the hypothesis (8.2.6).

@seismicisolation
@seismicisolation
8.3 Scheme of Continuous Methods 561

Lemma 8.3.13 It is assumed that the hypotheses.(H 7) and.(H 8) are satisfied. Then,
Scheme 3 has a unique solution .u h such that
1
|u h |h,k + (|μn x + νn y | 2 u h )h,k ≤ C| f |h,k ,
. (8.3.25)

where the constant .C does not depend on .h. ♦

Proof Let . FK be the transformation defined in Lemma 4.9.1. We define

-
u (ξ, η) = u h (x, y),
. h

with .(x, y) = FK (ξ, η). Using the relations (8.3.1) and (8.3.2), we can write

E
k E k ( ( ))
∂u h K ∂-
uh ∂ y ∂-
uh ∂ y
. X= wiKj u h (gi j ) = - wj -
wi - uh − (gi , g j )
i, j=1
∂x i, j=1
∂ξ ∂η ∂η ∂ξ

and, we have

∂y 1 ξ
. = (y1 + y2 − y3 − y4 ) + (y1 − y2 + y3 − y4 ).
∂η 4 4

The polynomial .-u h ∂-uh

∂ξ
has degree .≤ 2k − 1 for the variable .ξ . The .gi , .1 ≤ i ≤ k,
being the abscissas of Gauss-Legendre, and the .- wi , .1 ≤ i ≤ k, being the associated
weights, it comes

E
k ( ) E
k { 1( )
∂-uh ∂ y 1 ∂-uh
. -
wi -
wj -uh (gi , g j ) = (y1 + y2 − y3 − y4 ) -
wj -uh (ξ, g j ) dξ +
∂ξ ∂η 4 −1 ∂ξ
i, j=1 j=1

E k ( )
1 ∂-
uh
. + (y1 − y2 + y3 − y4 ) -
wi -
wj -
uh ξ (gi , g j ).
4 i, j=1
∂ξ
On the other hand, we have
E k { ( )
1 1
∂-
uh
. (y1 + y2 − y3 − y4 ) -
wj -
uh (ξ, g j ) dξ =
4 j=1 −1 ∂ξ

1 E k
1 E k
. = (y1 − y4 ) - u 2h (1, g j ) + (y3 − y2 )
w j- -
w j-
u 2h (−1, g j ) −
4 j=1
4 j=1

1 E k
. − (y1 − y2 + y3 − y4 ) -
w j (-
u 2h (1, g j ) + -
u 2h (−1, g j )).
8 j=1

@seismicisolation
@seismicisolation
562 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

By combining the last two equalities, it comes

E
k ( ) E
k E
k
∂-uh ∂ y 1 1
. -
wi -
wj -uh (gi , g j ) = (y1 − y4 ) -
w j-
u 2h (1, g j ) + (y3 − y2 ) -
w j-
u 2h (−1, g j ) +
∂ξ ∂η 4 4
i, j=1 j=1 j=1

⎛ ⎞
E
k ( ) E
k
1 ⎝ ∂-uh
. + (y1 − y2 + y3 − y4 ) 2 -
wi -
wj -uh ξ (gi , g j ) − -
w j (- u h (−1, g j ))⎠ .
u h (1, g j ) + -
2
8 ∂ξ
i, j=1 j=1

E ( )
∂y
If we, similarly, develop the expression . i,k j=1 - wj -
wi - u h ∂-uh
∂η ∂ξ
(gi , g j ), after the
E ( )
expression . i,k j=1 wiKj u h ∂u
∂y
h
(giKj ), we obtain

E ( )
1 E μ(yi+1 − yi ) − ν(xi+1 − xi )
k 4
∂u h ∂u h
. wiKj μ +ν u h (giKj ) = −−−−→ (u h )2h,Ai Ai+1 ,k +
∂x ∂y 2 | Ai Ai+1 |
i, j=1 i=1

⎛ ⎞
E
k ( )
∂-
uh ∂-
uh
. + δ ⎝2 -
wi -
w j-
uh ξ− η (gi , g j )⎠ +
i, j=1
∂ξ ∂η

⎛ ⎞
E
k E
k
. + δ ⎝− -
w j (- u 2h (−1, g j ))⎠ +
u 2h (1, g j ) + - -
wi (-
u 2h (gi , 1) + -
u 2h (gi , −1)),
j=1 i=1

where .δ = 18 (μ(y1 − y2 + y3 − y4 ) − ν(x1 − x2 + x3 − x4 )). The last relation and

the hypotheses .(H 7) and .(H 8) lead to the existence of constants .ci , .1 ≤ i ≤ 4,
independent of the quadrilateral . K , such that

E
k
1 E μ(yi+1 − yi ) − ν(xi+1 − xi )
4
. wiKj (Au h )u h (giKj ) ≥ −−−−→ (1 + ci h K )(u h )2h,Ai Ai+1 ,k + c|u h |2h,K ,
2 | Ai Ai+1 |
i, j=1 i=1
(8.3.26)
where .c is a constant .> 0 independent of .h and, where we have

(μ(yi+1 − yi ) − ν(xi+1 − xi ))ci < 0, 1 ≤ i ≤ 4.

.

From the previous inequality, we deduce the existence of a unique solution .u h for
Scheme 3. By combining the relations (8.3.24) and (8.3.26) for all the quadrilaterals
. K ∈ τh , we get the inequality (8.3.25). Q.E.D.

Reasoning as in Theorem 8.3.4 and, using Lemma 8.3.13 and the definition of
Scheme 3, we obtain
Theorem 8.3.14 (Lesaint [1]) Let .(τh )h be a regular family of triangulations of
G. We assume that the hypotheses .(H 7) and .(H 8) are satisfied, and that .h is small
.

enough. Let.u h ∈ X hk be the solution of Scheme 3, then, if the solution.u of the problem
(8.1.7)–(8.1.8) belongs to the space .C 1 (G, R), we have

@seismicisolation
@seismicisolation
8.4 Discontinuous Finite Element Methods 563

1
( 1
)
|u − u h |h,k + (|B| 2 (u − u h ))h,k ≤ C inf
. |A(u − vh )|h,k + (|B| 2 (u − vh ))h,k ,
vh ∈X hk

where .C is a constant .> 0 independent of .h. ♦

Theorem 8.3.15 (Lesaint [1]) Let .(τh )h be a regular family of triangulations in

arbitrary quadrilaterals . K . We assume that the hypotheses .(H 7) and .(H 8) are
satisfied and that .h is small enough. Let .u h ∈ X hk be the solution of Scheme 3. Then,
if the solution .u of the problem (8.1.7)–(8.1.8) belongs to the space . H k+2 (G), we
have
1
.|u − u h |h,k + (|B| 2 (u − u h ))h,k ≤ Ch ||u||k+2,2,G ,
k+1
(8.3.27)

where .C is a constant .> 0 independent of .h. ♦

Proof By using the Theorem 8.3.14, the Lemma 8.2.8 and the inequalities (8.3.20)
and (8.3.21), we obtain
1
|u − u h |h,k + (|B| 2 (u − u h ))h,k ≤ C(Z h k−1 |u|k+1,2,G + h k+1 |u|k+2,2,G ),
.

where . Z = max{Z K , K ∈ τh }. The previous inequality and the hypothesis .(H 7)

lead to the upper bound (8.3.27). Q.E.D.

8.4 Discontinuous Finite Element Methods

We assume the open set .G to be polyhedral. Let .τh be a triangulation of .G in finite

elements . K and let . PK be a finite-dimensional space defined on . K , for . K ∈ τh .
| |
Definition 8.4.1 Discontinuous methods. We are looking for .u h ∈ Wh = PK
K ∈τh
such that, for all .vh ∈ Wh
{ {
. (Au h − f )vh d xd y − (μn x + νn y )(u h − th )vh ds = 0, (8.4.1)
K ∂− K

where { n
0 on ∂− K ∂− G n
t =
. h (8.4.2)
outer trace of u h on ∂− K \(∂− K ∂− G).

Remember that if . S designates a face of an element . K not included in .∂G, the outer
trace on . S of a function .vh ∈ Wh is the value of .vh along . S, outside . K . ♦

@seismicisolation
@seismicisolation
564 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

8.4.1 Existence and Uniqueness Solution

Theorem 8.4.2 ([20]) We assume that . f ∈ L 2 (G, R). Then, the problem (8.4.1)–
(8.4.2) has a unique solution .u h ∈ Wh . ♦

Proof The finite element method (8.4.1)–(8.4.2) is equivalent to a linear system

whose number of equations is equal to the dimension of .Wh . It is therefore sufficient
to show the uniqueness of the solution .u h . We, therefore, assume that . f = 0 and, we
will show that we, necessarily, have .u h = 0. Let . K 1 , . K 2 , . . . , K N be a numbering
of the . K ∈ τh satisfying the conditions of Lemma 8.1.3. If .u h = 0 on
U elements
U U
.K1 K 2 · · · K i−1 , then .th = 0 on .∂− K i and the equation (8.4.1) becomes for
the element . K i
{ {
.− (Bu h )vh ds + (Au h )vh d xd y = 0, for all vh ∈ PK i .
∂− K i Ki

Choosing .vh = u h and applying Green’s formula

{ ( ) {
∂u h ∂u h 1
. μ +ν u h d xd y = (μn x + νn y )u 2h ds,
Ki ∂x ∂y 2 ∂ Ki

we obtain { { {
. Bu 2h ds − Bu 2h ds + σ u 2h d xd y = 0.
∂+ K i ∂− K i Ki

According to the definition of .∂+ K i (resp. .∂− K i ), we have the inequality . B > 0 on
∂ K i (resp. . B < 0 on .∂− K i ). On the other hand, we have .σ ≥ α > 0 (Remark 8.1.6).
. +
We deduce that .u h is zero on . K i . By reasoning by induction, we show that .u h is zero
on .G. Q.E.D.

8.4.2 Neutron Conservation Property

We assume that the inclusion . P0 ⊂ PK is satisfied for all . K ∈ τh . Let .u h ∈ Wh be

the solution of the problem (8.4.1). We can replace .vh by .1 in the equality (8.4.1),
it comes
{ { {
. Au h d xd y − (μn x + νn y )(u h − th ) ds = f d xdy.
K ∂− K K

It is deduced that
{ { { {
. Bu h ds + Bth ds + σ u h d xd y = f d xdy. (8.4.3)
∂+ K ∂− K K K

@seismicisolation
@seismicisolation
8.4 Discontinuous Finite Element Methods 565

This relation expresses the conservation of neutrons element by element. By summing

relations such as (8.4.3) for all . K ∈ τh elements and, using equality

u = 0 on ∂− G,
. h

we get the equality

{ { {
. (μn x + νn y )u h ds + σ u h d xd y = f d xdy. (8.4.4)
∂+ G G G

So, we have the following.

Lemma 8.4.3 We assume that the inclusion . P0 ⊂ PK is satisfied for all . K ∈ τh .

Then, the neutron conservation property is satisfied in the following sense: let .u h ∈
Wh be the solution of the problem (8.4.1), then we have the equality (8.4.4). ♦

8.4.3 General Bound of the Error

Let .vh ∈ PK and .q ∈ L 2 (∂− K , R). We pose

w = u h − vh ∈ PK ,
. ξ = th − q. (8.4.5)

We consider the expression

{ {
Y := −
. h B(w − ξ )w ds + (Aw)w d xdy.
∂− K K

Using Green’s formula, it comes

{ { {
1
.Y h = Bw ds − B(w − ξ )w ds + σ w2 d xd y.
2
2 ∂K ∂− K K

Because
1 2
. (w − ξ )w = (w − ξ 2 + (w − ξ )2 ).
2
We obtain
{ { { {
1 1 1
.Y h = Bw2 ds + Bξ 2 ds − B(w − ξ )2 ds + σ w2 d xd y.
2 ∂+ K 2 ∂− K 2 ∂− K K
(8.4.6)
We, now, show a general bound of the error .u − u h .

@seismicisolation
@seismicisolation
566 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

Lemma 8.4.4 ([20]) For any . K ∈ τh , for any .vh ∈ PK and for any function .q ∈
L 2 (∂− K , R), we have the relation
{ {
1 1
. B(u h − vh ) ds +
2
B(th − q)2 ds−
2 ∂+ K 2 ∂− K
{ { {
1
. − B((u h − vh ) − (th − t))2 ds + σ (u h − v)2 d xd y = B(u − v)(u h − vh ) ds +
2 ∂− K K ∂+ K

{ {
. B(u − q)(u h − vh ) ds + (u − vh )A∗ (u h − vh ) d xd y,
∂− K K
(8.4.7)
where . A∗ is the formal adjoint of operator . A, i.e.,
∂ ∂
. A∗ = −μ −ν + σ. ♦
∂x ∂y

Proof Let .vh ∈ PK and .q ∈ L 2 (∂− K , R) and let .w be defined in (8.4.5). Using the
equality (8.4.1), we have
{ {
Y =
. h B(vh − q)w ds + A(u − vh )w d xdy.
∂− K K

We have the following Green formula

{ { {
. A(u − vh )w d xdy = (u − vh )A∗ w d xdy + B(u − v)w ds.
K K ∂K

It is deduced that
{ { {
.Y h = B(u − vh )w ds + B(u − q)w ds + (u − vh )A∗ w d xdy.
∂+ K ∂− K K
(8.4.8)
By combining the relations (8.4.5), (8.4.6) and (8.4.8), we obtain the increase (8.4.7).
Q.E.D.

To obtain more explicit error bounds, it is necessary to define the finite-dimensional

spaces. PK . More precisely, let. K be any convex quadrilateral of.τh and let. FK ∈ (Q 1 )2
be the transformation (defined in the Lemma 4.9.1) which sends the reference square
.K- = [−1, 1] × [−1, 1] on quadrilateral . K . On each quadrilateral . K , we choose as
in (8.2.8),
. PK = PK = { p = -
k
p ◦ FK−1 , ∀-
p ∈ Q k }. (8.4.9)

We first consider the case of arbitrary quadrilaterals, in the general case. On the
-, we define the operator .| |
reference element . K -, R), Q k ) which to any
- ∈ L(L 2 ( K
2 - --
v ∈ L ( K , R) associates .| |
.- v ∈ Q k by

@seismicisolation
@seismicisolation
8.4 Discontinuous Finite Element Methods 567
{ {
. --
(| | w dξ dη =
v)- -w dξ dη, for all -
v- w ∈ Qk .
-
K -
K

For all . K ∈ τh , we define .| |kK ∈ L(L 2 (K , R), PKk ) by

-
| | - v for all v ∈ L 2 (K , R),
K v = | |-
k
.

where .- v = v ◦ FK (we can notice that when the element . K is a rectangle, we find
exactly the projection operator defined in (8.3.10)). Then, for all .v ∈ L 2 (G, R), we
define .πh v as the unique function of .Wh whose restriction to each . K ∈ τh is equal to
.| | K v. Let . K 1 , . K 2 , . . . , K I be a given numbering of the elements . K of .τh satisfying
k

the conditions of Lemma 8.1.3. For all .i = 1, . . . , I , we set

| | | |
.Gi = K 1 ··· Ki

and, we define .∂− Gi and .∂+ Gi as in (8.1.9) and (8.1.12), respectively.

Theorem 8.4.5 ([20]) Let .(τh )h be a regular family of triangulations from .G into
quadrilateral elements. K . We assume that the solution.u of the problem (8.1.7)-(8.1.8)
belongs to . H k+2 (G). Then, we have for all .i = 1, . . . , I ,

.||u − u h ||0,Gi ≤ Ch k ||u||k+1,2,Gi , (8.4.10)

({ ) 21
. B(u − u h ) ds
2
≤ Ch k ||u||k+1,2,Gi , (8.4.11)
∂+ Gi

and ⎛ ⎞ 21
i {
E
. ⎝− B(u h − th )2 ds ⎠ ≤ Ch k ||u||k+1,2,Gi , (8.4.12)
j=1 ∂− K j

where .C is a constant .> 0 independent of .h. ♦

Proof For all . K ∈ τh , we set

{ n
0 on ∂− K ∂− G n
q =
. h (8.4.13)
outer trace of πh u on ∂− K \(∂− K ∂− G).

@seismicisolation
@seismicisolation
568 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

In the equality (8.4.7), we replace .vh by .πh u, .q by .qh and we estimate the terms
of the corresponding right-hand side. Reasoning as in [24], we have the following
result, analogous to those of Lemma 4.1.14

.||u − πh u||0,2,K ≤ Ch k+1

K ||u||k+1,2,K ,

and
k+ 1
. ||u − πh u||0,2,S ≤ Ch K 2 ||u||k+1,2,K , (8.4.14)

where . S is any side of . K . According to Lemma 4.1.13, we have

||u h − πh u||1,2,K ≤ Ch −1
. K ||u h − πh u||0,2,K ,

and
−1
. ||u h − πh u||0,2,S ≤ Ch K 2 ||u h − πh u||0,2,K .

Let . S be any face of . K . From these four inequalities, we deduce that

{
. (u − πh u)A∗ (u h − πh u) d xd y ≤ C||u − πh u||0,2,K ||u h − πh u||1,2,K
K
≤ Ch kK ||u||k+1,2,K ||u h − πh u||0,2,K ,
{
. B(u − πh u)(u h − πh u) ds ≤ Ch kK ||u||k+1,2,K ||u h − πh u||0,2,K ,
∂+ K

{
. B(u − qh )(u h − πh u) ds ≤ Ch kD K ||u||k+1,2,D K ||u h − πh u||0,2,K ,
∂− K

where . D K is the union of the elements of .τh that have included in .∂− K and, where

. h D K = max{h K , K ⊂ D K }.

The relation (8.4.7) combined with these last three inequalities becomes
{ {
1 1
. B(u h − πh u)2 ds − B((u h − πh u) − (th − qh ))2 ds + α||u h − πh u||20,2,K ≤
2 ∂+ K 2 ∂− K

{
1
. ≤− B(th − qh )2 ds + Ch k ||u||k+1,2,D K ||u h − πh u||0,2,K .
2 ∂− K

@seismicisolation
@seismicisolation
8.4 Discontinuous Finite Element Methods 569

Summing over all . K ⊂ Gi elements, and, using (8.4.2) and (8.4.13), it comes
{ i {
1 1E
. B(u h − πh u)2 ds − B((u h − πh u) − (th − qh ))2 ds + α||u h − πh u||0,2,Gi ≤
2 ∂+ Gi 2 ∂− K j
j=1

. ≤ Ch k ||u||k+1,2,Gi ||u h − πh u||0,2,Gi . (8.4.15)

From the inequalities (8.4.10) and (8.4.15), we deduce

||u − u h ||0,2,Gi ≤ ||u − πh u||0,2,Gi + ||u h − πh u||0,2,Gi

.

≤ Ch k+1 ||u||k+1,2,Gi ,

i.e., inequality (8.4.15). From the relation (8.4.15), we derive

({ ) 21
. B(u h − πh u)2 ds ≤ Ch k ||u||k+1,2,Gi ,
∂+ Gi

and, according to the inequality (8.4.14), we have

({ ) 21
1
. B(u − πh u) ds 2
≤ Ch k+ 2 ||u||k+1,2,Gi .
∂+ Gi

These last two relations lead to the inequality (8.4.11). Finally, we can write

u − th = (u h − πh u − (th − qh )) + (πh u − u) + (u − qh ).
. h

The inequality (8.4.15) implies that

⎛ ⎞ 21
i {
E
. ⎝− B((u h − πh u) − (th − qh ))2 ds ⎠ ≤ Ch k ||u||k+1,2,Gi .
j=1 ∂− K j

The last two relations and the inequality (8.4.14) lead to the bound (8.4.12). Q.E.D.

Remark 8.4.6 The upper bound (8.4.12) shows that when the solution .u of problem
(8.1.7)–(8.1.8) is fairly regular (i.e., belongs to . H k+1 (G) for an integer .k ≥ 1), the
jumps .u h − th of the approximate solution along the sides of the elements . K tend to
zero when .h tends to zero. ♦

In [20], we show that when the elements . K ∈ τh are rectangles, we have the bound

. ||u − u h ||0,G = o(h k+1 ).

@seismicisolation
@seismicisolation
570 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

We will show that this bound is in fact valid as soon as the hypotheses (8.2.6) (i.e., each
quadrilateral . K ∈ τh has two illuminated faces) and (8.2.16) (i.e., for any element
. K ∈ τh , we have the inequality . Z K ≤ ch K ).
2

• On the interval .[−1, 1], we consider the quadrature formula

.

{ 1 E
k+1
. f (ξ ) dξ ∼ -
wir f (ri ), (8.4.16)
−1 i=1

where the .ri , .1 ≤ i ≤ k + 1, .rk+1 = 1 are the abscissas of Gauss-Radau ([22, 25])
and the .-
wi are the weights such that the quadrature formula (8.4.16) is exact for the
polynomials of . P2k .
• On the reference square . K
. - = [−1, 1] × [−1, 1], we consider the points .-
ri j , with
coordinates (.ri , r j ), .1 ≤ i, j ≤ k + 1.

We will assume that the two lit faces of the quadrilateral . K with vertices . A1 , . A2 ,
. A3 and . A4 are the sides . A2 A3 and . A3 A4 . Given a function .-
v ∈ C 0(K-, R), we define
-v as the unique polynomial of . Q k which interpolates .-
. R- v at points .-
ri j . Then, for all
.v ∈ C (K , R), we define . R K v ∈ PK
0 k

- -v,
R K v = R-
.

v = v ◦ F. Finally, for any .v ∈ C 0 (G, R), we define . Rh v ∈ Wh as the unique

where .-
function of .Wh such that

. Rh v|K = R K v, for all K ∈ τh .

By reasoning as in [24], we can show the following result, analogous to

Lemma 4.1.14.

@seismicisolation
@seismicisolation
8.4 Discontinuous Finite Element Methods 571

Lemma 8.4.7 Let .(τh )h be a regular family of triangulations of .G. Then,

.||v − R K v||0,2,K ≤ Ch k+1

K ||v||k+1,2,K for all v ∈ H
k+1
(K ), (8.4.17)

and
k+ 3
||v − R K v||0,2,S ≤ Ch K 2 ||v||k+1,∞,K for all v ∈ W k+1,∞ (K ),
.

for any face . S included in .∂+ K . ♦

For all .v ∈ C 0 (G, R), we define the semi-norm .||v||h,∞ by
( )
||v||h,∞ = max max{|v(ri,k+1
.
K
)|, |v(rk+1,
K
j )|, 1 ≤ i, j ≤ k + 1} ,
K ∈τh

where .riKj = FK (-
ri j ), .1 ≤ i, j ≤ k + 1, for all . K ∈ τh . For all . K ∈ τh , we set
{ n
0 on ∂− K ∂− G n
. h q =
outer trace of Rh u on ∂− K \(∂− K ∂− G).

We use the relation (8.4.7) with .vh = Rh u, .q = qh . The corresponding right-hand

side can be written
. E K (u, u h − Rh u) = μX K (u, u h − Rh u) + νY K (u, u h − Rh u)+
{
. σ (u − Rh u)(u h − Rh u) d xd y, (8.4.18)
K

where
{ { {
∂w
. X K (u, w) = (u − Rh u)wn x ds + (u − qh )wn x ds − (u − Rh u) d xd y
∂+ K ∂− K K ∂x

and, where .Y K (u, w) is defined analogously. Using the one-to-one correspondence

v −→ -
. v = v ◦ FK , we have

. X K (u, w) = -
X K-,ξ (- w) + -
u, - X K-,η (-
u, -
w),

with
{ 1 { 1
y1 − y4 y3 − y2
.- -u )-
X- K ,ξ (-
u, -
w) = (-
u − R- w(1, η)
2
dη + (-
u −-
q )-
w(−1, η)
2
dη −
−1 −1

{ {
w ∂y
-u ) ∂-
1 1
. (-
u − R- dξ dη,
−1 −1 ∂ξ ∂η

@seismicisolation
@seismicisolation
572 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …
{ 1 { 1
y2 − y1 y4 − y3
.- -u )-
X- K ,η (-
u, -
w) = (-
u − R- w(ξ, 1)
2
dξ + (-
u −-
q )-
w(ξ, −1)
2
dξ +
−1 −1

{ {
w ∂y
-u ) ∂-
1 1
. (-
u − R- dξ dη.
−1 −1 ∂η ∂ξ
We, now, use the following essential result.
Lemma 8.4.8 (Lesaint [1]) Let .(τh )h be a regular triangulation family .G in quadri-
lateral elements. We assume that the hypothesis (8.2.6) is satisfied, and that . PK is
chosen as in (8.4.9), for all . K ∈ τh . We suppose on n
the other hand that the solution .u
of the problem (8.1.7)-(8.1.8) belongs to . H k+2 (G) W k+1,∞ (G). Then, there exists
a constant .C > 0 independent of .h such that for all .w ∈ PKk

.|X K (u, w)| + |Y K (u, w)| ≤ C(Z K h k−1

K ||u||k+1,2,K + h K ||u||k+2,2,K )||w||0,K .
k+1

(8.4.19)
♦
X - (- -
u ,-
w)
Proof The linear map .- u −→ K ,ξh K is continuous from . H k+2 ( K-) into .R, of norm
- u ||k+2,2, K- ||-
less than or equal to .C||- w||0, K- , and it is identically zero for all .-
u ∈ Qk ,
according to the definition of the interpolate . R- -u . For .-
u = ξ , we have
k+1

-
. -u (1, η) = -
u (1, η) − R- u (−1, η) − -
q (−1, η) = 0.

On the other hand, . ∂-

w
has degree .≤ k − 1 with respect to .ξ , .- -u is a polynomial
u − R-
∂ξ
in .ξ of degree .k + 1 which vanishes at the Gauss-Radau abscissas, since

∂y 1 ξ
. = (y1 + y2 − y3 − y4 ) + (y1 − y2 + y3 − y4 ),
∂η 4 4

it comes
{ 1 { 1 { 1 { 1
. (-
u − R-
w ∂y
-u ) ∂- 1
dξ dη = (y1 − y2 + y3 − y4 ) (- -u ) ∂-
u − R-
w
ξ dξ dη.
−1 −1 ∂ξ ∂η 4 −1 −1 ∂ξ

We deduce that for .-

u = ξ k+1 , we have
| | | k+1 |
|- w) || |∂ - u ||
| X K-,ξ (-
u, -
-ZK |
.| |≤C | ∂ξ k+1 | ||-
w||0, K- . (8.4.20)
| hK | hK

For .-
u = ηk+1 , .- -u is a polynomial of degree .k + 1 with respect to .η does not
u − R-
depend on the variable .ξ . So, we have

@seismicisolation
@seismicisolation
8.4 Discontinuous Finite Element Methods 573

[{ 1
1
.-
X K-,ξ (- w) = (y1 − y2 + y3 − y4 )
u, - (- -u (1, η))-
u (1, η) − R- w(1, η)+
4 −1

{ { ]
-u ) ∂-
1 1
w
. (-
u (−1, η) − -
q (−1, η))-
w(−1, η) dη − (-
u − R- ξ dξ dη ,
−1 −1 ∂ξ
gold
{ 1 { 1
1
.-
X K-,ξ (-
u, -
w) = (y1 − y2 + y3 − y4 ) (- -u )-
u − R- w dξ dη.
4 −1 −1

We deduce that for .-

u = ηk+1 , we have
| | || k+1 ||
|- w) || || ∂ - u ||
| X K-,ξ (-
u, -
-ZK || ||
.| |≤C || ∂ηk+1 || - ||-
w||0, K- . (8.4.21)
| hK | hK 0, K

From the inequalities (8.4.20) and (8.4.21), we get

| |
|- w) ||
| X K-,ξ (-
u, -
- Z K [-
.| |≤C u ]k+1,2, K- ||-
w||0, K- ,
| hK | hK
U
for all .-
w ∈ Q k and all .-
u ∈ Qk Pk+1 . By applying the Lemma 3.4.17, it comes
| | (( )( [ 2 ] ) )
|- w) ||
| X K-,ξ (-
u, - ZK ∂ -
u ZK
.| |≤C 1+ u ]k+2,2, K- +
[- + [-
u ]k+1,2, K- ||-
w||0, K- .
| hK | hK ∂ξ ∂η k,2, K- hK
(8.4.22)
According to Lemma 4.1.7, we have

(JK (ξ, η))− 2 h kK |u|k,2,K ,

1
[-
.u ]k,2, K- ≤ C inf (8.4.23)
−1≤ξ,η≤1

and
[ ]
∂ 2-
u
(JK (ξ, η))− 2 (h kK Z K |u|k+1,2,K + h k+2
1
. ≤C inf K |u|k+2,2,K ).
∂ξ ∂η -
k,2, K −1≤ξ,η≤1
(8.4.24)
By combining the relations (8.4.22), (8.4.23), and (8.4.24) with the Lemma 4.1.5,
we obtain the inequality
[ ]
.| - ≤ C (h k−1 k−2 2 k+1
X K-,ξ (-
u, -
w)| K Z K + h K Z K )|u|k+1,2,K + (h K + h K Z K )|u|k+2,2,K ||w||0,K .
k

On the other hand, we, necessarily, have the inequality . Z K ≤ h K . We can deduce

|-
.X K-,ξ (-
u, -
w)| ≤ C(h k−1
K Z K |u|k+1,2,K + h K |u|k+2,2,K )||w||0,K .
k+1
(8.4.25)

@seismicisolation
@seismicisolation
574 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

We can show in the same way that the inequality (8.4.25) is still valid if we replace
|-
.X K-,ξ (- w)| by .| -
u, - X K-,η (-
u, -
w)|. We deduce that the inequality (8.4.25) is valid when
the left-hand side is replaced by .|X K (u, w)|. The term .|Y K (u, w)| increases in the
same way and, we thus have the result. Q.E.D.

By using the relations (8.4.17), (8.4.18) and (8.4.19), we obtain, for all . K ∈ τh

E K (u, u h − Rh u) ≤ C(Z K h k−1

. K ||u||k+1,2,K + h K ||u||k+2,2,K )||u h − Rh u||0,K .
k+1

(8.4.26)
We can, then, show the following theorem.
Theorem 8.4.9 (Lesaint [1]) Let.(τh )h be a regular triangulation family.G in quadri-
lateral elements. We assume that the hypothesis (8.2.6) is satisfied, and that . PK is
chosen as in (8.4.9), for all . K ∈ τh . We suppose on n
the other hand that the solution .u
of the problem (8.1.7)–(8.1.8) belongs to . H k+2 (G) W k+1,∞ (G). Then, there exists
a constant .C > 0 independent of .h such that for all .i = 1, . . . , I , we have

||u − u h ||0,2,Gi ≤ C(Z h k−1 ||u||k+1,2,Gi + h k+1 ||u||k+2,2,Gi ),

. (8.4.27)
({ ) 21
. B(u − u h )2 ds ≤ C(Z h k−1 ||u||k+1,2,Gi +
∂+ Gi

. h k+1 (||u||k+2,2,Gi + ||u||k+1,∞,Gi )), (8.4.28)

and
1 3
|||B| 2 (u − u h )||h,∞ ≤ C(Z h k− 2 ||u||k+1,2,G + h k+1 (||u||k+2,2,G + ||u||k+1,∞,G )).
.

(8.4.29)
♦

Proof By combining the relation (8.4.7), with .vh = Rh u, .q = qh , and the inequality
(8.4.26), we have
{ {
1 1
. B(u h − Rh u) ds + α||u h − Rh u||0,K ≤ −
2 2
B(th − qh )2 ds +
2 ∂+ K 2 ∂− K

K ||u||k+1,2,K + h K ||u||k+2,2,K )||u h − Rh u||0,K .

C(Z K h k−1 k+1
.

By summing over all the elements . K j , .1 ≤ j ≤ i, it comes

{
1
. B(u h − Rh u)2 ds + α||u h − Rh u||20,Gi ≤
2 ∂+ Gi

K ||u||k+1,2,Gi + h K ||u||k+2,2,Gi )||u h − Rh u||0,Gi . (8.4.30)

C(Z K h k−1 k+1
.

@seismicisolation
@seismicisolation
References 575

The estimates (8.4.27) and (8.4.28) follow directly from the inequality (8.4.30) and
from Lemma 8.4.7. The function .(u h − Rh u)2 is a polynomial of . P2k on each face . S
included in .∂+ Gi . The integral
{
. B(u h − Rh u)2 ds
∂+ Gi

is, therefore, calculated exactly by using on each face. S ⊂ ∂+ Gi a quadrature formula

built from abscissas of Gauss-Radau. At these points, we have .u = Rh u. On the
other hand, the weights associated with the formula are reduced by the product of
the length of the side . S by a constant independent of .h. We deduce the inequality
(8.4.29). Q.E.D.

Corollary 8.4.10 We assume that the hypotheses of the Theorem 8.4.9 are satisfied,
and that the inequality (8.2.16) is verified, for all . K ∈ τh . Then,

||u − u h ||0,Gi ≤ Ch k+1 ||u||k+2,2,Gi ,

.

({ ) 21
. B(u − u h ) ds
2
≤ Ch k+1 (||u||k+2,2,Gi ) + ||u||k+1,∞,Gi ),
∂+ Gi

and
1 1
. |||B| 2 (u − u h )||h,∞ ≤ Ch k+ 2 (||u||k+2,2,G ) + ||u||k+1,∞,G ). ♦

References

1. P. Lesaint, Sur la résolution des systèmes hyperboliques du premier ordre by des méthodes
d’éléments finis (Thèse, Pierre et Marie Curie, Paris, 1975)
2. B. Abdelmoumen, A. Dehici, A. Jeribi, M. Mnif, Some new properties in Fredholm theory,
Schechter essential spectrum, and application to transport theory. J. Inequal. Appl. Art. ID
852676, 14 pp (2008)
3. A. Ben Amar, A. Jeribi, M. Mnif, Some applications of the regularity and irreducibility on
transport theory. Acta Appl. Math. 110(1), 431–448 (2010)
4. S. Charfi, A. Elleuch, A. Jeribi, On the time asymptotic behavior of a transport operator
with bounce-back boundary condition. Applied mathematics in Tunisia, 329–358. Springer
Proceedings in Mathematics and Statistics, vol. 131 (Springer, Cham, 2015)
5. S. Charfi, A. Jeribi, N. Moalla, Time asymptotic behavior of the solution of an abstract Cauchy
problem given by a one-velocity transport operator with Maxwell boundary condition. Collect.
Math. 64(1), 97–109 (2013)
6. A. Dehici, A. Jeribi, K. Latrach, Spectral analysis of a transport operator arising in growing
cell populations. Acta Appl. Math. 92(1), 37–62 (2006)
7. A. Jeribi, Quelques remarques sur les opérateurs de Fredholm et application à l’équation de
transport. (French) [Some remarks on Fredholm operators and application to the transport
equation] C. R. Acad. Sci. Paris Sér. I Math. 325(1), 43–48 (1997)

@seismicisolation
@seismicisolation
576 8 Approximation of the Transport Equation in Plane Two-Dimensional Geometry …

8. A. Jeribi, Some remarks on the Schechter essential spectrum and applications to transport
equations. J. Math. Anal. Appl. 275(1), 222–237 (2002)
9. A. Jeribi, K. Latrach, Quelques remarques sur le spectre essentiel et application à l’équation de
transport. (French) [Some remarks on the essential spectrum and application to the transport
equation] C. R. Acad. Sci. Paris Sér. I Math. 323(5), 469–474 (1996)
10. A. Jeribi, I. Walha, Gustafson, Weidmann, Kato, Wolf, Schechter and Browder essential spectra
of some matrix operator and application to two-group transport equation. Math. Nachr. 284(1),
67–86 (2011)
11. K. Latrach, A. Jeribi, On the essential spectrum of transport operators on L1-spaces. J. Math.
Phys. 37(12), 6486–6494 (1996)
12. K. Latrach, A. Jeribi, H. Megdiche, Time asymptotic behavior of the solution to a Cauchy
problem governed by a transport operator. J. Integral Eqs. Appl. 17(2), 121–139 (2005)
13. N. Moalla, M. Damak, A. Jeribi, Essential spectra of some matrix operators and application to
two-group transport operators with general boundary conditions. J. Math. Anal. Appl. 323(2),
1071–1090 (2006)
14. G.I. Bell, S. Glasstone, Nuclear Reactor Theory (van Nostrand Reinhold Company, 1970)
15. K.D. Lathrop, Spacial differencing of the transport equation, Positivity versus accuracy. J.
Comput. Phys. 4, 475–498 (1969)
16. K. D. Lathrop, B. Carlson, Transport theory. The method of discrete ordinates. Computing
methods in reactor physics (H. Greenspan, C.N. Kelerb, D. Okrent, eds.) (Gordon and Breach,
1968), pp. 165–266
17. W.H. Reed, K.D. Latrop, Truncation error analysis of finite difference approximations to the
transport equation. Nucl. Scie. Eng. 41, 237–248 (1970)
18. K.O. Friedrichs, Symmetric positive linear differential equations. Commun. Pure Appl. Math.
11, 333–418 (1958)
19. R.S. Philipps, L. Sarason, Singular symmetric positive first order differential operators. J. Math.
Mech. 15, 235–271 (1966)
20. P. Lesaint, P.A. Raviart, On a finite element method for solving the neutron transport equation.
Mathematical aspects of finite elements in partial differential equations (Proceedings of the
Symposium Mathematics Research Center, University of Wisconsin, Madison, Wis., 1974), pp.
89–123. Publication No. 33, (Mathematics and its Research Center, University of Wisconsin-
Madison, Academic Press, New York, 1974)
21. W.H. Reed, T.R. Hill, F. Brinkley, K.D. Lathrop, A two dimensional multigroup triangular
mesh planar geometry explicit transport code, LA 5428 MS. Los Alamos Scientific Laboratory
of the University of California (1974)
22. V.I. Krylov, Approximation calculation of integrals (Macmillan New York, London, 1962)
23. K.D. Lathrop, B. Carlson, Numerical solution of the Boltzmann transport equation. J. Comput.
Phys. 1, 173–197 (1967)
24. P.G. Ciarlet, P.-A. Raviart, General Lagrange and Hermite interpolation in.Rn with applications
to finite elements methods. Arch. Rational Mech. Anal. 46, 177–199 (1972)
25. R. Radau, Etude sur les formules d’approximation qui servent à calculer la valeur numérique
d’une intégrale définie. J. Math. Pures Appl. 3(6), 283–336 (1880)

@seismicisolation
@seismicisolation
Chapter 9
Exercises with Solutions

9.1 Exercises

Exercise 9.1 Demonstrate that if .(xn )n and .(yn )n are two sequences contained in
the unit ball of a prehilbertian space such that

. lim (xn , yn ) = 1,
n→∞

then
. ||xn − yn || → 0 when n → ∞.

♦
Exercise 9.2 ([1]) Let .C ([0, 1], R) be the vector space of continuous functions on
0

[0, 1], with real values. We pose for . f ∈ C 0 ([0, 1], R),
.

{ 1
|| f ||1 =
. | f (t)|dt
0

and
.|| f ||∞ = sup | f (t)|.
t∈[0,1]

1. Show that .|| · ||1 and .|| · ||∞ are two norms on .C 0 ([0, 1], R).
2. Checked that .|| f ||1 ≤ || f ||∞ but that these two norms are not equivalent.
3. We pose ⎧
⎪ 1
⎨ n if 0 ≤ x ≤ 2
. f n (x) :=
n
⎪ 1 1
⎩ √ if 2 ≤ x ≤ 1.
x n

( f n )n is it a Cauchy sequence. .(C 0 ([0, 1], R), || · ||1 ) is it complete.

. ♦
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 577
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_9
@seismicisolation
@seismicisolation
578 9 Exercises with Solutions

Exercise 9.3 Let .x1 , .. . ., .xn be elements given in a Hilbert space . X , and let .λ1 , .. . .,
λ , be .n real given. Show that
. n

. M = {x ∈ X such that (x, xk ) ≤ λk , k = 1, . . . , n}

is a closed convex set of . X . ♦

Exercise 9.4 ([1]) 1. Let .C be a non-empty, closed, convex set, of a Hilbert space
. X . Show that .C has one element and only one with the smallest norm.
2. We consider in .l 2 the sequence
⎛ ⎞
1
. X = ⎝0, . . . , 0, 1 + , 0, . . .⎠ n ≥ 1.
n
, ,, , n
(n−1)

.(a) Show that . F := {X n , n ∈ N∗ } is closed in .l 2 .

.(b) Deduce that . F has no element with the smallest norm. ♦
Exercise 9.5 ([1]) Let .C 0 ([0, 1], R) be the vector space of continuous functions on
.[0, 1] and with values in .R, provided with the following norm

|| f || = | f (0)| + || f ||1
.
{ 1
= | f (0)| + | f (t)|dt.
0

Let . f 0 be the function everywhere equal to .1 and let

. F = { f ∈ C 0 ([0, 1], R) checking f (0) = 0}.

Calculate .dist( f 0 , F). Is there a function . f ∈ F such that

|| f 0 − f || = dist( f 0 , F).
.

♦
Exercise 9.6 ([1]) Let . X be a Hilbert space and .a be a non-zero element of . X . We
denote by . X a the vector subspace generated by .a.
1. Express the orthogonal projection on . X a of an element .x ∈ X , and its orthogonal
projection on . X a⊥ .
2. Deduce the value of .dist(x, X a⊥ ).
3. Let . X = L 2 (]0, 1[, R), . f be the constant function equal to .1 and
{ { 1 }
. F = g ∈ X such that g(x) d x = 0 .
0

Calculate .dist(g0 , F), where .g0 is the function defined by .g0 (x) = e x . ♦

@seismicisolation
@seismicisolation
9.1 Exercises 579

Exercise 9.7 Let .m, .n ∈ N∗ such that .0 < m < n and let .C be a rectangular matrix
.(m, n) of rank .m. Let

. M = {v ∈ Rn such that Cv = 0}.

1. Show that . M ⊥ = {t Cv, v ∈ Rm } with .t C is the transpose of the matrix .C.

2. Show that the matrix .C t C is invertible.
3. Deduce the expression of . PM ⊥ and of . PM . ♦

Exercise 9.8 ([1]) Let . E be a non-empty set and . X be a vector space of functions
defined on . E with values in .R, provided with a structure of Hilbert space. We notice
.( f, g) the inner product of two elements . f , . g of . X , and .|| f || the norm of . f . We
assume that for all .x ∈ E, there is a constant .C x > 0 such that

| f (x)| ≤ C x || f || for all f ∈ X.

.

1. Show that for all .x ∈ E, there is .k x ∈ X unique such that

. f (x) = ( f, k x ) for all f ∈ X.

2. Check that the family .(k x )x∈E is total in . X , i.e.,

{k x , x ∈ E}⊥ = {0}.
.

3. For .(x, y) ∈ E 2 , we pose . K (y, x) = k y (x).

.(a) Verify that for all .(x 1 , . . . , x n ) ∈ E and any .(α1 , . . . , αn ) ∈ R , we have
n n

E
. K (xi , x j )αi α j ≥ 0.
1≤i, j≤n

(b) Check that if .(en )n≥1 is a Hilbert basis of . X , then

.

∞
E
. K (y, x) = en (x)en (y).
n=1

4. Let .(x1 , . . . , xn ) ∈ E n be such that the determinant

. det(K (xi , x j ))1≤i, j≤n /= 0.

Let .(a1 , . . . , an ) ∈ Rn , and note .c1 , . . . , cn the unique solution of the system of
equations E
. K (xi , x j )ci = a j j = 1, . . . , n.
1≤i≤n

@seismicisolation
@seismicisolation
580 9 Exercises with Solutions

Verify that among the functions . f ∈ X checking . f (x j ) = a j for . j = 1, . . . , n, the

function E
. f0 = ci k xi
1≤i≤n

is the one with the smallest norm. ♦

Exercise 9.9 Let . E := C ([0, 1], R) be the space of continuous functions of .[0, 1]
0

on .R, provided with inner product

{ 1
( f, g) =
. f (x) g(x) d x,
0

for all . f , .g ∈ E. We pose

{ a
.U( f ) = x 2 f (x) d x 0 < a < 1.
0

1. Show that .U is a continuous linear form on . E.

2. Show that /
a5
.||U || ≤ .
5

3. Let .( f n )n be the defined function

⎧ 2
⎪
⎪ x if 0 ≤ x ≤ a
⎪
⎨ 1
. f n (x) =
−na 2 x + na 3 + a 2 if a ≤ x ≤ a +
⎪ n
⎪
⎪ 1
⎩0 if a + ≤ x ≤ 1.
n

.(a) Calculate .U ( f n ).
.(b) Give an increase of .|| f n ||.
.(c) Deduce .||U ||.
4. Show that there does not exist a vector .g ∈ E such that

. U ( f ) = ( f, g)

for all . f ∈ E. ♦
Exercise 9.10 Let . X be a separable real Hilbert space and .(wn )n be a Hilbert basis
of . X . Let .m ∈ N∗ and let’s pose

. Vm = (w1 , . . . , wm ).

Let .a(·, ·) be a continuous bilinear form and coercive on . X × X and . L(·) be a

continuous linear form on . X .

@seismicisolation
@seismicisolation
9.1 Exercises 581

1. Show directly (without using the Lax-Milgram theorem) that for all .m ∈ N, the
problem: Find .u m ∈ Vm solution of

a(u m , v) = L(v)
.

for all .v ∈ Vm admits a unique solution.

2. Prove that the sequence .(u m )m is bounded in . X . Deduce a new proof of the
Lax-Milgram theorem. ♦

Exercise 9.11 Prove that a vector subspace . E of a Hilbert space . X is dense in . X if,
and only if, . E ⊥ = {0}. ♦

Exercise 9.12 ([1]) Let . X be a Hilbert space on .R, .(en )n≥1 be a Hilbert basis of
X and let . E be the vector subspace of finite linear combinations of .(en )n≥1 and of
.
element
E∞
en
.a = .
n=1
n

1. Show that . E is dense in . X .

2. Show that if . y ∈ E and .(y, en ) = 0 for all .n = 2, 3, . . ., then . y = 0.
3. Do we have .(en )n≥2 total in . E. Conclude. ♦

Exercise 9.13 Let. X be a Hilbert space provided with an inner product.(·, ·) inducing
the norm .|| · ||. Let .a(·, ·) be a continuous bilinear form on . X × X and . L(·) be a
continuous linear form. We define the function . R by

. R(v) = sup (a(v, w) − L(w)).

||w||=1

1. Show that . R(v) is defined for all .v ∈ X and . R(·) is bounded below but the lower
bound of . R(·) is not reached in general.
2. Let . X h be a subspace of . X of finite dimension. Show that there is at least .u h ∈ X h
checking
. R(u h ) = inf R(vh ).
vh ∈X h

3. We assume that .a(·, ·) is coercive. Show that there is .u ∈ X unique such that

. R(u) = inf R(v) = 0.

v∈X

4. Establish existence of a constant .c such that

||u − u h || ≤ c inf ||u − vh ||.

.
vh ∈X h

@seismicisolation
@seismicisolation
582 9 Exercises with Solutions

Exercise 9.14 Let . K be a cone .( .λK ⊂ K for all .λ ≥ 0 .), non-empty, closed, convex
set, of a Hilbert space . X .
1. Show that the projection . PK ( f ) of a point . f ∈ X is characterized by
.(i) .( f − PK ( f ), PK ( f )) = 0,
.(ii) for all .v ∈ K , .( f − PK ( f ), v) ≤ 0.
2. Let us show that the projection . PK of best approximation of . X by . K satisfies for
all . f ∈ X and for all .λ ≥ 0,
. PK (λ f ) = λPK ( f )

and, for all . f ∈ X ,

|| f ||2 = ||PK ( f )||2 + || f − PK ( f )||2 .

.

♦
Exercise 9.15 Let . X be a Hilbert space, .a(·, ·) be a bilinear form on . X × X , sym-
metrical continuous and coercive, . L(·) be a continuous linear form on . X and . K be
a non-empty, closed, convex set, of . X . Show that there is only one .u ∈ K such that

. J (u) ≤ J (v)

for all .v ∈ K , with

1
. J (v) = a(v, v) − L(v).
2
Furthermore, .u is characterized by

.a(u, v − u) ≥ L(v − u)

for all .v ∈ K . ♦
Exercise 9.16 1. Let . K be a subset non-empty, closed, convex, of a Hilbert space
. X and let . x 0 ∈ X be such that . x 0 ∈
/ K . Show that there is .a ∈ X checking

. sup(x, a) < (a, x0 ).

x∈K

2. Let . K 1 be a closed
N convex set and . K 2 be a compact convex set of a Hilbert space
X . Show that . K 1 K 2 = ∅ if, and only if, there is a linear form .x −→ (a, x) such
.

that
. sup (x, a) < inf (x, a)
x∈K 1 x∈K 2

or
. sup (x, a) < inf (x, a).
x∈K 2 x∈K 1

@seismicisolation
@seismicisolation
9.1 Exercises 583

Exercise 9.17 Let .(X n )n be a sequence of Hilbert spaces and let . X be the space
{ ∞
}
E
. X = x = (x1 , x2 , . . . , xn , . . .) such that xn ∈ X n and ||xn || < ∞ .
2

n=1

. X is equipped with operations

. x + y = (x1 + y1 , x2 + y2 , . . . , xn + yn , . . .)

λx = (λx1 , λx2 , . . . , λxn , . . .)

.

and, with the following inner product

∞
E
. (x, y) = (xn , yn ).
n=1

1. Show that . X is a Hilbert space.

2. Let . X n = R for all .n ∈ N and . X = l 2 (R). Show that . X is a separable space. ♦

Exercise 9.18 We provide the space .C 0 ([a, b], R) with the norm

|| f ||∞ = sup | f (t)|.

.
t∈[0,1]

Let .C be the set

{ { 1 { }
2 1
.C= f ∈ C ([a, b], R) such that
0
f (t) dt − f (t) dt = 1 .
1
0 2

We define the sequence .( f n )n by

⎧ 1 1 1
⎪
⎪ 1+ if 0 ≤ t ≤ −
⎪
⎪ +
⎨ (n + 1)2 (
⎪ )
n 2 n 1
1 1 1 1 1
. f n (t) := − t− if − ≤t ≤ +
⎪
⎪ ( n ) 2 2 n+1 2 n+1
⎪
⎪
⎪
⎩− 1 +
1 1
if +
1
≤ t ≤ 1.
n 2 n+1

1. Show that .C is convex.

2. Show that .C is closed.
3. Show that .( f n )n ∈ C.
4. Calculate .|| f n ||∞ .
5. Show that
. inf || f ||∞ = 1.
f ∈C

@seismicisolation
@seismicisolation
584 9 Exercises with Solutions

6. Show that the origin has no best approximation, i.e., .C has no element with the
smallest norm. ♦
Exercise 9.19 Let . X be a real Hilbert space provided of the norm .|| · || and .a(·, ·)
be a continuous bilinear form on . X × X . Show that the following assertions are
equivalent:
(i) For all . L(·) ∈ X , , the problem: Find .u ∈ X solution of the problem
.

.a(u, v) = L(v)

for all .v ∈ X admits one solution and only one.

(ii) The bilinear form .a(·, ·) check the conditions :
.
.(a) For all .v ∈ X \{0}, there is .u ∈ X such that .a(u, v) / = 0.
.(b) . inf sup a(u, v) > 0. ♦
||u||=1 ||v||=1

Exercise 9.20 Let .Ω be an open bounded domain of .Rn and let .(am )m be a sequence
of points of .Ω such that .(am )m converges to .a in .Ω. Show that .(δam )m converges
to .δa . ♦
Exercise 9.21 Let . H be the Heaviside function defined on .R by
{
1 if x > 0
. H (x) :=
0 if x < 0.

Show that
dTH
. = δ.
dx
♦
Exercise 9.22 Let . f be a defined function on .R and admitting a derivative . f ,
continuous and bounded in the open set

R \ {x j }kj=1
.

with .x1 < x2 < · · · < xk . Consider

s = f (x j + 0) − f (x j − 0)
. j

the jump of . f at point .x j . Show that

( ), E
k
. Tf = Tf, + s j δx j ,
j=1

with .δx j is the mass of Dirac at points .x j . j = 1, . . . , k. ♦

@seismicisolation
@seismicisolation
9.1 Exercises 585

Exercise 9.23 We consider the following sequence of function .(vm )m

vm (x)
m

x0 − 1 x0 x0 + 1
2m 2m

Show that .(vm (x))m converges to .δx0 . ♦

Exercise 9.24 Let .Ω be an open of .R and let .[a, b] be a bounded closed interval
contained in .Ω. We pose .v = χ[a,b] . Calculate . ddvx in the sens of .D , (Ω). Deduce that
.v ∈
/ H 1 (Ω).
♦

Exercise 9.25 Let . f : [0, 1] −→ R be the function defined by

{
x if x ∈ [0, 21 ]
. f (x) =
1 − x if x ∈] 21 , 1].

1. Show that . f ∈ H01 (]0, 1[).

2. Show that there is a polynomial . p of degree less than or equal to .2 and only one
checking

({ 1 { 1 ) 21 ({ 1 { 1 ) 21
. ( f − p)2 d x + ( f , − p , )2 d x = inf ( f − q)2 d x + ( f , − q , )2 d x ,
0 0 q∈P2 0 0
(9.1.1)

where . P2 designates the real vector space of the polynomials of degree less than or
equal to .2.
3. Give the expression of the polynomial . p of degree less than or equal to .2 and
checking (9.1.1). ♦

Exercise 9.26 We consider the problem

d 2u
. − + u = f if x ∈]0, 1[ (9.1.2)
dx2

@seismicisolation
@seismicisolation
586 9 Exercises with Solutions

u(0) = u , (1) = 0,
. (9.1.3)

with . f ∈ L 2 (]0, 1[, R). This problem corresponds to a metal bar heated by means of
an amount of heat. f . The function.u representing temperature. We fix the temperature
,
.u = 0 in . x = 0 and, we assume that the heat flow .u = 0 in . x = 1. Put this problem in
variational form. Choose the space .V and apply the Lax-Milgram theorem. ♦

Exercise 9.27 We consider the problem at the limits: Find .u such that
⎧ ( )
⎪ d du
⎨− χ + μu = f on ]0, 1[
.
dx dx (9.1.4)
⎪
⎩ u(0) = 0, du (1) = 1,
dx

where . f ∈ L 2 (]0, 1[, R), .χ ∈ C 1 ([0, 1], R), .μ ∈ L ∞ (0, 1) with

. χ (x) ≥ α > 0

and
μ(x) ≥ β > 0,
.

for all .x ∈]0, 1[. This problem corresponds to a metal bar heated by means of a
quantity of heat. f , the function.u representing the temperature. We fix the temperature
.u = 0 in . x = 0 and, we assume that the heat flow

du
. (1) = 1 in x = 1.
dx

.1. Write the variational formulation of the boundary problem (9.1.4).(define the space
.V ).
.2. Show that the variational problem associated to (9.1.4) admit a unique solution in

an appropriate Hilbert space .V .

.3. Show that the variational problem is equivalent to the problem (9.1.4).
.4. When the functions .χ and .μ are fairly regulars .(for example in .C ([0, 1],
1

R) .) and . f ∈ H (0, 1), show that the solution of the variational problem belongs to
1

.C ([0, 1], R). Deduce that the boundary problem (9.1.4) admits a classic solution
2

and only one. ♦

Exercise 9.28 Let .a < b. For .u, .v ∈ H 1 (]a, b[), we pose

{ b { b { b
∂u ∂v
.a(u, v) = dx + u dx v d x.
a ∂x ∂x a a

1. Show that .a(·, ·) is coercive.

2. Deduce that for all . f ∈ L 2 (]a, b[, R), there is a unique solution .u in . H 1 (]a, b[)
of the problem

@seismicisolation
@seismicisolation
9.1 Exercises 587

{ b
a(u, v) =
. f v dx
a

for all .v ∈ H 1 (]a, b[). Show that

{ b { b
1
. u dx = f d x.
a b−a a

3. If { b
. f d x = 0,
a

then what is the partial derivatives equation who checks .u. ♦

Exercise 9.29 1. Put the problem

⎧
⎨ d 2u
− 2 + c(x)u = f if x ∈]0, 1[
. dx
⎩ u(0) = u(1) = 0

in variational form and show that it has a unique solution in an appropriate Hilbert
space.V with.c(·) ∈ C 1 ([0, 1], R),.c(x) ≥ 0 for all.x ∈ [0, 1], and. f ∈ L 2 (]0, 1[, R).
2. When . f ∈ H 1 (]0, 1[), check that the solution .u obtained is solution in the usual
sense.
3. For .c(x) = 1 for all .x ∈ [0, 1], adapt the current analysis to obtain an approxima-
tion .u h of .u. In particular, we will characterize the approximation space .Vh of which
we will specify a basis.
4. For .c(·) ∈ C 0 ([0, 1], R), in the general case, and .c(x) ≥ 0 for all .x ∈ [0, 1], we
divide the interval .[0, 1] into . I intervals of length .h. We pose .xi = i h, .0 ≤ i ≤ I
and, we approach
{ 1
. l(x) d x
0

by the following trapezoid formula

{ 1 ( )
1 1
. l(x) d x ∼ h l(x0 ) + l(x1 ) + · · · + l(x I −1 ) + l(x I ) .
0 2 2

.(i) Using the trapezoid formula, adapt the analysis performed in progress to obtain
an approximation .u h of .u. Show that the approximate problem is equivalent to a
linear system . Ah u h = bh .
.(ii) Show that . A h is positive definite.

.(iii) Show that . A h is monotone.

@seismicisolation
@seismicisolation
588 9 Exercises with Solutions

.(iv) Let . A0h be the matrix . Ah with .c(x) = 0 for all .x ∈ [0, 1]. Show that . A0h is
monotone.
.(v) Show that . A h ≥ A h .
0
0 −1
.(vi) Show that .(A h ) ≥ A−1
h . ⎛ ⎞
1
0 −1 0 −1 ⎜ .. ⎟
.(vii) Show that .||(A h ) ||∞ = ||(Ah ) e||∞ with .e = ⎝ . ⎠ .
1
(viii) Show that .h(A0h )−1 e is the approximate solution to the following boundary
.
problem

. − u ,, = 1 on ]0, 1[

u(0) = u(1) = 0.
.
♦
Exercise 9.30 We consider the boundary problem: Find .u such that
⎧ 2
⎪
⎨ − d u + μu = f in ]0, 1[
. dx2 (9.1.5)
⎪
⎩ u(0) = 0, du (1) + τ u(1) = 0,
dx

where . f ∈ L 2 ([0, 1], R), .μ ∈ L ∞ ([0, 1], R) with .μ(x) ≥ β > 0, for all .x ∈]0, 1[
and .τ > 0 is a real number.
1. Write the variational formulation of the boundary problem (9.1.5) (define the space
V ).
.

2. Show that the variational problem associated with (9.1.5) admits a unique solution
in an appropriate Hilbert space .V .
3. Show that the variational problem is equivalent to the problem (9.1.5).
4. When the function .μ is fairly regular (for example in .C 1 ([0, 1], R)) and . f ∈
H 1 (0, 1) show that the solution of the variational problem belongs to .C 2 ([0, 1], R).
Deduce that the boundary problem (9.1.5) admits a classical solution and a alone. ♦
Exercise 9.31 1. Put the problem
⎧
⎨ d 2u
− 2 + u = f if x ∈]0, 1[
. dx
⎩ u(0) = u(1), u , 0) = u , (1),

in variational form and show that it has a unique solution in an appropriate Hilbert
space .V with . f ∈ L 2 (]0, 1[, R).
2. When . f ∈ H 1 (]0, 1[), check that the solution .u obtained is solution in the sense
usual.

@seismicisolation
@seismicisolation
9.1 Exercises 589

3. Adapt the analysis performed in progress to obtain an approximation .u h of .u. In

particular, we will characterize the approximation space .Vh for which we will specify
a basis.
♦

Exercise 9.32 We consider the problem

⎧
⎨ d 2u
− 2 + au = f if x ∈]0, 1[
. dx
⎩ u , (0) = u(1) = 0,

with .a > 0 and . f ∈ L 2 (]0, 1[, R).

1. Put this problem in variational form (define the space .V ).
2. Define a subspace .Vh of .V .
3. Write the linear system corresponding to the approximate problem when we seek
.u h ∈ Vh . ♦

Exercise 9.33 Let . X be a real Hilbert space with norm .|| · ||, and .a(·, ·) be a contin-
uous bilinear form on . X × X with . M for constant of continuity. Let . L(·) ∈ X , . It is
assumed that there is an .u ∈ X of the problem

a(u, v) = L(v)
.

for all .v ∈ X. Let .Vh be a subspace of . X of finite dimension for which there is a
constant .βh > 0 such that

. inf sup a(vh , wh ) ≥ βh .

||vh ||=1 ||wh ||=1

1. Show that the problem: Find .u h ∈ Vh solution of

a(u h , vh ) = L(vh )
.

for all .vh ∈ Vh admits a unique solution.

2. Establish the general estimates for the errors
( )
M
.||u − u h || ≤ 1+ inf ||u − vh ||.
βh vh ∈Vh

Exercise 9.34 Consider the following triangulation of a domain .Ω, where the
vertices are numbered as in the figure opposite

@seismicisolation
@seismicisolation
590 9 Exercises with Solutions

6 5 4 3

7 2
12 11

8 9 10 1

We solve the problem

⎧
⎪ −Δu = f in Ω
⎨
.
u = 0 on ∂Ω\[A9 A10 ]
⎩ ∂u = 0 on [A9 A10 ].
⎪
∂ν
We use a finite element method with 3 nodes (degree of freedom at the vertices), give
the linear system corresponding to this problem (we will suppose that the horizontal
and vertical sides of the triangles are of lengths .h). ♦

Exercise 9.35 Let . P = P1 + {ξ η(1 − ξ − η)}.

(0, 1)

4 ( 13 , 13 )

3 1

(0, 0) (1, 0)

@seismicisolation
@seismicisolation
9.1 Exercises 591

1. Is there a unique . p ∈ P polynomial taking values given at points .1, .2, .3 and .4.
2. Can we then with this finite element, construct a space
n
. Vh ⊂ C 0 (Ω, R) H01 (Ω),
U
where .Ω := (triangles) and, then give an error estimates for the problem

. − Δu = f

with Dirichlet conditions. ♦

Exercise 9.36 Let .Ω ⊂ R2 be an open bounded and let .V = H01 (Ω) be the Hilbert
space provided of the norm .|| · ||1,Ω . Let .a(·, ·) be the bilinear form defined on .V × V
by
2 {
E ∂u ∂v
a(u, v) =
. ai j d x1 d x2 ,
i, j=1 Ω
∂ xi ∂ x j

where the .ai j ∈ L ∞ (Ω, R) and satisfy the inequality

E
2
. ai j (x1 , x2 )λi λ j ≥ α(λ21 + λ22 ),
i, j=1

with .α > 0, for all .(x1 , x2 ) ∈ Ω and for all .(λ1 , λ2 ) ∈ R2 .

1. .(i) Show that the bilinear form .a(·, ·) is continuous on .V × V and is .V -elliptic.
(ii) Let . L(·) be the linear form defined by
.

{
. L(v) = f v d x1 d x2 ,
Ω

where . f ∈ L 2 (Ω, R). Show that . L(·) is continuous on .V .

(iii) Deduce that the problem: Find .u ∈ V such that .a(u, v) = L(v) for all .v ∈ V
.

admits a unique solution .u ∈ V .

2. We suppose that the domain .Ω is polygonal and is decomposed into the union of
triangles .T of diameters .h(T ) ≤ h. Let .Wh be the space

. Wh = {vh ∈ C 0 (Ω, R) such that vh |T ∈ P1 , for all T ⊂ Ω}.

.(i) Show that n

. Wh ⊂ C 0 (Ω, R) H 1 (Ω).

(ii) Let .Vh be the subspace of the functions of .Wh equal to zero at the degrees of
.
freedom attached to the boundary .∂Ω. Show that

@seismicisolation
@seismicisolation
592 9 Exercises with Solutions

. Vh ⊂ H01 (Ω).

(iii) Show that the following approximate problem: Find .u h ∈ Vh such that
.

a(u h , vh ) = L(vh )
.

for all .vh ∈ Vh admits one solution and only one .u h ∈ Vh .

N
3. It is now assumed that the solution .u belongs to . H 2 (Ω) H01 (Ω). We define
.r h u ∈ Vh , . Vh -interpolated function of .u by

r u |T = r T u
. h

for all .T ∈ Ω, where .r T u is the . P1 -interpolated function of .u.

(i) Show that
.

. ||u − r T u||0,T ≤ Ch(T )2 |u|2,T

h(T )2
|u − r T u|1,T ≤ C
. |u|2,T .
ρ(T )

Deduce that
⎛ ⎞ 21
E h(T )4
||u − u h ||1,Ω
. ≤C⎝ |u|2 ⎠ .
ρ(T )2 2,T
T ⊂Ω

(iii) We further assume that

.
h(T )
. ≤α
ρ(T )

for all .T ∈ Ω. Show that

||u − u h ||1,Ω ≤ Ch|u|2,Ω .

.

Exercise 9.37 Let . P = {1, x, y, x y, x 2 , x 2 y}.

@seismicisolation
@seismicisolation
9.1 Exercises 593

2 h 5
1

−h h x

3 −h 6 4

1. Is there a unique polynomial . p ∈ P taking values given at points .1, .2, .3, .4, .5, and
6. Indication: We can show that the function of basis .ϕ1 (·, ·) is given by
.

x
ϕ (x, y) =
. 1 (x + h)(y + h).
4h 3

Likewise, we, can, find .ϕ2 (·, ·), .ϕ3 (·, ·), and .ϕ4 (·, ·). In fact,
1
ϕ (x, y) = −
. 2 x(x − h)(y + h),
4h 3

1
ϕ (x, y) =
. 3 x(x − h)(y − h),
4h 3

1
ϕ (x, y) = −
. 4 x(x + h)(y − h).
4h 3

On the other hand, .ϕ5 (·, ·) is given by

1
ϕ (x, y) =
. 5 (h 2 − x 2 )(y + h).
2h 3

Likewise, we, can, find .ϕ6 (·, ·). In fact,

1
ϕ (x, y) = −
. 6 (h 2 − x 2 )(y − h).
2h 3
Hence,

@seismicisolation
@seismicisolation
594 9 Exercises with Solutions

. p(x, y) = p(A1 )ϕ1 (x, y) + p(A2 )ϕ2 (x, y) + p(A3 )ϕ3 (x, y)

. + p(A4 )ϕ4 (x, y) + p(A5 )ϕ5 (x, y) + p(A6 )ϕ6 (x, y).

for the uniqueness of . p, it suffices to consider the mapping

Φ : P −→ R6
.

p −→ ( p(A1 ), p(A2 ), p(A3 ), p(A4 ), p(A5 ), p(A6 ))

and to show that .Φ is bijective.

2. .(i) Show that
.||u − r h u||0,R ≤ ch |u|2,R ,
2

with .rh u denoting the interpolated in . P of .u.

(ii) Show that
. || ||
|| d ||
|| ||
.
|| d x (u − rh u)|| ≤ ch|u|2,R .
0,R

(iii) Show that

. || ||
|| d ||
|| ||
.
|| d x (u − rh u)|| ≤ ch 2 |u|3,R .
0,R

(iv) Show that

. || ||
|| d ||
|| ||
.
|| dy (u − rh u)|| ≤ ch|u|2,R .
0,R

(v) Show that

.

({ [ ]2 ) 21
h
d 3
. (u − rh u)(x, h) dx ≤ ch 2 |u|3,R .
−h dx

(vi) Let .wh ∈ P. Show that

.

{ h( )
d dwh
. (u − rh u) (x, h) d x ≤ ch|u|3,R |wh |1,R .
−h d x dx

Exercise 9.38 Consider the operator . A defined by

dw
w −→ Aw =
. + w, for 0 < x < 1.
dx

Show that we can choose .v ∈ C 1 ([0, 1], R) so that .(Av, v)0,(0,1) < 0, where for all
u, v ∈ C 1 ([0, 1], R)
. ♦

@seismicisolation
@seismicisolation
9.1 Exercises 595

{ 1
(u, v)0,(0,1) =
. uv d x.
0

Exercise 9.39 Which boundary conditions would give the systematic choice
(7.5.1). ♦

Exercise 9.40 Consider the system (7.5.5) with

{ 1
. f d x = 0.
0

1. Give the matrices. M(0) and. M(1), admissible, to take into account the Neumann
conditions. Show that the problem has a weak solution. We can consider the space

.W = {ϕ = (ϕ̊1 , ϕ2 ) ∈ C 1 ([0, 1], R)|R × C 1 ([0, 1]), R) such that ϕ2 (0) = ϕ2 (1) = 0}.

Check that this weak solution is strong, and satisfies the expected boundary con-
ditions.
2. We consider the system (7.5.5). Give the matrices . M(0) and . M(1), admissible, to
take into account the Dirichlet conditions. Show, without making the system positive,
that the problem admits a weak solution that we will define. Interpret. ♦

Exercise 9.41 In Example 7.12.7, determine the admissible matrix . M, allowing to

take into account the Neumann boundary conditions. ♦

Exercise 9.42 Show that we can choose the signs of .α and .β to obtain an increase
in energy
{ 1 { 1
. (v2 + w2 )(x, t) d x ≤ (v2 + w2 )(x, 0) d x.
0 0

Compare these values of .α and .β with those obtained by the Friedrichs method. ♦

Exercise 9.43 Either the system

( ) ( ) ( ) ( ) ( )( ) ( )
10 ∂ u a1 0 ∂ u 10 u f1
. + + = ,
0 1 ∂t v 0 −a2 ∂ x v 01 v f2

where .a1 and .a2 are two strictly positive constants. We want to be able to express in
an admissible way the conditions

u(0) = λv(0), v(1) = μv(1).

. (9.1.6)

1. Discuss according to the values of .a1 , .a2 , .λ and .μ.

2. We premultiply by the matrix

@seismicisolation
@seismicisolation
596 9 Exercises with Solutions
( )
b(x) 0
. .
0 c(x)

Can we choose .b(·) and .c(·) to be able to express the boundary conditions (9.1.6) in
case of impossibility. ♦

Exercise 9.44 Let .W = {wh ∈ Vh ; (wh , p)(L 2 (Ω,R)) p = (A∗ vh , p)(L 2 (Ω,R)) p for all
p ∈ Vh , for vh ∈ X h∗ }. Let .{ψi } be a basis of . X h∗ . We assume that the operator . A
is positive and that the matrix . M satisfies . M + M ∗ semi-definite positive. Show that
the .{wi } ∈ Vh such that .(wi , p)(L 2 (Ω,R)) p = (A∗ ψi , p)(L 2 (Ω,R)) p for all . p ∈ Vh form
a basis of .W . ♦

Exercise 9.45 Prove that the problem (7.7.2) admits at least one solution. ♦

Exercise 9.46 Either the problem

. − u ,, + u = f, 0 < x < 1

. u(0) = u(1) = 0.

1. After having written this problem in the form of a Friedrichs system, explain
the equations for the problem (7.7.2), when .Vh = Z h × Z h , where . Z h is the space
of continuous functions, polynomials of degree .≤ 1 on each interval .[xi , xi+1 ], .0 ≤
i ≤ I − 1, where . I h = 1, and .xi = i h, .0 ≤ i ≤ I . Conclude.
2. Repeat 1 by looking for the solution .u h in the space . X h defined in (7.7.1). ♦

Exercise 9.47 We consider the problem

du
. + σ u = f, 0 < x < 1,
dx

u(0) = 0.
.

Using piecewise affine continuous functions for this problem, define the spaces
. X h and . X h∗ and write the equations satisfied by the .u i = u h (xi ). ♦

Exercise 9.48 We consider the problem

. − u ,, + u = f, 0 < x < 1,

u(0) = u(1) = 0,
.

written as a Friedrichs system. Using polynomials of degree .≤ 1, write the lin-

ear system satisfied by the unknowns, using the method defined by the problem
(7.7.5). ♦

@seismicisolation
@seismicisolation
9.1 Exercises 597

Exercise 9.49 We consider the scheme (7.9.8), (7.9.9), (7.9.10) with .m i = 0, .1 ≤

i ≤ I − 1.
1. Show that we can assign the values .u i+ 21 to points . yi+ 21 ∈ [xi , xi+1 ] so that the
truncation error is in . O(h) for all the equations (we will assume that .h i = h for
all .i).
2. Carry out directly by finite difference techniques in the upper bound of the error
for the scheme of 1. ♦

Exercise 9.50 We assume that the domain .Ω is a rectangle meshed with triangles
generated by three families of parallel lines. We think that
| |
. Wh = P0 (T ),
T ⊂Ω

where . P0 (T ) designates the constants on .T . Write the diagram obtained by the

discontinuous method for the transport equation

∂ϕ ∂ϕ
.μ +ν + σ ϕ = 0, in Ω. ♦
∂x ∂y

Exercise 9.51 We consider the problem

du
. + σ u = f, 0 < x < 1,
dx

u(0) = λ.
.

Let.Vh be the space of continuous functions on.[0, 1], affine on each interval.[xi , xi+1 ],
0 ≤ i ≤ I − 1, with .xi = i h, . I h = 1.
.

1. Write in this
| | particular case the method of Petrov Galerkin.
2. Let.Wh = T ⊂Ω Pk (T ), where. Pk (T ) is the space of polynomials of degree.≤ k on
the triangle .T (we can generalize to higher dimensions). We are looking for .u h ∈ Wh
such that

E{ { ((
BT − M T
) ) }
. (Au h − f, vh + h Avh )(L 2 (T,R)) p − h − u h ), vh
(u int ext int
ds = 0,
∂T 2 2
T ⊂Ω
(9.1.7)
for all .vh ∈ Wh . This is the discontinuous Petrov Galerkin method. Show that the
problem (9.1.7) admits a unique solution .u h and perform the error bound. ♦

@seismicisolation
@seismicisolation
598 9 Exercises with Solutions

9.2 Corrected

Solution: Exercise 9.1

We have .||xn || ≤ 1 and .||yn || ≤ 1. Hence,

||xn − yn ||2 = (xn − yn , xn − yn )

.

= ||xn ||2 + ||yn ||2 − 2(xn , yn )

≤ 2(1 − (xn , yn )).

By crossing the limit, we find the result.

Solution: Exercise 9.2
1. Let us show that .|| · ||1 and .|| · ||∞ are norms on .C 0 ([0, 1], R). In fact, .|| f ||1 = 0,
so { 1
. | f (t)|dt = 0.
0

Suppose there is .x0 ∈]0, 1[ (the same if .x0 = 0 or .x0 = 1) such that . f (x0 ) /= 0. For
example . f (x0 ) > 0. Furthermore, . f is continuous in .x0 , hence for all .ε > 0, there
is .α ∈]0, x0 [ such that for all .x checking .|x − x0 | < α, we have

| f (x) − f (x0 )| < ε.

.

f (x0 )
For .ε = 2
> 0, there is .α ∈]0, x0 [ such that for all .x checking .|x − x0 | < α, we
have
f (x0 ) f (x0 )
.0< = f (x0 ) −
2 2
< f (x)
f (x0 )
< f (x0 ) + .
2
Thus,
{
α f (x0 ) x0
. f (x0 ) = dt
2 2 x0 −α
{ 1
< | f (t)|dt
0
= 0.

Which proves that . f (x0 ) < 0 and this is absurd because . f (x0 ) > 0. Hence, . f = 0.
On the other hand,
{ 1
||λ f ||1 =
. |λ f (t)|dt
0

@seismicisolation
@seismicisolation
9.2 Corrected 599

= |λ||| f ||1

and for all .t ∈ [0, 1], we have

| f (t) + g(t)| ≤ | f (t)| + |g(t)|.

.

So,
{ 1 { 1 { 1
. | f (t) + g(t)|dt ≤ | f (t)|dt + |g(t)|dt.
0 0 0

It follows that

|| f + g||1 ≤ || f ||1 + ||g||1 .

.

Thus, .|| · ||1 is a norm on .C 0 ([0, 1], R).

Likewise, we can easily show that .|| · ||∞ is a norm on .C 0 ([0, 1], R).
2. .| f (t)| ≤ || f ||∞ for all .t ∈ [0, 1]. Passing to the integral, we find

. || f ||1 ≤ || f ||∞ .

For .gn (t) = t n , we have

1
||gn ||1 =
.
n+1

and
.||gn ||∞ = 1.

If .|| · ||1 and .|| · ||∞ are two equivalent norms, then there is .α > 0 such that

. α||gn ||∞ ≤ ||gn ||1 .

Hence,
1
α≤
.
n+1

for all .n ∈ N. By passing to the limit when .n tend to .∞, we find

.α ≤ 0.

This is absurd because .α > 0.

3. Let . p, .q ∈ N such that . p ≥ q, we have
{ 1
. || f p − f q ||1 = | f p (x) − f q (x)| d x
0

@seismicisolation
@seismicisolation
600 9 Exercises with Solutions

{ 1 { 1 ( )
p2 q2 1
= ( p − q) d x + √ −q dx
0 1 x
p2

1 1
= − .
q p

Hence,
|| f p − f q ||1 → 0
.

when . p, .q → ∞. Thus, .( f n )n is a Cauchy sequence. Let’s reason by the absurd.

Suppose that .(C 0 ([0, 1], R), || · ||1 ) is complete. Since .( f n )n is Cauchy, then .( f n )n
is a convergent sequence. Let . f ∈ E be the limit of .( f n )n when .n tend to infinity.
Hence,
.|| f n − f ||1 → 0

when .n → ∞. Furthermore,
{ 1
.|| f n − f ||1 = | f n (t) − f (t)|dt
0
{ { | |
| 1 |
1
n2
1
= |n − f (t)| d x + | √ − f (t)| d x.
1
| x |
0 n2

Which prove | |
{ 1 | 1 |
| √ − f (t)| d x → 0 when n → ∞.
.
1
| x |
n2

So,
1
. f (x) = √
x

for all.x ∈]0, 1[. Furthermore,. f cannot be extended by continuity in.0 so. f cannot
be in . E. Which is absurd. Thus, . E is not complete.
Solution: Exercise 9.3
Convexity of M: Let .x, . y ∈ M and .t ∈ [0, 1]. Let us show that .t x + (1 − t)y ∈ M.
.
We have .x ∈ M so

(x, xk ) ≤ λk ,
.

for all .k = 1, . . . , n and . y ∈ M. Hence,

(y, xk ) ≤ λk ,
.

for all .k = 1, . . . , n. Let .k ∈ [1, n], we have

@seismicisolation
@seismicisolation
9.2 Corrected 601

(t x + (1 − t)y, xk ) = t(x, xk ) + (1 − t)(y, xk )

.

≤ tλk + (1 − t)λk
= λk .

Thus, . M is convex.
Closure of M: Consider, for .k ∈ [1, n], the function
.

. k ϕ : X −→ R
x −→ ϕk (x) = (x, xk ).

ϕ is a continuous linear form. In fact,

. k

|ϕk (x)| = |(x, xk )|

.

≤ ||xk ||||x||.

We can easily write . M under the form

n
. M= ϕk−1 (] − ∞, λk ]).
1≤k≤n

] − ∞, λk ] is a closed one of .R, .ϕk is continuous. Hence, the reciprocal range of

.
closed by a continuous mapping is closed. As a result

ϕ −1 (] − ∞, λk ])
. k

is closed of . X . Since the finite intersection of closed is closed so . M is closed of . X .

Solution: Exercise 9.4
1. .0 ∈ X , according to the closed convex projection theorem (Theorem 2.12.3), there
is .x0 ∈ C unique such that

||0 − x0 || = inf ||0 − x||.

.
x∈C

Thus,
. ||x0 || ≤ ||x||

for all .x ∈ C. Hence, .C has an element and only one having the smallest norm.
2. .(a) For .n /= m, we have
( ) ( )
1 2 1 2
. ||X n − X m ||2 = 1 + + 1+ ≥ 2.
n m

It is clear to see that . F is closed since it is formed of isolated points.

@seismicisolation
@seismicisolation
602 9 Exercises with Solutions

(b) We have
.

( )
1 2
.||X || = 1+ .
n 2
n

Hence,

. inf ||X n || = 1 < ||X m ||

n∈N∗

for all .m ∈ N∗ . Thus, . F does not have any element having the smallest norm.
Solution: Exercise 9.5
Let . f ∈ F. We have
{ 1
|| f − f 0 || = 1 +
. | f (x) − 1| d x ≥ 1.
0

Thus,

. || f − f 0 || ≥ 1.

Let .n ∈ N∗ . Consider the mapping

{ [ ]
nx if x ∈ 0, n1
. n f (x) := [ ]
1 if x ∈ n1 , 1 .

We have . f n ∈ F and
{ 1
|| f 0 − f n || = 1 +
. |1 − f n (x)| d x
0
{ 1
n
= 1+ |1 − nx| d x
0
1
1 2
= 1+ −nn
n 2
1 1
= 1+ −
n 2n
1
= 1+ .
2n
Since

dist( f 0 , F) ≤ inf∗ || f 0 − f n ||
.
n∈N

@seismicisolation
@seismicisolation
9.2 Corrected 603
( )
1
= inf∗ 1 +
n∈N 2n
= 1,

then
. dist( f 0 , F) = 1.

Let’s reason by the absurd. Suppose there is a function . f ∈ F, for which,

.|| f 0 − f || = dist( f 0 , F) = 1.

Hence, { 1
. |1 − f (x)| d x = 0.
0

Thus,
. f (x) = 1

almost everywhere in .[0, 1]. Furthermore, . f is continuous on .[0, 1], hence . f (x) = 1
everywhere in .[0, 1]. In particular, for .x = 0, we have

. f (0) = 1.

Which is absurd because. f (0) = 0. Thus, there is no one function. f ∈ F such that

|| f 0 − f || = dist( f 0 , F).
.

It follows that .C 0 ([a, b], R) is a Banach space, .|| · || is not that induced by an
inner product although . F is non-empty, closed, convex.
Solution: Exercise 9.6
1. . X a is the vector subspace generated by .a. Hence,

. dim X a = 1 < ∞.

Thus, . X a is a closed vector subspace of . X , hence this is a Hilbert space. It follows

that

. PX a (x) = λa

with .λ is a scalar. Furthermore,

. x = PX a (x) + PX a⊥ (x).

Hence,

@seismicisolation
@seismicisolation
604 9 Exercises with Solutions

. x − PX a (x) = PX a⊥ (x) ∈ X a⊥ .

Thus,

(x − λa, a) = 0.
.

It thus appears that

(x, a)
λ=
. .
||a||2

Which proves
(x, a)
. PX a (x) = a.
||a||2

So, we have
(x, a)
. PX a⊥ (x) = x − a.
||a||2

2. We have

. dist(x, X a⊥ ) = ||x − PX a⊥ (x)||

|(x, a)|
= .
||a||

3. It suffices to see that

. F = {g ∈ X such that (g, f ) = 0} = X ⊥f .

According to 2, we have

dist(g0 , F) = dist(g0 , X ⊥f )
.

|(g0 , f )|
= .
|| f ||

On the other hand,

{ 1
(g0 , f ) =
. ex d x
0
= e−1≥0

and

@seismicisolation
@seismicisolation
9.2 Corrected 605
{ 1
|| f || =
. d x = 1.
0

Thus,

dist(g0 , F) = e − 1.
.

Solution: Exercise 9.7

1. Let . F = {t Cv, v ∈ Rm }. Let us show that . F = M ⊥ . Let .w ∈ F, then

w = t Cv,
.

with .v ∈ Rm . Let .u ∈ M, we have

(w, u) = (t Cv, u)
.

= (v, Cu)
= 0.

Thus, .w ∈ M ⊥ . It follows that . F ⊂ M ⊥ . On the other hand,

O
.M M ⊥ = Rn .

So, .dim M ⊥ = n − dim M. Let

ϕ : Rn −→ Rm
.

v −→ ϕ(v) = Cv.

We have . K er (ϕ) = M and

O
.Rn = K er (ϕ) I m(ϕ)
O
=M I m(ϕ).

So,
. dim M = n − m

since

rank(C) = dim I m(C)

.

= dim I m(ϕ)
= m.

Thus,

@seismicisolation
@seismicisolation
606 9 Exercises with Solutions

. dim M ⊥ = m.

Let us show that .dim F = m. For that, consider the mapping

ψ : Rm −→ Rn
.

v −→ ψ(v) =t Cv.

ψ is injective. In fact, .ψ is linear, so if .ψ(v) = 0, then

.

t
. Cv = 0.

Since .rank(C) = m, we can easily show that

v = 0.
.

Thus, .ψ is injective. We have

O
Rm = K er (ψ)
. I m(ψ).

Hence,

m = dim K er (ψ) + dim I m(ψ).

.

Thus,
. dim I m(ψ) = m.

Furthermore, . I m(ψ) = F, so
. dim F = m.

Since . F ⊂ M ⊥ and .dim F = dim M ⊥ , then

. F = M ⊥.

2. .C t C is an invertible square matrix of order .n. In fact, if .C t Cv = 0, then

(c t Cv, v) = (t Cv,t Cv)

.

= ||t Cv||2
= ||ψ(v)||2
= 0.

Since .ψ is injective, then

v = 0.
.

@seismicisolation
@seismicisolation
9.2 Corrected 607

Thus, .C t C is an invertible matrix.

3. We have O
. Rn = M M ⊥.

Hence, for all .v ∈ Rn ,

. v = PM v + (v − PM v)
= PM v + PM ⊥ v.

Furthermore, . PM ⊥ v ∈ M ⊥ , hence

. PM ⊥ v =t Cw

with .w ∈ Rm . Thus,

Cv = C PM ⊥ v
.

= C t Cw.

Furthermore, .C t C is an invertible matrix. It follows that

. w = (C t C)−1 Cv.

Which proves
. PM ⊥ v =t C(C t C)−1 Cv

and thus,

. PM v = (I − PM ⊥ )v
= v −t C(C tC)−1 Cv.

Solution: Exercise 9.8

1. Let .x ∈ X and consider the linear mapping .ϕx defined by

. x ϕ : E −→ R
f −→ f (x).

We have

.|ϕx ( f )| = | f (x)|
≤ C x || f ||.

@seismicisolation
@seismicisolation
608 9 Exercises with Solutions

Thus, .ϕx is a continuous linear form on . E. According to Riesz Fréchet’s rep-

resentation theorem (Theorem 2.14.1), there is .k x ∈ E unique such that for all
. f ∈ E,

ϕ ( f ) = f (x)
. x

= ( f, k x ).

2. Let . f ∈ {k x , x ∈ E}⊥ . Then, for all .x ∈ E,

. ( f, k x ) = 0.

Hence, for all .x ∈ E,

. f (x) = 0.

So,
. f = 0E .

Thus,

{k x , x ∈ E}⊥ = {0}.
.

3. .(a) We have
E E
. K (xi , x j )αi α j = k xi (x j )αi α j
1≤i, j≤n 1≤i, j≤n
E
= (k xi , k x j )αi α j
1≤i, j≤n
/ n \
E E
n
= αi k xi , α j kx j
i=1 j=1
|| n ||2
||E ||
|| ||
= || αi k xi || ≥ 0.
|| ||
i=1

(b) According to Parseval identity

.

∞
E
(k y , k x ) =
. (k y , en )(k x , en ).
n=1

Thus,
∞
E
. K (y, x) = (k y , en )(k x , en )
n=1

@seismicisolation
@seismicisolation
9.2 Corrected 609

∞
E
= en (x)en (y).
n=1

4. We pose

. F = { f ∈ X such that f (x j ) = a j for j = 1, . . . , n}.

It is clear that . F is a convex set. Let us show that . F is closed. In fact, consider the
linear mapping

ψ : X −→ Rn
.

f −→ ( f (x1 ), . . . , f (xn )).

On .Rn , we consider the norm (because in a finite-dimensional space all the norms
are equivalent)
.||y||∞ = sup |yi |.
1≤i≤n

We have

||ψ( f )||∞ = sup | f (xi )|

.
1≤i≤n

≤ sup C xi || f ||.
1≤i≤n

Hence, .ψ is continuous. Furthermore,

. F = ψ −1 {(a1 , . . . , an )}

is a closed subspace. In addition,

f (x j ) = a j
. 0

for . j = 1, . . . , n, so . f 0 ∈ F and . F is non-empty. Since . F is closed, convex, .0 E ∈

X , then according to projection theorem (Theorem 2.12.3), there is . f 1 ∈ F unique
such that

||0 E − f 1 || = dist(0 E , F)
.

= min ||0 E − f ||.

f ∈F

On the other hand, for all . f ∈ F, we have

/ \
E
n
( f, f 0 ) =
. f, ci k xi
i=1

@seismicisolation
@seismicisolation
610 9 Exercises with Solutions

E
n
= ci ( f, k xi )
i=1
En
= ci ai .
i=1

In particular, for . f = f 0 , we obtain

|| f 0 ||2 = ( f 0 , f 0 )
.

E n
= ci ai .
i=1

By using Cauchy-Schwarz inequality, for all . f ∈ F, we have

.|| f 0 ||2 = ( f, f 0 )
≤ || f |||| f 0 ||.

Hence, for all . f ∈ F

.|| f 0 || ≤ || f ||.

So,

. || f 0 || = min || f ||.
f ∈F

Thus,
. 1 f = f0 .

So, . f 0 is the one with the smallest norm.

Solution: Exercise 9.9
1. .U is linear. According to Cauchy-Schwarz inequality,
|{ a |2
| |
|U ( f )|2 = ||
. x 2 f (x) d x ||
0
{ a { a
≤ x 4d x | f (x)|2 d x
0 0
a5
≤ || f ||2 .
5
Thus, .U is continuous.
2. It’s clear that according to 1 that

@seismicisolation
@seismicisolation
9.2 Corrected 611
/
a5
. ||U || ≤ .
5
3. .(a) We have . f n ∈ E and

a5
. U ( fn ) = .
5
(b) We have
.

{ a { a+ n1
|| f n ||2 =
. x 4d x + (−na 2 x + na 3 + a 2 )2 d x
0 a
{ a+ 1
a5 n
≤ + dx
5 a
a5 1
≤ + .
5 n

(c) For all .n ∈ N∗ , we have

.

|U ( f n )|
||U || ≥
.
|| f n ||
a5
≥/ 5
.
a5
5
+ 1
n

Which proves for all .n ∈ N∗ ,

a5
||U || ≥ /
.
5
.
a5
5
+ 1
n

By making .n tend to infinity, we find

/
a5
. ||U || ≥ .
5
Thus,
/
a5
||U || =
. .
5
4. It is assumed that there is an .g ∈ E such that for all . f ∈ E,

U ( f ) = ( f, g).
.

@seismicisolation
@seismicisolation
612 9 Exercises with Solutions

Hence, for all . f ∈ E,

{ 1 { 1
. f (x)x 2 χ[0,a] (x) d x = f (x)g(x) d x.
0 0

Let’s pose
g (x) = x 2 χ[0,a] (x).
. 0

We have .g0 ∈ L 2 (]0, 1[, R) and for all . f ∈ E,

( f, g0 ) = ( f, g).
.

Thus, for all . f ∈ E,

. ( f, g0 − g) = 0.

Since . E is dense in . L 2 (]0, 1[, R), then by using Lemma 2.9.11,

g − g ∈ E ⊥ = {0 L 2 (]0,1[,R) }.
. 0

Which proves
. g(x) = g0 (x)

almost everywhere in .[0, 1]. However, .g cannot, therefore, be continuous at the point
x = a. Therefore, there is no vector .g ∈ E such that
.

. U ( f ) = ( f, g)

for all . f ∈ E. Thus, .(C 0 ([0, 1], R), (·, ·)) is not a Hilbert space.
Solution: Exercise 9.10
1. Let .m ∈ N∗ and .u ∈ Vm . Consider the continuous linear form

a(u, ·) : Vm −→ R.
.

According to Riesz Fréchet’s representation theorem (Theorem 2.14.1), there is

. Au ∈ Vm unique such that

a(u, v) = (Au, v)
.

for all .v ∈ Vm . The operator

. A : Vm −→ Vm

@seismicisolation
@seismicisolation
9.2 Corrected 613

is linear and continuous. In fact, let .u, .v ∈ Vm , .α, .β ∈ R, .αu + βv ∈ Vm and

a(αu + βv, ·) ∈ Vm, (the dual of.Vm ), then according to Riesz Fréchet’s representation
.
theorem (Theorem 2.14.1), there is . A(αu + βv) ∈ Vm unique such that

a(αu + βv, w) = (A(αu + βv), w)

.

for all .w ∈ Vm . On the other hand, .u ∈ Vm and .a(u, ·) ∈ Vm, , then according to Riesz
Fréchet’s representation theorem (Theorem 2.14.1), there is . Au ∈ Vm unique such
that

a(u, w) = (Au, w)
.

for all .w ∈ Vm . Since .v ∈ Vm and .a(v, ·) ∈ Vm, , according to Riesz Fréchet’s

representation theorem (Theorem 2.14.1), there is . Av ∈ Vm unique such that

. a(v, w) = (Av, w)

for all .w ∈ Vm . Thus, . A is linear. Let’s show the continuity of . A. In fact, according
to Riesz Fréchet’s representation theorem (Theorem 2.14.1),

.||Au|| = sup |(Au, v)|

||v||=1

= sup |a(u, v)|

||v||=1

≤ c||u||

because .a(·, ·) is continuous. Thus, . A is continuous. On the other hand, . L(·) ∈ Vm,
so according to Riesz Fréchet’s representation theorem (Theorem 2.14.1), there is
. f ∈ Vm unique such that
. L(v) = ( f, v)

for all .v ∈ Vm . Hence, the formulation

a(u, v) = L(v)
.

for all .v ∈ Vm is equivalent to

(Au, v) = ( f, v)
.

for all .v ∈ Vm . Thus,

. Au = f.

Let us show that the operator . A is bijective. In fact, the bilinear form .a(·, ·) is
coercive hence there is .α > 0 such that

@seismicisolation
@seismicisolation
614 9 Exercises with Solutions

. a(u, u) ≥ α||u||2 .

Which proves if . Au = 0, then .u = 0. In fact, if . Au = 0, then

.0 = (Au, u) ≥ α||u||2 .

So, .u = 0. Thus, . A is linear continuous injective if, and only if, . A is bijective. Hence,

. Au = f

if, and only if,

u = A−1 f.
.

Thus, we have the existence and the uniqueness of the problem: Find .u m ∈ Vm
solution of

. a(u m , v) = L(v)

for all .v ∈ Vm .
2. Since .a(·, ·) is coercive, then there is .α > 0 such that

a(u, u) ≥ α||u||2
.

for all .u ∈ Vm . In particular, .u = u m ,

.α||u m ||2 ≤ a(u m , u m )

= L(u m )
≤ ||L||||u m ||.

Hence,
1
||u m || ≤
. ||L||.
α
Thus, .(u m )m is bounded. Hence, we can extract a subsequence .(u ϕ(m) )m of .(u m )m
which converges to a some .u. We have

. a(u ϕ(m) , w j ) = L(w j )

for all . j = 1, . . . , ϕ(m). By passing to the limit when .m tend to infinity, we obtain

a(u, w j ) = L(w j )
.

for all . j ∈ N∗ . However, the sequence .(w j ) j is dense in . X , hence

@seismicisolation
@seismicisolation
9.2 Corrected 615

. a(u, v) = L(v)

for all .v ∈ X. Thus, a new proof of the Lax-Milgram theorem.

Solution: Exercise 9.11
For any vector subspace . E, it is always true that . E ⊥ = (E)⊥ . Let . E be a dense
subspace, i.e., . E = X . Then, . E ⊥ = X ⊥ = {0}
O(see Lemma 2.9.11). Conversely, if
⊥
.E = {0}, then .(E)⊥ = 0. Since . X = (E)⊥ E, then it follows that . E = X , and
so . X is dense in . X .
Solution: Exercise 9.12
1. Let .x ∈ X . Since .(en )n is a Hilbert basis, we can write .x under the form
∞
E
. x= xn en
n=1

with
∞
E
. |xn |2 < ∞.
n=1

If .x ∈ E ⊥ , then

(x, en ) = 0
.

for all .n ≥ 2 and

. (x, a) = 0.

Furthermore,

x =0
. n

for all .n ≥ 2. Thus,

. x = x1 e1 .

Hence,

x (e1 , a) = 0.
. 1

It follows that
x = 0.
. 1

Which shows that

@seismicisolation
@seismicisolation
616 9 Exercises with Solutions

. E ⊥ = {0}

and so . E is dense in . X (see Lemma 2.9.11).

2. Let . y ∈ E, we have

E
m
. y = λa + λjej.
j=1

If for all .n ≥ 2, .(y, en ) = 0, then

(y, em+1 ) = 0.
.

Hence,
(y, en ) = 0
.

for all .n = 2, 3, . . . , m. Which leads to

λ
. = 0.
m+1

Thus,

λ =0
. n

for all .n = 2, 3, . . . , m. Hence,

. y = 0.

3. It is clear that.a ∈ E and .a ∈

/ ((en )n≥2 ). This suffices to conclude that the family
is not total in . E, although, according to 2.,

{(en )n≥2 }⊥ = {0}.

.

Thus, . E it’s not complete. Hence, in a prehilbertian space, for an orthonormal system
to be a basis, it is not enough that the only element orthonormal to all the elements
of the system is a null element.
Solution: Exercise 9.13
1. Let .v ∈ X . We have .a(v, ·) ∈ X , . According to Riesz Fréchet’s representation
theorem (Theorem 2.14.1), there is . Av ∈ X unique such that for all .w ∈ X ,

a(v, w) = (Av, w).

.

. A is linear and continuous. In fact, . A is linear. For the continuity of . A, we have

a(u, v) = (Au, v)
.

@seismicisolation
@seismicisolation
9.2 Corrected 617

≤ ||Au||||v||.

Hence
. sup a(u, v) ≤ ||Au||.
||v||=1

On the other hand,

( ) / \
Au Au
. a u, = Au,
||Au|| ||Au||
= ||Au||.

So,

. sup a(u, v) = ||Au||

||v||=1

≤ M||u||.

Thus, . A is continuous. On the other hand, . L(·) ∈ X , , so according to Riesz Fréchet’s

representation theorem (Theorem 2.14.1), there is . f ∈ X unique such that for all
.w ∈ X ,
. L(w) = ( f, w).

Hence,

a(v, w) − L(w) = (Av − f, w)

.

for all .w ∈ X . However,

.a(v, ·) − L(·) ∈ X , ,

so according to Riesz Fréchet’s representation theorem (Theorem 2.14.1), there is

g ∈ X unique such that for all .w ∈ X ,
.

.a(v, w) − L(w) = (g, w).

This .g verify

. ||g|| = sup (a(v, w) − L(w)).

||w||=1

According to the uniqueness

. g = Av − f.

We have

@seismicisolation
@seismicisolation
618 9 Exercises with Solutions

0 ≤ ||Av − f ||
.

= sup (a(v, w) − L(w))

||w||=1

= R(v).

Hence,
. R(v) = ||Av − f || < ∞.

We have . R(v) ≥ 0 so . R(·) is reduced by .0. Hence, . R(·) has a lower bound. Hence,

. R(v) = || f − Av||
= inf ||w − f ||.
w∈I m(A)

We have
. A : X −→ X

is a continuous linear mapping, . I m(A) is a vector subspace of . X and . I m(A) is

not necessarily closed. Let . f ∈ I m(A) and . f ∈
/ I m(A). Hence, there is a sequence
.(wm )m ∈ I m(A) such that .(wm )m converges to . f in . X . Which proves

.0≤ inf ||w − f || ≤ ||wn − f || → 0 when n → ∞.

w∈I m(A)

Thus,

. inf ||w − f || = 0
w∈I m(A)

yet . f ∈
/ I m(A). Hence, the inf is not necessarily reached.
2. We have

. inf || f − Avh || = inf || f − w||.

vh ∈X h w∈I m(A|X h )

. A |X h : X h −→ X h , and . X h is a vector subspace of finite dimension. . I m(A |X h ) is

a closed vector subspace of . X h . According to the closed convex projection theorem
(Theorem 2.12.3), there is .b ∈ I m(A|X h ) unique such that

|| f − b|| =
. inf || f − w||.
w∈I m(A|X h )

Hence, there is .u h ∈ X h such that

. Au h = b.

Hence,

@seismicisolation
@seismicisolation
9.2 Corrected 619

. R(u h ) = || f − Au h ||
= inf || f − Aw||.
w∈X h

3. Show that . A : X −→ X is an isomorphism. In fact, let . f ∈ X . Consider the

linear form
. L(v) = ( f, v).

According to the Lax-Milgram theorem (Theorem 2.17.1), there is .u ∈ X unique

such that

a(u, v) = L(v)
.

for all .v ∈ X . Hence,

.(Au, v) = ( f, v)

for all .v ∈ X . Thus,

. Au = f.

So, for all . f ∈ V , there is .u ∈ X unique such that

. Au = f.

This implies that . A is a continuous isomorphism on . X . Hence,

. I m(A) = X.

Thus,

. inf R(v) = inf ||Av − f ||

v∈X v∈X
= inf ||w − f ||
w∈I m(A)X
= inf ||w − f ||
w∈X
=0

because . f ∈ X = R(u), with .u is the only element of . X who checks

. Au = f.

4. We have

.0 = R(u) = inf ||Av − f ||

v∈X

@seismicisolation
@seismicisolation
620 9 Exercises with Solutions

= inf ||w − f ||
w∈I m(A)=X
= ||Au − f ||.

According to the characterization of .u

( f − Au, v) = 0
.

for all .v ∈ X . Hence,

( f − Au h , vh ) = 0
.

for all .vh ∈ I m(A|X h ). Again,

( f − Au h , Avh ) = 0
.

for all .vh ∈ X h . It follows that

. (Au h − Au, Avh ) = 0

for all .vh ∈ X h . It thus appears that

(A(u h − u), Avh ) = 0

.

for all .vh ∈ X h . Thus,

.(A(u h − u), A(u h − vh )) = 0

for all .vh ∈ X h . It follows that

||A(u h − u)||2 = (A(u h − u), A(vh − u)).

.

Hence,

.||A(u h − u)||2 ≤ ||A||||vh − u||||A(u h − u)||.

So,

α||u h − u|| ≤ ||A(u h − u)||

.

≤ ||A||||vh − u||.

It thus appears that

||A||
||u h − u|| ≤
. ||vh − u||
α

@seismicisolation
@seismicisolation
9.2 Corrected 621

for all .vh ∈ X h . Thus,

||A||
||u h − u|| ≤
. inf ||vh − u||.
α vh ∈X h

Solution: Exercise 9.14

1. For . f ∈ X , according to the closed convex projection theorem (Theorem 2.12.3),
there is . PK ( f ) ∈ K unique such that

( f − PK ( f ), v − PK ( f )) ≤ 0
. (9.2.1)

for all .v ∈ K . Let us show the equivalence between (9.2.1) and .(i)–.(ii). Let .v ∈ K .
We have
.( f − PK ( f ), v − PK ( f )) ≤ 0.

By taking .v = 0 ∈ K and .v = 2PK ( f ) ∈ K , we obtain

( f − PK ( f ), −PK ( f )) ≤ 0
.

and

. ( f − PK ( f ), PK ( f )) ≤ 0.

So,

( f − PK ( f ), PK ( f )) = 0.
.

Thus, .(i). Furthermore, since for all .v ∈ K ,

( f − PK ( f ), v − PK ( f )) ≤ 0,
.

we have

.( f − PK ( f ), v) ≤ ( f − PK ( f ), PK ( f )) = 0.

Hence, for all .v ∈ K ,

.( f − PK ( f ), v) ≤ 0.

So, .(ii).
Reciprocally, let .v ∈ K ,

( f − PK ( f ), v − PK ( f )) = ( f − PK ( f ), v) − ( f − PK ( f ), PK ( f ))
.

= ( f − PK ( f ), v) ≤ 0

@seismicisolation
@seismicisolation
622 9 Exercises with Solutions

for all .v ∈ K .
2. Let . f ∈ X and .λ ≥ 0, we have

( f − PK ( f ), PK ( f )) = 0
.

and

( f − PK ( f ), v) ≤ 0
.

for all .v ∈ K . Hence,

(λ f − λPK ( f ), λPK ( f )) = 0
.

and

(λ f − λPK ( f ), v) ≤ 0
.

for all .v ∈ K . The projection of .λ f on . K is .λPK ( f ). However, the projection of .λ f

on . K is . PK (λ f ). According to the uniqueness of the projection, we have

. PK (λ f ) = λPK ( f ).

Let . f ∈ X , since
( f − PK ( f ), PK ( f )) = 0,
.

using the Pythagorean relation, we obtain

|| f ||2 = ||PK ( f )||2 + || f − PK ( f )||2 .

.

Solution: Exercise 9.15

Let us show that .a(·, ·) is an inner product on . X . In fact, let us show that .a(·, ·) is
positive definite. In fact, .a(·, ·) is coercive so

a(u, u) ≥ α||u||2 ,
.

for all .u ∈ X with .α > 0. Hence,

a(u, u) ≥ 0
.

and .a(u, u) = 0 if, and only if, .u = 0. We pose

/
- = a(u, u).
.||u||

@seismicisolation
@seismicisolation
9.2 Corrected 623

It is clear to see that .-

|| · || defined a norm on . X equivalent to the norm .|| · ||. In
fact, for .u ∈ X ,
√ / √
. α||u|| ≤ ||u||- = a(u, u) ≤ c||u||.

According to Riesz Fréchet’s representation theorem (Theorem 2.14.1), there is

. f ∈ X unique such that for all .v ∈ X ,

a( f, v) = L(v).
.

Find an .u ∈ K such that

. J (v) ≥ J (u)

for all .v ∈ K . Comes back to find an .u ∈ K such that

.a(u, v − u) − L(v − u) ≥ 0

for all .v ∈ K . Comes back to find an .u ∈ K such that

a(u − f, v − u) − L(v − u) ≥ 0
.

for all .v ∈ K . Comes back to find an .u ∈ K such that

. || -
f − u|| ≤ || -
f − v||

for all .v ∈ X . Comes back to find an .u ∈ K such that

. a( f − u, f − u) ≤ a( f − v, f − v)

for all .v ∈ X .
Solution: Exercise 9.16
1. We have .x0 ∈ X , so according to the closed convex projection theorem (Theorem
2.12.3), there is . PK (x0 ) ∈ K unique such that

. ||x0 − PK (x0 )|| = inf ||x0 − y||.

y∈K

This is equivalent to

. (x0 − PK (x0 ), y − PK (x0 )) ≤ 0

for all . y ∈ K . This is equivalent to

.(x0 − PK (x0 ), y) ≤ (x0 − PK (x0 ), PK (x0 ))

@seismicisolation
@seismicisolation
624 9 Exercises with Solutions

for all . y ∈ K . This is equivalent to

(x0 − PK (x0 ), y) ≤ (x0 − PK (x0 ), x0 ) − ||x0 − PK (x0 )||2

.

for all . y ∈ K . Which proves

. sup(x0 − PK (x0 ), y) ≤ (x0 − PK (x0 ), x0 ) − ||x0 − PK (x0 )||2

y∈K

< (x0 − PK (x0 ), x0 )

because .x0 ∈
/ K and .x0 /= PK (x0 ). Taking

a = x0 − PK (x0 ),
.

we find

. sup(x, a) < (a, x0 ).

x∈K

N
2. Note that . K 1 K 2 = ∅ if, and only if, .0 ∈/ K 1 − K 2 and .0 ∈
/ K 2 − K 1 . In fact,
if .0 ∈ K 1 − K 2 , so there is .x ∈ K 1 and . y ∈ K 2 such that

. x − y = 0.

Which proves
. x = y.
N
Thus, .x ∈ K 1 K 2 . Which is absurd.
Reciprocally, suppose that n
.K1 K 2 /= ∅.
N
Let .z ∈ K 1 K 2 , so
. z − z = 0 ∈ K1 − K2.

Which is absurd. We have. K 1 convex and. K 2 convex, so. K 1 − K 2 is convex.. K 1 − K 2

is closed because the sum of closed with compact is closed. Thus, since . X is a Hilbert
space, . K 1 − K 2 is a non-empty, closed, convex set,

. 0 x =0∈
/ K1 − K2,

so according to 1, there is .a ∈ X such that

. sup (x, a) < (a, x0 ) = 0.

x∈K 1 −K 2

@seismicisolation
@seismicisolation
9.2 Corrected 625

Hence, there is .a ∈ X such that

. sup (x, a) < 0.

x∈K 1 −K 2

Likewise, for .0 ∈
/ K 2 − K 1 , . K 2 − K 1 is a non-empty, closed, convex set, so there
is .b ∈ X such that

. sup (x, b) < 0.

x∈K 2 −K 1

Hence,

. sup (x2 − x1 , b) < 0.

x1 ∈K 1 , x2 ∈K 2

Thus,

. sup sup (x2 − x1 , b) < 0.

x1 ∈K 1 x2 ∈K 2

Which prove

. sup sup ((x2 , b) − (x1 , b)) < 0.

x1 ∈K 1 x2 ∈K 2

So,

. sup (x2 , b) + sup −(x1 , b) < 0.

x2 ∈K 2 x1 ∈K 1

It follows that

. sup (x2 , b) − inf (x1 , b) < 0.

x2 ∈K 2 x1 ∈K 1

Thus,

. sup (x2 , b) < inf (x1 , b).

x2 ∈K 2 x1 ∈K 1

Likewise, the inequality

. sup (x, a) < 0

x∈K 1 −K 2

gives

@seismicisolation
@seismicisolation
626 9 Exercises with Solutions

. sup (x1 , a) < inf (x2 , a).

x1 ∈K 1 x2 ∈K 2

Solution: Exercise 9.17

Let .(x p ) p be a sequence of Cauchy elements of . X . Let us show that .(x p ) p is con-
vergent in . X . In fact, for all .ε > 0, there is . N > 0, such that for all . p, .q > N , we
have

||x p − x q || < ε.
.

Let .ε > 0, there is . N > 0, such that for all . p, .q > N , we have
∞
E
. ||xnp − xnq ||2 < ε2 .
n=1

Hence, there is . N > 0, such that for all . p, .q > N , for all .n ∈ N∗ , we have

||xnp − xnq || < ε.

.

Let .n ∈ N∗ . The sequence .(xn ) p is a Cauchy sequence on . X n , . X n Hilbert, so

p

is convergent and converges to .xn ∈ X n . Hence, for all .n ∈ N∗ ,

p
.(x n ) p

E
n
p q
. ||xk − xk ||2 < ε2 .
k=1

Which proves

E
n
p q
. lim ||xk − xk ||2 < ε2 .
q→∞
k=1

Hence,

E
n
p
. ||xk − xk ||2 < ε2 .
k=1

For all .ε > 0, there is . N > 0, such that for all . p > N , we have

E
n
p
. ||xk − xk ||2 < ε2
k=1

for all .n ∈ N∗ . So,

@seismicisolation
@seismicisolation
9.2 Corrected 627

E
n
p
. lim ||xk − xk ||2 < ε2 .
n→∞
k=1

Thus,
∞
E
. ||xnp − xn ||2 < ε2 .
n=1

Hence, for all .ε > 0, there is . N > 0, such that for all . p > N , we have

||x p − x|| < ε.

.

Let us show that .x ∈ X . In fact,

(∞ ) 21
E
||x|| =
. ||xn || 2

n=1

≤||x − x N +1 || + ||x N +1 ||
≤ε + ||x N +1 || < ∞.

Hence,
∞
E
. ||xn ||2 < ∞.
n=1

Thus, .x ∈ X . It follows that . X is a Hilbert space.

2. Consider the sequence elements of . X

e p = (0, . . . , 0, 1, 0, . . .) = (enp )n .
.
, ,, ,
( p−1)

We have
(∞ ) 21
E
||e || =
.
p
||enp ||2 = 1.
n=1

For . p /= q, we have
∞
E
(e p , eq ) =
. enp enq = 0.
n=1

Hence, .(e p ) p is an orthonormal sequence. Let us show that .(e p ) p is maximal. In

fact, let .u ∈ X , .u = (xn )n and

@seismicisolation
@seismicisolation
628 9 Exercises with Solutions

∞
E
. |xn |2 < ∞
n=1

such that .(u, e p ) = 0 for all . p ∈ N∗ . So,

∞
E
(u, e p ) =
. xn enp = x p = 0
n=1

for all . p ∈ N∗ . Thus,

.u = 0.

This implies that .(e p ) p is maximal. It follows that .(e p ) p is a Hilbert basis.
Solution: Exercise 9.18
1. Let .λ ∈ [0, 1], . f and .g ∈ C. Let us show that

.λ f + (1 − λ)g ∈ C.

In fact, . f ∈ C, so
{ 1 { 1
2
. f (t) dt − f (t) dt = 1.
1
0 2

. g ∈ C, so
{ 1 { 1
2
. g(t) dt − g(t) dt = 1.
1
0 2

Which proves
{ 1 { 1
2
. λ f (t) dt − λ f (t) dt = λ
1
0 2

and
{ 1 { 1
2
. (1 − λ)g(t) dt − (1 − λ)g(t) dt = 1 − λ.
1
0 2

By adding up
{ 1 { 1
2
. (λ f (t) + (1 − λ)g(t)) dt − (λ f (t) + (1 − λ)g(t)) dt = λ f (t) + (1 − λ) = 1.
1
0 2

@seismicisolation
@seismicisolation
9.2 Corrected 629

So,
.λ f + (1 − λ)g ∈ C.

Thus, .C is convex.
2. Consider the mapping

.ψ : C 0 ([a, b], R) −→ R
{ 1 { 1
2
f −→ f (t) dt − f (t) dt − 1.
1
0 2

ψ is continuous. In fact, let . f , .g ∈ C 0 ([a, b], R), we have

.

|{ 1 { 1 |
| 2 |
| |
.|ψ( f ) − ψ(g)| = | ( f (t) − g(t)) dt − ( f (t) − g(t)) dt |
| 0 1
2
|
≤ sup | f (t) − g(t)|
t∈[0,1]

≤ || f − g||∞ .

Furthermore,
C = ψ −1 ({0})
.

and .ψ is continuous, so the reciprocal range of closed by a continuous mapping is

closed. Thus, .C is closed.
3. We have
{ 1 { 1
2
. f n (t) dt − f n (t) dt
1
0 2

{ ( ) { 1 ( )
2 − n+1
1 1
1 2 (n + 1)2 1
. = 1+ dt + − t− dt
0 n 2 − n+1
1 1 n 2
{ 1+ 1 ( ) { 1 ( )
2 n+1 (n + 1)2 1 1
− − t− dt − − 1+ dt
1 n 2 2 + n+1
1 1 n
( 2
)( ) ( )( )
1 1 1 1 1 1 1 1
= − 1+ + + + − 1+
2 n+1 n 2n 2n 2 n+1 n
= 1.

Thus, .( f n )n ∈ C.
4. .|| f n ||∞ = 1 + 1
n
for all .n ∈ N∗ .
5. Let . f ∈ C, we have

@seismicisolation
@seismicisolation
630 9 Exercises with Solutions

{ 1 { 1
2
1=
. f (t) dt − f (t) dt.
1
0 2

Hence,
|{ 1 { 1 |
| 2 | 1
| | 1
.1 ≤ | f (t) dt − f (t) dt | ≤ || f ||∞ + || f ||∞ .
| 0 1
2
| 2 2

So,

1 ≤ || f ||∞ .
.

Thus,

. inf || f ||∞ ≥ 1.
f ∈C

On the other hand,

. inf || f ||∞ ≤ inf∗ || f n ||∞

f ∈C n∈N
( )
1
= inf∗ 1 +
n∈N n
= 1.

Thus,
. inf || f ||∞ = 1.
f ∈C

6. Suppose there is .g ∈ C such that

||g||∞ = inf || f ||∞ ,

.
f ∈C

i.e.,
{ 1 { 1
2
. g(t) dt − g(t) dt = 1
1
0 2

and

. ||g||∞ = 1.

Let us show that the only cases of .g are

@seismicisolation
@seismicisolation
9.2 Corrected 631
{
1 if 0 ≤ t ≤ 21
. g(t) :=
−1 if 21 < t ≤ 1

or
{
−1 if 0 ≤ t ≤ 21
. g(t) :=
1 if 21 < t ≤ 1.

In both cases .g ∈
/ C 0 ([a, b], R). In fact, suppose there is .t0 ∈ [0, 21 ] such that
. g(t0 ) < 1. Hence,

{ 1
2 1
. g(t) dt < .
0 2

Furthermore,
|{ |
| 1 | 1
| |
.| g(t) dt | ≤
| 21 | 2

so
{ 1 { 1
2
. g(t) dt − g(t) dt < 1.
1
0 2

Which is absurd. Hence, .g(t) = 1 for all .t ∈ [0, 21 ]. Thus,

{ 1
1
.− g(t) dt = .
1
2
2

Hence,
{ 1
1
. g(t) dt = − .
1
2
2

Which proves
. g(t) = −1

for all .t ∈ [ 21 , 1]. Which is absurd. Hence, the origin has no best approximation. It
follows that .C 0 ([a, b], R) is a Banach space, .|| · ||∞ is not that induced by an inner
product although .C is non-empty, closed, convex.
Solution: Exercise 9.19
(i) ⇒ (ii) Let . L(·) ∈ X , . According to the Riesz Fréchet theorem, there is .v0 ∈ X
.
such that
. L(v) = (v0 , v)

@seismicisolation
@seismicisolation
632 9 Exercises with Solutions

for all .v ∈ X . According to .(i), the problem

a(u, v) = L(v)
.

for all .v ∈ X admits one solution and only one .u ∈ X . In particular, for .v = v0 ,

.a(u, v0 ) = L(v0 ) = ||v0 ||2 /= 0.

Thus, for all .v0 ∈ X \{0}, there is .u ∈ X such that

a(u, v0 ) /= 0.
.

Which proves .(a). It remains to show .(b). Let .u ∈ X , consider .a(u, ·). We have
a(u, ·) ∈ X , . Hence, according to the Riesz Fréchet theorem, there is . Au ∈ X unique
.
such that
.a(u, v) = (Au, v)

for all .v ∈ X . We define the mapping

. A : X −→ X, v −→ Av.

. A is linear (it’s not too difficult). Let us show that . A is continuous. In fact,

. sup a(u, v) = sup (Au, v).

||v||=1 ||v||=1

Furthermore,

(Au, v) ≤ ||Au||||v||.
.

Hence,

. sup a(u, v) ≤ ||Au||.

||v||=1

On the other hand,

/ \
Au
. Au, = ||Au||.
||Au||

Thus,

. sup a(u, v) = ||Au||.

||v||=1

So, since the bilinear form .a(·, ·) is continuous so there is .c > 0 such that

@seismicisolation
@seismicisolation
9.2 Corrected 633

.a(u, v) ≤ c||u||||v||

for all .u, .v ∈ X . Thus,

. ||Au|| ≤ c||u||.

It follows that . A is continuous. Let us show that . A is bijective. In fact, let . f ∈ X .

If . f /= 0, then consider the following not zero linear form

. L(v) = ( f, v)

for all .v ∈ X . Then, there is .u ∈ X unique such that

a(u, v) = L(v)
.

for all .v ∈ X . Which proves

. (Au, v) = ( f, v)

for all .v ∈ X . So,

(Au − f, v) = 0
.

for all .v ∈ X . Thus,

. Au = f.

If . f = 0 consider the zero linear form .θ ∈ X , , then there is .u ∈ X unique such that

. a(u, v) = 0

for all .v ∈ X . Hence,

. (Au, v) = 0

for all .v ∈ X . So,

. Au = 0.

However, . A(0) = 0 so .u = 0. Thus, . A realized a continuous bijection from . X into

X . Thus, . A is an algebraic isomorphism from . X into . X . Let us show that
.

. inf ||Av|| > 0.

||v||=1

In fact, if not there is a sequence .(u n )n ∈ X of norm .||u n || = 1 and

||Au n || → 0
.

@seismicisolation
@seismicisolation
634 9 Exercises with Solutions

when .n → ∞. Consider

.ψ : X −→ R, x −→ ||A−1 x||.

Using the Han Banach theorem the function.ψ is continuous on. X . Hence,.ψ(Au n )
converges to .ψ(0) in .R. Thus, .||u n || → 0. Furthermore, .||u n || = 1, which is absurd.
It thus appears that
. inf ||Av|| > 0.
||v||=1

Which proves that

. inf sup a(u, v) > 0.
||u||=1 ||v||=1

Thus, .(b).
(ii) ⇒ (i) Let . L(·) ∈ X , . According to the Riesz Fréchet theorem, there is . f ∈ X
.
such that
. L(v) = ( f, v)

for all .v ∈ X . Let .u ∈ X , consider .a(u, ·). We have .a(u, ·) ∈ X , . Hence, according
to the Riesz Fréchet theorem, there is . Au ∈ X unique such that

.a(u, v) = (Au, v)

for all .v ∈ X . Hence, find .u ∈ X such that

. a(u, v) = L(v)

for all .v ∈ X and is equivalent to finding .u ∈ X such that . Au = f. We have

. α := inf sup a(u, v)

||u||=1 ||v||=1

= inf ||Au||.
||u||=1

Hence, for all .v ∈ X of norm .||v|| = 1, .||Av|| ≥ α. Thus, .||Av|| ≥ α||v|| for all .v ∈ X .
A is injective. In fact, if . Av = 0, then since
.

||Av|| ≥ α||v||,
.

we have .||v|| = 0. Hence, .v = 0. Thus, . A is an algebraic isomorphism from . X into

A(X ). Let us show that . A(X ) = X . In fact, let .v0 ∈ A(X )⊥ . Then, for all .w ∈ A(X )
.

(w, v0 ) = 0.
.

Again, for all .u ∈ X , .(Au, v0 ) = 0. Hence,

@seismicisolation
@seismicisolation
9.2 Corrected 635

.a(u, v0 ) = 0

for all .u ∈ X . However, according to .(a), we have for all .v /= 0, there is .u ∈ X

such that
.a(u, v) / = 0.

If.v0 /= 0, then there is.u ∈ X such that.a(u, v0 ) /= 0. Which is absurd. Hence,.v0 = 0.

Thus, . A(X )⊥ = {0} and . A(X ) = X . It follows that . A is an algebraic isomorphism
from . X into . X and so for all . f 0 ∈ X , there is .u ∈ X unique such that

. Au = f 0 .
Solution: Exercise 9.20
Let .ϕ ∈ D(Ω). We have

|(δam , ϕ) − (δa , ϕ)| = |ϕ(am ) − ϕ(a)|.

.

Further, .ϕ ∈ D(Ω), so .ϕ is continuous and since .(am )m converges to .a, we have

(ϕ(am ))m converges to .ϕ(a). Thus, .(δam )m converges to .δa .
.

Solution: Exercise 9.21

Let .ϕ ∈ D(Ω). We have
/ \
dTH
. , ϕ = −(TH , ϕ , )
dx
{
= − H (x)ϕ , (x) d x
{R∞
=− ϕ , (x) d x
0
= ϕ(0)
= (δ, ϕ).

Thus,
dTH
. = δ.
dx

Solution: Exercise 9.22

See Example 2.18.2 .(b).
Solution: Exercise 9.23
Let .ϕ ∈ D(Ω). We have
{
. (vm (x), ϕ) = vm (x)ϕ(x) dx
R
{ x0 + 1
2m
= mϕ(x) d x.
x0 − 21m

@seismicisolation
@seismicisolation
636 9 Exercises with Solutions

From the formula of the mean, there is .cm ∈]x0 − 1

2m
, x0 + 1
2m
[ such that

{ x0 + 21m
1
. mϕ(x) d x = mϕ(cm )
x0 − 21m m
= ϕ(cm ).

Thus,
.(vm (x), ϕ) = ϕ(cm ).

when .m → ∞, .cm tends to .x0 . Since .ϕ is continuous,

. lim (vm (x), ϕ) = lim ϕ(cm )

m→∞ m→∞
= ϕ(x0 )
= (δx0 , ϕ).

Thus, .(vm (x))m converges to .δx0 .

Solution: Exercise 9.24
Let .ϕ ∈ D(Ω). We have

(v, , ϕ) = −(v, ϕ , )
.
{
= − v(x)ϕ , (x) d x
R
{ b
=− ϕ , (x) d x
a
= ϕ(a) − ϕ(b)
= (δa , ϕ) − (δb , ϕ)
= (δa − δb , ϕ).

Thus,
v, = δa − δb .
.

Since .δa ∈
/ L 2 (]a, b[, R) (see Proposition 2.18.10), we have .v ∈
/ H 1 (Ω).
Solution: Exercise 9.25
1. . f ∈ L 2 (]0, 1[, R) and
{ [ ]
,
1 if x ∈ 0, 21
. f (x) = ] ] ∈ L 2 (]0, 1[, R).
−1 if x ∈ 21 , 1

Thus, . f ∈ H 1 (]0, 1[). We have . f (0) = f (1) = 0 so . f ∈ H01 (]0, 1[).

@seismicisolation
@seismicisolation
9.2 Corrected 637

2. . P2 is a vector subspace of . H 1 (]0, 1[) of finite dimension equal to .3. Hence, it is

a non-empty, closed, convex subspace. According to the closed convex projection
theorem (Theorem 2.12.3), since . f ∈ H01 (]0, 1[), we have . f ∈ H 1 (]0, 1[), so there
is a polynomial . p of degree less than or equal to .2 and only one checking (9.1.1).
3. According to Corollary 2.12.5, the . p checks . p ∈ P2 and .( f − p, v)1,[0,1] = 0 for
all .v ∈ P2 . In particular, for the elements of basis for . P2 . Hence, .( f − p, 1)1,[0,1] =
0, .( f − p, x)1,[0,1] = 0, and .( f − p, x 2 )1,[0,1] = 0. We pose . p(x) = a + bx + cx 2 .
Which proves
{ 1 { 1
2 , ,
. ( f − a − bx − cx ) 1 d x + ( f − a − bx − cx 2 ) d x = 0
0 0

{ 1 { 1
2 , ,
. ( f − a − bx − cx ) x d x + ( f − a − bx − cx 2 )xd x = 0
0 0

and
{ 1 { 1
,
. ( f − a − bx − cx 2 ), x 2 d x + ( f − a − bx − cx 2 )x 2 d x = 0.
0 0

Hence,

{ 1 { 1
2
. (−a − (b − 1)x − cx ) d x + 2
(1 − a − (b + 1)x − cx 2 ) d x = 0
1
0 2

{ 1 { 1 { 1
2 2
. (1 − b − 2cx) d x + (−1 − b − 2cx) d x + (−a − (b − 1)x − cx 2 )xd x +
1
0 2 0

{ 1
. (1 − a − (b + 1)x − cx 2 )xd x = 0
1
2

and
{ 1 { 1 { 1
2 2
. 2(1 − b − 2cx)xd x + 2(−1 − b − 2cx)xd x + (−a − (b − 1)x − cx 2 )x 2 d x +
1
0 2 0

{ 1
. (1 − a − (b + 1)x − cx 2 )x 2 d x = 0.
1
2

A simple calculation leads to

@seismicisolation
@seismicisolation
638 9 Exercises with Solutions

1 1 1
a+ b+ c=
.
2 3 4

1 4 5 1
. a+ b+ c=
2 3 4 8
and
1 5 23 41
. a+ b+ c=− .
3 4 15 96
We obtain the following linear system of three equations with three unknowns .a, .b
and .c
⎛ 1 1 ⎞⎛ ⎞ ⎛ 1 ⎞
1 2 3 a 4
.⎝
1 4 5 ⎠⎝ ⎠
2 3 4
b = ⎝ 18 ⎠ .
1 5 23
3 4 15
c − 41
96

A simple calculation leads to

1 735 735 2
. p(x) = − + x− x .
976 488 488
Solution: Exercise 9.26
We notice in the boundary conditions of the problem (9.1.2)–(9.1.3), there is a con-
dition on .u (.u(0) = 0) and a condition on the derivative of .u (.u , (1) = 0). Hence,
according to the lesson the appropriate Hilbert space is

. V := {v ∈ H 1 (0, 1) such that v(0) = 0}.

Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(0) = 0} be a test function. Then,

{ 1 2 { 1 { 1
d u
.− v dx + uv d x = f v d x.
2
0 dx 0 0

Using Theorem 3.3.3, we deduce

{ 1 { 1 { 1
du dv , 1
. d x − [u v]0 + uv d x = f v d x.
0 dx dx 0 0

Which proves
{ 1 { 1 { 1
du dv
. d x − u , (1)v(1) + u , (0)v(0) + uv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have

@seismicisolation
@seismicisolation
9.2 Corrected 639
{ 1 { 1 { 1
du dv
. dx + uv d x = f v d x. (9.2.2)
0 dx dx 0 0

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + uϕ d x = f ϕ d x.
0 dx dx 0 0

This implies that

{ 1 { 1 { 1
d 2u
. − ϕ dx + uϕ d x = f ϕ d x.
0 dx2 0 0

Thus,
{ 1 ( 2 )
d u
. − 2 + u − f ϕ d x = 0.
0 dx

So,
/ \
d 2u
. − 2 + u − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx

It follows that
d 2u
. − +u− f =0
dx2

on .D , (]0, 1[). Moreover, . f ∈ L 2 (]0, 1[, R), so

d 2u
. − +u = f
dx2

on . L 2 (]0, 1[, R) (see (2.18.5)). It thus appears that

d 2u
. − +u = f
dx2
almost everywhere in .]0, 1[. Further, .u ∈ V , so

u(0) = 0.
.

It remains to show that .u , (1) = 0. In fact, let .v ∈ V , using Green’s formula to the
variational problem, we find
{ 1 2 { 1 { 1
d u , 1
.− v d x + [u v] 0 + uv d x = f v d x.
2
0 dx 0 0

@seismicisolation
@seismicisolation
640 9 Exercises with Solutions

Furthermore,
d 2u
. − + u = f,
dx2
hence
u , (1)v(1) = 0
.

for all .v ∈ V . We choose a .v ∈ V such that .v(1) /= 0. Thus,

u , (1) = 0.
.

Thus, the equivalence between the strong problem and the variational formulation.
We consider the following symmetrical bilinear form

a : V × V −→ R
.
{ {
1
∂u ∂v 1
(u, v) −→ a(u, v) = dx + uv d x
0 ∂x ∂x 0

and, the following linear form

. L : V −→ R
{ 1
v −→ L(v) = f v d x.
0

. V is a Hilbert space. In fact, consider the mapping

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(0).

ϕ is linear.
.

Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,

.

|ϕ(v)| = |v(0)|
.

≤ ||v||∞ .

Further, . H 1 (]0, 1[) continuously injects into .C 0 ([0, 1], R), so

. ||v||∞ ≤ c||v||1,]0,1[ .

Thus,
|ϕ(v)| ≤ c||v||1,]0,1[ .
.

@seismicisolation
@seismicisolation
9.2 Corrected 641

Hence, .ϕ is continuous. Since .V = ϕ −1 ({0}) and .{0} is closed of .R, the reciprocal
range of closed by a continuous map is closed. Thus, .V is a closed subspace of
. H (]0, 1[). Thus, . V is a Hilbert space.
1

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

|a(u, v)||(u, v)1,]0,1[ |

.

≤ ||u||1,]0,1[ ||v||1,]0,1[ .

Hence, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

a(u, u) = (u, u)1,]0,1[

.

= ||u||21,]0,1[
1
≥ ||u||21,]0,1[ .
2
So, .a(·, ·) is coercive.
Continuity of L(·): Let .v ∈ V . Then,
.

{ 1
|L(v)| ≤
. | f | |ϕ| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (9.2.2) admits
one solution and only one .u ∈ V .
Solution: Exercise 9.27
1. Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(0) = 0} be a test function. Then,
{ 1 ( ) { 1 { 1
d du
. − χ v dx + μuv d x = f v d x.
0 dx dx 0 0

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv
. χ d x − [χ u , v]10 + μuv d x = f v d x.
0 dx dx 0 0

This proves

@seismicisolation
@seismicisolation
642 9 Exercises with Solutions
{ 1 { 1 { 1
du dv
. χ d x − χ (1)u , (1)v(1) + χ (0)u , (0)v(0) + μuv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. χ dx + μuv d x − χ (1)v(1) = f v d x. (9.2.3)
0 dx dx 0 0

2. We consider the following symmetrical bilinear form

a : V × V −→ R
.
{ 1 { 1
du dv
(u, v) −→ a(u, v) = χ dx + μuv d x
0 dx dx 0

and, the following linear form

. L : V −→ R
{ 1
v −→ L(v) = f ϕ d x + χ (1)v(1).
0

. V is a Hilbert space. In fact, consider the mapping

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(0).

ϕ is linear.
.

Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,

.

.|ϕ(v)| = |v(0)|
≤ ||v||∞ .

Gold, . H 1 (]0, 1[) continuously injects into .C 0 ([0, 1], R), so

. ||v||∞ ≤ c||v||1,]0,1[ .

Thus,
|ϕ(v)| ≤ c||v||1,]0,1[ .
.

Hence, .ϕ is continuous. Furthermore, .V = ϕ −1 ({0}). .{0} is closed of .R, the recip-

rocal range of closed by a continuous map is closed. So, .V is a closed subspace of
. H (]0, 1[). Thus, . V is a Hilbert space.
1

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

@seismicisolation
@seismicisolation
9.2 Corrected 643

|a(u, v)| ≤ ||χ ||∞ |u|1,]0,1[ |v|1,]0,1[ + ||μ||∞ ||u||0,]0,1[ ||v||0,]0,1[

.

≤ (||χ ||∞ + ||μ||∞ )||u||1,]0,1[ ||v||1,]0,1[ .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

{ 1 ( )2 { 1
du
. a(u, u) = χ dx + μu 2 d x
0 dx 0
≥ α|u|21,]0,1[ + β||u||20,]0,1[
≥ min(α, β)||u||21,]0,1[ .

Since .min(α, β) > 0, then .a(·, ·) is coercive.

Continuity of L(·): Let .v ∈ V . Then,
.

{ 1
.|L(v)| ≤ | f | |v| d x + |χ (1)||v(1)|
0
≤ || f ||0,]0,1[ ||v||0,]0,1[ + ||χ ||∞ ||v||∞ .

Furthermore, . H 1 (]0, 1[) continuously injects into .C 0 ([0, 1], R), so

. ||v||∞ ≤ c||v||1,]0,1[ .

Hence,

. |L(v)| ≤ || f ||0,]0,1[ ||v||1,]0,1[ + c||χ ||∞ ||v||1,]0,1[

= (|| f ||0,]0,1[ + c||χ ||∞ )||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (9.2.3) admits
one solution and only one .u ∈ V .
3. Let .ϕ ∈ D(]0, 1[), we have
{ 1 { 1 { 1
du dϕ
. χ dx + μuϕ d x − χ (1)ϕ(1) = f ϕ d x.
0 dx dx 0 0

Further, .ϕ(1) = 0. This proves

{ 1 ( ) { 1 { 1
d du
.− χ ϕ dx + μuϕ d x = f ϕ d x.
0 dx dx 0 0

Thus,

@seismicisolation
@seismicisolation
644 9 Exercises with Solutions

{ 1 ( ( ) )
d du
. − χ + μu − f ϕ d x = 0.
0 dx dx

So,
/ ( ) \
d du
. − χ + μu − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx dx

It follows that ( )
d du
.− χ + μu − f = 0
dx dx

on .D , (]0, 1[). On the other hand, . f ∈ L 2 (]0, 1[, R), so

( )
d du
.− χ + μu = f
dx dx

on . L 2 (]0, 1[, R) (see (2.18.5)). It thus appears that

( )
d du
.− χ + μu = f
dx dx

almost everywhere in the domain .]0, 1[. Further, .u ∈ V , so .u(0) = 0. It remains

to show that .u , (1) = 1. In fact, let .v ∈ V , using Green’s formula to the variational
problem, we find

{ 1 ( ) { 1 { 1
d du
. − χ v d x + [χ u , v]10 + μuv d x − χ (1)v(1) = f v d x.
0 dx dx 0 0

Hence,
{ 1( ( ) )
d du
. − χ + μu − f v d x + χ (1)(u , (1) − 1)v(1) = 0.
0 dx dx

On the other hand, ( )

d du
.− χ + μu = f,
dx dx

so .χ (1)(u , (1) − 1)v(1) = 0 for all .v ∈ V . Further, .χ (1) /= 0 and, if we choose a

,
.v ∈ V such that .v(1) / = 0, then .u (1) = 1. Thus, the equivalence between the strong
problem and the variational formulation.
4. Since ( )
d du
. − χ + μu = f,
dx dx

@seismicisolation
@seismicisolation
9.2 Corrected 645

then ( )
d du
. − χ = f − μu.
dx dx

Furthermore, .u, .μ(·) ∈ C 1 ([0, 1], R), and . f ∈ H 1 (]0, 1[), so

( )
d du
. − χ ∈ H 1 (]0, 1[).
dx dx

Hence,
du
χ
. ∈ H 2 (]0, 1[).
dx

Since .χ > 0, we have .u ∈ H 3 (]0, 1[). Further, .3 > 21 + 2, so according to Theorem

3.4.9, . H 3 (]0, 1[) continuously injects into .C 2 ([0, 1], R). Hence, .u ∈ C 2 ([0, 1], R).
Thus, the obtained solution .u is solution in the usual sense.

Solution: Exercise 9.28

1. Show that .a(·, ·) is coercive amounts to showing that there is .α > 0 such that

.a(u, u) ≥ α||u||21,]a,b[

for all .u ∈ H 1 (]a, b[). Let’s reason by the absurd, suppose there is a sequence
.(u n )n ∈ H (]a, b[) such that .||u n ||1,]a,b[ = 1 and .a(u n , u n ) → 0 when .n → ∞. Fur-
1

ther,. H (]a, b[) is a separable Hilbert space, so we can extract a subsequence.(u ϕ(n) )n
1

converging to .u in . H 1 (]a, b[). This proves .||u ϕ(n) ||1,]a,b[ = 1 and

.a(u ϕ(n) , u ϕ(n) ) → 0 when n → ∞.

By crossing the limit when .n tends to .∞, we find .||u||1,]a,b[ = 1 and .a(u, u) = 0.
Thus,
{ b | |2 ({ b )2
| ∂u |
| | dx + u d x = 0.
.
|∂x |
a a

It follows that { | |2
b | ∂u |
| | dx = 0
.
|∂x |
a

and ({ )2
b
. u dx = 0.
a

Hence,
∂u
. = 0 in L 2 (]a, b[, R)
∂x

@seismicisolation
@seismicisolation
646 9 Exercises with Solutions

and { b
. u d x = 0.
a

Furthermore, . H 1 (]a, b[) ⊂ C 0 ([a, b], R) hence .u = cte and

{ b
. cte d x = 0,
a

so .cte = 0. Thus, .u = 0, absurd because .||u||1,]a,b[ = 1.

2. .a(·, ·) is a continuous and coercive bilinear form on . H 1 (]a, b[) and . L(·) is a
continuous linear form. Hence, according to the Lax-Milgram theorem (Theorem
2.17.1), the problem: Find .u ∈ H 1 (]a, b[) such that
{ b
.a(u, v) = f v dx
a

for all .v ∈ H 1 (]a, b[), admits one solution and only one .u ∈ H 1 (]a, b[). On the other
hand,
{ b { b { b { b
∂u ∂v
. dx + u dx v dx = f v d x.
a ∂x ∂x a a a

For .v = 1 ∈ H 1 (]a, b[), we have

{ b { b { b
. u dx dx = f d x.
a a a

Thus,
{ b { b
1
. u dx = f d x.
a b−a a

3. If { b
. f d x = 0,
a

then using the question 2, our system becomes

{ b { b
∂u ∂v
. dx = f v d x.
a ∂x ∂x a

Let .ϕ ∈ D(]a, b[). We have

{ b { b
, ,
. u ϕ dx = f ϕ d x.
a a

@seismicisolation
@seismicisolation
9.2 Corrected 647

Hence,
{ b { b
. − u ,, ϕ d x = f ϕ d x.
a a

This proves
{ b
. (−u ,, − f )ϕ d x = 0
a

for all .ϕ ∈ D(]a, b[). Thus, .−u ,, − f = 0 on .D , (]a, b[). Further, . f ∈ L 2 (]a, b[, R),
so .−u ,, = f on . L 2 (]a, b[, R). It thus appears that .−u ,, = f almost everywhere in
.]a, b[. Using Green’s formula for the variational problem, we find

{ b 2 { b
d u , b
.− v d x + [u v]a = f v d x.
2
a dx a

Furthermore,
d 2u
. − = f,
dx2
hence
u , (b)v(b) − u , (a)v(a) = 0
.

for all .v ∈ H 1 (]a, b[). If we choose .v ∈ H 1 (]a, b[) such that .v(b) /= 0 and .v(a) = 0,
then
,
.u (b) = 0.

If we choose .v ∈ H 1 (]a, b[) such that .v(b) = 0 and .v(a) /= 0, then

u , (a) = 0.
.

Hence, .u verify the boundary problem

. − Δu = f in ]a, b[
{ b
u d x = 0, u , (a) = u , (b) = 0.
a

Solution: Exercise 9.29

1. Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(0) = v(1) = 0} := H01 (0, 1) be a test
function. Then,
{ 1 { 1 { 1
d 2u
. − v dx + c(x)uv d x = f v d x.
0 dx2 0 0

@seismicisolation
@seismicisolation
648 9 Exercises with Solutions

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv
. d x − [u , v]10 + c(x)uv d x = f v d x.
0 dx dx 0 0

This proves

{ 1 { 1 { 1
du dv
. d x − u , (1)v(1) + u , (0)v(0) + c(x)uv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. dx + c(x)uv d x = f v d x. (9.2.4)
0 dx dx 0 0

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + c(x)uϕ d x = f ϕ d x.
0 dx dx 0 0

This implies that

{ 1 { 1 { 1
d 2u
. − ϕ dx + c(x)uϕ d x = f ϕ d x.
0 dx2 0 0

Thus,
{ 1 ( )
d 2u
. − + c(x)u − f ϕ d x = 0.
0 dx2

So,
/ \
d 2u
. − + c(x)u − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx2

It follows that
d 2u
. − + c(x)u − f = 0
dx2

on .D , (]0, 1[). Since . f ∈ L 2 (]0, 1[, R), then

d 2u
. − + c(x)u = f
dx2

on . L 2 (]0, 1[, R). It thus appears that

@seismicisolation
@seismicisolation
9.2 Corrected 649

d 2u
. − + c(x)u = f
dx2
almost everywhere in .]0, 1[. Moreover, .u ∈ V , then

. u(0) = u(1) = 0.

Thus, the equivalence between the strong problem and the variational formulation.
We consider the following symmetrical bilinear form

a : V × V −→ R
.
{ {
1
∂u ∂v 1
(u, v) −→ a(u, v) = dx + c(x)uv d x
0 ∂x ∂x 0

and, the following linear form

. L : V −→ R
{ 1
v −→ L(v) = f ϕ d x.
0

. V is a Hilbert space.
Continuity of a(·, ·): Let .u, .v ∈ V . Then,
.

| || |
{ { 1
| du | | dv |
1
.|a(u, v)| ≤ | | | | dx + |c(x)||u||v| d x
|dx | |dx |
0 0
{ 1
≤ |u|1,]0,1[ |v|1,]0,1[ + ||c||∞ |u||v| d x
0
≤ |u|1,]0,1[ |v|1,]0,1[ + ||c||∞ ||u||0,]0,1[ ||v||0,]0,1[
≤ (1 + ||c||∞ )||u||1,]0,1[ ||v||1,]0,1[ .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

{ | |2 { 1
1 | du |
a(u, u) = | | dx + c(x)u 2 d x
.
|dx |
0 0
≥ |u|21,]0,1[ .

Further, in the space .V , .| · |1,]0,1[ and .|| · ||1,]0,1[ are two equivalent norms. Hence,
there is .α > 0 such that
.|u|1,]0,1[ ≥ α||u||1,]0,1[ .

@seismicisolation
@seismicisolation
650 9 Exercises with Solutions

So,

a(u, u) ≥ α 2 ||u||21,]0,1[ .
.

Thus, .a(·, ·) is coercive.

Continuity of L(·): Let .v ∈ V . Then,
.

{ 1
|L(v)| ≤
. | f | |ϕ| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (9.2.4) admits
one solution and only one .u ∈ V .
2. Since
.
d 2u
. − + c(x)u = f,
dx2
then
d 2u
. − = f − c(x)u.
dx2

Moreover, .u, . f ∈ H 1 (]0, 1[) and .c(·) ∈ C 1 ([0, 1], R), so

d 2u
. − ∈ H 1 (]0, 1[).
dx2

This proves .u ∈ H 3 (]0, 1[). On the other hand, .3 > 21 + 2, so according to Theorem
3.4.9, . H 3 (]0, 1[) continuously injects into .C 2 ([0, 1], R). Hence, .u ∈ C 2 ([0, 1], R).
Thus, the obtained solution .u is solution in the usual sense.
3. We divide the interval .into .intervals of length.. We pose .xi = i h, .0 ≤ i ≤ I . Let
. Wh be the space
. Wh = {vh ∈ C 0 ([0, 1], R) such that vh |(xi ,xi+1 ) ∈ P1 , vh is fully determined
.by its values vi = vh (x i ) at points x i 0 ≤ i ≤ I }.

We pose

. Vh = {vh ∈ Wh such that vh (0) = vh (1) = 0}.

A basis of the space .Vh is composed of functions .ϕ j (x) ∈ Vh equal to .1 at .x j and

to zero in all .xi .i /= j. We obtain the hats functions. Thus,

@seismicisolation
@seismicisolation
9.2 Corrected 651

. Vh = {ϕ j , 1 ≤ j ≤ I − 1}.

We consider the approximate problem: Find .u h ∈ Vh solution of

.a(u h , vh ) = L(vh )

for all .vh ∈ Vh . This is equivalent to find .u h ∈ Vh solution of

. a(u h , ϕ j ) = L(ϕ j )

for all . j = 1, . . . , I − 1. Moreover,

I −1
E
. hu = u i ϕi .
i=1

Thus,
I −1
E
. u i a(ϕi , ϕ j ) = L(ϕ j )
i=1

for all . j = 1, . . . , I − 1. We have

a(ϕi , ϕ j ) = 0
.

if .|i − j| ≥ 2. It suffices to calculate .a(ϕi , ϕi ) for .i = 1, . . . , I − 1, and .a(ϕi , ϕi+1 )

for .i = 1, . . . , I − 2.
Calculation of a(ϕi , ϕi ) i = 1, . . . , I − 1: Let .i ∈ {1, . . . , I − 1}, we have
.

{ 1 ( )2 { 1
dϕi
.a(ϕi , ϕi ) = dx + ϕi2 d x.
0 dx 0

This proves
{ xi+1 ( )2 { xi+1
dϕi
.a(ϕi , ϕi ) = dx + ϕi2 d x.
xi−1 dx xi−1

Moreover,
⎧
⎪
⎪ 0 if x ∈ [x0 , xi−1 ]
⎪
⎪1
⎪
⎨ x +1−i if x ∈ [xi−1 , xi ]
.ϕi (x) := h
⎪
⎪
1
− x +1+i if x ∈ [xi , xi+1 ]
⎪
⎪
⎪
⎩ h
0 if x ∈ [xi+1 , x I ].

@seismicisolation
@seismicisolation
652 9 Exercises with Solutions

We have
{ xi+1 ( )2 { xi ( )2 { xi+1 ( )2
dϕi dϕi dϕi
. dx = dx + dx
xi−1 dx xi−1 dx xi dx
2
= .
h
On the other hand,
{ xi+1 { xi { xi+1
. ϕi2 d x = ϕi2 d x + ϕi2 d x
xi−1 xi−1 xi
{ xi (
)2 { xi+1 ( )2
1 1
= x +1−i dx + − x +1+i dx
xi−1 h xi h
[ ( )3 ]xi [ ( )3 ]xi+1
h 1 h 1
= x +1−i − − x +1+i
3 h 3 h
xi−1 xi
2h
= .
3
Thus, for all .i = 1, . . . , I − 1,

2 2h
a(ϕi , ϕi ) =
. + .
h 3

Calculation of a(ϕi , ϕi+1 ) i = 1, . . . , I − 2: Let .i ∈ {1, . . . , I − 2}, we have

.

{ 1 { 1
dϕi dϕi+1
.a(ϕi , ϕi+1 ) = dx + ϕi ϕi+1 d x.
0 dx dx 0

Then,
{ xi+1 { xi+1
dϕi dϕi+1
.a(ϕi , ϕi+1 ) = dx + ϕi ϕi+1 d x
xi dx dx xi
{ xi+1 { xi+1 ( )( )
11 1 1
=− dx + − x +1+i x − i dx
xi hh xi h h
=
{ xi+1 ( )2 { xi+1 ( )
1 1 1
=− + − x −i dx + x − i dx
h xi h xi h
[( )3 ]xi+1 [( )2 ]xi+1
1 h 1 h 1
=− − x −i + x −i
h 3 h 2 h
xi xi
1 h h
=− − +
h 3 2

@seismicisolation
@seismicisolation
9.2 Corrected 653

1 h
=− + .
h 6
Thus,
⎛ { 1 ⎞
⎛ ⎞⎛ ⎞ ⎜ 0 f ϕ 1 d x ⎟
1 + h −1 + h 0 0 ··· 0 u1 ⎜ { 1 ⎟
⎜ h1 3h 2 h 2h6 1 ⎟⎜ ⎜ ⎟
−
⎜ h + + − + h 0 · · · 0 ⎟⎜ 2 ⎟ ⎜
u ⎟ ⎜ f ϕ2 d x ⎟
⎜ 6 h 3 h 6 ⎟⎜ . ⎟ ⎜ 0 ⎟
⎜ .. .. .. .. .. ⎟⎜ . ⎟ ⎜ ⎟
⎜ . . . . . ⎟⎜ . ⎟ ⎜ . ⎟
⎜ 0 ⎟⎜ .. ⎟
.⎜ . . . . . ⎟⎜ . ⎟ ⎟ = ⎜
⎜
⎟.
⎟
⎜ .. .. .. .. .. .
⎟⎜ . ⎟ ⎜ .. ⎟
⎜ 0 ⎟⎜
⎜ . . . . ⎟⎜ . ⎟ ⎟
⎜
⎜
. ⎟
⎟
⎜ . . . . ⎟⎝ . ⎠ ⎜ .. ⎟
⎝ . . . . 1
− h + h6 ⎠ . ⎜ ⎟
⎜ . ⎟
0 − h + 6 h2 + 2h ⎝{ 1
··· ··· 1 h u I −1
0 3 ⎠
f ϕ I −1 d x
0

4. .(i) .Calculation of a(ϕi , ϕi ) i = 1, . . . , I − 1: Let .i ∈ {1, . . . , I − 1}, we have

{ 1 ( )2 { 1
dϕi
.a(ϕi , ϕi ) = dx + c(x)ϕi2 d x.
0 dx 0

Then,
{ xi+1
2
.a(ϕi , ϕi ) = + c(x)ϕi2 d x.
h xi−1

Using the trapezoid formula, we have

{ xi+1
. c(x)ϕi2 d x = hc(xi ).
xi−1

Thus, for all .i = 1, . . . , I − 1,

2
a(ϕi , ϕi ) =
. + hc(xi ).
h

Calculation of a(ϕi , ϕi+1 ) i = 1, . . . , I − 2: Let .i ∈ {1, . . . , I − 2}, we have

.

{ 1 { 1
dϕi dϕi+1
.a(ϕi , ϕi+1 ) = dx + c(x)ϕi ϕi+1 d x.
0 dx dx 0

Which proves
{ xi+1 { xi+1
dϕi dϕi+1
a(ϕi , ϕi+1 ) =
. dx + c(x)ϕi ϕi+1 d x
xi dx dx xi

@seismicisolation
@seismicisolation
654 9 Exercises with Solutions

1
=− .
h

Calculation of L(ϕi ) i = 1, . . . , I − I : Using the trapezoid formula, we have

.
L(ϕi ) = h f (xi ). Thus, the approximate problem is written under the form. Ah u h = bh
.

again,
⎛2 ⎞⎛ ⎞ ⎛ ⎞
h
+ hc(x1 ) − h1 0 0 ··· 0 u1 h f (x1 )
⎜ − h1 2
+ hc(x2 ) − h1 0 ··· 0 ⎟⎜ u2 ⎟ ⎜ h f (x2 ) ⎟
⎜ h ⎟⎜ ⎟ ⎜ ⎟
⎜ .. .. .. .. .. ⎟⎜ .. ⎟ ⎜ .. ⎟
⎜ 0 . . . . . ⎟⎜ . ⎟ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
.⎜ .. .. .. .. .. ⎟⎜ .. ⎟=⎜ .. ⎟.
⎜ . . . . . 0 ⎟⎜ . ⎟ ⎜ . ⎟
⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ .. .. .. .. ⎟⎜ .. ⎟ ⎜ .. ⎟
⎝ . . . . − h1 ⎠⎝ . ⎠ ⎝ . ⎠
0 ··· ··· 0 − h1 h2 + hc(x I −1 ) u I −1 h f (x I −1 )

(ii) Let, us, show that . Ah is positive definite. In fact, let .v ∈ R I −1 , we have
.

1[
(Ah v, v) =
. (2 + c(x1 )h 2 )v12 − v1 v2 + (−v1 v2 + ((2 + c(x2 )h 2 )v22 − v3 v2 )
h
+ (−v2 v3 + (2 + c(x3 )h 2 )v32 − v4 v3 ) + · · ·
+ v I −2 (−v I −3 + (2 + c(x I −2 )h 2 )v I −2 − v I −1 )
]
+v I −1 (−v I −2 + (2 + c(x I −1 )h 2 )v I −1
1[
= (2 + c(x1 )h 2 )v12 − 2v1 v2 + (2 + c(x2 )h 2 )v22 − 2v3 v2 )
h
+ (2 + c(x3 )h 2 )v32 + · · · − 2v I −2 v I −3 + (2 + c(x I −2 )h 2 )v2I −2
]
−2v I −1 v I −2 + (2 + c(x I −1 )h 2 )v2I −1
1[
= (1 + c(x1 )h 2 )v12 + c2 (x2 )h 2 v22 + · · · + c(x I −2 )h 2 v2I −2
h
+ (1 + c(x I −1 )h 2 )v2I −1 + (v1 − v2 )2 + (v2 − v3 )2 + · · ·
1[
= + (2 + c(x1 )h 2 )v12 − 2v1 v2 + (2 + c(x2 )h 2 )v22 − 2v3 v2 )
h ]
+ (v I −2 − v I −1 )2
≥ 0.

Thus, . Ah is positive. Hence, .(Ah v, v) = 0 if, and only if, .v1 = v2 = · · · = v I −1 = 0.

It follows that . Ah is positive definite.
(iii) Let .v ∈ {v ∈ Rn such that Av ≥ 0}. Let us show that .v ∈ {v ∈ Rn such that v ≥
.

0}. In fact, if all .c(xi ) > 0 for all .i = 1, . . . , I − 1, consider then

.k = min vi .
1≤i≤I −1

@seismicisolation
@seismicisolation
9.2 Corrected 655

If .k = v1 , then
(2 + c(x1 )h 2 )v1 − v2 ≥ 0.
.

Which proves
(2 + c(x1 )h 2 )v1 ≥ v2 ≥ v1 .
.

Hence,
(1 + c(x1 )h 2 )v1 ≥ 0.
.

Thus,
. 1v ≥ 0.

It follows that
v ≥0
. i

for all .i = 1, . . . , I − 1. If .k = v I −1 , then

. − v I −2 + (2 + c(x I −1 )h 2 )v I −1 ≥ 0.

Hence,
(2 + c(x I −1 )h 2 )v I −1 ≥ v I −2 ≥ v I −1 .
.

Thus proves
v
. I −1 ≥ 0.

Thus,
v ≥0
. i

for all .i = 1, . . . , I − 1. If .k = v j with .1 < j < I − 1, then

. − v j−1 + (2 + c(x j )h 2 )v j − v j+1 ≥ 0.

Hence,
. − v j−1 + (2 + c(x j )h 2 )v j ≥ v j+1 ≥ v j .

Which proves
. − v j−1 + (1 + c(x j )h 2 )v j ≥ v j+1 ≥ 0.

Thus,
.(v j − v j−1 ) + c(x j )h 2 v j ≥ 0.

Since .v j − v j−1 ≤ 0, then

c(x j )h 2 v j ≥ 0.
.

@seismicisolation
@seismicisolation
656 9 Exercises with Solutions

Since .c(x j ) ≥ 0, then

v ≥ 0.
. j

Thus, for all .i = 1, . . . , I − 1,

v ≥ 0.
. i

It thus appears that

.v ≥ 0.

If there is . j ∈ [1, I − 1] such that .c(x j ) = 0, then we pose

. Bh (α) = Ah (c(x j )) := c(x j ) + α

with .α > 0. . Bh (α) is so monotone. Thus,

(Bh (α))−1 ≥ 0
.

for all .α > 0. Since the function

α −→ (Bh (α))−1
.

is continuous, we have

. lim (Bh (α))−1 ≥ 0.

α→0

Which shows that

. A−1
h ≥0

and so . Ah is monotone.
(iv) If the function .c(·) is null, then the matrix . Ah become
.

⎛ ⎞
2 −1 0 0 · · · 0
⎜ −1 2 −1 0 · · · 0 ⎟
⎜ ⎟
⎜ . . . . . . . . .. ⎟
1⎜ ⎜ 0 . . . . . ⎟
⎟
. Ah = ⎟.
0
⎜ .. . . . . . . . .
h⎜ . . . . . 0 ⎟
⎜ ⎟
⎜ . . . . ⎟
⎝ .. . .
. . . −1 ⎠.
0 · · · · · · 0 −1 2

Let us show that . A0h is monotone. In fact, we have shown in .(iii) that . Ah is
monotone for .c(·) ≥ 0. In particular, for the function null .c(·). Thus, . A0h is monotone.

@seismicisolation
@seismicisolation
9.2 Corrected 657

(v) Let us show that . Ah ≥ A0h . In fact,

.

⎛ ⎞
c(x1 )h 0 0 ··· 0
⎜ 0 c(x2 )h 0 ··· 0 ⎟
⎜ ⎟
⎜ .. .. .. .. .. ⎟
. Ah − Ah = ⎜ . . . ⎟ ≥ 0.
⎜ . .
0
⎟
⎜ . .. .. ⎟
⎝ .. . . 0 ⎠
0 ··· · · · 0 c(x I −1 )h

(vi) Let us show that .(A0h )−1 ≥ A−1

. h . In fact,

(A0h )−1 − A−1

.
0 −1 −1
h = (A h ) (A h − A h )A h ≥ 0.
0

Thus,
.(A0h )−1 ≥ A−1
h .

(vii) Let us show that .||(A0h )−1 ||∞ = ||(A0h )−1 e||∞ . In fact,
.

I −1
E
||(A0h )−1 ||∞ = max
. |bi j |
i
j=1
I −1
E
= max bi j car A0h est monotone
i
j=1

= max |((A0h )−1 e)i |

i
= ||(A0h )−1 e||∞ .

(viii) We have . A0h u h = bh with .bh = he. Since . A0h is invertible, then
.

u = (A0h )−1 bh
. h

= h(A0h )−1 e.

Hence, .h(A0h )−1 e is the approximate solution of the following boundary problem

. − u ,, = 1 on ]0, 1[

u(0) = u(1) = 0.
.

@seismicisolation
@seismicisolation
658 9 Exercises with Solutions

Solution: Exercise 9.30

1. Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(0) = 0} be a test function. Then,
{ 1 2 { 1 { 1
d u
.− v d x + μuv d x = f v d x.
2
0 dx 0 0

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv , 1
. d x − [u v]0 + μuv d x = f v d x.
0 dx dx 0 0

Which proves
{ 1 { 1 { 1
du dv
. d x + τ u(1)v(1) + μuv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. d x + τ u(1)v(1) + μuv d x = f v d x. (9.2.5)
0 dx dx 0 0

2. We consider the following symmetrical bilinear form

.a : V × V −→ R
{ 1 { 1
du dv
(u, v) −→ a(u, v) = d x + τ u(1)v(1) + μuv d x
0 dx dx 0

and, the following linear form

. L : V −→ R
{ 1
v −→ L(v) = f ϕ d x.
0

. V is a Hilbert space. In fact, consider the mapping

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(0).

ϕ is linear.
.

Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,

.

|ϕ(v)| = |v(0)|
.

≤ ||v||∞ .

@seismicisolation
@seismicisolation
9.2 Corrected 659

Further, . H 1 (]0, 1[) continuously injects into .C 0 ([0, 1], R), so

. ||v||∞ ≤ c||v||1,]0,1[ .

Thus,

|ϕ(v)| ≤ c||v||1,]0,1[ .
.

Hence, .ϕ is continuous. Further, .V = ϕ −1 ({0}). .{0} is closed of .R, the recipro-

cal range of closed by a continuous map is closed. So, .V is a closed subspace of
. H (]0, 1[). Thus, . V is a Hilbert space.
1

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

|{ 1 { 1 |
| du dv |
.|a(u, v)| = | d x + τ u(1)v(1) + μuv d x ||
| d x d x
0 0
{ 1 | || | { 1
| du | | dv |
≤ | | | | d x + τ |u(1)||v(1)| + μ|u||v| d x
|dx | |dx |
0 0
≤ |u|1,]0,1[ |v|1,]0,1[ + τ ||u||∞ ||v||∞ + ||μ||∞ ||u||0,]0,1[ ||v||0,]0,1[ .

Since . H 1 (]0, 1[) continuously injects into .C 0 ([0, 1], R), so

. ||v||∞ ≤ c||v||1,]0,1[ .

So,

|a(u, v)| ≤ (1 + ||μ||∞ )||u||1,]0,1[ ||v||1,]0,1[ + τ c2 ||u||1,]0,1[ ||v||1,]0,1[

.

≤ (1 + ||μ||∞ + τ c2 )||u||1,]0,1[ ||v||1,]0,1[ .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

{ 1 ( )2 { 1
du
.a(u, u) = d x + τ u(1)2 + μu 2 d x
0 dx 0
≥ |u|21,]0,1[ + β||u||20,]0,1[
≥ min(1, β)||u||21,]0,1[ .

Thus, since .min(1, β) > 0, then .a(·, ·) is coercive.

@seismicisolation
@seismicisolation
660 9 Exercises with Solutions

Continuity of L(·): Let .v ∈ V . Then,

.

{ 1
|L(v)| ≤
. | f | |v| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (9.2.5) admits
one solution and only one .u ∈ V .
3. Let .ϕ ∈ D(]0, 1[), we have
{ 1 { 1 { 1
du dϕ
. d x + τ u(1)ϕ(1) + μ(x)uϕ d x = f ϕ d x.
0 dx dx 0 0

This implies that

{ 1 { 1 { 1
d 2u
. − ϕ dx + μ(x)uϕ d x = f ϕ d x.
0 dx2 0 0

Thus,
{ 1 ( 2 )
d u
. − 2 + μ(x)u − f ϕ d x = 0.
0 dx

So,
/ \
d 2u
. − 2 + μ(x)u − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx

It follows that
d 2u
. − + μ(x)u − f = 0
dx2

on .D , (]0, 1[). Since . f ∈ L 2 (]0, 1[, R), then

d 2u
. − + μ(x)u = f
dx2

on . L 2 (]0, 1[, R). It thus appears that

d 2u
. − + μ(x)u = f
dx2

@seismicisolation
@seismicisolation
9.2 Corrected 661

almost everywhere in .]0, 1[. Hence, (9.1.5). Moreover, .u ∈ V , then

u(0) = 0.
.

It remains to show that .u , (1) = −τ u(1). In fact, let .v ∈ V , using Green’s formula
for the variational problem, we find
{ 1 2 { 1 { 1
d u , 1
.− v d x + [u v] 0 + μuv d x + τ u(1)v(1) = f v d x.
2
0 dx 0 0

Hence,
{ 1 ( 2 )
d u
. − 2 + μu − f v d x + (u , (1) + τ u(1))v(1) = 0.
0 dx

On the other hand,

d 2u
. − + μu = f,
dx2

so .(u , (1) + τ u(1))v(1) = 0 for all .v ∈ V . Further, if we choose a .v ∈ V such that

,
.v(1) / = 0, then .u (1) = −τ u(1). Thus, the equivalence between the strong problem

and the variational formulation.

4. Since
d 2u
. − + μu = f,
dx2
then
d 2u
. − = f − μu.
dx2

Furthermore, .u, .μ(·) ∈ C 1 ([0, 1], R), and . f ∈ H 1 (]0, 1[), so

d 2u
. − ∈ H 1 (]0, 1[).
dx2

Hence, .u ∈ H 3 (]0, 1[). Further, .3 > 21 + 2, so according to Theorem 3.4.9,

. H (]0, 1[) continuously injects into .C ([0, 1], R). Hence, .u ∈ C ([0, 1], R). Thus,
3 2 2

the obtained solution .u is solution in the usual sense.

Solution: Exercise 9.31

1. Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(0) = v(1)} be a test function. Then,
{ 1 2 { 1 { 1
d u
.− v d x + uv d x = f v d x.
2
0 dx 0 0

@seismicisolation
@seismicisolation
662 9 Exercises with Solutions

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv
. d x − [u , v]10 + uv d x = f v d x.
0 dx dx 0 0

Which proves
{ 1 { 1 { 1
du dv , ,
. d x − u (1)v(1) + u (0)v(0) + uv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. dx + uv d x = f v d x. (9.2.6)
0 dx dx 0 0

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + uϕ d x = f ϕ d x.
0 dx dx 0 0

Then,
{ 1 { 1 { 1
d 2u
. − ϕ dx + uϕ d x = f ϕ d x.
0 dx2 0 0

Thus,
{ 1 ( 2 )
d u
. − 2 + u − f ϕ d x = 0.
0 dx

So,
/ \
d 2u
. − + u − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx2

It follows that
d 2u
. − +u− f =0
dx2

on .D , (]0, 1[). Moreover, . f ∈ L 2 (]0, 1[, R), so

d 2u
. − +u = f
dx2

@seismicisolation
@seismicisolation
9.2 Corrected 663

on . L 2 (]0, 1[, R). It thus appears that

d 2u
. − +u = f
dx2
almost everywhere in .]0, 1[. Moreover, .u ∈ V , so

u(0) = u(1).
.

It remains to show that .u , (0) = u , (1). In fact, let .v ∈ V , using Green’s formula to
the variational problem, we find
{ 1 2 { 1 { 1
d u , 1
.− v d x + [u v]0 + uv d x = f v d x.
2
0 dx 0 0

Moreover,
d 2u
. − + u = f,
dx2
hence
u , (1)v(1) − u , (0)v(0) = 0
.

for all .v ∈ V . Thus,

v(0)(u , (1) − u , (0)) = 0.
.

If we choose a .v ∈ V such that

v(0) /= 0,
.

then
u , (1) = u , (0).
.

Thus, the equivalence between the strong problem and the variational formulation.
We consider the following symmetrical bilinear form

.a : V × V −→ R
{ {
1
∂u ∂v 1
(u, v) −→ a(u, v) = dx + uv d x
0 ∂x ∂x 0

and, the following linear form

. L : V −→ R
{ 1
v −→ L(v) = f ϕ d x.
0

@seismicisolation
@seismicisolation
664 9 Exercises with Solutions

. V is a Hilbert space. In fact, consider the mapping

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(0) − v(1).

ϕ is linear.
.
Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,
.

|ϕ(v)| = |v(0) − v(1)|

.

≤ 2||v||∞ .

Further, . H 1 (]0, 1[) continuously injects into .C 0 ([0, 1], R), so

||v||∞ ≤ c||v||1,]0,1[ .
.

Thus,

|ϕ(v)| ≤ 2c||v||1,]0,1[ .
.

Hence, .ϕ is continuous. Further, .V = ϕ −1 ({0}). .{0} is closed of .R, the recipro-

cal range of closed by a continuous map is closed. So, .V is a closed subspace of
. H (]0, 1[). Thus, . V is a Hilbert space.
1

Continuity of a(·, ·): Let .u, .v ∈ V . Then,

.

|a(u, v)| = |(u, v)1,]0,1[ |

.

≤ ||u||1,]0,1[ ||v||1,]0,1[ .

Thus, .a(·, ·) is continuous.

Coercivity of a(·, ·): Let .u ∈ V . Then,
.

. a(u, u) = (u, u)1,]0,1[

= ||u||21,]0,1[
1
≥ ||u||21,]0,1[ .
2
Thus, .a(·, ·) is coercive.

@seismicisolation
@seismicisolation
9.2 Corrected 665

Continuity of L(·): Let .v ∈ V . Then,

.

{ 1
.|L(v)| ≤ | f | |v| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (9.2.6) admits
one solution and only one .u ∈ V .
2. Since
.
d 2u
. − + u = f,
dx2
then
d 2u
. − = f − u.
dx2

Moreover, .u, . f ∈ H 1 (]0, 1[), so

d 2u
. − ∈ H 1 (]0, 1[).
dx2

This proves .u ∈ H 3 (]0, 1[). Further, .3 > 21 + 2, so according to Theorem 3.4.9,

. H (]0, 1[) continuously injects into .C ([0, 1], R). Hence, .u ∈ C ([0, 1], R). Thus,
3 2 2

the obtained solution .u is solution in the usual sense.

3. We divide the interval .[0, 1] into . I intervals of length .h. We pose .xi = i h, .0 ≤
i ≤ I . Let .Wh be the space

Wh = {vh ∈ C 0 ([0, 1], R) such that vh |(xi ,xi+1 ) ∈ P1 , vh is fully determined

.
by its values vi = vh (xi ) at points xi 0 ≤ i ≤ I }.

We pose

. Vh = {vh ∈ Wh such that vh (0) = vh (1)}.

A basis of the space .Vh is composed of functions .ϕ j (x) ∈ Vh equal to .1 at .x j and to

zero in all .xi .i /= j. We obtain the hats functions. Thus,
| |
. Vh = {ϕ j , 1 ≤ j ≤ I − 1} {ψ0 },

@seismicisolation
@seismicisolation
666 9 Exercises with Solutions

where .ϕ j are the hats functions and .ψ0 ∈ P1 (xi , xi+1 ) is the function defined by
⎧
⎨ 1 in x0
.ψ0 := ϕ0 + ϕ I := 0 in x1 , . . . , x I −1
⎩
1 in x I .

Thus,
⎧
⎪
⎪
1
⎨ − h x + 1 if x ∈ [x0 , x1 ]
⎪
.ψ0 (x) := 0 if x ∈ [x1 , x I −1 ]
⎪
⎪
⎪ 1
⎩ x + 1 − I if x ∈ [x I −1 , x I ].
h

We consider the approximate problem: Find .u h ∈ Vh solution of .a(u h , vh ) = L(vh )

for all .vh ∈ Vh . This is equivalent to find .u h ∈ Vh solution of

a(u h , ϕ j ) = L(ϕ j )
.

for all . j = 1, . . . , I − 1 and

a(u h , ψ0 ) = L(ψ0 ).
.

Further, .u h ∈ Vh , so .u h is written under the form

I −1
E
u = u 0 ψ0 +
. h u i ϕi .
i=1

Thus,
I −1
E
u a(ψ0 , ϕ j ) +
. 0 u i a(ϕi , ϕ j ) = L(ϕ j )
i=1

for all . j = 1, . . . , I − 1 and

I −1
E
u a(ψ0 , ψ0 ) +
. 0 u i a(ϕi , ψ0 ) = L(ψ0 ).
i=1

We have.a(ϕi , ϕ j ) = 0 if.|i − j| ≥ 2. It suffices to calculate.a(ϕi , ϕi ).i = 1, . . . , I −

1, .a(ϕi , ϕi+1 ) .i = 1, . . . , I − 2, .a(ϕi , ψ0 ) .i = 1, . . . , I − 1, and .a(ψ0 , ψ0 ).
Calculation of a(ϕi , ϕi ) i = 1, . . . , I − 1: Let .i ∈ {1, . . . , I − 1}, we have
.

{ 1 ( )2 { 1
dϕi
a(ϕi , ϕi ) =
. dx + ϕi2 d x.
0 dx 0

@seismicisolation
@seismicisolation
9.2 Corrected 667

Hence,
{ xi+1 ( )2 { xi+1
dϕi
a(ϕi , ϕi ) =
. dx + ϕi2 d x.
xi−1 dx xi−1

Further,
⎧
⎪
⎪ 0 if x ∈ [x0 , xi−1 ]
⎪
⎪
⎪ 1
⎨ x +1−i if x ∈ [xi−1 , xi ]
.ϕi (x) := h
⎪
⎪
1
− x +1+i if x ∈ [xi , xi+1 ]
⎪
⎪
⎪
⎩ h
0 if x ∈ [xi+1 , x I ].

We have
{ xi+1 ( )2 { xi ( )2 { xi+1 ( )2
dϕi dϕi dϕi
. dx = dx + dx
xi−1 dx xi−1 dx xi dx
2
= .
h
On the other hand,
{ xi+1 { xi { xi+1
. ϕi d x =
2
ϕi2 dx + ϕi2 d x
xi−1 xi−1 xi
{ xi ( )2 { xi+1 ( )2
1 1
= x +1−i dx + − x +1+i dx
xi−1 h xi h
[ ( )3 ]xi [ ( )3 ]xi+1
h 1 h 1
= x +1−i − − x +1+i
3 h 3 h
xi−1 xi
2h
= .
3
Thus, for all .i = 1, . . . , I − 1,

2 2h
a(ϕi , ϕi ) =
. + .
h 3

Calculation of a(ϕi , ϕi+1 ) i = 1, . . . , I − 2: Let .i ∈ {1, . . . , I − 2}, we have

.

{ 1 { 1
dϕi dϕi+1
.a(ϕi , ϕi+1 ) = dx + ϕi ϕi+1 d x.
0 dx dx 0

@seismicisolation
@seismicisolation
668 9 Exercises with Solutions

Then,
{ xi+1 { xi+1
dϕi dϕi+1
a(ϕi , ϕi+1 ) =
. dx + ϕi ϕi+1 d x
xi dx dx xi
{ xi+1 { xi+1 ( )( )
11 1 1
=− dx + − x +1+i x − i dx
xi hh xi h h
=
{ xi+1 ( )2 { xi+1 ( )
1 1 1
=− + − x −i dx + x − i dx
h xi h xi h
[( )3 ]xi+1 [( )2 ]xi+1
1 h 1 h 1
=− − x −i + x −i
h 3 h 2 h
xi xi
1 h h
=− − +
h 3 2
1 h
=− + .
h 6

Calculation of a(ϕi , ψ0 ) i = 1, . . . , I − 1: .a(ϕi , ψ0 ) = 0 for .i = 2, . . . , I − 2.

.

It remains to calculate .a(ϕ1 , ψ0 ) and .a(ϕ I −1 , ψ0 ). We have

a(ϕ1 , ψ0 ) = a(ϕ1 , ϕ0 )
.

1 h
=− + ,
h 6
with { 1
− x + 1 if x ∈ [x0 , x1 ]
ϕ (x) :=
. 0 h
0 if x ∈ [x1 , x I ].

Likewise,

a(ϕ I −1 , ψ0 ) = a(ϕ I −1 , ϕ I )
.

1 h
=− +
h 6
with {
0 if x ∈ [x0 , x I −1 ]
ϕ I (x) :=
. 1
h
x + 1 − I if x ∈ [x I −1 , x I ].

Calculation of a(ψ0 , ψ0 ):
.

.a(ψ0 , ψ0 ) = a(ϕ0 , ϕ0 ) + a(ϕ I , ϕ I )

2 2h
= + .
h 3

@seismicisolation
@seismicisolation
9.2 Corrected 669

Thus,
⎛ { 1 ⎞
⎛ ⎞⎛ ⎞ ⎜ 0 f ψ 0 d x ⎟
2 + 2h − 1 + h 0 ··· 0 − h1 + h6 u0 ⎜ { 1 ⎟
⎜ h 1 3h 2 h 2h6 1 ⎟⎜ ⎜ ⎟
⎜−h + 6 h + 3 −h + 6 0 h ··· 0 ⎟ ⎜ u1 ⎟ ⎜ f ϕ d x ⎟
⎜ ⎟⎜ . ⎟ ⎟
⎜
⎜
1 ⎟
⎟
⎜ .. .. .. .. .. ⎟⎜ . ⎟ ⎜ 0
⎟
⎜ 0 . . . . . ⎟⎜ . ⎟ ⎜ .. ⎟
⎜ ⎟⎜ ⎟ ⎜ . ⎟.
.⎜ . . . . . ⎟⎜ . ⎟ = ⎜ ⎟
⎜ .. .. .. .. .. ⎟ ⎜ .. ⎟ ⎜ . ⎟
⎜ 0 ⎟⎜ ⎟ ⎜ .
. ⎟
⎜ .. .. .. ⎟⎜ . ⎟ ⎜ ⎟
⎜ ⎟⎝ . ⎠ ⎜ .. ⎟
⎝ 0 . . . − h1 + h6 ⎠ . ⎜ ⎟
1 h ⎜{ . ⎟
−h + 6 0 ··· 0 − h + 6 h2 + 2h
1 h
3
u I −1 ⎝ 1 ⎠
f ϕ I −1 d x
0

Solution: Exercise 9.32

1. Let .v ∈ V := {v ∈ H 1 (0, 1) such that v(1) = 0} be a test function. Then,
{ 1 2 { 1 { 1
d u
.− v d x + a uv d x = f v d x.
2
0 dx 0 0

Using Theorem 3.4.4, we deduce

{ 1 { 1 { 1
du dv
. d x − [u , v]10 + a uv d x = f v d x.
0 dx dx 0 0

So,
{ 1 { 1 { 1
du dv
. d x − u , (1)v(1) + u , (0)v(0) + a uv d x = f v d x.
0 dx dx 0 0

Hence, the variational problem is: Find .u ∈ V such that for all .v ∈ V , we have
{ 1 { 1 { 1
du dv
. dx + a uv d x = f v d x. (9.2.7)
0 dx dx 0 0

Let .ϕ ∈ D(]0, 1[), we have

{ 1 { 1 { 1
du dϕ
. dx + a uϕ d x = f ϕ d x.
0 dx dx 0 0

Hence,
{ 1 { 1 { 1
d 2u
. − ϕ dx + a uϕ d x = f ϕ d x.
0 dx2 0 0

@seismicisolation
@seismicisolation
670 9 Exercises with Solutions

Thus,
{ 1 ( )
d 2u
. − + au − f ϕ d x = 0.
0 dx2

So,
/ \
d 2u
. − 2 + au − f, ϕ = 0 for all ϕ ∈ D(]0, 1[).
dx

It follows that
d 2u
. − + au − f = 0
dx2

on .D , (]0, 1[). Further, . f ∈ L 2 (]0, 1[, R), so

d 2u
. − + au = f
dx2

on . L 2 (]0, 1[, R). It thus appears that

d 2u
. − + au = f
dx2
almost everywhere in .]0, 1[. Moreover, .u ∈ V , so

. u(1) = 0.

It remains to show that

u , (0) = 0.
.

In fact, let .v ∈ V , using Green’s formula to the variational problem, we find

{ 1 2 { 1 { 1
d u , 1
.− v d x + [u v] 0 + a uv d x = f v d x.
2
0 dx 0 0

Since
d 2u
. − + au = f,
dx2

then .u , (1)v(1) − u , (0)v(0) = 0 for all .v ∈ V . So, .v(0)u , (0) = 0. If we choose a

,
.v ∈ V such that .v(0) / = 0, then .u (0) = 0. Thus, the equivalence between the strong
problem and the variational formulation.

@seismicisolation
@seismicisolation
9.2 Corrected 671

We consider the following symmetrical bilinear form

.a : V × V −→ R
{ {
1
∂u ∂v 1
(u, v) −→ a(u, v) = dx + a uv d x
0 ∂x ∂x 0

and, the following linear form

. L : V −→ R
{ 1
v −→ L(v) = f ϕ d x.
0

. V is a Hilbert space. In fact, consider the mapping

ϕ : H 1 (]0, 1[) −→ R
.

v −→ ϕ(v) = v(1).

ϕ is linear.
.

Continuity of ϕ(·): Let .v ∈ H 1 (]0, 1[). Then,

.

|ϕ(v)| = |v(1)|
.

≤ ||v||∞ .

Since . H 1 (]0, 1[) continuously injects into .C 0 ([0, 1], R), then

. ||v||∞ ≤ c||v||1,]0,1[ .

Thus,

|ϕ(v)| ≤ c||v||1,]0,1[ .
.

Hence, .ϕ is continuous. Gold, .V = ϕ −1 ({0}). .{0} is closed of .R, the reciprocal range
of closed by a continuous map is closed. Thus, .V is a closed subspace of . H 1 (]0, 1[).
So, .V is a Hilbert space.
Continuity of a(·, ·): Let .u, .v ∈ V . Then,
.

| || |
{ { 1
| du | | dv |
1
.|a(u, v)| ≤ | | | |
|dx | |dx | dx + a |u||v| d x
0 0
≤ (1 + a)||u||1,]0,1[ ||v||1,]0,1[ .

Thus, .a(·, ·) is continuous.

@seismicisolation
@seismicisolation
672 9 Exercises with Solutions

Coercivity of a(·, ·): Let .u ∈ V . Then,

.

{ 1 ( )2 { 1
du
.a(u, u) = dx + a u2 d x
0 dx 0
≥ min(1, a)||u||21,]0,1[ .

Thus, .a(·, ·) is coercive.

Continuity of L(·): Let .v ∈ V . Then,
.

{ 1
.|L(v)| ≤ | f | |ϕ| d x
0
≤ || f ||0,]0,1[ ||v||0,]0,1[
≤ || f ||0,]0,1[ ||v||1,]0,1[ .

Thus, . L(·) is continuous.

According to the Lax-Milgram theorem (Theorem 2.17.1), the problem (9.2.7) admits
one solution and only one .u ∈ V .
2. We divide the interval .[0, 1] into . I intervals of length .h. We pose .xi = i h, .0 ≤
.

i ≤ I . Let .Wh be the space

. Wh = {vh ∈ C 0 ([0, 1], R) such that vh |(xi ,xi+1 ) ∈ P1 , vh is fully determined
.by its values vi = vh (x i ) at points x i 0 ≤ i ≤ I }.

We pose
. Vh = {vh ∈ Wh such that vh (1) = 0}.

3. A basis of the space .Vh is composed of functions

ϕ j (x) ∈ Vh
.

equal to .1 at .x j and to zero in all .xi .i /= j. We obtain the hats functions. Thus,
Vh = {ϕ j , 0 ≤ j ≤ I − 1}. We consider the approximate problem: Find .u h ∈ Vh
.

solution of .a(u h , vh ) = L(vh ) for all .vh ∈ Vh . This is equivalent to find .u h ∈ Vh

solution of

a(u h , ϕ j ) = L(ϕ j )
.

for all . j = 0, . . . , I − 1. Gold,

I −1
E
. h u = u i ϕi .
i=0

@seismicisolation
@seismicisolation
9.2 Corrected 673

Thus,
I −1
E
. u i a(ϕi , ϕ j ) = L(ϕ j )
i=0

for all. j = 0, . . . , I − 1. We have.a(ϕi , ϕ j ) = 0 if.|i − j| ≥ 2. It suffices to calculate

a(ϕi , ϕi ) .i = 0, . . . , I − 1, .a(ϕi , ϕi+1 ) .i = 0, . . . , I − 2.
.

Calculation of a(ϕi , ϕi ) i = 0, . . . , I − 1: We start with the calculation of .a(ϕ0 , ϕ0 ).

.
We have
{ 1 ( )2 { 1
dϕ0
a(ϕ0 , ϕ0 ) =
. dx + a ϕ02 d x.
0 dx 0

Hence,
{ x1 ( )2 { x1
dϕ0
a(ϕ0 , ϕ0 ) =
. dx + a ϕ02 d x
x0 dx x0
1 h
= +a .
h 3
Let .i ∈ {1, . . . , I − 1}, we have
{ ) ({ 1
dϕi 2 1
.a(ϕi , ϕi ) = dx + a ϕi2 d x
0 dx 0
2 2h
= +a .
h 3

Calculation of a(ϕi , ϕi+1 ) i = 0, . . . , I − 2: Let .i ∈ {0, . . . , I − 2}, we have

.

{ 1 { 1
dϕi dϕi+1
.a(ϕi , ϕi+1 ) = dx + a ϕi ϕi+1 d x
0 dx dx 0
1 h
= − +a .
h 6
Thus,
⎛ { 1 ⎞
⎛ ⎞⎛ ⎞
1
h+ a h3 − h1 + a h6 0 0 ··· 0 u0 ⎜ f ϕ0 d x ⎟
⎜ − 1 + a h 2 + a 2h − 1 + a h ··· ⎟ ⎜ u ⎟ ⎜ {0 1 ⎟
⎜ h 6 h 3 h 6 0 0 ⎟⎜ 1 ⎟ ⎜ ⎟
⎜ .. .. .. .. .. ⎟⎜ . ⎟ ⎜ f ϕ1 d x ⎟
⎜ ⎟⎜ . ⎟ ⎜ ⎟
⎜ 0 . . . . . ⎟⎜ . ⎟ ⎜ 0 ⎟
⎜ ⎟⎜ ⎜ . ⎟
.⎜ .. . . .. .. ⎟⎜ . ⎟ ⎟ =⎜ .. ⎟.
⎜ . . . . . . . .
⎟⎜ . ⎟ ⎜ ⎟
⎜ 0 ⎟⎜ ⎟ ⎜ ⎟
⎜ .. ⎟⎜ . ⎟ ⎜ .. ⎟
⎜ .. .. .. ⎟⎝ . ⎠ ⎜ . ⎟
⎝ . . . . − h1 + a h6 ⎠ . ⎜{ ⎟
⎝ 1 ⎠
0 ··· ··· 0 − h + a 6 h + a 2h
1 h 2
3
u I −1 f ϕ I −1 d x
0

@seismicisolation
@seismicisolation
674 9 Exercises with Solutions

Solution: Exercise 9.33

1. .Vh is a vector subspace of . X of finite dimension. Hence, .Vh is closed. Thus, .Vh
is a Hilbert space. We have . L(·) ∈ X , . According to Riesz Fréchet’s representation
theorem (Theorem 2.14.1), there is . f h ∈ Vh unique such that . L(vh ) = ( f h , vh ) for
all .vh ∈ Vh . Let .u h ∈ Vh . The mapping .a(u h , ·) ∈ X , . According to Riesz Fréchet’s
representation theorem (Theorem 2.14.1), there is . Ah u h ∈ Vh unique such that
.a(u h , vh ) = (A h u h , vh ) for all .vh ∈ Vh . Comes back to find .u h ∈ Vh such that
. A h u h = f h . . A h is a linear mapping from . Vh into . Vh who is continuous. Let us

show that . Ah is injective. In fact, we have

. inf sup a(vh , wh ) ≥ βh .

||vh ||=1 ||wh ||=1

Hence,

. inf sup (Ah vh , wh ) ≥ βh .

||vh ||=1 ||wh ||=1

for all .wh ∈ Vh such that .||wh || = 1, we have

||Ah wh || ≥ inf ||Ah vh ||.

.
||vh ||=1

Gold,

. inf ||Ah vh || ≥ βh .
||vh ||=1

We have, for all .wh /= 0,

. ||Ah wh || ≥ βh ||wh ||.

Thus, . Ah is injective and so . Ah is bijective.

2. We have

.||Ah (vh − u h )|| = sup (Ah (vh − u h ), wh ).

||wh ||=1

Hence,

||Ah (vh − u h )|| = sup a(vh − u h , wh ).

.
||wh ||=1

Since

a(vh − u h , wh ) = a(vh , wh ) − a(u h , wh ),

.

@seismicisolation
@seismicisolation
9.2 Corrected 675

then

a(vh − u h , wh ) = a(vh , wh ) − L(wh ).

.

Hence,

a(vh − u h , wh ) = a(vh , wh ) − a(u, wh ).

.

Thus,

a(vh − u h , wh ) = a(vh − u, wh ).
.

It follows that

a(vh − u h , wh ) ≤ M||vh − u||||wh ||.

.

On the other hand,

||Ah (vh − u h )|| ≥ βh ||vh − u h ||.

.

Hence,

β ||vh − u h || ≤ M||vh − u||.

. h

So,
M
||vh − u h || ≤
. ||u − vh ||.
βh

Thus, for all .vh ∈ Vh , we have

( )
M
.||u − u h || ≤ 1+ ||u − vh ||.
βh

Thus proves the result.

Solution: Exercise
{ 9.34 }
Let .v ∈ V := v ∈ H 1 (Ω) such that v|∂Ω\[A9 A10 ] = 0 , be a test function. We have
{ {
. −Δu v d x = f v d x.
Ω Ω

Using Theorem 3.4.4, we deduce

{ { {
∂u
. ∇u · ∇v d x − v dσ = f v dx
Ω L ∂ν Ω

@seismicisolation
@seismicisolation
676 9 Exercises with Solutions

with

∂u E ∂u n
. = νi .
∂ν i=1
∂ xi |L

Thus,

{ { { {
∂u ∂u
. ∇u · ∇v d x − v dσ − v dσ = f v d x.
Ω ∂Ω\[A9 A10 ] ∂ν [A9 A10 ] ∂ν Ω

It follows that for all .v ∈ V

{ {
. ∇u · ∇v d x = f v d x.
Ω Ω

We pose
{
a(u, v) =
. ∇u · ∇v d x
Ω

and
{
. L(v) = f v d x.
Ω

We pose
{ }
. Vh = vh ∈ C 0 (Ω, R) such that vh |T ∈ P1 , vh = 0 in 1, . . . , 8 .

We have

. dim Vh = 4.

A basis of.Vh is composed of functions.ϕ9 ,.ϕ10 ,.ϕ11 , and.ϕ12 . Consider the problem:
Find .u h ∈ Vh such that

.a(u h , v) = L(v)

for all .v ∈ Vh . This is equivalent to find .u h ∈ Vh such that

. a(u h , ϕi ) = L(ϕi )

@seismicisolation
@seismicisolation
9.2 Corrected 677

for all .i = 9, 10, 11, 12. Let’s seek .a(u h , ϕ9 ).

12 11

T2

T3 T1

8 9 10

{
a(u h , ϕ9 ) =
. ∇u h ∇ϕ9 d xd y
suppϕ9
{ ( )
∂u h ∂ϕ9 ∂u h ∂ϕ9
= + d xd y.
suppϕ9 ∂ x ∂ x ∂y ∂y

For T1 :
.
∂u h u 10 − u 9
. =
∂x h

∂ϕ9 ϕ9 (10) − ϕ9 (9) 1

. = =−
∂x h h

∂ϕ9 ϕ9 (11) − ϕ9 (10)

. = =0
∂y h

∂u h u 11 − u 10
. = .
∂y h

For T2 :
.

∂u h u 11 − u 12
. =
∂x h

∂ϕ9
. =0
∂x

∂ϕ9 1
. =−
∂y h

@seismicisolation
@seismicisolation
678 9 Exercises with Solutions

∂u h u 12 − u 9
. = .
∂y h

For T3 :
.

∂u h u9 − u8
. =
∂x h

∂ϕ9 1
. =
∂x h

∂ϕ9 1
. =−
∂y h

∂u h u 12 − u 9
. = .
∂y h

Thus,
[ ]
h2 u 9 − u 10 u 9 − u 12 u9 − u8 u 9 − u 12
.a(u h , ϕ9 ) = + + + .
2 h2 h2 h2 h2

It follows that
1
a(u h , ϕ9 ) =
. (4u 9 − u 8 − u 10 − 2u 12 ).
2
Further, .u 8 = 0 because .u h |∂Ω\[A9 A10 ] = 0. Thus,
{
1
.a(u h , ϕ9 ) = 2u 9 − u 10 − u 12 = f ϕ9 d xd y.
2 Ω

Likewise, we find others

{
1
.a(u h , ϕ10 ) = 2u 10 − u 11 − u 9 = f ϕ10 d xd y
2 Ω

{
a(u h , ϕ11 ) = 4u 11 − u 12 − u 10 =
. f ϕ11 d xd y
Ω

{
. a(u h , ϕ12 ) = 4u 12 − u 11 − u 9 = f ϕ12 d xd y.
Ω

@seismicisolation
@seismicisolation
9.2 Corrected 679

Hence, we obtain the following linear system:

⎧ {
⎪
⎪ 2u 9 − 2 u 10 − u 12 =
1
f ϕ9 d xd y
⎪
⎪
⎪
⎪ {Ω
⎪
⎪
⎪
⎨ 2u 10 − u 11 − 21 u 9 = f ϕ10 d xd y
. {Ω
⎪
⎪ 4u 11 − u 12 − u 10 = f ϕ11 d xd y
⎪
⎪
⎪
⎪ {Ω
⎪
⎪
⎪
⎩ 4u 12 − u 11 − u 9 = f ϕ12 d xd y.
Ω

Again,
⎛{ ⎞
f ϕ9 d xd y ⎟
⎛ ⎞⎛ ⎞ ⎜ ⎜{Ω ⎟
2 − 21 0 −1 u9 ⎜ ⎟
⎜−1 −1 0 ⎟ ⎜ ⎟ ⎜ f ϕ10 d xd y ⎟
.⎜ 2
2 ⎟ ⎜ u 10 ⎟ = ⎜
⎜ {Ω
⎟
⎟.
⎝ 0 −1 −1 ⎠ ⎝ u 11 ⎠ ⎜
4
⎜ f ϕ11 d xd y ⎟
⎟
−1 0 −1 4 u 12 ⎜ {Ω ⎟
⎝ ⎠
f ϕ12 d xd y
Ω

To solve the last system we can see the book of BenHamadou and Jeribi [2].
Solution: Exercise 9.35
1. To answer this question we will look for the basic elements. The first function of
basis is .ϕ1 (ξ, η) and such that .ϕ1 (-
A1 ) = 1 and .ϕ1 = 0 at other points. Let

ϕ (ξ, η) = a + bξ + cη + dξ η(1 − ξ − η).

. 1

We have .ϕ1 (-
A2 ) = 0. Hence, .a + c = 0. .ϕ1 (-
A3 ) = 0 so .a = c = 0. .ϕ1 (-
A1 ) = 1
thus
.b = 1.

ϕ (-
. 1A4 ) = 0 so ( )
1 d 1 1
. + 1− − = 0.
3 9 3 3

Thus,

ϕ (ξ, η) = ξ − 9ξ η(1 − ξ − η).

. 1

By symmetry

@seismicisolation
@seismicisolation
680 9 Exercises with Solutions

ϕ (ξ, η) = η − 9ξ η(1 − ξ − η).

. 2

Likewise, for the other two

ϕ (ξ, η) = (1 − ξ − η) − 9ξ η(1 − ξ − η)
. 3

and

ϕ (ξ, η) = 27ξ η(1 − ξ − η).

. 4

Hence, any polynomial . p of . P is written in a unique way

. p(ξ, η) = αϕ1 (ξ, η) + βϕ2 (ξ, η) + γ ϕ3 (ξ, η) + δϕ4 (ξ, η).

2. We have . p|[2,3] = βϕ2 (ξ, η) + γ ϕ3 (ξ, η). So, we have continuity

N with the neigh-
boring triangle. Hence, we can construct a space .Vh ⊂ C 0 (Ω, R) H01 (Ω), .- r-p=- p
for all . -
p ∈ P1 ⊂ P. So, according to the course an increase of the error is given of
the order .o(h), i.e., .||u − u h || ≤ ch.
Solution: Exercise 9.36
1. .(i) .Continuity of a(·, ·): Let .u, .v ∈ V . Then,

2 {
E | || |
| ∂u | | ∂v |
|a(u, v)| ≤ |ai j | || || | d x1 d x2
∂ xi | | ∂ x j |
.
i, j=1 Ω

E 2 { | || |
| ∂u | | ∂v |
≤ ||ai j ||∞ | || |
| ∂ x | | ∂ x | d x1 d x2
i, j=1 Ω i j

2 ||
E || || ||
|| ∂u || || ∂v ||
≤ ||ai j ||∞ || || || ||
|| ∂ x || || ∂ x ||
i, j=1 i 0,Ω j 0,Ω

≤ 4||ai j ||∞ |u|1,Ω |v|1,Ω .

Thus, .a(·, ·) is continuous.

@seismicisolation
@seismicisolation
9.2 Corrected 681

Coercivity of a(·, ·): Let .u ∈ V . Then,

.

2 {
E | |
| ∂u |2
a(u, u) =
. ai j || | d x1 d x2
|
i, j=1 Ω ∂ x i
({ | | { | | )
| ∂u |2 | ∂u |2
≥α | | | |
| | d x1 d x2 + | | d x1 d x2
Ω ∂ x1 Ω ∂ x2

= α|u|21,Ω .

Thus, .a(·, ·) is coercive.

(ii) .Continuity of L(·): Let .v ∈ V . Then,
.

{
|L(v)| ≤
. | f | |v| d x1 d x2
Ω
≤ || f ||0,Ω ||v||0,Ω .

Further, according to the Poincaré inequality (Theorem 3.2.2),.||v||0,Ω ≤ C(Ω)|v|1,Ω .

Hence,
.|L(v)| ≤ C(Ω)|| f ||0,Ω |v|1,Ω .

Thus, . L(·) is continuous.

(iii) According to the Lax-Milgram theorem (Theorem 2.17.1), the problem: Find
.
u ∈ V such that .a(u, v) = L(v) for all .v ∈ V admits a unique solution .u ∈ V .
.

2. .(i) Immediate consequence of Lemma

.
N 3.3.6.
(ii) Let .T be a triangle such that .T ∂Ω = [Ai A j ]. Hence, .[Ai A j ] is beside of
.
triangle .T who belongs to the border .∂Ω. For .vh ∈ Vh , we have .vh |T ∈ P1 and

v (Ai ) = vh (A j ) = 0.
. h

However, a polynomial of degree 1 vanishes at two points, it then vanishes over the
entire line containing these two points. Therefore, .vh |[Ai A j ] = 0. Thus, .Vh ⊂ H01 (Ω).
.(iii) .dim Vh is equal to the number of vertices triangles that do not belong to the
border .∂Ω. Hence, .Vh is a subspace of . H01 (Ω) of finite dimension. Hence, .Vh is
closed. Thus, .Vh is a Hilbert space. We have .a(·, ·) is a continuous bilinear form and
coercive on . H01 (Ω), so .a(·, ·) is a continuous bilinear form and coercive on .Vh and
. L(·) is a continuous linear form on . H0 (Ω), so . L(·) is a continuous linear form on
1

. Vh . Hence, according to the Lax-Milgram theorem (Theorem 2.17.1), the problem:

to find .u h ∈ Vh such that .a(u h , vh ) = L(vh ) for all .vh ∈ Vh admits one solution and
only one .u h ∈ Vh .
3. See Sect. 4.6.9.

@seismicisolation
@seismicisolation
682 9 Exercises with Solutions

Solution: Exercise 9.37

1. The functions of basis are given by
x
ϕ (x, y) =
. 1 (x + h)(y + h),
4h 3

1
ϕ (x, y) = −
. 2 x(x − h)(y + h),
4h 3

1
. 3ϕ (x, y) = x(x − h)(y − h),
4h 3

1
ϕ (x, y) = −
. 4 x(x + h)(y − h),
4h 3

1
. 5 ϕ (x, y) = (h 2 − x 2 )(y + h),
2h 3

1
ϕ (x, y) = −
. 6 (h 2 − x 2 )(y − h).
2h 3
Hence,
. p(x, y) = p(A1 )ϕ1 (x, y) + p(A2 )ϕ2 (x, y) + p(A3 )ϕ3 (x, y)

. + p(A4 )ϕ4 (x, y) + p(A5 )ϕ5 (x, y) + p(A6 )ϕ6 (x, y).

2. .(i) We have
{
4h 2
||u
. − rh u||20,R = (-
u −-
r-
u )2 dξ dη
-
R:=[−1,1]×[−1,1] 4
{
=h 2
(-
u −-
r-
u )2 dξ dη.
-
R

Consider the continuous linear mapping .- - −→ L 2 ( R,

r : H 2 ( R) - R). .-
r-
p=- p for all
.-
p ∈ P1 ⊂ P. In fact, if. -
p = 1, then .-
r-
p = 1. If. -
p = ξ , then .-
r-
p = ξ and if. -
p = η, then
.-
r-
p = η. Then, according to Theorem 3.4.12, we have

.||-
u −-
r-
u ||0, R- ≤ c|-
u |2, R- ≤ ch|u|2,R .

Thus,
||u − rh u||0,R ≤ ch 2 |u|2,R .
.

@seismicisolation
@seismicisolation
9.2 Corrected 683

(ii) We have
.

|| ||2 { ( )2
|| d || 1 d
|| ||
.
|| d x (u − rh u)|| =
- h2 dξ
(-
u −- r-
u ) h 2 dξ dη
0,R R
{ ( )2
d
= (-
u −- r-
u ) dξ dη
- dξ
R
≤ |-
u −-
r-
u |21, R-
≤ ||-
u −-
r-
u ||21, R-.

Consider the continuous linear mapping .- - −→ H 1 ( R)

r : H 2 ( R) - and .-
r-
p=-
p for all
.-
p ∈ P1 . Then, according to Theorem 3.4.12, we have

.||-
u −-
r-
u ||1, R- ≤ c|-
u |2, R-.

Hence,
|| ||2
|| d ||
|| ||
.
|| d x (u − rh u)|| ≤ c|-
u |22, R-
0,R
≤ ch 2 |u|22,R .

Thus, || ||
|| d ||
|| ||
.
|| d x (u − rh u)|| ≤ ch|u|2,R .
0,R

(iii) We can do better, i.e.,

.

|| ||
|| d ||
. || (u − r u) || ≤ ch 2 |u|3,R .
|| d x h ||
0,R

In fact,
|| || || ||
|| d || ||
−1 || d
||
|| || u )||
.
|| d x (u − rh u)|| = hh || dξ (-
u −-
r- || - .
0,R 0, R

The idea is that instead of working with. P1 , we interpolate with. P2 . If.-

u = ξ 2 , then
.-
r-
u = ξ and if .-u = η , then .-r-
u = 1. Consider the mapping
2 2

d
. L :-
u −→ (-
u −-
r-
u)
dξ

- into.R and. L(-

is linear continuous from. H 3 ( R) u ) = 0 for all.-
u ∈ P2 . Then, according
to Theorem 3.4.10, we have .|L(- u )| ≤ c|-
u |3, R-. Hence,

@seismicisolation
@seismicisolation
684 9 Exercises with Solutions
{
. |L(- - u |2 -.
u )|2 dξ dη ≤ c mes( R)|- 3, R
-
R

Thus,
|| ||
|| d ||
|| (- u − -
r -
u ) || ≤ c|-u |3, R-
.
|| dξ || -
0, R
≤ ch 2 |u|3,R .

It follows that || ||
|| d ||
|| (u − rh u)|| ≤ ch 2 |u|3,R .
.
|| d x ||
0,R

(iv) We have
.

|| ||2 { ( )2
|| d || 1 d
|| (u − r u) || = (-
u − -
r -
u ) h 2 dξ dη.
.
|| dy h || - h 2 dη
0,R R

Using the same reasoning as .(ii), we obtain

|| ||
|| d ||
|| ||
.
|| dy (u − rh u)|| ≤ ch|u|2,R .
0,R

We can’t do better because if .-

u = η2 , then .-
r-
u = 1, .-
u −-
r-
u = η2 − 1 and
d
. (-
u −-
r-
u ) = 2η /= 0.
dη

(v) We have
.

({ [ ]2 ) 21 ({ [ ]2 ) 21
h 1
d 1 d 2h
. (u − rh u)(x, h) dx = 2 dξ
(-
u −-r-
u )(ξ, 1) dξ
−h dx −1 h 2
({ [ ]2 ) 21
1
d
= h− 2
1
(-
u −-r-
u )(ξ, 1) dξ .
−1 dξ

Consider the linear continuous mapping .- r : H 3 (] − 1, 1[) −→ H 1 (] − 1, 1[). Fur-

thermore, .-r-
p=- p for all . -
p ∈ P2 only in .ξ because .-
r-p=- p for all . -
p ∈ P1 and if
.-
p = ξ 2 , then .-
r-
p = ξ 2 . Hence, according to Theorem 3.4.12, we have
({ [ ]2 ) 21
1
d
. (-
u −-
r-
u )(ξ, 1) dξ ≤ c|-
u (·, 1)|3,(−1,1) .
−1 dξ

@seismicisolation
@seismicisolation
9.2 Corrected 685

Furthermore,
({ [ ]2 ) 21
1
d3
|-
.u (·, 1)|3,(−1,1) = 3
-
u (·, 1) dξ
−1 dξ
|| 3 ||
|| d ||
≤ || u ||
|| dξ 3 - || -
0, R
≤ ||-
u ||3, R-
≤ h −1 h 3 ||u||3,R .

Thus,
({ [ ]2 ) 21
h
d 3
. (u − rh u)(x, h) dx ≤ ch 2 |u|3,R .
−h dx

.(vi) Let .wh ∈ P. We have

{ ( ) ({ ( )2 ) 21
h h
d dwh d
. (u − rh u) (x, h) d x ≤ (u − rh u (x, h) d x
−h dx dx −h dx
({ ( )2 ) 21
h
dwh
. × (x, h) d x .
−h dx

On the other hand,

{ ( )2 { ( )
h
dwh 1
1 d-wh 2
. (x, h) d x = (ξ, 1)hdξ
−h dx −1 h dξ
{ 1 ( )
d-wh 2
= h −1 (ξ, 1) dξ.
−1 dξ

The mapping
({ ( )2 ) 21
1
d-wh
-
wh −→
. (ξ, 1) dξ
−1 dξ

is a norm on . P/P0 and .-

wh −→ |-
wh |1, R- is a norm on . P/P0 . Since

. dim(P/P0 ) < ∞,

then since in a finite-dimensional space all the norms are equivalent, there is .c > 0
such that

@seismicisolation
@seismicisolation
686 9 Exercises with Solutions

({ ( )2 ) 21
1
d-wh
. (ξ, 1) dξ ≤ c|-
wh |1, R-.
−1 dξ

Thus,
({ ( )2 ) 1 ({ ( )2 )1
h 2 h 2
d dwh 1 3
. (u − rh u (x, h) d x (x, h) d x ≤ ch − 2 |-
wh |1, R-h 2 |u|3,R .
−h dx −h dx

Further,

wh |1, R- ≤ hh −1 |wh |1,R ,

|-
.

hence
{ h ( )
d dwh
. (u − rh u) (x, h) d x ≤ ch|u|3,R |wh |1,R .
−h dx dx

Other method: We have

{ h ( ) { 1 ( )
d dwh d d-wh
. (u − rh u) (x, h) d x = hh −2 (-
u −-
r-
u) (ξ, 1) dξ.
−h dx dx −1 dξ dξ

We pose
{ 1 ( )
d d-wh
. u, -
G(- wh ) = (-
u −-
r-
u) (ξ, 1) dξ.
−1 dξ dξ

The mapping.- u −→ G(- u, - - into.R of norm less

wh ) is linear continuous from. H 3 ( R)
wh |1, R- and is identically null for all .-
than or equal to .c|- u ∈ P2 (because if .-
u = ξ 2 , then
.-
r-
u = ξ 2 and so .G(ξ 2 , - wh ) = 0 and if .- u = η2 , then .-
r-u = 1 and so .G(η2 , -
wh ) = 0).
Hence, according to Theorem 3.4.10, we have

. |G(-
u, -
wh )| ≤ c|-
wh |1, R-|-
u |3, R-.

Thus,
{ h ( )
d dwh
. (u − rh u) (x, h) d x ≤ ch|u|3,R |wh |1,R .
−h dx dx

@seismicisolation
@seismicisolation
9.2 Corrected 687

Solution: Exercise 9.38

dw
. Aw = + w. We have
dx
{ 1
2(Aw, w)0,(0,1) = 2
. w2 d x + w2 (1) − w2 (0).
0

Let .v(x) = (1 − x)2 . We, then, have

2 3
.2(Av, v)0,(0,1) = − 1 = − < 0.
5 5
Solution: Exercise 9.39 Either the system
( ) ( ) ( )( ) ( )
0 −1 ∂ u 10 u f
. + = .
−1 0 ∂x v 01 v 0

We have
( ) ( )( )( )
01 1 1 1 1 0 1 1
. B(0) = = .
10 2 1 −1 0 −1 1 −1

The systematic choice for the matrix . M gives us

( )( )( )
1 1 1 10 1 1
. M(0) =
2 1 −1 01 1 −1
( )
10
= .
01

Hence, ( )
−1 1
(B − M)(0) =
.
1 −1

let there be boundary conditions .−u(0) + v(0) = 0.

Solution: Exercise 9.40 1. Consider
( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + = .
−1 0 ∂x v 01 v 0

To express the Neumann conditions in an admissible way, we write

( ) ( )
01 0 −1
. B(0) = , M(0) = , β ≥ 0,
10 1 β
( )
0 2
. (B − M)(0) = ,
0 −β

@seismicisolation
@seismicisolation
688 9 Exercises with Solutions

( )
00
(B + M)(0) =
. ,
2β
( ) ( )
0 −1 0 1
. B(1) = , M(1) = , γ ≥ 0,
−1 0 −1 γ
( )
0 −2
. (B − M)(1) = ,
0 −γ
( )
0 0
. (B + M)(1) = .
−2 γ

We have
( )
0 2
(B + M)(0) =
.
t
0 −β

and
( )
0 −2
. (B + M)(1) =
t
.
0 −γ

Let

. W = {(ϕ̊1 , ϕ2 ) ∈ C 1 ([0, 1], R)|R × C 1 ([0, 1], R) such that ϕ2 (0) = ϕ2 (1) = 0}

and

. V = {v = A∗ ϕ, ϕ ∈ W }.

Let us show that for all .v ∈ V , there exists unique .ϕ ∈ W such that . A∗ ϕ = v. Just
check that . A∗ ϕ = 0, .ϕ ∈ W , leads to .ϕ = 0. Let .ϕ1 and .ϕ2 be such that
dϕ2 dϕ1
. = 0, + ϕ2 = 0.
dx dx

We have .ϕ2 = a, .ϕ1 = −ax + b. Gold, .ϕ2 (0) = ϕ2 (1) = 0, hence .ϕ2 = a =
0 and .ϕ1 = b. So, .ϕ̊1 = 0. Space .V is included in space
{ { 1 }
2 ∗
.(L ) = (v1 , v2 ) ∈ (L ) v1 d x = 0 ,
2 2
such that
0

@seismicisolation
@seismicisolation
9.2 Corrected 689

because for all .v ∈ V , there is .ϕ ∈ W such that

dϕ2 dϕ1
. = v1 , + ϕ2 = v2 .
dx dx
Hence
{ 1
ϕ (1) − ϕ2 (0) =
. 2 v1 d x = 0.
0

The subspace .(L 2 )∗ endowed with the norm induced by .(L 2 )2 is a Hilbert space.
Let .v ∈ V be any element, we associate .ϕ ∈ W to it, unique solution of . A∗ ϕ = v and
we set
{ 1
.v −→ L(v) = f ϕ1 d x,
0

where .ϕ1 is a representative of class .ϕ̊1 . This linear form is well defined because
{ 1
. f dx = 0
0

and .ϕ̊1 ∈ C 1 ([0, 1], R)|R . Let us show that this form is continuous for the norm of
2 ∗
.(L ) . In fact,

.|L(v)| ≤ || f ||0,(0,1) ||ϕ1 ||0,(0,1)

for any representative .ϕ1 of class .ϕ̊1 . So,

.|L(v)| ≤ || f ||0,(0,1) ||ϕ̊1 || L 2|R

|| ||
|| dϕ1 ||
||
≤ c|| f ||0,(0,1) || || .
d x ||0,(0,1)

Gold, || ||
|| dϕ1 ||
|| || ≤ ||v2 ||0,(0,1) + ||ϕ2 ||0,(0,1) .
.
|| d x ||
0,(0,1)

On the other hand,

{ x
ϕ (x) = ϕ2 (0) +
. 2 v1 dt
{ x 0

= v1 dt.
0

@seismicisolation
@seismicisolation
690 9 Exercises with Solutions

Hence,
||ϕ2 ||0,(0,1) ≤ ||v1 ||0,(0,1) .
.

We can deduce

|L(v)| ≤ c|| f ||0,(0,1) (||v1 ||0,(0,1) + ||v2 ||0,(0,1) )

.
√
≤ c 2|| f ||0,(0,1) ||v||(L 2 )2 .

The map.v −→ L(v) from.V into.R is extended according

√ to Hahn Banach’s theorem
into a map . L(·) from .(L 2 )∗ into .R of norm .≤ c 2|| f ||0,(0,1) . By Riesz’s theorem,
there exists .u = (u 1 , u 2 ) ∈ (L 2 )∗ such that

||u||(L 2 )∗ = ||L|| and L(v) = (u, v) L 2 ×L 2 .

.

We therefore have the existence of an element .u ∈ (L 2 )∗ such that

{ 1
. f ϕ1 d x = L(v)
0
= (u, v) L 2 ×L 2
= (u, A∗ ϕ) L 2 ×L 2

for all .ϕ ∈ W . We have

{ 1 ( ) { 1
dϕ2 dϕ1
. u1 + u2 dx = f ϕ1 d x
0 dx dx 0

for all .ϕ ∈ W . Hence,

. Au = f on D , × D , .

So, .u 2 ∈ H 1 (0, 1) and .u 1 ∈ H 2 (0, 1)|R . Applying Green’s formula, it then comes
.u 2 (0) = u 2 (1) = 0.

2. Consider
( ) ( ) ( )( ) ( )
0 −1 ∂ u 00 u f
. + = .
−1 0 ∂x v 01 v 0

To express the boundary conditions of Neumann, in an admissible way, we write

( ) ( )
01 a 1
. B(0) = , M(0) = , a ≥ 0.
10 −1 0
( ) ( )
0 −1 a1 −1
. B(1) = , M(1) = , a1 ≥ 0.
−1 0 1 0

@seismicisolation
@seismicisolation
9.2 Corrected 691

We have
( )
−a 0
(B +t M)(0) = (B − M)(0) =
.
2 0

and
( )
−a1 0
.(B +t M)(1) = (B − M)(1) = .
−2 0

Let

. W = {ϕ = (ϕ1 , ϕ2 ) ∈ C 1 ([0, 1], R) × C 1 ([0, 1], R) such that ϕ1 (0) = ϕ1 (1) = 0}

and

. V = {v = A∗ ϕ, ϕ ∈ W }.

For all .v ∈ V , we have

dϕ2 dϕ1
. = v1 , + ϕ2 = v2 .
dx dx

To any element .v ∈ V , we associate .ϕ ∈ W in a unique way because . A∗ ϕ = 0

if, and only if, . dϕ
dx
2
= 0 and . dϕ
dx
1
+ ϕ2 = 0. Hence .ϕ2 = λ and .ϕ1 = −λx + μ, with
.ϕ1 (0) = ϕ1 (1) = 0. So, .ϕ1 = ϕ2 = 0. Let the function . L(·) defined on . V by

{ 1
. L(v) = f ϕ1 d x.
0

This linear functional is well defined on .V . Let us show that it is continuous in

the sense of the norm . L 2 × L 2 . We have

|L(v)| ≤ || f ||0,(0,1) ||ϕ1 ||0,(0,1) .

.

{ 1 { 1 ( )
dϕ1
. v2 d x = + ϕ2 dx
0 0 dx
{ 1
= ϕ2 d x
0
= ϕ2 (ξ )

and

@seismicisolation
@seismicisolation
692 9 Exercises with Solutions
{ x
ϕ (x) = ϕ2 (ξ ) +
. 2 v1 dt
ξ
{ 1 { x
= v2 d x + v1 dt.
0 ξ

We can deduce
. ||ϕ2 ||0,(0,1) ≤ ||v2 ||0,(0,1) + ||v1 ||0,(0,1) .

On the other hand,

{ x
ϕ (x) =
. 1 (v2 − ϕ2 ) dt.
0

Hence

.||ϕ1 ||0,(0,1) ≤ ||v2 ||0,(0,1) + ||ϕ2 ||0,(0,1)

≤ 2||v2 ||0,(0,1) + ||v1 ||0,(0,1)
√
≤ 5||v|| L 2 ×L 2 .

The function . L(·) is therefore continuous. We extend this functional defined on .V

into a functional . L(·) defined on . L 2 × L 2 with

||L|| = ||L||.
.

Using Riesz’s theorem, we deduce the existence of.u = (u 1 , u 2 ) ∈ L 2 × L 2 such that

(u, v)0,(0,1) = L(v)

.
{ 1
= f ϕ1 d x
0
= (u, A∗ ϕ)0,(0,1)

for all .ϕ ∈ W . If we interpolate this relation in the sense of .D , (]0, 1[), we have

du 2 du 1
. − = f, − + u 2 = 0.
dx dx
N
Hence .u 2 ∈ H 1 (0, 1) and .u 1 ∈ H 2 (0, 1) H01 (0, 1). Green’s formula allows us
to conclude .u 1 (0) = u 1 (1) = 0.
Solution: Exercise 9.41 We have
⎛ ⎞
0 −νx −ν y
. B = ⎝ −ν x 0 0 ⎠,
−ν y 0 0

@seismicisolation
@seismicisolation
9.2 Corrected 693

where .νx and .ν y are the components of the exterior normal .ν to .∂Ω. We choose
⎛ ⎞
0 νx ν y
. M = ⎝ −ν x 0 0 ⎠ .
−ν y 0 0

Hence,
⎛ ⎞
0 −2νx −2ν y
.(B − M) = ⎝ 0 0 0 ⎠.
0 0 0

.(B − M)u = 0 is equivalent to .vν x + wν y = 0, which is the homogeneous

Neumann condition on .∂Ω. On the other hand, we check that

. K er (B + M) = {(u, v, w) ∈ R3 such that u = 0}.

So,

. K er (B − M) + K er (B + M) = R3 .

Solution: Exercise 9.42 We have

∂v ∂w
. − = 0, we multiply by v
∂t ∂x
and
∂w ∂v
. − = 0, we multiply by w.
∂t ∂x
We get

{ 1 { 1
. (v2 + w2 )(x, t) d x − (v2 + w2 )(x, 0) d x − 2(v, w)(1, t) + 2vw(0, t) = 0
0 0

v(0, t) = −αw(0, t), v(1, t) = −βw(1, t).

.

Hence,

{ 1 { t
. (v2 + w2 )(x, t) d x + 2β(w(1, t))2 − 2α(w(0, t))2 = (v2 + w2 )(x, 0) d x.
0 0

@seismicisolation
@seismicisolation
694 9 Exercises with Solutions

Thus, .β ≥ 0 and .α ≤ 0. Friedrichs formalism allows us to write .−av + (1 −

b)w = 0 for .x = 0 with .a ≥ 0 and .|b| ≤ 1. So,
b−1
.v+ w=0
a
and
b−1
. ≤ 0.
α=
a
( ) ( ) ( )
0 −1 a1 b1 −a1 −1 − b1
In.1,. B(1) = ,. M(1) = ,.(B − M)(1) = ,
−1 0 c1 d1 −1 − c1 −d1
( )
a1 −1 + b1
and .(B + M)(1) = . Hence, .a1 d1 − (1 + b1 )(1 + c1 ) = 0,
−1 + c1 d1
.b1 + c1 = 0, .4a1 d1 − (b1 + c1 ) ≥ 0, .a1 + d1 ≥ 0, and .a1 d1 = 1 − b1 . So, .|b1 | ≤ 1,
2 2

.c1 = −b1 , .−a1 v − (1 + b1 )w = 0, .v + w = 0, and .β = a1 ≥ 0. The values

1+b1 1+b1
a1
of .α and .β thus found by the Friedrichs method therefore make it possible to obtain
the increase in energy.

Solution: Exercise 9.43 1. We have

( ) ( ) ( ) ( ) ( )( ) ( )
10 ∂ u a1 0 ∂ u 10 u f
. + + = .
0 1 ∂t v 0 −a2 ∂ x v 01 v 0

We have
( ) ( )
−a1 0 ab
. B(0, t) = and M(0, t) = .
0 a2 cd

Necessarily,.dim K er (B − M)(0, t) ≥ 1 and.dim K er (B + M)(0, t) ≥ 1. Hence

. − a1 a2 + ad − bc = 0, −aa2 + da1 = 0. (9.2.8)

A necessary and sufficient condition for . M +t M to be positive semi-definite is that

.a + d ≥ 0, 4ad − (b + c)2 ≥ 0. (9.2.9)

( )
u
The condition .(B − M)(0, t) = 0 must be equivalent to the condition .u(0) =
v
λv(0), which is written
.(a1 + a)λ + b = 0. (9.2.10)

The relations (9.2.8), (9.2.9) and (9.2.10) allow to write

a2
d=a
. , b = −(a1 + a)λ. (9.2.11)
a1

@seismicisolation
@seismicisolation
9.2 Corrected 695

Hence,
a2
. − a1 a2 + a 2 + (a1 + a)λc = 0.
a1

So,
a2 (a1 − a)
.c= .
λa1

By replacing in the 2nd inequality (9.2.9), we get

( ) ( )( ) ( )
a2 2 a2 a2 a2 2
.a
2
λ −
2
+ 2a1 λ −
2
λ +
2
a + a1 λ −
2 2
≤ 0. (9.2.12)
a1 a1 a1 a1

We must have .a and .d are positive. Hence, .a ≥ 0, according to relations (9.2.11).

We therefore pose the following problem: does there exist a real .a ≥ 0 such that the
inequality (9.2.12) is satisfied. If .λ2 − aa21 = 0, then the relationship (9.2.12) is
satisfied. Otherwise, the relation (9.2.12) is equivalent to the relation

λ2 + a2
a1
.a 2 + 2a1 a + a12 = ϕ(a) ≤ 0.
λ2 − a2
a1

We have
4λ2 a1 a2
Δ, = (
. )2 ≥ 0.
λ2 − aa21

If .λ2 − aa21 > 0, the equation .ϕ(ξ ) = 0 has two strictly negative real roots, so for
all .a ≥ 0 , we have .ϕ(a) > 0.
If .λ2 − aa21 < 0, the equation .ϕ(ξ ) = 0 has two positive real roots .ξ1 and .ξ2 , so
for all .a such that .0 < ξ1 ≤ a ≤ ξ2 , we have .ϕ(a) ≤ 0. We can therefore admissibly
examine the condition .u(0, t) = λv(0, t) if, and only if,
/
a2
.|λ| ≤ .
a1

Similarly, we can express in an admissible way the condition .u(1, t) = μv(1, t)

if, and only if,
/
a1
.|μ| ≤ .
a2

@seismicisolation
@seismicisolation
696 9 Exercises with Solutions

2. Premultiply the system by the matrix

( )
b(x) 0
. .
0 c(x)

We get the system

( ) ( ) ( ) ( ) ( )( ) ( )
b0 ∂ u ba1 0 ∂ u b0 u b f1
. + + = .
0c ∂t v 0 −ca2 ∂ x v 0c v c f2

The system is positive in the sense of Friedrichs if, and only if, the matrix
⎛ ⎞
∂b
⎜ −a 1 + 2b 0 ⎟
.⎝ ∂x ⎠
∂c
0 a2 + 2c
∂x

is positive definite. We choose .b and .c such that .b(x) and .c(x) are strictly positive
for all .x ∈ [0, 1], and
/
a2 c(0)
.|λ| ≤
a1 b(0)

and
/
a1 b(1)
|μ| ≤
.
a2 c(1)

or
b(0) a2 1
. ≤ ,
c(0) a 1 λ2

and
b(1) a2
. ≥ μ2 .
c(1) a1

If we choose .b(x) = 1, then

a1 2 a1 β
c(x) =
. λ (1 − x α ) + x ,
a2 a2 μ2

@seismicisolation
@seismicisolation
9.2 Corrected 697

with .α > 1 and .β > 1. We have good

b(0) 1
. =
c(0) c(0)
a2
=
a 1 λ2

and
b(1) 1
. =
c(1) c(1)
a2 μ2
= .
a1

On the other hand,

∂c a1
a
. 2 (0) + 2c = −αλ2 a1 x α−1 + β 2 x β−1 + 2c.
∂x μ

Hence,
∂c
a
. 2 (0) + 2c(0) = 2c(0) > 0
∂x
and
( )
∂c β
.a2 (1) + 2c(1) = 2c(1) + β 2 − αλ a1
2
∂x μ

and we can always choose .β ≥ αλ2 μ2 . The system is therefore positive.

Solution: Exercise 9.44 Let .{ψi } be a basis of . X h∗ , then the .wi ∈ Vh such that
∗
.(wi , p)(L 2 (Ω,R)) p = (A ψi , p)(L 2 (Ω,R)) p constitute a basis of . W because otherwise,
E
let . i λi wi = 0, the .λi not all being zero, then for all . p ∈ Vh , we have
( ( ) )
E
∗
. A λi ψi , p = 0.
i (L 2 (Ω,R)) p

In particular,
( ( ) )
E E
∗
. A λi ψi , λi ψi = 0.
i i (L 2 (Ω,R)) p

@seismicisolation
@seismicisolation
698 9 Exercises with Solutions

The positivity of . A∗ leads to

E
. λi ψi = 0.
i

Solution: Exercise 9.45 Let .W = {w ∈ Vh , (w, p)(L 2 (Ω,R)) p = (A∗ vh , p)(L 2 (Ω,R)) p
for all p ∈ Vh , with vh ∈ X h∗ }. .W is strictly included in .Vh if, and only if, . K er (B +t
M) does not reduce to the zero element. On .W , we define the linear form

w −→ L(w) = (F, vh )(L 2 (Ω,R)) p ,

.

where .vh ∈ X h∗ is uniquely defined by

(w, p)(L 2 (Ω,R)) p = (A∗ vh , p)(L 2 (Ω,R)) p

.

for all . p ∈ Vh . We have

|L(w)| ≤ ||F||(L 2 (Ω,R)) p ||vh ||(L 2 (Ω,R)) p

.

1
≤ ||F||(L 2 (Ω,R)) p ||w||(L 2 (Ω,R)) p .
α

We extend .Vh by a linear form noted . L such that

1
.||L|| ≤ ||F||(L 2 (Ω,R)) p .
α
This linear form is expressed using the scalar product induced by the scalar product
in .(L 2 (Ω, R)) p

. L(wh ) = (u h , wh )(L 2 (Ω,R)) p for all wh ∈ Vh

and, we have

||u h ||(L 2 (Ω,R)) p = ||L||

.

.
So, there exists .u h ∈ Vh such that

.( f, vh )(L 2 (Ω,R)) p = L(wh )

= (u h , w)(L 2 (Ω,R)) p
= (u h , A∗ vh )(L 2 (Ω,R)) p , for all vh ∈ X h∗ .

@seismicisolation
@seismicisolation
9.2 Corrected 699

Solution: Exercise 9.46

1. ( ) ( ) ( )( ) ( )
0 −1 ∂ u 10 u f
. + = ,
−1 0 ∂x v 01 v 0

. B − M = B +t M.

{ 1( ( ) ( )) { 1
dψh dϕh
. uh + ϕh + vh + ψh dx = f ϕh d x, for all (ϕh , ψh ) ∈ (Vh )2 ,
0 dx dx 0

with .ϕh (0) = ϕh (1) = 0. We have .2(I + 1) unknown (. I + 1 + I + 1) and .2I

equations (. I − 1 + I + 1):

{ 1
1 1 1
.− (vi+1 − vi−1 ) + (u i+1 + 4u i + u i−1 ) = f ϕi d x, 1 ≤ i ≤ I − 1,
2h 6 h 0

1 1
. − (u i+1 − u i−1 ) + (vi+1 + 4vi + vi−1 ) = 0, 1 ≤ i ≤ I − 1,
2h 6

1 1
. − (u 1 + u 0 ) + (2v0 + v1 ) = 0,
2h 6

1 1
. (u I + u I −1 ) + (2v I + v I −1 ) = 0.
2h 6
If we fix .u 0 and .u I , then we have a unique solution.
2. Add the two equations .u 0 = 0 and .u I = 0.

Solution: Exercise 9.47

du
. + σ u = f, 0 < x < 1.
dx

B(0) = −1 = −M(0), .(B − M)(0) = −2, .(B +t M)(0) = 0, . B(1) = 1 =

.
M(1), .(B − M)(1) = 0, .(B +t M)(1) = 2, . X h = {vh ∈ Z h such that vh (0) = 0},
and . X h∗ = {vh ∈ Z h such that vh (1) = 0}. The approximate problem is written: Find
.u h ∈ Z h such that

{ 1 ( ) { 1
dϕh
. uh − + σ ϕh d x = f ϕh d x, for all ϕh ∈ Z h∗ ,
0 dx 0

@seismicisolation
@seismicisolation
700 9 Exercises with Solutions

i.e.,

{
u i+1 − u i−1 σ 1 1
. + (u i+1 + 4u i + u i−1 ) = f ϕi d x, 1 ≤ i ≤ I − 1,
2h 6 h 0

{ 1
1 1
. − (u 1 + u 0 ) + σ (2u 0 + u 1 ) = f ϕ0 d x,
h h 0

u = 0,
. 0

where the .ϕi are the usual hat functions, and where .ϕ0 is the half hat function.
Solution: Exercise 9.48 The problem

. − u ,, + u = f, 0 < x < 1,

u(0) = u(1) = 0,
.

is written in the form

( ) ( ) ( )( ) ( )
0 −1 ∂ u 10 u f
. + = , 0 ≤ x ≤ 1.
−1 0 ∂x v 01 v 0

To take this into account in the limits, we choose

( ) ( )
a 1 b −1
. M(0) = , a ≥ 0, M(1) = , b ≥ 0.
−1 0 1 0

The problem (7.7.5) is written: Find .(u h , vh ) ∈ Z h × Z h such that

.

{ 1 (( ) ( ) )
dvh du h a
. − + uh − f ph + − + vh qh d x + u h (0) ph (0)
0 dx dx 2
b
− u h (0)qh (0) + u h (1) + u h (1)qh (1) = 0,
2

for all .( ph , qh ) ∈ Z h × Z h . By replacing . ph and .qh by the usual basic functions, we

obtain { xi+1
1 1
.− (vi+1 − vi−1 ) + (u i+1 + 4u i + u i−1 ) = f ϕi d x,
2h 6 xi−1

1 1
. − (u i+1 − u i−1 ) + (vi+1 + 4vi + vi−1 ) = 0,
2h 6

@seismicisolation
@seismicisolation
9.2 Corrected 701

for .1 ≤ i ≤ I − 1 (with . I h = 1, .xi = i h). So,

{ h
1 1 a 1
.− (v1 − v0 ) + (u 1 − 2u 0 ) + u0 = f ϕ0 d x,
h 3 2h h 0

{ h
1 1 b 1
. − (v I − v I −1 ) + (u I −1 + 2u I ) + uI = f ϕ I d x,
h 3 2h h 0

1 1 u0
. − (u 1 − u 0 ) + (v1 − 2v0 ) − = 0,
h 3 h
1 1 uI
. − (u I − u I −1 ) + (v I −1 − 2v I ) + = 0.
h 3 h
Solution: Exercise 9.49 1. We have
1
. h(σ u i+ 21 − f i+ 21 ) + (u i+ 23 − u i− 21 ) = 0, 1 ≤ i ≤ I − 2,
2
1( )
. h(σ u 21 − f 21 ) + u 23 + u 21 − μ = 0,
2
1
. h(σ u I − 21 − f I − 21 ) + (u I − 21 − u I − 23 ) = 0.
2

If . I is even, we assign .u 21 to .0, .u 23 to .2h, .u 5 to .2h, .u i+ 21 to .xi+ 21 + (−1)i−1 h2 . We

2
therefore
[ ( have, ( for the truncation ) error )
1 h
. hσ u xi+ 21 + (−1)i−1 − f i+ 21
h 2
( ( ) ( ))]
1 h h
. + u xi+ 23 − (−1)i−1 − u xi− 21 − (−1)i−1 = O(h),
2 2 2
[ ( ) 1 ]
1
. h σ u(0) − f 12 + (u(2h) + u(0)) − μ = O(h),
h 2
[ ( ( ) ) ( ( ) ( ))]
1 h 1 h h
. hσ u x I − 1 + − fI− 1 + u xI− 1 + − u xI− 3 − = O(h).
h 2 2 2 2 2 2 2 2

If . I is odd, we assign .u 21 to .h, .u 23 to .h, .u i+ 21 to .xi+ 21 + (−1)i h2 ,

[ ( ( ) )
1 h
. hσ u xi+ 21 + (−1)i − f i+ 21
h 2
( ( ) ( ))]
1 ih ih
. + u xi+ 23 − (−1) − u xi− 21 − (−1) = O(h),
2 2 2

@seismicisolation
@seismicisolation
702 9 Exercises with Solutions

[ ( ) 1 ]
1
. h σ u(h) − f 21 + (u(h) + u(h)) − μ = O(h),
h 2

[ ( ( ) ) ( ( ) ( ))]
1 h 1 h h
. hσ u x I − 1 + − fI− 1 + u xI− 1 + − u xI− 3 − = O(h).
h 2 2 2 2 2 2 2 2

2. For .1 ≤ i ≤ I − 2, we have

2(u i+ 23 − u i− 21 )u i+ 21 = u i+
.
2
3 − u
2
i− 1
− (u i+ 23 − u i− 21 )2 + (u i+ 21 − u i− 21 )2 .
2 2

We sum

u 2I − 1 − u 2I − 5 − (u I − 21 − u I − 23 )2 + (u I − 23 − u I − 5 )2
.
2 2 2

u 2I − 3 − u 2I − 7 − (u I − 23 − u I − 5 )2 + (u I − 5 − u I − 27 )2
.
2 2 2 2

u 2I − 5 − etc.
.
2

. u 23 − u 23 − (u 23 − u 5 )2 + (u 5 − u 23 )2
2 2 2 2

u 25 − u 21 − (u 5 − u 23 )2 + (u 23 − u 21 )2 .
.
2 2 2

He stays
1( 2 )
. u I − 1 + u 2I − 7 − (u I − 21 − u I − 23 )2 − u 23 − u 21 + (u 23 − u 21 )2 +
4 2 2 2 2
( )
1 1
. (u 3 + u 21 ) − μ u 21 + (u I − 21 − u I − 23 )u I − 21 .
2 2 2

@seismicisolation
@seismicisolation
9.2 Corrected 703

He comes

E I −1
1 2 1 1
. u I − 1 + u 21 − μu 12 + 2
u i+ 1 − f i+ 1 u i+ 1 = 0.
2 2 2 2 2 i=0
2 2 2

Let .vi+ 21 be the interpolation of .u in the sense of 1, and .wi+ 21 = u i+ 21 − vi+ 21 . We

have

E I −1 E I −1
1 2 1
. w I − 1 + w21 + h 2
wi+ 1 = h
2
wi+ 21 .
2 2 2 2 i=0
2
i=0

Hence,
( I −1
) 21
E
. h (u i+ 21 − vi+ 21 )2 = O(h).
i=0

So,

v
. i+ 1
2
− u(xi+ 21 ) = O(h)

and
I −1
E
. h(vi+ 21 − u(xi+ 21 ))2 = O(h 2 ).
i=0

The triangle inequality implies

( I −1
) 21
E
. h (u i+ 21 − u(xi+ 21 ))2 = O(h).
i=0

Solution: Exercise 9.50 We assume that .μ > 0 and .ν > 0. We pose

ΔxΔy Δx Δy 2
. A= , α=σ , β=σ , l = Δx 2 + Δy 2 .
2 μ γ

The discontinuous method, using constants per triangle, is written

{ {
. σ ϕh d xd y − (μn x + νn y )(ϕhint − ϕhext ) ds = 0.
T ∂− T

The exterior normal to triangle .1, along side . S16 (separating triangles .1 and .6) has
components

@seismicisolation
@seismicisolation
704 9 Exercises with Solutions

Δy Δx
n =
. x , ny = .
l l
Call .ϕ j the value of .ϕh on the triangle . j. For the triangle .6, we have the equation

σ Aϕ6 + νΔx(ϕ6 − ϕ5 ) = 0.
. (9.2.13)

For the triangle .1, we have

. σ Aϕ1 + μΔy(ϕ1 − ϕ2 ) + (−μΔy + νΔx)(ϕ1 − ϕ6 ) = 0.

For the other triangles, one or the other of these two relations is obtained.

1
2 6

3 5
4 7

(μ, ν)

In particular, we can write

σ Aϕ5 + μΔy(ϕ5 − ϕ4 ) + (−μΔy + νΔx)(ϕ5 − ϕ7 ) = 0.

. (9.2.14)

From the relation (9.2.13), we derive

νΔxϕ5 = (νΔx + σ A)ϕ6 .

.

By reporting in (9.2.14), we obtain

1
. (σ A + νΔx)ϕ6 − μΔyϕ4 − (−μΔy + νΔ x)ϕ7 = 0.
νΔx
Dividing by .ΔxΔy, we get
( 2 )
σ Δy ν μ μ ν
. +σ + ϕ6 − ϕ4 + ϕ7 − ϕ7 = 0,
4ν Δy Δx Δx Δy

@seismicisolation
@seismicisolation
9.2 Corrected 705

what is still being written

μ ν σ 2 Δy
. (ϕ7 − ϕ4 ) + (ϕ6 − ϕ7 ) + σ ϕ6 + ϕ6 = 0.
Δx Δy 4ν

In the sense of finite differences, this scheme is of order .1. If we eliminate the
unknowns .ϕ2 and .ϕ6 , we obtain the same

μ ν σ 2 Δy
. (ϕ5 − ϕ3 ) + (ϕ1 − ϕ5 ) + σ ϕ1 + ϕ1 = 0.
Δx Δy 4ν

It is of course advisable to modify the equations according to the slope of the char-
acteristic direction .(μ, ν). In the case of the particular mesh above, we therefore have
1
a method in . O(h) (to be compared with the value .h 2 provided by the general theo-
rem).

Solution: Exercise 9.51 1. We consider the problem

du
. + σ u = f, 0 < x < 1,
dx

u(0) = λ.
.

Petrov Galerkin’s method is written

I −1 {
E xi+1 ( )( ( ))
du h dvh
. + σ uh − f vh + h + σ vh d x + (u h (0+ ) − λ)vh =0,
i=0 xi dx dx

for all .vh . Using the trapezium formula to calculate the integrals, we get
u i+1 − ai−1
. + σ u i − f (xi )(1 + σ h)
2h
( )
u i+1 − 2u i + u i−1 u i+1 − u i−1 f (xi+1 ) − f (xi−1 )
. +h − − σ + =1
h2 2h 2h

.1 > i > J − 1.
( )
u1 − u0
. + σ u 0 − f (x0 ) (1 + 0h)
h
( )
−u 1 + u 0 u1 + u0 f (x1 ) − f (x0 )
.+h −σ + + (u 0 − λ) = 0.
h h h

@seismicisolation
@seismicisolation
706 9 Exercises with Solutions

| |
2. Given the problem: Find .u h ∈ Wh = T ⊂Ω Pk (T ) such that

E{ { ((
BT − M T
) ) }
. (Au h − f, vh + h Avh )(L 2 (T,R)) p − (u int
h − u h ), vh
ext int
ds = 0,
∂T 2 2
T ⊂Ω

for all .vh ∈ Wh , where .u h ext

|∂Ω = 0. We post, for all .v and .w whose restriction to . T
belongs to .C 1 (T, R)

E{ { ((
BT − M T
) ) }
. E(v, w) = (Av, w + h Aw)(L 2 (T,R)) p − (vint − v ext ), wint ds .
∂T 2 2
T ⊂Ω

Green’s formula leads

E 1
. E(vh , vh ) = ((A + A∗ )vh , vh )(L 2 (T,R)) p + h||Avh ||20,T +
T ⊂Ω
2
{
1 ( )
. (BT vhext , vhint )2 + (MT (vhint − vhext ), vhint )2 ds.
2 ∂T

We have
E{
. (BT vhext , vhint )2 ds = 0
T ⊂Ω ∂T

and
E{
. (MT (vhint − vhext ), vhint )2 ds ≥ 0,
T ⊂Ω ∂T

provided you choose the matrices . MT as in Theorem 7.9.1. So, we have

E
. E(vh , vh ) ≥ α||vh ||0,T + h||Avh ||20,T ,
2

T ⊂Ω

which proves the existence and uniqueness of the approximate solution .u h . For the
error markup, we set .e = u h − vh . Using Green’s sum, we write, using the notations
of Theorem 7.9.1,

E {
1 E
. E(e, e) ≥ α||e||20,T + h||Ae||20,T + (M S (eint − eext ), eint − eext )2 ds.
T ⊂Ω
2 S
S⊂S

@seismicisolation
@seismicisolation
References 707

By reasoning in Theorem 7.9.7 (choice of .vh and integration by parts and

hypotheses on . M S ), we obtain
1
||e||0,Ω = O(h k+ 2 ),
.

and consequently
1
||u − u h ||0,Ω = O(h k+ 2 ).
.

References

1. S. Kallel, Analyse hilbertienne, Exercices corrigés, Centre de Publication Universitaire,

Collection M/ Sciences Fondamentales (2001)
2. M. BenHamadou et A. Jeribi, Analyse numérique matricielle - Méthodes et algorithmes,
exercices et problèmes corrigés, Collection références sciences, Ellipses (2020)

@seismicisolation
@seismicisolation
Chapter 10
Revision Issues

10.1 Exercises and Problems

Exercise 10.1 Let the problem be .−u ,, + u = f , .0 < x < 1, .u(0) = u(1) = 0.
After having written this problem in the form of a Friedrichs system, explain the
equations for the problem (7.7.2), when .Vh = Z h × Z h , or . Z h is the space of contin-
uous functions, polynomials of degree.≤ 1 on each interval.[xi , xi+1 ],.0 ≤ i ≤ I − 1,
or . I h = 1 and .xi = i h, .0 ≤ i ≤ I . Conclude. ♦
Exercise 10.2 Repeat Exercise 10.1 by looking for the solution .u h in the space . X h
defined in (7.7.1). ♦
Exercise 10.3 1. We consider the scheme (7.9.8), (7.9.9), (7.9.10) with .m i = 0,
1 ≤ i ≤ I − 1. Show that we can assign the values .u i+ 21 to points . yi+ 21 ∈ [xi , xi+1 ]
.
so that the truncation error is in . O(h) for all equations (we will assume that .h i = h
for all .i).
2. Perform directly by finite difference techniques the increase of the error for the
diagram of 1. ♦
Exercise 10.4 We consider the problem of the elastic cord attached at its two extrem-
ities .0 and .1. We act on this string with a force of density . f . It is assumed that the
internal energy of the string is proportional to the variation in length of the string.
1. Write the total energy of the string for a displacement .v.
2. Simplify the expression for energy by assuming small motions.
3. Write this problem in the form .a(u, v) = L(v) and apply the Lax-Milgram theo-
rem.
4. Interpret the problem (differential equation). ♦
Exercise 10.5 Determine the eigenfunctions and eigenvalues of the operator

d2
. −
dx2
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 709
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
https://doi.org/10.1007/978-981-97-5710-7_10
@seismicisolation
@seismicisolation
710 10 Revision Issues

associated with the Dirichlet conditions, i.e.,

d 2ϕ
. − = λϕ, x ∈]0, 1[
dx2

ϕ(0) = ϕ(1) = 0.
.

♦
Exercise 10.6 Let .Vh be the space generated by the functions
(( ) )
1
w j (x) = cos
. j+ π x + 1 − x, 0 ≤ j ≤ N .
2

Calculate the coefficients { 1

dw j dwl
a jl =
. d x.
0 dx dx

Conclude. ♦
Exercise 10.7 Consider a regular open set .Ω. Using Green’s formula, show that
||Δv||0,Ω = |v|2,Ω for all .v ∈ H02 (Ω).
. ♦
Exercise 10.8 1. Show the following inequality
{ 1 { 1 ( )2
dv
. v dx ≤ c
2
dx
0 0 dx

for all .v ∈ C 1 ([0, 1], R), such that .v(0) = v(1) = 0.

2. Let .{vi }i=0
I
be a sequence of scalars such that .v0 = v I = 0. Using the relationship

x v2
. i+1 i+1 − xi vi2 = xi (vi+1 − vi )(vi+1 + vi ) + hvi+1
2
,

where .xi = i h, .0 ≤ i ≤ I , . I h = 1, show inequality (discrete Poincaré)

I −1
E I −1 (
E )
vi+1 − vi 2
. h vi2 ≤ 4h
i=1 i=0
h

take inspiration from 1.

3. We divide the interval .[0, 1] into . I equal intervals of length .h. We set .xi = i h,
.0 ≤ i ≤ I . Let . Wh be the space introduced in Example 4.3.4.
.(i) Calculate
{ 1( )
dvh 2
. dx
0 dx

for .vh ∈ Wh , depending on the values .vi = vh (xi ), .0 ≤ i ≤ I .

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 711

(ii) Calculate
.
{ xi+1
. vh2 d x
xi

depending on the values .vi and .vi+1 . Show that there exists a constant .c independent
of .vh and .h such that { xi+1
. vh2 d x ≥ ch(vi2 + vi+1
2
).
xi

.(iii) Let .Vh = {vh ∈ Wh such that vh (0) = vh (1) = 0}. Using 1. (Poincaré inequal-
ity) and the previous questions, find the inequality demonstrated at 2.
.(iv) Show that there exists a constant .c > 0 and independent of .h such that

({ ( )2 ) 21 ({ ) 21
1 1
dvh −1
. dx ≤ ch vh2 dx ,
0 dx 0

for all .vh ∈ Vh (this inequality as under the name of inverse inequality generalizes to
higher dimensions in space and to polynomials of higher degree).
4. We consider the problem (4.3.1), (4.3.2) discretized as in Example 4.3.4.
.(i) Using the discrete Poincaré inequality, show the inequality

⎛ ⎞ 21
I −1
E
.||u h ||h ≤ c ⎝h ( f (x j ))2 ⎠ ,
j=1

.|| · ||h is defined in Remark 4.3.4 (we will use an integration formula by discrete part).
.(ii) Using the truncation error definition (Remark 4.3.3) write the numerical scheme
linking the .e j = u h (x j ) − u(x j ), .0 ≤ j ≤ I to . E j , .1 ≤ j ≤ I − 1. Deduce that if
.u ∈ C ([0, 1], R), then .||e||h = O(h ), where .e = (e0 , . . . , e I ). ♦
4 2

Exercise 10.9 We consider the problem

( )
d du du
.− a(x) + b(x) + c(x)u = f, x ∈]0, 1[
dx dx dx

u(0) = u(1) = 0.
.

Give sufficient conditions on the functions .a, .b, and .c so that the problem admits a
solution .u ∈ H01 (0, 1). Same question with Neumann boundary conditions. ♦

Exercise 10.10 1. Let .ψ be a function of . H 1 (0, 1) checking, at direction of distri-

butions on .(0, 1), .ψ ,, − ψ = 0. Show that .ψ ∈ C 2 ([0, 1], R).
2. Let .V = {v ∈ H 1 (0, 1) such that v(0) = 0}. Determine the orthogonal of .V in
. H (0, 1) of which we will specify a basis of this space. ♦
1

@seismicisolation
@seismicisolation
712 10 Revision Issues

Exercise 10.11 Let . f α be the function defined on .]1, +∞[ by . f α (x) := x α .(α ∈
R). For which values of .α is . f α a function of . H 1 (]1, +∞[). Same question but
considering this time . f α defined on .]0, 1[. ♦
Exercise 10.12 Let .x0 ∈]0, 1[. Show that the problem: Find .u ∈ H01 (]0, 1[) such
that .−u ,, = δx0 admits a unique solution, where .δx0 is the mass of Dirac at point .x0 .
Calculate, then, this solution. ♦
Exercise 10.13 Let. X be a real Hilbert space of norm.|| · || and let. K be a non-empty,
closed, convex set, of . X .
1. Show that for all . f ∈ X , there is .u ∈ X unique such that .|| f − u|| ≤ || f − v|| for
all .v ∈ K .
2. Show that .u is characterized by the relation .( f − u, v − u) ≤ 0 for all .v ∈ K .
3. Let .a(·, ·) be a bilinear form on . X × X , symmetrical continuous and coercive and
. L(·) be a continuous linear form on . X . Show that there is one and only one .u ∈ K
checking
. J (u) ≤ J (v)

for all .v ∈ K , where

1
. J (v) = a(v, v) − L(v).
2
Furthermore, .u is characterized by .a(u, v − u) ≥ L(v − u) for all .v ∈ K . ♦
Exercise 10.14 Let . X be a real Hilbert space of norm .|| · ||, .a(·, ·) be a continuous
bilinear form on . X × X with . M for constant of continuity, and . L(·) ∈ X , . It is
assumed that there is .u ∈ X of the problem .a(u, v) = L(v) for all .v ∈ X . Let .Vh be
a subspace of . X of finite dimension for which there is a constant .βh > 0 such that

. inf sup a(u h , wh ) ≥ βh .

||vh ||=1 ||wh ||=1

1. Show that the problem: Find .u h ∈ Vh solution of

a(u h , vh ) = L(vh )
.

for all .vh ∈ Vh admits a unique solution.

2. Establish the general error estimates
( )
M
||u − u h || ≤ 1 +
. inf ||u − vh ||.
βh vh ∈Vh ♦

Exercise 10.15 Let . H and .V be two real Hilbert spaces. We note .(·, ·) H and .(·, ·)V
their, respective, inner products and, .|| · || H and .|| · ||V , the norms associated with
these scalar products. We assume that .V is a dense subspace of . H and that there
exists .C > 0 such that

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 713

.||v|| H ≤ C||v||V for all v ∈ V.

Let.a(·, ·) be a continuous bilinear form on. H × H and.b(·, ·) be a continuous bilinear

form on .V × V . We suppose that there is .α > 0 and .β > 0 such that

. a(v, v) ≥ α||v||2H , for all v ∈ H

b(v, v) + ||v||2H ≥ β||v||2V for all v ∈ V.

.

Finally, let . f ∈ H be fixed.

1. Show that there is .u ∈ H unique such that

a(u, v) = ( f, v) H for all v ∈ H.

.

2. Show that if we choose.n 0 large enough, then for all.n ≥ n 0 , there is.u n ∈ V unique
such that
1
.a(u n , v) + b(u n , v) = ( f, v) H for all v ∈ V.
n

3. Show that the sequence .(u n )n≥n 0 is bounded in . H and deduce that the sequence
( )
1
. √ un
n n≥n 0

is bounded in .V . ♦

Exercise 10.16 Recall that the Frobenius norm of a matrix .n × n

. A = (ai j )1≤i, j≤n

is given by [
|E
| n 2
.||A|| = | ai j
i, j=1

and that this norm check

||AB|| ≤ ||A||||B||
.

and
||A + B|| = ||A|| + ||B||
.

if, and only if, . A = λB .(we do not ask to demonstrate these relations again.).
1. For .s vector of .Rn , we denote by .s t the transpose of .s. Show that

||ss t || = (s, s),

.

@seismicisolation
@seismicisolation
714 10 Revision Issues

where .(s, s) designates the inner product of .s by himself.

2. Let . B be a matrix .n × n, .s /= 0 and . y be two vectors of .Rn . We pose

(y − Bs)s t
. B̄ = B + .
(s, s)

Check that . B̄ is solution of the problem

|| B̄ − B|| = inf{||C − B|| among the matrices C checking Cs = y}.

. (10.1.1)

3. Show that the problem (10.1.1) admits a unique minimum. ♦

Exercise 10.17 Let.a(·, ·) be a bilinear form symmetrical and coercive on the Hilbert
space .V . Let .Vh be a subspace of .V of finite dimension . N and let .ϕ1 , ϕ2 , . . . , ϕ N be
a basis of .Vh . Show that the matrix

. A = (a(ϕi , ϕ j ))1≤i, j≤N

is symmetrical positive definite. ♦

Exercise 10.18 We denote by . L 1 (] − 1, 1[, R), the space of integrable functions on
.] − 1, 1[. For . g ∈ L (] − 1, 1[, R), we pose
1

{ 1
||g||1 :=
. |g(x)|d x.
−1

1. Let. f ∈ L 1 (] − 1, 1[, R). Show that there exists. p ∈ Pn (the set of polynomials of
degree ≤ n) such that

.|| f − p||1 = inf{|| f − q||1 , q ∈ Pn }..

2. Determine the polynomial . p when .n = 0, . f (x) = 1 if .x ∈]0, 1[ and . f (x) = −1

if .x ∈] − 1, 0[. ♦
Exercise 10.19 We consider the problem at the limits: Find .u such that
⎧
⎨ d 2u
− + u = f on ]0, 1[ (10.1.2)
. dx2
⎩ u(0) = 0, u(1) = 0,

where . f ∈ L 2 (]0, 1[, R).

1. Write the variational formulation of the boundary problem (10.1.2)
(.define the spaceV ).
2. Show that the variational problem associated with (10.1.2) admits a unique solution
in an appropriate Hilbert space .V .
3. Show that the variational problem is equivalent to the problem (10.1.2).

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 715

4. When . f ∈ H 1 (0, 1), show that the solution of the problem variational belongs to
.C ([0, 1], R). Deduce that the boundary problem (10.1.2) admits a classical solution
2

and only one.

5. Given an integer . N greater than or equal to 1, we set .h = N 1+1 and we define a
regular subdivision .(xi )0≤i≤N +1 of the interval .[0, 1] of constant step .h. Let
{ }
. Wh = vh ∈ C 0 ([0, 1], R) such that vh|(xi ,xi+1 ) ∈ P1 (xi , xi+1 ) 0 ≤ i ≤ N .

.(a) Define a subspace .Vh of .V .

.(b) Determine .dim Vh .
.(c) Write the corresponding linear system to the approximate problem when looking
for .u h ∈ Vh . ♦

Exercise 10.20 We consider the boundary problem: find .u such that

⎧ 2
⎪ d u
⎨ − 2 + μu = f in ]0, 1[
. dx (10.1.3)
⎪
⎩ u(0) = 0, du (1) + τ u(1) = 0,
dx

where . f ∈ L 2 (0, 1), μ ∈ L ∞ (0, 1) with .μ(x) ≥ β > 0, for all .x ∈]0, 1[ and .τ > 0
is a real number.
1. Write the variational formulation of the boundary problem (10.1.3) (define the
space .V ).
2. Show that the variational problem associated with (10.1.3) admits a unique solution
in an appropriate Hilbert space .V .
3. Show that the variational problem is equivalent to the problem (10.1.3).
4. When the function .μ is fairly regular (for example, in .C 1 ([0, 1], R)) and . f ∈
H 1 (0, 1) show that the solution of the variational problem belongs to .C 2 ([0, 1], R).
Deduce that the boundary problem (10.1.3) admits a classical solution and a alone.
5. Given an integer . N greater than or equal to 1, we set .h = δ N 1+1 and we define a
regular subdivision .(xi )0≤i≤N +1 of the interval .[0, 1] of constant step .h. Let’s ask
{ }
. Wh = vh ∈ C 0 ([0, 1], R) such that vh|(xi ,xi+1 ) ∈ P1 (xi , xi+1 ) 0 ≤ i ≤ N .

.(i) Define a subspace .Vh of .V .

.(ii) Determine .dim Vh .
.(iii) Write the corresponding linear system to the approximation of the problem
when looking for .u h ∈ Vh . ♦

Exercise 10.21 We consider the weak formulation

{
Find u ∈ H01 (0, 1) such that
. (10.1.4)
a(u, v) = L(v) for all v ∈ H01 (0, 1),

@seismicisolation
@seismicisolation
716 10 Revision Issues

with { {
1 1
.a(u, v) = u , v, d x + u , v d x,
0 0

{ 1
. L(v) = f v d x,
0

and . f ∈ L 2 (]0, 1[, R).

.1. Show that .a(., .) is a continuous bilinear form on . H0 (0, 1).
1
{ 1
.2. Check that . u , u d x = 0 for all .u ∈ H01 (0, 1).
0
.3. Deduce that .a(., .) is coercive.
.4. Show that the problem (10.1.4) admits a unique solution noted .u.

.5. Show that .u checks a boundary problem

{
−u ,, + u , = f on (0, 1)
. (10.1.5)
u(0) = u(1) = 0.

.6. Show the equivalence between (10.1.4) and (10.1.5).

.7. When the function . f ∈ H 1 (0, 1), show that the solution of the variational problem
(10.1.4) belongs to .C 2 ([0, 1], R) and the boundary problem (10.1.5) admits only one
classic solution.
.8. Given an integer . N greater than or equal to 1, we set .h =
1
N +1
and we define a
regular subdivision .(xi )0≤i≤N +1 of the interval .[0, 1] of constant step .h. Let
{ }
. Wh = vh ∈ C 0 ([0, 1], R) such that vh|(xi ,xi+1 ) ∈ P1 (xi , xi+1 ) 0 ≤ i ≤ N .

.(a) Define the discrete subspace .Vh of . H01 (0, 1).

.(b) Determine .dim Vh .

.(c) Write the corresponding linear system to the approximate problem when looking
for .u h ∈ Vh . ♦

Exercise 10.22 We consider the weak formulation

{
Find u ∈ H01 (0, 1) such that
. (10.1.6)
a(u, v) = L(v) for all v ∈ H01 (0, 1),
{ 1 { 1 { 1
with .a(u, v)= χ u , v, d x + μu , v d x, . L(v) = f v d x, .χ (·) ∈ C 1 ([0, 1], R),
0 0 0
.χ (x) > 0 for all .x ∈ [0, 1], .μ(·) ∈ L ∞ (0, 1) and . f ∈ L 2 (]0, 1[, R).
.1. Show that .a(., .) is a continuous bilinear form on . H0 (0, 1).
1
{ 1
.2. Check that . χ u , u d x = 0 for all .u ∈ H01 (0, 1).
0
.3. Deduce that .a(., .) is coercive.
.4. Show that the problem (10.1.6) admits a unique solution noted .u.

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 717

5. Show that .u satisfies a boundary problem

.

{
−(χ u , ), + μu , = f on (0, 1)
. (10.1.7)
u(0) = u(1) = 0.

.6. Show the equivalence between (10.1.6) and (10.1.7).

.7. When the function . f ∈ H 1 (0, 1) and the function .μ(·) ∈ C 1 [0, 1], show that the
solution of the variational problem (10.1.6) belongs to .C 2 ([0, 1], R) and that the
boundary problem (10.1.7) admits only one classical solution.
.8. Given an integer . N greater than or equal to 1, we set .h =
1
N +1
and we define a
regular subdivision .(xi )0≤i≤N +1 of the interval .[0, 1] of constant step .h. Let
{ }
. Wh = vh ∈ C 0 ([0, 1], R) such that vh|(xi ,xi+1 ) ∈ P1 (xi , xi+1 ) 0 ≤ i ≤ N .

.(a) Define the discrete subspace .Vh of . H01 (0, 1).

.(b) Determine .dim Vh .
.(c) For .χ (x) = μ(x) = 1, write the corresponding linear system to the approximate
problem when looking for .u h ∈ Vh . ♦

Exercise 10.23 I. Let.Ω be a bounded open of.Rn with a fairly regular.∂Ω boundary.
We decompose .Ω in the form
| | | |
Ω = Ω1
. Ω2 E

(like the figure opposite)

H1 E H2

Ωk being an open of .Rn of border .Γk , .k = 1.2. We assume that the interface .E is
.

fairly regular .(C 1 per piece) so that for .v = (v1 , v2 ) ∈ H 1 (Ω1 ) × H 1 (Ω2 ), we can
define the traces .v1|E and .v2|E .
1. Let .v ∈ C 0 (Ω, R). Say why, we have
{ {
. vn 1 ds = − vn 2 ds,
E E

where .n 1 (resp., .n 2 ) is the external normal to .∂Ω1 (resp., .∂Ω2 ).

2. Let .v ∈ C 0 (Ω, R) be a function such that for each .k = 1, 2 the restriction of .v to
.Ωk , .v|Ωk , belongs to . H (Ωk ).
1

For .1 ≤ i ≤ n, we define .vi ∈ L 2 (Ω, R) by

@seismicisolation
@seismicisolation
718 10 Revision Issues

∂ ( )
v
. i |Ωk = v|Ωk , k = 1, 2.
∂ xi

(i) Show that for all .ϕ ∈ D(Ω), we have

.

2 {
E E 2 {
∂v
. (vi , ϕ) = ϕ dx − vϕn i ds.
k=1 Ωk ∂ xi k=1 Γk

(ii) Deduce that

.
∂v
v =
. i
∂ xi

in .D , (Ω).
.(iii) Deduce that .v ∈ H (Ω).
1

3. Let .Ω be a bounded open with border .Γ , .C 1 per piece. Assume that

| |
p
Ω=
. Ωk ,
k=1

where
.• Ωk is an open of .R contained in .Ω, whose border .γk is .C per piece .1 ≤ k ≤ p.
n 1
n
.• Ωk Ωs = ∅ for .k /= s.
Finally, let .v ∈ C 0 (Ω, R) be a function such that for all .k = 1, . . . , p the restriction
of .v to .Ωk , .v|Ωk , belongs to . H 1 (Ωk ).
Drawing on the above reasoning, show that .v ∈ H 1 (Ω). ♦

Exercise 10.24 We consider in dimension .n = 2, the function

( ) /
1 k 1
U = log
. for r = x 2 + y2 ≤ and k > 0.
r 2

1. Show that we can choose .k so that the function .U belongs to . H 1 (Ω), with
{ / }
1
Ω = (x, y) ∈ R2 such that x 2 + y 2 <
. .
2

2. Deduce that the inclusion

. H 1 (Ω) ⊂ C 0 (Ω, R)

is not, in general, for .n ≥ 2. ♦

Exercise 10.25 Throughout the problem .Ω denotes a regular bounded open edge
Γ = ∂Ω.
.

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 719

I. We denote the functional . J defined on the Sobolev space . H 1 (Ω) by

{ {
1
. J (v) = (|∇v|2 + v2 ) d x − f v dx
2 Ω Ω

and let . K be the subset of . H 1 (Ω)

. K = {v ∈ H 1 (Ω) such that v ≥ 0 on Γ }.

1. Show that . J has a unique minimum .u over . K .

2. Determine the partial differential equation satisfied by .u on .Ω
n (we will admit that
.u ∈ H (Ω)). Show that there exists a partition of .Γ : .Γ = Γ0 Γ1 with
2

∂u
on .Γ0 : .u = 0 ≥0.
∂n
∂u
on .Γ1 : .u ≥ 0 . = 0.
∂n
II. Approximation of an optimization problem:
Let .V be a Hilbert space, . J be a differentiable .α-convex functional defined on .V and
K be a closed convex of .V . We give ourselves .Vh and consider the problems
.

.(P) min J (v)

v∈K

and
. (Ph ) min J (v).
vh ∈K h

1. Show that the problems .(P) and .(Ph ) have a unique solution (respectively, .u and
u ) and write the Euler inequalities verified by .u and .u h .
. h

We assume that the approximation of.V and. K by.Vh and. K h satisfies both hypotheses
(i) For all .v ∈ K , there is .vh ∈ K h such that .vh → v strongly when .h tends to .0.
.
(ii) .vh ∈ K h and .vh converge weakly to .v implies .v ∈ K .
.

2. Show that the sequence . J (u h ) is bounded, then that the sequence .vh is bounded.
Deduce that .u h converges weakly, then strongly towards .u when .h tends towards .0.
III. We return to the problem of I.: We assume here that .Ω is a polygon of .R2 and
we consider a triangulation of .Ω by triangles .T of diameters less than or equal at .h.
We pose

. Vh = {vh ∈ C 0 (Ω, R), vh is a polynomial of degree 1 (vh ∈ P1 ) on each triangle T }

and

. K h = {vh ∈ Vh , vh (ai ) ≥ 0 at any vertex ai of the triangulation located on Γ }.

@seismicisolation
@seismicisolation
720 10 Revision Issues

1. Show . K h ⊂ K .
2. Quickly express why the hypothesis .(i) is verified here. Show that .(ii) is verified.
Show that .u h solution of the problem .(Ph ) strongly converges to .u solution of .(P).
♦

Exercise 10.26 Let .]a, b[ be a bounded interval of .R and .χ (·) be a function of

H 2 (]a, b[) satisfying
.

.χ (a) ≤ 0

and
χ (b) ≤ 0.
.

n
1. .(a) Show that there is .u ∈ H 2 (]a, b[) H01 (]a, b[) such that

. − u ,, = −χ ,, on ]a, b[.

.(b) Show that .u ≥ χ almost everywhere on .]a, b[.

2. We pose . K = {v ∈ H01 (]a, b[) such that v ≥ χ almost everywhere on ]a, b[}.
.(a) Show that the set . K is a non-empty convex.

.(b) Show that .u ≥ χ almost everywhere on .]a, b[.

3. We set . K = {v ∈ H01 (]a, b[) such that v ≥ χ almost everywhere on ]a, b[}.
.(a) Show that the set . K is a non-empty convex.

.(b) Show that . K is closed.

4. Let . f ∈ L 2 (]a, b[, R). We set for .u, .v ∈ H 1 (]a, b[)
{ b
a(u, v) =
. u , v, d x
a

and { b
. L(v) = f v d x.
a

Show that the problem: Find .u ∈ K such that

.a(u, u − v) ≤ L(u − v) for all v ∈ K (10.1.8)

admits one solution and only one.

5. Let .u be the solution of the problem (10.1.8). We assume that .u ∈ H 2 (]a, b[) and
we set
. I = {x ∈]a, b[ such that u(x) > χ (x)}.

(a) Show that . I is an open set of .R.

.

(b) Assuming that for all .ϕ ∈ D(I ), .u + tϕ ∈ K for all .|t| > 0 small enough, show
.
that
,,
.(−u − f )(u − χ ) = 0 almost everywhere on ]a, b[.

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 721

(c) Knowing that the set of positive functions of .D(]a, b[) is dense in the set of
.

positive functions of . L 2 (]a, b[, R), show that

. − u ,, ≥ f almost everywhere on ]a, b[.

♦
Exercise 10.27 We consider the problem at the limits: Find .u such that
⎧ ( )
⎪ d du
⎨− χ + μu = f on ]0, 1[
.
dx dx (10.1.9)
⎪
⎩ u(0) = 0, du (1) + τ u(1) = 0,
dx

where . f ∈ L 2 (]0, 1[, R), .χ ∈ C 1 ([0, 1], R), .μ ∈ L ∞ (0, 1) with

χ (x) ≥ α > 0,
.
μ(x) ≥ β > 0,

for all .x ∈]0, 1[ and .τ ≥ 0 is a real number.

1. Write the variational formulation of the boundary problem (10.1.9) (define the
space V).
2. Show that the variational problem associated with (10.1.9) admits a unique solution
in an appropriate Hilbert space .V .
3. Show that the variational problem is equivalent to the problem (10.1.9).
4. When the functions .χ and .μ are fairly regular .(for example in C 1 ([0, 1], R)
and . f ∈ H 1 (0, 1) show that the solution of the variational problem belongs to
.C ([0, 1], R). Deduce that the boundary problem (10.1.9) admits a classical solution
2

and only one.

5. Given an integer . N greater than or equal to 1, we set .h = N 1+1 and we define a
regular subdivision .(xi )0≤i≤N +1 of the interval .[0, 1] of constant step .h. Let
{ }
. Wh = vh ∈ C 0 ([0, 1], R) such that vh|(xi ,xi+1 ) ∈ P1 (xi , xi+1 ) 0 ≤ i ≤ N .

.(a) Define a subspace .Vh of .V .

.(b) Determine .dim Vh .
.(c) For .χ (x) = 1 ∀x ∈]0, 1[, write the corresponding linear system to the approx-
imate problem when we seek .u h ∈ Vh . ♦
Exercise 10.28 We consider the problem at the limits: Find .u such that
⎧ ( )
⎪ d du
⎨− χ + μu = f on ]0, 1[
.
dx dx (10.1.10)
⎪
⎩ du (0) − τ u(0) = 0, u(1) = 0,
dx

where . f ∈ L 2 (]0, 1[, R), χ ∈ C 0 ([0, 1], R), μ ∈ L ∞ (0, 1) with

@seismicisolation
@seismicisolation
722 10 Revision Issues

.χ (x) ≥ a > 0,

. μ(x) ≥ b > 0,

for all .x ∈]0, 1[ and .τ ≥ 0 is a real number.

1. Write the variational formulation of the boundary problem (10.1.10) (define the
space .V ).
2. Show that the variational problem associated with (10.1.10) admits a unique solu-
tion in an appropriate Hilbert space .V .
3. Show that the variational problem is equivalent to the problem (10.1.10).
4. When the functions .χ and .μ are fairly regular .(for example, in .C 1 ([0, 1], R))
and . f ∈ H 1 (0, 1) show that the solution of the variational problem belongs to space
.C ([0, 1], R). Deduce that the boundary problem (10.1.10) admits a classical solution
2

and only one.

5. Given an integer . N greater than or equal to 1, we set .h = N 1+1 and we define a
regular subdivision .(xi )0≤i≤N +1 of the interval .[0, 1] of constant step .h. Let
{ }
. Wh = vh ∈ C 0 ([0, 1], R) such that vh|(xi ,xi+1 ) ∈ P1 (xi , xi+1 ) 0 ≤ i ≤ N .

.(a) Define a subspace .Vh of .V .

.(b) Determine .dim Vh .
.(c) For .χ (x) = 1 ∀x ∈]0, 1[, write the corresponding linear system to the approx-
imate problem when we seek .u h ∈ Vh . ♦

Exercise 10.29 I. Let .a(·, ·) be a continuous and coercive bilinear form on the
Hilbert space, of infinite dimension .V , and let . L(·) be a continuous linear form over
. V . We consider the following two problems:

Find .u ∈ V such that

. J (u) = inf J (v), (10.1.11)

v∈V

where
1
. J (v) := a(v, v) − L(v).
2
Find .u ∈ V such that
a(u, v) = L(v) for all v ∈ V.
. (10.1.12)

1. Let .u be a solution of the problem (10.1.11).

.

(a) Let .ε be a scalar and .w be an any function of .V . Show that

.

. J (u) ≤ J (u + εw).

(b) Deduce that .u is solution of the problem (10.1.12).

.
2. Let .u be a solution of the problem (10.1.12).
.

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 723

(a) Show that for all .v ∈ V ,

.

1
. J (v) = J (u) + a(u, v − u) − L(v − u) + a(v − u, v − u).
2
.(b) Deduce that .u is solution of the problem (10.1.11).
.3. Show that the problem (10.1.11) admits a unique solution.
.4. To solve the problem numerically (10.1.11), we are led to look for an approximate
solution .u h belonging to a subspace .Vh ⊂ V , of finite dimension, i.e., find .u h ∈ Vh
such that
. J (u h ) = inf J (vh ). (10.1.13)
vh ∈Vh

(a) Show that the problem (10.1.13) admits a unique solution .u h ∈ Vh .

.
(b) Show that if .a(·, ·) is symmetric, then
.

/
M
.||u − u h || ≤ inf ||u − vh ||,
α vh ∈Vh

where . M is the continuity constant of .a(·, ·) and .α is the coercivity constant of .a(·, ·).
II. We consider the boundary problem: Find .u such that
⎧ ( )
⎪ d du
⎨− χ + μu = f in]0.1[
.
dx dx (10.1.14)
⎪
⎩ u(0) = 0, du (1) = 1,
dx

where . f ∈ L 2 (]0, 1[, R), .χ ∈ C 0 ([0, 1], R), .μ ∈ L ∞ (0, 1) with

χ (x) ≥ α > 0
.

and
μ(x) ≥ β > 0,
.

for all .x ∈]0.1[. This problem corresponds to a metal bar heated by means of a
quantity of heat. f , the function.u representing the temperature. We fix the temperature
.u = 0 in . x = 0 and we suppose that the heat flow

du
. (1) = 1
dx
in .x = 1.
.1. Write the variational formulation of the boundary problem (10.1.14) (define the
space .V ).
.2. Show that the variational problem associated with (10.1.14) admits a unique solu-

tion in an appropriate Hilbert space .V .

@seismicisolation
@seismicisolation
724 10 Revision Issues

.3. Show that the variational problem is equivalent to the problem (10.1.14).
.4. When the functions .χ and .μ are fairly regular (.for example in C 1 ([0, 1], R)) and
. f ∈ H (0, 1) show that the solution of the variational problem belongs to space
1

.C ([0, 1], R). Deduce that the boundary problem (10.1.14) admits a classical solution
2

and only one.

.5. Given an integer . N greater than or equal to 1, we set .h =
1
N +1
and we define a
regular subdivision .(xi )0≤i≤N +1 of the interval .[0, 1] of constant step .h. Let
{ }
. Wh = vh ∈ C 0 ([0, 1], R) such that vh|(xi ,xi+1 ) ∈ P1 (xi , xi+1 ) 0 ≤ i ≤ N .

.(a) Define a subspace .Vh of .V .

.(b) Determine .dim Vh .
.(c) For .χ (x) = 1 ∀x ∈]0, 1[, write the corresponding linear system to the approx-

imate problem when we seek .u h ∈ Vh .

III. Let . E be a real Hilbert space and . F be a closed vector subspace of . E.
1. Let .x ∈ E. Show equivalence
.

. y = PF (x) if, and only if, y ∈ F and y − x ∈ F ⊥ ,

where . PF is the projection operator from . E into . F.

.2. Show that the map . PF is a continuous linear from . E into . F. Calculate its norm
and its kernel.
⊥
.3. Show that . E = F ⊕ F .
( ⊥ )⊥
.4. Show that . F = F. ♦
Exercise 10.30 Let . P = P1 + {ξ η(1 − ξ − η)}.
η

(0, 1)

1
2 3 2

4 ( 31 , 13 )

1
1 ξ
2

(0, 0) (1, 0)

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 725

1. Is there a unique polynomial . p ∈ P taking values given at points .1, .2, .3, and .4.
Indication: we, can, find the basis functions

ϕ (ξ, η) = 1 − 2η − 9ξ η(1 − ξ − η),

. 1

ϕ (ξ, η) = 1 − 2(1 − ξ − η) − 9ξ η(1 − ξ − η),

. 2

ϕ (ξ, η) = 1 − 2ξ − 9ξ η(1 − ξ − η),

. 3

ϕ (ξ, η) = 27ξ η(1 − ξ − η).

. 4

Hence,

. -1 ) + ϕ2 (ξ, η) p( A
p(ξ, η) = ϕ1 (ξ, η) p( A -2 ) + ϕ3 (ξ, η) p( A
-3 ) + ϕ4 (ξ, η) p( A
-4 ).

For the uniqueness of . p, it suffices to consider the mapping

Φ : P −→ R4
.

p −→ ( p(-
A1 ), p(-
A2 ), p(-
A3 ), p(-
A4 ))

and to show that .Φ is bijective.

2. Notice, according to question 1.

. -1 ) + p( A
p(ξ, η)|ξ =0 = p(0, η) = (1 − 2η) p( A -3 ) + (2η − 1) p( A
-2 ).

So, there is no continuity between two neighboring triangles. So, this is a non-
conforming method. Study the non-conforming method associated with this element,
the existence of an approximate solution and the convergence of the approximate
solution towards the exact solution.
3. Same questions for . P = P1 + {ξ 2 + η2 + (1 − ξ − η)2 }. ♦
Exercise 10.31 Consider the following triangulation of a domain .Ω, where the
vertices are numbered as in the figure opposite.
11 10 9 8 7 6 5

27 26 25 24 23
12 4

13 3
28 35 34 33 22

14 2
29 30 31 32 21

15 16 17 18 19 20 1

@seismicisolation
@seismicisolation
726 10 Revision Issues

We solve the problem ⎧

⎪ −Δu = f in Ω
⎨
.
u = 0 on ∂Ω \ [A1 A15 ]
⎩ ∂u = 0 on [A1 A15 ].
⎪
∂ν
We use a finite element method with six nodes (degree of freedom at the vertices and
the midpoints of the sides): give the linear system corresponding to this problem (we
will suppose that the horizontal and vertical sides of the triangles are lengths .h). ♦

Exercise 10.32 I. Let . P = P2 + {ξ η(1 − ξ − η)}.

(0, 1)

5 4

7 ( 13 , 13 )

3 6 1

(0, 0) (1, 0)

1. Is there a unique polynomial . p ∈ P taking values given at points .1, .2, .3, .4, .5, .6,
and .7.
2. Can we then, with this finite element, construct a space

. Vh ⊂ H01 (Ω),
u
where .Ω := (triangles) and then give an estimates of the error for the problem

. − Δu = f

with Dirichlet boundary conditions.

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 727

II. Let . P = P1 + {ξ η(1 − ξ − η)}.

1. Is there a unique polynomial . p ∈ P taking values given at points .1, .2, .3, and .7.
Note: The basic functions can be searched for at points .1, .2, .3, and .7 which are given
by
.ϕ1 (ξ, η) = ξ − 9ξ η(1 − ξ − η),

ϕ (ξ, η) = η − 9ξ η(1 − ξ − η),

. 2

ϕ (ξ, η) = 1 − ξ − η − 9ξ η(1 − ξ − η),

. 3

ϕ (ξ, η) = 27ξ η(1 − ξ − η).

. 7

2. Can we then with this finite element construct a space

n
. Vh ⊂ C 0 (Ω, R) H01 (Ω),
u
where .Ω := (triangles) and then give an estimate of the error for the problem

. − Δu = f

with Dirichlet conditions.

3. Is there a unique polynomial . p ∈ P taking values given at points .4, .5, .6, and .7.
Note: The basic functions can be searched for at points .4, .5, .6, and .7 which are given
by
ϕ4 (ξ, η) = 1 − 2(1 − ξ − η) − 9ξ η(1 − ξ − η), ϕ5 (ξ, η) = 1 − 2ξ − 9ξ η(1 − ξ − η),
.
ϕ6 (ξ, η) = 1 − 2η − 9ξ η(1 − ξ − η), ϕ7 (ξ, η) = 27ξ η(1 − ξ − η).

4. Study the non-conforming method associated with this element; existence of the
approximate solution and convergence of the approximate solution to the exact solu-
tion.
5. Same question with

. P = P1 + {ξ 2 + η2 + (1 − ξ − η)2 }.

Exercise 10.33 Consider the following 10-node triangle.

@seismicisolation
@seismicisolation
728 10 Revision Issues

Find the abscissa .a and the weights to obtain an exact formula for . P3 , from a
triangle with 10 nodes. ♦

Exercise 10.34 Let .(K , E K , PK ) be the following finite element

E K = {values at midpoints of sides}

.

. PK = P1 .

a2 a1

a3

Show that the set .E K is . PK -unisolvent. Give the expression of a function of . PK

equal to . pi at points .ai .1 ≤ i ≤ 3. Is the element thus defined of class .C 0 . ♦

Exercise 10.35 Let

. P = P2 + {ξ η(1 − ξ − η)}.

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 729

5 4

6 7 3
axis of symmetry

1 2

Is there a unique polynomial . p ∈ P taking values given at points .1, .2, .3, .4, .5, .6,
and .7. Conclude. Indication: It must be shown that this polynomial is not unique. So,
.(T riangle, P) is not a finite element. ♦
Exercise 10.36 Let . P = P3 .

6 5

7 10 4
axis of symmetry

3 8 9 1

1. Is there a unique polynomial . p ∈ P taking values given at points .1, .2, .3, .4, .5, .6,
7, .8, .9, and .10.
.
2. Can we then, with this finite element, construct a space
n
. Vh ⊂ C 0 (Ω, R) H01 (Ω),
u
where .Ω := (triangles) and then give an estimate of the error for the problem

@seismicisolation
@seismicisolation
730 10 Revision Issues

. − Δu = f

with Dirichlet boundary conditions. ♦

Exercise 10.37 Let the square .Ω =]0, 1[×]0, 1[.

y

5 2
3

T1

6 9 8

T2

4 7 1 x

1. Let .Wh = {vh ∈ C 0 (Ω, R); vh |Ti ∈ P2 , i = 1, 2, vh determined by its values

. at the vertices and midpoints of sides}.
What is the dimension
n of .Wh .
2. Let .Vh = Wh H01 (Ω). We search .u h ∈ Vh such that .a(u h , vh ) = L(vh ), for all
.vh ∈ Vh , where {
. L(vh ) = f vh d xd y
Ω

and {
.a(vh , wh ) = ∇vh ∇wh d xd y.
Ω

Write the corresponding linear system.

n
u Same question when .Vh = Wh V , where .V = {v ∈ H (Ω); v = 0 on [2, 3]
1
3.
[3, 4]}.
4. Same question when the space .Vh = Wh ; discuss then the existence of possible
solutions. ♦

Exercise 10.38 Let .Ω =]0, 2[×]0, 1[ be the domain.

@seismicisolation
@seismicisolation
10.1 Exercises and Problems 731

2 x

Let .Wh = {vh ∈ L 2 (Ω, R); vh |T ∈ P1 , vh determined by its values

. in the middle of the sides}.
What is the dimension of .Wh .
2. Let .Vh = {vh ∈ Wh ; vh = 0 in the middle of sides ⊂ ∂Ω}. We search .u h ∈ Vh
such that
.ah (u h , vh ) = L(vh ),

E{
where
a (vh , wh ) =
. h ∇vh ∇wh d xd y
T ⊂Ω T

and {
. L(vh ) = f vh d xd y.
Ω

Write the linear system satisfied by the components of .u h (we will first number the
points).
3. Same question when .Vh = {vh ∈ Wh ; vh = 0 for the points such that y = 1}.
4. Same question when the space .Vh = Wh . What can we say then. ♦
- = [−1, 1] × [−1, 1], we consider
Exercise 10.39 1. On the square of reference . R
-
the space . P = {1, ξ, η, ξ − η }. Show that there exists a unique polynomial
2 2

.- - determined by its mean values along the sides of . R.

p ∈ P, - Give the expression of
the corresponding basic functions.
2. We want to use the preceding non-conforming finite element to solve the problem

. − Δu = f on Ω =]0, 1[×]0, 1[
u = 0 on ∂Ω.

@seismicisolation
@seismicisolation
732 10 Revision Issues

Define an approximate problem. Check the existence of the approximate solution

u .
. h
3. By appropriately rewriting the expression
E{ ∂u
. E(u, wh ) = wh ds,
T ⊂Ω ∂T ∂n T

give an increase in it according to .h, .u and of

( ) 21
E
.||wh ||h = |wh |21,T ,
T ⊂Ω

then of .||u − u h ||h .

4. Study, in this case, the Aubin-Nitsche duality process to obtain an increase in norm
. L with the Dirichlet conditions .u = 0 on .∂Ω, then with the following conditions
2
∂u
.u = 0 for . x = 0 and . x = 1 and for . y = 0, . = 0 for . y = 1.
∂y
5. With the preceding finite element, define a Nodal method. Is there equivalence
between this Nodal method and the previous non-conforming method? ♦

Exercise 10.40 Let . P = {1, ξ, η, ξ η}.

η

-2
A 1 -1
A

-
R

−1 1 ξ

-3
A −1 -4
A

-i , .i = 1, .2, .3,
1. Is there a unique polynomial . p ∈ P taking values given at points . A
and .4. Indication: We can show that

@seismicisolation
@seismicisolation
10.2 Problems and Exams 733

(1 + ξ )(1 + η) - (1 − ξ )(1 + η) - (1 − ξ )(1 − η) -

. p(ξ, η) = p( A1 ) + p( A2 ) + p( A3 )
4 4 4
(1 + ξ )(1 − η) -
+ p( A4 ).
4

For the uniqueness of . p, it suffices to consider the mapping

.Φ : P −→ R4
p −→ ( p(-
A1 ), p(-
A2 ), p(-
A3 ), p(-
A4 ))

and to show that .Φ is bijective.

Let .Ω be the following domain:
6 5 4 3

7 2
12 11

8 9 10 1
n
uwith this finite element construct a space .Vh ⊂ C (Ω, R) H01 (Ω),
0
2. Can we then
where .Ω := (r ectangles) and then give an error increase for the

. − Δu = f

problem with Dirichlet boundary conditions. ♦

10.2 Problems and Exams

Exercise 10.41 I. Consider the segment .[−h, h]. Let . P2 := {1, x, x 2 }.

1. Determine the functions .x −→ ϕi (x) ∈ P2 , .1 ≤ i ≤ 3 such that .ϕi (a j ) = 0 for
. j / = i, .ϕi (ai ) = 1, where .a1 = −h, .a2 = α, and .a3 = h, where .α is a real (discuss
in terms of .α; what happens in particular if .α → h).
2. Study the increase of the expression as a function of .α
{ h ( )
d dwh
. (v − rh v) (x) d x,
−h dx dx

@seismicisolation
@seismicisolation
734 10 Revision Issues

as a function of .h, .v, and .||wh ||1,(−h,h) , where .wh ∈ P2 and where .rh v denotes the
polynomial of . P2 taking the same values as .v in .−h, .α, and .h.
II. Let the segment .[−1, 1]. Let . P3 := {1, ξ, ξ 2 , ξ 3 }.
1. To any function .- v ∈ C 0 ([−1, 1], R), we associate its interpolated .-r-v ∈ P3 , such
as .-
r-
v =-v at points .−1, .+1, .+β, and .−β. What can be said about .- r-v −- v if .-
v ∈ P3 .
Give the expression of .- v −- r-
v if .-
v = ξ 4 . Deduce an increase of
{ 1
d d-
w
. (-
v −-
r-
v) dξ,
−1 dξ dξ

where .-w ∈ P3 , depending on .- v and .||-

w||1,(−1,1) (discuss depending on .β).
2. What would become of this increase on a current segment .[x j , x j+1 ], as a function
of .h = x j+1 − x j .
3. To solve the problem .−u ,, = f , .0 < x < 1, .u(0) = u(1) = 0, we divide the inter-
val .[0, 1] into . I intervals of length .h. We set .x j = j h. We define

. Vh = {vh ∈ C 0 ([0, 1], R), vh ||(x j ,x j+1 ) ∈ P3 , vh determined on each segment by

its values in x j , x j+1 and to the ranges of ± β by the transformation
1+ξ 1−ξ
x= x j+1 + x j }.
2 2
Define an approximate problem allowing to find .u h ∈ Vh . Give an increase of .||u h −
rh u||1,(0,1) according to .β. ♦

Exercise 10.42 I. Consider the problem: Find .u ∈ V such that .a(u, v) = L(v) for
all .v ∈ V , where .V := {v ∈ H 1 (−1, 1) such that v(0) = 0, v(1) + v(−1) = 0},
{ 1
a(u, v) =
. u , v, d x,
−1

and { 1
. L(v) = f v d x,
−1

with . f ∈ L 2 (] − 1, 1[, R).

1. Check that this problem admits a unique solution.
2. Write with precision the differential equations satisfied by .u, as well as the bound-
ary conditions. In the case, where . f = 1, calculate .u exactly. What can we say about
the regularity of .u.
3. To discretize the problem, we choose an integer. I and we set.h = 1I . We set.xi = i h,
.−I ≤ i ≤ I . Define using hat functions a basis of . Vh , space of continuous functions,

piecewise affine, with .Vh ⊂ V . Make sure that the problem: Find .u h ∈ Vh such that
.a(u h , vh ) = L(vh ) for all .vh ∈ Vh admits a unique solution. Write the associated
linear system.
4. Let .W = {v ∈ H 1 (−1, 1) such that v(0) = 0}. We set

@seismicisolation
@seismicisolation
10.2 Problems and Exams 735

. K = {v ∈ W such that v(1) + v(−1) = 0}.

We consider the problem: Find .-

u ∈ K such that

. J (-
u ) = inf{J (v), v ∈ K },

where
1
. J (v) = a(v, v) − L(v).
2
Compare .- u and .u.
5. For .ε > 0, we set
Φ(v) = (v(1) + v(−1))2
.

and
1
J (v) = J (v) + Φ(v).
. ε
ε
Study the problem: Find .u ε ∈ W such that

J (u ε ) = inf{Jε (v) v ∈ W }.
. ε

Convergence of .u ε to .u.
6. Try to introduce a Lagrange multiplier . p ∈ R2 , corresponding to the constraint

v(1) + v(−1) = 0.
.

Define if possible a Lagrange for the above problem and an Uzawa algorithm to deal
with this problem. What would become of this algorithm in the discrete case?
II. Let .V := {v ∈ H 1 (−1, 1) such that v(0) = 0}. We pose

1
. J (v) = a(v, v) − L(v),
2
with { 1
a(u, v) =
. u , v, d x,
−1

{ 1
. L(v) = f v d x,
−1

and || ( )||
|| 1 ||
. j (v) = || ||
||v 2 || .

1. Check that the problem: Find .u ∈ V such that

@seismicisolation
@seismicisolation
736 10 Revision Issues

. J (u) + j (u) = inf{J (v) + j (v), v ∈ V }

admits a unique solution.

2. Define a regularized function . jε (v), for . j (v), then study the problem: Find .u ε ∈ V
such that
. J (u ε ) + jε (u ε ) = inf{J (v) + jε (v), v ∈ V }.

Convergence of .u ε to .u.
3. Try to define for this problem a Lagrange multiplier, then an Uzawa-type algorithm.
4. Let . I be an integer and .h be such that .2I h = 1. We set .xi = i h, .−I ≤ i ≤ I . Write
the preceding problems in the discrete case, giving as many arguments as possible.
♦

Exercise 10.43 I. We pose

{ 1 { 1
. J (v) = (v,2 + bv2 ) d x − f v d x,
0 0

|{ |
| 1 |
. j (v) = || v d x || ,
0

W = H 1 (Ω), .V = H01 (Ω), where .b ≥ 0.

.
1. To study in terms of .b the problem: Find .u ∈ W (resp. .V ) such that

. J (u) + j (u) = inf{J (v) + j (v), for all v ∈ W (resp. V )}

admit a unique solution.

2. We consider here the case of the space .V , with .b = 0 and
{ 1
. f d x = 0.
0

Can we use the relation

{ 1 { 1
.ψ =ψ− ψ dx + ψ dx
0 0

to characterize the solution .u by an equality. Comment.

3. For .ε > 0, we pose
(({ )2 ) 21
1
. ε j (v) = v dx + ε2 .
0

Consider the problem: Find .u ε ∈ V such that

@seismicisolation
@seismicisolation
10.2 Problems and Exams 737

. J (u ε ) + jε (u ε ) = inf{J (v) + jε (v), for all v ∈ V }.

Study the existence and uniqueness of .u ε . Calculate the gradient of . jε (Cakes,

Fréchet). Highlight a Lagrange multiplier . p ∈ R such that
{ 1
. a(u, v) − L(v) + p v d x = 0,
0

for all .v ∈ V. What are the relationships satisfied by . p?

4. Let .Vh ⊂ V be the space generated by the hats functions .ϕ j , piecewise affine
(the pieces being the segments .[xi , xi+1 ], .xi = i h, . I h = 1), .ϕ j (x j ) = 1, .ϕ j (xi ) = 0,
.i / = j. Let .u h be the solution of the approximate problem. Write the inequalities
satisfied by the components
.u j = u h (x j ),

1 ≤ j ≤ I − 1. In the case, where . I = 2, calculate exactly .u 1 and . p, first when

.
f = 0, then when . f (x1 ) > 0 and finally . f (x1 ) < 0 .(we assume that the calculation
.

of {
. f ϕj

is performed by the trapezoidal method on each interval). We will use results similar
to those of question 3. Can we say something for . I > 2.
5. In the case of the previous question, define an Uzawa algorithm to approach the
.u j , .1 ≤ j ≤ I − 1, and . p. What can we say when . I = 2, . I = 3 and in general.

II. Let
{ ( ) }
1
.K = v ∈ H 2 (−1, 1) such that v(0) = v, (0) = 0, v = b, v(1) = β, v, (1) = γ .
2

1. Show that the problem: Find .u ∈ K such that

{ 1 {{ 1 }
,, 2 ,, 2
. J (u) = (u ) d x = inf (v ) d x, v ∈ K
0 0

admits a unique solution. Calculate this solution.

2. Let. K ∗ = {v ∈ H 2 (−1, 1) such that v(0) = v, (0) = 0, v(1) = β, v, (1) = γ }. We
pose
( ( ) )2
1
.Φ(v) = v −b .
2

Let .ε > 0. Show that the problem: Find .u ε ∈ K ∗ such that

{ }
1 1 ∗
. J (u ε ) + Φ(u ε ) = inf J (v) + Φ(v), v ∈ K
ε ε

@seismicisolation
@seismicisolation
738 10 Revision Issues

admits a unique solution. Give the results of convergence .u ε towards .u. Write the
relations satisfied by .u ε . By choosing a suitable . pε , show the existence of . p ∈ R
such that ( )
1
.a(u, v − u) + p(v − u) ≥ 0,
2

for all .v ∈ K ∗ , where { 1

a(u, v) =
. u ,, v,, d x.
0

Calculate .u then . p according to the derivatives of .u.

3. Consider the two previous questions with
{ 1
. J (v) = ((v,, )2 + (v, )2 ) d x.
0

4. We still want to go back to

{ 1
. (v,, )2 d x.
0

ntry to check the value of .v in . 2 . We pose the problem: Find .U = (-

u, -
α) ∈
1
For that, we
∗
(K × R) K , such that

. E(U ) = inf{E(W ), W ∈ K },

where { ( ) }
1 ∗
.K = W = (w, α) ∈ K × R such as w =b+α
2

and { 1
. E(W ) = (w,, )2 d x + να 2 .
0

Check that this problem has a unique solution .U ∈ K . By penalizing the condition
( )
1
.w = b + α,
2

try to highlight a state aide . P. Study as a function of .ν the relation between

{ 1
. (u ,, )2 d x
0

and { 1
. u ,, )2 d x,
(-
0

@seismicisolation
@seismicisolation
10.2 Problems and Exams 739

where .u is the solution found in 1 and .-

u is the solution found in 4. ♦

Exercise 10.44 Let . P = P3 . We consider a finite element of the Serendipity family:

-2 -12
1A
A -1
A

≫
R

-23
A -14
A
≫
−1 1 ξ

-3
A -34
−1 A -4
A

1. Is there a unique polynomial . p ∈ P taking values given at points . A-i .1 ≤ i ≤ 4,

-
and . Ai j .1 ≤ i < j ≤ 4 and the means of the third derivatives .vξ and .vη of . ∂∂x 3 and
3

∂3
. 3 . Indication: we can show that
∂y
ξ η(1 + ξ )(1 + η) - ξ η(1 − ξ )(1 + η) - ξ η(1 − ξ )(1 − η) -
. p(ξ, η) = p( A1 ) − p( A2 ) + p( A3 ) −
4 4 4
ξ η(1 + ξ )(1 − η) - (ξ + 1)(η − 1)(1 + η) - (ξ + 1)(η + 1)(ξ − 1) -
p( A4 ) − p( A14 ) − p( A12 ) +
4 2 2
(ξ − 1)(η − 1)(1 + η) - (ξ + 1)(η − 1)(ξ − 1) - (ξ − ξ )
3 (η − η)
3
p( A23 )− p( A34 )+ vξ + vη .
2 2 6 6
2. Show that, we have a non-conforming finite element.
3. Let .Ω be a rectangle cut into rectangles. We consider the following problem: Find
.u ∈ H0 (Ω) such that
1

a(u, v) = L(v)

for all .v ∈ H01 (Ω), with {

.a(u, v) = ∇u · ∇v d x
Ω

and {
. L(v) = f v d x.
Ω

@seismicisolation
@seismicisolation
740 10 Revision Issues

Let . FR ∈ Q 21 be a transformation given by

1 1 1 1
.x = (1 + ξ )(1 + η)x1 + (1 − ξ )(1 + η)x2 + (1 − ξ )(1 − η)x3 + (1 + ξ )(1 − η)x4
4 4 4 4

1 1 1 1
.y = (1 + ξ )(1 + η)y1 + (1 − ξ )(1 + η)y2 + (1 − ξ )(1 − η)y3 + (1 + ξ )(1 − η)y4 .
4 4 4 4

-i ) = Ai .1 ≤ i ≤ 4, . FR ( A
Show that . FR ( A -i j ) = Ai j .1 ≤ i < j ≤ 4 and . FR ( R)
- = R.
4. We set .-
v(ξ, η) = v(x(ξ, η), y(ξ, η)) and . PR = { p = - −1
p ◦ FR , - -
p ∈ P}. Consider
E{
a (u h , vh ) =
. h ∇u h ∇vh d xd y
R⊂Ω R

and {
. L(vh ) = f vh d xd y,
Ω

with .u h and .vh ∈ Vh with

{
. Vh := vh ∈ L (Ω, R) such as vh |R ∈ P3 (R), vh is entirely determined by the
2

. values at the vertices and midpoints of the sides and by the means of third derivat-
∂3
ives 3
∂x
∂3 }
. and 3 on each element R ., vh = 0 at vertices and midpoints of sides ⊂ ∂Ω .
∂y
Show that the mapping
1
v −→ ||vh ||h := (ah (vh , vh )) 2
. h

is a norm over .Vh .

5. Deduce that the problem: Finding .u h ∈ Vh such that

a (u h , vh ) = L(vh )
. h

for all .vh ∈ Vh admits a unique solution.

Δy
6. We set .h = max(Δx, Δy) and we assume that .c < Δx
< C. Show that

a (u − rh u, u − rh u) ≤ ch 3 |u|4,R ||u h − rh u||h .

. h

7. Show that .||u − rh u||h ≤ c(h|u|2,Ω + h 3 |u|4,Ω ).

3
8. Show that .||u − u h ||h ≤ ch 3 |u|4,Ω + ch 2 |u|4,Ω + ch 2 |u|3,Ω .
Exercise 10.45 We consider the problem

d 2uε
. −ε + u ,ε + αu ε = f on ]0.1[,
dx2

@seismicisolation
@seismicisolation
10.2 Problems and Exams 741

u (0) = 0, u ε (1) = 1,
. ε

where .ε > 0 is intended to tend towards zero and where .α is a strictly positive
constant or zero.
1. We set .a(ε)u ,ε = vε , where the function

.ε −→ a(ε)

is to be chosen. Write the preceding equation as a symmetric Friedrichs system,

positive if possible. Express the matrices . Mε (0) and . Mε (1) to take into account in an
admissible fashion the boundary conditions. Give as a function of .α and .ε surcharges
for .||u ε ||0,(0,1) , .||vε ||0,(0,1) , etc. What can be said when .ε → √
0.
2. We assume that .α = 0 and that we have chosen .a(ε) = ε. Can we premultiply
the system to make it (symmetric) and positive within the meaning of Friedrichs,
while expressing in an admissible way the boundary conditions.
3. We assume that .α = 0 and . f = 0. Give the expression of .u ε (x) as a function of .x
and .ε. Study the limit of .u ε , .ε → 0.
Consider the problem
,
.u = 0 0 < x < 1, (10.2.1)

u(0) = 0.
. (10.2.2)

Study .u ε − u. How to best define the matrices . Mε (0) and . Mε (1) as a function of
.a(ε) so that when .ε → 0 the condition at the limit of the problem (10.2.1)–(10.2.2)
is respected. √
4. We assume that . f = 0, .α = 0, and .a(ε) = ε. Without premultiplying the
obtained Friedrichs system, give a discretization using continuous finite elements
. P1 for .u ε and .vε . Study the corresponding linear system.
Then, give a discretization using discontinuous finite elements . P1 or . P0 (your
choice) for .u ε and .vε . Study the linear system.

Exercise 10.46 Consider the problem

d 4u
. − αu ,, + βu = f on ]0.1[,
dx4

u(0) = u , (0) = u(1) = u , (1) = 0.

.

1. Write this problem in the form of a Friedrichs system; is it positive. Without pre-
multiplying to make it positive, can we express admissible the boundary conditions?
Can we get an existence theorem.
2. Try to premultiply to make the system positive.

@seismicisolation
@seismicisolation
742 10 Revision Issues

Exercise 10.47 We consider the problem

du
. = f on ]0.1[,
dx

u(0) = γ ,
.

where .γ ∈ R is given.
1. Is this problem symmetrical positive within the meaning of Friedrichs. Can the
boundary conditions be taken into account admissiblely.
2. We premultiply the equation by .a(x). Give conditions on .a(x) so that the problem
becomes positive. Can we always express the condition in .x = 0.
3. We choose .a(x) = α(1 − x) + βx. Study as a function of .α and .β the positivity
of the system and the possibility of taking into account the boundary conditions.
4. We assume that .α = 1 and .β = 0. Write explicitly the numerical method when
using polynomials of degree .≤ 1 continuous.
5. With .α = L, .β = 0, study a discontinuous method.

Exercise 10.48 1. We set . Au = du

dx
, . A∗ u = − ddux , and

. W = {z ∈ H 1 (0, 1) such that z(1) = 0}.

We pose the problem: Find .u ∈ L 2 ((0, 1), R) such that

{ 1
.(u, A∗ z) = f zd x (10.2.3)
0

for all .z ∈ W . Let

. V = {A∗ z, z ∈ W }

and the map

.v −→ L(v)

from .V into .R, defined by

{ 1
. L(v) = f zd x,
0

where .v = A∗ z. Finish the proof of existence of a solution.

2. Let
. Z h = {ϕ j , functions hats and half hats}

and
. Wh = {z h ∈ Z h such that z h (1) = 0}.

Study the problems: Find .u h ∈ Z h such that

@seismicisolation
@seismicisolation
10.2 Problems and Exams 743
{ 1
∗
(u h , A z h ) =
. f z h d x + γ z h (0)
0

for all .z h ∈ Wh . Find .u h ∈ Z h such that

{ 1
1 1 1
. (Au h , z h ) + (u h , A∗ z h ) + ((u h − γ )z h )(0) + (u h z h )(1) = f zh d x
2 2 2 0

for all .z h ∈ Wh . Write the corresponding matrices for these problems; invertible or
not (discuss).
3. Write for the problem (10.2.3) a discontinuous method using polynomials of . P0
at intervals. What can we say then.

Exercise 10.49 Consider the problem

. − u ,,, + u = f, 0 < x < 1.

1. Put this problem in the form of a symmetrical system of the first order. Is this
system positive. If not, can we make it positive by premultiplying it with a suitable
matrix?
2. Without premultiplying, can we take into account the boundary conditions

.u(0) = u , (0) = u(1) = 0.

3. Can we show an existence theorem of a weak solution.

@seismicisolation
@seismicisolation
Index

Symbols .h(T ), 234, 239

./\u, 107 . j, 86
.T-, 235, 236 2
.l , 21, 75
. Vh , 303 .λi (M), 226
.| · |h,k , 544 .λi (x, y), 226
.| · |h,K ,k , 544 .<·, ·>a , 75, 188
.| · |m, p,g , 111 .<·, ·>h,k , 544
.|v|1,g , 97 .<·, ·>h,K ,k , 544
.|v|k+1,O , 109 .<·, ·>h,S,k , 544
∂u .<·>h,k , 544
. , 107
∂ν .<·>h,S,k , 544
t
. O, 77 .< f, g>0,g , 27
.δa , 83 .<u, v>1,g , 92
.δnm , 36 .<u, v>m,g , 104
.δ, 88 .[·]m, p,g , 112
.det A, 39 .(·, ·)2 , 381
.dim X , 18 .( f, u), 112
.dist(x, y), 19 .(ϕm )m converges to .ϕ on .D (g), 82
.||c||∞ , 172 .(u, ϕ)h , 487
∗
.|| · ||k+1,O , 109 .(u, v)2 , 381
.|| · ||m, p,g , 111 .(u, v)(L 2 (g,R)) p , 381
.|| f ||0,g , 28 .(H 1), 258
∗
.|| f ||m, p,g , 112 .(H 2), 433
.|| f || L ∞ (g,R) , 27 .(H 3), 435
.|| f || L p (g,R) , 27 .(H 4), 540
∗
.||L||k+1,O , 109 .(H 5), 540
.||T ||, 19 .(H 6), 540
.||vh ||h , 281 .(H 7), 546
.||v||1,g , 92 .(H 8), 546
.||v||m,g , 104 ,
.(Tm )m converges to . T on .D (g), 83
.||x||1 , 18 .l, 231
.||x||2 , 18 .∇u · ∇v, 92, 123
.||x||∞ , 19 .n-simplexes of type .k, 184
.γ0 , 101 . p F (x), 35
.γ1 , 106
α
.∂ ϕ, 82
.γ , 106
α
.∂ T , 84
.h K , 178
α
.∂ v, 110
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer
Nature Singapore Pte Ltd. 2024 745
A. Jeribi, Problems in Finite Element Methods, Infosys Science Foundation Series,
@seismicisolation
https://doi.org/10.1007/978-981-97-5710-7
@seismicisolation
746 Index

.ϕ j , 195, 196, 201, 230 . Wh , 194, 200, 227, 275

.r h v, 201, 304 .W
m, p (g),111
,,
.r R v, 304 . X , 20
o
.ρ K , 178 .X = M N , 30
,
.ρ(T ), 234, 239 . X , 20
.D (g), 82 . X -elliptical, 59
,
.D (g), 83 . Z K , 181
.L(X, Y ), 19 . x n → x, 18
.Mn (R), 39 5-point Laplacian equation, 280
.On (R), 40
.span(V ), 30
.tr ( A), 39 A
.τh , 224, 272 Additional orthogonal, 35
⊥
.u , 31 Adjacent, 181
.u⊥v, 30 Adjoint boundary conditions, 10, 380
. B(a, r ), 82 Adjoint matrix, 9, 380
.C ([a, b], R), 22
0 Approximate solution, 186
1
.C by parts, 379 Arithmetic-geometric mean inequality, 24
.C (g, R), 109
s A side of . K i+1 is attached to . K i , 533
. FT , 235 A side of . K i is attached to . K i+1 , 533
. H0 (g), 95
1 A triangulation is of class .C 0 , 183
. H0 (g), 108
2 Aubin-Nitsche lemma, 191
Aubin Nitsche’s duality process, 219, 248,
. H (g), 92
1
287
. H (g), 104
m

. I m(U ), 32
. J K , 180
B
. JS , 180
Banach space, 19
. JT , 285
Banach-Steinhaus theorem, 19
. K er (A), 79
Barycentric coordinates, 226
. K er (U ), 32
Basic functions of the finite element, 176
. L (g, R), 27
p
Basis, 18
. L (]a, b[, R), 22
2
Bilinear form, 20
. L (g, R), 27
2
⊥
Boundary conditions are admissible, 393
. M , 31 Bounded, 19
. P1 , 26 Bounded linear functionals, 20
. P2 , 26
. P3 , 27
. Pk , 109 C
. PK f , 43 Canonical injection, 86
k
. PK , 542 Cauchy-Schwarz inequality, 24
. PM ⊥ , 47 Cauchy sequence, 18
. P-interpolated function, 179 Class .C 0 , 182, 183
. P-unisolvent, 175 Classical solution, 380
. Q 1 , 266 Coercive, 59, 380
. Q k , 109, 184, 508 Coercivity constant, 59
. T f , 84 Compact, 19
. Ti , 224, 272 Complete, 18
. TH , 88 Cone, 582
,
. T , 83 Conforme, 3
. Vh , 195, 201, 227, 275 Conforming finite element, 468, 472
. Vh -interpolated function, 182, 201 Continuous, 59
0
. V , 79 Continuous linear functionals, 20
,
. Vh , 303 Continuous methods, 10

@seismicisolation
@seismicisolation
Index 747

Contraction, 54 Green’s formula, 103, 382

Convex, 23
Coordinate of .x, 33
Corner singularities at re-entrant corners, H
162 Hahn-Banach, 20
Hats functions, 195
Heaviside function, 88
D Hereafter element of reference, 179
Defined positive, 39 Hilbert basis, 36
Derivative within the meaning of distribu- Hilbert space, 22
tions, 83 Homogeneous Dirichlet boundary condi-
Direct complement, 30 tion, 117
Direct decomposition, 30 Homogeneous Neumann boundary condi-
Direct method of the calculus of variations, tion, 130, 135
66
Direct sum, 30
Dirichlet conditions, 159 I
Dirichlet problem, 117 Indirect method, 66
Dirichlet’s principle, 124 Inf-sup-condition, 189
Discontinuous methods, 10 Inner product, 21
Discontinuous Petrov Galerkin method, 597 Isometry, 23
Isomorphism, 80
Distance, 35
Distance preserving map, 23
Distribution space, 83
L
Dual, 112
Lagrange finite element, 175
Duality, 112
Laplacian of .u, 107
Dual norm, 112
Laplacian operator, 6
Dual space, 20
Lax-Milgram theorem, 73
Limit, 18
Linear combination, 30
E
Linear finite element of Lagrange, 224
Edges of the triangulation, 233
Element of reference, 179 Linearity, 21
Elliptic, 158 Linearly dependent, 18
Energy norm, 59 Linearly independent, 18, 30
Equivalent, 179 Linearly independent family, 30
Estimates for the errors, 200, 236 Linear operator, 19
Euclidean space, 21
Euler equation, 124
Euler-Lagrange equation, 124 M
Euler’s retrograde method, 477 Macro-elements method, 3
Mass of Dirac, 83, 88
Maximal, 37
F Mixed problem, 134
Friedrichs symmetric systems, 379 Monotone, 40
Frontier, 532

N
G Neumann boundary condition, 135
Galerkin method, 186 Neumann conditions, 162
Galerkin orthogonality, 188 Neumann problem, 130, 137
Gram-Schmidt orthonormalization method, Nodes, 175
36 Nodes of triangulation, 182, 230, 233

@seismicisolation
@seismicisolation
748 Index

Non-conforming finite element method, 3, 8, Range space, 32

275, 300, 342, 348, 365, 374 Reference hypercube, 184
Norm, 18 Reference segment, 202
Normed space, 18 Reference triangle, 236
Normed vector space, 19 Reflexive, 20
Not . Q 1 -unisolvent, 321 Regular, 97, 118, 178, 258
Null space, 32 Regular distribution, 85
Numbering algorithm, 233 Riesz Fréchet’s representation theorem, 55
Robin’s condition, 163, 164

O
Orthogonal, 30 S
Orthogonal family, 30 Schwarz distribution space, 83
Orthogonal group, 40 Second dual space, 20
Orthogonal matrix, 40, 77 Semi-admissible, 384
Orthogonal of . M, 31 Semi-frontier, 532
Orthogonal of .u, 31 Separable, 19
Orthogonal projection, 35, 43 Simpliciel type, 178
Orthonormal basis, 33 Sobolev space of order .1, 92
Orthonormal family, 33 Sobolev space of order .m, 104
Sobolev spaces, 2
Span, 30
P Stampacchia theorem, 60
Parallelogram, 176 Strictly convex, 23
Parallelogram identity, 25 Strip matrix, 231
Parallelotope, 176 Strong formulation, 91
Patch test, 3, 309 Strong problem, 117
PDEs, 158 Strong solution, 380
Penalization, 473 Subspace, 18
Petrov Galerkin-type methods, 519 Symbol of Kronecker, 36
Poincaré inequality, 95 Symmetric, 21
Poincaré-Wirtinger inequality, 96 Symmetry, 21
Poisson equation, 122 System stiffness matrix, 233
Polar, 79
Polarization identity, 26
Polygonal domain, 224, 271, 288 T
Polyhedral open set, 181 Theorem of Stampacchia, 60
Positive in the sense of Friedrichs, 383 Third boundary condition, 164
Positive semi-definite, 39 Topological dual, 20
Positivity, 21 Trace mapping, 101
Prehilbertian space, 21 Trace of .v on .l, 101
Projection, 34, 40 Trace theorem, 98, 101
Projection theorem on a closed convex, 41 Transpose of the matrix, 77
Pseudo-topological, 82, 83 Triangles, 224, 272
Pythagorean relation, 26, 30 Triangular inequality, 25
Triangulation, 181, 224, 235, 272
Truncation error, 199
Q
Quadrilateral, 176
Quadrilateral of type .k, 184 U
Uniformly elliptic, 157
Uzawa algorithm, 475
R
Range, 79

@seismicisolation
@seismicisolation
Index 749

V Vertices, 230
Variational equation, 68
Variational formulation, 118
Variational inequality, 68 W
Vector normed, 33 Weak approach, 6
Vectors, 17 Weak formulation, 6, 91, 118
Vector space, 17 Weakly convergent, 29
Vector unitary, 33 Weak solution, 91, 118, 385
Vertex numbering, 233 Wilson Brick, 300

@seismicisolation
@seismicisolation