Studies in Systems, Decision and Control 111

Michael Z. Zgurovsky
Pavlo O. Kasyanov

Qualitative and
Quantitative
Analysis of
Nonlinear Systems
Theory and Applications
Studies in Systems, Decision and Control

Volume 111

Series editor
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
e-mail: [email protected]
About this Series

The series “Studies in Systems, Decision and Control” (SSDC) covers both new
developments and advances, as well as the state of the art, in the various areas of
broadly perceived systems, decision making and control-quickly, up to date and
with a high quality. The intent is to cover the theory, applications, and perspectives
on the state of the art and future developments relevant to systems, decision
making, control, complex processes and related areas, as embedded in the fields of
engineering, computer science, physics, economics, social and life sciences, as well
as the paradigms and methodologies behind them. The series contains monographs,
textbooks, lecture notes and edited volumes in systems, decision making and
control spanning the areas of Cyber-Physical Systems, Autonomous Systems,
Sensor Networks, Control Systems, Energy Systems, Automotive Systems,
Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace
Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power
Systems, Robotics, Social Systems, Economic Systems and other. Of particular
value to both the contributors and the readership are the short publication timeframe
and the world-wide distribution and exposure which enable both a wide and rapid
dissemination of research output.

More information about this series at http://www.springer.com/series/13304

Michael Z. Zgurovsky Pavlo O. Kasyanov
•

Qualitative and Quantitative

Analysis of Nonlinear
Systems
Theory and Applications

123
Michael Z. Zgurovsky Pavlo O. Kasyanov
Igor Sikorsky Kyiv Polytechnic Institute Institute for Applied System Analysis
Kyiv Igor Sikorsky Kyiv Polytechnic Institute
Ukraine Kyiv
Ukraine

ISSN 2198-4182 ISSN 2198-4190 (electronic)

Studies in Systems, Decision and Control
ISBN 978-3-319-59839-0 ISBN 978-3-319-59840-6 (eBook)
DOI 10.1007/978-3-319-59840-6
Library of Congress Control Number: 2017944288

© Springer International Publishing AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
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, express 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.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
“A mathematician is a person who can find
analogies between theorems; a better
mathematician is one who can see analogies
between proofs and the best mathematician
can notice analogies between theories.”
Stefan Banach
Preface

Scope

In an abstract form, the evolutional nonlinear system is a mathematical model that

describes how physical, chemical, biological, economic, or even mathematical
phenomena evolve in time. As a rule, it contains ordinary/partial/stochastic
differential equations or inclusions that tell us how the system at hand changes
“from one instant to the next.” The main goal is to gain information about solutions
of this system and then translate this mathematical information into the scientific
context. The main challenge addressed by this book is to take this short-term
information and obtain information about long-term overall behavior. The study of
nonlinear systems has three parts: exact methods, quantitative methods and quali-
tative methods. But even if we solve the system symbolically, the question of
computing values remains.
In this book, we concentrate on the following topics, specific for nonlinear
systems:
(a) constructive existence results and regularity theorems for all weak (generalized)
solutions;
(b) convergence results for solutions and their approximations in strongest
topologies of the natural phase and extended phase spaces;
(c) uniform global behavior of solutions in time;
(d) pointwise behavior of solutions for autonomous problems with possible gaps by
the phase variables.
With numerous applications including nonlinear parabolic equations of diver-
gent form, parabolic problems with nonpolynomial growth, nonlinear stochastic
equations of parabolic type, unilateral problems with possibly nonmonotone
potential, nonlinear problems on manifolds with or without boundary, contact
piezoelectric problems with nonmonotone potential, viscoelastic problems with
nonlinear “reaction-displacement” and “reaction-velocity” laws as well as particular
examples like a model of conduction of electrical impulses in nerve axons, a climate

vii
viii Preface

energy balance model, FitzHugh–Nagumo system, Lotka–Volterra system with

diffusion, Ginzburg–Landau equations, Belousov–Zhabotinsky equations, and the
3D Navier–Stokes equations. This book is also distinguished with the solutions of a
number of applied problems in physics, chemistry, biology, economics, etc.

Contents

This book consists of three parts: Existence and Regularity Results, Quantitative
Methods and Their Convergence (Part I), Convergence Results in Strongest
Topologies (Part II), and Uniform Global Behavior of Solutions: Uniform
Attractors, Flattening and Entropy (Part III). Part I presents several numerical
methods for approximate solution of nonlinear systems, their convergence, and
regularity results and also discusses recent advances in regularity problem for the
3D Navier–Stokes equations. Part II covers three major topics: (1) strongest con-
vergence results for weak solutions of nonautonomous reaction–diffusion equations
with Carathéodory's nonlinearity with applications to FitzHugh–Nagumo systems,
Lotka–Volterra systems with diffusion, Ginzburg–Landau equations, Belousov–
Zhabotinsky equations, etc; (2) strongest convergence results for weak solutions of
feedback control problems with applications to impulse feedback control
mechanical problems and mathematical problems of biology and climatology; and
(3) strongest convergence results for weak solutions of differential-operator equa-
tions and inclusions with applications to nonlinear parabolic equations of divergent
form, parabolic problems with nonpolynomial growth, nonlinear stochastic equa-
tions of parabolic type, general parabolic and hyperbolic problems, unilateral
problems with possibly nonmonotone operators, etc. Part III discusses general
methodology for the global qualitative and quantitative investigation of dissipative
dynamical systems, first- and second-order operator differential equations and
inclusions, and evolutional variational inequalities with possibly nonmonotone
potential with several applications. Indirect Lyapunov method for autonomous
dynamical systems, exponential attractors, and Kolmogorov entropy are also
established. All case studies are closely related to theoretical Parts I and II and are
examples of applications to solutions of problems (a) and (b).

Audience

This book is aimed at practitioners working in the areas of nonlinear mechanics,

mathematical biology, control theory, differential equations, nonlinear boundary
value problems, and decision making. It can serve as a quick introduction into the
novel methods of qualitative and quantitative analysis of nonlinear systems for the
graduate students, engineers, and mathematicians interested in analysis and control
of nonlinear processes and fields, mathematical modeling, and dynamical systems
Preface ix

in infinite-dimensional spaces, to mention just a few. It can also be used as a

supplementary reading for a number of graduate courses including but not limited
to those of nonlinear PDEs, control and optimization, stochastic partial differential
equations, advanced numerical methods, systems analysis, and advanced engi-
neering economy.

Acknowledgements

We express our gratitude to editors of the “Springer” Publishing House for the
constant support of our publishing initiatives and everybody who took part in
preparation of the manuscript. We thank Distinguished Profs. Joseph S.B. Mitchell
and Eugene A. Feinberg, the Stony Brook University, USA; Prof. Oleksiy V.
Kapustyan, National Taras Shevchenko University of Kyiv, Ukraine; and Professor
José Valero Cuadra, Universidad Miguel Hernández de Elche, Spain, for providing
valuable comments. We want to express the special gratitude to Olena L. Poptsova,
Nataliia V. Gorban, Liliia V. Paliichuk, and Olha V. Khomenko, Institute for
Applied System Analysis at the Igor Sykorsky Kyiv Polytechnic Institute, Ukraine,
for a technical support of this manuscript.

Kyiv, Ukraine Michael Z. Zgurovsky

February 2017 Pavlo O. Kasyanov
Contents

Part I Existence and Regularity Results, Quantitative Methods

and Their Convergence
1 Qualitative Methods for Classes of Nonlinear Systems:
Constructive Existence Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 First Order Differential-Operator Equations and Inclusions . . . . . . . 3
1.1.1 Setting of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Main Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.3 Special Basis and Approximations for Multi-valued
Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 8
1.1.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 10
1.2 Second Order Operator Differential Equations and Inclusions .... 30
1.3 Evolutional Variational Inequalities: Penalty Method
and Strong Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 33
1.4 Nonlinear Parabolic Equations of Divergent Form . . . . . . . . . .... 42
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 43
2 Regularity of Solutions for Nonlinear Systems . . . . . . . . . . . . . . .... 47
2.1 Regularity of All Weak Solutions for a Parabolic Feedback
Control Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 47
2.2 Artificial Control Method for Nonlinear Partial Differential
Equations and Inclusions: Regularity of All Weak Solutions . .... 50
2.3 Regularity of All Weak Solutions for Nonlinear
Reaction-Diffusion Systems with Nonlinear Growth . . . . . . . . . . . . 52
2.3.1 Reaction-Diffusion Equations . . . . . . . . . . . . . . . . . . . . . . . 52
2.3.2 Systems of Reaction-Diffusion Equations . . . . . . . . . . . . . . 55
2.4 Examples of Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.4.1 A Parabolic Feedback Control Problem . . . . . . . . . . . . . . . 60
2.4.2 A Model of Conduction of Electrical Impulses
in Nerve Axons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 61
2.4.3 Climate Energy Balance Model . . . . . . . . . . . . . . . . . . .... 61

xi
xii Contents

2.4.4 FitzHugh–Nagumo System . . . . . . . . . . . . . . . . . . . . . . . . . 64

2.4.5 A Model of Combustion in Porous Media . . . . . . . . . . . . . 65
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3 Advances in the 3D Navier-Stokes Equations . . . . . . . . . . ......... 69
3.1 Weak, Leray-Hopf and Strong Solutions . . . . . . . . . . . ......... 69
3.2 Leray-Hopf Property for a Weak Solution of the 3D
Navier-Stokes System: Method of Artificial Control . . . ......... 72
3.3 The Existence of Strong Solutions and 1-Dimensional
Dynamical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . ......... 74
3.4 Extremal Solutions: Existence and Continuity Results
in Strongest Topologies . . . . . . . . . . . . . . . . . . . . . . . . ......... 79
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ......... 86

Part II Convergence Results in Strongest Topologies

4 Strongest Convergence Results for Weak Solutions
of Non-autonomous Reaction-Diffusion Equations
with Carathéodory’s Nonlinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.1 Translation-Compact, Translation-Bounded and Translation
Uniform Integrable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.2 Setting of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3 Preliminary Properties of Weak Solutions . . . . . . . . . . . . . . . . . . . 91
4.4 Strongest Convergence Results in C Loc ðR þ ; HÞ . . . . . . . . . . . . . . . 95
4.5 Strongest Convergence Results for Solutions in the Natural
Extended Phase Space ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.6 Examples of Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.6.1 Non-autonomous Complex Ginzburg–Landau Equation . . . 106
4.6.2 Non-autonomous Lotka–Volterra System
with Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5 Strongest Convergence Results for Weak Solutions
of Feedback Control Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 111
5.1 Setting of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 111
5.2 Regularity of All Weak Solutions and Their Additional
Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.3 Convergence of Weak Solutions in the Strongest Topologies . . . . . 113
5.4 Examples of Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.4.1 A Model of Combustion in Porous Media . . . . . . . . . . . . . 116
5.4.2 A Model of Conduction of Electrical Impulses
in Nerve Axons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 117
5.4.3 Climate Energy Balance Model . . . . . . . . . . . . . . . . . . .... 118
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 122
Contents xiii

6 Strongest Convergence Results for Weak Solutions

of Differential-Operator Equations and Inclusions . . . . . . . . . . . .... 125
6.1 First Order Differential-Operator Equations and Inclusions . . . .... 125
6.1.1 Convergence Results for Autonomous Evolution
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 125
6.1.2 Convergence Results for Nonautonomous Evolution
Inclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.2 Second Order Operator Differential Equations and Inclusions . . . . 139
6.3 Examples of Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.3.1 Nonlinear Parabolic Equations of Divergent Form . . . . . . . 151
6.3.2 Nonlinear Non-autonomous Problems on Manifolds
with and Without Boundary: A Climate Energy
Balance Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 152
6.3.3 A Model of Conduction of Electrical Impulses
in Nerve Axons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 153
6.3.4 Viscoelastic Problems with Nonlinear
“Reaction-Displacement” Law . . . . . . . . . . . . . . . . . . . .... 154
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 157

Part III Uniform Global Behavior of Solutions: Uniform

Attractors, Flattening and Entropy
7 Uniform Global Attractors for Non-autonomous Dissipative
Dynamical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.1 General Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.2 Main Constructions and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.3 Proof of Theorem 7.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
7.4 Example of Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.4.1 Autonomous Evolution Problem . . . . . . . . . . . . . . . . . . . . . 170
7.4.2 Non-autonomous Evolution Problem . . . . . . . . . . . . . . . . . . 170
7.4.3 Non-autonomous Differential-Operator Inclusion. . . . . . . . . 171
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
8 Uniform Trajectory Attractors for Non-autonomous Nonlinear
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
8.1 Uniform Trajectory Attractor for Non-autonomous
Reaction-Diffusion Equations with Carathéodory’s
Nonlinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
8.2 Structure of Uniform Global Attractor for Non-autonomous
Reaction-Diffusion Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
8.3 Uniform Trajectory Attractors for Nonautonomous Dissipative
Dynamical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
8.4 Notes on Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
xiv Contents

9 Indirect Lyapunov Method for Autonomous Dynamical Systems . . . 211

9.1 First Order Autonomous Differential-Operator Equations
and Inclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
9.2 Second Order Autonomous Operator Differential Equations
and Inclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.3 Examples of Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
9.3.1 A Model of Combustion in Porous Media . . . . . . . . . . . . . 220
9.3.2 A Model of Conduction of Electrical Impulses
in Nerve Axons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
9.3.3 Viscoelastic Problems with Nonlinear
“Reaction-Displacement” Law . . . . . . . . . . . . . . . . . . . . . . . 221
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Introduction: Special Classes of Extended Phase
Spaces of Distributions

Abstract In this introduction, we briefly establish special classes of extended phase

spaces of distributions. We consider sums and intersections of Banach spaces;
Gelfand triples; special classes of Bochner integrable functions; generalized
derivatives; and basic properties of extended phase spaces.

If it is necessary to describe a nonstationary process that evolve in some domain

X  Rn during the time interval ½s; T, we may deal with functions that correspond
to each pair fx; tg 2 X  S the real number or vector uðx; tÞ. In this approach, the
time and the space variables are equivalent. But there is a more convenient
approach to the mathematical description for evolution processes [1, 2]: For each
point in time t, it is mapped the state function uð  ; tÞ (e.g., for each point of time
we put the temperature distribution or velocity distribution in the domain X)

differential - operator equation / inclusion

mathematical model ðPDEÞ y : ½s; T ! ðX ! Rm Þ
physical process or field ) )
y : X  ½s; T ! Rm yðt; : 2 ðX ! Rm Þ
t 2 ½s; T 

Thus, we consider the functions defined on ½s; T with values in the state
functions space (e.g., in the space H01 ðXÞ). Therefore, to investigate the evolution
problem, it is natural to consider the space of functions acting from the time interval
½s; T into some infinite-dimensional space V. In particular, it is natural to consider
the spaces of integrable and differentiable functions. In this book, we consider only
real vector spaces.

xv
xvi Introduction: Special Classes of Extended Phase Spaces of Distributions

In this chapter, we introduce the classes of function spaces used for qualitative
and quantitative analysis of nonlinear distributed systems:

Lu þ AðuÞ 3 f ; u 2 DðLÞ; ð1Þ


where A :! 2X is possibly multi-valued mapping with nonempty values, X is a
Banach space, X  is its dual space, L : DðLÞ  X ! X  is a linear operator defined
on DðLÞ; and f 2 X  . Moreover, in this chapter, we refer to the basic properties for
this spaces (see, e.g., [1, 2] and references therein for details).
For Banach spaces X; Y; the following denotation

XY

means the embedding in both the set-theoretic and the topological senses.

The following two theorems are frequently used in the qualitative and
quantitative analysis of nonlinear systems in infinite-dimensional spaces. The
main idea is in the following: the uniform prior estimates for solutions of
approximative problems and the following theorems allow ones to obtain at
least weak convergence (up to a subsequence in the general situation) of these
approximations to the exact solution of the problem in hands.

Theorem 1 (The reflexivity criterion) A Banach space E is reflexive if and only

ifeach bounded in E sequence has a subsequence that weakly converges in E.

Stefan Banach (March 30, 1892–August 31, 1945) was a Polish, Ukrainian,
and Soviet mathematician who is generally considered one of the world’s
most important and influential twentieth-century mathematicians. He was one
of the founders of modern functional analysis and an original member of the
Lviv School of Mathematics. His major work was the 1932 book, Théorie des
opérations linéaires (Theory of Linear Operations), the first monograph on the
general theory of functional analysis (Fig. 1).
Born in Kraków, Banach attended IV Gymnasium, a secondary school, and
worked on mathematics problems with his friend Witold Wilkosz. After
graduating in 1910, Banach moved to Lviv. However, and during World
War I Banach returned to Kraków, where he befriended Hugo Steinhaus.
After Banach solved some mathematics problems which Steinhaus consid-
ered difficult, they published their first joint work. In 1919, with several other
mathematicians, Banach formed a mathematical society. In 1920, he received
an assistantship at the Lviv Polytechnic.
He soon became a professor at the Lviv Polytechnic and a member of the
Polish Academy of Learning. He organized the “Lviv School of
Introduction: Special Classes of Extended Phase Spaces of Distributions xvii

Fig. 1 Stefan Banach

Mathematics.” Around 1929, he began writing his Théorie des opérations

linéaires.
After the outbreak of World War II, in September 1939, Lviv was taken
over by the Soviet Union. Banach became a member of the Academy of
Sciences of Ukraine and was dean of Lviv University’s Department of
Mathematics and Physics.
In 1941, when the Germans took over Lviv, all institutions of higher
education were closed to Poles. As a result, Banach was forced to earn a
living as a feeder of lice at Rudolf Weigl's Institute for Study of Typhus and
Virology. While the job carried the risk of infection with typhus, it protected
him from being sent to slave labor in Germany and from other forms of
repression. When the Soviets recaptured Lviv in 1944, Banach re-established
the university. However, because the Soviets were removing Poles from
Soviet-annexed formerly Polish territories, Banach prepared to return to
Kraków. Before he could do so, he died in August 1945, having been
diagnosed seven months earlier with lung cancer.
Some of the notable mathematical concepts that bear Banach’s name
include Banach spaces, Banach algebras, the Banach–Tarski paradox, the
Hahn–Banach theorem, the Banach–Steinhaus theorem, the Banach–Mazur
game, the Banach–Alaoglu theorem, and the Banach fixed-point theorem.

Theorem 2 (Banach–Alaoglu) In reflexive Banach space, each bounded sequence

has a subsequence that weakly converges

Leonidas (Leon) Alaoglu (March 19, 1914–August 1981) was a mathe-

matician, known for his result, called Alaoglu’s theorem on the weak-star
compactness of the closed unit ball in the dual of a normed space, also known
as the Banach–Alaoglu theorem; Fig. 2.
xviii Introduction: Special Classes of Extended Phase Spaces of Distributions

Fig. 2 Leonidas (Leon)

Alaoglu’s grave

Sums and Intersections of Banach Spaces

Let us consider the sums and intersections of Banach spaces. Such objects naturally
appear under the investigation of number of anisotropic problems. Let n  2 be a
natural number and fXi gni¼1 be a family of Banach spaces. Let us introduce the
definition of the interpolation family of Banach spaces.
Definition 1 If there exists a vector topological space (LTS) Y such that

Xi  Y

for each i ¼ 1. . .n; then the family of Banach spaces fXi gni¼1 is called an inter-
polation family. If n ¼ 2; then the interpolation family is called an interpolation
pair.

In the field of mathematical analysis, an interpolation space is a space which

lies “in between” two other Banach spaces. The main applications are in
Sobolev spaces, where spaces of functions that have a noninteger number of
derivatives are interpolated from the spaces of functions with integer number
of derivatives.
The theory of interpolation of vector spaces began by an observation of
Józef Marcinkiewicz, later generalized and now known as the Riesz-Thorin
theorem. In simple terms, if a linear function is continuous on a certain space
Lp and also on a certain space Lq , then it is also continuous on the space Lr ,
for any intermediate r between p and q. In other words, Lr is a space which is
intermediate between Lp and Lq .
In the development of Sobolev spaces, it became clear that the trace spaces
were not any of the usual function spaces (with integer number of derivatives),
and Jacques-Louis Lions discovered that indeed these trace spaces were
constituted of functions that have a noninteger degree of differentiability.
Introduction: Special Classes of Extended Phase Spaces of Distributions xix

Further, let fXi gni¼1 be an interpolation family of Banach spaces. Similar to [1,
p. 23], we endow the vector space X ¼ \ ni¼1 Xi with the following norm:

X
n
kxkX :¼ kxkXi 8x 2 X; ð2Þ
i¼1

where k  kXi is the norm in Xi .

Proposition 1 Let fX; Y; Zg be an interpolation family. Then

X \ ðY \ ZÞ ¼ ðX \ YÞ \ Z ¼ X \ Y \ Z; X \Y ¼ Y \X

both in the sense of equality of sets and in the sense of equality of norms.
Let us consider also the vector space
( )
X
n X
n
Z :¼ Xi ¼ xi : xi 2 Xi ; i ¼ 1. . .n
i¼1 i¼1

with the norm

( )
X
n
kzkZ :¼ inf max kxi kXi : xi 2 Xi ; xi ¼ z 8z 2 Z: ð3Þ
i¼1...n
i¼1

Proposition 2 Let fXi gni¼1 be an interpolation family. Then X ¼ \ ni¼1 Xi and Z ¼

P n
i¼1 Xi are Banach spaces. Moreover,

X  Xi  Z ð4Þ

for each i ¼ 1. . .n:

Remark 1 Let Banach spaces X and Y satisfy the following conditions

X  Y; X is dense in Y;
kxkY  ckxkX 8x 2 X; c ¼ const :

Then

Y   X; kf kX   ckf kY  8f 2 Y  :

Moreover, if X is reflexive, then Y  is dense in X  .

xx Introduction: Special Classes of Extended Phase Spaces of Distributions

Józef Marcinkiewicz (March 30, 1910–1940) was a Polish mathematician.

He was a student of Antoni Zygmund and later worked with Juliusz Schauder
and Stefan Kaczmarz. He was a Professor of the Stefan Batory University in
Wilno.
Marcinkiewicz was taken as a Polish POW to a Soviet camp in
Starobielsk. The exact place and date of his death remain unknown, but it is
believed that he died in the Katyn massacre on the mass murder site near
Smolensk. His parents, to whom he gave his manuscripts before the begin-
ning of World War II, were transported to the Soviet Union in 1940 and later
died of hunger in a camp (Fig. 3).

Let fXi gni¼1 be an interpolation family. Assume that the Banach space X :¼
\ ni¼1 Xi ;
endowed with the norm defined in (2) is dense in Xi for each i ¼ 1. . .n.
Remark 1 yields that each space Xi may be considered as a subspace of X  . Therefore,
P
the vector space ni¼1 Xi is well-defined, and the following embedding holds:

Fig. 3 Józef Marcinkiewicz

Introduction: Special Classes of Extended Phase Spaces of Distributions xxi

X
n  
Xi  \ ni¼1 Xi : ð5Þ
i¼1

P
Since X is dense in Z :¼ ni¼1 Xi for each i ¼ 1. . .n; then each Xi is dense in Z:
According to Remark 1, the space Z  can be considered both as a subspace of Xi for
each i ¼ 1. . .n and as a subspace of \ ni¼1 Xi , that is,
!
X
n
Xi  \ ni¼1 Xi : ð6Þ
i¼1

G. Olof Thorin (February 23, 1912–February 14, 2004) was a Swedish

mathematician working on analysis and probability, who introduced the
Riesz–Thorin theorem (Fig. 4).

Fig. 4 G. Olof Thorin

xxii Introduction: Special Classes of Extended Phase Spaces of Distributions

Proposition 3 Let fXi gni¼1 be an interpolation family of Banach spaces such that
the space X :¼ \ ni¼1 Xi endowed with the norm (2) is dense in Xi for each i ¼ 1. . .n.
Then the following equalities hold:
!
X
n   X
n
Xi ¼ \ ni¼1 Xi and Xi ¼ \ ni¼1 Xi
i¼1 i¼1

in the sense of both equalities of sets and norms.

Sergei Lvovich Sobolev (October 6, 1908–January 3, 1989) was a Soviet

mathematician working in mathematical analysis and partial differential
equations. Sobolev introduced the notions that are now fundamental for
several areas of mathematics. Sobolev spaces can be defined by some growth
conditions on the Fourier transform. They and their embedding theorems are
an important subject in functional analysis. Generalized functions (later
known as distributions) were first introduced by Sobolev in 1935 for weak
solutions and further developed by Laurent Schwartz. Sobolev abstracted the
classical notion of differentiation, so expanding the range of application
of the technique of Newton and Leibniz. The theory of distributions is
considered now as the calculus of the modern epoch (Fig. 5).

Gelfand Triple

Let V be a real reflexive separable Banach space V with the norm k  kV and H be a
real Hilbert space with the inner product ð;  and respective norm k  kH . Assume
that

V  H; V is dense in H;
ð7Þ
9c [ 0 : kvkH  ckvkV 8v 2 V:

Remark 1 and conditions (7) yield that the dual space H  to H is a subspace
of the dual space V  to V. Since the Banach space V is reflexive and the set V is
dense in the space H; then the set H  is dense in the space V  and the following
inequality holds:

kf kV   ckf kH  8f 2 H  ;

where k  kV  and k  kH  are the norms in spaces V  and H  , respectively. By

applying the Riesz representation theorem, we can identify H  with H. Therefore,
Introduction: Special Classes of Extended Phase Spaces of Distributions xxiii

Fig. 5 Sergei Lvovich

Sobolev

H  is identified with some subspace of V  ; that is, each element y 2 H is identified

with some fy 2 V  such that

ðy; xÞ ¼ hfy ; xiV 8x 2 V;

where h; iV is the canonical pairing between V  and V. Since the elements y and fy
are identified, then conditions (7) imply that the restriction of the pairing h; iV on
H  V coincides with the inner product ð; Þ on H restricted on the same set. After
this identification of H and H  , we obtain the following tuple of the continuous and
dense embeddings

V  H  V :
xxiv Introduction: Special Classes of Extended Phase Spaces of Distributions

Definition 2 The tuple of spaces (V; H; V  ) satisfying the above conditions is

called the evolution triple (sometimes Gelfand triple).

Israel Moiseevich Gelfand (September 2 [O.S. 20 August], 1913–October 5,

2009) was a prominent Soviet and American mathematician. He made sig-
nificant contributions to many branches of mathematics, including group
theory, representation theory, and functional analysis. The recipient of many
awards, including the Order of Lenin and the Wolf Prize, he was a Fellow
of the Royal Society and Professor at Moscow State University and, after
immigrating to the USA shortly before his 76th birthday, at Rutgers
University (Fig. 6).
His legacy continues through his students, who include Endre Szemerédi,
Alexandre Kirillov, Edward Frenkel, and Joseph Bernstein, as well as his own
son, Sergei Gelfand.

Fig. 6 Israel Moiseevich

Gelfand
Introduction: Special Classes of Extended Phase Spaces of Distributions xxv

Special Classes of Bochner Integrable Functions

Let us consider classes of distributions with values in a Banach space. Let Y be a

real Banach space, Y  be its dual space, and S be a compact time interval. We
consider the classes of functions defined on S and taking values in Y (or in Y  ,
respectively).
In mathematics, the Bochner integral, named for Salomon Bochner, extends the definition
of Lebesgue integral to functions that take values in a Banach space, as the limit of integrals
of simple functions.

Let 1  p  þ 1. The set Lp ðS; YÞ of all Bochner measurable functions (see [1])
such that
Z 1=p
kykLp ðS;YÞ ¼ kyðtÞkpY dt \1
S

is a Banach space. If p ¼ þ 1; then the norm on L1 ðS; YÞ is defined as follows

k y kL1 ðS;YÞ ¼ ess supp k yðtÞ kY :

t2S

Salomon Bochner (August 20, 1899–May 2, 1982) was an American

mathematician of Austrian–Hungarian origin, known for work in mathe-
matical analysis, probability theory, and differential geometry (Fig. 7).

The following theorem establishes the sufficient conditions for the identification
of the dual space ðLp ðS; YÞÞ to Lp ðS; YÞ, 1  p\ þ 1, with Lq ðS; Y  Þ, where q is
such that p 1 þ q 1 ¼ 1. Sometimes, the following theorem is called the Riesz
representation theorem for spaces of Bochner integrable functions. We note that
1=1 :¼ 0.
Theorem 3 Let Y be a reflexive and separable Banach space, 1  p\ þ 1; and
q [ 1 be such that p 1 þ q 1 ¼ 1. Then for each f 2 ðLp ðS; YÞÞ there exists a
unique n 2 Lq ðS; Y  Þ such that
Z
f ðyÞ ¼ hnðtÞ; yðtÞiY dt
S
xxvi Introduction: Special Classes of Extended Phase Spaces of Distributions

Fig. 7 Salomon Bochner

for each y 2 Lp ðS; YÞ: Moreover, this correspondence f ! n is linear and

kf kðLp ðS;YÞÞ ¼ knkLq ðS;Y  Þ ;

that is, this mapping is isometric isomorphism.

Frigyes Riesz (January 22, 1880–February 28, 1956) was a Hungarian

mathematician who made fundamental contributions to functional analysis.
He was the Rector and a Professor at the University of Szeged, as well as a
member of the Hungarian Academy of Sciences. He was the older brother
of the mathematician Marcel Riesz (Fig. 8).

Let us consider the sums and intersections of Banach spaces of Bochner inte-
grable functions. These spaces are important for the investigation of nonlinear
Introduction: Special Classes of Extended Phase Spaces of Distributions xxvii

Fig. 8 Frigyes Riesz

anisotropic problems and the respective differential-operator equations and inclu-

sions. Let pi and ri , i ¼ 1; 2 be real numbers such that 1\pi  ri  þ 1 and
pi \ þ 1. Define the real numbers qi  ri0  1 as follows:

pi 1 þ qi 1
¼ ri 1 þ ri0 1 ¼ 1; i ¼ 1; 2:

Let (Vi ; H; Vi ), i ¼ 1; 2; be evolution triple such that

the set V1 \ V2 is dense in the spaces V1 ; V2 and H: ð8Þ

Consider the following Banach spaces (see Proposition 2):

Xi ¼ Xi ðSÞ ¼ Lqi ðS; Vi Þ þ Lri0 ðS; HÞ; i ¼ 1; 2

X ¼ XðSÞ ¼ Lq1 ðS; V1 Þ þ Lq2 ðS; V2 Þ þ Lr20 ðS; HÞ þ Lr10 ðS; HÞ

with the following respective norms

n
kykXi ¼ inf maxfky1 kLq ðS;V  Þ ; ky2 kLr 0 ðS;HÞ g :

i i i

: y1 2 Lqi ðS; Vi Þ; y2 2 Lri0 ðS; HÞ; y ¼ y1 þ y2 ;

xxviii Introduction: Special Classes of Extended Phase Spaces of Distributions

for all y 2 Xi , and

kykX ¼ inffmaxfky1i kLq ðS;V  Þ ; ky2i kLr 0 ðS;HÞ g : y1i 2 Lqi ðS; Vi Þ;
i¼1;2 i i i

y2i 2 Lri0 ðS; HÞ; i ¼ 1; 2; y ¼ y11 þ y12 þ y21 þ y22 g;

for each y 2 X.
If ri \ þ 1, then Proposition 1 and Theorem 3 imply that the space Xi is
reflexive. Similarly, if maxfr1 ; r2 g\ þ 1, then the space X is reflexive. Moreover,
for i ¼ 1; 2 the dual space Xi ¼ Xi ðSÞ, we identify with Lri ðS; HÞ \ Lpi ðS; Vi Þ,
where

kykX  ¼ kykLr ðS;HÞ þ kykLp ðS;Vi Þ

i i i

for each y 2 Xi : Similarly, for the dual space X  ¼ X  ðSÞ we identify with

Lr1 ðS; HÞ \ Lr2 ðS; HÞ \ Lp1 ðS; V1 Þ \ Lp2 ðS; V2 Þ;

where

kykX  ðSÞ ¼ kykLr ðS;HÞ þ kykLr ðS;HÞ þ kykLp ðS;V1 Þ þ kykLp ðS;V2 Þ
1 2 1 2

for each y 2 X  . The pairing on XðSÞ  X  ðSÞ is defined by

Z Z
hf ; yi ¼hf ; yiS ¼ ðf11 ðsÞ; yðsÞds þ ðf12 ðsÞ; yðsÞds
Z S
Z S Z
þ hf21 ðsÞ; yðsÞiV1 ds þ hf22 ðsÞ; yðsÞiV2 ds ¼ ðf ðsÞ; yðsÞÞds
S S S

for each f 2 X and y 2 X  ; where f ¼ f11 þ f12 þ f21 þ f22 , f1i 2 Lri0 ðS; HÞ,
f2i 2 Lqi ðS; Vi Þ, i ¼ 1; 2.
If maxfr1 ; r2 g\ þ 1; then we will always use the following “standard”
denotations [1, p. 171]: for the spaces X  , X1 , and X2 , we will denote as X, X1 , and
X2 , respectively, and vice versa; for the spaces X, X1 , and X2 , we will denote as X  ,
X1 , and X2 , respectively. These denotations are correct, because Proposition 3 and
Theorem 3 yield that these spaces and their dual spaces are reflexive. The following
statement directly follows from Proposition 3 and Theorem 3.
Proposition 4 If maxfr1 ; r2 g\ þ 1, then the Banach spaces X, X1 and X2 are
reflexive.
Introduction: Special Classes of Extended Phase Spaces of Distributions xxix

Generalized Derivatives

Let S be a time interval. The space DðSÞ of test functions on S is defined as follows.
A function u : S ! R is said to have compact support if there exists a compact
subset K of S such that uðxÞ ¼ 0 for all x 2 SnK. The elements of DðSÞ are the
infinitely differentiable functions u : S ! R with compact support—also known as
bump functions. This is a real vector space. It can be given a topology by defining
the limit of a sequence of elements of DðSÞ. A sequence fuk gk  1  DðSÞ is said to
converge to u 2 DðSÞ if the following two conditions hold:
(i) There is a compact set K  S containing the supports of all uk :
[ k supp ðuk Þ  K;
(ii) For each multi-index a, the sequence of partial derivatives @ a uk tends uni-
formly to @ a u:
With this definition, DðSÞ becomes a complete locally convex topological vector
space satisfying the Heine–Borel property.
Let Y be a real reflexive Banach space. The distribution on S with values in Y is a
continuous linear mapping acting from DðSÞ into Y endowed with the weak
topology. The space of all distributions on S with values in Y is denoted by D ðS; YÞ:
For each f 2 D ðS; YÞ, its generalized derivative f 0 is well defined as follows:

f 0 ðuÞ ¼ f ðu0 Þ

for each u 2 DðSÞ:

We note that each locally integrable in the Bochner sense function u (i.e.,
1 ðS; YÞ if and only if u 2 L1 ðK; YÞ for each compact interval K  S), we can
u 2 Lloc
identify with the distribution fu 2 D ðS; YÞ defined as follows:
Z
fu ðuÞ ¼ uðuÞ ¼ uðtÞuðtÞdt; ð9Þ
S

for each u 2 DðSÞ; where the integral is regarded in the Bochner sense. Therefore,

1 ðS; YÞ as a subspace of D ðS; YÞ, and regular distributions (the
we interpret Lloc
distributions that admit the representation (9) via the locally Bochner integrable
function) are considered as functions from (S ! Y). We also note that the following
operation f ! f 0 is continuous in D ðS; YÞ [1, p.169].

Laurent-Moïse Schwartz (March 5, 1915–July 4, 2002) was a French

mathematician. He pioneered the theory of distributions, which gives a
well-defined meaning to objects such as the Dirac delta function. He was
awarded the Fields Medal in 1950 for his work on the theory of distributions.
For several years, he taught at the École Polytechnique (Fig. 9).
xxx Introduction: Special Classes of Extended Phase Spaces of Distributions

Fig. 9 Laurent-Moïse
Schwartz

Definition 3 Let Cm ðS; YÞ, m  0; be a family of all functions y : S ! Y such that

each strong derivative yðiÞ of order i ¼ 1; 2; . . .; m is continuous (we note that
yð0Þ ¼ y). If S is a compact interval, then C m ðS; YÞ is a Banach space with the norm

X
m X
m
kykCm ðS;YÞ ¼ sup kyðiÞ ðtÞkY ¼ max kyðiÞ ðtÞkY :
t2S
i¼0 t2S i¼0

Extended Phase Spaces

Let (Vi ; H; Vi ), i ¼ 1; 2; be evolution triple such that assumption (8) holds. Let S be
a finite time interval and X ¼ XðSÞ and X  ¼ X  ðSÞ be the spaces introduced in
Sect. 3. The extended phase space W  ¼ W  ðSÞ, where the real (generalized)
solutions of nonlinear evolution systems belongs, is defined as follows:

W  ðSÞ ¼ fy 2 X  ðSÞ : y0 2 XðSÞg;

where the derivative y0 of y 2 X  is considered in the sense of the distributions

space D ðS; V  Þ.
Introduction: Special Classes of Extended Phase Spaces of Distributions xxxi

By the analogy with Sobolev spaces, it is necessary to establish basic structure

properties, embedding and approximations theorems as well as some “rules of
work” with the elements of such spaces.
Theorem 4 The set W  with the natural operations and graph norm for y0 :
kykW  ¼ kykX  þ ky0 kX 8y 2 W 

is Banach space.
Theorem 5 The set C1 ðS; VÞ \ W0 is dense in W0 .
Theorem 6 W0  CðS; HÞ with continuous embedding. Moreover, for every
y; n 2 W0 and s; t 2 S, the next formula of integration by parts takes place
Z t
ðyðtÞ; nðtÞÞ ðyðsÞ; nðsÞÞ ¼ fðy0 ðsÞ; nðsÞÞ þ ðyðsÞ; n0 ðsÞÞgds: ð10Þ
s

In particular, when y ¼ n, we have:

Z t
1
ðkyðtÞk2H kyðsÞk2H Þ ¼ ðy0 ðsÞ; yðsÞÞds: ð11Þ
2 s

Corollary 1 W   CðS; HÞ with continuous embedding. Moreover, for every

y; n 2 W  and s; t 2 S formula (10) takes place.
Remark 2 When maxfr1 ; r2 g\ þ 1, due to the standard denotations [1, p. 173],
we will denote the space W  as W; “  ” will direct on nonreflexivity of the spaces X
and W .

Jacques-Louis Lions (May 3, 1928–May 17, 2001) was a French mathe-

matician who made contributions to the theory of partial differential equations
and to stochastic control, among other areas. He received the SIAM’s John
von Neumann prize in 1986 and numerous other distinctions. Lions is listed
as an ISI highly cited researcher.
After being part of the French Résistance in 1943 and 1944, J.-L. Lions
entered the École Normale Supérieure in 1947. He was a Professor of math-
ematics at the Université of Nancy, the Faculty of Sciences of Paris, and the
École Polytechnique. He joined the prestigious Collège de France as well as
the French Academy of Sciences in 1973. In 1979, he was appointed director
of the Institut National de la Recherche en Informatique et Automatique
(INRIA), where he taught and promoted the use of numerical simulations using
finite elements integration. Throughout his career, Lions insisted on the use of
mathematics in industry, with a particular involvement in the French space
program, as well as in domains such as energy and the environment. This
xxxii Introduction: Special Classes of Extended Phase Spaces of Distributions

eventually led him to be appointed director of the Centre National d'Etudes

Spatiales (CNES) from 1984 to 1992.
Lions was elected President of the International Mathematical Union in
1991 and also received the Japan Prize and the Harvey Prize that same year.
In 1991, Lions became a foreign member of the National Academy of
Sciences of Ukraine. In 1992, the University of Houston awarded him an
honorary doctoral degree. He was elected President of the French Academy
of Sciences in 1996 and was also a Foreign Member of the Royal Society
(ForMemRS) and numerous other foreign academies.
He has left a considerable body of work, among this more than 400
scientific articles, 20 volumes of mathematics that were translated into
English and Russian, and major contributions to several collective works,
including the 4000 pages of the monumental Mathematical Analysis and
Numerical Methods for Science and Technology (in collaboration with
Robert Dautray), as well as the Handbook of Numerical Analysis in 7 vol-
umes (with Philippe G. Ciarlet).
His son Pierre-Louis Lions is also a well-known mathematician who was
awarded a Fields Medal in 1994. In fact, both Father and Son have also both
received recognition in the form of Honorary Doctorates from Heriot-Watt
University in 1986 and 1995, respectively; Fig. 10.

Fig. 10 Jacques-Louis Lions

Introduction: Special Classes of Extended Phase Spaces of Distributions xxxiii

References

1. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution inclusions and variation
inequalities for earth data processing I. Springer, Heidelberg (2010). Doi:10.1007/978-3-
642-13837-9
2. Zgurovsky, M.Z., Mel'nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation
Inequalities for Earth Data Processing II. Springer, Heidelberg (2010). Doi:10.1007/978-3-
642-13878-2
3. Zgurovsky, M.Z., Mel'nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation
Inequalities for Earth Data Processing II. Springer, Heidelberg (2010). Doi:10.1007/978-3-
642-13878-2
 Part I
Existence and Regularity Results,
Quantitative Methods and Their
Convergence
Chapter 1
Qualitative Methods for Classes
of Nonlinear Systems: Constructive
Existence Results

Abstract In this chapter we establish the existence results for classes of nonlin-
ear systems. Section 2.1 devoted to the first order differential-operator equations
and inclusions. In Sect. 2.2 we consider the second order operator differential equa-
tions and inclusions in special classes of infinite-dimensional spaces of distributions.
Section 2.3 devoted to the existence of strong solutions for evolutional variational
inequalities with nonmonotone potential. The penalty method for strong solutions
is justified. A nonlinear parabolic equations of divergent form are considered as
examples of applications in Sect. 2.4.

1.1 First Order Differential-Operator Equations

and Inclusions

1.1.1 Setting of the Problem

Let (V1 , ·V1 ) and (V2 , ·V2 ) be real reflexive separable Banach spaces continuously
embedded in a Hilbert space (H, (·, ·)). Assume that

the set V : = V1 ∩ V2 is dense in spaces V1 , V2 and H. (1.1)

After the identification H ≡ H ∗ we obtain the following tuples of continuous and

dense embeddings:
V1 ⊂ H ⊂ V1∗ , V2 ⊂ H ⊂ V2∗ , (1.2)

where (Vi∗ ,  · Vi∗ ) is the dual space to Vi , i = 1, 2, with respect to the pairing

·, ·Vi : Vi∗ × Vi → R

which coincides on H × V with the inner product (·, ·) on H .

Let S = [0, T ], 0 < T < +∞, 1 < pi ≤ ri < +∞, i = 1, 2. For i = 1, 2 we
consider the reflexive Banach space

© Springer International Publishing AG 2018 3

M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_1
4 1 Qualitative Methods for Classes of Nonlinear …

X i = L ri (S;H ) ∩ L pi (S;Vi )

with the norm y X i = y L pi (S;Vi ) + y L ri (S;H ) , y ∈ X i ; see section “Special
Classes of Bochner Integrable Functions”. The Banach space X = X 1 ∩ X 2 with
the norm y X = y X 1 + y X 2 is also reflexive (see section “Special Classes of
Bochner Integrable Functions”). We identify the spaces L qi (S;Vi∗ ) + L ri (S;H ) and
X i∗ . Similarly,

X ∗ = X 1∗ + X 2∗ ≡ L q1 (S;V1∗ ) + L q2 (S;V2∗ ) + L r1 (S;H ) + L r2 (S;H ),

where ri−1 + ri−1 = pi−1 + qi−1 = 1. Let us define the duality form on X ∗ × X
  
 f, y = ( f 11 (τ ), y(τ )) H dτ + ( f 12 (τ ), y(τ )) H dτ +  f 21 (τ ), y(τ )V1 dτ +
S S S
 
+  f 22 (τ ), y(τ )V2 dτ = ( f (τ ), y(τ ))dτ,
S S

where f = f 11 + f 12 + f 21 + f 22 , f 1i ∈ L ri (S;H ), f 2i ∈ L qi (S;Vi∗ ).

Assume that there is a separable Hilbert space Vσ such that Vσ ⊂ V1 , Vσ ⊂ V2
with continuous and dense embedding, Vσ ⊂ H with compact and dense embedding.
Then the following tuples of continuous and dense embeddings hold:

Vσ ⊂ V1 ⊂ H ⊂ V1∗ ⊂ Vσ∗ , Vσ ⊂ V2 ⊂ H ⊂ V2∗ ⊂ Vσ∗ .

For i = 1, 2 we set

X i,σ = L ri (S;H ) ∩ L pi (S;Vσ ), X σ = X 1,σ ∩ X 2,σ ,

∗
X i,σ = L ri (S;H ) + L qi (S;Vσ∗ ), X σ∗ = X 1,σ
∗ ∗
+ X 2,σ ,
∗
Wi,σ = {y ∈ X i | y ∈ X i,σ }, Wσ = W1,σ ∩ W2,σ .

For multi-valued (in the general case) map A : X ⇒ X ∗ let us consider the following
problem: 
u + A(u)  f,
(1.3)
u(0) = a, u ∈ W ⊂ C(S;H ),

where a ∈ H and f ∈ X ∗ are arbitrary fixed elements. The main purpose of

this section is to establish sufficient conditions for the existence of a solution of
Problem (1.3) via the Faedo–Galerkin method.
1.1 First Order Differential-Operator Equations and Inclusions 5

1.1.2 Main Assumptions

Let d ∈ X ∗ and E ⊂ S be a measurable set. Further we will use the following

denotations:

1, τ ∈ E;
(dχ E )(τ ) = d(τ )χ E (τ ) for a.e. τ ∈ S; χ E (τ ) =
0, elsewhere.

We recall that the set B belongs to H (X ∗ ) if for each measurable set E ⊂ S and
u, v ∈ B the following inclusion u + (v − u)χ E ∈ B holds.
Lemma 1.1 ([45]) B ∈ H (X ∗ ) if and only if for each n ≥ 1, {d i }i=1 ⊂ B and a
n

family of pairwise disjoint measurable sets {E j } j=1 ⊂ S such that nj=1 E j = S we

n

have that nj=1 d j χ E j ∈ B.

Remark 1.1 We note that ∅, X ∗ ∈ H (X ∗ ); { f } ∈ H (X ∗ ) for each f ∈ X ∗ ; if

K : S ⇒ V ∗ is an arbitrary multi-valued map, then

{ f ∈ X ∗ | f (t) ∈ K (t) for a.e. t ∈ S} ∈ H (X ∗ ).

On the other hand, if v ∈ V ∗ \0̄, then the closed convex set B = { f ∈ X ∗ | f ≡

αv, α ∈ [0, 1]} does not belong to H (X ∗ ), because g(·) = v · χ[0;T /2] (·) ∈
/ B.

Let Y be a reflexive Banach space, Y ∗ be its dual, ·, ·Y : Y ∗ × Y → R be a

pairing, A : Y ⇒ Y ∗ be a strict multi-valued map, that is, A(y) = ∅ for each y ∈ Y.
Define the upper and lower support functions:

[A(y), z]+ := sup d, zY , [A(y), z]− := inf d, zY ;
d∈A(y) d∈A(y)

and the upper and lower norms:

A(y)+ := sup d X ∗ , A(y): = inf d X ∗ ,

d∈A(y) d∈A(y)

y, z ∈ Y . For a nonempty set B ⊂ Y ∗ let coB denotes its convex hull, and coB
denotes the closed convex hull of the set B (see Fig. 1.1), that is,

coB = ∩ B⊂C, C∈C1 (Y ∗ ) C, coB = ∩ B⊂C, C∈C2 (Y ∗ ) C,

where C1 (Y ∗ ) (C2 (Y ∗ )) is the family of all nonempty convex (nonempty closed and
convex respectively) subsets of Y ∗ . Consider the following multi-valued mappings:
coA : Y ⇒ Y ∗ and coA : Y ⇒ Y ∗ such that

(coA)(y) = co(A(y)) and (coA(y)) = co(A(y)),

6 1 Qualitative Methods for Classes of Nonlinear …

Fig. 1.1 Closed convex hull

for each y ∈ Y. Each strict multi-valued maps A, B : Y ⇒ Y ∗ satisfy the following

properties [20, 31, 50, 53]:
(i) [A(y), v1 + v2 ]+ ≤ [A(y), v1 ]+ + [A(y), v2 ]+ , [A(y), v1 + v2 ]− ≥ [A(y), v1 ]−
+ [A(y), v2 ]− ∀y, v1 , v2 ∈ Y ;
(ii) [A(y), v]+ = −[A(y), −v]− , [A(y) + B(y), v]+(−) = [A(y), v]+(−) +
[B(y), v]+(−) ∀y, v ∈ Y ;
(iii) [A(y), v]+(−) = [coA(y), v]+(−) ∀y, v ∈ Y ;
(iv) [A(y), v]+(−) ≤ A(y)+(−) vY , A(y) + B(y)+ ≤ A(y)+ + B(y)+
∀y ∈ Y ;
(v) the inclusion d ∈ coA(y) holds if and only if

[A(y), v]+ ≥ d, vY ∀v ∈ Y ;

(vi) if D ⊂ Y and a(·, ·) : D × Y → R, then for each y ∈ D the function

w → a(y, w) is positively homogeneous convex and lower semi-continuous if
and only if there exists a multi-valued map A : Y ⇒ Y ∗ such that D(A) :=
{y ∈ Y : A(y) = ∅} = D and

a(y, w) = [A(y), w]+ ∀y ∈ D(A), ∀w ∈ Y.

Therefore, the following equalities hold:

[A(y), v]+(−) = [coA(y), v]+(−) and A(y)+(−) = coA(y)+(−)

for each y, v ∈ Y.
Further, the denotation
yn  y in Y

will mean that yn converges weakly to y in a Banach space Y . The family of all
nonempty convex closed (weakly star) and bounded subsets of the dual space Y ∗ (to
Y ) we denote by Cv (Y ∗ ).
1.1 First Order Differential-Operator Equations and Inclusions 7

Let W be a normed space such that W ⊂ Y with the continuous embedding.

Consider the basic classes of multi-valued maps acting from Y into Y ∗ (see also [51]
and references therein).

Definition 1.1 A strict multi-valued map A : Y ⇒ Y ∗ is called:

• pseudomonotone on W , if for each sequence {yn , dn }n≥0 ⊂ W × Y ∗ satisfying
dn ∈ coA(yn ) for each n ≥ 1, yn  y0 in W , dn d0 in Y ∗ as n → +∞, and

lim sup dn , yn − y0 Y ≤ 0, (1.4)

n→∞

there exists a subsequence {yn k , dn k }k≥1 ⊂ {yn , dn }n≥1 such that the following
inequality holds:

lim inf dn k , yn k − wY ≥ [A(y0 ), y0 − w]− (1.5)

k→∞

for each w ∈ Y ;
• bounded, if for every L > 0 there exists l > 0 such that A(y)+ ≤ l for each
y ∈ Y with yY ≤ L;
• coercive, if there exists the real function γ : R+ → R such that γ (s) → +∞ as
s → +∞ and
inf d, yY ≥ γ (yY )yY ∀y ∈ Y ;
d∈A(y)

• demi-closed, if for each sequence {yn , dn }n≥0 ⊂ W × Y ∗ satisfying dn ∈ coA(yn )

for each n ≥ 1, yn  y0 in W , dn d0 in Y ∗ as n → +∞, it follows that
d ∈ coA(y).

Definition 1.2 A multi-valued map A : Y ⇒ Y ∗ satisfies the property Sk on W , if

for each sequence {yn , dn }n≥0 ⊂ W × Y ∗ satisfying dn ∈ coA(yn ) for each n ≥ 1,
yn  y0 in W , dn d0 in Y ∗ as n → +∞, and

lim dn , yn − y0 Y = 0,
n→∞

it follows that d0 ∈ coA(y0 ).

Definition 1.3 A strict multi-valued map A : X ⇒ X ∗ is called the Volterra type

operator (see Fig. 1.2) if for each u, v ∈ X and t ∈ S satisfying the equality u(s) =
v(s) for a.e. s ∈ (0, t), it follows that

[A(u), ξt ]+ = [A(v), ξt ]+

for each ξt ∈ X such that ξt (s) = 0 for a.e. s ∈ S \ [0, t].

8 1 Qualitative Methods for Classes of Nonlinear …

Fig. 1.2 Volterra type operator

1.1.3 Special Basis and Approximations for Multi-valued

Mappings

Let us consider the complete vectors system {h i }i≥1 ⊂ V such that

(α1 ) {h i }i≥1 orthonormal in H ;
(α2 ) {h i }i≥1 orthogonal in V ;
(α3 ) (h i , v)V = λi (h i , v) for each i ≥ 1 and v ∈ V ,
where 0 ≤ λ1 ≤ λ2 , · · · , λ j → ∞ as j → ∞, (·, ·)V is the natural inner product
in V . This system {h i }i≥1 is called a special basis. Let for each m ≥ 1 Hm =
span {h i }i=1
m
, on which we consider the inner product induced from H that we again
denote by (·, ·). Due to the equivalence of H ∗ and H it follows that Hm∗ ≡ Hm ;
X m = L p0 (S;Hm ), X m∗ = L q0 (S;Hm ), p0 = max{r1 , r2 }, q0 > 1: 1/ p0 + 1/q0 = 1,
·, · X m = ·, · X | X m∗ ×X m , Wm := {y ∈ X m | y ∈ X m∗ }, where y is the derivative of an
element y ∈ X m in the sense of distributions from D ∗ (S, Hm ).
Let us consider multi-valued maps that act from X m into X m∗ , m ≥ 1. Let us
remark that embeddings X m ⊂ Ym ⊂ X m∗ are continuous, and the embedding Wm
into X m is compact.
Definition 1.4 The multi-valued map A : X m → Cv (X m∗ ) is called (Wm , X m∗ )-
weakly closed if from that fact that yn  y in Wm , dn  d in X m∗ , dn ∈ A (yn )
∀n ≥ 1 it follows that d ∈ A (y).
Lemma 1.2 The multi-valued map A : X m → Cv (X m∗ ) satisfies the property Sk on
Wm if and only if A : X m → Cv (X m∗ ) is (Wm , X m∗ )-weakly closed.
Proof Let us prove the necessity. Let yn  y in Wm , dn  d in X m∗ , where
dn ∈ A (yn ) ∀n ≥ 1. Then yn → y in X m and dn , yn − y X m → 0 as n → +∞.
Therefore, in virtue of A satisfies the Sk property on Wm , we obtain that d ∈ A (y).
1.1 First Order Differential-Operator Equations and Inclusions 9

Let us prove sufficiency. Let yn  y in Wm , dn  d in X m∗ , dn , yn − y X m ≤ 0

as n → +∞, where dn ∈ A (yn ) ∀n ≥ 1. Then yn → y in X m and d ∈ A (y).
The lemma is proved.
Corollary 1.1 If the multi-valued map A : X m → Cv (X m∗ ) satisfies the property Sk
on Wm , then A is pseudomonotone on Wm .
Let further I : X → X ∗ be the canonical embedding. Let us fix λ ∈ R and set
ϕλ (t) = e−λt , t ∈ S. For an arbitrary y ∈ X ∗ let us define yλ (as a map from S into
V ∗ ) as follows: yλ (t) = ϕλ (t)y(t) for a.e. t ∈ S. Let us remark that (yλ )−λ = y, for
all y ∈ X ∗ . Also we define the element ϕλ y by (ϕλ y)(t) = y(t)ϕλ (t) for a.e. t ∈ S.
Lemma 1.3 The map y → yλ is an isomorphism and an homeomorphism as a map
acting from X m into X m (respectively from X m∗ into X m∗ , from Wm into Wm , from
X into X , from X ∗ into X ∗ , from Ym into Ym , from Y into Y ). Moreover, the map
Wm  y −→ yλ ∈ Wm is weakly-weakly continuous, i.e. from the fact that yn  y
in Wm it follows that yn,λ  yλ in Wm . Also, we have yλ = ϕλ y + ϕλ y ∈ X m∗ ,
∀y ∈ Wm .
Let us consider the multi-valued map A : X → Cv (X ∗ ). Let us define the set
Aλ (yλ ) ∈ Cv (X ∗ ) for fixed y ∈ X by the next relation

[Aλ (yλ ), ω]+ = [A (y) + λy, ωλ ]+ , ∀ω ∈ X.

Let us remark that as the functional ω −→ [A (y) + λy, ωλ ]+ is semiadditive,

positively homogeneous and lower semicontinuous (as the supremum of linear and
continuous functionals), Aλ (yλ ) is defined correctly.
Lemma 1.4 If the map A : X → Cv (X ∗ ) is bounded, then Aλ : X → Cv (X ∗ ) is
bounded.
Lemma 1.5 If A : X m → Cv (X m∗ ) is (Wm , X m∗ )-weakly closed, then Aλ is
(Wm , X m∗ )-weakly closed.
Proof Let yn,λ  yλ in Wm , dn  d in X m∗ , dn ∈ Aλ (yn,λ ). Then, in virtue of
Lemma 1.3 we obtain that yn := (yn,λ )−λ  y := (yλ )−λ in Wm , yn,λ → yλ in X m
and yn → y in X m . Since [A (yn ) + λyn , ωλ ]+ ≥ dn , ω X m , for any ω ∈ X m , then
dn,−λ ∈ A (yn ) + λyn . Therefore, gn := dn,−λ − λyn ∈ A (yn ). Let us remark that
dn,−λ = (dn )−λ  d−λ in X m∗ , and since X m ⊂ X m∗ continuously, we have gn  g
in X m∗ for some g ∈ X m∗ . Due to the fact that A is (Wm , X m∗ )-weakly closed we have
that g ∈ A (y). Therefore, dn,−λ − λyn  g in X m∗ , so that dn,−λ  λy + g in X m∗ ,
and then dn = (dn,−λ )λ  λyλ + gλ in X m∗ . Therefore,

d, ω X m = λyλ + gλ , ω X m = λy + g, ωλ  X m ≤ [A (y) + λy, ωλ ]+ ,

for all ω ∈ X m . Therefore,

d ∈ Aλ (yλ ).

The lemma is proved.

10 1 Qualitative Methods for Classes of Nonlinear …

Since the embedding Wm into X m is compact, then Lemmas 1.2 and 1.5 yield the
following corollary.
Corollary 1.2 If the multi-valued map A : X m → Cv (X m∗ ) satisfies the property Sk
on Wm , then Aλ is pseudomonotone on Wm .

1.1.4 Results

The main solvability results for Problem (1.3) are provided in Theorems 1.1 and 1.2,
Corollaries 1.3, 1.4, and 1.5, and Proposition 1.1 (see also Fig. 1.3 and [4–11, 14–16,
18, 21, 23, 26, 27, 30, 32, 33, 35–47]).
Theorem 1.1 Let a = 0̄, A : X → Cv (X ∗ ) ∩ H (X ∗ ) be coercive bounded map
of the Volterra type that satisfies the property Sk on Wσ . Then for arbitrary f ∈ X ∗
there exists at least one solution of Problem (1.3) that can be obtained via the Faedo–
Galerkin method.

Fig. 1.3 Sufficient conditions of multi-valued mapping for the existence of a weak solution for
differential-operator equation/inclusion via the FG method
1.1 First Order Differential-Operator Equations and Inclusions 11

Proof From coercivity for A : X ⇒ X ∗ it follows that ∀ y ∈ X

inf d, y X ≥ γ (y X )y X .

d∈A(y)

So, ∃ r0 > 0 : γ (r0 ) >  f  X ∗ ≥ 0. Therefore,

∀y ∈ X : y X = r0 [A(y) − f, y]− ≥ 0. (1.6)

T h e s o l v a b i l i t y o f a p p r o x i m a t e p r o b l e m s.
Let us consider the complete vectors system {h i }i≥1 ⊂ V such that
(α1 ) {h i }i≥1 orthonormal in H ;
(α2 ) {h i }i≥1 orthogonal in V ;
(α3 ) ∀ i ≥ 1 (h i , v)V = λi (h i , v) ∀v ∈ V ,
where 0 ≤ λ1 ≤ λ2 , · · · , λ j → ∞ as j → ∞, (·, ·)V is the natural inner product in
V , i.e. {h i }i≥1 is a special basis. Let for each m ≥ 1 Hm = span {h i }i=1
m
, on which
we consider the inner product induced from H that we again denote by (·, ·). Due to
the equivalence of H ∗ and H it follows that Hm∗ ≡ Hm ; X m = L p0 (S;Hm ), X m∗ =
L q0 (S;Hm ), p0 = max{r1 , r2 }, q0 > 1: 1/ p0 + 1/q0 = 1, ·, · X m = = ·, · X | X m∗ ×X m ,
Wm := {y ∈ X m | y ∈ X m∗ }, where y is the derivative of an element y ∈ X m is
considered in the sense of D ∗ (S, Hm ). For any m ≥ 1 let Im ∈ L (X m ; X ) be the
canonical embedding of X m in X , Im∗ be the adjoint operator to Im . Then

∀ m ≥ 1 Im∗ L (X σ∗ ;X σ∗ ) = 1. (1.7)

Let us consider such maps:

Am : = Im∗ ◦ A ◦ Im : X m → Cv (X ∗ ), f m := Im∗ f.

Therefore, (1.6) and Corollary 1.1 yield that

( j1 ) Am is pseudomonotone on Wm ;
( j2 ) Am is bounded;
( j3 ) [Am (y) − f m , y]+ ≥ 0 ∀y ∈ X m : y X = r0 .
Let us consider the operator L m : D(L m ) ⊂ X m → X m∗ with the definition domain

D(L m ) = {y ∈ Wm | y(0) = 0} = Wm0 ,

that acts by the rule:

∀y ∈ Wm0 L m y = y ,

where the derivative y we consider in the sense of the distributions space D ∗ (S;Hm ).
The operator L m satisfies the following properties:
( j4 ) L m is linear;
( j5 ) ∀ y ∈ Wm0 L m y, y ≥ 0;
( j6 ) L m is maximal monotone.
12 1 Qualitative Methods for Classes of Nonlinear …

Therefore, conditions ( j1 )–( j6 ) and [51] guarantees the existence at least one
solution ym ∈ D(L m ) of the problem:

L m (ym ) + Am (ym )  f m , ym  X ≤ r0 ,

that can be obtained by the method of singular perturbations. This means that ym is
the solution of such problem:

ym + Am (ym )  f m
(1.8)
ym (0) = 0̄, ym ∈ Wm , ym  X ≤ R,

where R = r0 .
P a s s i n g t o t h e l i m i t.
From the inclusion from (1.8) it follows that ∀m ≥ 1 ∃dm ∈ A(ym ) :

Im∗ dm = f m − ym ∈ Am (ym ) = Im∗ A(ym ). (1.9)

1◦ . The boundedness of {dm }m≥1 in X ∗ follows from the boundedness of A and

from (1.8). Therefore,

∃c1 > 0 : ∀m ≥ 1 dm  X ∗ ≤ c1 . (1.10)

2◦ . Let us prove the boundedness {ym }m≥1 in X σ∗ . From (1.9) it follows that ∀m ≥ 1
ym = Im∗ ( f − dm ), and, taking into account (1.7), (1.8) and (1.10) we have:

ym  X σ∗ ≤ ym Wσ ≤ c2 < +∞. (1.11)

In virtue of (1.8) and the continuous embedding Wm ⊂ C(S;Hm ) we obtain the

existence of c3 > 0 such that

∀m ≥ 1, ∀t ∈ S ym (t) H ≤ c3 . (1.12)

3◦ . In virtue of estimations from (1.10)–(1.12), due to the Banach–Alaoglu

theorem, taking into account the compact embedding W ⊂ Y , it follows the existence
of subsequences

{ym k }k≥1 ⊂ {ym }m≥1 , {dm k }k≥1 ⊂ {dm }m≥1

and elements y ∈ W , d ∈ X ∗ , for which the following converges hold:

ym k  y in W, dm k  d in X ∗

ym k (t)  y(t) in H for each t ∈ S (1.13)

ym k (t) → y(t) in H for a.e. t ∈ S as k → ∞.

1.1 First Order Differential-Operator Equations and Inclusions 13

Therefore, since ym k (0) = 0̄ for each k ≥ 1, then y(0) = 0̄.

4◦ . Let us prove that
y = f − d. (1.14)

Let ϕ ∈ D(S), n ∈ N and h ∈ Hn . Then ∀k ≥ 1: m k ≥ n we have:

 
ϕ(τ )(ym k (τ ) + dm k (τ ))dτ, h = ym k + dm k , ψ,
S

where ψ(τ ) = h · ϕ(τ ) ∈ X n ⊂ X . Let us remark that here we use the property of
Bochner integral [12, Theorem IV.1.8, p. 153]. Since for m k ≥ n Hm k ⊃ Hn , then
ym k + dm k , ψ =  f m k , ψ. Therefore, ∀k ≥ 1 : m k ≥ n
 
 f m k , ψ = ϕ(τ ) f (τ )dτ, h .
S

Hence, for all k ≥ 1: m k ≥ n

 

ϕ(τ )ym k (τ )dτ, h = f − dm k , ψ →
S

 
→ ϕ(τ )( f (τ ) − d(τ )dτ, h as k → ∞. (1.15)
S

The last follows from the weak convergence dm k to d in X ∗ .

From convergence (1.13) we have:
 
ϕ(τ )ym k (τ )dτ, h → y (ϕ), h as k → ∞, (1.16)
S

where 
∀ ϕ ∈ D(S) y (ϕ) = −y(ϕ ) = − y(τ )ϕ (τ )dτ.
S

Therefore, from (1.15) and (1.16) it follows that

 
∀ϕ ∈ D(S) ∀h ∈ Hm (y (ϕ), h) = ϕ(τ )( f (τ ) − d(τ ))dτ, h .
m≥1 S


Since Hm is dense in V we have that
m≥1


∀ϕ ∈ D(S) y (ϕ) = ϕ(τ )( f (τ ) − d(τ ))dτ.
S
14 1 Qualitative Methods for Classes of Nonlinear …

Therefore, y = f − d ∈ X ∗ .
5◦ . In order to prove that y is the solution of Problem (1.3) it remains to show that
y satisfies the inclusion y + A(y)  f . In virtue if identity (1.14), it is sufficient to
prove that d ∈ A(y).
From (1.13) it follows the existence of {τl }l≥1 ⊂ S such that τl  T as l → +∞
and
∀ l ≥ 1 ym k (τl ) → y(τl ) in H as k → ∞. (1.17)

Let us show that for any l ≥ 1

d, w ≤ [A(y), w]+ ∀w ∈ X : w(t) = 0 for a e. t ∈ [τl , T ]. (1.18)

Let us fix an arbitrary τ ∈ {τl }l≥1 . For i = 1, 2 let us set

X i,σ (τ ) = L ri (τ, T ; H ) ∩ L pi (τ, T ; Vσ ), X σ (τ ) = X 1,σ (τ ) ∩ X 2,σ (τ ),

∗
X i,σ (τ ) = L ri (τ, T ; H ) + L qi (τ, T ; Vσ∗ ), X σ∗ (τ ) = X 1,σ
∗ ∗
(τ ) + X 2,σ (τ ),
∗
Wi,σ (τ ) = {y ∈ X i (τ ) | y ∈ X i,σ (τ )}, Wσ (τ ) = W1,σ (τ ) ∩ W2,σ (τ ),

a0 = y(τ ), ak = ym k (τ ), k ≥ 1.

Similarly we introduce X (τ ), X ∗ (τ ), W (τ ). From (1.17) it follows that

ak → a0 in H as k → +∞. (1.19)

For any k ≥ 1 let z k ∈ W (τ ) be such that


z k + J (z k )  0̄,
(1.20)
z k (τ ) = ak ,

where J : X (τ ) → Cv (X ∗ (τ )) be the duality (in general multi-valued) mapping, i.e.

[J (u), u]+ = [J (u), u]− = u2X (τ ) = J (u)2+ = J (u)2− , u ∈ X (τ ).

We remark that Problem (1.20) has a solution z k ∈ W (τ ) because J is monotone,

coercive, bounded and demiclosed (see [1, 3, 12, 26]). Let us also note that for any
k≥1
z k (T )2H − ak 2H = 2z k , z k  X (τ ) + 2z k 2X (τ ) = 0.

Hence,
1
∀k ≥ 1 z k  X ∗ (τ ) = z k  X (τ ) ≤ √ ak  H ≤ c3 .
2
1.1 First Order Differential-Operator Equations and Inclusions 15

Due to (1.19), similarly to [12, 26], as k → +∞, z k weakly converges in W

to the unique solution z 0 ∈ W of Problem (1.20) with initial time value condition
z(0) = a0 . Moreover,
z k → z 0 in X (τ ) as k → +∞ (1.21)

because lim sup z k 2X (τ ) ≤ z 0 2X (τ ) , z k  z 0 in X (τ ), and X (τ ) is a Hilbert space.

k→+∞
For any k ≥ 1 let us set
 
ym k (t), if t ∈ [0, τ ], dm k (t), if t ∈ [0, τ ],
u k (t) = gk (t) =
z k (t), elsewhere, d̂k (t), elsewhere,

where d̂k ∈ A(u k ) is an arbitrary. As {u k }k≥1 is bounded, A : X ⇒X ∗ is bounded,

then {d̂k }k≥1 is bounded in X ∗ . In virtue of (1.21), (1.13), (1.17)
 τ
lim gk , u k − u = lim (dk (t), yk (t) − y(t)) dt =
k→+∞ k→+∞ 0
 τ  τ
= lim f (t) − yk (t), yk (t) − y(t) dt = lim yk (t), y(t) − yk (t) dt =
k→+∞ 0 k→+∞ 0

 τ
1
= lim yk (0)2H − yk (τ )2H + lim yk (t), y(t) dt =
k→+∞ 2 k→+∞ 0
 τ
1
= y(0)2H − y(τ )2H + y (t), y(t) dt = 0.
2 0

So,
lim gk , u k − u = 0. (1.22)
k→+∞

Let us show that gk ∈ A(u k ) ∀k ≥ 1. For any w ∈ X let us set

 
w(t), if t ∈ [0, τ ], τ 0̄, if t ∈ [0, τ ],
wτ (t) = w (t) =
0̄, elsewhere, w(t), elsewhere.

In virtue of A is the Volterra type operator we obtain that

gk , w = dm k , wτ  + d̂k , wτ  ≤ [A(ym k ), wτ ]+ + d̂k , wτ [A(u k ), wτ ]+ + d̂k , wτ 

≤ [A(u k ), wτ ]+ + [A(u k ), wτ ]+ .

Due to A(u k ) ∈ H (X ∗ ), similarly to [45], we obtain that

[A(u k ), wτ ]+ + [A(u k ), wτ ]+ = [A(u k ), w]+ .

16 1 Qualitative Methods for Classes of Nonlinear …

Since w ∈ X is an arbitrary, then gk ∈ A(u k ) ∀k ≥ 1. Due to {u k }k≥1 is bounded

in X , then {gk }k≥1 is bounded in X ∗ . Thus, up to a subsequence {u k j , gk j } j≥1 ⊂
{u k , gk }k≥1 , for some u ∈ W , g ∈ X ∗ the following convergence takes place

u k j  u in Wσ , gk j  g in X ∗ as j → ∞. (1.23)

We remark that

u(t) = y(t), g(t) = d(t) for a.e. t ∈ [0, τ ]. (1.24)

In virtue of (1.22), (1.23), as A satisfies the property Sk on Wσ , we obtain that

g ∈ A(u). Hence, due to (1.24), as A is the Volterra type operator, for any w ∈ X
such that w(t) = 0 for a.e. t ∈ [τ, T ] we have

d, w = g, w ≤ [A(u), w]+ = [A(y), w]+ .

As τ ∈ {τl }l≥1 is an arbitrary, we obtain (1.18).

From (1.18), due to the functional w → [A(y), w]+ is convex and lower semi-
continuous on X (hence it is continuous on X ) we obtain that for any w ∈ X
d, w ≤ [A(y), w]+ . So, d ∈ A(y).
The theorem is proved.

The following corollary to Theorem 1.1 establishes sufficient conditions for solv-
ability of Problem (1.3) with nonzero initial conditions; see [28].

Corollary 1.3 Let A : X → Cv (X ∗ ) ∩ H (X ∗ ) be bounded map of the Volterra type

that satisfies the property Sk on Wσ . Moreover, let for some c > 0

inf d∈A(y) d, y X − cA(y)+

→ +∞ (1.25)
y X

as y X → +∞. Then for each a ∈ H and f ∈ X ∗ there exists at least one solution
of Problem (1.3) that can be obtained via the Faedo–Galerkin method.
a2H
Proof Let us set ε = 2c2
. We consider w ∈ W such that

w + ε J (w) = 0̄,
w(0) = a,

where J : X → Cv (X ∗ ) be the duality map. Hence w X ≤ c. We define  : X →

Cv (X ∗ ) ∩ H (X ∗ ) by the rule: Â(z) = A(z + w), z ∈ X . Let us set fˆ = f − w ∈ X ∗ .
If z ∈ W is the solution of the problem

z + Â(z)  f,
z(0) = 0̄,
1.1 First Order Differential-Operator Equations and Inclusions 17

then y = z + w is the solution of Problem (1.3). It is clear that  is a bounded map

of the Volterra type that satisfies the property Sk on W . Thus, Theorem 1.1 yields
that it is sufficient to verify the coercivity for the map Â. This property follows from
the following estimates:

[ Â(z), z]− ≥ [A(z + w), z + w]− − [A(z + w), w]+ ≥

≥ [A(z + w), z + w]− − cA(z + w)+ ,

z X ≥ z + w X − c.

The corollary is proved.

Analyzing the proof of Theorem 1.1 the following convergence result holds.

Corollary 1.4 A : X → Cv (X ∗ ) ∩ H (X ∗ ) be bounded map of the Volterra type

that satisfies the property Sk on Wσ , {an }n≥0 ⊂ H : an → a0 in H as n → +∞,
yn ∈ W , n ≥ 1 be the corresponding to initial data an solution of Problem (1.3). If
yn y0 in X , as n → +∞, then y ∈ W is the solution of Problem (1.3) with initial
data a0 . Moreover, up to a subsequence, yn  y0 in Wσ ∩ C(S;H ).

Now let V and H be real Hilbert spaces, V1 = V2 = Vσ := V ; pi = ri = 2,

i = 1, 2. Let us set Y = L 2 (S;H ). Then, according to the identification H ∗ ≡ H,
the spaces Y ∗ and L 2 (S;H ) are identified.
We note that the vector space W = {y ∈ X | y ∈ X ∗ } is a Hilbert space with the
norm yW = y X + y  X ∗ , where y is the derivative of y ∈ X in the sense of
the space of distributions D ∗ (S;V ∗ ) [12]. For any v ∈ X and f ∈ X ∗ consider the
pairing 
 f, v =  f (τ ), v(τ )V dτ,
S

where ·, ·V : V ∗ × V → R is the canonical pairing, which coincides with the inner
product (·, ·) in H on H × V. Hence,  f, v = S ( f (τ ), v(τ )) dτ if f ∈ Y. In the
sequel, to simplify the conclusions, we shall use the last notation even if f ∈ X ∗ .
In the following theorem we justify the Faedo–Galerkin method for solutions of
Problem (1.3) when the multi-valued mapping A is possibly noncoercive (see also
Fig. 1.4).

Theorem 1.2 Let a = 0̄, A : X → Cv (X ∗ ) ∩ H (X ∗ ) be a bounded map of the

Volterra type, which satisfies the property Sk on W . Moreover, let for some λ ≥ 0 the
map A + λI be coercive. Then for arbitrary f ∈ X ∗ there exists at least one solution
of Problem (1.3), which can be obtained via the Faedo–Galerkin method.
18 1 Qualitative Methods for Classes of Nonlinear …

Fig. 1.4 Sufficient conditions for the existence of a solution

Proof We shall provide the proof in several steps.

Step 1: A priory estimate.
At first let us show that there exists a real nondecreasing function γ : R+ → R such
that γ (r ) → +∞ as r → +∞, the function in hands is bounded from below on
bounded sets and the following inequality holds:
 T
inf e−2λτ (d(τ ) + λy(τ ), y(τ ))dτ ≥ γ (y X )y X , (1.26)
d∈A(y) 0

for each y ∈ X. For an arbitrary r > 0 we set

d + λy, y X
γ (r ) =
 inf inf
y∈X, y X =r d∈A(y) y X

and 
γ (0) := 0. The following properties hold:

(a) As A is bounded and the embedding X ⊂ X ∗ is continuous, we have  γ (r ) >

−∞.
(b) From the construction of the function 
γ we have that for all y ∈ X ,
1.1 First Order Differential-Operator Equations and Inclusions 19

[A(y) + λy, y]− ≥ 

γ (y X )y X . (1.27)

In virtue of the boundedness of A it follows that  γ is bounded from below on

bounded sets.
(c) From the coercivity of A + λI it follows that 
γ (r ) → +∞ as r → +∞.
(d) From (a)–(c) we have inf 
γ (r ) =: a > −∞.
r ≥0

For an arbitrary b > a let us consider the nonempty bounded set of R+ given
by Ab = {c ≥ 0 |  γ (c) ≤ b}. Let cb = inf c∈Ab c, b > a. Let us remark that
cb2 ≤ cb1 < +∞, for all b1 > b2 > a, and cb → +∞ as b → +∞. Let us set

a, t ∈ [0, ca+1 ],
γ (t) =

a + k, t ∈ (ca+k , ca+k+1 ], k ≥ 1.

γ : R+ → R is a bounded from below function on bounded sets of R+ , it is

Then, 
a nondecreasing function such that γ (r ) → +∞, as r → ∞, and  γ (t) ≤ 
γ (t), for
any t ≥ 0.
Let us fix an arbitrary y ∈ X. Since A is the operator of the Volterra type, then
 t
inf d(τ ) + λy(τ ), y(τ ) dτ
d∈A(y) 0
 T
= inf d(τ ) + λyt (τ ), yt (τ ) dτ
d∈A(y) 0

≥
γ (yt  X )yt  X = 
γ (y X t )y X t ,

y(τ ), τ ∈ [0, t],
for all t ∈ S, where y X t = yt  X , yt (τ ) = Let for an arbitrary
0̄, else.
d ∈ A(y)
gd (τ ) = d(τ ) + λy(τ ), y(τ ) , for a.e. τ ∈ S,
h(t) = γ (y X t )y X t , t ∈ S.

Let us remark that h(t) ≥ min{

γ (0), 0}y X and
 t
inf gd (τ )dτ ≥ h(t), t ∈ S.
d∈A(y) 0

Let us show that  T

inf e−2λτ (d(τ ) + λy(τ ), y(τ ))dτ ≥
d∈A(y) 0

 T
−2λT
≥e inf (d(τ ) + λy(τ ), y(τ ))dτ + (1.28)
d∈A(y) 0
20 1 Qualitative Methods for Classes of Nonlinear …
 T
+ inf (e−2λτ − e−2λT )(d(τ ) + λy(τ ), y(τ ))dτ.
d∈A(y) 0

Let us set ϕ(τ ) = e−2λ(T −τ ) , τ ∈ [0, T ] (so ϕ ∈ (0, 1]). For any n ≥ 1 we put

n−1
ϕn (τ ) = ϕ inT χ i T , (i+1)T  (τ ), τ ∈ [0, T ]. Then, ϕ inT d1 + 1 − ϕ inT d2 ∈
i=0 n n

A(y), ∀d1 ∈ A(y), ∀d2 ∈ A(y), ∀i = 0, n − 1. Let us remark that |ϕn (τ ) − ϕ(τ )| ≤
2λT
n
, ∀τ ∈ [0, T ]. Lemma 1.1 implies that


n−1     
iT iT
d= (ϕ d1 + 1 − ϕ d2 )χ[ti ,ti+1 ) (τ ) ∈ A(y),
i=0
n n

where ti = iT
n
. Therefore,
 T
inf e−2λτ (d(τ ) + λy(τ ), y(τ ))dτ ≥
d∈A(y) 0

 T
≥ (d(τ ) + λy(τ ), y(τ )e−2λτ )dτ =
0

 T
= ϕn (τ )(d1 (τ ) + λy(τ ), y(τ )e−2λτ )dτ +
0

 T
+ (1 − ϕn (τ ))(d2 (τ ) + λy(τ ), y(τ )e−2λτ )dτ ≥
0

 T
−2λT
≥e (d1 (τ ) + λy(τ ), y(τ ))dτ +
0

 T
+ e−2λτ − e−2λT (d2 (τ ) + λy(τ ), y(τ ))dτ −
0

4λT
− A(y)+ y X + λy2Y .
n

If n → +∞, then taking the infimum with respect to d1 ∈ A(y) and d2 ∈ A(y) in
the last inequality we will obtain (1.28). From (1.28) it follows that
 T
inf e−2λτ (d(τ ) + λy(τ ), y(τ ))dτ ≥
d∈A(y) 0

 T  s
−2λT −2λs
≥e h(T ) + 2λ inf e gd (τ )dτ ds ≥
d∈A(y) 0 0
1.1 First Order Differential-Operator Equations and Inclusions 21

≥ e−2λT h(T )+
 s
+2λT inf inf e−2λs (d(τ ) + λy(τ ), y(τ ))dτ.
d∈A(y) s∈S 0

Let us show that

 s
inf inf e−2λs (d(τ ) + λy(τ ), y(τ ))dτ ≥ −c1 y X ,
d∈A(y) s∈S 0

where c1 = max{− γ (0), 0} ≥ 0 does not depend on y ∈ X. Let y ∈ X is fixed. For

s ∈ S, d ∈ A(y) let us set
 s
−2λs
ϕ(s, d) = e d(τ ) + λy(τ ), y(τ ) dτ,
0
a = inf inf ϕ(s, d), Sd = {s ∈ S | ϕ(s, d) ≤ a}.
d∈A(y) s∈S

From the continuity of ϕ(·, d) on S it follows that Sd is a nonempty closed set for an
arbitrary d ∈ A(y). Indeed, for any fixed d ∈ A(y) there exists sd ∈ S such that

ϕ(sd , d) = min ϕ(ŝ, d) ≤ a.

ŝ∈S

From the continuity of ϕ(·, d) on S it follows that Sd is closed.

Let us prove now that the system {Sd }d∈A(y) is centered. For fixed {di }i=1
n
⊂ A(y),
n ≥ 1, let us set

ψi (·) = di (·) + λy(·), y(·) ,

ψ(·) = max ψi (·),
i∈{1,...,n}
E 0 = ∅,
   
j−1
E j = τ ∈ S \ ∪i=0 E i | ψ j (τ ) = ψ(τ ) ,

for j = 1, n, and

n
d(·) = d j (·)χ E j (·).
j=1

Let us remark that E j is measurable for any j = 1, n, ∪nj=1 E j = S, E i ∩ E j = ∅,

∀i = j, i, j = 1, n. Also, d ∈ X ∗ . Moreover,
 s  s
ϕ(s, di ) = e−2λs ψi (τ )dτ ≤ e−2λs ψ(τ )dτ =
0 0

= ϕ(s, d), s ∈ S, i = 1, n.
22 1 Qualitative Methods for Classes of Nonlinear …

Therefore, in virtue of Lemma 1.1 we have d ∈ A(y) and for some sd ∈ S,

ϕ(sd , di ) ≤ ϕ(sd , d) = min ϕ(ŝ, d) ≤ a, i = 1, n.

ŝ∈S

So, sd ∈ ∩i=1
n
Sdi = ∅.
Since S is compact, and the system of closed sets {Sd }d∈A(y) is centered, we obtain
the existence of s0 ∈ S such that s0 ∈ ∩d∈A(y) Sd . This implies that
 s
−2λs
inf inf e d(τ ) + λy(τ ), y(τ ) dτ
d∈A(y) s∈S
0 s0
≥ inf e−2λs0 d(τ ) + λy(τ ), y(τ ) dτ
d∈A(y) 0
 s0
= e−2λs0 inf gd (τ )dτ ≥ e−2λs0 h(s0 )
d∈A(y) 0

≥ e−2λs0 min{
γ (0), 0}y X
≥ − max{− γ (0), 0}y X = −c1 y X .

Therefore, for all y ∈ X ,

 T
inf e−2λτ (d(τ ) + λy(τ ), y(τ ))dτ ≥
d∈A(y) 0

≥ e−2λT 
γ (y X ) − 2λc1 T y X .

If we set γ (r ) = e−2λT γ (r ) − 2λc1 T, then we will obtain (1.26).

From (1.26), the properties of the real function γ and the conditions of the theorem
it follows the existence of r0 > 0 such that γ (r0 ) >  f λ  X ∗ ≥ 0 and also that for
any y ∈ X ,
[Aλ (yλ ), yλ ]− ≥ γ (y X )y X ≥ γ (yλ  X )yλ  X .

Therefore, for all y ∈ X satisfying yλ  X = r0 we have

[Aλ (yλ ) − f λ , yλ ]− ≥ (γ (r0 ) −  f λ  X ∗ )r0 ≥ 0,

that is,
[Aλ (yλ ) − f λ , yλ ]− ≥ 0. (1.29)

Step 2: Finite-dimensional approximations.

We shall consider now a sequence of finite-dimensional approximative problems via
the Faedo–Galerkin method.
For any m ≥ 1 let Im ∈ L (X m ; X ) be the canonical embedding of X m into X ,
and Im∗ be the adjoint operator to Im . Then
1.1 First Order Differential-Operator Equations and Inclusions 23

Im∗ L (X ∗ ; X ∗ ) = 1, ∀m ≥ 1. (1.30)

Let us consider the following maps [25]:

Am := Im∗ ◦ A ◦ Im : X m → Cv (X ∗ ),
Aλ,m := Im∗ ◦ Aλ ◦ Im : X m → Cv (X ∗ ),
Am,λ := (Am )λ : X m → Cv (X ∗ ),
f m := Im∗ f, f λ,m := Im∗ f λ , f m,λ := ( f m )λ .

Let us remark that

Aλ,m = Am,λ , f λ,m = f m,λ . (1.31)

Indeed, in virtue of Lemma 1.3 for any y, w ∈ X m ,

[Aλ,m (yλ ), w]− = [(Im∗ ◦ Aλ )(yλ ), w]− = [Aλ (yλ ), w]−

= [A(y) + λy, wλ ]− = [Im∗ ◦ (A + λI )(y), wλ ]− =

= [(Am )λ (yλ ), w]− = [Am,λ (yλ ), w]− .

So, from (1.29), (1.31), Lemma 1.4, Corollary 1.2, and the conditions of the theorem,
applying similar arguments as in [25, pp. 115–117], [19, pp. 197–198], we obtain
the following properties:
(j1) Aλ,m is pseudomonotone on Wm ;
(j2) Aλ,m is bounded;
(j3) [Aλ,m (yλ ) − f λ,m , yλ ]− ≥ 0 for all yλ ∈ X m such that yλ  X = r0 .
We note that (j3) is a consequence of (1.29) and the definition of Aλ,m , f λ,m ,
whereas (j2) follows from Lemma 1.4 and the boundednesss of Im , Im∗ . Finally, (j1)
is obtained in the following way: since A satisfies the property Sk in W , for Am the
same property holds on Wm ; hence, by Corollary 1.2 the operator Am,λ = (Am )λ is
pseudomonotone in Wm , and then (1.31) implies (j1).
Let us consider the operator L m : D(L m ) ⊂ X m → X m∗ with domain

D(L m ) = {y ∈ Wm | y(0) = 0} = Wm0 ,

which is defined by the rule: L m y = y , ∀y ∈ Wm0 , where the derivative y we

consider in the sense of the space of distributions D ∗ (S;Hm ). From [25, Lemma 5,
p. 117] for the operator L m the next properties are true:
(j4) L m is linear;
(j5) L m y, y ≥ 0, ∀y ∈ Wm0 ;
(j6) L m is maximal monotone.
Therefore, conditions ( j1 )–( j6 ) and Theorem 3.1 from [26] guaranty the existence
of at least one solution z m ∈ D(L m ) of the problem:
24 1 Qualitative Methods for Classes of Nonlinear …

L m (z m ) + Aλ,m (z m )  f λ,m , z m  X ≤ r0 ,

which can be obtained by the method of singular perturbations. This means (see
(1.31)) that ym := (z m )−λ ∈ Wm is the solution of the problem

ym + Am (ym )  f m ,
(1.32)
ym (0) = 0̄, ym ∈ Wm , ym  X ≤ R,

where R = r0 eλT .
Step 3: Passing to the limit.
From (1.32) it follows that for any m ≥ 1 there exists dm ∈ A(ym ) such that

Im∗ dm = f m − ym ∈ Am (ym ) = Im∗ A(ym ). (1.33)

Let us prove now that (up to a subsequence) the sequence of solutions of (1.32)
converges to a solution of (1.3). Again, we divide this proof in some substeps.
Step 3a.
The boundedness of A and (1.32) imply that {dm }m≥1 is bounded in X ∗ . Therefore,
there exists c1 > 0 such that

dm  X ∗ ≤ c1 ∀m ≥ 1. (1.34)

Step 3b.
Let us prove the boundedness of {ym }m≥1 in X ∗ . From (1.33) it follows that ym =
Im∗ ( f − dm ), ∀m ≥ 1, and taking into account (1.30), (1.32) and (1.34) we have

ym  X ∗ ≤ ym W ≤ R +  f  X ∗ + c1 =: c2 . (1.35)

In virtue of the continuous embedding W ⊂ C(S;H ) we obtain the existence of

c3 > 0 such that
ym (t) H ≤ c3 ∀m ≥ 1, ∀t ∈ S. (1.36)

Step 3c.
In virtue of estimates (1.34)–(1.36), due to the Banach–Alaoglu theorem, and taking
into account the continuous embedding W ⊂ C(S;H ) and the compact embedding
W ⊂ Y , it follows the existence of subsequences

{ym k }k≥1 ⊂ {ym }m≥1 , {dm k }k≥1 ⊂ {dm }m≥1

and elements y ∈ W , d ∈ X ∗ , for which the next convergences take place:

ym k  y in W, dm k  d in X ∗ ,
ym k (t)  y(t) in H for each t ∈ S, (1.37)
ym k (t) → y(t) in H for a.e. t ∈ S as k → ∞.
1.1 First Order Differential-Operator Equations and Inclusions 25

From here, as ym k (0) = 0̄, ∀k ≥ 1, we have y(0) = 0̄.

Step 3d.
Let us prove that
y = f − d. (1.38)

Let ϕ ∈ D(S), n ∈ N, and h ∈ Hn . Then for all k ≥ 1 such that m k ≥ n we have

 
ϕ(τ )(ym k (τ ) + dm k (τ ))dτ, h = ym k + dm k , ψ,
S

where ψ(τ ) = h · ϕ(τ ) ∈ X n ⊂ X . Let us remark that here we use the properties of
Bochner’s integral (see [12], Theorem IV.1.8). Since Hm k ⊃ Hn , for m k ≥ n we get
ym k + dm k , ψ =  f m k , ψ. Therefore, for all k ≥ 1 such that m k ≥ n
 
 f m k , ψ = ϕ(τ ) f (τ )dτ, h .
S

Hence, for all k ≥ 1 such that m k ≥ n

 
ϕ(τ )ym k (τ )dτ, h =  f − dm k , ψ →
S
 
→ ϕ(τ )( f (τ ) − d(τ ))dτ, h as k → ∞. (1.39)
S

The last convergence follows from the weak convergence dm k to d in X ∗ . From (1.37)
we have  
ϕ(τ )ym k (τ )dτ, h → y (ϕ), h as k → +∞, (1.40)
S

where 
y (ϕ) = −y(ϕ ) = − y(τ )ϕ (τ )dτ, ∀ϕ ∈ D(S).
S

Therefore, from (1.39) and (1.40) it follows that for all ϕ ∈ D(S), h ∈ m≥1 Hm ,
 
y (ϕ), h = ϕ(τ )( f (τ ) − d(τ ))dτ, h .
S

Since Hm is dense in V we have that
m≥1


y (ϕ) = ϕ(τ )( f (τ ) − d(τ ))dτ , ∀ϕ ∈ D(S).
S

Therefore, y = f − d ∈ X ∗ .
26 1 Qualitative Methods for Classes of Nonlinear …

Step 3e.
To prove that y is a solution of Problem (1.3) it remains to verify that y satisfies
the inclusion y + A(y)  f . Thus, according to (1.38), it is sufficient to prove that
d ∈ A(y).
From (1.37) it follows the existence of {τl }l≥1 ⊂ S such that τl  T as l → +∞,
and
ym k (τl ) → y(τl ) in H ∀l ≥ 1 as k → +∞. (1.41)

Let us show that

d, w ≤ [A(y), w]+ (1.42)

for each l ≥ 1 and w ∈ X such that w(t) = 0 for a.e. t ∈ [τl , T ].

Let us fix an arbitrary τ ∈ {τl }l≥1 . Let us set

X (τ ) = L 2 (τ, T ; V ), X ∗ (τ ) = L 2 (τ, T ; V ∗ ),
 T
u, v X (τ ) = u(s), v(s)V ds
τ

for u ∈ X (τ ), v ∈ X ∗ (τ ), and

W (τ ) = {u ∈ X (τ ) | u ∈ X ∗ (τ )},

a0 = y(τ ), ak = ym k (τ ), k ≥ 1.

From (1.41) it follows that

ak → a0 in H as k → +∞. (1.43)

For any k ≥ 1 let z k ∈ W (τ ) be such that


z k + J (z k ) = 0̄,
(1.44)
z k (τ ) = ak ,

where J : X (τ ) → X ∗ (τ ) is the duality mapping (which is single-valued, as X (τ )

is a Hilbert space), i.e.

J (u), u X (τ ) = u2X (τ ) = J (u)2X ∗ (τ ) , u ∈ X (τ ).

We remark that Problem (1.44) has a solution z k ∈ W (τ ) because J : X (τ ) → X ∗ (τ )

is monotone, coercive, bounded and demicontinuous (see [1, 3, 12, 26]). Let us also
note that for any k ≥ 1,

z k (T )2H − ak 2H = 2z k , z k  X (τ ) = −2z k 2X (τ ) .

1.1 First Order Differential-Operator Equations and Inclusions 27

Hence,
1
z k  X ∗ (τ ) = z k  X (τ ) ≤ √ ak  H ≤ c3 , ∀k ≥ 1.
2

Due to (1.43), similarly to [12, 26], z k converges weakly in W as k → +∞ to

the unique solution z 0 ∈ W of Problem (1.44) with initial condition z(τ ) = a0 .
Moreover,
z k → z 0 in X (τ ) as k → +∞, (1.45)

because lim sup z k 2X (τ ) ≤ z 0 2X (τ ) , z k  z 0 in X (τ ), and X (τ ) is a Hilbert space.

k→+∞
For any k ≥ 1 let us set

ym k (t), if t ∈ [0, τ ],
u k (t) =
z k (t), elsewhere,

dm k (t), if t ∈ [0, τ ],
gk (t) =
d̂k (t), elsewhere,

where d̂k ∈ A(u k ) be an arbitrary. As {u k }k≥1 is bounded and A : X ⇒ X ∗ is

bounded, we obtain that {d̂k }k≥1 is bounded in X ∗ . In virtue of (1.45), (1.37), (1.41)
we have
lim gk , u k − u =
k→+∞

 τ
= lim dm k (t), ym k (t) − y(t) dt =
k→+∞ 0

 τ
= lim f (t) − ym k (t), ym k (t) − y(t) dt =
k→+∞ 0

 τ
= lim ym k (t), y(t) − ym k (t) dt =
k→+∞ 0

1
= lim ym k (0)2H − ym k (τ )2H +
k→+∞ 2
 τ
+ lim ym k (t), y(t) dt =
k→+∞ 0

 τ
1
= y(0)2H − y(τ )2H + y (t), y(t) dt = 0.
2 0

So,
lim gk , u k − u = 0. (1.46)
k→+∞
28 1 Qualitative Methods for Classes of Nonlinear …

Let us show that gk ∈ A(u k ), ∀k ≥ 1. For any w ∈ X let us set


w(t), if t ∈ [0, τ ],
wτ (t) =
0̄, elsewhere,

τ 0̄, if t ∈ [0, τ ],
w (t) =
w(t), elsewhere.

Since A is an operator of the Volterra type we obtain that

gk , w = dm k , wτ  + d̂k , wτ  ≤

≤ [A(ym k ), wτ ]+ + d̂k , wτ  =

= [A(u k ), wτ ]+ + d̂k , wτ  ≤

≤ [A(u k ), wτ ]+ + [A(u k ), wτ ]+ .

Since A(u k ) ∈ H (X ∗ ), similarly to the proof of (1.28) we obtain that

[A(u k ), wτ ]+ + [A(u k ), wτ ]+ = [A(u k ), w]+ .

Since w ∈ X is an arbitrary, then gk ∈ A(u k ) for all k ≥ 1. Since {u k }k≥1 is bounded in

X , {gk }k≥1 is bounded in X ∗ . Thus, up to a subsequence {u k j , gk j } j≥1 ⊂ {u k , gk }k≥1 ,
for some u ∈ W , g ∈ X ∗ the next convergence holds

u k j  u in W, gk j  g in X ∗ as j → ∞. (1.47)

We remark that
u(t) = y(t), g(t) = d(t) for a.e. t ∈ [0, τ ]. (1.48)

In virtue of (1.46), (1.47), as A satisfies the property Sk on W , we obtain that g ∈ A(u).

Hence, due to (1.48), as A is the Volterra type operator, for any w ∈ X such that
w(t) = 0 for a.e. t ∈ [τ, T ] we get

d, w = g, w ≤ [A(u), w]+ = [A(y), w]+ .

As τ ∈ {τl }l≥1 is an arbitrary, we obtain (1.42).

From (1.42), as the functional w → [A(y), w]+ is convex and lower semicontin-
uous on X (hence it is continuous on X ), we obtain that d, w ≤ [A(y), w]+ for
each w ∈ X . Therefore, d ∈ A(y).
The theorem is proved.

Analyzing the proofs of Theorem 1.2 and Corollary 1.3 (see [28]) the following
proposition holds.
1.1 First Order Differential-Operator Equations and Inclusions 29

Proposition 1.1 Let A : X → Cv (X ∗ ) ∩ H (X ∗ ) be bounded map of the Volterra

type which satisfies the property Sk on W . Moreover, let for some λ A ≥ 0 and c > 0

[A(y), y]− − cA(y)+ + λ A y2Y

→ +∞ (1.49)
y X

as y X → +∞. Then for any a ∈ H , f ∈ X ∗ there exists at least one solution of
Problem (1.3), which can be obtained via the Faedo–Galerkin method.
a2H
Proof Let us set ε = 2c2
. We consider w ∈ W such that

w + ε J (w) = 0̄,
w(0) = a,

where J : X → X ∗ is the duality map. Hence w X ≤ c. We define  : X →

Cv (X ∗ ) ∩ H (X ∗ ) by the rule: Â(z) = A(z + w), z ∈ X . Let us set fˆ = f − w ∗ . If
z ∈ W is a solution of the problem

z + Â(z)  
f,
z(0) = 0̄,

then y = z + w is a solution of Problem (1.3). It is clear that  is a bounded map of

the Volterra type which satisfies the property Sk on W . So, due to Theorem 1.2 it is
enough to prove the coercivity for the map  + λ A I . This property follows from the
estimates:
[ Â(z), z]+ ≥ [A(z + w), z + w]+ − [A(z + w), w]+ ≥

≥ [A(z + w), z + w]+ − cA(z + w)+ ,

z2Y ≥ z + w2Y − c2 − 2w X ∗ z X .

z X ≥ z + w X − c.

The proposition is proved.

Analyzing the proof of Theorem 1.2 we can obtain the following convergence
result.

Corollary 1.5 Let A : X → Cv (X ∗ ) ∩ H (X ∗ ) be a bounded map of the Volterra

type which satisfies the property Sk on W . We consider a sequence {an }n≥0 ⊂ H such
that an → a0 in H as n → +∞. Let yn ∈ W , n ≥ 1, be solutions of Problem (1.3)
corresponding to the initial data an . If yn  y0 in X as n → +∞, then y0 ∈ W
is solution of Problem (1.3) with initial data a0 . Moreover, up to a subsequence,
yn  y0 in W ⊂ C(S;H ).
30 1 Qualitative Methods for Classes of Nonlinear …

1.2 Second Order Operator Differential Equations

and Inclusions

In this section we provide the existence results for the second order operator differ-
ential equations and inclusions. Since the problem in hands naturally considers as
the first order integro-differential-operator inclusion, we recall some of them.
Let the following conditions hold:
(H1 ) V , Z , H are Hilbert spaces; H ∗ ≡ H and we have such chain of dense and
compact embeddings:

V ⊂ Z ⊂ H ≡ H ∗ ⊂ Z ∗ ⊂ V ∗;

(H2 ) f 0 ∈ V ∗ ;
(A1 ) ∃c > 0 : ∀u ∈ V, ∀d ∈ A0 (u) dV ∗ ≤ c(1 + uV );
(A2 ) ∃α, β > 0 : ∀u ∈ V , ∀d ∈ A0 (u) d, uV ≥ αu2V − β;
(A3 ) A0 = A1 + A2 , where A1 : V → V ∗ is linear, selfconjugated, positive operator,
A2 : V ⇒ V ∗ satisfies such conditions:
(a) there exists a Hilbert space Z such that the embedding V ⊂ Z is dense and
compact, and the embedding Z ⊂ H is dense and continuous;
(b) for any u ∈ Z the set A2 (u) is nonempty, convex, and weakly compact in Z ∗ ;
(c) A2 : Z ⇒ Z ∗ is a bounded map, that is, A2 converts bounded sets from Z into
bounded sets in the space Z ∗ ;
(d) A2 : Z ⇒ Z ∗ is a demiclosed map, i.e. if u n → u in Z , dn → d weakly in
Z ∗ , n → +∞, and dn ∈ A2 (u n ) ∀n ≥ 1, then d ∈ A2 (u);
(B1 ) B0 : V → V ∗ is a linear selfconjugated operator;
(B2 ) ∃γ > 0 : B0 u, uV ≥ γ u2V .
Here ·, ·V : V ∗ × V → R is the duality in V ∗ × V, coinciding on H × V with the
inner product (·, ·) in Hilbert space H.
Note that from (A1 )–(A3 ), [34, 51] it follows that the map A0 satisfies such
condition:
(A3 ) A0 : V ⇒ V ∗ is (generalized) pseudomonotone operator, that is,
(a) for any u ∈ V the set A0 (u) is nonempty, convex, and weakly compact one in
V ∗;
(b) if u n → u weakly in V, n → +∞, dn ∈ A0 (u n ) ∀n ≥ 1, and lim supdn , u n −
n→∞
uV ≤ 0, then ∀ω ∈ V ∃d(ω) ∈ A0 (u) :

lim inf dn , u n − ωV ≥ d(ω), u − ωV ;

n→+∞

(c) the map A0 is upper semicontinuous one that acts from an arbitrary finite-
dimensional subspace of V into V ∗ endowed with weak topology.
Thus, we investigate the dynamic of all weak solutions of the second order non-
linear autonomous differential-operator inclusion
1.2 Second Order Operator Differential Equations and Inclusions 31

y (t) + A0 (y (t)) + B0 (y(t))  f 0 for a.e. t > 0 (1.50)

as t → +∞, which are defined as t ≥ 0, where parameters of the problem satisfy

conditions (H1 ), (H2 ), (A1 )–(A3 ), (B1 )–(B2 ).
As a weak solution of evolution inclusion (1.50) on the interval [τ, T ] we consider
a pair of elements (u(·), u (·))T ∈ L 2 (τ, T ; V × V ) such that for some d(·) ∈
L 2 (τ, T ; V ∗ )

d(t) ∈ A0 (u (t)) for a.e. t ∈ (τ, T ),

T T
− τ (ζ (t), u (t))dt + τ d(t), ζ (t)V dt+ (1.51)
+ τ B0 u(t), ζ (t)V dt = τ  f 0 , ζ (t)V ∀ζ ∈ C0∞ ([τ, T ]; V ),
T T

where u is the derivative of the element u(·) in the sense of the space of distributions
D ∗ ([τ, T ]; V ∗ ).
Note that the abstract theorems on the existence of solutions for such problems
under weaker conditions were considered in [34, 51]. Here we consider Problem 2
from [34], for which we can (as follows from results of the further chapters) have
not only the abstract result on existence of weak solution but we can investigate
the behavior of all weak solutions as t → +∞ in the phase space V × H and
study the structure of the global and trajectory attractors. Underline that results
concerning multi-valued dynamic of displacements and velocities can be applied to
hemivariational inequalities.
√ Further, without loss the generality we consider the equivalent norm uV =
B0 u, uV , u ∈ V, in the space V . The given norm is generated by the inner
product (u, v)V = B0 u, vV , u, v ∈ V. For fixed τ < T let us consider
∗
X τ,T = L 2 (τ, T ; V ), X τ,T = L 2 (τ, T ; V ∗ ), Wτ,T = {u ∈ X τ,T |u ∈ X τ,T
∗
},
∗ ∗
Aτ,T : X τ,T ⇒ X τ,T , Aτ,T (y) = {d ∈ X τ,T |d(t) ∈ A0 (y(t)) for a.e. t ∈ (τ, T )},
∗
Bτ,T : X τ,T → X τ,T , Bτ,T (y)(t) = B0 (y(t)) for a.e. t ∈ (τ, T ),
∗
f τ,T ∈ X τ,T , f τ,T (t) = f 0 for a.e. t ∈ (τ, T ).

Note that the space Wτ,T is the Hilbert space with the graph norm of the derivative
(cf. [50, 51]):
u2Wτ,T = u2X τ,T + u 2X τ,T
∗ , u ∈ Wτ,T . (1.52)

∗
From [34, Lemma 7, p. 516], (A1 ), (A2 ), (A3 ) it follows that Aτ,T : X τ,T ⇒ X τ,T
satisfies the next conditions:
(N1 ) ∃C1 > 0: d X τ,T
∗ ≤ C1 (1 + y X τ,T ) ∀y ∈ X τ,T , ∀d ∈ Aτ,T (y);
(N2 ) ∃C2 , C3 > 0: d, y X τ,T ≥ C2 y2X τ,T − C3 ∀y ∈ X τ,T , ∀d ∈ Aτ,T (y);
∗
(N3 ) Aτ,T : X τ,T ⇒ X τ,T is (generalized) pseudomonotone on Wτ,T , that is,
(a) for any y ∈ X τ,T the set Aτ,T (y) is a nonempty, convex, and weakly compact
32 1 Qualitative Methods for Classes of Nonlinear …

∗
one in X τ,T ;
(b) Aτ,T is the upper semicontinuous map as the map that acts from an arbitrary finite
∗
dimensional subspace from X τ,T into X τ,T endowed by the weak topology;
∗
(c) if yn → y weakly in Wτ,T , dn ∈ Aτ,T (yn ) ∀n ≥ 1, dn → d weakly in X τ,T , and

lim supdn , yn − y X τ,T ≤ 0,

n→+∞

then d ∈ Aτ,T (y) and lim dn , yn  X τ,T = d, y X τ,T .

n→+∞
∗ ∗
Here ·, · X τ,T : X τ,T × X τ,T → R is the pairing in X τ,T × X τ,T coinciding on
L 2 (τ, T ; H ) × X τ,T with the inner product in L 2 (τ, T ; H ), that is,
 T
∀u ∈ L 2 (τ, T ; H ), ∀v ∈ X τ,T u, v X τ,T = (u(t), v(t))dt.
τ

Note also (cf. [12, Theorem IV.1.17, P. 177]) that the embedding Wτ,T ⊂ C([τ,
T ]; H ) is continuous and dense. Moreover,
 T 
(u(T ), v(T )) − (u(τ ), v(τ )) = u (t), v(t)V + v (t), u(t)V dt, (1.53)
τ

for each u, v ∈ Wτ,T .

The definition of derivative in the sense of D([τ, T ]; V ∗ ) and equality (1.51) yield
the following statement.
Lemma 1.6 Each weak solution (y(·), y (·))T of Problem (1.50) on the interval
[τ, T ] belongs to the space C([τ, T ]; V ) × Wτ,T . Moreover,

y + Aτ,T (y ) + Bτ,T (y)  f τ,T . (1.54)

Vice versa, if y(·) ∈ C([τ, T ]; V ), y (·) ∈ Wτ,T , and y(·) satisfies (1.54), then
(y(·), y (·))T is a weak solution of (1.50) on [τ, T ].
A weak solution of Problem (1.50) with initial data

y(τ ) = a, y (τ ) = b (1.55)

on the interval [τ, T ] exists for any a ∈ V, b ∈ H. It follows from [34, Theorem 11,
p. 523]. Thus, the following lemma holds.
Lemma 1.7 For any τ < T, a ∈ V, b ∈ H Cauchy Problem (1.50), (1.55) has
a weak solution (y, y )T ∈ X τ,T × X τ,T . Moreover, each weak solution (y, y )T
of Cauchy Problem (1.50), (1.55) on the interval [τ, T ] belongs to the space
C([τ, T ]; V ) × Wτ,T and y satisfies (1.54).
The similar statement holds also for non-autonomous problems. For this purpose
additional measurability assumption for A2 is claimed.
1.3 Evolutional Variational Inequalities: Penalty Method and Strong Solutions 33

1.3 Evolutional Variational Inequalities: Penalty Method

and Strong Solutions

In investigating unilateral problems, problems on Riemannian manifolds with or

without boundary, semi-penetration problems, and in the analysis and control of
processes and fields of different natures on the boundary of a domain, a demand
arises for the consideration of evolutionary variational inequalities with nonlinear
nonmonotone operators in infinite-dimensional spaces. To describe the state func-
tions of such objects, the concepts of strong and weak solutions are naturally intro-
duced. A strong solution does not always adequately describe a system state since,
in the majority of cases, such classes of solutions prevents the effects of breaks or
unilateral semi-penetration, that is, are too regular to adequately describe states of
the processes and fields being investigated. The proof of the existence of strong
solutions (especially for equations with nonmonotone reaction laws) is problematic.
The concept of a weak solution is too general (this solution not always adequately
describes a state function, that is, this class of solutions can formally include not
only physical solutions) and, at the same time, is insufficiently regular to adequately
implement the numerical analysis of the problems being investigated. Note that a
strong solution of an evolutionary variational inequality is, as a rule, a weak solu-
tion. A demand arises for the introduction of a new intermediate class of physical
solutions to such problems that, on the one hand, must satisfy natural energy equal-
ities and, on the other hand, provide the possibility of substantiation of constructive
(and at the same time physical) methods of their existence (for example, the artificial
viscosity method for problems of classical hydroaeromechanics in an incompressible
continuous medium).
This section introduces the concept of a physical solution on a finite time inter-
val for classes of autonomous evolutionary variational inequalities with nonlinear
nonmonotone (in general cases) mappings defined on convex cones. This concept is
based on natural energy equalities and continuous dependence of state functions in
the phase space on the time variable. For approximate searching for physical solu-
tions, the classical penalty method is used. For the solutions obtained, the possibility
of a global description of the behavior of such systems is substantiated on the basis
of the results of [13, 17] in their natural phase space with respect to the topology of
strong convergence by finite algorithms up to an arbitrary small parameter.
For an evolutionary triple (V ; H ; V ∗ ) a nonlinear (in the general case) mapping
A : V → V ∗ , and a convex closed cone K ⊆ V , the problem of investigating
the dynamics of the following autonomous evolutionary variational inequality is
considered in the phase space H of all physical solutions y : R+ → V , y(t) ∈ K
for a.a. t > 0:

y (t) + A(y(t)), v − y(t)V ≥ 0 for all v ∈ K and for a.a. t > 0 (1.56)

in which the parameters of the problem satisfy the following conditions.

34 1 Qualitative Methods for Classes of Nonlinear …

Assumption 1 p ≥ 2 and q > 1: 1

p
+ 1
q
= 1 and the embedding of V into H is
compact.
p−1
Assumption 2 ∃c > 0: A(u)V ∗ ≤ c(1 + uV ) ∀u ∈ V .
p
Assumption 3 ∃α, β > 0: A(u), uV ≥ αuV − β ∀u ∈ V .

Assumption 4 A : V → V ∗ is a pseudomonotone operator satisfying the (S)-

property, that is, since u n → u weakly in V , and lim A(u n ), u n − uV ≤ 0, we
n→+∞
obtain that u n → u in V and lim A(u n ), u n − ωV ≥ A(u), u − ωV ∀ω ∈ V .
n→+∞

Assumption 5 K ⊆ V is a convex closed cone such that intVσ K σ = 0, where Vσ ⊆

V is a real reflexive separable Banach space continuously and densely embedded
into V , K σ := K ∩ Vσ .

Here, ·, ·V : V ∗ × V → R is a pairing in V ∗ × V , and this pairing coinsides on

H × V with the scalar product (·, ·) in a Hilbert space H . Note that a space Vσ∗ is the
conjugate of Vσ with respect to the canonical pairing ·, ·Vσ : Vσ∗ × Vσ → R that
coincides on H × Vσ with the scalar product (·, ·) in H . Then we obtain the following
chain of such continuous and dense embeddings: Vσ ⊂ V ⊂ H ⊂ V ∗ ⊂ Vσ∗ .
Let 0 ≤ τ < T < +∞. We set

K τ,T := {y ∈ L p (τ, T ; V ) : y(t) ∈ K for a.a. t ∈ (τ, T )}.

By a physical solution of evolutionary variational inequality (1.56) on an interval

[τ, T ] we understand an element y that belongs to the space K τ,T ∩ C([τ, T ]; H )
such that
 T  T
− (ξ (t), y(t))dt + A(y(t)), ξ(t)V dt ≥ 0 ∀ξ ∈ C0∞ ([τ, T ]; V ) ∩ K τ,T ,
τ τ
 (1.57)
t2
||y(t2 )||2H − ||y(t1 )||2H +2 A(y(t)), y(t)V dt = 0 ∀t1 , t2 ∈ [τ, T ]; (1.58)
t1

see also Fig. 1.5.

Note that the concept of a physical solution naturally weakens the concept of a
strong solution of unilateral Problem (1.56). The concept of a physical solution is
a weak solution of Problem (1.56) that is continuous as a mapping from the time
interval [τ, T ] into the phase space H and satisfies energy equality (1.58). Of course,
each strong solution of Problem (1.56) is a physical solution of this problem. At
present, for Assumptions 1, 2, 3, 4 and 5, only the fact of existence of weak solutions
is well known.
Now, with the help of the penalty method, we establish the fact of existence of
physical solutions to Problem (1.56) for arbitrary initial data from K . The obtained
results will be applied in the next sections to the investigation of the dynamics of
processes and fields of different nature under unilateral constraints.
1.3 Evolutional Variational Inequalities: Penalty Method and Strong Solutions 35

Fig. 1.5 Classes of solutions

for evolution variation
inequalities

For a fixed 0 ≤ τ < T < +∞ we consider

∗ ∗
X τ,T = L p (τ, T ; V ), X τ,T = L q (τ, T ; V ∗ ), Wτ,T = {y ∈ X τ,T : y ∈ X τ,T },
∗
Aτ,T : X τ,T → X τ,T , (Aτ,T (y))(t) = A(y(t)) for a.a. t ∈ (τ, T ),
Yτ,T,σ = L 1 (τ, T ; Vσ∗ ), Wτ,T,σ = {y ∈ X τ,T : y ∈ Yτ,T,σ },

where y is the derivative of an element y ∈ X τ,T in the sense of the space of distri-
butions D([τ, T ]; V ∗ ) (see, for example [12, Definition IV.1.10, p. 168]). Note that
the space Wτ,T is a reflexive Banach space with the following derivative graph norm
(see, for example, [48, Statement 4.2.1, p. 291]): uWτ,T = u X τ,T +u  X τ,T ∗ , u∈
Wτ,T .
It follows from [52, Sect. 2.2], [46, pp. 152–157], and Assumptions 1, 2, 3 and 4
∗
that Aτ,T : X τ,T → X τ,T satisfies the following properties.
Property 1. ∃C1 > 0: Aτ,T (y) X τ,T ∗ ≤ C1 (1 + y Xp−1τ,T ) ∀y ∈ X τ,T .
Property 2. ∃C2 , C3 > 0: Aτ,T (y), y X τ,T ≥ C2 y Xp τ,T − C3 ∀y ∈ X τ,T .
∗
Property 3. Aτ,T : X τ,T → X τ,T is (generalized) pseudomonotone on Wτ,T,σ and
satisfies the (S)-property, that is, the facts that yn → y weakly in X τ,T , {yn }n=1,2,... is
∗
bounded in Yτ,T,σ , Aτ,T (yn ) → d weakly in X τ,T , and lim Aτ,T (yn ), yn − y X τ,T ≤
n→+∞
0, imply that d = Aτ,T (y) and yn → y in X τ,T .
In particular, the following equalities hold:

lim Aτ,T (yn ), yn  X τ,T = d, y X τ,T ,

n→+∞

 T
lim |A(yn (t)), yn (t) − y(t)V |dt = 0.
n→+∞ τ
36 1 Qualitative Methods for Classes of Nonlinear …

Fig. 1.6 Sufficient conditions for the existence of a solution for evolution variation inequality

∗ ∗
Here, ·, · X τ,T : X τ,T × X τ,T → R is the pairing in X τ,T × X τ,T that coincides with
the scalar product in L 2 (τ, T ; H ) on L 2 (τ, T ; H ) × X τ,T , that is,
 T
∀u ∈ L 2 (τ, T ; H ), ∀v ∈ X τ,T u, v X τ,T = (u(t), v(t))dt.
τ

Note also that (see [12, Theorem IV.1.17, p. 177]) the embedding Wτ,T ⊂ C([τ,
T ]; H ) is continuous and dense. Moreover,
 T
(u(T ), v(T )) − (u(τ ), v(τ )) = [u (t), v(t)V + v (t), u(t)V ]dt, (1.59)
τ

for each u, v ∈ Wτ,T .

The main result of this section has the following formulation; Fig. 1.6.

Theorem 1.3 Let Assumptions 1, 2, 3, 4 and 5 be satisfied, 0 ≤ τ < T < +∞.

Then, for any yτ ∈ K there is at least one physical solution y of Problem (1.56) on
[τ, T ], and this solution is such that y(τ ) = yτ .
1.3 Evolutional Variational Inequalities: Penalty Method and Strong Solutions 37

Proof We use the penalty method. Let PK be the operator of orthogonal projection
of an arbitrary element of the space V onto the convex cone K . Let J : V → V ∗
be the dual operator, that is, the mapping satisfying the following two equalities:
p−1
J (v)V ∗ vV = J (v), vV and J (v)V ∗ = vV for an arbitrary v ∈ V . By the
Asplund theorem, the space V can be renormalized by an equivalent strict norm so
that the corresponding norm in the conjugated space V ∗ is also strict and equivalent
to its natural norm. Therefore, the operator J can be considered as single-valued.
We will use β(v) = J (v − PK v), v ∈ V , in the capacity of the penalty operator.
Note that β(v) = 0̄ if and only if v ∈ K . Moreover, β(αv) = α|α| p−2 β(v) and
β(v), vV = 0 for some arbitrary v ∈ V and α ∈ R. Hereafter, we consider that
(β(y))(t) = β(y(t)) for a.a. t ∈ (τ, T ) for all y ∈ X τ,T .
∗
Since β : X τ,T → X τ,T is a bounded monotone demicontinuous operator, for an
arbitrary ε > 0, the mapping Aε (y) := Aτ,T (y) + 1ε β(y), y ∈ X τ,T , is generalized
pseudomonotone on Wτ,T (satisfies property 3). Moreover, the penalty operator def-
inition (defining the properties of the dual mapping J ) and properties 1 and 2 for the
operator Aτ,T imply properties 1 and 2 for the new operator Aε acting from X τ,T to
∗
X τ,T . Thus, [51, Theorem 2.4, p. 123] implies the existence of a solution yε ∈ Wτ,T
to the following problem:

1
yε + Aτ,T (yε ) + β(yε ) = 0̄, yε (τ ) = yτ . (1.60)
ε
Moreover, formula (1.59), the monotonicity of β, and the fact that K is a cone, imply
the following relationships:

− ξ , yε  X τ,T + Aτ,T (yε ), ξ  X τ,T ≥ 0 ∀ξ ∈ C0∞ ([τ, T ]; V ) ∩ K τ,T , (1.61)

 t2
yε (t2 )2H − yε (t1 )2H + 2 A(yε (t)), yε (t)V dt = 0 ∀t1 , t2 ∈ [τ, T ]. (1.62)
t1

Assumptions 2 and 3 imply the existence a constant C4 > 0 such that

yε C([τ,T ];H ) ≤ C4 , yε  X τ,T ≤ C4 , Aτ,T (yε ) X τ,T

∗ ≤ C4 ∀ε > 0. (1.63)

Let us prove the existence of a constant C5 > 0 such that

yε Yτ,T,σ ≤ C5 ∀ε > 0. (1.64)

Assumption 5 implies the existence of vσ ∈ K σ and rσ > 0 such that {v ∈ Vσ :

v−vσ Vσ ≤ rσ } ⊂ K σ . Since K σ = K ∩ Vσ is a cone, without loss of generality, we
can consider that vσ Vσ = 1 and rσ ≤ 1. We put M := {v ∈ K σ : v − vσ Vσ ≤ 1}
and N := (M −vσ )∩(vσ − M). The set N is convex, closed, absorbing, and balanced.
Thus, for the set N the Minkowski functional ρ N (ω) := inf{t > 0 : ωt ∈ N }, ω ∈ Vσ
is correctly defined. Moreover, ρ N satisfies the following three properties:
38 1 Qualitative Methods for Classes of Nonlinear …

(1) ωVσ ≤ ρ N (ω) ≤ r1σ ωVσ for any ω ∈ Vσ ;

(2) ρ N (vσ ) = 1;
(3) {ω ∈ Vσ : ρ N (ω − vσ ) ≤ 1} ⊂ K σ .
We put ρ N∗ (g) := sup{g, ωVσ : ω ∈ Vσ , ρ N (ω) ≤ 1}, g ∈ Vσ∗ . Property (1) for
ρ N provides the equivalence between the norm ρ N∗ and the natural norm of the space
Vσ∗ . Consider now K σ− := {g ∈ Vσ∗ : g, ωVσ ≤ 0 ∀ω ∈ K σ }. Properties (2) and (3)
of ρ N imply
ρ N∗ (g) = −g, vσ Vσ ∀g ∈ K σ− . (1.65)

Since K ⊆ V is a cone, the monotonicity of β : V → V ∗ guarantees that β(ω) ∈ K σ−

for an arbitrary ω ∈ V . Therefore, statements (1.59), (1.60), and (1.65) provide the
fulfillment of the equalities
T
1
ε
β(yε )Yτ,T,σ = − 1ε τ β(yε (t)), vσ V dt
T
= τ A(yε (t)), vσ V dt + (yε (T ) − yε (τ ), vσ ) H

for any ε > 0. Thus, from inequalities (1.63), we obtain

1 1
β(yε )Yτ,T,σ ≤ C4 vσ V (T − τ ) p + 2C4 vσ  H ∀ε > 0. (1.66)
ε
Finally, inequality (1.64) follows from Problem (1.60) and inequalities (1.63) and
∗
(1.66) since the embedding X τ,T ⊂ Yτ,T,σ is continuous and dense.
The following equality holds:

β(yε ), yε  X τ,T = 0 ∀ε > 0, (1.67)

since β(v), vV = 0 for an arbitrary v ∈ V . Moreover, from the monotonicity of β

and the BanachSteinhaus theorem, we obtain

∃C5 > 0 : β(yε ) X τ,T

∗ ≤ C5 ∀ε ∈ (0, 1). (1.68)

In fact, for an arbitrary ω ∈ X τ,T , the monotonicity of β, estimate (1.63) and equality
(1.67) imply the following:

sup β(yε ), ω X τ,T ≤ sup β(yε ), ω − yε  X τ,T + sup β(yε ), yε  X τ,T

ε∈(0,1) ε∈(0,1) ε∈(0,1)
= sup β(ω), ω − yε  X τ,T ≤ β(ω) X τ,T
∗ (ω
X τ,T + C 4 ) < ∞.
ε∈(0,1)

Hence, we obtain inequality (1.68) from the Banach–Steinhaus theorem.

From a priori estimates (1.63), (1.64), and (1.66), and the lemma on the com-
pactness of the embedding Wτ,T ⊂ L 2 (τ, T ; H ) (by virtue of the compactness of
the embedding V ⊂ H ), we obtain as a corollary of Banach–Alaoglu theorem that
∗
there is a sequence εn 0, n → ∞, and elements y ∈ X τ,T and d ∈ X τ,T such that
1.3 Evolutional Variational Inequalities: Penalty Method and Strong Solutions 39

y(τ ) = yτ and the following convergences take place:

yεn  y in X τ,T , yεn (t) → y(t) in H for a.a. t ∈ (τ, T ),

∗ (1.69)
yεn (T )  y(T ) in H, Aτ,T (yεn )  d in X τ,T n → ∞.

Moreover, from inequalities (1.66) and (1.68), we obtain that

∗
β(yεn ) → 0 in X τ,T n → ∞. (1.70)

Let us show that y ∈ K τ,T . It follows from convergences (1.69) and (1.70) that
lim β(yεn ), yεn − y X τ,T = 0. Since the monotone demicontinuous operator β is
n→∞
pseudomonotone, taking into account convergence (1.70), the following inequality
holds:
0 = lim β(yεn ), yεn − ω X τ,T ≥ β(y), y − ω X τ,T ,
n→∞

∀ω ∈ X τ,T . Thus, β(y(t)) ∈ K for a.a. t ∈ (τ, T ). Therefore, y ∈ K τ,T since

y ∈ X τ,T .
Let us show that
lim Aτ,T (yεn ), yεn − y X τ,T ≤ 0. (1.71)
n→∞

In fact, Problem (1.60), the monotonicity of β and formula (1.59) imply

Aτ,T (yεn ), yεn − v X τ,T = 1ε β(yεn ), v − yεn  X τ,T + yεn , v − yεn  X τ,T
≤ 1ε β(yεn ), v − yεn  X τ,T + v , v − yεn  X τ,T (1.72)
≤ ε β(v), v − yεn  X τ,T + v , v − yεn  X τ,T ≤ v , v − yεn  X τ,T ,
1

for arbitrary n = 1, 2, . . . and v ∈ Wτ,T ∩ K τ,T since β(v) = 0̄. Thus, convergence
(1.69) implies the inequality

lim Aτ,T (yεn ), yεn  X τ,T ≤ d, v X τ,T + v , v − y X τ,T , (1.73)

n→∞

for all v ∈ Wτ,T ∩ K τ,T . Since 0̄ ∈ K τ,T − ωτ for ωτ ≡ yτ ∈ K τ,T , [29, p. 284]
implies the existence of a sequence {v j } j=1,2,... ⊂ (K τ,T − ωτ ) ∩ Wτ,T such that

(a) v j (τ ) = 0̄ for all j = 1, 2, . . . ;

(b) v j → y − ωτ in X τ,T as j → ∞;
(c) lim v j , v j + ωτ − y X τ ,T ≤ 0.
j→∞

Putting v = v j + ωτ ∈ K τ,T ∩ Wτ,T , j = 1, 2, . . . in inequality (1.73), we obtain that

lim Aτ,T (yεn ), yεn  X τ,T ≤ d, y X τ,T . The last inequality together with convergences
n→∞
(1.69) and inequalities (1.72) implies inequality (1.71). We will use the pseudo-
monotonicity of Aτ,T on Wτ,T,σ . It follows from inequality (1.64), convergences
(1.69), and inequality (1.71) that d = Aτ,T (y), lim Aτ,T (yεn ), yεn  X τ,T = d, y X τ,T ,
n→∞
40 1 Qualitative Methods for Classes of Nonlinear …

and  T  
A(yε (t)), yε (t) − y(t)V  dt → 0 n → ∞. (1.74)
n n
τ

Moreover, inequalities (1.72) additionally imply inequality (1.57).

To complete the verification of the fact that y is a physical solution to Problem
(1.56) on [τ, T ], it remains to verify that y ∈ C([τ, T ]; H ) and that y satisfies energy
equality (1.58).
It follows from formulas (1.59) and (1.60) (see also formula (1.62)) that

d
yεn (t)2H = −A(yεn (t)), yεn (t)V
dt

for a.a. t ∈ (τ, T ). We obtain from formula (1.74) and the last convergence in
formula (1.69) that the sequence {t → dtd yεn (t)2H }n=1,2,... of measurable real-
valued functions on (τ, T ) is uniformly integrable, that is, there is a subsequence
{yεm }m ⊆ {yεn }n such that the sequence {t → dtd yεm (t)2H }m weakly converges in
L 1 (τ, T ) to an element −A(y( · )), y( · )V ∈ L 1 (τ, T ). Hence, the sequence {t →
d
dt
yεm (t)2H }m , on the one hand, converges in the space D ∗ (τ, T ) (of generalized
functions on [τ, T ]) to a regular generalized function −A(y( · )), y( · )V ∈
L 1 (τ, T ). On the other hand, the sequence {t → yεm (t)2H }m converges in the
space D ∗ (τ, T ) to the measurable function y( · )2H essentially bounded on (τ, T ).
Thus, the sequence {t → dtd yεm (t)2H }m converges in the space D ∗ (τ, T ) to the
generalized function dtd y( · )2H . Thus, by virtue of the uniqueness of the limit in
the space D ∗ (τ, T ), dtd y( · )2H = −A(y( · )), y( · )V ∈ L 1 (τ, T ), which, in
view of formula (1.59), implies a priori estimate (1.58).
Let us show that
y(t) − yτ  H → 0 t  τ+ . (1.75)

Since yτ ∈ K and β(v), vV = 0 for arbitrary v ∈ V , formula (1.60) implies

t
(yεn (t), yτ ) − yτ 2H = τ (yεn (s), yτ )ds
t t
= − τ A(yεn (s)), yτ V dt − 1ε τ β(yεn (s)), yτ V dt
t 1 t
= − τ A(yεn (s)), yτ V dt + ε τ β(yεn (s)), yεn (s) − yτ V dt
t
≥ − τ A(yεn (s)), yτ V dt

for arbitrary n = 1, 2, . . . and t ∈ (τ, T ). Otherwise,

(yεn (t), yτ ) − yτ 2H ≤ yτ  H (yεn (t) H − yτ  H )

for arbitrary n = 1, 2, . . ..
Thus,
t
− τ A(yεn (s)), yτ V dt ≤ (yεn (t), yτ ) − yτ 2H
≤ yτ  H (yεn (t) H − yτ  H ), n = 1, 2, . . . , t ∈ (τ, T ),
1.3 Evolutional Variational Inequalities: Penalty Method and Strong Solutions 41

and, taking into account convergences (1.69), we obtain

t
− τ A(y(s)), yτ V dt ≤ (y(t), yτ ) − yτ 2H
≤ yτ  H (y(t) H − yτ  H ) for a.a. t ∈ (τ, T ).

Since A(y( · )), yτ V ∈ L 1 (τ, T ), energy equality (1.58) provides the last two
inequalities for all t ∈ [τ, T ]. Hence

(y(t), yτ ) − yτ 2H → 0 as t  τ+ . (1.76)

To complete the proof of property (1.75) note that

||y(t) − yτ ||2H = ||y(t)||2H + ||yτ ||2H − 2(y(t), yτ ), t ∈ [τ, T ].

Hence, energy equality (1.58) and property (1.76) imply property (1.75).
From the monotonicity of β and formulas (1.59) and (1.60) we obtain the
inequality

yεn (t + h) − yεn (t)2H

≤ yεn (τ + h) − yτ 2H − 2 τt A(yεn (s + h)) − A(yεn (s)), yεn (s + h) − yεn (s)V dt

for arbitrary t ∈ (τ, T − h) and h ∈ (0, T − τ ). From property 3, convergences

(1.69) and (1.74), and the last inequality we obtain

y(t + h) − y(t)2H
t
≤ y(τ + h) − yτ 2H− 2 τ A(y(s + h)) − A(y(s)), y(s + h) − y(s)V dt
(1.77)
for a.a. t ∈ (τ, T −h) and arbitrary h ∈ (0, T −τ ). With allowance for energy equality
(1.58), inequality (1.77) takes place for all t ∈ (τ, T − h) and h ∈ (0, T − τ ). Thus,
formulas (1.75) and (1.77) imply the property of continuity of y as a mapping from
a time interval [τ, T ] into the phase space H .
The theorem is proved.
For fixed τ < T , we introduce the notation:

Dτ,T (yτ ) = {y(·) | y is a physical solution to inequality (1.56) on [τ, T ], y(τ ) = yτ },

yτ ∈ K .

It follows from Theorem 1.3 that Dτ,T (yτ ) = ∅ and Dτ,T (yτ ) ⊂ C([τ, T ]; H )
∀τ < T, yτ ∈ K . Moreover, the conditions imposed on the parameters of Problem
(1.56) and the generalized Gronwall–Bellman lemma [2] imply the existence of C4 ,
C5 , C6 , C7 > 0 such that, for any finite time interval [τ, T ] each physical solution
y to Problem (1.56) on [τ, T ] satisfies the following estimate ∀t ≥ s, t, s ∈ [τ, T ]:
 t
p
y(t)2H + C4 y(ξ )V dξ ≤ y(s)2H + C5 (t − s), (1.78)
s
42 1 Qualitative Methods for Classes of Nonlinear …

y(t)2H ≤ y(s)2H e−C6 (t−s) + C7 . (1.79)

Moreover, the translation and concatenation of physical solutions to Problem (1.56)

on finite time intervals are physical solutions of this problem on the corresponding
intervals. It follows from the proof of Theorem 1.3 that the penalty method guarantees
the existence of physical solutions to Problem (1.56) on a finite time interval that are
equicontinuous as mappings from a time interval [τ, T ] into the phase space H if
they start from a bounded subset of the natural phase space H (that is, the statements
of the theorems on the strong convergence of solutions from [24]). Thus (see [24]),
physical solutions (a) can be extended to global solutions defined on the positive time
semiaxis; (b) tends uniformly to a small (compact) subset of the natural phase space
H (as time t → +∞), and this subset is independent of the bounded set from which
they have started. Proceeding from the results of [24], such an attracting set consists
of complete trajectories of Problem (1.56) that are defined on the entire real line.
Thus, the results of (1.56) allow one to globally describe the dynamics of solutions
to such problems by finite algorithms up to an arbitrary small parameter.

1.4 Nonlinear Parabolic Equations of Divergent Form

Consider now an example of the class of nonlinear boundary problems for which the
dynamics of solutions can be investigated as t → +∞. Note that our consideration
does not pretend to generality.
Assume that n ≥ 2, m ≥ 1, p ≥ 2, 1 < q ≤ 2, 1p + q1 = 1, and Ω ⊂ Rn is a
bounded domain with a sufficiently smooth boundary Γ = ∂Ω, N1 (N2 accordingly)
is the number of differentiations with respect to x of order of ≤ m − 1 (of order of
= m accordingly). Let Aα (x, η; ξ ) be the family of real functions (|α| ≤ m) that are
defined in Ω × R N1 × R N2 and satisfy the following conditions:
(i) for a.a. x ∈ Ω a function (η, ξ ) → Aα (x, η, ξ ) is continuous in R N1 × R N2 ;
(ii) ∀(η, ξ ) ∈ R N1 × R N2 a function x → Aα (x, η, ξ ) is measurable in Ω;
(iii) there are c1 ≥ 0 and k1 ∈ L q (Ω) such that, for a.a. x ∈ Ω and ∀(η, ξ ) ∈
R N1 × R N2
|Aα (x, η, ξ )| ≤ c1 [|η| p−1 + |ξ | p−1 + k1 (x)];

(iv) there are c2 > 0 and k2 ∈ L 1 (Ω) such that, for a.a x ∈ Ω and ∀(η, ξ ) ∈
R N1 × R N2 
Aα (x, η, ξ )ξα ≥ c2 |ξ | p − k2 (x);
|α|=m

(v) there is an increasing real-valued function υ such that, for a.a. x ∈ Ω, ∀η ∈ R N1 ,

and ∀ξ, ξ ∗ ∈ R N2 , ξ = ξ ∗ the following inequality holds:
1.4 Nonlinear Parabolic Equations of Divergent Form 43

(Aα (x, η, ξ ) − Aα (x, η, ξ ∗ ))(ξα − ξα∗ ) ≥ (υ(ξα ) − υ(ξα∗ ))(ξα − ξα∗ ).
|α|=m

We introduce the denotations D k u = {D β u, |β| = k} and δu = {u, Du, . . . , D m−1 u}

[48, p. 194].
For an arbitrary fixed external force f ∈ L 2 (Ω), we will investigate the dynamics
of the following problem for all weak (generalized) solutions defined on [0, +∞) as
t → +∞:

∂ y(x, t) 
+ (−1)|α| D α Aα (x, δy(x, t), D m y(x, t)) = f (x) in Ω × (0, +∞),
∂t |α|≤m
(1.80)
D α y(x, t) = 0 on Γ × (0, +∞), |α| ≤ m − 1,
(1.81)
y(x, t) ≥ 0 for a.a. (x, t) ∈ Ω × (0, +∞).
m, p
We introduce the following denotations [48, p. 195]: H = L 2 (Ω), V = W0 (Ω),
m+σ, p m, p
Vσ = W0 (Ω), σ  1, is the Sobolev real space, K = {y ∈ W0 (Ω) : y(x) ≥
0 for a.a. x ∈ Ω}, and
 
a(u, ω) = Aα (x, δu(x), D m u(x))D α ω(x)d x, u, ω ∈ V.
|α|≤m Ω

Taking into account conditions (i)–(v) and [29, pp. 192–199], the operator A : V →
V ∗ defined by the formula A(u), ωV = a(u, ω) ∀u, ω ∈ V satisfies the basic
assumptions. Therefore, we can pass from Problems (1.80), (1.81) to corresponding
Problem (1.56). Note that

A(u) = (−1)|α| D α Aα (x, δu, D m u) ∀u ∈ C0∞ (Ω).
|α|≤m

Thus, for physical solutions to Problems (1.80), (1.81), all the statements from the
previous sections hold.

References

1. Aubin, J.P., Ekeland, I.: Applied Nonlinear Analysis. Mir, Moscow (1988)
2. Ball, J.M.: Continuity properties and global attractors of generalized semiflows and the Navier–
Stokes equations. J. Nonlinear Sci. 7(5), 475–502 (1997)
3. Barbu, V.: Nonlinear Semigroups and Differential Equations in Banach Spaces. Editura Acad,
Bucuresti (1976)
4. Bearman, P.W., Obasaju, E.D.: An experimental study of pressure fluctuations on fixed and
oscillating square-section cylinders. J. Fluid Mech. 119, 297–321 (1982)
5. Browder, F.E., Hess, P.: Nonlinear mappings of monotone type in Banach spaces. J. Funct.
Anal. (1972). doi:10.1016/0022-1236(72)90070-5
44 1 Qualitative Methods for Classes of Nonlinear …

6. Browder, F.E.: Pseudomonotone operators and nonlinear elliptic boundary value problems on
unbounded domains. Proc. Natl. Acad. Sci. 74, 2659–2661 (1977)
7. Carl, S., Motreanu, D.: Extremal solutions of quasilinear parabolic inclusions with generalized
Clarke’s gradient. J. Differ. Equ. (2003). doi:10.1016/S0022-0396(03)00022-6
8. Chikrii, A.A.: Conflict-Controlled Processes. Kluver Academic Publishers, Boston (1997)
9. Davis, R.W., Moore, E.F.: A numerical study of vortex shedding from rectangles. J. Fluid Mech.
116, 475–506 (1982)
10. Denkowski, Z., Migorski, S., Papageorgiou, N.S.: An Introduction to Nonlinear Analysis.
Kluwer Academic Publishers, Boston (2003)
11. Duvaut, G., Lions, J.L.: Inequalities in Mechanics and in Phisycs. Nauka, Moskow (1980)
12. Gajewski, H., Gröger, K., Zacharias, K.: Nichtlineare Operatorgleichungen und Operatordif-
ferentialgleichungen. Akademie, Berlin (1974)
13. Gorban, N.V., Kapustyan, O.V., Kasyanov, P.O.: Uniform trajectory attractor for non-
autonomous reaction-diffusion equations with Carathéodory’s nonlinearity. Nonlinear Anal.
98, 13–26 (2014)
14. Guan, Z., Karsatos, A.G., Skrypnik, I.V.: Ranges of densely defined generalized
pseudomonotone perturbations of maximal monotone operators. J Differ. Equ. (2003). doi:10.
1016/S0022-0396(02)00066-9
15. Hu, S., Papageorgiou, N.S.: Handbook of Multi-Valued Analysis, Vol. I: Theory. Kluwer Aca-
demic Publishers, Dordrecht (1997)
16. Hu, S., Papageorgiou, N.S.: Handbook of Multi-Valued Analysis, Vol. II: Applications. Kluwer
Academic Publishers, Dordrecht (1997)
17. Kalita, P., Lukaszewicz, G.: Global attractors for multi-valued semiflows with weak continuity
properties. Nonlinear Anal. Theory Methods Appl. 101, 124–143 (2014)
18. Kapustyan, V.O., Kasyanov, P.O., Kogut, O.P.: On solvability for one class of parameterized
operator inclusions. Ukr. Math. J. (2008). doi:10.1007/s11253-009-0179-z
19. Kasyanov, P.O., Mel’nik, V.S., Toscano, S.: Periodic solutions for nonlinear evolution equations
with Wλ0 -pseudomonotone maps. Nonlinear Oscil. (2006). doi:10.1007/s11072-006-0037-y
20. Kasyanov, P.O., Melnik, V.S., Valero, J.: On the method of approximation for evolu-
tionary inclusions of pseudo monotone type. Bull. Aust. Math. Soc. (2008). doi:10.1017/
S0004972708000130
21. Kasyanov, P.O., Melnik, V.S., Yasinsky, V.V.: Evolution Inclusions and Inequalities in Banach
Spaces with wλ -Pseudomonotone Maps. Naukova Dumka, Kiev (2007)
22. Kasyanov, P.O., Mel’nik, V.S., Piccirillo, A.M.: Local subdifferentials and multivariational
inequalities in Banach and Frechet spaces. Opusc. Math. 28, 295–311 (2008)
23. Kasyanov, P.O., Mel’nik, V.S., Toscano, S.: Solutions of Cauchy and periodic problems for
evolution inclusions with multi-valued wλ0 -pseudomonotone maps. J. Differ. Equ. 249(6),
1258–1287 (2010)
24. Kasyanov, P.O.: Multi-valued dynamics of solutions of an autonomous differential-operator
inclusion with pseudomonotone nonlinearity. Cybern. Syst. Anal. 47(5), 150–163 (2011)
25. Kasyanov, P.O., Melnik, V.S.: Faedo–Galerkin method differential-operator inclusions in
Banach spaces with maps of wλ0 -pseudomonotone type. Natl. Acad. Sci. Ukr. Kiev, Inst.
Math. Prepr. 1, 82–105 (2005). Part 2
26. Kasyanov, P.O., Melnik, V.S.: On solvabbility of differential-operator inclusions and evolution
variation inequalities generated by wλ0 -pseudomonotone maps type. Ukr. Math. Bull. 4, 535–
581 (2007)
27. Kuttler, K.: Non-degenerate implicit evolution inclusions. Electron. J. Differ. Equ. 34, 1–20
(2000)
28. Lions, J.L.: Quelques methodes de resolution des problemes aux limites non lineaires. Dunod
Gauthier-Villars, Paris (1969)
29. Lions, J.-L.: Some Methods of Solution of Nonlinear Boundary-Value Problems [Russian
Translation]. Mir, Moskow (1972)
30. Liu, Z., Migórski, S.: Noncoercive Damping in Dynamic Hemivariational Inequality with
Application to Problem of Piezoelectricity. Discret. Contin. Dyn. Syst. Ser. B (2008). doi:10.
3934/dcdsb.2008.9.129
References 45

31. Mel’nik, V.S.: Critical points of some classes of multi-valued mappings. Cybern. Syst. Anal.
(1997). doi:10.1007/BF02665895
32. Mel’nik, V.S.: Multivariational inequalities and operator inclusions in banach spaces with
mappings of the class (S)+ . Ukr. Mat. Zh. (2000). doi:10.1023/A:1010431221039
33. Mel’nik, V.S.: Topological methods in the theory of operator inclusions in Banach spaces. I.
Ukr. Math. J. (2006). doi:10.1007/s11253-006-0062-0
34. Migórski, S.: Boundary hemivariational inequalities of hyperbolic type and applications. J.
Global Optim. (2005). doi:10.1007/s10898-004-7021-9
35. Minewitsch, S., Franke, R., Rodi, W.: Numerical investigation of laminar vortex-shedding flow
past a square cylinder oscillating in line with the mean flow. J. Fluids Struct. 8, 787–802 (1994)
36. Naniewicz, Z., Panagiotopoulos, P.D.: Mathematical Theory of Hemivariational Inequalities
and Applications. Marcel Dekker, New York (1995)
37. Okajima, A.: Strouhal number of rectangular cylinders. J. Fluid Mech. 123, 379–398 (1982)
38. Panagiotopoulos, P.D.: Inequality Problems in Mechanics and Applications. Convex and Non-
convex Energy Functions. Birkhauser, Basel (1985)
39. Panagiotopoulos, P.D.: Hemivariational Inequalities, Applications in Mechanics and Engineer-
ing. Springer, Berlin (1993)
40. Perestyuk, N.A., Plotnikov, V.A., Samoilenko, A.M., Skrypnik, N.V.: Impulse differential equa-
tions with multi-valued and discontinuous right-hand side. Institute of mathematics NAS of
Ukraine, Kyiv (2007)
41. Perestyuk, M.O., Kasyanov, P.O., Zadoyanchuk, N.V.: On Faedo-Galerkin method for evolution
inclusions with wλ0 -pseudomonotone maps. Mem. Differ. Equ. Math. Phys. 44, 105–132
(2008)
42. Skrypnik, I.V.: Methods of Investigation of Nonlinear Elliptic Boundary Problems. Nauka,
Moscow (1990)
43. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Springer,
New York (1988)
44. Vickery, B.J.: Fluctuating lift and drag on a long cylinder of square cross-section in a smooth
and in a turbulent stream. J. Fluid Mech. 25, 481–494 (1966)
45. Zadoyanchuk, N.V., Kas’yanov, P.O.: Faedo-Galerkin method for second-order evolution inclu-
sions with Wλ -pseudomonotone mappings. Ukr. Math. J. (2009). doi:10.1007/s11253-009-
0207-z
46. Zgurovsky, M.Z., Kasyanov, P.O.: Multi-valued dynamics of solutions for autonomous operator
differential equations in strongest topologies. In: Continuous and Distributed Systems: Theory
and Applications. Solid Mechanics and Its Applications, vol. 211, pp. 149–162 (2014)
47. Zgurovsky, M.Z., Melnik, V.S.: Ky Fan inequality and operational inclusions in Banach spaces.
Cybern. Syst. Anal. (2002). doi:10.1023/A:1016391328367
48. Zgurovsky, M.Z., Kasyanov, P.O., Melnik, V.S.: Differential-Operator Inclusions and Varia-
tional Inequalities in Infinitely Dimensional Spaces [in Russian]. Naukova Dumka, Kyiv (2008)
49. Zgurovsky, M.Z., Kasyanov, P.O., Melnik, V.S.: Differential-Operator Inclusions and Variation
Inequalities in Infinitedimensional Spaces (in Russian). Naukova Dumka, Kyiv (2008)
50. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation
Inequalities for Earth Data Processing I. Springer, Heidelberg (2010). doi:10.1007/978-3-
642-13837-9
51. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation
Inequalities for Earth Data Processing II. Springer, Springer (2010). doi:10.1007/978-3-642-
13878-2
52. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolu-
tion Inclusions and Variation Inequalities for Earth Data Processing III. Series: Advances in
Mechanics and Mathematics, vol. 27. Springer, Berlin (2012)
53. Zgurovsky, M.Z., Melnik, V.S.: Nonlinear Analysis and Control of Physical Processes and
Fields. Springer, Berlin (2004)
Chapter 2
Regularity of Solutions for Nonlinear
Systems

Abstract In this chapter we establish sufficient conditions for regularity of all weak
solutions for nonlinear systems. We note that the respective Cauchy problems may
have nonunique weak solution. In Sect. 2.1 we establish regularity of all weak solu-
tions for parabolic feedback control problems. Section 2.2 devoted to artificial con-
trol method for nonlinear partial differential equations and inclusions. The regularity
of all weak solutions is obtained. In Sect. 2.3 we consider regularity results of all
weak solutions for nonlinear reaction-diffusion systems with nonlinear growth. In
Sect. 2.4 we consider the following examples of applications: a parabolic feedback
control problem; a model of conduction of electrical impulses in nerve axons; a cli-
mate energy balance model; FitzHugh–Nagumo System; a model of combustion in
porous media.

2.1 Regularity of All Weak Solutions for a Parabolic

Feedback Control Problem

Let Ω ⊂ Rn , n ≥ 1, be bounded and open subset with a smooth boundary ∂Ω, f ,

f : R → R are some real functions. We consider the semilinear reaction-diffusion
inclusion:

u t − u + [ f (u), f (u)]  0 in Ω × (τ, T ), (−∞ < τ < T < +∞), (2.1)

with boundary condition 

u ∂Ω = 0, (2.2)

where [a, b] = {αa + (1 − α)b | α ∈ [0, 1]}, a, b ∈ R. We suppose that f = [ f , f ] :

R → 2R \ {∅} satisfies the growth condition

∃c0 > 0 : −c0 (1 + |u|) ≤ f (u) ≤ f (u) ≤ c0 (1 + |u|) ∀u ∈ R. (2.3)

Suppose also that f is lower semi-continuous, and f is upper semi-continuous.

© Springer International Publishing AG 2018 47
M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_2
48 2 Regularity of Solutions for Nonlinear Systems

We shall use the following standard notations: H = L 2 (Ω), V = H01 (Ω), V is

the dual space of V . The function u(·) ∈ L 2 (τ, T ; V ) is a weak solution of Problem
(2.1) and (2.2) on [τ, T ], if there exists a measurable function d : Ω × (τ, T ) → R
such that

d(x, t) ∈ [ f (u(x, t)), f (u(x, t))] for a.e. (x, t) ∈ Ω × (τ, T ); (2.4)
 T    T  T
dξ
− u, dt + (∇u, ∇ξ ) d xdt + (d, ξ ) d xdt = 0 (2.5)
τ dt τ Ω τ Ω

for all ξ ∈ C0∞ (Ω × (τ, T )), where ·, · denotes the pairing in the space V .
We note that Problem (2.1) and (2.2) arises in many important models for distrib-
uted parameter control problems and that large class of identification problems enter
this formulation. Let us indicate a problem which is one of motivations for the study
of the autonomous evolution inclusion (2.1) (cf. [37, 56] and references therein). In
a subset Ω of R3 , we consider the nonstationary heat conduction equation (Figs. 2.1
and 2.2):
∂y
− y = f in Ω × (0, +∞)
∂t

with initial conditions and suitable boundary ones. Here y = y(x, t) represents the
temperature at the point x ∈ Ω and time t > 0. It is supposed that f = f 1 + f 2 ,
where f 2 is given and f 1 is a known function of the temperature of the form

− f 1 (x, t) ∈ ∂ j (x, y(x, t)) a.e. (x, t) ∈ Ω × (0, +∞);

Figure 2.3 Here ∂ j (x, ξ ) denotes generalized gradient of Clarke (see [12]) with
respect to the last variable of a function j : Ω × R → R which is assumed to be
locally Lipschitz in ξ (cf. [37] and references therein). The multi-valued function
∂ j (x, ·) : R → 2R is generally nonmonotone and it includes the vertical jumps. In a
physicist’s language it means that the law is characterized by the generalized gradient
of a nonsmooth potential j (cf. [39]).

Fig. 2.1 Diffusion processes

2.1 Regularity of All Weak Solutions for a Parabolic Feedback Control Problem 49

Fig. 2.2 Idealized physical

setting for heat conduction in
a rod with homogeneous
boundary conditions

Fig. 2.3 Feedback control

diagram

Another motivations connected with parabolic equations with a discontinuous

nonlinearity. In [43] it is considered the case, when f is the difference of maxi-
mal monotone maps. Global attractor in phase space H for such type equations is
considered there. Obtained inclusion is a particular case of an abstract differential
inclusion generated by a difference of subdifferential maps of proper convex lower
semicontinuous functionals [38]. Models of physical interest includes also the next
(cf. [3] and references therein):
• a model of combustion in porous media;
• a model of conduction of electrical impulses in nerve axons;
• a climate energy balance model;
etc. The main purpose of this subsection is to investigate regularity properties of all
globally defined weak solutions for Problem (2.1) and (2.2) with initial data u τ ∈ H
under listed above assumptions.
Further we need to consider the restriction of v : [τ, T ] → V ∗ on [s, T ], s ∈
(τ, T ), τ < T . To simplify conclusions denote it by the same symbol v(·).
Theorem 2.1 Let u(·) be an arbitrary weak solution of Problem (2.1) and (2.2)
on [τ, T ]. Then for any ε ∈ (0, T − τ ) u(·) ∈ C([τ + ε, T ]; V ) ∩ L 2 (τ + ε, T ;
H 2 (Ω) ∩ V ) and u t (·) ∈ L 2 (τ + ε, T ; H ).
Proof Let u(·) be an arbitrary weak solution of Problem (2.1) and (2.2) on [τ, T ].
Then there exists a measurable function d : Ω × (τ, T ) → R such that u(·) and d(·)
satisfy (2.4) and (2.5). As u(·) ∈ L 2 (Ω × (τ, T )) and the growth condition (2.3)
holds, then d(·) ∈ L 2 (Ω × (τ, T )). The set

D := {s ∈ (τ, T ) | u(s) ∈ V }

is dense in [τ, T ]. For any arbitrary fixed s ∈ D we note that u(·) is the unique weak
solution on [s, T ] of the problem
50 2 Regularity of Solutions for Nonlinear Systems
⎧
⎨ z t − z = −d(x, t) in Ω × (s, T ),
z  = 0, (2.6)
⎩ ∂Ω
z(x, s) = u(x, s) in Ω.

Moreover, u(·) ∈ L 2 (s, T ; H 2 (Ω) ∩ V ) ∩ C([s, T ]; V ) and u t (·) ∈ L 2 (s, T ; H ),

s ∈ D (cf. [40, Chap. 4.I], [42, Chap. III] and references therein). Thus for any ε ∈
(0, T − τ ) u(·) ∈ C([τ + ε, T ]; V ) ∩ L 2 (τ + ε, T ; H 2 (Ω) ∩ V ) and u t (·) ∈ L 2
(τ + ε, T ; H ).
The theorem is proved.

2.2 Artificial Control Method for Nonlinear Partial

Differential Equations and Inclusions: Regularity
of All Weak Solutions

Let (V ; H ; V ∗ ) be evolution triple , where V be a real Hilbert space, such that V ⊂ H

with compact imbedding. Let A : V → V ∗ be a linear symmetric operator such that
∃c > 0 : Av, vV ≥ cv2V , for each v ∈ V and let D(A) = {u ∈ V : Au ∈ H }. We
note that the mapping v → Av H defines the equivalent norm on D(A); Temam [42,
Chap. III]. Let B : R × V → 2 H \ {∅} be set-valued (in the general case) mapping
such that the following assumption holds: there exists c1 > 0 such that y H ≤
c1 (1 + uV ), for a.e. t and each u ∈ V and y ∈ B(t, u).
For a set D ⊂ H let coD be a closed convex hull of a set D. We consider the
differential-operator inclusion:

du
+ Au(t) + B(t, u(t))  0̄ (−∞ < τ < T < +∞). (2.7)
dt

The function u(·) ∈ L 2 (τ, T ; V ) is called a weak solution of Problem (2.7) on

[τ, T ], if there exists a Bochner-measurable function d : (τ, T ) → H such that

d(t) ∈ coB(t, u(t)) for a.e. t ∈ (τ, T ); and (2.8)

 T
− u, v ξ (t) + Au, v ξ(t) + d, v ξ(t) dt = 0, (2.9)
τ

for all ξ ∈ C0∞ (τ, T ) and for all v ∈ V , where ·, · denotes the pairing in the space V .
The main regularity result of this section has the following formulation.

Theorem 2.2 Let −∞ < τ < T < +∞ and u τ ∈ H . If u(·) is a weak solution of
Problem (2.7) on [τ, T ], then u(·) ∈ C([τ + ε, T ]; V ) ∩ L 2 (τ + ε, T ; D(A)) and
du
dt
(·) ∈ L 2 (τ + ε, T ; H ) for each ε ∈ (0, T − τ ).
2.2 Artificial Control Method for Nonlinear Partial Differential … 51

Proof Let u(·) be an arbitrary weak solution of Problem (2.7) on [τ, T ]. According
to the definition of a weak solution of Problem (2.7) on [τ, T ], there exist d ∈
L 2 (τ, T ; H ) such that u(·) ∈ L 2 (τ, T ; V ) and d(·) satisfy (2.8) and (2.9). Note that
the set
D := {s ∈ (τ, T ) | u(s) ∈ V }

is dense in [τ, T ]. For an arbitrary fixed s ∈ D we remark that u(·) is the unique
weak solution on [τ, T ] of the problem
dz
dt
+ Az(t) = −d(t) on (s, T ),
(2.10)
z(s) = u(s).

Therefore, u(·) ∈ L 2 (s, T ; D(A)) ∩ C([s, T ]; V ) and du

dt
(·) ∈ L 2 (s, T ; H ), s ∈ D
(cf. [40, Chap. 4.I], [42, Chap. III] and references therein). Thus u(·) ∈ C([τ +
ε, T ]; V ) ∩ L 2 (τ + ε, T ; D(A)) and du
dt
(·) ∈ L 2 (τ + ε, T ; H ) for any ε ∈ (0,
T − τ ).
The theorem is proved.

Remark 2.1 Theorem 2.2 implies that each weak solution of Problem (2.7) on [τ, T ]
is regular, that is, u(·) ∈ L 2 (ε, T ; D(A)) ∩ C([ε, T ]; V ) and du
dt
(·) ∈ L 2 (ε, T ; H ),
for each ε ∈ (0, T − τ ).

Let B(t, u) := ∂ J1 (u) − ∂ J2 (u) for each u ∈ V and t ∈ R, where Ji : H → R

be a convex, lower semi-continuous function such that the following assumptions
hold: (i) (growth condition) there exists c1 > 0 such that y H ≤ c1 (1 + u H ), for
each u ∈ H and y ∈ ∂ Ji (u) and i = 1, 2; (ii) (sign condition) there exist c2 > 0,
λ ∈ (0, c) such that (y1 − y2 , u) H ≥ −λu2H − c2 , for each yi ∈ ∂ Ji (u), u ∈ H ,
where ∂ Ji (u) the subdifferential of Ji (·) at a point u. Note that u ∗ ∈ ∂ Ji (u) if and
only if u ∗ (v − u) ≤ Ji (v) − Ji (u) ∀v ∈ H ; i = 1, 2. For such B Problem (2.7) has
the following formulation:

du
+ Au(t) + ∂ J1 (u(t)) − ∂ J2 (u(t))  0̄ (−∞ < τ < T < +∞). (2.11)
dt

We recall that the function u(·) ∈ L 2 (τ, T ; V ) is called a weak solution of Problem
(2.11) on [τ, T ], if there exist Bochner measurable functions di : (τ, T ) → H ; i =
1,2, such that
di (t) ∈ ∂ Ji (u(t)) for a.e. t ∈ (τ, T ), i=1,2; and (2.12)
 T
− u, v ξ (t) +  Au, v ξ(t) + d1 , v ξ(t) − d2 , v ξ(t) dt = 0, (2.13)
τ

for all ξ ∈ C0∞ (τ, T ) and for all v ∈ V .

The following theorem provides sufficient conditions for the existence and regu-
larity of all weak solutions for Problem (2.11).
52 2 Regularity of Solutions for Nonlinear Systems

Theorem 2.3 Let −∞ < τ < T < +∞ and u τ ∈ H . If u(·) is a weak solution of
Problem (2.11) on [τ, T ], then u(·) ∈ C([τ + ε, T ]; V ) ∩ L 2 (τ + ε, T ; D(A)) and
du
dt
(·) ∈ L 2 (τ + ε, T ; H ) for any ε ∈ (0, T − τ ).

Proof The regularity of each weak solution follows from Theorem 2.2.
The theorem is proved.

2.3 Regularity of All Weak Solutions for Nonlinear

Reaction-Diffusion Systems with Nonlinear Growth

In this section we establish sufficient conditions for regularity of weak solutions

for both reaction-diffusion equations (Sect. 2.3.1) as well as systems of reaction-
diffusion equations (Sect. 2.3.2) separately.

2.3.1 Reaction-Diffusion Equations

In a bounded domain Ω ⊂ R3 with sufficiently smooth boundary ∂Ω we consider

the problem ⎧
⎨ u t − Δu + f (u) = h, x ∈ Ω, t > 0,
u|∂Ω = 0, (2.14)
⎩
u (0) = u 0 ,

where
h ∈ L 2 (Ω),
f ∈ C(R), (2.15)
| f (u)| ≤ C1 (1 + |u| p−1 ), ∀u ∈ R,

with 2 ≤ p ≤ 3, C1 , C2 , α > 0.
We denote by A the operator −Δ with Dirichlet boundary conditions, so that
D (A) = H 2 (Ω) ∩ H01 (Ω) . As usual, denote the eigenvalues and the eigenfunc-
tions of A by λi , ei , i = 1, 2 . . .
Denote F(u) = 0 f (s)ds. From (2.15) we have that lim inf f (u)
u
u
= ∞, and for
|u|→∞
some D1 0,
|F(u)| ≤ D1 (1 + |u| p ), ∀u ∈ R. (2.16)

In what follows we denote H = L 2 (Ω), V = H01 (Ω), and  · , (·, ·) will be the
norm and the scalar product in L 2 (Ω). We denote by  ·  X the norm in the abstract
Banach space X , whereas (·, ·)Y will be the scalar product in the abstract Hilbert
space Y . Also, P (X ) will be the set of all non-empty subsets of X.
On the other hand, we define the usual sequence of spaces
2.3 Regularity of All Weak Solutions for Nonlinear Reaction-Diffusion … 53

∞
V 2α = D (Aα ) = {u ∈ H : λi2α |(u, ei )|2 < ∞},
i=1

where α ≥ 0. We recall the following well known result, which is a particular case
of [40, Lemma 37.8] for our operator A = −Δ in a three-dimensional domain.

Lemma 2.1 D (Aα ) ⊂ W k,q (Ω) whenever q ≥ 2 and k is an integer such that

3 3
k− < 2α − .
q 2

Also, it is well known that V s ⊂ 

H s (Ω) for all s ≥ 0 (see [49, Chap. IV] or [34]).
p
A function u ∈ L loc
2
(0, +∞; V ) L loc (0, +∞; L p (Ω)) is called a weak solution
of (2.14) on (0, +∞) if for all T > 0 , v ∈ V, η ∈ C0∞ (0, T )

T T
− (u, v)ηt dt + ((u, v)V + ( f (u), v) − (h, v)) ηdt = 0.
0 0

It is well known [1, Theorem 2] or [9, p. 284] that for any u 0 ∈ there exists at
least one weak solution of (2.14) with u(0) = u 0 (and it may be non unique) and
that any weak solution of (2.14) belongs to C ([0, +∞); H ). Moreover, the function
t → u(t)2 is absolutely continuous and

1 d
u(t)2 + u(t)2V + ( f (u(t)), u(t)) − (h, u(t)) = 0 a.e. (2.17)
2 dt
 p
The function u ∈ L loc
2
(0, +∞; V ) L loc (0, +∞; L p (Ω)) is called a regular
solution of (2.14) on (0, +∞) if for all T > 0, v ∈ V, and η ∈ C0∞ (0, T ) we have

T T
− (u, v)ηt dt + ((u, v)V + ( f (u), v) − (h, v)) ηdt = 0, (2.18)
0 0

and

u ∈ L ∞ (ε, T ; V ) , (2.19)
u t ∈ L (ε, T ; H ) , ∀0 < ε < T.
2
(2.20)

Any regular solution u satisfies

u ∈ L 2 (ε, T ; D (A)) . (2.21)

In this section we will prove that every weak solution is in fact a regular solution.
54 2 Regularity of Solutions for Nonlinear Systems

Theorem 2.4 Assume that 2 ≤ p ≤ 3 in condition (2.15). Then any weak solution
u (·) satisfies u ∈ C ([ε, T ]; V ) ∩ L 2 (ε, T ; D (A)), u t ∈ L 2 (ε, T ; H ) for all ε > 0,
that is, it is a regular solution.

Proof From
 
p
| f (u (x, t))| p−1 d x ≤ C1 + C2 |u (x, t)| p d x
Ω Ω

we obtain that
2 p−2
 f (u (t))2 p ≤ C3 + C4 u (t) L p (Ω) .
L p−1 (Ω)

 
Using the Sobolev embedding H r (Ω) ⊂ L p (Ω) if r = 3
2
− 3
p
≤ 1
2
(as p ≤ 3)
and the Gagliardo–Nirenberg inequality

1 1
v H r (Ω) ≤ C5 v H 21 ≤ C6 v 2 v H2 1 (Ω) ,
(Ω)

we have
p−1
 f (u (t))2 p ≤ C3 + C7 u (t) p−1 u (t) H 1 (Ω)
L p−1 (Ω)
≤ C8 + C9 u (t)2 u (t)2H 1 (Ω) .

Thus,
 
 f (u)  p
 ≤ C10 1 + uC([0,T ];H ) u L 2 (0,T ;H 1 (Ω)) .
L 2 0,T ;L p−1 (Ω)

 p

Set d (x, t) = f (u (x, t)) for (x, t) ∈ (0, T ) × Ω. Then d ∈ L 2 0, T ; L p−1 (Ω)
   
⊂ L 2 0, T ; H −r (Ω) ⊂ L 2 0, T ; V −r .
We consider the problem
⎧
⎨ vt − Δv = −d (x, t) + h (x) , x ∈ Ω, t > 0,
v|∂Ω = 0,
⎩
v (τ ) = u (τ ) .

We note that u (τ ) ∈ V ⊂ V r for a.a. τ > 0. For such τ in view of [40, p. 163,
Theorem 42.12] there exists a unique weak solution v (·) such that  v ∈ C([τ, T ];
V r ) ∩ L 2 τ, T ; V r +1 . Hence, u ∈ C([ε, T ]; V r ) ∩ L 2 ε, T ; V r +1 for all ε > 0.
We shall prove that f (u (·)) ∈ L 2 (ε, T ; H ). As this is obvious if p = 2, we
consider that 2 < p ≤ 3. We  notethat V ⊂ H (Ω) ⊂ L (Ω). Also, by Lemma
r r p

2.1 with α = r +1 2
, r = 3 21 − 1p , k = 1 we obtain that V r +1 ⊂ W 1,q (Ω) for
any q < p. On the other hand, by the Sobolev embedding theorems we have
2.3 Regularity of All Weak Solutions for Nonlinear Reaction-Diffusion … 55

W 1,q (Ω) ⊂ L q (Ω), for q < 3− 3p

. Thus, the inequality p ( p − 1) < 3− 3p
, for all
p   p
2 ≤ p ≤ 3, implies that u ∈ C([ε, T ]; L (Ω)) ∩ L ε, T ; L
p 2 p( p−1)
(Ω) . By (2.15)
we have
 
 f (u (t))2 = | f (u (x, t))|2 d x ≤ C11 + C12 |u (x, t)|2( p−1) d x
Ω Ω
p−1 p−1
≤ C13 + C14 u (t) L p (Ω) u (t) L p( p−1) (Ω) .

Therefore, f (u (·)) ∈ L 2 (ε, T ; H ). Then standard results imply that u ∈ C([ε, T ];

V ) ∩ L 2 (ε, T ; D (A)) and u t ∈ L 2 (ε, T ; H ).
The lemma is proved.

Remark 2.2 Theorem 2.4 was proved in [23].

2.3.2 Systems of Reaction-Diffusion Equations

Let us consider the following reaction-diffusion system (RD-system for short)

u t = aΔu − f (u) + h(x), x ∈ Ω, t > 0,

(2.22)
u|∂Ω = 0,

where u = u(x, t) = (u 1 (x, t), . . . , u N (x, t)) is unknown vector-function, a is a

real N × N matrix with positive symmetric part 21 (a + a ∗ ) ≥ β I , β > 0, h =
(h 1 , . . . , h N ), f = ( f 1 , . . . , f N ) are given functions,

h ∈ (L 2 (Ω)) N , f ∈ C(R N ; R N ),

and for given numbers C1 , C2 ≥ 0, γ > 0, pi ≥ 2, i = 1, N the following condi-

tions hold:
N N
∀v∈R N
| f i (v)| ≤ C1 (1 +
qi
|vi | pi ), (2.23)
i=1 i=1

N N
∀v∈R N
f i (v)v ≥ γ
i
|vi | pi − C2 , (2.24)
i=1 i=1

where p1i + q1i = 1, i = 1, N . In further arguments we will use the standard func-
tional spaces
 N
H = (L (Ω)) with the norm |v| =
2 N 2
|vi (x)|2 d x,
Ω i=1
56 2 Regularity of Solutions for Nonlinear Systems

 N
V = (H01 (Ω)) N with the norm v2 = |∇vi (x)|2 d x.
Ω i=1

Let us denote V = H −1 (Ω)) N , p = ( p1 , . . . , p N ), L p (Ω) = L p1 (Ω) × · · ·

× L p N (Ω),
p
W = L loc (0, +∞; L p (Ω)) ∩ L loc
2
(0, +∞; V ).

Definition 2.1 The function u = u(x, t) ∈ W is called a (global) weak solution of

Problem (2.22) on (0, +∞) if for all T > 0, v ∈ V ∩ L p (Ω),
   
d
u(x, t)v(x)d x + a∇u(x, t)∇v(x) + f (u(x, t))v(x) − h(x)v(x) d x = 0
dt
Ω Ω
(2.25)
in the sense of scalar distributions on (0, T ).

From (2.23) and Sobolev embedding theorem we see that every solution of (2.22)
satisfies u t ∈ L loc (0, +∞; H −r (Ω)), where r = (r1 , . . . , r N ), rk = max{1, n( 12 −
q

1
pk
)}. The well-known result on global resolvability of (2.22) for initial conditions
from the phase space H established in [9]. Under conditions (2.23), (2.24) for every
u 0 ∈ H there exists at least one weak solution of (2.22) on (0, +∞) with u(0) = u 0 .
Every weak solution of (2.22) belongs to C ([0, +∞); H ), the function t → |u(t)|2
is absolutely continuous and for a.a. t ≥ 0 the following energy equality holds

1 d
|u(t)|2 + (a∇u(t), ∇u(t)) + ( f (u(t)), u(t)) = (h, u(t)). (2.26)
2 dt

The function u = u(x, t) ∈ W is called a regular solution of Problem (2.22) on

(0, +∞) if it is weak solution on (0, +∞) and, additionally,
 
u ∈ L ∞ ε, T ; V ∩ L p (Ω) , (2.27)
u t ∈ L (ε, T ; H ) ∀ 0 < ε < T.
2
(2.28)

Let us consider the following additional condition on vector-function f [51]:

∀ v ∈ R N f (v) = ∇ F(v) + g(v), (2.29)

where ∇ F satisfies (2.23), (2.24), and g ∈ C(R N ; R N ) is such that for some constants
C3 ≥ 0, C4 ≥ 0,

|g(v)|2 ≤ C3 F(v) + C4 (|v|2 + 1), ∀ v ∈ R N . (2.30)

v
If N = 1 (scalar case), then (2.29), (2.30) hold with F(v) = f (s)ds, g ≡ 0.
0
Conditions (2.29), (2.30) also take place if
2.3 Regularity of All Weak Solutions for Nonlinear Reaction-Diffusion … 57

f i (v) = αi vi |vi | pi −2 + gi (v), i = 1, N ,

where αi > 0, g ∈ C(R N ; R N ), and |g(v)| ≤ C4 (1 + |v|). Another example is the

FitzHugh–Nagumo system (see the example in Sect. 3.4.4 below).
Let us briefly analyze conditions (2.29), (2.30).
Using the equality
1
1 
(|v|+1)2 1
F(v) − F(0) = ∇ F(sv) · vds = ∇ F(sv) · vds + ∇ F(sv) · vds
0 0 1
(|v|+1)2

and condition (2.24), we deduce that for some α > 0

N
∀ v ∈ R N F(v) ≥ α |vi | pi − C5 . (2.31)
i=1

1
Again using the equality F(v) − F(0) = ∇ F(sv)vds, Young’s inequality and con-
0
dition (2.23), we obtain
N
|F(v)| ≤ C6 ( |vi | pi + 1). (2.32)
i=1

Theorem 2.5 Under conditions (2.23), (2.24), (2.29), (2.30) for every u 0 ∈ H there
exists at least one regular solution u(·) of (2.22) such that u (0) = u 0 , and for some
positive constants C(g), D(g), which depend on the function g but not on u(·), the
following energy inequality holds for a.e. s > 0 and each t ≥ s

 t t
E (u (t)) + |u r | dr ≤ E (u (s)) + C(g)
2
E(u( p))dp + D(g)(t − s),
s
s
(2.33)
where E (u (t)) = u (t)2 + 2 (F (u (t)) , 1) − 2 (h, u (t)) . Moreover, C(g) =
D(g) = 0 if in condition (2.29) we have g ≡ 0.

Proof We take as in [9, p.281] the Galerkin approximations using the basis of eigen-
functions {w j (x) , j ∈ N}, of the Laplace operator with Dirichlet boundary condi-
tions. Let X n = {w1 , . . . , wn } and let Pn be the orthogonal projector from H onto X n .
Then u n (x, t) = nj=i a j,m (t) w j (x) will be a solution of the system of ordinary
differential equations

du n  
= Pn Δu n − Pn f u n + Pn h, u n (0) = Pn u 0 . (2.34)
dt
58 2 Regularity of Solutions for Nonlinear Systems

It is proved in [9, p.281] that (2.34) is globally resolved, and for every T > 0 passing
to a subsequence u n converges to a weak solution u of (2.22) in C([0, T ]; H ),
weakly  in L (0,
p
T ; Lp (Ω)) and weakly in L 2 (0, T ; V ). Also, u nt → u t weakly
−r
in L 0, T ; H (Ω) .
q

Multiplying the equation in (2.34) by u nt we get

d  n 2 
u  + 2(F(u n ), 1) − 2(h, u n ) + 2|u nt |2 = −2(g(u n ), u nt ). (2.35)
dt
Using (2.30), we deduce from (2.35) that

d  n 2 
u  + 2(F(u n ), 1) − 2(h, u n ) + |u nt |2
dt
 
≤ C7 (g) u n 2 + 2(F(u n ), 1) − 2(h, u n ) + C8 (g). (2.36)

In particular, u n satisfies (2.33) ∀ t ≥ s ≥ 0. We note that if g ≡ 0, then C7 (g) =

C8 (g) = 0, so that C(g) = D(g) = 0 holds.
On the other hand, multiplying (2.22) by u n and using (2.23) in a standard way
we obtain

d n2  2 N
 n  pi
|u | + λ1 |u n |2 + u n  + γ u  p ≤ K + |h|2 .
i L i (2.37)
dt i=1

By Gronwall’s lemma we obtain

1  
|u n (t)|2 ≤ e−λ1 t |u n0 |2 + K + |h|2 . (2.38)
λ1

Thus integrating (2.37) over (t, t + r ) with r > 0 we have

 
t+r  n 2 t+r N
 n  pi
|u n (t + r ) |2 + u  ds + γ u (s) p ds (2.39)
i L i
t t i=1
 
≤ |u n (t) |2 + r K + |h|2
 
1  
≤ e−λ1 t |u n0 |2 + + r K + |h|2 .
λ1

Then from (2.32),

2.3 Regularity of All Weak Solutions for Nonlinear Reaction-Diffusion … 59
 t+r 
 n 2      
u  + 2 F u n (s) , 1 − 2 h, u n ds
t
 t+r  t+r N  t+r
 n 2  n  pi
≤  u  ds + 2C6   2
u i (s) L pi ds + r |h| + |u n |2 ds + 2C6 |Ω|r
t t i=1 t

≤ C9 (e−λ1 t |u n0 |2 + r + 1). (2.40)

Now we can apply uniform Gronwall Lemma [46] to inequality (2.36) and obtain
 n       
u (t + r )2 + 2 F u n (t + r ) , 1 − 2 h, u n (t + r ) (2.41)
−λ1 t
e |u n0 |2 +1
≤ C10 ( + 1)er for all 0 ≤ t ≤ t + r.
r
From the last inequality and (2.31) we have

 n  N
 n 
u (t + r )2 + u (t + r ) pip (2.42)
i L i
i=1
 
e−λ1 t |u n0 |2 + 1
≤ C11 ( + 1)er + 1 for all 0 ≤ t ≤ t + r.
r

Therefore, the sequence u n (·) is bounded in L ∞ (r, T ; V ∩ L p (Ω)) for all

0 < r < T.
Integrating (2.36) over (r, T ), we have

 T 
T  n 2      
|u nt |2 dt ≤ C7 u (s) + 2 F u n (s) , 1 − 2 h, u n (s) ds (2.43)
r
r
 2      
+ u n (r ) + 2 F u n (r ) , 1 − 2 h, u n (r ) + C8 (T − r ) + 2C5 |Ω| + |h|2 + |u n (T )|2 .

So from (2.38), (2.40), (2.41) and the last inequality we deduce that u nt is bounded
in L 2 (r, T ; H ) for all 0 < r < T .
Thus for the limit function u we can claim that it is regular solution of (2.22) and
u(0) = u 0 .
Let us prove that u satisfies the energy inequality (2.33). As u n is bounded in
L ∞ (r, T ; L p (Ω)), so f (u n ) is bounded in L ∞ (r, T ; L q (Ω)). Therefore from [45]
up to subsequence

u n → u in L 2 (r, T ; V ) ∩ L p (r, T ; L p (Ω)). (2.44)

and, in particular,
u n (t) → u(t) in V for a.a. t ∈ (r, T ).
60 2 Regularity of Solutions for Nonlinear Systems

Also, it is standard to check that u n → u in C([r, T ], H ), for all 0 < r < T, and
that u n (t) → u(t) weakly in V for all 0 < t ≤ T .
Then by the dominated convergence theorem F(u n (t)) → F(u(t)) in L 1 (Ω) for
a.a. t ∈ [r, T ]. Also, for any 0 < t ≤ T we have F(u n (x, t)) → F(u(x, t)) for a.a.
x. Then F(u n (x, t)) ≥ −C5 and Fatou’s lemma imply
 
F(u(x, t))d x ≤ lim inf F(u n (x, t))d x
Ω Ω

and
E(u(t)) ≤ lim inf E(u n (t)).

Hence, we can pass to the limit in (2.33) and obtain the required result.
The theorem is proved.
Remark 2.3 Theorem 2.5 yields only existence but not regularity of each weak solu-
tion of Problem (2.22). This theorem was proved in [24].

2.4 Examples of Applications

In this section we provide examples of applications to theorems established in

Sects. 2.1–2.3. We consider a parabolic feedback control problem (Sect. 2.4.1), a
model of conduction of electrical impulses in nerve axons (Sect. 2.4.2), a climate
energy balance model (Sect. 2.4.3); FitzHugh–Nagumo system (Sect. 2.4.4); and a
model of combustion in porous media (Sect. 2.4.5).

2.4.1 A Parabolic Feedback Control Problem

Let Ω be an open and bounded subset of R3 . Let us consider the following nonsta-
tionary heat conduction equation

∂y
− y = f in Ω × R (2.45)
∂t
with initial condition and Dirichlet homogeneous boundary condition. Here y =
y(x, t) represents the temperature at the point x ∈ Ω and time t > 0.
Let j : Ω × R → R be a locally Lipschitz function in ξ (cf. [37] and references
therein) and ∂ j (x, ξ ) denotes generalized gradient of Clarke (see [12]) with respect to
the last variable. Note that the multi-valued function ∂ j (x, ·) : R → 2R is generally
nonmonotone and it includes the vertical jumps.
We assume that f = f 1 + f 2 , where f 2 = f 2 (x) is given and f 1 is a known
function of the temperature of the form
2.4 Examples of Applications 61

− f 1 (x, t) ∈ ∂ j (x, y(x, t)) a.e. (x, t) ∈ Ω × R. (2.46)

In a physicist’s language it means that the law is characterized by the generalized

gradient of a nonsmooth potential j (cf. [39]).
Assume also that ∂ j satisfies the growth condition

∃c0 > 0 : | p| ≤ c0 (1 + |u|) for a.e. x ∈ Ω, and each u ∈ R, and d ∈ ∂ j (x, u);

and the sign condition

inf d∈∂ j (x,u) d supd∈∂ j (x,u) d

lim > −λ1 ; lim > −λ1 ,
u→+∞ u u→−∞ u

where λ1 is the first eigenvalue of − in H01 (Ω). According to Theorem 2.2, for any
−∞ < τ < T < +∞ each weak solution u τ ∈ L 2 (Ω) of Problem (2.45) and (2.46)
on [τ, T ] belongs to C([τ + ε, T ]; H01 (Ω)) ∩ L 2 (τ + ε, T ; H 2 (Ω) ∩ H01 (Ω)) and
du
dt
(·) ∈ L 2 (τ + ε, T ; L 2 (Ω)) for each ε ∈ (0, T − τ ).

2.4.2 A Model of Conduction of Electrical Impulses

in Nerve Axons

Consider the problem:

∂u
− ∂∂ xu2 + u ∈ λH (u − a), (x, t) ∈ (0, π ) × R,
2

∂t (2.47)
u(0, t) = u(π, t) = 0, t ∈ R,
 
where a ∈ 0, 21 ; Terman [47, 48]. Since Problem (2.47) is a particular case of
Problem (2.1) and (2.2), then for each −∞ < τ < T < +∞ and a weak solution
u τ ∈ L 2 ((0, π )) of Problem (2.47) on [τ, T ] belongs to C([τ + ε, T ]; H01 ((0, π ))) ∩
L 2 (τ + ε, T ; H 2 ((0, π )) ∩ H01 ((0, π ))) and du
dt
(·) ∈ L 2 (τ + ε, T ; L 2 ((0, π ))) for
each ε ∈ (0, T − τ ); Figs. 2.4, 2.5, 2.6, and 2.7.

2.4.3 Climate Energy Balance Model

Let (M , g) be a C ∞ compact connected oriented two-dimensional Riemannian mani-

fold without boundary (as, e.g. M = S 2 the unit sphere of R3 ). Consider the problem:

∂u
∂t
− Δu + Re (x, u) ∈ QS(x)β(u), (x, t) ∈ M × R, (2.48)
62 2 Regularity of Solutions for Nonlinear Systems

Fig. 2.4 Graphics of solutions of problem (2.47) with a = 0.49, λ = 2, n = 10, h = 0, 001, N =
100 in a moment a t = 0; b t = 0.8; c t = 1.6; d t = 2.4; e t = 3.2; f t = 4

Fig. 2.5 Screenlist of animation for dynamics of solutions of problem (2.47) in 2D

where Δu = divM (∇M u) ; ∇M is understood in the sense of the Riemannian metric

g. Note that (2.48) is the so-called climate energy balance model. It was proposed in
Budyko [8] and Sellers [41] and examined also in Díaz et al. [13–15]. The unknown
u(x, t) represents the average temperature of the Earth’s surface. In Budyko [8] the
energy balance is expressed as

heat variation = Ra − Re + D.
2.4 Examples of Applications 63

Fig. 2.6 Screenlist of animation for dynamics of solutions of problem (2.47) in 3D

Fig. 2.7 Screenlist of animation for dynamics of solutions of problem (2.47) in section
64 2 Regularity of Solutions for Nonlinear Systems

Here Ra = QS(x)β(u). It represents the solar energy absorbed by the Earth, Q > 0 is
a solar constant, S(x) is an insolation function, given the distribution of solar radiation
falling on upper atmosphere, β represents the ratio between absorbed and incident
solar energy at the point x of the Earth’s surface (so-called co-albedo function). The
term Re represents the energy emitted by the Earth into space, as usual, it is assumed
to be an increasing function on u. The term D is heat diffusion, we assume (for
simplicity) that it is constant.
As usual, the term Re may be chosen according to the Newton cooling law as
linear function on u, Re = Bu + C (here B, C are some positive constants) [8], or
according to the Stefan-Boltzman law, Re = σ u 4 [41]. In this subsection we consider
Re = Bu as in Budyko [8].
Let S : M → R be a function such that S ∈ L ∞ (M ) and there exist S0 , S1 > 0
such that
0 < S0 ≤ S(x) ≤ S1 .

Suppose also that β is a bounded maximal monotone graph of R2 , that is there exist
m, M ∈ R such that for all s ∈ R and z ∈ β(s)

m ≤ z ≤ M.

Let us consider real Hilbert spaces

H := L 2 (M ), V := {u ∈ L 2 (M ) : ∇M u ∈ L 2 (T M )}

with respective standard norms  ·  H ,  · V , and inner products ( · , · ) H , ( · , · )V ,

where T M represents the tangent bundle and the functional spaces L 2 (M ) and
L 2 (T M ) are defined in a standard way; see, for example, Aubin [2]. According
to Theorem 2.2, for any −∞ < τ < T < +∞ each weak solution u τ ∈ L 2 (Ω) of
Problem (2.48) on [τ, T ] belongs to C([τ + ε, T ]; H01 (Ω)) ∩ L 2 (τ + ε, T ; H 2 (Ω) ∩
H01 ((0, π ))) and du
dt
(·) ∈ L 2 (τ + ε, T ; L 2 (Ω)) for each ε ∈ (0, T − τ ).

2.4.4 FitzHugh–Nagumo System

Let us consider generalized FitzHugh–Nagumo system [46]:

u t = d1 Δu − f 1 (u) − v, (2.49)

vt = d2 Δv + δu − γ v, (2.50)

u|∂Ω = v|∂Ω = 0, (2.51)

where Ω = (0, L), d1 , d2 , δ, γ are positive constants, f 1 ∈ C(R),

2.4 Examples of Applications 65

Fig. 2.8 Trajectories of

FitzHugh–Nagumo system

| f 1 (u)| ≤ C1 (1 + |u|3 ); f 1 (u)u ≥ α|u|4 − C2 . (2.52)

For the vector-function  

f 1 (u) + v
f (u, v) =
−δu + γ v

conditions (2.23), (2.24) hold with p1 = 4, p2 = 2.Moreover,

 f = ∇ F + g, where
u
γ 2 v
F = F(u, v) = f 1 (s)ds + 2 v , g = g(u, v) = and conditions (2.29),
0
−δu
(2.30) also hold. Then all statements of Theorem 2.5 hold; Fig. 2.8.

2.4.5 A Model of Combustion in Porous Media

Let us consider the following problem:

∂u
− ∂∂ xu2 − f (u) ∈ λH (u − 1), (x, t) ∈ (0, π ) × R,
2

∂t (2.53)
u(0, t) = u(π, t) = 0, t ∈ R,

where f : R → R is a continuous and nondecreasing function satisfying growth and

sign assumptions, λ > 0, and H (0) = [0, 1], H (s) = I{s > 0}, s = 0; Feireisl and
Norbury [17]. Since Problem (2.53) is a particular case of Problem (2.1) and (2.2),
then for any −∞ < τ < T < +∞ each weak solution u τ ∈ L 2 ((0, π )) of Problem
(2.53) on [τ, T ] belongs to C([τ + ε, T ]; H01 ((0, π ))) ∩ L 2 (τ + ε, T ; H 2 ((0, π )) ∩
H01 ((0, π ))) and du
dt
(·) ∈ L 2 (τ + ε, T ; L 2 ((0, π ))) for each ε ∈ (0, T − τ ); Fig. 2.9.
66 2 Regularity of Solutions for Nonlinear Systems

Fig. 2.9 Graphics of solutions with f (u) = u, λ = 2, ε = 0.1, M = 100 in moment a t = 0;

b t = 0.8; c t = 1.6; d t = 2.4; e t = 3.2; f t = 4

References

1. Anguiano M., Caraballo T., Real J., Valero J.: Pullback attractors for reaction-diffusion equa-
tions in some unbounded domains with an H −1 -valued non-autonomous forcing term and
without uniqueness of solutions. Discret. Contin. Dyn. Syst. Ser. B, 14, 307–326 (2010)
2. Aubin, T.: Nonlinear Analysis on Manifolds. Monge-Ampére Equations. Springer, Berlin
(1980)
3. Balibrea, F., Caraballo, T., Kloeden, P.E., Valero, J.: Recent developments in dynamical sys-
tems: three perspectives. Int. J. Bifurc. Chaos (2010). doi:10.1142/S0218127410027246
4. Ball, J.M.: Continuity properties and global attractors of generalized semiflows and the Navier-
Stokes equations. Nonlinear Sci. 7, 475–502 (1997). (Erratum, ibid 8:233,1998. Corrected
version appears in ‘Mechanics: from Theory to Computation’. pp. 447–474. Springer, Berlin
(2000))
5. Ball, J.M.: Global attractors for damped semilinear wave equations. DCDS 10, 31–52 (2004)
6. Babin, A.V., Vishik, M.I.: Attractors of Evolution Equations. Nauka, Moscow (1989)
7. Barbu, V.: Nonlinear Semigroups and Differential Equations in Banach Spaces. Editura Acad-
emiei, Bucuresti (1976)
8. Budyko, M.I.: The effects of solar radiation variations on the climate of the Earth. Tellus 21,
611–619 (1969)
9. Chepyzhov, V.V., Vishik, M.I.: Attractors for Equations of Mathematical Physics. American
Mathematical Society, Providence, Rhode Island (2002)
10. Chepyzhov, V.V., Vishik, M.I.: Evolution equations and their trajectory attractors. J. Math. Pure
Appl. (1997). doi:10.1016/S0021-7824(97)89978-3
11. Chepyzhov, V.V., Vishik, M.I.: Trajectory and global attractors of three-dimensional Navier-
Stokes systems. Math. Notes (2002). doi:10.1023/A:1014190629738
12. Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley, Inc., New York (1983)
13. H. Díaz, J. Díaz, On a stochastic parabolic PDE arising in climatology. Rev. R. Acad. Cien.
Ser. A Mat. 96, 123–128 (2002)
14. J. Díaz, J. Hernández, L. Tello, On the multiplicity of equilibrium solutions to a nonlinear
diffusion equation on a manifold arising in climatology. J. Math. Anal. Appl. 216, 593–613
(1997)
References 67

15. Díaz, J., Hernández, J., Tello, L.: Some results about multiplicity and bifurcation of stationary
solutions of a reaction diffusion climatological model. Rev. R. Acad. Cien. Ser. A. Mat. 96(3),
357–366 (2002)
16. Díaz, J., Tello, L.: Infinitely many stationary solutions for a simple climate model via a shooting
method. Math. Methods Appl. Sci. 25, 327–334 (2002)
17. Feireisl, E., Norbury, J.: Some existence and nonuniqueness theorems for solutions of parabolic
equations with discontinuous nonlinearities. Proc. R. Soc. Edinburgh A. 119(1–2), 1–17 (1991)
18. Gajewski, H., Gröger, K., Zacharias, K.: Nichtlineare operatorgleichungen und operatordiffer-
entialgleichungen. Akademie-Verlag, Berlin (1974)
19. Kalita, P.: Convergence of Rothe scheme for hemivariational inequalities of parabolic type. Int.
J. Numer. Anal. Model. 10, 445–465 (2013)
20. Kalita, P.: Regularity and Rothe method error estimates for parabolic hemivariational inequality.
J. Math. Anal. Appl. 389, 618–631 (2012)
21. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Pullback attractors for a class of extremal solutions
of the 3D Navier-Stokes equations. J. Math. Anal. Appl. 373(2), 535–547 (2011). doi:10.1016/
j.jmaa.2010.07.040
22. Kapustyan, O.V., Valero, J.: Comparison between trajectory and global attractors for evo-
lution systems without uniqueness of solutions. Int. J. Bifurc. Chaos (2010). doi:10.1142/
S0218127410027313
23. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Structure of the global attractor for weak solutions
of reaction-diffusion equation. Appl. Math. Inf. Sci. 9(5), 2257–2264 (2015). doi:10.12785/
amis/090506
24. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Regular solutions and global attractors for reaction-
diffusion systems without uniqueness. Commun. Pure Appl. Anal. 13(5), 1891–1906 (2014).
doi:10.3934/cpaa.2014.13.1891
25. Kasyanov, P.O.: Multi-valued dynamics of solutions of autonomous operator differential equa-
tions with pseudomonotone nonlinearity. Math. Notes 92(1–2), 205–218 (2012). doi:10.1134/
S0001434612070231
26. Kasyanov, P.O.: Multi-valued dynamics of solutions of an autonomous differential-operator
inclusion with pseudomonotone nonlinearity. Cybern. Syst. Anal. (2011). doi:10.1007/s10559-
011-9359-6
27. Kas’yanov, P.O., Mel’nyk, V.S.: On properties of subdifferential mappings in Frechet spaces.
Ukrainian Math. J. 57(10), 1621–1634 (2005)
28. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Long-time behaviour of solutions
for autonomous evolution hemivariational inequality with multidimensional “Reaction-
Displacement” law. Abstr. Appl. Anal. 2012(450984), 21 (2012) doi:10.1155/2012/450984
29. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Regularity of weak solutions and their attrac-
tors for a parabolic feedback control problem. Set-Valued Var. Anal. 21, 271–282 (2013).
doi:10.1007/s11228-013-0233-8
30. Ladyzhenskaya, O.A.: Dynamical system, generated by Navier-Stokes equations. Zap Nauch
Sem LOMI. 27, 91–115 (1972)
31. Ladyzhenskaya, O.A.: Some comments to my papers on the theory of attractors for abstract
semigroups. Zap Nauchn Sem LOMI. 188, 102–112 (1990)
32. Ladyzhenskaya, O.A.: Attractors for Semigroups and Evolution Equations. University Press,
Cambridge (1991)
33. Lions, J.L.: Quelques méthodes de résolution des problémes aux limites non linéaires. Dunod,
Paris (1969)
34. Lions, J.L., Magenes, E.: Problèmes aux limites non-homogènes et applications. Dunod, Paris
(1968)
35. Melnik, V.S., Valero, J.: On attractors of multi-valued semiflows and differential inclusions.
Set Valued Anal. (1998). doi:10.1023/A:1008608431399
36. Migórski, S.: On the existence of solutions for parabolic hemivariational inequalities. J. Com-
put. Appl. Math. 129, 77–87 (2001)
68 2 Regularity of Solutions for Nonlinear Systems

37. Migórski, S., Ochal, A.: Optimal Control of Parabolic Hemivariational Inequalities. J. Glob.
Optim. 17, 285–300 (2000)
38. Otani, M., Fujita, H.: On existence of strong solutions for du dt (t) + ∂ϕ (u(t)) − ∂ϕ (u(t)) 
1 2

f (t). J. Fac. Sci. Univ. Tokyo. Sect. 1 A, Math. 24(3), 575–605 (1977)
39. Panagiotopoulos, P.D.: Inequality Problems in Mechanics and Applications. Convex and Non-
convex Energy Functions. Birkhauser, Basel (1985)
40. Sell, G.R., You, Yu.: Dynamics of Evolutionary Equations. Springer, New York (2002)
41. Sellers, W.D.: A global climatic model based on the energy balance of the Earth-atmosphere
system. J. Appl. Meteorol. 8, 392–400 (1969)
42. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Springer,
New York (1988)
43. Valero, J.: Attractors of parabolic equations without uniqueness. J. Dyn. Diff. Equ. (2001).
doi:10.1023/A:1016642525800
44. Valero, J., Kapustyan, A.V.: On the connectedness and asymptotic behaviour of solutions of
reaction-diffusion systems. J. Math. Anal. Appl. (2006). doi:10.1016/j.jmaa.2005.10.042
45. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong trajectory attractor of dissipative reaction-
diffusion system. Doklady RAN 435(2), 155–159 (2010)
46. Temam, R.: InfInite-dimensional Dynamical Systems in Mechanics and Physics. Springer,
New York (1997)
47. Terman, D.: A free boundary problem arising from a bistable reactiondiffusion equation. Siam
J. Math. Anal. 14, 1107–1129 (1983)
48. Terman, D.: A free boundary arising from a model for nerve conduction. J. Diff. Equ. 58(3),
345–363 (1985)
49. Vishik, M.I., Fursikov, A.V.: “Mathematical Problems of Statistical Hydromechanics. Kluwer
Academic Publishers, Dordrecht (1988)
50. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong Trajectory Attractor for Dissipative Reaction-
Diffusion System. Doclady Math. (2010). doi:10.1134/S1064562410060086
51. Zelik S.: The attractor for a nonlinear reaction-diffusion system with a supercritical nonlinearity
and it’s dimension. Rend. Accad. Naz. Sci. XL Mem. Mat. Appl. 24, 1–25 (2000)
52. Zgurovsky, M.Z., Kasyanov, P.O., Valero, J.: Noncoercive evolution inclusions for Sk type
operators. Int. J. Bifurc. Chaos (2010). doi:10.1142/S0218127410027386
53. Zgurovsky, M.Z., Kasyanov, P.O., Zadoianchuk, N.V.: Long-time behavior of solutions for qua-
silinear hyperbolic hemivariational inequalities with application to piezoelectricity problem.
Appl. Math. Lett. 25(10), 1569–1574 (2012). doi:10.1016/j.aml.2012.01.016
54. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation Inequali-
ties for Earth Data Processing I. Springer, Berlin (2010). doi:10.1007/978-3-642-13837-9
55. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation
Inequalities for Earth Data Processing II. Springer, Berlin (2010). doi:10.1007/978-3-642-
13878-2
56. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
Inclusions and Variation Inequalities for Earth Data Processing III. Springer, Berlin (2012).
doi:10.1007/978-3-642-28512-7
Chapter 3
Advances in the 3D Navier-Stokes Equations

Abstract In this chapter we provide a criterion for the existence of global strong
solutions for the 3D Navier-Stokes system for any regular initial data. Moreover,
we establish sufficient conditions for Leray-Hopf property of a weak solution for
the 3D Navier-Stokes system. Under such conditions this weak solution is rightly
continuous in the standard phase space H endowed with the strong convergence
topology.

3.1 Weak, Leray-Hopf and Strong Solutions

Let Ω ⊂ R3 be a bounded domain with rather smooth boundary Γ = ∂Ω, and [τ, T ]
be a fixed time interval with −∞ < τ < T < +∞. We consider 3D Navier-Stokes
system in Ω × [τ, T ]
⎧
⎨ ∂y
− νy + (y · ∇)y = −∇ p + f, div y = 0,
⎩ y∂t = 0,  (3.1)
Γ
y t=τ = yτ ,

where y(x, t) means the unknown velocity, p(x, t) is the unknown pressure, f (x, t)
is the given exterior force, and yτ (x) is the given initial velocity with t ∈ [τ, T ],
x ∈ Ω, ν > 0 means the viscosity constant; see also Figs. 3.1 and 3.2.
Throughout this note we consider generalized setting of Problem (3.1). For this
purpose define the usual function spaces

V = {u ∈ (C0∞ (Ω))3 : div u = 0}, Vσ = cl(H0σ (Ω))3 V , σ ≥ 0,

where cl X denotes the closure in the space X . Set H := V0 , V := V1 . It is well known

that each Vσ , σ > 0, is a separable Hilbert space and identifying H and its dual H ∗
we have Vσ ⊂ H ⊂ Vσ∗ with dense and compact embedding for each σ > 0. We
denote by (·, ·), · and ((·, ·)), · V the inner product and norm in H and V ,

© Springer International Publishing AG 2018 69

M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_3
70 3 Advances in the 3D Navier-Stokes Equations

Fig. 3.1 Alternating turbulence

Fig. 3.2 Poincare

intersections of vortices in
groove in perturbed flow

respectively; ·, · will denote pairing between V and V ∗ that coincides on H × V

with the inner product (·, ·). Let Hw be the space H endowed with the weak topology.
For u, v, w ∈ V we put
 
3
∂v j
b(u, v, w) = ui w j d x.
Ω i, j=1 ∂ xi

It is known that b is a trilinear continuous form on V and b(u, v, v) = 0, if u, v ∈ V .

Furthermore, there exists a positive constant C such that

|b(u, v, w)| ≤ C u V v V w V, (3.2)

3.1 Weak, Leray-Hopf and Strong Solutions 71

for each u, v, w ∈ V ; see, for example, Sohr [18, Lemma V.1.2.1] and references
therein.
Let f ∈ L 2 (τ, T ; V ∗ ) + L 1 (τ, T ; H ) and yτ ∈ H . Recall that the function y ∈
dy
L 2 (τ, T ; V ) with ∈ L 1 (τ, T ; V ∗ ) is a weak solution of Problem (3.1) on [τ, T ],
dt
if for all v ∈ V
d
(y, v) + ν((y, v)) + b(y, y, v) = f, v (3.3)
dt
in the sense of distributions, and

y(τ ) = yτ . (3.4)

The weak solution y of Problem (3.1) on [τ, T ] is called a Leray-Hopf solution of

Problem (3.1) on [τ, T ], if y satisfies the energy inequality:

Vτ (y(t)) ≤ Vτ (y(s)) for all t ∈ [s, T ], a.e. s > τ and s = τ, (3.5)

where
ς ς
1
Vτ (y(ς )) := y(ς ) 2
+ν y(ξ ) 2
V dξ − f (ξ ), y(ξ ) dξ, ς ∈ [τ, T ].
2
τ τ
(3.6)

For each f ∈ L 2 (τ, T ; V ∗ ) + L 1 (τ, T ; H ) and yτ ∈ H there exists at least one

Leray-Hopf solution of Problem (3.1); see, for example, Temam [19, Chapter III]
dy
∈ L 3 (τ, T ; V ∗ ) +
4
and references therein. Moreover, y ∈ C([τ, T ], Hw ) and
dt
dy
L 1 (τ, T ; H ). If f ∈ L 2 (τ, T ; V ∗ ), then, additionally, ∈ L 3 (τ, T ; V ∗ ). In par-
4

dt
ticular, initial condition (3.4) makes sense.
Let A : V → V ∗ be the linear operator associated to the bilinear form ((u, v)) =
Au, v . Then A is an isomorphism from D(A) onto H with D(A) = (H 2 (Ω))3 ∩ V.
We recall that the embedding D(A) ⊂ V is dense and continuous. Moreover, we
assume Au H as the norm on D(A), which is equivalent to the one induced by
(H 2 (Ω))3 . Problem (3.3) can be rewritten as
 dy
dt
+ ν Ay + B(y, y) = f in V ∗ ,
(3.7)
y(τ ) = yτ ,

where the first equation we understand in the sense of distributions on (τ, T ). Now
we write

D(yτ , f ) = {y : y is a weak solution of Problem (3.3) on [τ, T ]}.

72 3 Advances in the 3D Navier-Stokes Equations

It is well known (cf. [19]) that if f ∈ L 2 (τ, T ; V ∗ ), and if yτ ∈ H , then D(yτ , f ) is

not empty.
A weak solution y of Problem (3.3) on [τ, T ] is called a strong one, if it addi-
tionally belongs to Serrin’s class L 8 (τ, T ; (L 4 (Ω))3 ). We note that any strong solu-
tion y of Problem (3.3) on [τ, T ] belongs to C([τ, T ]; V ) ∩ L 2 (τ, T ; D(A)) and
dy
dt
∈ L 2 (τ, T ; H ) (cf. [18, Theorem 1.8.1, p. 296] and references therein).
For any f ∈ L ∞ (τ, T ; H ) and yτ ∈ V it is well known the only local existence
of strong solutions for the 3D Navier-Stokes equations (cf. [18–20] and references
therein).

3.2 Leray-Hopf Property for a Weak Solution of the 3D

Navier-Stokes System: Method of Artificial Control

Let −∞ < τ < T < +∞. We consider the following space of parameters:

Uτ,T := (L 2 (τ, T ; V )) × L 2 τ, T ; V ∗ + L 1 (τ, T ; H ) × H.

Each triple (u, g, z τ ) ∈ Uτ,T is called admissible for the following auxiliary control
problem.
dz
Problem (C) on [τ, T ] with (u, g, z τ ) ∈ Uτ,T : find z ∈ L 2 (τ, T ; V ) with ∈
dt
∗
L (τ, T ; V ) such that z(τ ) = z τ and for all v ∈ V
1

d
(z, v) + ν((z, v)) + b(u, z, v) = g, v (3.8)
dt
in the sense of distributions; cf. Kapustyan et al. [10, 11]; Kasyanov et al. [12, 13];
Melnik and Toscano [15]; Zgurovsky et al. [20, Chap. 6].
As usual, let A : V → V ∗ be the linear operator associated with the bilinear form
((u, v)) = Au, v , u, v ∈ V . For u, v ∈ V we denote by B (u, v) the element of V ∗
defined by B (u, v) , w = b(u, v, w), for all w ∈ V . Then Problem (C) on [τ, T ] with
dz
(u, g, z τ ) ∈ Uτ,T can be rewritten as: find z ∈ L 2 (τ, T ; V ) with ∈ L 1 (τ, T ; V ∗ )
dt
such that
dz
+ ν Az + B (u, z) = g, in V ∗ , and z(τ ) = z τ . (3.9)
dt

We recall, that {w1 , w2 , · · · } ⊂ V is the special basis if ((w j , v)) = λ j (w j , v) for

each v ∈ V and j = 1, 2, · · · , where 0 < λ1 ≤ λ2 ≤ · · · is the sequence of eigen-
values. Let Pm be the projection operator of H onto Hm := span{w1 , · · · , wm }, that
m
is Pm v = i=1 (v, wi )wi for each v ∈ H and m = 1, 2, · · · . Of course we may con-
sider Pm as a projection operator that acts from Vσ onto Hm for each σ > 0 and,
3.2 Leray-Hopf Property for a Weak Solution of the 3D Navier-Stokes System … 73

since Pm∗ = Pm , we deduce that Pm L (Vσ∗ ;Vσ∗ ) ≤ 1. Note that (w j , v)Vσ = λσj (w j , v)
for each v ∈ Vσ and j = 1, 2, · · · .
The following theorem establishes sufficient conditions for the existence of an
unique solution for Problem (C). This is the main result of this section.

Theorem 3.1 Let −∞ < τ < T < +∞, yτ ∈ H , f ∈ L 2 (τ, T ; V ∗ ) + L 1 (τ, T ;

H ), and y be a weak solution of Problem (3.1) on [τ, T ]. If Problem (C) on [τ, T ]
with (u, 0̄, 0̄) ∈ Uτ,T has the unique solution z ≡ 0̄, then (y, f, yτ ) ∈ Uτ,T and Prob-
lem (C) on [τ, T ] with (y, f, yτ ) ∈ Uτ,T has the unique solution z = y. Moreover, y
satisfies inequality (3.5).

Before the proof of Theorem 3.1 we remark that AC([τ, T ]; Hm ), m = 1, 2, · · · ,

will denote the family of absolutely continuous functions acting from [τ, T ] into Hm ,
m = 1, 2, · · · .
Proof of Theorem 3.1. Prove that z = y is the unique solution of Problem (C)
on [τ, T ] with (y, f, yτ ) ∈ Uτ,T . Indeed, y is the solution of Problem (C) on [τ, T ]
with (y, f, yτ ) ∈ Uτ,T , because y is a weak solution of Problem (3.1) on [τ, T ].
Uniqueness holds, because if z is a solution of Problem (C) on [τ, T ] with (y, f, yτ ) ∈
Uτ,T , then z − y ≡ 0̄ is the unique solution of Problem (C) on [τ, T ] with (y, 0̄, 0̄) ∈
Uτ,T .
The rest of the proof establishes that y satisfies inequality (3.5). We note that y
can be obtained via standard Galerkin arguments, that is, if ym ∈ AC([τ, T ]; Hm )
d
with ym ∈ L 1 (τ, T ; Hm ), m = 1, 2, · · · , is the approximate solution such that
dt dym
+ ν Aym + Pm B (y, ym ) = Pm f, in Hm , ym (τ ) = Pm y(τ ), (3.10)
dt

then the following statements hold:

(i) ym satisfy the following energy equality:
 t1  t1
1
ym (t1 ) 2
+ν ym (ξ ) 2
V dξ − f (ξ ), ym (ξ ) dξ
2 s s
 t2  t2
1
= ym (t2 ) 2
+ν ym (ξ ) V dξ −
2
f (ξ ), ym (ξ ) dξ,
2 s s
(3.11)
for each t1 , t2 ∈ [τ, T ], for each m = 1, 2, · · · ;
(ii) there exists a subsequence {ym k }k=1,2,··· ⊆ {ym }m=1,2,··· such that the following
convergence (as k → ∞) hold:
(ii)1 ym k → y weakly in L 2 (τ, T ; V );
(ii)2 ym k → y weakly star in L ∞ (τ, T ; H );
(ii)3 Pm k B y, ym k → B (y, y) weakly in L 2 (τ, T ; V 3∗ );
2
(ii)4 Pm k f → f strongly in L 2 (τ, T ; V ∗ ) + L 1 (τ, T ; H );
dym k dy
(ii)5 → weakly in L 2 (τ, T ; V 3∗ ) + L 1 (τ, T ; H ).
dt dt 2
74 3 Advances in the 3D Navier-Stokes Equations

Indeed, convergences (ii)1 and (ii)2 follow from (3.11) (see also Temam [19,
Remark III.3.1, pp. 264, 282]) and Banach-Alaoglu theorem. Since there exists
1
1
C1 > 0 such that |b(u, v, w)| ≤ C u V w V v V2 v 2 , for each u, v, w ∈ V (see,
for example, Sohr [18, Lemma V.1.2.1]), then (ii)1 , (ii)2 and Banach-Alaoglu theorem
imply (ii)3 . Convergence (ii)4 holds, because of the basic properties of the projec-
tion operators {Pm }m=1,2,··· . Convergence (ii)5 directly follows from (ii)3 , (ii)4 and
(3.10). We note that we may not pass to a subsequence in (ii)1 –(ii)5 , because z = y
is the unique solution of Problem (C) on [τ, T ] with (y, f, yτ ) ∈ Uτ,T .
Moreover, there exists a subsequence {yk j } j=1,2,··· ⊆ {ym k }k=1,2,··· such that

yk j (t) → y(t) strongly in H for a.e. t ∈ (τ, T ) and t = τ, j → ∞. (3.12)

Indeed, according to (3.10), (3.11) and (ii)3 , the sequence {ym k }k=1,2,··· is bounded in
a reflexive Banach space Wτ,T := {w ∈ L 2 (τ, T ; V ) : dtd w ∈ L 1 (τ, T ; V 3∗ )}. Com-
2
pactness lemma yields that Wτ,T ⊂ L 2 (τ, T ; H ) with compact embedding. There-
fore, (ii)1 –(ii)5 imply that ym k → y strongly in L 2 (τ, T ; H ) as k → ∞. Thus, there
exists a subsequence {yk j } j=1,2,··· ⊆ {ym k }k=1,2,··· such that (3.12) holds.
Due to convergences (ii)1 –(ii)5 and (3.12), if we pass to the limit in (3.11) as
m k j → ∞, then we obtain that y satisfies the inequality
 t  t
1 1
y(t) 2
+ν y(ξ ) 2
V dξ − f (ξ ), y(ξ ) dξ ≤ y(τ ) 2 , (3.13)
2 s s 2

for a.e. t ∈ (s, T ), a.e. s ∈ (τ, T ) and s = τ .

Since y ∈ L ∞ (τ, T ; H ) ∩ C([τ, T ]; V ∗ ) and H ⊂ V ∗ with continuous embed-
ding, then y ∈ C([τ, T ]; Hw ); Temam [19, Chap. III]. Thus, equality (3.13) yields
 t  t
1 1
y(t) 2
+ν y(ξ ) 2
V dξ − f (ξ ), y(ξ ) dξ ≤ y(τ ) 2 ,
2 s s 2

for each t ∈ [τ, T ], a.e. s ∈ (τ, T ) and s = τ . Therefore, y satisfies inequality (3.5).
The theorem is proved.

3.3 The Existence of Strong Solutions and 1-Dimensional

Dynamical Systems

Let T > 0. The main result of this section has the following formulation (see also
Figs. 3.3 and 3.4).

Theorem 3.2 Let f ∈ L 2 (0, T ; H ) and y0 ∈ V . Then either for any λ ∈ [0, 1] there
is an yλ ∈ C([0, T ]; V ) ∩ L 2 (0, T ; D(A)) such that yλ ∈ D(λy0 , λ f ), or the set
3.3 The Existence of Strong Solutions and 1-Dimensional Dynamical Systems 75

Fig. 3.3 Relations between

the types of solutions for the
3D Navier-Stokes system

{y ∈ C([0, T ]; V ) ∩ L 2 (0, T ; D(A)) : y ∈ D(λy0 , λ f ), λ ∈ (0, 1)} (3.14)

is unbounded in L 8 (0, T ; (L 4 (Ω))3 ).

In the proof of Theorem 3.2 we use an auxiliary statement connected with con-
tinuity property of strong solutions on parameters of problem (3.3) in Serrin’s class
L 8 (0, T ; (L 4 (Ω))3 ).

Theorem 3.3 Let f ∈ L 2 (0, T ; H ) and y0 ∈ V . If y is a strong solution for Prob-

lem (3.3) on [0, T ], then there exist L , δ > 0 such that for any z 0 ∈ V and
g ∈ L 2 (0, T ; H ), satisfying the inequality

z 0 − y0 2
V + g− f 2
L 2 (0,T ;H ) < δ, (3.15)

the set D(z 0 , g) is one-point set {z} which belongs to C([0, T ]; V ) ∩ L 2 (0, T ; D(A)),
and
ν
z−y 2
C([0,T ];V ) + z−y 2
D(A) ≤L z 0 − y0 2
V + g− f 2
L 2 (0,T ;H ) .
4
(3.16)

Remark 3.1 We note that from Theorem 3.3 with z 0 ∈ V and g ∈ L 2 (0, T ; H ) with
sufficiently small z 0 2V + g 2L 2 (0,T ;H ) , Problem (3.3) has only one global strong
solution.

Remark 3.2 Theorem 3.3 provides that, if for any λ ∈ [0, 1] there is an yλ ∈
L 8 (0, T ; (L 4 (Ω))3 ) such that yλ ∈ D(λy0 , λ f ), then the set
76 3 Advances in the 3D Navier-Stokes Equations

Fig. 3.4 Existence of solutions for the 3D Navier-Stokes System

{y ∈ C([0, T ]; V ) ∩ L 2 (0, T ; D(A)) : y ∈ D(λy0 , λ f ), λ ∈ (0, 1)}

is bounded in L 8 (0, T ; (L 4 (Ω))3 ).

If Ω is a C ∞ -domain and if f ∈ C0∞ ((0, T ) × Ω)3 , then any strong solution y of
Problem (3.3) on [0, T ] belongs to C ∞ ((0, T ] × Ω)3 and p ∈ C ∞ ((0, T ] × Ω) (cf.
[18, Theorem 1.8.2, p. 300] and references therein). This fact directly provides the
next corollary of Theorems 3.2 and 3.3.
Corollary 3.1 Let Ω be a C ∞ -domain, f ∈ C0∞ ((0, T ) × Ω)3 . Then either for any
y0 ∈ V there is a strong solution of Problem (3.3) on [0, T ], or the set

{y ∈ C ∞ ((0, T ] × Ω)3 : y ∈ D(λy0 , λ f ), λ ∈ (0, 1)}

is unbounded in L 8 (0, T ; (L 4 (Ω))3 ) for some y0 ∈ C0∞ (Ω)3 .

3.3 The Existence of Strong Solutions and 1-Dimensional Dynamical Systems 77

Proof of Theorem 3.3. Let f ∈ L 2 (0, T ; H ), y0 ∈ V , and y ∈ C([0, T ]; V ) ∩

L (0, T ; D(A)) be a strong solution of Problem (3.3) on [0, T ]. Due to [17], [19,
2

Chap. 3] the set D(y0 , f ) = {y}. Let us now fix z 0 ∈ V and g ∈ L 2 (0, T ; H ) satis-
fying (3.15) with
  
ν  −2T C 27c4 77 c8 2
δ = min 1; e , C = max ; y 4
C([0,T ];V ) +1 , (3.17)
4 2ν 3 29 ν 7

c > 0 is a constant from the inequalities (cf. [18, 19])

1 1
|b(u, v, w)| ≤ c u V v 2
V v 2
D(A) w H ∀u ∈ V, v ∈ D(A), w ∈ H ; (3.18)

3 1
|b(u, v, w)| ≤ c u 4
D(A) u 4
V v V w H ∀u ∈ D(A), v ∈ V, w ∈ H. (3.19)

The auxiliary problem

 dη
dt
+ ν Aη + B(η, η) + B(y, η) + B(η, y) = g − f in V ∗ ,
(3.20)
η(0) = z 0 − y0 ,

has a strong solution η ∈ C([0, T ]; V ) ∩ L 2 (0, T ; D(A)) with dη

dt
∈ L 2 (0, T ; H ),
i.e.

d
(η, v) + ν((η, v)) + b(η, η, v) + b(y, η, v) + b(η, y, v) = g − f, v for all v ∈ V,
dt

in the sense of distributions on (0, T ). In fact, let {w j } j≥1 ⊂ D(A) be a special basis
(cf. [19]), i.e. Aw j = λ j w j , j = 1, 2, ..., 0 < λ1 ≤ λ2 ≤ ..., λ j → +∞, j → +∞.
We consider Galerkin approximations ηm : [0, T ] → span{w j }mj=1 for solutions of
Problem (3.20) satisfying

d
(ηm , w j ) + ν((ηm , w j )) + b(ηm , ηm , w j ) + b(y, ηm , w j ) + b(ηm , y, w j ) = g − f, w j ,
dt

with (ηm (0), w j ) = (z 0 − y0 , w j ), j = 1, m. Due to (3.18), (3.19) and Young’s

inequality we get

ν 4
2 g − f, Aηm ≤ 2 g − f H ηm D(A) ≤ ηm 2
+ f −g H;
2
4 D(A)
ν
3 3 ν 27c4
−2b(ηm , ηm , Aηm ) ≤ 2c ηm V
2
ηm 2
D(A) ≤ ηm 2
D(A) + ηm V;
6
2 2ν 3
78 3 Advances in the 3D Navier-Stokes Equations

1 3 ν 27c4
−2b(y, ηm , Aηm ) ≤ 2c y V ηm 2
V ηm 2
D(A) ≤ ηm 2
D(A) + y 4
C([0,T ];V ) ηm V;
2
2 2ν 3

7 1 ν 77 c 8
−2b(ηm , y, Aηm ) ≤ 2c ηm 4
ηm 4
y V ≤ ηm 2
+ y 8
C([0,T ];V ) ηm V.
2
D(A) V
2 D(A)
29 ν 7

Thus,
d ν 4
ηm 2
+ ηm 2
≤ C( ηm 2
+ ηm V)
6
+ g− f H,
2
dt V
4 D(A) V
ν
where C > 0 is a constant from (3.17). Hence, the absolutely continuous function
ϕ = min{ ηm 2V , 1} satisfies the inequality dtd ϕ ≤ 2Cϕ + ν4 g − f 2H , and there-
fore ϕ ≤ L( z 0 − y0 2V + g − f 2L 2 (0,T ;H ) ) < 1 on [0, T ], where L = δ −1 . Thus,
{ηn }n≥1 is bounded in L ∞ (0, T ; V ) ∩ L 2 (0, T ; D(A)) and { dtd ηn }n≥1 is bounded in
L 2 (0, T ; H ). In a standard way we get that the limit function η of ηn , n → +∞, is
a strong solution of Problem (3.20) on [0, T ]. Due to [17], [19, Chapter 3] the set
D(z 0 , g) is one-point z = y + η ∈ L 8 (0, T ; (L 4 (Ω))3 ). So, z is strong solution of
Problem (3.3) on [0, T ] satisfying (3.16).
The theorem is proved.
Proof of Theorem 3.2. Let f ∈ L 2 (0, T ; H ) and y0 ∈ V . We consider the 3D
controlled Navier-Stokes system (cf. [10, 15])
 dy
dt
+ ν Ay + B(z, y) = f,
(3.21)
y(0) = y0 ,

where z ∈ L 8 (0, T ; (L 4 (Ω))3 ).

By using standard Galerkin approximations (see [19]) it is easy to show that for any
z ∈ L 8 (0, T ; (L 4 (Ω))3 ) there exists an unique weak solution y ∈ L ∞ (0, T ; H ) ∩
L 2 (0, T ; V ) of Problem (3.21) on [0, T ], that is,

d
(y, v) + ν((y, v)) + b(z, y, v) = f, v , for all v ∈ V, (3.22)
dt

in the sense of distributions on (0, T ). Moreover, by the inequality

1 7 ν
|b(u, v, Av)| ≤ c1 u (L 4 (Ω))3 v 4
V v 4
D(A) ≤ v 2
D(A) + c2 u 8
(L 4 (Ω))3 v V,
2
2
(3.23)
for all u ∈ (L 4 (Ω))3 and v ∈ D(A), where c1 , c2 > 0 are some constants that do
not depend on u, v (cf. [19]), we find that y ∈ C([0, T ]; V ) ∩ L 2 (0, T ; D(A))
and B(z, y) ∈ L 2 (0, T ; H ), so dy
dt
∈ L 2 (0, T ; H ) as well. We add that, for any z ∈
L (0, T ; (L (Ω)) ) and corresponding weak solution y ∈ C([0, T ]; V ) ∩ L 2 (0, T ;
8 4 3

D(A)) of (3.21) on [0, T ], by using Gronwall inequality, we obtain

3.3 The Existence of Strong Solutions and 1-Dimensional Dynamical Systems 79

t 8
2c2 z(t) (L 4 (Ω))3
dt
y(t) 2
≤ y0 2V e 0 , ∀t ∈ [0, T ];
V
⎡ ⎤
T (3.24)
T 2c2 z(t) 8
dt
ν ≤ y0 ⎣1 + 2c2 e (L 4 (Ω))3
⎦
L 8 (0,T ;(L 4 (Ω))3 ) .
2 2 8
y(t) D(A) dt V
0 z
0

Let us consider the operator F : L 8 (0, T ; (L 4 (Ω))3 ) → L 8 (0, T ; (L 4 (Ω))3 ),

where F(z) ∈ C([0, T ]; V ) ∩ L 2 (0, T ; D(A)) is the unique weak solution of (3.21)
on [0, T ] corresponded to z ∈ L 8 (0, T ; (L 4 (Ω))3 ).
Let us check that F is a compact transformation of Banach space L 8 (0, T ;
(L (Ω))3 ) into itself (cf. [7]). In fact, if {z n }n≥1 is a bounded sequence in L 8 (0, T ;
4

(L 4 (Ω))3 ), then, due to (3.23) and (3.24), the respective weak solutions yn , n =
1, 2, ..., of Problem (3.21) on [0, T ] are uniformly bounded in C([0, T ]; V ) ∩
L 2 (0, T ; D(A)) and their time derivatives dy dt
n
, n = 1, 2, ..., are uniformly bounded in
L (0, T ; H ). So, {F(z n )}n≥1 is a precompact set in L 8 (0, T ; (L 4 (Ω))3 ). In a standard
2

way we deduce that F : L 8 (0, T ; (L 4 (Ω))3 ) → L 8 (0, T ; (L 4 (Ω))3 ) is continuous

mapping. Since F is a compact transformation of L 8 (0, T ; (L 4 (Ω))3 ) into itself,
Schaefer’s Theorem (cf. [7, p. 133] and references therein) and Theorem 3.3 provide
the statement of Theorem 3.2. We note that Theorem 3.3 implies that the set {z ∈
L 8 (0, T ; (L 4 (Ω))3 ) : z = λF(z), λ ∈ (0, 1)} is bounded in L 8 (0, T ; (L 4 (Ω))3 ) iff
the set defined in (3.14) is bounded in L 8 (0, T ; (L 4 (Ω))3 ).
The theorem is proved.

3.4 Extremal Solutions: Existence and Continuity Results

in Strongest Topologies

We consider the 3D controlled Navier-Stokes system

 dy
dt
+ Ay + B(u, y) = f,
(3.25)
y(τ ) = yτ ∈ H,

where f ∈ H and
⎧
⎪
⎪ u ∈ L ∞ (τ, +∞; H ) ∩ L loc
2 (τ, +∞; V ) ∩ L ∞ (τ, +∞; L4 (Ω)),
loc
⎨ +∞

u(·) ∈ Uτ = u( p) 2 e−δp dp < ∞, |u( p)| ≤ R0 for a.a. p ≥ τ,
⎪
⎪ τ
⎩
u(t) L4 ≤ α for a.a. t > τ,
(3.26)

+∞
Jτ (u, y) = y( p) − u( p) 2 e−δp dp → inf, (3.27)
τ
80 3 Advances in the 3D Navier-Stokes Equations

|f|
with δ = λ1 ν, R0 = νλ 1
, and where λ1 is the first eigenvalue of the Stokes operator
A and α > 0 is some constant.
By using standard Galerkin approximations it is easy to show that for any
yτ ∈ H and u(·) ∈ Uτ there exists a unique weak solution y(·) ∈ L ∞ (τ, +∞; H ) ∩
2 (τ, +∞; V ) of (3.25), that is,
L loc

d
(y, v) + ν((y, v)) + b(u, y, v) = f, v , for all v ∈ V. (3.28)
dt
Indeed, let us prove existence of a weak solution of (3.25). Let {wi } ⊂ D (A)
be the sequence of eigenfunctions of A, which are an orthonormal basis of H . Let
m
y m (t) = i=1 gim (t) wi be the Galerkin approximations of (3.25), i.e.
 dy m
+ ν Ay m + Pm B (u, y m ) = Pm f,
dt (3.29)
y m (τ ) = yτm ,

where Pm is the projection onto the finite dimensional subspace generated by the set
{w1 , ..., wm }. Also, yτm belongs to this subspace and yτm → yτ in H .
We need to obtain some a priori estimates for the approximative functions {y m }.
Multiplying (3.29) by y m we obtain

1 d m2  2
|y | + ν  y m  = f, y m , (3.30)
2 dt
where we have used the equalities

Pm B u, y m , y m = B u, y m , y m = b u, y m , y m = 0.

Also from (3.30) we obtain for all p ∈ [s, T ] , s ∈ [τ, T ] that

p 
1 m  m  p
1
|y ( p) |2 + ν  y (τ )2 dτ ≤ f (τ ) , y m (τ ) dτ + |y m (s) |2 .
2 s 2
s
(3.31)

In view of (3.31) we conclude that {y m } is bounded in L 2 (τ, T ; V ) ∩ L ∞ (τ, T ; H ) .

Therefore, passing to a subsequence we obtain y m → y weakly in L 2 (τ, T ; V )
and weakly star in L ∞ (τ, T ; H ). From the inequalities

|b(u, y m , w)| ≤ d u L4 ym w , ∀w ∈ V,

and
   
 Pm B u, y m  ∗ ≤  B u, y m  ∗ ,
V V
3.4 Extremal Solutions: Existence and Continuity Results in Strongest Topologies 81

due to the choice of the spacial basis, we immediately obtain that Pm B (u, y m ) is
bounded in L 2 (τ, T ; V ∗ ). Then

d m d
y → y weakly in L 2 τ, T ; V ∗ , m → ∞,
dt dt

so that y ∈ C([τ, T ]; H ) and by the Compactness Lemma we have

y m → y strongly in L 2 (τ, T ; H ) , m → ∞.

Hence, y m (t) → y(t) strongly in H for a.e. t ∈ (τ, T ) , m → ∞. Since one can eas-
ily prove using the Ascoli-Arzelà theorem that y m → y, m → ∞, in C ([τ, T ]; V ∗ ),
a standard argument implies that y m (t) → y (t) weakly in H for all t ∈ [τ, T ] , m →
∞. In particular, y(τ ) = yτ .
On the other hand, from

 m 2 T  m 2
u i y  ≤ ui 2 y  dt ≤ C
j L 2 (τ,T ;L 2 (Ω)) L 4 (Ω) j L 4 (Ω)
τ

we obtain u i y mj → u i y j weakly in L 2 (τ, T ; L 2 (Ω)) , m → ∞, so that

 3 
 
T T
∂w j
b(u, y − y, w)dt = −
m
u i y mj − y j d xdt → 0, m → ∞,
τ i, j=1 τ Ω ∂ xi

for any w ∈ L 2 (τ, T ; V ). This implies

B(u, y m ) → B(u, y) weakly in L 2 τ, T ; V ∗ , m → ∞.

So we can pass to the limit in (3.29) and deduce that y is solution of (3.25). To prove
uniqueness we should note that if y1 , y2 are solutions of (3.25), corresponding the
same control function u, then

d d(y1 − y2 )
|y1 − y2 |2 = 2( , y1 − y2 ),
dt dt

b(u, y1 − y2 , y1 − y2 ) = 0.

So after simple calculations we have

d
|y1 − y2 |2 ≤ C|y1 − y2 |2 ,
dt
and therefore y1 ≡ y2 .
82 3 Advances in the 3D Navier-Stokes Equations

Moreover, by the inequality

|b(u, y, v)| = |b(u, v, y)| ≤ c1 u L4 v y L4 ≤ c2 c1 u L4 v y , ∀u, y, v ∈ V,

∗ ∗
and (3.26) we have B(u (·) , y (·)) ∈ L loc
2 (τ, +∞; V ), so dt ∈ L 2 (τ, +∞; V ) as
dy loc

well. Hence, it follows that y(·) ∈ C([τ, +∞); H ) (so the initial condition y(τ ) = yτ
makes sense for any yτ ∈ H ) and standard arguments imply that for all t ≥ s ≥ τ ,

F(y(t)) := |y(t)|2 − R02 eδt ≤ F(y(s)), (3.32)

t t
1
Vτ (y(t)) := |y(t)|2 + ν y( p) dp −
2
( f, y( p))dp ≤ Vτ (y(s)), (3.33)
2
τ τ

t
| f |2
|y(t)| + ν 2
y( p) 2 dp ≤ |yτ |2 + (t − τ ). (3.34)
νλ1
τ

Indeed, multiplying the equation by y (t) and using the property b (u, y, y) = 0
we obtain
1 d
|y|2 + ν y 2
= ( f, y) . (3.35)
2 dt
After integration over (s, t) we obtain
 t  t
1 1
|y (t)|2 + ν y ( p) 2
dp = ( f, y ( p)) dp + |y (s)|2 , (3.36)
2 s s 2

and then (3.33) follows. Taking s = τ in (3.36) and using the inequality

1 | f |2 ν
( f, y ( p)) ≤ | f | |y ( p)| ≤ √ | f | y ( p) ≤ + y ( p) 2
λ1 2λ1 ν 2

we have
 t
| f |2
|y (t)| + ν
2
y ( p) 2
dp ≤ |y (τ )|2 + (t − τ ) .
s λ1 ν

Finally, from (3.35) we obtain

d | f |2
|y|2 + λ1 ν |y|2 ≤ .
dt λ1 ν
3.4 Extremal Solutions: Existence and Continuity Results in Strongest Topologies 83

Multiplying the last inequality by eνλ1 t and integrating we get

| f |2 λ1 νt
|y (t)|2 eνλ1 t ≤ |y (s)|2 eνλ1 s + (e − eνλ1 s ),
λ21 ν 2

and then (3.32) holds.

So, for all n ≥ 0,

τ +(n+1)

y( p) − u( p) 2 e−δp dp ≤
τ +n


τ +(n+1) 
τ +(n+1)
−δ(n+τ )
≤ 2e y( p) dp + 2
2
u( p) 2 e−δp dp
τ +n τ +n


τ +(n+1)
2 | f |2
≤ e−δ(n+τ ) (|yτ |2 + )+2 u( p) 2 e−δp dp.
ν νλ1
τ +n

From this

∞
 
τ +(n+1)

Jτ (u, y) = y( p) − u( p) 2 e−δp dp
n=0 τ +n

∞ ∞ 
τ +(n+1)
2e−δτ | f |2  −δn 
≤ (|yτ |2 + ) e +2 u( p) 2 e−δp dp < ∞.
ν νλ1 n=0 n=0 τ +n

Therefore, the functional Jτ and the optimal control problem (3.25), (3.26) and
(3.27) is correctly defined.

Lemma 3.1 For any τ ∈ R and yτ ∈ H the optimal control problem (3.25), (3.26)
and (3.27) has at least one solution {y(·), u(·)}, and, moreover, dy
dt
∈ L loc
2 (τ, +∞;
∗
V ), y(·) ∈ C([τ, +∞); H ) and (3.32), (3.33) and (3.34) hold.

Proof Let {yn , u n } be a minimizing sequence such that

+∞
1
yn ( p) − u n ( p) 2 e−δp dp ≤ d + , ∀n ≥ 1 ,
n
τ
84 3 Advances in the 3D Navier-Stokes Equations

where d = inf Jτ (u, y). Thus, for all T > τ and n ≥ 1

T
yn ( p) − u n ( p) 2 e−δp dp ≤ d + n1 ,
τ
(3.37)
T
yn ( p) − u n ( p) 2 dp ≤ (d + n1 )eδT .
τ

From (3.32), (3.33) and (3.34) we obtain that {yn } is bounded in L ∞ (τ, T ; H ) ∩
L 2 (τ, T ; V ). Hence, (3.37) implies that {u n } is bounded in L 2 (τ, T ; V ) and from the
definition of Uτ it follows that

|u n ( p)| ≤ R0 , ∀ p ≥ τ,
u n ( p) L4 ≤ α for a.e. p > τ and for all n ≥ 1.

Therefore, there exist u ∈ L ∞ (τ, T ; H ) ∩ L 2 (τ, T ; V ) ∩ L ∞ (τ, T ; L4 (Ω)) and y ∈

L ∞ (τ, T ; H ) ∩ L 2 (τ, T ; V ) such that

un → u weakly in L 2 (τ, T ; V ),
un → u weakly star in L ∞ (τ, T ; H ),
un → u weakly star in L ∞ (τ, T ; L4 (Ω)),
yn → y weakly in L 2 (τ, T ; V ),
yn → y weakly star in L ∞ (τ, T ; H ), n → ∞.

Moreover, B(u n , yn ) V ∗ ≤ c1 yn u n L4 . Hence, dy dt

n
is bounded in L 2 (τ, T ; V ∗ ).
From this using standard arguments, we obtain that y(·) ∈ C([τ, T ]; H ) is the solu-
tion of (3.25) with control u(·), y(·) satisfies (3.32), (3.33) and (3.34), and for this
control the following relations hold:

|u( p)| ≤ R0 , for a.a. p ≥ τ,

u( p) L4 ≤ α for a.a. p > τ,

u ∈ L 2 (τ, T ; V ),

T
y( p) − u( p) 2 e−δp dp ≤ d.
τ

The fact that y (·) is a solution with control u (·) is proved in a standard way.
Indeed, as dy
dt
n
is bounded in L 2 (τ, T ; V ∗ ), up to subsequence

d d
yn → y weakly in L 2 τ, T ; V ∗ , n → ∞.
dt dt
3.4 Extremal Solutions: Existence and Continuity Results in Strongest Topologies 85

Thus, y ∈ C([τ, T ]; H ) and arguing as in the proof of the existence of solution for
(3.25) we obtain

yn → y strongly in L 2 (τ, T ; H ) ,
yn (t) → y(t) strongly in H for a.a t ∈ (τ, T ) ,
yn (t) → y(t) weakly in H for all t ∈ [τ, T ], n → ∞.

From

 n n 2 T  n 2  n 2
u y  ≤ u  y  dt ≤ C
i j L 2 (τ,T ;L 2 (Ω)) i L 4 (Ω) j L 4 (Ω)
τ

we obtain u i y mj → u i y j weakly in L 2 (τ, T ; L 2 (Ω)) , n → ∞, so that

 3 
  
T T ∂w j T
b(u n , y n , w)dt = − u in y nj d xdt → b(u, w, y)dt, n → ∞,
τ ∂ xi
i, j=1 τ Ω τ

for any w ∈ L 2 (τ, T ; V ). This implies

B(u, y m ) → B(u, y) weakly in L 2 τ, T ; V ∗ , n → ∞.

Hence we can pass to the limit in (3.25) and obtain that {u, y} is a solution. Also,
y (τ ) = yτ .
By using a standard diagonal procedure we can claim that y(·) and u(·) are defined
on [τ, +∞), yn → y, u n → u in the previous sense on every [τ, T ], n → ∞, and

+∞
y( p) − u( p) 2 e−δp dp ≤ d. (3.38)
τ

By (3.34), arguing as before,

 ∞ ∞ 
 τ +n+1
2 −δp 2 −δp
y ( p) e dp = y ( p) e dp
τ n=0 τ +n
  ∞
e−δτ | f |2  −δn
≤ |yτ |2 + e < ∞.
ν νλ1 n=0

and from (3.38) we have

+∞
u( p) 2 e−δp dp < ∞.
τ
86 3 Advances in the 3D Navier-Stokes Equations

It follows that u(·) ∈ Uτ and from (3.38) we obtain that {y(·), u(·)} is an optimal
pair of problem (3.25), (3.26) and (3.27).
The lemma is proved.
Remark 3.3 Lemma 3.1 was proved in [10].

References

1. Ball, J.M.: Continuity properties and global attractors of generalized semiflows and the Navier-
Stokes equations. Nonlinear Sci. 7, 475–502 (1997). Erratum, ibid 8:233, 1998. Corrected
version appears in Mechanics: from Theory to Computation. pp. 447–474. Springer (2000)
2. Balibrea, F., Caraballo, T., Kloeden, P.E., Valero, J.: Recent developments in dynamical sys-
tems: three perspectives. Int. J. Bifurc. Chaos (2010). doi:10.1142/S0218127410027246
3. Barbu, V., Rodrigues, S.S., Shirikyan, A.: Internal exponential stabilization to a nonstation-
ary solution for 3D NavierStokes equations. SIAM J. Control Optim. (2011). doi:10.1137/
100785739
4. Cao, Ch., Titi, E.S.: Global regularity criterion for the 3D NavierStokes equations involving one
entry of the velocity gradient tensor. Arch. Ration. Mech. Anal. (2011). doi:10.1007/s00205-
011-0439-6
5. Chepyzhov, V.V., Vishik, M.I.: Trajectory and global attractors of three-dimensional Navier-
Stokes systems. Math. Notes (2002). doi:10.1023/A:1014190629738
6. Cheskidov, A., Shvydkoy, R.: A unified approach to regularity problems for the 3D Navier-
Stokes and euler equations: the use of kolmogorovs dissipation range. J. Math. Fluid Mech.
(2014). doi:10.1007/s00021-014-0167-4
7. Cronin, J.: Fixed Points and Topological Degree in Nonlinear Analysis. American Mathematical
Society, Providence, RI (1964)
8. Gajewski, H., Gröger, K., Zacharias, K.: Nichtlineare Operatorgleichungen Und Operatordif-
ferentialgleichungen. Akademie-Verlag, Berlin (1978)
9. Halmos, P.R.: Measure Theory. Springer, NewYork (1974)
10. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Pullback attractors for a class of extremal solutions
of the 3D Navier-Stokes system. J. Math. Anal. Appl. 373, 535–547 (2011)
11. Kapustyan, O.V., Melnik, V.S., Valero, J.: A weak attractor and properties of solutions for the
three-dimensional Bénard problem. Discrete Contin. Dyn. Syst. 18, 449–481 (2007)
12. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Topological properties of strong solutions for
the 3D Navier-Stokes equations. Solid Mech. Appl. 211, 181–187 (2014)
13. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: A criterion for the existence of strong solutions
for the 3D Navier-Stokes equations. Appl. Math. Lett. 26, 15–17 (2013)
14. Kloeden, P.E., Marin-Rubio, P., Valero, J.: The envelope attractor of non-strict multi-valued
dynamical systems with application to the 3D Navier-Stokes and reaction-diffusion equations.
Set-Valued Var. Anal. 21, 517–540 (2013). doi:10.1007/s11228-012-0228-x
15. Melnik, V.S., Toscano, L.: On weak extensions of extreme problems for nonlinear operator
equations. Part I. weak solutions. J. Autom. Inf. Sci. 38, 68–78 (2006)
16. Royden, H.L.: Real Analysis, 2nd edn. Macmillan, New York (1968)
17. Serrin, J.: The initial value problem for the Navier-Stokes equations. In: Langer, R.E. (ed.)
Nonlinear Problems, pp. 69–98. University of Wisconsin Press, Madison (1963)
18. Sohr, H.: The Navier-Stokes Equations. An Elementary Functional Analytic Approach.
Birkhäuser, Basel (2001)
19. Temam, R.: Navier-Stokes Equations. North-Holland, Amsterdam (1979)
20. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
Inclusions and Variation Inequalities for Earth Data Processing III. Springer, Berlin (2012)
 Part II
Convergence Results in Strongest
Topologies
Chapter 4
Strongest Convergence Results for Weak
Solutions of Non-autonomous
Reaction-Diffusion Equations
with Carathéodory’s Nonlinearity

Abstract In this chapter we consider the problem of uniform convergence results

for all globally defined weak solutions of non-autonomous reaction-diffusion sys-
tem with Carathéodory’s nonlinearity satisfying standard sign and polynomial growth
assumptions. The main contributions of this chapter are: the uniform convergence
results for all globally defined weak solutions of non-autonomous reaction-diffusion
equations with Carathéodory’s nonlinearity and sufficient conditions for the conver-
gence of weak solutions in strongest topologies.
Let N , M = 1, 2, . . ., Ω ⊂ R N be a bounded domain with sufficiently smooth
boundary ∂Ω. We consider a problem of long-time behavior of all globally defined
weak solutions for the non-autonomous parabolic problem (named RD-system)

yt = aΔy − f (x, t, y), x ∈ Ω, t > 0,
(4.1)
y|∂Ω = 0,

as t → +∞, where y = y(x, t) = (y (1) (x, t), . . . , y (M) (x, t)) is unknown vector-
function, f = f (x, t, y) = ( f (1) (x, t, y), . . . , f (M) (x, t, y)) is given function, a is
real M × M matrix with positive symmetric part.

4.1 Translation-Compact, Translation-Bounded

and Translation Uniform Integrable Functions

To introduce the assumptions on parameters of Problem (4.1) we need to present

some additional constructions. Let γ ≥ 1 and E be a real separable Banach space.
γ (R+ ; E ) we consider the Fréchet space of all locally integrable functions with
As L loc
γ (R+ ; E ) if and only if for any finite interval [τ, T ] ⊂ R+
values in E , i.e. ϕ ∈ L loc
the restriction of ϕ on [τ, T ] belongs to the space L γ (τ, T ; E ). If E ⊆ L 1 (Ω), then
any function ϕ from L locγ (R+ ; E ) can be considered as a measurable mapping that
acts from Ω × R+ into R. Further, we write ϕ(x, t), when we consider this mapping
as a function from Ω × R+ into R, and ϕ(t), if this mapping is considered as an
γ (R+ ; E ); cf. Gajewski et al. [3, Chap. III]; Temam [23]; Babin and
element from L loc
Vishik [1]; Chepyzhov and Vishik [5]; Zgurovsky et al. [28] and references therein.
© Springer International Publishing AG 2018 89
M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_4
90 4 Strongest Convergence Results …

A function ϕ ∈ L loc
γ (R+ ; E ) is called translation bounded in L γ (R+ ; E ), if
loc

t+1
γ
sup ϕ(s)E ds < +∞; (4.2)
t≥0
t

Chepyzhov and Vishik [7, p. 105]. A function ϕ ∈ L loc 1 (R+ ; L 1 (Ω) is called trans-
lation uniform integrable (t.u.i.) in L loc
1 (R+ ; L 1 (Ω)), if

t+1
lim sup |ϕ(x, s)|χ{|ϕ(x,s)|≥K } d xds = 0. (4.3)
K →+∞ t≥0
t Ω

Dunford–Pettis compactness criterion provides that a function ϕ ∈ L loc 1 (R+ ; L 1 (Ω))

1 (R+ ; L 1 (Ω)) if and only if for every sequence of elements {τn }n≥1 ⊂
is t.u.i. in L loc
R+ the sequence {ϕ( · + τn )}n≥1 contains a subsequence which converges weakly
1 (R+ ; L 1 (Ω)). We note that for any γ > 1 Hölder’s and Chebyshev’s inequal-
in L loc
ities imply that every translation bounded in L loc γ (R+ ; L γ (Ω)) function is t.u.i. in
L loc
1 (R + ; L 1 (Ω)), because


t+1 
t+1
1
|ϕ(x, s)|χ{|ϕ(x,s)|≥K } d xds ≤ sup |ϕ(x, s)|γ d xds → 0 as K → +∞.
K γ −1 t≥0
t Ω t Ω

4.2 Setting of the Problem

Throughout this chapter we suppose that the listed below assumptions hold.
Assumption I. Let pi ≥ 2 and qi > 1 are such that p1i + q1i = 1, for any i =
1, 2, . . . , M. Moreover, there exists a positive constant d such that 21 (a+a ∗ ) ≥ d I ,
where I is unit M × M matrix, a ∗ is a transposed matrix for a.
Assumption II. The interaction function f = ( f (1) , . . . , f (M) ) : Ω × R+ ×
R M → R M satisfies the standard Carathéodory’s conditions, i.e. the mapping
(x, t, u) → f (x, t, u) is continuous in u ∈ R M for a.e. (x, t) ∈ Ω × R+ , and it is
measurable in (x, t) ∈ Ω × R+ for any u ∈ R M .
Assumption III. (Growth Condition). There exist a t.u.i. in L loc1 (R+ ; L 1 (Ω)) func-
tion c1 : Ω × R+ → R+ and a constant c2 > 0 such that


M
 (i)  
M
 (i)  pi
 f (x, t, u)qi ≤ c1 (x, t) + c2 u 
i=1 i=1
4.2 Setting of the Problem 91

for any u = (u (1) , . . . , u (M) ) ∈ R M , and a.e. (x, t) ∈ Ω × R+ .

Assumption IV. (Sign Condition). There exists a constant α > 0 and a t.u.i. in
1 (R+ ; L 1 (Ω)) function β : Ω × R+ → R+ such that
L loc


M 
M
 (i)  pi
f (i) (x, t, u)u (i) ≥ α u  − β(x, t)
i=1 i=1

for any u = (u (1) , . . . , u (M) ) ∈ R M , and a.e. (x, t) ∈ Ω × R+ .

In further arguments we will use standard functional Hilbert spaces H =
(L 2 (Ω)) M , V = (H01 (Ω)) M , and V ∗ = (H −1 (Ω)) M with standard respective inner
products and norms (·, ·) H and  ·  H , (·, ·)V and  · V , and (·, ·)V ∗ and  · V ∗ , vector
notations p = ( p1 , p2 , . . . , p M ) and q = (q1 , q2 , . . . , q M ), and the spaces

Lp (Ω) := L p1 (Ω) × · · · × L p M (Ω), Lq (Ω) := L q1 (Ω) × · · · × L q M (Ω),

Lp (τ, T ; Lp (Ω)) := L p1 (τ, T ; L p1 (Ω)) × · · · × L p M (τ, T ; L p M (Ω)),
Lq (τ, T ; Lq (Ω)) := L q1 (τ, T ; L q1 (Ω)) × · · · × L q M (τ, T ; L q M (Ω)), 0 ≤ τ < T < +∞.

Let 0 ≤ τ < T < +∞. A function y = y(x, t) ∈ L2 (τ, T ; V ) ∩ Lp (τ, T ; Lp (Ω))

is called a weak solution of Problem (4.1) on [τ, T ], if for any function ϕ = ϕ(x) ∈
(C0∞ (Ω)) M , the following identity holds
 
d
y(x, t) · ϕ(x)d x + {a∇ y(x, t) · ∇ϕ(x) + f (x, t, y(x, t)) · ϕ(x)}d x = 0
dt Ω Ω
(4.4)
in the sense of scalar distributions on (τ, T ).
In the general case Problem (4.1) on [τ, T ] with initial condition y(x, τ ) = yτ (x)
in Ω has more than one weak solution with yτ ∈ H (cf. Balibrea et al. [2] and
references therein).

4.3 Preliminary Properties of Weak Solutions

Let ·, · : (V ∗ + Lq (Ω)) × (V ∩ Lp (Ω)) → R be the pairing in (V ∗ + Lq (Ω)) ×

(V ∩ Lp (Ω)), that coincides on H × (V ∩ Lp (Ω)) with the inner product (·, ·) H on
the Hilbert space H , i.e. u, v = (u, v) H for any u ∈ H and v ∈ V ∩ Lp (Ω).
For fixed nonnegative τ and T , τ < T , let us consider the spaces
(i)
X τ,T = L 2 (τ, T ; H01 (Ω)) ∩ L pi (τ, T ; L pi (Ω)),
(i) ∗
X τ,T = L 2 (τ, T ; H −1 (Ω)) + L qi (τ, T ; L qi (Ω)),
(1) (M) ∗ (1) ∗ (M) ∗
X τ,T = X τ,T × · · · × X τ,T , X τ,T = X τ,T × · · · × X τ,T ,
(i) (i) (i) ∗ (1) (M)
Wτ,T = {y ∈ X τ,T | y  ∈ X τ,T }, Wτ,T = Wτ,T × · · · × Wτ,T ,
92 4 Strongest Convergence Results …

(i)
where y  is a derivative of an element y ∈ X τ,T (y ∈ X τ,T ) in the sense of
D([τ, T ]; H −1 (Ω) + L q (Ω)) (D([τ, T ]; V ∗ + Lq (Ω)) respectively); Gajewski
et al. [3, Definition IV.1.10]. Note that the space Wτ,T is a reflexive Banach space with
the graph norm of a derivative uWτ,T = u X τ,T +u   X τ,T∗ ,u ∈ W
τ,T . Let ·, · X τ,T :
∗ ∗
X τ,T × X τ,T → R be the pairing in X τ,T × X τ,T , that coincides on L 2 (τ, T ; H )× X τ,T
T
with the inner product in L 2 (τ, T ; H ), i.e. u, v X τ,T = (u(t), v(t)) H dt for any
τ
u ∈ L 2 (τ, T ; H ) and v ∈ X τ,T . Gajewski et al. [3, Theorem IV.1.17] provide that
(i)
the embedding Wτ,T ⊂ C([τ, T ]; L 2 (Ω)) is continuous and dense, i = 1, 2, . . . , M.
Thus, the embedding Wτ,T ⊂ C([τ, T ]; H ) is continuous and dense. Moreover,

T  
(u(T ), v(T )) H − (u(τ ), v(τ )) H = u  (t), v(t) + v (t), u(t) dt, (4.5)
τ

for any u, v ∈ Wτ,T .

If y(x, t) ∈ Lp (τ, T ; Lp (Ω)), then Assumptions I–III yield

f (x, t, y(x, t)) ∈ Lq (τ, T ; Lq (Ω)),

and

M
 f (i) (y( · )) Liq (τ,T ;L q (Ω))
q
i i
i=1
 (4.6)

M
(i) p
≤c2 y ( · ) Lip (τ,T ;L p (Ω)) + c1 (x, t)d xdt.
i i
i=1 Ω×(τ,T )

Moreover, if y(x, t) ∈ L2 (τ, T ; V ), then ay(x, t) ∈ L2 (τ, T ; V ∗ ).

Assumptions I–IV and Chepyzhov and Vishik [7, pp. 283–284] (see also
Zgurovsky et al. [27, Chap. 2] and references therein) provide the existence of a
weak solution of Cauchy problem (4.1) with initial data y(τ ) = y (τ ) on the interval
[τ, T ], for any y (τ ) ∈ H . The proof is provided by standard Faedo–Galerkin approxi-
mations and using local existence Carathéodory’s theorem instead of classical Peano
results. A priori estimates are similar. Formula (4.4) and definition of the derivative
for an element from D([τ, T ]; V ∗ + Lq (Ω)) yield that each weak solution y ∈ X τ,T
of Problem (4.1) on [τ, T ] belongs to the space Wτ,T . Moreover, each weak solution
of Problem (4.1) on [τ, T ] satisfies the equality:
 T
∂ y(x, t)
· ψ(x, t) + a∇ y(x, t) · ∇ψ(x, t) + f (x, t, y(x, t)) · ψ(x, t) d xdt = 0,
τ Ω ∂t
(4.7)
for any ψ ∈ X τ,T . For fixed τ and T , such that 0 ≤ τ < T < +∞, we denote

Dτ,T (y (τ ) ) = {y(·) | y is a weak solution of (4.1) on [τ, T ], y(τ ) = y (τ ) }, y (τ ) ∈ H.

4.3 Preliminary Properties of Weak Solutions 93

We remark that Dτ,T (y (τ ) ) = ∅ and Dτ,T (y (τ ) ) ⊂ Wτ,T , if 0 ≤ τ < T < +∞

and y (τ ) ∈ H . Moreover, the concatenation of Problem (4.1) weak solutions is a
weak solutions too, i.e. if 0 ≤ τ < t < T , y (τ ) ∈ H , y(·) ∈ Dτ,t (y (τ ) ), and
v(·) ∈ Dt,T (y(t)), then 
y(s), s ∈ [τ, t],
z(s) =
v(s), s ∈ [t, T ],

belongs to Dτ,T (y (τ ) ); cf. Zgurovsky et al. [28, pp. 55–56].

Listed above properties of solutions and Gronwall lemma provide that for any
finite time interval [τ, T ] ⊂ R+ each weak solution y of Problem (4.1) on [τ, T ]
satisfies estimates
 t M  t
  t
y (i) (ξ ) Li (Ω) dξ + 2d
p
y(t)2H − 2 β(x, ξ )d xdξ + 2α y(ξ )2V dξ
pi
τ Ω i=1 s s
 s
≤ y(s)2H − 2 β(x, ξ )d xdξ, (4.8)
τ Ω

 t
y(t)2H ≤ y(s)2H e−2dλ1 (t−s) + 2 β(x, ξ )e−2dλ1 (t−ξ ) d xdξ, (4.9)
s Ω

for any t, s ∈ [τ, T ], t ≥ s, where λ1 is the first eigenvalue of the scalar operator −Δ
with Dirichlet boundary conditions; cf. Chepyzhov and Vishik [7, p. 285]; Vishik
et al. [28, p. 56]; Valero and Kapustyan [24] and references therein. We note that the
same term with β appears both on the left and right hand side of inequality (4.9).
This was done on purpose to comply the inequality with the definition (4.18) of J
and Jk below.
Therefore, any weak solution y of Problem (4.1) on a finite time interval [τ, T ] ⊂
R+ can be extended to a global one, defined on [τ, +∞). For arbitrary τ ≥ 0 and
y (τ ) ∈ H let Dτ (y (τ ) ) be the set of all weak solutions (defined on [τ, +∞)) of Problem
(4.1) with initial data y(τ ) = y (τ ) . Let us consider the family Kτ+ = ∪ y (τ ) ∈H Dτ (y (τ ) )
of all weak solutions of Problem (4.1) defined on the semi-infinite time interval
[τ, +∞).
In further arguments as a Banach space Ft1 ,t2 we consider either C([t1 , t2 ]; H ) or
Wt1 ,t2 with respective topologies of strong convergence, where 0 ≤ t1 < t2 < +∞.
Consider the Fréchet space

F loc (R+ ) := {y : R+ → H : Πt1 ,t2 y ∈ Ft1 ,t2 for any [t1 , t2 ] ⊂ R+ },

where Πt1 ,t2 is the restriction operator to the interval [t1 , t2 ]; Chepyzhov and Vishik [5,
p. 918]. We remark that the sequence { f n }n≥1 converges (converges weakly respec-
tively) in F loc (R+ ) towards f ∈ F loc (R+ ) as n → +∞ if and only if the sequence
{Πt1 ,t2 f n }n≥1 converges (converges weakly respectively) in Ft1 ,t2 towards Πt1 ,t2 f as
n → +∞ for any finite interval [t1 , t2 ] ⊂ R+ .
94 4 Strongest Convergence Results …

We denote T (h)y(·) = yh (·), where yh (t) = y(t + h) for any y ∈ F loc (R+ ) and
t, h ≥ 0.
In the autonomous case, when f (x, t, y) does not depend on t, the long-time
behavior of all globally defined weak solutions for Problem (4.1) is described by
using trajectory and global attractors theory; Chepyzhov and Vishik [7, Chap. XIII];
Vishik et al. [25]; Melnik and Valero [19]; Kasyanov [11, 12], Zgurovsky et al. [28,
Chap. 2] and references therein; see also Balibrea et al. [2]. In this situation the set
K + := K0+ is translation semi-invariant, i.e. T (h)K + ⊆ K + for any h ≥ 0.
As trajectory attractor it is considered a classical global attractor for translation
semigroup {T (h)}h≥0 , that acts on K + .
In the non-autonomous case we notice that T (h)K0+  K0+ . Therefore, we
need to consider united trajectory space that includes all globally defined on any
[τ, +∞) ⊆ R+ weak solutions of Problem (4.1) shifted to τ = 0:

K∪+ := y( · + τ ) ∈ W loc (R+ ) : y( · ) ∈ Kτ+ , (4.10)

τ ≥0

Note that T (h){y( · + τ ) : y ∈ Kτ+ } ⊆ {y( · + τ + h) : y ∈ Kτ++h } for any

τ, h ≥ 0. Therefore,
T (h)K∪+ ⊆ K∪+

for any h ≥ 0. Further we consider extended united trajectory space for Problem
(4.1) (see Fig. 4.1):  
KF+loc (R+ ) = clF loc (R+ ) K∪+ , (4.11)

Fig. 4.1 The extended united trajectory space construction scheme

4.3 Preliminary Properties of Weak Solutions 95

where clF loc (R+ ) [ · ] is the closure in F loc (R+ ). We note that

T (h)KF+loc (R+ ) ⊆ KF+loc (R+ )

for each h ≥ 0, because

ρF loc (R+ ) (T (h)u, T (h)v) ≤ ρF loc (R+ ) (u, v) for any u, v ∈ F loc (R+ ),

where ρF loc (R+ ) is a standard metric on Fréchet space F loc (R+ ); cf. Vishik, Zelik,
and Chepyzhov [25]; Chepyzhov and Vishik [5]; Vishik et al. [25].

R+ ; H)
4.4 Strongest Convergence Results in C Loc (R

Let us provide the result characterizing the compactness properties of shifted solu-
tions of Problem (4.1) in the induced topology from C loc (R+ ; H ).

Theorem 4.1 Let Assumptions I–IV hold. If {yn }n≥1 ⊂ KC+loc (R+ ;H ) be an arbi-
trary sequence, which is bounded in L ∞ (R+ ; H ), then there exist a subsequence
{yn k }k≥1 ⊆ {yn }n≥1 and an element y ∈ KC+loc (R+ ;H ) such that

Πτ,T yn k − Πτ,T yC([τ,T ];H ) → 0, k → +∞, (4.12)

for any finite time interval [τ, T ] ⊂ (0, +∞). Moreover, for any y ∈ KC+loc (R+ ;H ) the
estimate holds
y(t)2H ≤ y(0)2H e−c3 t + c4 , (4.13)

for any t ≥ 0, where positive constants c3 and c4 do not depend on y ∈ KC+loc (R+ ;H )
and t ≥ 0.

Proof Assume that {yn }n≥1 ⊂ K∪+ be an arbitrary sequence, which is bounded in
L ∞ (R+ ; H ). Let us fix n ≥ 1. Formula (4.10) provides the existence of τn ≥ 0 and
z n ( · ) ∈ Kτn+ such that yn ( · ) = z n ( · + τn ). Then, formulas (4.8) and (4.9) yield that

 t M  t
  t
(i) p
yn (t)2H − 2 βn (x, ξ )d xdξ +2α yn (ξ ) Li (Ω) dξ + 2 yn (ξ )2V dξ
pi
0 Ω i=1 s s
 s
≤ yn (s)2H − 2 βn (x, ξ )d xdξ, (4.14)
0 Ω

 t
yn (t)2H ≤ yn (s)2H e−2dλ1 (t−s) + 2 βn (x, ξ )e−2dλ1 (t−ξ ) d xdξ, (4.15)
s Ω

for any t ≥ s ≥ 0, where βn (x, t) := β(x, t + τn ) for a.e. (x, t) ∈ Ω × R+ .

96 4 Strongest Convergence Results …

Note that formula (4.15) and t.u.i. of β in L loc

1 (R+ ; L 1 (Ω)) provide formula (4.13)
for any y ∈ K∪+ , where positive constants c3 and c4 do not depend on respective y and
t; cf. Chepyzhov and Vishik [7, p. 35]. Formula (4.13) holds for any y ∈ KC+loc (R+ ;H ) ,
because the set K∪+ is dense in KC+loc (R+ ;H ) endowed with strong local convergence
topology of C loc (R+ ; H ). Therefore, the second statement of the theorem (estimate
(4.13)) is proved.
Let us continue the proof of the first statement of the theorem (formula (4.12)).
Further, to simplify arguments we set

dn (x, t) := f (x, t + τn , yn (x, t)) for a.e. (x, t) ∈ Ω × R+ and n ≥ 1.

Estimates (4.14) and (4.15), formula (4.7), t.u.i. of β and c1 in L loc 1 (R+ ; L 1 (Ω)),
and Assumptions III and IV, provide that the sequence {yn , dn }n≥1 is bounded in
W loc (R+ ) × Lq (τ, T ; Lq (Ω)). Banach–Alaoglu theorem (cf. Zgurovsky et al. [27,
Chap. 1]; Kasyanov [11] and references therein) yields that there exist a subsequence
{yn k , dn k }k≥1 ⊆ {yn , dn }n≥1 and elements (y, d) ∈ W loc (R+ )×Lq (τ, T ; Lq (Ω)), and
1 (R+ ; L 1 (Ω)) such that
β̄ ∈ L loc

(yn k , dn k ) → (y, d) weakly in W loc (R+ ) × Lqloc (R+ ; Lq (Ω)),

yn k → y weakly in C loc (R+ ; H ),
yn k → y 2 (R+ ; H ),
in L loc (4.16)
yn k (t) → y(t) in H for a.e. t > 0,
βn k → β̄ 1 (R+ ; L 1 (Ω)), k → +∞.
weakly in L loc

Note that the second convergence holds, because the embedding W loc (R+ ) ⊂
C loc (R+ ; H ) is continuous, the third one follows from the compact embedding of
W loc (R+ ) into L loc
2 (R+ ; H ) (cf. Zgurovsky et al. [27, Chap. 1]), the fourth con-
vergence follows from the third one, and the last statement in (4.16) follows from
Dunford–Pettis compactness criterion.
Let us prove that

yn k (t) → y(t) in H for any t > 0, as k → +∞. (4.17)

We consider continuous and nonincreasing (by formula (4.14)) functions on R+ :

 t  t
Jk (t) = yn (t)2H − 2 βn (x, ξ )d xdξ, J (t) = y(t)2H − 2 β̄(x, ξ )d xdξ, k ≥ 1;
0 Ω 0 Ω
(4.18)
cf. Kapustyan et al. [14]. The fourth and the last statements in (4.16) imply

Jk (t) → J (t), as k → +∞, for a.e. t > 0. (4.19)

Similarly to Zgurovsky et al. [28, p. 57] (see book and references therein) we show
that
4.4 Strongest Convergence Results in C Loc (R + ; H ) 97

lim sup Jk (t) ≤ J (t) ∀t > 0. (4.20)

k→+∞

Indeed, formula (4.19) and continuity of J imply that for any t > 0 and ε > 0 there
exists t¯ ∈ (0, t) such that |J (t¯) − J (t)| < ε and lim Jk (t¯) = J (t¯). Hence,
k→+∞

Jk (t) − J (t) ≤ Jk (t¯) − J (t) ≤ |Jk (t¯) − J (t¯)| + |J (t¯) − J (t)| < ε + |Jk (t¯) − J (t¯)|,

for any k ≥ 1. Therefore, lim sup Jk (t) ≤ J (t) + ε, for each t > 0 and ε > 0. Thus,
k→+∞
inequality (4.20) holds.
Formula (4.20) and last statement of (4.16) yield the inequality

lim sup yn k (t)2H ≤ y(t)2H ∀t > 0.

k→+∞

Convergence (4.17) holds, because of the last inequality and the pointwise weak
convergence in H of the sequence {yn k }k≥1 towards y, as k → +∞ (see the second
statement in (4.16)).
Let us prove (4.12). By contradiction suppose the existence of a positive constant
L > 0, a finite interval [τ, T ] ⊂ (0, +∞), and a subsequence {yk j } j≥1 ⊆ {yn k }k≥1
such that

∀j ≥ 1 max yk j (t) − y(t) H = yk j (t j ) − y(t j ) H ≥ L .

t∈[τ,T ]

Suppose also that t j → t0 ∈ [τ, T ], as j → +∞. Continuity of Πτ,T y : [τ, T ] → H

implies
lim inf yk j (t j ) − y(t0 ) H ≥ L . (4.21)
j→+∞

On the other hand we prove that

yk j (t j ) → y(t0 ) in H, j → +∞. (4.22)

For this purpose we firstly note that

yk j (t j ) → y(t0 ) weakly in H, j → +∞. (4.23)

Indeed, for a fixed h ∈ (C0∞ (Ω)) M from (4.16) it follows that the sequence of real
functions {(Πτ,T yn k ( · ), h) H : [τ, T ] → R}k≥1 is uniformly bounded and equicon-
tinuous. Taking into account the boundedness of {Πτ,T yn k }k≥1 in Wτ,T and the density
of the set (C0∞ (Ω)) M in H we obtain that yn k (t) → y(t) weakly in H uniformly on
[τ, T ], as k → +∞. So, we obtain (4.23).
Secondly we prove that
98 4 Strongest Convergence Results …

lim sup yk j (t j ) H ≤ y(t0 ) H . (4.24)

j→+∞

We consider continuous nonincreasing functions J and Jk j , j ≥ 1, defined in (4.18).

Let us fix an arbitrary ε > 0. Continuity of J and (4.19) provide the existence of
t¯ ∈ (τ, t0 ) such that lim Jk j (t¯) = J (t¯) and |J (t¯) − J (t0 )| < ε. Then,
j→+∞

Jk j (t j ) − J (t0 ) ≤ |Jk j (t¯) − J (t¯)| + |J (t¯) − J (t0 )| ≤ |Jk j (t¯) − J (t¯)| + ε,

for rather large j ≥ 1. Thus, lim sup Jk j (t j ) ≤ J (t0 ) and inequality (4.24) holds.
j→+∞
Thirdly note that the convergence (4.22) holds, because of (4.23) and (4.24); cf.
Gajewski et al. [3, Chap. I]. Finally we remark that statement (4.22) contradicts
to assumption (4.21). Therefore, the first statement of the theorem holds for any
sequence {yn }n≥1 ⊂ K∪+ .
To finish the proof of the theorem we consider the first statement in the general
case. Let {yn }n≥1 ⊂ KC+loc (R+ ;H ) be an arbitrary sequence, which is bounded in
L ∞ (R+ ; H ). Since the set K∪+ is dense in a Polish space KC+loc (R+ ;H ) we have that
for any n ≥ 1 there exists ȳn ∈ K∪+ such that ρC loc (R+ ;H ) (yn , ȳn ) ≤ n1 . A priori
estimate (4.13) provides that the sequence { ȳn }n≥1 is bounded in L ∞ (R+ ; H ). The
first statement of the theorem, applied for the sequence { ȳn }n≥1 ⊂ K∪+ , yields that
there exist a subsequence { ȳn k }k≥1 ⊂ { ȳn }n≥1 and an element y ∈ KC+loc (R+ ;H ) such
that Πτ,T ȳn k − Πτ,T yC loc ([τ,T ];H ) → 0, as k → +∞, for any finite time interval
[τ, T ] ⊂ (0, +∞). Therefore, formula (4.12) holds for any [τ, T ] ⊂ (0, +∞).

4.5 Strongest Convergence Results for Solutions

in the Natural Extended Phase Space

For convergence results in the strong topology of the natural extended phase
space W loc (R+ ) it is necessary to claim that additional assumption holds (see
Example 8.1). To formulate this additional assumption we provide some auxil-
iary constructions. A function ϕ ∈ L loc 1 (R+ ; L 1 (Ω)) is called translation-compact
(tr.-c.) in L loc
1 (R + ; L 1 (Ω)), if the set {ϕ( · + h) : h ≥ 0} is precompact in
L loc
1 (R+ ; L 1 (Ω)); cf. Chepyzhov and Vishik [5, p. 917]. Note that a function
ϕ ∈ L loc (R ;
+ 1L (Ω)) is tr.-c. in L loc
1 +(R ; L 1 (Ω)) if and only if two conditions
1
 t+h
hold: a) the set t ϕ(s)ds : t ≥ 0 is precompact in L 1 (Ω) for any h > 0; b)
there exists a function ψ(s), ψ(s) → 0+ as s → 0+ such that
 t+1 
|ϕ(x, s) − ϕ(x, s + h)|d xds ≤ ψ(|h|) for any t ≥ 0 and h ≥ −t;
t Ω

Chepyzhov and Vishik [5, Proposition 6.5].

4.5 Strongest Convergence Results for Solutions in the Natural … 99

Assumption V. Let the conditions hold:

(i) the functions c1 and β from Assumptions (III) and (IV) respectively are tr.-c.
1 (R
in L loc  +; L 1 (Ω)); 
t+h
(ii) the set h1 t f ( · , s, u)ds : t ≥ 0, h ∈ (0, h 0 ), uR M ≤ R is precompact
in (L 1 (Ω)) M for any R > 0 and some h 0 = h 0 (R) > 0;
(iii) for any r > 0 there exist a nondecreasing function ψ(s, r ) : R2+ → R+ ,
ψ(s, r ) → 0+ as s → 0+, and h 0 = h 0 (r ) > 0 such that

M    
1  t+h 1  (i) 
 f (x, s, u) − f (i) (x, s + h 2 , v) d xds ≤ ψ(|h 2 | + u − vR M , r )
h1 t Ω
i=1

for each t ≥ 0, h 1 ∈ (0, h 0 ), h 2 ≥ −t, and u, v ∈ R M such that

uR M , vR M ≤ r .

Remark 4.1 Let us discuss sufficient conditions for Assumption V.

(i) The autonomous case. Let f does not depend on the time variable t and it sat-
isfies Assumptions I–IV with c1 , β ∈ L 1 (Ω) (in particular, assumptions from Vishik
et al. [25] hold). Then Assumption V hold. Indeed, Assumptions V(i) holds, because
c1 and β do not depend on t; Assumptions II, III and the dominated convergence
theorem imply Assumption V(ii). Assumption  V(iii) follows from Heine–Cantor
theorem and continuity of the mapping u → Ω f (x, u)d x. The last follows from
the dominated convergence theorem and Assumptions I–III.
(ii) The non-autonomous case. Let f = f (t, u) is jointly continuous mapping,
it satisfies Assumptions I–IV with positive constants c1 and β, and f being tr.-c. in
C loc (R+ ; C(R M )), that is

 f (t, u) − f (s, v) R M ≤ ω (|t − s| + u − vR M , K ) ,

for all t, s ∈ R+ , uR M , vR M ≤ K , K > 0, where ω (l, K ) → 0, as l → 0+;

see, for example, Chepyzhov and Vishik [7, p.105], Kapustyan and Valero [16, 24],
where uniform global in H and uniform trajectory in C loc (R+ ; H ) attractors were
investigated. Then Assumption V holds.
(iii) The sufficient condition for Assumption V(iii) is: for any r > 0 there exist a
nondecreasing function ψ(s, r ) : R2+ → R+ , ψ(s, r ) → 0+ as s → 0+, such that

M 
  (i) 
 f (x, t, u) − f (i) (x, t + h, v) d xds ≤ ψ(|h| + u − vR M , r )
i=1 Ω

for each t ≥ 0, h ≥ −t, and u, v ∈ R M such that uR M , vR M ≤ r .

Note that Assumption V is a generalization of the above assumptions to the
case when f depends on the space, time and state variables simultaneously and
it is not necessarily continuous by t. Meanwhile, Example 8.1 below provide piece-
wise continuous function f that satisfies Assumptions I–IV, but it does not satisfy
100 4 Strongest Convergence Results …

Assumption V. Moreover, the statement of Theorem 4.2 below does not hold for
Problem (4.1) with such interaction function.
Now let us provide the result characterizing the compactness properties of shifted
solutions of Problem (4.1) in the induced topology from W loc (R+ ).
Theorem 4.2 Let Assumptions I–V hold. If {yn }n≥1 ⊂ KW+loc (R+ ) be an arbi-
trary sequence, which is bounded in L ∞ (R+ ; H ), then there exist a subsequence
{yn k }k≥1 ⊆ {yn }n≥1 and an element y ∈ KW+loc (R+ ) such that

Πτ,T yn k − Πτ,T yWτ,T → 0, as k → +∞, (4.25)

for any finite time interval [τ, T ] ⊂ (0, +∞). Moreover, for any y ∈ KW+loc (R+ ) the
estimate (4.13) holds for any t ≥ 0, where positive constants c3 and c4 do not depend
on y ∈ KW+loc (R+ ) and t ≥ 0.
Proof The embedding KW+loc (R+ ) ⊆ KC+loc (R+ ;H ) and Theorem 4.1 yield the second
statement of the theorem. Let us provide the first one.
We consider an arbitrary sequence {yn }n≥1 ⊂ KW+loc (R+ ) , which is bounded in
L ∞ (R+ ; H ). Since the set K∪+ is dense in a Polish space KW+loc (R+ ) and the estimate
(4.13) holds for any y ∈ KW+loc (R+ ) , there exists a sequence { ȳn }n≥1 ⊂ K∪+ , which
is bounded in L ∞ (R+ ; H ) and ρW loc (R+ ) (yn , ȳn ) ≤ n1 for any n ≥ 1. Therefore,
to provide the first statement of the theorem, we may additionally suppose that the
sequence {yn }n≥1 belongs to K∪+ .
Note that Assumptions III, IV, and V, and Young’s inequality yield that there exist
positive constants α  , β  > 0 and a tr.-c. in L loc 
1 (R+ ; L 1 (Ω)) function c : Ω ×R+ →
R+ such that


M 
M
 (i)  pi 
M
 (i)  pi
f (i) (x, t, u)(u (i) − v(i) ) ≥ α  u  − β  v  − c (x, t), (4.26)
i=1 i=1 i=1

for any u, v ∈ R M and a.e. (x, t) ∈ Ω × (0, +∞). Let H (c ) := cl L loc 1 (R+ ;L 1 (Ω))
{c ( · + h) : h ≥ 0} be the hull of tr.-c. function c in L loc1 (R+ ; L 1 (Ω)). This is a
compact set in L loc
1 (R+ ; L 1 (Ω)); Chepyzhov and Vishik [5].
Let us fix an arbitrary n ≥ 1. Formula (4.10) provides the existence of τn ≥
0 and z n ( · ) ∈ Kτn+ such that yn ( · ) = z n ( · + τn ). Following to the statement
of Theorem 4.1 and its proof (see formula (4.16) and conclusions above it), there
exist a subsequence {yn k , dn k }k≥1 ⊆ {yn , dn }n≥1 and elements (y, d) ∈ W loc (R+ ) ×
Lq (τ, T ; Lq (Ω)), and β̄ ∈ L loc 1 (R+ ; L 1 (Ω)) such that convergences (4.16) and
(4.12) hold. Here we again use the notation:

dn (x, t) := f (x, t + τn , yn (x, t)) for a.e. (x, t) ∈ Ω × R+ and n ≥ 1.

Since the sets H (c1 ) and H (c ) are compact in L loc

1 (R+ ; L 1 (Ω)), taking into
account the third statement of (4.16), we may additionally claim (passing to a
4.5 Strongest Convergence Results for Solutions in the Natural … 101

subsequence if necessary) the existence of elements c̄1 and c̄ from L loc
1 (R+ ; L 1 (Ω))
such that

yn k (x, t) → y(x, t)
c1,n k (x, t) → c̄1 (x, t),
cn k (x, t) → c̄ (x, t), as k → +∞, for a.e. (x, t) ∈ Ω × R+ ; (4.27)
cn k → 
c̄ ,
c1,n k → c̄1 , 1 (R+ ; L 1 (Ω)), as k → +∞,
in L loc

where c1,n k := c1 (x, t + τn k ) and cn k (x, t) := c (x, t + τn k ) for a.e. (x, t) ∈ Ω × R+
and any k ≥ 1.
Assumption III yields that


M
 (i)  
M
 (i) 
d (x, t)qi ≤ c1,n (x, t) + c2  y (x, t) pi (4.28)
nk k nk
i=1 i=1

for a.e. (x, t) ∈ Ω × R+ and any k ≥ 1. Therefore, the first two statements of (4.27)
provide

dn k (x, t) · (yn k − y)(x, t) → 0 as k → +∞, for a.e. (x, t) ∈ Ω × R+ . (4.29)

Now let fix an arbitrary finite time interval [τ, T ] ⊂ (0, +∞). Prove that

Πτ,T yn k − Πτ,T y2X τ,T = Πτ,T yn k − Πτ,T y2L 2 (τ,T ;V )

(4.30)
+Πτ,T yn k − Πτ,T y2Lp (τ,T ;Lp (Ω)) → 0, as k → +∞.

Formulas (4.7) and (4.5) yield

(yn k (τ ), yn k (τ ) − y(τ )) H − (yn k (T ), yn k (T ) − y(T )) H

 T  T
= a∇ yn k (x, t) · ∇(yn k − y)(x, t)d xdt + dn k (x, t) · (yn k − y)(x, t)d xdt,
τ Ω τ Ω
(4.31)

for any k ≥ 1. Formula (4.12) provides

(yn k (τ ), yn k (τ ) − y(τ )) H − (yn k (T ), yn k (T ) − y(T )) H → 0, as k → +∞. (4.32)

The first statement of (4.16) implies

 T
lim inf a∇ yn k (x, t) · ∇(yn k − y)(x, t)d xdt
k→+∞ τ Ω
 T  T
= lim inf a∇ yn k (x, t) · ∇ yn k (x, t)d xdt − a∇ y(x, t) · ∇ y(x, t)d xdt ≥ 0,
k→+∞ τ Ω τ Ω
(4.33)
102 4 Strongest Convergence Results …

and
     
T M
 (i)  T M
 (i) 
lim inf  y (x, t) pi d xdt ≥  y (x, t) pi d xdt, (4.34)
nk
k→+∞ τ Ω i=1 τ Ω i=1

because the sequence {Πτ,T yn k }k≥1 converges weakly to Πτ,T y in X τ,T as k → +∞.
Therefore, formulas (4.31)–(4.33) yield
 T 
lim sup dn k (x, t) · (yn k − y)(x, t)d xdt ≤ 0. (4.35)
k→∞ τ Ω

Let us apply Fatou’s lemma to the sequence {ψk }k≥1 of Lebesgue integrable non-
negative (see formula (4.26)) functions


M
 (i) 
ψk (x, t) := dn k (x, t) · (yn k − y)(x, t) − α  y (x, t) pi 
nk
i=1


M
 (i) 
+β   y (x, t) pi + c (x, t), (x, t) ∈ Ω × R+ , k ≥ 1.
n,k
i=1

We obtain
 T  
M   pi  T  
M   pi  T 
 (i)   (i) 
− α y (x, t) d xdt + β  y (x, t) d xdt + c̄ (x, t)d xdt
τ Ω i=1 τ Ω i=1 τ Ω
   M 

T   pi M 
  pi
 (i)   (i) 
= lim inf dn k (x, t) · (yn k − y)(x, t) − α  yn k (x, t) + β  y (x, t) + c̄n k (x, t) d xdt
τ Ω k→∞ i=1 i=1
    
M 
T T  pi
 (i) 
≤ lim inf dn k (x, t) · (yn k − y)(x, t)d xdt − α  yn k (x, t) d xdt
k→∞ τ Ω τ Ω i=1
  
M    
T  pi T
  (i) 
+β y (x, t) d xdt + cn k (x, t)d xdt
τ Ω i=1 τ Ω
 T  
M   pi  T  
M   pi  T 
 (i)   (i) 
≤ − α  lim sup yn k (x, t) d xdt + β  y (x, t) d xdt + c̄ (x, t)d xdt,
k→∞ τ Ω i=1 τ Ω i=1 τ Ω

where the equality follows from the first and third statements of (4.27) and for-
mula (4.29); the first inequality follows from Fatou’s lemma applied to the sequence
{ψk }k≥1 ; the second inequality follows from (4.35) and the last statement of (4.27).
Therefore, due to inequality (4.34), we have
     
T M
 (i)  T M
 (i) 
 y (x, t) pi d xdt →  y (x, t) pi d xdt, as k → ∞.
nk
τ Ω i=1 τ Ω i=1
(4.36)
Moreover,
4.5 Strongest Convergence Results for Solutions in the Natural … 103
 T 
lim inf dn k (x, t) · (yn k − y)(x, t)d xdt ≥ 0. (4.37)
k→∞ τ Ω

Inequalities (4.35) and (4.37) provide

 T 
dn k (x, t) · (yn k − y)(x, t)d xdt → 0, as k → ∞. (4.38)
τ Ω

Passing to the limit as k → +∞ in formula (4.31), taking into account (4.32) and
(4.38), we obtain
 T  T
a∇ yn k (x, t) · ∇ yn k (x, t)d xdt → a∇ y(x, t) · ∇ y(x, t)d xdt, as k → ∞.
τ Ω τ Ω
(4.39)

Statement (4.30) holds, because the sequence {Πτ,T yn k }k≥1 converges weakly to
Πτ,T y in the uniformly convex (superreflexive) Banach space X τ,T = L 2 (τ, T ; V ) ∩
Lp (τ, T ; Lp (Ω)) as k → +∞ (see the first statement of (4.16)), and statements
(4.36) and (4.39) hold. We note that the mapping

 T 
z→ a∇z(x, t) · ∇z(x, t)d xdt
τ Ω

defines a norm, that is equivalent to the natural one, defined on Hilbert space
L 2 (τ, T ; V ). Statement (4.30) is proved.
To finish the proof of the theorem we provide that there exist a subsequence
{ykm }m≥1 ⊆ {yn k }k≥1 such that Πτ,T ∂t∂ ykm − Πτ,T ∂t∂ y L 2 (τ,T ;V ∗ )+Lq (τ,T ;Lq (Ω)) → 0,
as m → +∞, for any finite time interval [τ, T ] ⊂ (0, +∞). Since yn k → y in
∗
2 (R+ ; V ), as k → +∞ (see formula (4.30)), it is sufficient to prove that for any
L loc
finite time interval [τ, T ] ⊂ (0, +∞) the sequence {Πτ,T dn k }k≥1 is precompact in
Lq (τ, T ; Lq (Ω)) (see Problem (4.1)).
On the contrary assume that {Πτ,T dn k }k≥1 is not precompact in Lq (τ, T ; Lq (Ω))
for some finite time interval [τ, T ] ⊂ (0, +∞). Therefore, there exist a subsequence
of {Πτ,T dn k }k≥1 (we denote it again by {Πτ,T dn k }k≥1 ), a finite time interval [τ, T ] ⊂
(0, +∞), and ε∗ > 0 such that

Πτ,T dn k − Πτ,T dn m Lq (τ,T ;Lq (Ω)) ≥ ε∗ , for any k, m ≥ 1. (4.40)

The last statement of (4.27), statements (4.28) and (4.30), and dominated conver-
gence theorem yield
104 4 Strongest Convergence Results …

T  
M  qi
 (i) 
lim sup dn k (x, t) χ Ak,K (x, t)d xdt
K →+∞ k≥1
τ Ω i=1
T   M 
  pi

 (i) 
≤ lim sup c1,n k (x, t) + c2 yn k (x, t) χ Ak,K (x, t)d xdt = 0,
K →+∞ k≥1
τ Ω i=1

where Ak,K := {(x, t) ∈ Ω × (τ, T ) : yn k (x, t)R M ≥ K }, K > 0, k ≥ 1.

Therefore, without loss of generality, we may additionally assume that there exists
a rather large K > 0 such that

yn k (x, t) R M ≤ K for a.e. (x, t) ∈ Ω × (τ, T ), and any k ≥ 1. (4.41)

Krasnosel’skii [17, Chap. 1] (see book and references therein) implies that for any
k = 1, 2, . . . , there exists a simple function z k : Ω × (τ, T ) → R M ,


Nk
z k (x, t) = b j,k χ B j,k (x, t) for a.e. (x, t) ∈ Ω × (τ, T ),
j=1

where Nk ≥ 1, {b j,k } Nj=1

k Nk
⊂ R M , {B j,k }k=1 ⊂ Ω × (τ, T ) be a family of disjoint
measurable sets, such that z k (x, t)R M ≤ K , for a.e. (x, t) ∈ Ω × (τ, T ), and

T  
M  qi
 (i)  1
Πτ,T yn k − z k Lp (τ,T ;Lp (Ω)) + dn k (x, t) − f (i) (x, t + τn k , z k (x, t)) d xdt ≤ .
k
τ Ω i=1
(4.42)

For any h ∈ (0, h 0 ), where h 0 = h 0 (K ) be a positive constant from Assumption

V, let us define the mapping Fh : Ω × R+ × R M → R M ,

t+h
1
Fh (x, t, u) := f (x, s, u)ds, (x, t, u) ∈ Ω × R+ × R M .
h
t

Assumption V(iii) yields that

T  
M  
 (i) (i) 
 f (x, t + τn k , z k (x, t)) − Fh (x, t + τn k , z k (x, t)) d xdt ≤ ψ(|h|, K )(T − τ )Nk ,
τ Ω i=1

for any k ≥ 1 and h ∈ (0, h 0 ). Since, ψ(s, K ) → 0+ as s → 0+, for each k ≥ 1

there exists h k ∈ (0, h 0 ) such that

1
ψ(|h k |, K )(T − τ )Nk ≤ .
k
4.5 Strongest Convergence Results for Solutions in the Natural … 105

Thus,

T  M  
 (i) (i)  1
 f (x, t + τn k , z k (x, t)) − Fh k (x, t + τn k , z k (x, t)) d xdt ≤ , (4.43)
i=1
k
τ Ω

for any k ≥ 1.
Assumptions V(ii) and V(iii) and Arzelà–Ascoli theorem (see Warga [26, Chap. I]
and references therein) provide the existence of a mapping G ∈ C([τ, T ] ×
B̄ K ; (L 1 (Ω)) M ), where B̄ K := {u ∈ R M : uR M ≤ K }, and a subsequence
{(t, u) → Fh km ( · , t + τkm , u)}m≥1 ⊆ {(t, u) → Fh k ( · , t + τn k , u)}k≥1 such that

 
M  
 (i)  1
sup Fh k (x, t + τkm , u) − G (i) (x, t, u) d x ≤ , for any m ≥ 1.
t∈[τ,T ], uR M ≤K m m Nkm
Ω i=1

Therefore,

T  
M  
 (i)  T −τ
Fh km (x, t + τkm , z km (x, t)) − G (i) (x, t, z km (x, t)) d xdt ≤ , for any m ≥ 1.
m
τ Ω i=1
(4.44)

Since G ∈ C([τ, T ] × B̄ K ; (L 1 (Ω)) M ), Heine–Cantor theorem provides the exis-

tence of nondecreasing function Ψ : R+ → R+ , Ψ (s) → 0+, as s → 0+, such
that
 
M
 (i) 
G (x, t1 , u 1 ) − G (i) (x, t2 , u 2 ) d x ≤ Ψ (|t1 − t2 | + u 1 − u 2 R M ),
Ω i=1

for any t1 , t2 ∈ [τ, T ] and u 1 , u 2 ∈ B̄ K . Therefore,

 

M
 (i) 
C · meas (x, t) ∈ Ω × (τ, T ) : G (x, t, u 1 ) − G (i) (x, t, u 2 ) ≥ C
i=1
T  
M
 (i) 
≤ G (x, t, u 1 ) − G (i) (x, t, u 2 ) d xdt ≤ (T − τ )Ψ (u 1 − u 2 R M )
τ Ω i=1
(4.45)

for any C > 0 and u 1 , u 2 ∈ B̄ K , where meas( · ) is a standard Lebeasgue measure

on Ω × (τ, T ).
We note that the sequence of mappings {(x, t) → G(x, t, z km (x, t))}m≥1 , defined
on Ω × (τ, T ), converges in measure towards the mapping (x, t) → G(x, t, y(x, t))
as m → +∞, i.e. for any C > 0 and ε > 0 there exists M̄ ≥ 1 such that
106 4 Strongest Convergence Results …

meas (Am,C ) ≤ ε, for each m ≥ M̄, where

⎧ ⎫
⎨ M 
  ⎬
 (i) 
Am,C := (x, t) ∈ Ω × (τ, T ) : G (x, t, z km (x, t)) − G (i) (x, t, y(x, t)) ≥ C .
⎩ ⎭
i=1

Indeed, there exists δ > 0 such that Ψ (δ)(T − τ ) ≤ Cε2

. Therefore, meas(Am,C \
{(x, t) ∈ Ω × (τ, T ) : z km (x, t) − y(x, t)R M ≥ δ}) ≤ 2ε for any m ≥ 1; see
formula (4.45). Since z km (x, t) → y(x, t) as m → +∞ for a.e. (x, t) ∈ Ω × (τ, T )
(see (4.42) and (4.27)), we obtain the necessary statement.
Since the sequence of mappings {(x, t) → G(x, t, z km (x, t))}m≥1 , defined on
Ω × (τ, T ), converges in measure towards the mapping (x, t) → G(x, t, y(x, t))
as m → +∞, inequalities (4.42)–(4.44) yield that the sequence {Πτ,T dkm }m≥1 con-
verges in measure towards the mapping (x, t) → G(x, t, y(x, t)) as m → +∞.
Thus, formulas (4.27), (4.28), (4.41) and dominated convergence theorem yield that

T  
M  qi
 (i) 
lim dkm (x, t) − G (i) (x, t, y(x, t)) d xdt = 0.
m→+∞
τ Ω i=1

This is a contradiction with (4.40).

4.6 Examples of Applications

As applications we may consider the following examples: FitzHugh-Nagumo sys-

tem (signal transmission across axons), complex Ginzburg–Landau equation (the-
ory of superconductivity), Lotka–Volterra system with diffusion (ecology mod-
els), Belousov-Zhabotinsky system (chemical dynamics) and many other reaction-
diffusion type systems (see Smoller [22]), whose dynamics are well studied in
autonomous case (see Temam [23], Chepyzhov and Vishik [7]) and in non-
autonomous case, when all coefficients are uniformly continuous on time variable
(see Chepyzhov and Vishik [7], Zgurovsky et al. [28] and references therein). Now
results of Theorems 4.1 and 4.2 allow us to study these systems with Carathéodory’s
nonlinearities.

4.6.1 Non-autonomous Complex Ginzburg–Landau

Equation

Let Ω ⊂ R N be an open bounded set with smooth boundary ∂Ω. Consider the
non-autonomous complex Ginzburg–Landau equation:
4.6 Examples of Applications 107
 ∂u
∂t
= (1 + ηi) Δu + R (t) u − (1 + iβ (t)) |u|2 u + g (x, t) ,
(4.46)
u |∂Ω = 0, u (x, τ ) = u τ (x) ,

where u = u (x, t) = u 1 (x, t) + iu 2 (x, t), (x, t) ∈ Ω × R+ ; g (t) = g 1 (t) +

g 2 (t) i ∈ L 2 (Ω; C) for a.e. t > 0; η, β (t) ∈ R; and R (t) > 0 for a.e. t > 0. We
 t+1
assume that g i ∈ L loc
2 (R+ ; L 2 (Ω)) with sup t>0 t g i ( · , s)2L 2 (Ω) ds < +∞ and
 R (t) and β (t) are measurable and essentially bounded.
also that the functions
For v = u 1 , u 2 , u = u 1 + iu 2 , Eq. (4.46) can be writen as the system
⎛ #   2 $  1 ⎞ 
 R (t) u 1 − u 1  + u 2 
2
u − β (t) u 2
∂v 1 −η ⎠ + g 2 (t, x)
1
= Δv+ ⎝ #    $ 
∂t η 1 R (t) u 2 − u 1  + u 2 
2 2
β (t) u 1 + u 2 g (t, x)

and Assumptions I–IV hold with p = (4, 4). Indeed, since

    
f (t, v) = −R (t) u 1 + |v|2 u 1 − β (t) u 2 , −R (t) u 2 + |v|2 β (t) u 1 + u 2 ,

then the Young’s inequality yields that

 1 4  4 # #  4   4 $
 f (t, v) 3 +  f 2 (t, v) 3 ≤ K 1 |R (t)| 43 u 1  3 + u 2  3

# $ #  4   4 $$ #  4  4 $
+ |v| 3 1 + |β (t)| 3 u 1  3 + u 2  3 ≤ K 2 u 1  + u 2  + K 3 ,
8 4

because R (t), β (t) are essentially bounded in R. Moreover,

 1 4  2 4
|v|4 u  + u 
( f (t, v) , v) = −R (t) |v| + |v| ≥
2 4
− K4 ≥ − K4.
2 2
Hence, all statements of Theorem 4.1 hold. Furthermore, if the functions R (t) and
β (t) satisfy

|β (t) − β (s)| ≤ a (|t − s|) , |R (t) − R (s)| ≤ b (|t − s|) , (4.47)

for all t, s ∈ R, where a (l) → 0, b (l) → 0, as l → 0+ , then, additionally,

Assumption V holds and, thus, all statements of Theorem 4.2 hold.

4.6.2 Non-autonomous Lotka–Volterra System with Diffusion

Let Di be positive constants, Ω ⊂ R3 be an open bounded subset with sufficiently

smooth boundary ∂Ω, and ai (t) , ai j (t) be positive measurable and bounded func-
tions on R+ . Consider the Lotka–Volterra system with diffusion:
108 4 Strongest Convergence Results …
⎧ ∂u 1  
⎨ ∂t2 = D1 Δu + u a1 (t) − u − a12 (t) u − a13 (t) u  ,
⎪ 1 1 1 2 3
∂u
∂t
= D2 Δu 2 + u 2 a2 (t) − u 2 − a21 (t) u 1 − a23 (t) u 3 , (4.48)
⎩ ∂u 3 = D Δu 3 + u 3 a (t) − u 3 − a (t) u 1 − a (t) u 2  ,
⎪
∂t 3 3 31 32

with Neumann boundary conditions ∂u | = ∂u ∂u 3

1 2

∂ν ∂Ω
| =
∂ν ∂Ω ∂ν
|∂Ω = 0, where u i =
u (x, t) ≥ 0. In this case the function f is given by
i

⎛  ⎞
−u 1 a1 (t) − u 1 − a12 (t) u 2 − a13 (t) u 3
⎜  ⎟
f (t, u) = ⎝ −u 2 a2 (t) − u 2 − a21 (t) u 1 − a23 (t) u 3 ⎠ .
 
−u 3 a3 (t) − u 3 − a31 (t) u 1 − a32 (t) u 2

Then Assumptions I–IV hold for u ∈ R3+ with p := (3, 3, 3). Indeed, since u i ≥ 0,
then the Young’s inequality implies that
 3  3  3  2  2  2
( f (t, u) , u) ≥ u#1 + u 2 + u 3 − a$1 (t) u 1 − a2 (t) u 2 − a3 (t) u 3
 3  3  3
≥ 21 u 1 + u 2 + u 3 − K1,

and


3
3
# 3  3  3   3   3   3
| f i (t, u)| 2 ≤ K 2 u1 + u2 + u3 + u1 2 + u2 2 + u3 2
i=1

 3  3  3 $ # 3  3  3 $
+ u1u2 2 + u2u3 2 + u1u3 2 ≤ K3 u1 + u2 + u3 + K4.

Hence, all statements of Theorem 4.1 hold. Furthermore, if the functions ai (t) and
ai j (t) satisfy (4.47), then, additionally, Assumption V holds and, thus, all statements
of Theorem 4.2 hold.

References

1. Babin, A.V., Vishik, M.I.: Attractors of Evolution Equations. Nauka, Moscow (1989). [in
Russian]
2. Balibrea, F., Caraballo, T., Kloeden, P.E., Valero, J.: Recent developments in dynamical sys-
tems: three perspectives. Int. J. Bifurc. Chaos (2010). doi:10.1142/S0218127410027246
3. Gajewski, H., Gröger, K., Zacharias, K.: Nichtlineare operatorgleichungen und operatordiffer-
entialgleichungen. Akademie-Verlag, Berlin (1978)
4. Chepyzhov V.V., Vishik M.I.: Trajectory attractors for evolution equations. C.R.Acad.Sci. Paris.
Serie I, vol. 321, pp. 1309–1314 (1995)
5. Chepyzhov, V.V., Vishik, M.I.: Evolution equations and their trajectory attractors. J. Math.
Pures Appl. 76, 913–964 (1997)
6. Chepyzhov, V.V., Vishik, M.I.: Trajectory and global attractors for 3D Navier-Stokes system.
Mat. Zametki. (2002). doi:10.1023/A:1014190629738
References 109

7. Chepyzhov, V.V., Vishik, M.I.: Attractors for equations of mathematical physics. American
Mathematical Society, Providence RI (2002)
8. Chepyzhov, V.V., Vishik, M.I.: Trajectory attractor for reaction-diffusion system with diffusion
coefficient vanishing in time. Discret. Contin. Dyn. Syst. 27(4), 1498–1509 (2010)
9. Gorban, N.V., Kapustyan, O.V., Kasyanov, P.O.: Uniform trajectory attractor for non-
autonomous reaction-diffusion equations with Caratheodory’s nonlinearity. Nonlinear Anal.
Theory Methods Appl. 98, 13–26 (2014). doi:10.1016/j.na.2013.12.004
10. Hale, J.K.: Asymptotic Behavior of Dissipative Systems. AMS, Providence, RI (1988)
11. Kasyanov, P.O.: Multi-valued dynamics of solutions of autonomous operator differential equa-
tions with pseudomonotone nonlinearity. Mathematical Notes 92, 205–218 (2012)
12. Kasyanov, P.O.: Multi-valued dynamics of solutions of an autonomous differential-operator
inclusion with pseudomonotone nonlinearity. Cybern. Syst. Anal. 47, 800–811 (2011)
13. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Regularity of Weak Solutions and Their
Attractors for a Parabolic Feedback Control Problem. Set-Valued Var. Anal. (2013). doi:10.
1007/s11228-013-0233-8
14. Kapustyan, O.V., Melnik, V.S., Valero, J.: Attractors of multi-valued dynamical processes,
generated by phase-field equations. Int. J. Bifurc. Chaos 6, 34–45 (2003)
15. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Pullback attractors for a class of extremal solutions
of the 3D Navier-Stokes equations. J. Math. Anal. Appl. (2011). doi:10.1016/j.jmaa.2010.07.
040
16. Kapustyan, A.V., Valero, J.: On the Kneser property for the complex Ginzburg-Landau equation
and the Lotka-Volterra system with diffusion. J. Math. Anal. Appl. 357, 254–272 (2009)
17. Krasnosel’skii, M.A.: Topological Methods in the Theory of Nonlinear Integral Equations.
International Series of Monographs on Pure and Applied Mathematics, vol. 45, pp. X + 395 S.
Oxford (1964)
18. Ladyzhenskaya, O.A.: Attractors for Semigroups and Evolution Equations. Cambridge Uni-
versity Press, Cambridge (1991)
19. Melnik, V.S., Valero, J.: On attractors of multi-valued semi-flows and generalized differential
equations. Set-Valued Anal. 6(1), 83–111 (1998)
20. Mel’nik, V.S., Valero J.: On global attractors of multi-valued semiprocesses and nonau-
tonomous evolution inclusions. Set-Valued Anal. (2000). doi:10.1023/A:1026514727329
21. Sell, G.R.: Global attractors for the three-dimensional Navier-Stokes equations. J. Dyn. Diff.
Eq. 8(12), 1–33 (1996)
22. Smoller, J.: Shock Waves and Reaction-Diffusion Equations (Grundlehren der mathematischen
Wissenschaften). Springer, New York (1983)
23. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics, vol. 68 Appl.
Math. Sci. Springer, New York (1988)
24. Valero, J., Kapustyan, A.V.: On the connectedness and asymptotic behaviour of solutions of
reaction-diffusion systems. J. Math. Anal. Appl. (2006). doi:10.1016/j.jmaa.2005.10.042
25. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong trajectory attractor for a dissipative reaction-
diffusion system. Doklady Math. 82(3), 869–873 (2010)
26. Warga J.: Optimal Control of Differential and Functional Equations. Academic Press, Cam-
bridge (1972)
27. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation Inequali-
ties for Earth Data Processing II. Springer, Berlin (2011)
28. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
Inclusions and Variation Inequalities for Earth Data Processing III. Springer, Berlin (2012)
29. Zgurovsky, M.Z., Kasyanov, P.O.: Zadoianchuk (Zadoyanchuk) N.V.: Long-time behavior of
solutions for quasilinear hyperbolic hemivariational inequalities with application to piezoelec-
tricity problem. Appl. Math. Lett. 25(10), 1569–1574 (2012)
Chapter 5
Strongest Convergence Results for Weak
Solutions of Feedback Control Problems

Abstract In this chapter we establish strongest convergence results for weak solu-
tions of feedback control problems. In Sect. 5.1 we set the problem. Section 5.2
devoted to the regularity of all weak solutions and their additional properties. In
Sect. 5.3 we consider convergence of weak solutions results in the strongest topolo-
gies. As examples of applications we consider a model of combustion in porous
media; a model of conduction of electrical impulses in nerve axons; and a climate
energy balance model.

5.1 Setting of the Problem

Let Ω ⊂ Rn , n ≥ 1, be bounded and open subset with a smooth boundary ∂Ω, f ,

f : R → R are some real functions. We consider the semilinear reaction-diffusion
inclusion

u t − u + [ f (u), f (u)]  0 in Ω × (τ, T ), (−∞ < τ < T < +∞), (5.1)

with boundary condition 

u ∂Ω = 0, (5.2)

where [a, b] = {αa + (1 − α)b | α ∈ [0, 1]}, a, b ∈ R. We suppose that f = [ f , f ] :

R → 2R \ {∅} satisfies the growth condition

∃c0 > 0 : −c0 (1 + |u|) ≤ f (u) ≤ f (u) ≤ c0 (1 + |u|) ∀u ∈ R, (5.3)

and the sign condition

f (u) f (u)
lim > −λ1 ; lim > −λ1 , (5.4)
u→+∞ u u→−∞ u

© Springer International Publishing AG 2018 111

M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_5
112 5 Strongest Convergence Results for Weak Solutions…

where λ1 is the first eigenvalue of − in H01 (Ω). Suppose also that f is lower
semi-continuous, and f is upper semi-continuous (see Sect. 2.1).
We shall use the following standard notations: H = L 2 (Ω), V = H01 (Ω), V is
the dual space of V and ·, · denotes the pairing in the space V .

5.2 Regularity of All Weak Solutions and Their Additional

Properties

Further by · E we denote the norm in a real Banach space E. Assumption (5.4) is
equivalent to the next one

∃λ ∈ (0, λ1 ), ∃c1 > 0 : f (u) · u ≥ −λu 2 − c1 ∀u ∈ R. (5.5)

Sign condition (5.5), the variational characterization of λ1 , and Gronwall-Bellman

inequality imply that for any τ < T and for any weak solution u(·) of Problems (5.1)
and (5.2) on [τ, T ] we have
∗ c2
u(t)2H ≤ u(s)2H e−2ε (t−s)
+ ∀τ ≤ s ≤ t ≤ T, (5.6)
ε∗

where ε∗ = λ1 − λ and c2 = c1 · meas(Ω) (cf. [56, p. 56]).

We note that the mapping v → v H defines an equivalent norm on V ∩ H 2 (Ω)
(cf. [42, Chapter III]). The next theorem provides additional a priory estimates for
all weak solutions of Problems (5.1) and (5.2).
Theorem 5.1 There exists C > 0 such that for any τ < T each weak solution
u(·) of Problems (5.1) and (5.2) on [τ, T ] belongs to C([τ + ε, T ]; V ) ∩ L 2 (τ +
ε, T ; D(A)) and du
dt
(·) ∈ L 2 (τ + ε, T ; H ) for each ε ∈ (0, T − τ ). Moreover, the
following inequality holds

t
(t − τ )u(t)2V + (s − τ )u(s)2H 2 (Ω)∩V ds ≤ C(1 + u(τ )2H + (t − τ )2 ) ∀t ∈ (τ, T ].
τ

Proof Let τ < T and u(·) be an arbitrary weak solution of Problems (5.1) and
(5.2) on [τ, T ]. We fix ε ∈ (0, T − τ ). Theorem 2.1 implies that u(·) ∈ C([τ +
ε, T ]; V ) ∩ L 2 (τ + ε, T ; H 2 (Ω) ∩ V ) and u t ∈ L 2 (τ + ε, T ; H ). Then u(·)2V
and u(·)2H are absolutely
 continuous on [τ + ε,T ] and for a.e. s ∈ (τ + ε, T )
we have ds d 1
2
u(s) 2
V = (u (s), −u(s)) and ds
d 1
2
u(s)2H = (u (s), u(s)) (cf.
[18, Chapter IV]). Thus due to grows and sign assumptions (5.3) and (5.5) for a.e.
s ∈ (τ + ε, T ) in a standard way we obtain
5.2 Regularity of All Weak Solutions and Their Additional Properties 113
 
d 1
(s − τ − ε)u(s)2V + u(s)2H + (s − τ − ε)u(s)2H 2 (Ω)∩V
ds 2

1
≤ u(s)2H c02 + + 2c02 (s − τ − ε) + c02 + 2c02 (s − τ − ε) meas(Ω)
2

1
≤ c02 + + 2c02 (s − τ − ε) u(s)2H + meas(Ω) (5.7)
2

1 ∗ c2
≤ c02 + + 2c02 (s − τ − ε) u(τ )2H e−2ε (s−τ ) + ∗ + meas(Ω) ,
2 ε

where the last inequality follows from (5.6). We fix an arbitrary t ∈ [τ + ε, T ].

Integrating the inequality (5.7) from τ + ε to t, we have

t
2 1 2 1 2
(t − τ − ε)u(t)V + u(t) H − u(τ + ε) H + (s − τ − ε)u(s)2H 2 (Ω)∩V ds
2 2
τ +ε
 
1 2 1 3 2 1 c2
≤ c + + c0 + (t − τ )2 × u(τ )2H + ∗ + meas(Ω) .
2 0 4 2 4 ε

Let ε  0 + . Then ∀t ∈ (τ, T ]

t
u(t)2V (t − τ) + (s − τ )u(s)2H 2 (Ω)∩V ds ≤ C((t − τ )2 + u(τ )2H + 1),
τ

where C > 0 is a constant that does not depend on τ, T, ε, and u(·).

The theorem is proved.

5.3 Convergence of Weak Solutions in the Strongest

Topologies

For each u τ ∈ H we set Dτ,T (u τ ) = {u(·) ∈ L 2 (τ, T ; V ) u(·) is a weak solution
of Problems (5.1) and (5.2) and u(τ ) = u τ }. We note that the existence of a weak
solution for this problem was considered in [56] (see also Sect. 1.1).
The compactness in V of global attractor and compactness in L loc 2
(R+ ; H 2 (Ω) ∩
V ) ∩ C(R+ ; V ) of trajectory attractor for Problems (5.1) and (5.2) with initial data
from H is based on properties of the family of weak solutions of Problems (5.1) and
(5.2), related to the asymptotic compactness of the generated m-semiflow of solutions
and its absorbing (cf. [10, 30–32, 35, 44] and references therein). Theorem 5.2 below
on dependence of weak solutions in V on initial data from H and Theorem 5.1 allow
us to investigate the dynamics of all weak solutions of Problems (5.1) and (5.2) in V
as t → +∞.
114 5 Strongest Convergence Results for Weak Solutions…

Theorem 5.2 Let τ < T and u τ,n → u τ weakly in H, u n (·) ∈ Dτ,T (u τ,n ) for any
n ≥ 1. Then there exist a subsequence {u n k (·)}k≥1 ⊂ {u n (·)}n≥1 and u(·) ∈ Dτ,T (u τ )
such that

∀ε ∈ (0, T − τ ) sup u n k (t) − u(t)V → 0, k → +∞. (5.8)

t∈[τ +ε,T ]

Proof Let τ < T , u τ,n → u τ weakly in H , u n (·) ∈ Dτ,T (u τ ) ∀n ≥ 1. Theorem 1

from [26] (cf. also [56, Theorem 2.1, p. 56]) implies the existence of a subsequence
{u n k (·)}k≥1 ⊆ {u n (·)}n≥1 and u(·) ∈ Dτ,T (u τ ) such that

∀ε ∈ (0, T − τ ) sup u n k (t) − u(t) H → 0, k → +∞. (5.9)

t∈[τ +ε,T ]

We fix an arbitrary ε ∈ (0, T − τ ). Theorem 2.1 implies that the restrictions of u n k (·)
and u(·) on [τ + ε, T ] belong to L 2 (τ + ε, T ; H 2 (Ω) ∩ V ) ∩ C([τ + ε, T ]; V ).
Moreover, u n k ,t (·) and u t (·) belong to L 2 (τ + ε, T ; H ). Theorem 5.1 imply that
{u n k (·)}k≥1 is bounded in C([τ + ε, T ]; V ) ∩ L 2 (τ + ε, T ; H 2 (Ω) ∩ V ). Moreover,
{u n k ,t (·)}k≥1 is bounded in L 2 (τ + ε, T ; H ). Thus in virtue of (5.9) and of the com-
pact and dense embedding H 2 (Ω) ∩ V ⊂ V ⊂ H ⊂ V ∗ , we have

u n k (·) → u(·) weakly in L 2 (τ + ε, T ; H 2 (Ω) ∩ V ),

(5.10)
u n k ,t (·) → u t (·) weakly in L 2 (τ + ε, T ; H ), k → +∞.

Moreover,
u n k (·) → u(·) in C([τ + ε, T ]; Vw ), k → +∞. (5.11)

Without loss of generality, in virtue of the compact embedding theorem (cf. [33,
Section 5.1]), the next convergences hold

u n k (t) → u(t) in V for a.e. t ∈ (τ + ε, T ),

(5.12)
u n k (·) → u(·) in L 2 (τ + ε, T ; V ), k → +∞.

We consider the dense subset of [τ, T ]:


D := {t ∈ [τ, T ]  u n k (t) → u(t) in V, k → +∞}.

Let us fix an arbitrary ε > 0 such that τ + ε ∈ D. Then

sup u n k (t) − u(t)V = u n k (tn k ) − u(tn k )V , (5.13)

t∈[τ +ε,T ]

where tn k ∈ [τ + ε, T ] for any k ≥ 1.

Let us show that

u n k (tn k ) − u(tn k )V → 0, k → +∞. (5.14)

5.3 Convergence of Weak Solutions in the Strongest Topologies 115

We prove this statement by contradiction. If (5.14) does not hold, then without loss
of generality we assume that for some t0 ∈ [τ + ε, T ]

tn k → t0 , k → +∞, (5.15)

and there exists δ ∗ > 0 such that

u n k (tn k ) − u(tn k )V ≥ δ ∗ ∀k ≥ 1. (5.16)

As u(·) ∈ C([τ + ε, T ]; V ) then (5.15) and (5.16) imply

u n k (tn k ) − u(t0 )V ≥ δ ∗ for k rather large. (5.17)

On the other hand from (5.12) we get

V (u n k )(s) → V (u)(s) for each s ∈ (τ + ε, T ) ∩ D, k → +∞, (5.18)

where for any weak solution z(·) of Problems (5.1) and (5.2) on [τ, T ] and any
s ∈ [τ + ε, T ]

s
V (z)(s) = z(s)2V − 2c02 meas(Ω)s − 2c02 u(ξ )2H dξ.
τ

We note that

V (u n k )(t) ≤ V (u n k )(s) ∀ τ ≤ s ≤ t ≤ T, ∀k ≥ 1. (5.19)

Let us prove the inequality

lim V (u n k )(tn k ) ≤ V (u)(t0 ). (5.20)

k→+∞

We need to consider two cases.

Case 1: t0 > τ + ε. Let us fix an arbitrary δ > 0. As u(·) ∈ C([τ + ε, T ]; V ),
then the density of D in [τ, T ] implies the existence of s̄ ∈ [τ + ε, t0 ) ∩ D such that

V (u)(s̄) − V (u)(t0 ) < δ.

In virtue of (5.15)–(5.19) for any δ > 0 we obtain

lim V (u n k )(tn k ) − V (u)(t0 ) ≤ lim V (u n k )(s̄) − V (u)(t0 ) = V (u)(s̄) − V (u)(t0 ) < δ.

k→+∞ k→+∞

Thus inequality (5.20) holds.

Case 2: t0 = τ + ε. As τ + ε ∈ D, then in virtue of (5.15)–(5.19) we obtain
116 5 Strongest Convergence Results for Weak Solutions…

lim V (u n k )(tn k ) − V (u)(t0 ) ≤ lim V (u n k )(t0 ) − V (u)(t0 ) = 0.

k→+∞ k→+∞

Thus inequality (5.20) is true. Therefore, (5.9), (5.11) and (5.15) imply

lim u n k (tn k )V ≤ u(t0 )V

k→+∞

that together with (5.11) provides

u n k (tn k ) → u(t0 ) in V, k → +∞,

which contradicts (5.17). Therefore (5.14) holds.

The theorem is proved.

5.4 Examples of Applications

In this section we provide examples of applications to theorems established in

Sects. 5.1–5.3. We consider a model of combustion in porous media (sect. 5.4.1),
a model of conduction of electrical impulses in nerve axons (sect. 5.4.2), a climate
energy balance model (sect. 5.4.2); and a model of combustion in porous media
(sect. 5.4.3).

5.4.1 A Model of Combustion in Porous Media

Let us consider the following problem:

 ∂u
− ∂∂ xu2 − f (u) ∈ λH (u − 1), (x, t) ∈ (0, π ) × R,
2

∂t (5.21)
u(0, t) = u(π, t) = 0, t ∈ R,

where f : R → R is a continuous and nondecreasing function satisfying growth

and sign assumptions, λ > 0, and H (0) = [0, 1], H (s) = I{s > 0}, s = 0; Feireisl
 5.1). For each u τ ∈ L ((0, π )) we set
2
and Norbury [17] (see also sect. 2.4.5 and Fig.
2 1 
Dτ,T (u τ ) = {u(·) ∈ L (τ, T ; H0 ((0, π ))) u(·) is a weak solution of
Problem (5.21) and u(τ ) = u τ }. Since Problem (5.21) is a particular case of Prob-
lems (5.1) and (5.2), then the following statement holds: if τ < T and u τ,n → u τ
weakly in L 2 ((0, π )), u n (·) ∈ Dτ,T (u τ,n ) for any n ≥ 1, then there exist a subse-
quence {u n k (·)}k≥1 ⊂ {u n (·)}n≥1 and u(·) ∈ Dτ,T (u τ ) such that

∀ε ∈ (0, T − τ ) sup u n k (t) − u(t) H01 ((0,π)) → 0, k → +∞.

t∈[τ +ε,T ]
5.4 Examples of Applications 117

Fig. 5.1 Porous media

Fig. 5.2 Structure of the peripheral nerve

5.4.2 A Model of Conduction of Electrical Impulses in Nerve

Axons

Consider the problem:

 ∂u
− ∂∂ xu2 + u ∈ λH (u − a), (x, t) ∈ (0, π ) × R,
2

∂t (5.22)
u(0, t) = u(π, t) = 0, t ∈ R,

where a ∈ 0, 21 ; Terman [47, 48] (see also sect. 2.4.2 and Fig. 5.2). Since Problems
(5.22) is a particular case of Problems (5.1) and (5.2), then then the following state-
ment holds: if τ < T and u τ,n → u τ weakly in L 2 ((0, π )), u n (·) ∈ Dτ,T (u τ,n ) for any
n ≥ 1, then there exist a subsequence {u n k (·)}k≥1 ⊂ {u n (·)}n≥1 and u(·) ∈ Dτ,T (u τ )
such that

∀ε ∈ (0, T − τ ) sup u n k (t) − u(t) H01 ((0,π)) → 0, k → +∞.

t∈[τ +ε,T ]
118 5 Strongest Convergence Results for Weak Solutions…

5.4.3 Climate Energy Balance Model

Let (M , g) be a C ∞ compact connected oriented two-dimensional Riemannian man-

ifold without boundary (as, e.g. M = S 2 the unit sphere of R3 ). Consider the Budyko
model (see also sect. 2.4.3 and Figs. 5.3 and 5.4):

∂u
∂t
− Δu + Bu ∈ Q S(x)β(u), (x, t) ∈ M × R, (5.23)

where Δu = divM (∇M u) ; ∇M is understood in the sense of the Riemannian metric

g (see sect. 2.4.3, Budyko [8] and Sellers [41]).
Let S : M → R be a function such that S ∈ L ∞ (M ) and there exist S0 , S1 > 0
such that
0 < S0 ≤ S(x) ≤ S1 .

Fig. 5.3 Budyko model

Fig. 5.4 Climate energy

5.4 Examples of Applications 119

Suppose also that β is a bounded maximal monotone graph of R2 , that is there exist
m, M ∈ R such that for all s ∈ R and z ∈ β(s)

m ≤ z ≤ M.

Let us consider real Hilbert spaces

H := L 2 (M ), V := {u ∈ L 2 (M ) : ∇M u ∈ L 2 (T M )}

with respective standard norms  ·  H ,  · V , and inner products ( · , · ) H , ( · , · )V ,

where T M represents the tangent bundle and the functional spaces L 2 (M ) and
L 2 (T M ) are defined in a standard way; see, for example, Aubin [2]. According
to Theorem 2.2, for any −∞ < τ < T < +∞ each weak solution u τ ∈ L 2 (Ω) of
Problem (5.23) on [τ, T ] belongs to C([τ + ε, T ]; H01 (Ω)) ∩ L 2 (τ + ε, T ; H 2 (Ω) ∩
H01 ((0, π ))) and du
dt
(·) ∈ L 2 (τ + ε, T ; L 2 (Ω)) for each ε ∈ (0, T − τ ). Consider the
generalized setting of Problem (5.23):

du
+ Au(t) + ∂ J1 (u(t)) − ∂ J2 (u(t))  0̄ on (−∞ < τ < T < +∞), (5.24)
dt

where A : V → V ∗ be a linear symmetric operator such that ∃c > 0 : Av, vV ≥

cv2V , for each v ∈ V and Ji : H → R be a convex, lower semi-continuous function
such, that the following assumptions hold: (i) (growth condition) there exists c1 > 0
such that y H ≤ c1 (1 + u H ), for each u ∈ H and y ∈ ∂ Ji (u) and i = 1, 2; (ii)
(sign condition) there exist c2 > 0, λ ∈ (0, c) such that (y1 − y2 , u) H ≥ −λu2H −
c2 , for each yi ∈ ∂ Ji (u), u ∈ H , where ∂ Ji (u) the subdifferential of Ji (·) at a point
u. Note that u ∗ ∈ ∂ Ji (u) if and only if u ∗ (v − u) ≤ Ji (v) − Ji (u) ∀v ∈ H ; i = 1, 2.
Let D(A) = {u ∈ V : Au ∈ H }. We note that the mapping v → Av H defines the
equivalent norm on D(A); Temam [42, Chapter III].
We recall that the function u(·) ∈ L 2 (τ, T ; V ) is called a weak solution of Problem
(5.24) on [τ, T ], if there exist Bochner measurable functions di : (τ, T ) → H ; i =
1, 2, such that

di (t) ∈ ∂ Ji (u(t)) for a.e. t ∈ (τ, T ), i = 1, 2; and (5.25)

 T  
− u, v ξ (t) +  Au, v ξ(t) + d1 , v ξ(t) − d2 , v ξ(t) dt = 0, (5.26)
τ

for all ξ ∈ C0∞ (τ, T ) and for all v ∈ V .

The following theorem provides sufficient conditions for the existence and regu-
larity of all weak solutions for Problem (5.24).
Theorem 5.3 Let −∞ < τ < T < +∞ and u τ ∈ H . Problem (5.24) has at least
one weak solution u(·) ∈ L 2 (τ, T ; V ) on [τ, T ] such that u(τ ) = u τ . Moreover, if
u(·) is a weak solution of Problem (5.24) on [τ, T ], then u(·) ∈ C([τ + ε, T ]; V ) ∩
L 2 (τ + ε, T ; D(A)) and du
dt
(·) ∈ L 2 (τ + ε, T ; H ) for any ε ∈ (0, T − τ ).
120 5 Strongest Convergence Results for Weak Solutions…

Proof We note that for any u τ ∈ H there exists at least one weak solution of Problem
(5.24) on [τ, T ] with initial condition u(τ ) = u τ ; see Kasyanov [26] and references
therein. The regularity of each weak solution follows from Theorem 2.3.
The theorem is proved.

Denote by K+ the family of all, globally defined on [0, +∞), weak solutions of
Problem (5.24). Let us set

1
E(u) = Au, u + J1 (u) − J2 (u), u ∈ V. (5.27)
2

Theorem 5.4 For each u ∈ K+ and all τ and T , 0 < τ < T < ∞, the energy equal-
ity holds
 T
du
E(u(T )) − E(u(τ )) = −  (s)2H ds. (5.28)
τ ds

Proof Suppose u(·) ∈ K+ be arbitrary fixed and let 0 < τ < T < +∞. To sim-
plify conclusions, let the symbol u(·) denotes the restriction of u(·) on [τ, T ].
Theorem 5.3 implies that u(·) ∈ C([τ, T ]; V ) ∩ L 2 (τ, T ; D(A)) and du dt
(·) ∈ L 2
(τ, T ; H ), because τ > 0. Barbu [7, Lemma 2.1, p. 189] yields that the functions
Ji (u(·)), i = 1, 2, are absolutely continuous on [τ, T ] and the equality holds:

d du
Ji (u(t)) = h i (t), (t) H , for a.e. t ∈ (τ, T ), (5.29)
dt dt

for all h i (·) ∈ L 2 (τ, T ; H ) such that h i (t) ∈ ∂ Ji (s)|s=u(t) for a.e. t ∈ (τ, T ), i=1,2.
We remark that the mapping t → Au(t), u(t)V is absolutely continuous on
[τ, T ] and the equality holds:

d du
Au(t), u(t) = 2Au(t), (t) H , for a.e. t ∈ (τ, T ) (5.30)
dt dt

Thus, the function E(u(·)) is absolutely continuous on [τ, T ] as the linear com-
bination of absolutely continuous on [τ, T ] functions. According to formulae (5.29)
and (5.30), dtd E(u(t)) = − du
dt
(t)2H for a.e. t ∈ (τ, T ).
The theorem is proved.

Repeating several lines from the proof of Theorem 5.1 we obtain that there exists
C > 0 such that for any τ < T and for each weak solution u(·) of Problem (5.24)
on [τ, T ] the inequality holds

t
(t − τ )u(t)2V + (s − τ )u(s)2D(A) ds ≤ C(1 + u(τ )2H + (t − τ )2 ), (5.31)
τ

for each t ∈ (τ, T ].

5.4 Examples of Applications 121

Let

Dτ,T (u τ ) = {u(·) ∈ L 2 (τ, T ; V ) u(·) is a weak solution of Problem (5.24) and u(τ ) = u τ },

for any u τ ∈ H . Let us provide the main convergence result for all weak solutions
of Problem (5.24) in the strongest topologies.
Theorem 5.5 Let τ < T , u τ,n → u τ weakly in H , u n (·) ∈ Dτ,T (u τ,n ) for any n ≥ 1.
Then there exists a subsequence {u n k (·)}k≥1 ⊆ {u n (·)}n≥1 and u(·) ∈ Dτ,T (u τ ) such
that
sup u n k (t) − u(t)V → 0, (5.32)
t∈[τ +ε,T ]

 T
du n k du
 (t) − (t)2H dt → 0, (5.33)
τ +ε dt dt

as k → +∞, for all ε ∈ (0, T − τ ).

Proof The inequality (5.31), Kasyanov et al. [29, Theorem 3], Banach-Alaoglu the-
orem, and Cantor diagonal arguments (alternatively we may repeat several lines
from the proof of Theorem 5.2) yield that there exist a subsequence {u n k (·)}k≥1 ⊆
{u n (·)}n≥1 and u(·) ∈ Dτ,T (u τ ) such that the following statements hold:
(a) the restrictions of u n k (·) and u(·) on [τ + ε, T ] belong to C([τ + ε, T ]; V ) ∩
du
L (τ + ε, T ; D(A)) and dtnk (·), du
2
dt
(·) ∈ L 2 (τ + ε, T ; H );
(b) the following convergence hold:

u n k (·) → u(·) weakly in L 2 (τ + ε, T ; D(A)),

u n k (·) → u(·) strongly in C([τ + ε, T ]; V ), (5.34)
du n k
dt
(·) → du
dt
(·) weakly in L 2 (τ + ε, T ; H ),

as k → ∞, for each ε ∈ (0, T − τ ), that imply statement (5.32). Let us prove (5.33).
Theorem 5.4 yields the following energy equalities
 T
du
 (t)2H dt = E(u(τ + ε)) − E(u(T )), (5.35)
τ +ε dt
 T
du n k
 (t)2H dt = E(u n k (τ + ε)) − E(u n k (T )), (5.36)
τ +ε dt

k ≥ 1, ε ∈ (0, T − τ ). Continuity of E on V and (5.32) imply

E(u n k (τ + ε)) − E(u n k (T )) → E(u(τ + ε)) − E(u(T )), m → ∞. (5.37)

Therefore, formulae (5.35)–(5.37) yield

122 5 Strongest Convergence Results for Weak Solutions…
 T  T
du n k du
 (t)2H dt →  (t)2H dt, (5.38)
τ +ε dt τ +ε dt

as k → ∞, for each ε ∈ (0, T − τ ). Since, L 2 (τ + ε; T ) is a Hilbert space, (5.34)

and (5.38) imply (5.33).
The theorem is proved.

References

1. Anguiano M., Caraballo T., Real J., Valero J.: Pullback attractors for reaction-diffusion equa-
tions in some unbounded domains with an H −1 -valued non-autonomous forcing term and
without uniqueness of solutions. Discret. Contin. Dyn. Syst. Ser. B 14, 307–326 (2010)
2. Aubin, T.: Nonlinear Analysis on Manifolds. Monge-Ampére Equations, Springer, Berlin
(1980)
3. Balibrea, F., Caraballo, T., Kloeden, P.E., Valero, J.: Recent developments in dynamical sys-
tems: three perspectives. Int. J. Bifurc. Chaos (2010). doi:10.1142/S0218127410027246
4. Ball, J.M.: Continuity properties and global attractors of generalized semiflows and the Navier-
Stokes equations. Nonlinear Sci. 7, 475–502 (1997). (Erratum, ibid 8:233,1998. Corrected
version appears in ‘Mechanics: from Theory to Computation’. pp. 447–474. Springer, Berlin
(2000))
5. Ball, J.M.: Global attractors for damped semilinear wave equations. DCDS. 10, 31–52 (2004)
6. Babin, A.V., Vishik, M.I.: Attractors of Evolution Equations. Nauka, Moscow (1989)
7. Barbu, V.: Nonlinear Semigroups and Differential Equations in Banach Spaces. Editura Acad-
emiei, Bucuresti (1976)
8. Budyko, M.I.: The effects of solar radiation variations on the climate of the Earth. Tellus 21,
611–619 (1969)
9. Chepyzhov, V.V., Vishik, M.I.: Attractors for Equations of Mathematical Physics. American
Mathematical Society, Providence, Rhode Island (2002)
10. Chepyzhov, V.V., Vishik, M.I.: Evolution equations and their trajectory attractors. J. Math. Pure
Appl. (1997). doi:10.1016/S0021-7824(97)89978-3
11. Chepyzhov, V.V., Vishik, M.I.: Trajectory and global attractors of three-dimensional Navier-
Stokes systems. Math. Notes (2002). doi:10.1023/A:1014190629738
12. Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley, New York (1983)
13. Díaz, H., Díaz, J.: On a stochastic parabolic PDE arising in climatology. Rev. R. Acad. Cien.
Ser. A Mat. 96, 123–128 (2002)
14. Díaz, J., Hernández, J., Tello, L.: On the multiplicity of equilibrium solutions to a nonlinear
diffusion equation on a manifold arising in climatology. J. Math. Anal. Appl. 216, 593–613
(1997)
15. Díaz, J., Hernández, J., Tello, L.: Some results about multiplicity and bifurcation of stationary
solutions of a reaction diffusion climatological model. Rev. R. Acad. Cien. Ser. A. Mat. 96(3),
357–366 (2002)
16. Díaz, J., Tello, L.: Infinitely many stationary solutions for a simple climate model via a shooting
method. Math. Meth. Appl. Sci. 25, 327–334 (2002)
17. Feireisl, E., Norbury, J.: Some existence and nonuniqueness theorems for solutions of parabolic
equations with discontinuous nonlinearities. Proc. Roy. Soc. Edinburgh A. 119(1–2), 1–17
(1991)
18. Gajewski, H., Gröger, K., Zacharias, K.: Nichtlineare Operatorgleichungen und Operatordif-
ferentialgleichungen. Akademie-Verlag, Berlin (1974)
19. Kalita, P.: Convergence of rothe scheme for hemivariational inequalities of parabolic type. Int.
J. Numer. Anal. Model. 10, 445–465 (2013)
References 123

20. Kalita, P.: Regularity and Rothe method error estimates for parabolic hemivariational inequality.
J. Math. Anal. Appl. 389, 618–631 (2012)
21. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Pullback attractors for a class of extremal solutions
of the 3D Navier-Stokes equations. J. Math. Anal. Appl. 373(2), 535–547 (2011). doi:10.1016/
j.jmaa.2010.07.040
22. Kapustyan, O.V., Valero, J.: Comparison between trajectory and global attractors for evo-
lution systems without uniqueness of solutions. Intern. J Bifur. Chaos (2010). doi:10.1142/
S0218127410027313
23. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Structure of the global attractor for weak solutions
of reaction-diffusion equation. Appl. Math. Inf. Sci. 9(5), 2257–2264 (2015). doi:10.12785/
amis/090506
24. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Regular solutions and global attractors for reaction-
diffusion systems without uniqueness. Commun. Pure Appl. Anal. 13(5), 1891–1906 (2014).
doi:10.3934/cpaa.2014.13.1891
25. Kasyanov, P.O.: Multi-valued dynamics of solutions of autonomous operator differential equa-
tions with pseudomonotone nonlinearity. Math. Notes. 92(1–2), 205–218 (2012). doi:10.1134/
S0001434612070231
26. Kasyanov, P.O.: Multi-valued dynamics of solutions of an autonomous differential-operator
inclusion with pseudomonotone nonlinearity. Cybern. Syst. Anal. (2011). doi:10.1007/s10559-
011-9359-6
27. Kas’yanov, P.O., Mel’nyk, V.S.: On properties of subdifferential mappings in Frechet spaces.
Ukr. Math. J. 57(10), 1621–1634 (2005)
28. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Long-time behaviour of solutions
for autonomous evolution hemivariational inequality with multidimensional “Reaction-
Displacement” law. Abstr. Appl. Anal. 2012(450984), 21 (2012). doi:10.1155/2012/450984
29. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Regularity of weak solutions and their attrac-
tors for a parabolic feedback control problem. Set-Valued Var. Anal. 21, 271–282 (2013).
doi:10.1007/s11228-013-0233-8
30. Ladyzhenskaya, O.A.: Dynamical system, generated by Navier-Stokes equations. Zap Nauch
Sem LOMI. 27, 91–115 (1972)
31. Ladyzhenskaya, O.A.: Some comments to my papers on the theory of attractors for abstract
semigroups. Zap Nauchn Sem LOMI. 188, 102–112 (1990)
32. Ladyzhenskaya, O.A.: Attractors for Semigroups and Evolution Equations. University Press,
Cambridge (1991)
33. Lions, J.L.: Quelques méthodes de résolution des problémes aux limites non linéaires. Dunod,
Paris (1969)
34. Lions, J.L., Magenes, E.: Problèmes Aux Limites Non-homogènes et Applications. Dunod,
Paris (1968)
35. Melnik, V.S., Valero, J.: On attractors of multi-valued semiflows and differential inclusions.
Set Valued Anal. (1998). doi:10.1023/A:1008608431399
36. Migórski, S.: On the existence of solutions for parabolic hemivariational inequalities. J. Com-
put. Appl. Math. 129, 77–87 (2001)
37. Migórski, S., Ochal, A.: Optimal control of parabolic hemivariational inequalities. J. Glob.
Optim. 17, 285–300 (2000)
38. Otani, M., Fujita, H.: On existence of strong solutions for dudt (t) + ∂ϕ (u(t)) − ∂ϕ (u(t)) 
1 2

f (t). J. Fac. Sci. Univ. Tokyo. Sect. 1 A, Math. 24(3), 575–605 (1977)
39. Panagiotopoulos, P.D.: Inequality Problems in Mechanics and Applications. Convex and Non-
convex Energy Functions. Birkhauser, Basel (1985)
40. Sell, G.R., You, Yu.: Dynamics of Evolutionary Equations. Springer, New York (2002)
41. Sellers, W.D.: A global climatic model based on the energy balance of the Earth-atmosphere
system. J. Appl. Meteorol. 8, 392–400 (1969)
42. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Springer,
New York (1988)
124 5 Strongest Convergence Results for Weak Solutions…

43. Valero, J.: Attractors of parabolic equations without uniqueness. J. Dyn. Differ. Equ. (2001).
doi:10.1023/A:1016642525800
44. Valero, J., Kapustyan, A.V.: On the connectedness and asymptotic behaviour of solutions of
reaction-diffusion systems. J. Math. Anal. Appl. (2006). doi:10.1016/j.jmaa.2005.10.042
45. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong trajectory attractor of dissipative reaction-
diffusion system. Doklady RAN 435(2), 155–159 (2010)
46. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Springer,
New York (1997)
47. Terman, D.: A free boundary problem arising from a bistable reaction-diffusion equation. SIAM
J. Math. Anal. 14, 1107–1129 (1983)
48. Terman, D.: A free boundary arising from a model for nerve conduction. J. Diff. Eqs. 58(3),
345–363 (1985)
49. Vishik, M.I., Fursikov, A.V.: Mathematical Problems of Statistical Hydromechanics. Kluwer
Academic, Dordrecht (1988)
50. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong trajectory attractor for dissipative reaction-
diffusion system. Doclady Math. (2010). doi:10.1134/S1064562410060086
51. Zelik, S.: The attractor for a nonlinear reaction-diffusion system with a supercritical nonlinearity
and it’s dimension. Rend. Accad. Naz. Sci. XL Mem. Mat. Appl. 24, 1–25 (2000)
52. Zgurovsky, M.Z., Kasyanov, P.O., Valero, J.: Noncoercive evolution inclusions for Sk type
operators. Int. J. Bifurc. Chaos (2010). doi:10.1142/S0218127410027386
53. Zgurovsky, M.Z., Kasyanov, P.O., Zadoianchuk, N.V.: Long-time behavior of solutions for qua-
silinear hyperbolic hemivariational inequalities with application to piezoelectricity problem.
Appl. Math. Letters 25(10), 1569–1574 (2012). doi:10.1016/j.aml.2012.01.016
54. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation Inequali-
ties for Earth Data Processing I. Springer, Berlin (2010). doi:10.1007/978-3-642-13837-9
55. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation
Inequalities for Earth Data Processing II. Springer, Berlin (2010). doi:10.1007/978-3-642-
13878-2
56. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
Inclusions and Variation Inequalities for Earth Data Processing III. Springer, Berlin (2012).
doi:10.1007/978-3-642-28512-7
Chapter 6
Strongest Convergence Results for Weak
Solutions of Differential-Operator
Equations and Inclusions

Abstract In this chapter we establish strongest convergence results for weak solu-
tions of differential-operator equations and inclusions. In Sect. 6.1 we consider first
order differential-operator equations and inclusions. Section 6.2 devoted to conver-
gence results for weak solutions of second order operator differential equations and
inclusions. In Sect. 6.3 we consider the following examples of applications: nonlin-
ear parabolic equations of divergent form; nonlinear problems on manifolds with
and without boundary: a climate energy balance model; a model of conduction of
electrical impulses in nerve axons; viscoelastic problems with nonlinear “reaction-
displacement” law.

6.1 First Order Differential-Operator Equations

and Inclusions

In this section we consider strongest convergence results for both the autonomous first
order differential-operator equations as well as nonautonomous evolution inclusions.

6.1.1 Convergence Results for Autonomous Evolution

Equations

Let us consider the first-order general nonlinear evolution equations of the form

u  (t) + A(u(t)) = 0̄, (6.1)

It is assumed that the nonlinear operator A : V → V ∗ , acts in a Banach space V ,

which is reflexive and separable and, for some Hilbert space H , the embeddings V 
H ≡ H ⊂ V ∗ are valid. Suppose that the nonlinear operator A is pseudomonotone
and satisfies dissipation conditions of the form
p
A(u), uV ≥ α u V − β ∀u ∈ V, (6.2)
© Springer International Publishing AG 2018 125
M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_6
126 6 Strongest Convergence Results for Weak Solutions …

where p ≥ 2, and α, β > 0, and also power growth conditions of the form
p−1
A(u) V∗ ≤ c(1 + u V ) ∀u ∈ V, (6.3)

for some c > 0. Here ·, ·V : V ∗ × V → R is the pairing in V ∗ × V coinciding on

H × V with the inner product(·, ·) in the Hilbert space H .
By a weak solution of operator differential equation (6.1) on a closed interval
[τ, T ] we mean an element u of the space L p (τ, T ; V ) such that

T T
∀ξ ∈ C0∞ ([τ, T ]; V ) − (ξ  (t), u(t))dt + A(u(t)), ξ(t)V dt = 0. (6.4)
τ τ

Many evolution partial differential equations in a domain Ω whose leading part

is a pth power nonlinear monotone differential operator and which may contain
lower (now nonmonotone) summands with subordinate nonlinearity growth can be
reduced to the form (6.1). In this case, the space V is a Sobolev space of the cor-
responding order, while the space H is H = L 2 (Ω). Such equations are very often
used to describe complicated evolution processes in various models in physics and
mechanics. For equations of the form (6.1), there is a well-developed technique
for constructing global (i.e., for all t ≥ 0) weak solutions u(t), t ≥ 0, from the
 ∗
p (R+ ; V ) such that u (·) ∈ L q (R+ ; V ) (here 1/ p + 1/q = 1). It is well
space L loc loc

known that such weak solutions u(t) are continuous functions with values in H , i.e.,
u(·) ∈ C(R+ ; H ).
The problem is to study the asymptotic behavior as t → +∞ of the families of
weak solutions {u(t)} of Problem (6.1) in the norm of H under the assumption that
the initial data {u(0)} constitute a bounded set in H (see also [1, 2, 4, 7, 9, 11,
19–21, 28]).
Note that, under certain additional conditions on the nonlinear operator A(u)
ensuring, for Problem (6.1), the unique solvability of the Cauchy problem u|t=0 = u 0
for any u 0 ∈ H , the study of the class of weak solutions under consideration involves
the highly fruitful theory of dynamical semigroups and their global attractors in
infinite-dimensional phase spaces. This theory has been successfully developed over
a period of more than thirty years; its foundations were created by Ladyzhenskaya,
Babin, Vishik, Hale, Temam and other well-known mathematicians [14, 15, 17, 18].
The problem becomes significantly more complicated if the corresponding Cauchy
problem is not uniquely solvable or the proof of the relevant theorem is not known.
Such a situation often occurs in complicated mathematical models. In this case,
the “classical” method based on unique semigroups and global attractors cannot be
applied directly. However, two approaches to the study of the dynamics of the corre-
sponding weak solutions are well known. The first method is based on the theory of
multi-valued semigroups; it was developed in ground-breaking papers of Babin and
Vishik (see, for example, [3]). The second approach uses the method of trajectory
attractors; it was proposed in the papers [5, 6] of Chepyzhov and Vishik as well as
in the independent work [25] of Sell.
6.1 First Order Differential-Operator Equations and Inclusions 127

The new results contained in the present section consist in the application of these
two approaches to the study of the strongest convergence results for weak solutions
of equations of the form (6.1) with general nonlinear pseudomonotone operator A(u)
satisfying (S)-property without any conditions guaranteeing the unique solvability
of the Cauchy problem.
For fixed τ < T let us set
∗
X τ,T = L p (τ, T ; V ), X τ,T = L q (τ, T ; V ∗ ), Wτ,T = {u ∈ X τ,T | u  ∈ X τ,T
∗
},

where u  is a derivative of an element u ∈ X τ,T in the sense of the space of distri-

butions D([τ, T ]; V ∗ ) (see, for example, [12, Definition IV.1.10, p. 168]). We note
that
A(u)(t) = A(u(t)), for any u ∈ X τ,T and a.e. t ∈ (τ, T ).

The space Wτ,T is a reflexive Banach space with the graph norm of a derivative (see,
for, example [15, Proposition 4.2.1, p. 291]):

u Wτ,T = u X τ,T + u ∗ ,
X τ,T u ∈ Wτ,T . (6.5)

Properties of A and (V, H, V ∗ ) provide the existence of a weak solution of Cauchy

problem (6.1) with initial data
u(τ ) = u τ (6.6)

on the interval [τ, T ] for an arbitrary yτ ∈ H . Therefore, the next result takes place:
According to Proposition 1.1, for any τ < T, yτ ∈ H Cauchy problem (6.1), (6.6)
has a weak solution on the interval [τ, T ]. Moreover, each weak solution u ∈ X τ,T of
Cauchy problem (6.1), (6.6) on the interval [τ, T ] belongs to Wτ,T ⊂ C([τ, T ]; H ).
For fixed τ < T we denote

Dτ,T (u τ ) = {u(·) | u is a weak solution of (6.1) on [τ, T ], u(τ ) = u τ }, u τ ∈ H.

From Proposition 1.1 it follows that Dτ,T (u τ ) = ∅ and Dτ,T (u τ ) ⊂ Wτ,T ∀τ <
T, u τ ∈ H .
We note that the translation and concatenation of weak solutions is a weak solution
too.
Lemma 6.1 (Zgurovsky et al. [15]) If τ < T , u τ ∈ H , u(·) ∈ Dτ,T (u τ ), then v(·) =
u(· + s) ∈ Dτ −s,T −s (u τ ) ∀s. If τ < t < T , u τ ∈ H , u(·) ∈ Dτ,t (u τ ) and v(·) ∈
Dt,T (u(t)), then 
u(s), s ∈ [τ, t],
z(s) =
v(s), s ∈ [t, T ]

belongs to Dτ,T (u τ ).
As a rule, the proof of the existence of compact global and trajectory attractors
for equations of type (6.1) is based on the properties of the set of weak solutions of
128 6 Strongest Convergence Results for Weak Solutions …

problem (6.1) related to the absorption of the generated m-semiflow of solutions and
its asymptotic compactness (see, for example, [24, 27] and the references therein).
The following lemma on a priori estimates of solutions and Theorem 6.1 on the
dependence of solutions on initial data will play a key role in the study of the dynamics
of the solutions of Problem (6.1) as t → +∞.

Lemma 6.2 (Zgurovsky et al. [15]) There exist c4 , c5 , c6 , c7 > 0 such that for any
finite interval of time [τ, T ] every weak solution u of problem (6.1) on [τ, T ] satisfies
estimates: ∀t ≥ s, t, s ∈ [τ, T ]

t
p
u(t) 2
H + c4 u(ξ ) V dξ ≤ u(s) 2
H + c5 (t − s), (6.7)
s

2 −c6 (t−s)
u(t) 2
H ≤ u(s) He + c7 . (6.8)

We recall that A : V → V ∗ satisfies (S)-property, if from u n → u weakly in V

and A(u n ), u n − uV → 0, as n → ∞, it follows that u n → u strongly in V , as
n → +∞.
Further we assume that A satisfies (S)-property.
Theorem 6.1 Let τ < T , {u n }n≥1 be an arbitrary sequence of weak solutions of (6.1)
on [τ, T ] such that u n (τ ) → η weakly in H . Then there exist {u n k }k≥1 ⊂ {u n }n≥1 and
u(·) ∈ Dτ,T (η) such that

T
p
∀ε ∈ (0, T − τ ) max u n k (t) − u(t) H + u n k (t) − u(t) V dt → 0, k → +∞.
t∈[τ +ε,T ]
τ +ε
(6.9)

Before the proof of Theorem 6.1 let us provide some auxiliary statements.

Lemma 6.3 Let τ < T , yn → y weakly in Wτ,T , and

lim supA(yn ), yn − y X τ,T ≤ 0. (6.10)

n→+∞

Then
T
lim |A(yn (t)), yn (t) − y(t)V | dt = 0. (6.11)
n→+∞
τ

Proof There exists a set of measure zero, Σ1 ⊂ (τ, T ) such that for t ∈
/ Σ1 , we have
that
yn (t) ∈ V for all n ≥ 1.

Similarly to [17, p. 7] we verify the following claim.

6.1 First Order Differential-Operator Equations and Inclusions 129

Claim: Let yn → y weakly in Wτ,T and let t ∈

/ Σ1 . Then

lim A(yn (t)), yn (t) − y(t)V ≥ 0.

n→+∞

Proof of the claim. Fix t ∈

/ Σ1 and suppose to the contrary that

lim A(yn (t)), yn (t) − y(t)V < 0. (6.12)

n→+∞

Then up to a subsequence {yn k }k≥1 ⊂ {yn }n≥1 we have

lim A(yn k (t)), yn k (t) − y(t)V = lim A(yn (t)), yn (t) − y(t)V < 0. (6.13)
k→+∞ n→+∞

Therefore, for all rather large k, growth and dissipation conditions imply
p p−1
α yn k (t) V − β ≤ A(yn k (t)) V∗ y(t) V ≤ c(1 + yn k (t) V ) y(t) V.

which implies that the sequences { yn k (t) V }k≥1 and consequently { A(yn k
(t)) V ∗ }k≥1 are bounded sequences. In virtue of the continuous embedding Wτ,T ⊂
C([τ, T ]; H ) we obtain that yn k (t) → y(t) weakly in H . Due to boundedness of
{yn k (t)}k≥1 in V we finally have

∀t ∈ [τ, T ]\Σ1 yn k (t) → y(t) weakly in V, k → +∞. (6.14)

The pseudomonotony of A, (6.12)–(6.14) imply that

lim A(yn (t)), yn (t) − y(t)V ≥ A(y(t)),

n→+∞

y(t) − y(t)V = 0 > lim A(yn (t)), yn (t) − y(t)V .

n→+∞

We obtain a contradiction.
The claim is proved.
Now let us continue the proof of Lemma 6.3. The claim provides that for a.e.
t ∈ [τ, T ], in fact for any t ∈
/ Σ1 , we have

lim A(yn (t)), yn (t) − y(t)V ≥ 0. (6.15)

n→+∞

Dissipation and growth conditions imply that, if ω ∈ X τ,T , then

p p−1
A(yn (t)), yn (t) − ω(t)V ≥ α yn (t) V − β − c(1 + yn (t) V ) ω(t) V
for a.e. t ∈ [τ, T ]\Σ1 .

Using p − 1 = qp , the right side of the above inequality equals to

130 6 Strongest Convergence Results for Weak Solutions …
p
p
α yn (t) V − β − c yn (t) q
V ω(t) V − c ω(t) V.

Now using Young’s inequality, we can obtain a constant c(c, α) depending on c, α

such that p
α p p
c yn (t) Vq ω(t) V ≤ yn (t) V + ω(t) V · c(c, α).
2

Letting c̄ = max{β + qc ; c(c, α) + cp } it follows that

p
A(yn (t)), yn (t) − ω(t)V ≥ −c̄(1 + ω(t) V) for a.e. t ∈ [τ, T ]. (6.16)

Letting ω = y, we can use Fatou’s lemma and we obtain

T
p
lim [A(yn (t)), yn (t) − y(t)V + c̄(1 + y(t) V )]dt ≥
n→+∞
0

T T
p p
≥ lim [A(yn (t)), yn (t) − y(t)V + c̄(1 + y(t) V )]dt ≥ c̄ (1 + y(t) V )dt.
n→+∞
0 0

Therefore,

T
0 ≥ lim supA(yn ), yn − y X τ,T ≥ lim A(yn (t)), yn (t) − y(t)V dt =
n→+∞ n→+∞
τ

T
= lim A(yn ), yn − y X τ,T ≥ lim A(yn (t)), yn (t) − y(t)V dt = 0,
n→+∞ n→+∞
τ

showing that
lim A(yn ), yn − y X τ,T = 0. (6.17)
n→+∞

From (6.16),

/ Σ1 0 ≤ A(yn (t)), yn (t) − y(t)−

p
∀n ≥ 1 ∀t ∈ V ≤ c̄(1 + y(t) V ),

where a − = max{0, −a}, for a ∈ R. Due to (6.15) we know that for a.e. t, A(yn (t)),
yn (t) − y(t)V ≥ −ε for all rather large n. Therefore, for such n, A(yn (t)), yn (t) −
y(t)− −
V ≤ ε, if A(yn (t)), yn (t) − y(t)V < 0 and A(yn (t)), yn (t) − y(t)V = 0, if
−
A(yn (t)), yn (t) − y(t)V ≥ 0. Therefore, lim A(yn (t)), yn (t) − y(t)V = 0 and
n→+∞
we can apply the dominated convergence theorem and from (6.15) we conclude that
6.1 First Order Differential-Operator Equations and Inclusions 131

T T
lim A(yn (t)), yn (t) − y(t)−
V = lim A(yn (t)), yn (t) − y(t)−
V dt = 0.
n→+∞ n→+∞
τ τ

Now by (6.17) and the above equation we have

T
lim A(yn (t)), yn (t) − y(t)+
V dt =
n→+∞
τ

T
= lim [A(yn (t)), yn (t) − y(t)V + A(yn (t)), yn (t) − y(t)−
V ]dt =
n→+∞
0

= lim A(yn ), yn − y X τ,T = 0.

n→+∞

Therefore,
T
lim |A(yn (t)), yn (t) − y(t)V |dt = 0.
n→+∞
τ

The lemma is proved.

Lemma 6.4 Let τ < T , yn → y weakly in Wτ,T , and (6.10) holds. Then there exists
a subsequence {yn k }k≥1 ⊂ {yn }n≥1 such that for a.e. t ∈ (τ, T ) we have that yn k (t) →
y(t) weakly in V , and A(yn k (t)), yn k (t) − y(t)V → 0, k → +∞.

Proof Let yn → y weakly in Wτ,T and

lim supA(yn ), yn − y X τ,T ≤ 0.

n→+∞

In virtue of Lemma 6.3 we obtain

T
lim |A(yn (t)), yn (t) − y(t)V |dt = 0. (6.18)
n→+∞
τ

Due to the continuous embedding Wτ,T ⊂ C([τ, T ]; H ) we have

∀t ∈ [τ, T ] yn (t) → y(t) weakly in H, n → +∞. (6.19)

From (6.18) it follows that there exists a subsequence {yn k }k≥1 ⊂ {yn }n≥1 such that

A(yn k (t)), yn k (t) − y(t)V → 0, k → +∞, for a.e. t ∈ [τ, T ].

132 6 Strongest Convergence Results for Weak Solutions …

Let Σ1 ⊂ [τ, T ] be a set of measure zero such that for t ∈

/ Σ1 yn k (t), y(t) are well-
defined ∀k ≥ 1, and

A(yn k (t)), yn k (t) − y(t)V → 0, k → +∞.

In virtue of growth and dissipation conditions we obtain

 
p p−1
/ Σ1 ∀k ≥ 1 lim sup α yn k (t)
∀t ∈ V − β − c(1 + yn k (t) V ) y(t) V ≤ 0.
k→+∞

Thus ∀t ∈
/ Σ1
p p
lim sup yn k (t) V ≤ c(c, α, β, p)(1 + y(t) V ).
k→+∞

Therefore, due to (6.19) we obtain that for a.e. t ∈ (τ, T ) yn k (t) → y(t) weakly in
V , k → +∞.
The lemma is proved.

Proof of Theorem 6.1. Let τ < T , {u n }n≥1 be an arbitrary sequence of weak

solutions of (6.1) on [τ, T ] such that u n (τ ) → η weakly in H . Theorem 1 from [15]
implies the existence of a subsequence {u n k }k≥1 ⊂ {u n }n≥1 and u(·) ∈ Dτ,T (η) such
that

∀ε ∈ (0, T − τ ) max u n k (t) − u(t) H → 0, k → +∞. (6.20)

t∈[τ +ε,T ]

Let us prove that

T
p
∀ε ∈ (0, T − τ ) u n k (t) − u(t) V dt → 0, k → +∞. (6.21)
τ +ε

On the contrary, without loss of generality we assume that for some ε ∈ (0, T − τ )
and δ > 0 it is fulfilled

T
p
u n k (t) − u(t) V dt ≥ δ, ∀k ≥ 1. (6.22)
τ +ε

In virtue of (6.7), without loss of generality we claim that

u n k → u weakly in Wτ +ε,T , k → +∞. (6.23)

Moreover, due to (6.20), we have

6.1 First Order Differential-Operator Equations and Inclusions 133

T
lim sup A(u n k (t)), u n k (t) − u(t)V dt ≤ 0. (6.24)
k→∞
τ +ε

Thus, Lemma 6.4 and (S)-property for A imply that up to a subsequence which we
denote again as {u n k }k≥1 for a.e. t ∈ (τ + ε, T ) we have that u n k (t) → u(t) strongly
in V , k → +∞. Moreover, Lemma 6.3 provides that

T
 
lim A(u n (t)), u n (t) − u(t)V  dt = 0.
k k
k→+∞
τ +ε

Dissipation and growth conditions follow the existence a constant C > 0 such that
 
+ A(u n k (t)), u n k (t) − u(t)V )
p p
u n k (t) − u(t) V ≤ C(1 + u(t) V

for a.e. t ∈ (τ + ε, T ) and any k ≥ 1. Therefore,

T
p
lim u n k (t) − u(t) V dt = 0.
k→+∞
τ +ε

We obtain a contradiction.
The theorem is proved.

6.1.2 Convergence Results for Nonautonomous Evolution

Inclusions

For evolution triple (Vi ; H ; Vi∗ )1 and multi-valued map Ai : R+ × V ⇒ V ∗ , i =

1, 2, . . . , N , N = 1, 2, . . . we consider a problem of long-time behavior of all glob-
ally defined weak solutions for nonautonomous evolution inclusion


N
y  (t) + Ai (t, y(t))  0̄, (6.25)
i=1

as t → +∞. Let ·, ·Vi : Vi ∗ × Vi → R be the pairing in Vi ∗ × Vi , that coincides

on H × Vi with the inner product (·, ·) in the Hilbert space H .

1 I.e.,
Vi is a real reflexive separable Banach space continuously and densely embedded into a real
Hilbert space H , H is identified with its topologically conjugated space H ∗ , Vi∗ is a dual space to Vi .
So, there is a chain of continuous and dense embeddings: Vi ⊂ H ≡ H ∗ ⊂ Vi∗ (see, for example,
Gajewski, Gröger, and Zacharias [12, Chap. I]).
134 6 Strongest Convergence Results for Weak Solutions …

Note that Problem (6.25) arises in many important models for distributed para-
meter control problems and that large class of identification problems enter this
formulation.
Throughout this subsection we suppose that the listed below assumptions hold:
Assumption I. Let pi ≥ 2, qi > 1 are such that p1i + q1i = 1, for each for
i = 1, 2, . . . , N , and the embedding Vi ⊂ H is compact one, for some for i =
1, 2, . . . , N .
Assumption II (Grows Condition). There exist a translation uniform integrable
1 (R+ ) function c1 : R+ → R+ and a constant c2 > 0 such that
(t.u.i.) function in L loc

N q

N
p
max di Vi ∗ ≤ c1 (t) + c2 u Vi
i=1
i=1

for any u ∈ Vi , di ∈ Ai (t, u), i = 1, 2, . . . , N , and a.e. t > 0.

Assumption III (Sign Assumption). There exist a constant α > 0 and a t.u.i. in
1 (R+ ) function β : R+ → R+ such that
L loc


N 
N
p
di , uVi ≥ α u Vi − β(t)
i=1 i=1

for any u ∈ Vi , di ∈ Ai (t, u), i = 1, 2, . . . , N , and a.e. t > 0.

Assumption IV (Strong Measurability). If C ⊆ Vi ∗ is a closed set, then the set
{(t, u) ∈ (0, +∞) × Vi : Ai (t, u) ∩ C = ∅} is a Borel subset in (0, +∞) × Vi .
Assumption V (Pointwise Pseudomonotonicity). Let for each i = 1, 2, . . . , N
and a.e. t > 0 two assumptions hold:
(a) for every u ∈ Vi the set Ai (t, u) is nonempty, convex, and weakly compact one
in Vi ∗ ;
(b) if a sequence {u n }n≥1 converges weakly in Vi towards u ∈ Vi as n → +∞,
dn ∈ Ai (t, u n ) for any n ≥ 1, and lim supdn , u n − uVi ≤ 0, then for any ω ∈ Vi
n→+∞
there exists d(ω) ∈ Ai (t, u) such that

lim inf dn , u n − ωVi ≥ d(ω), u − ωVi .

n→+∞

Let 0 ≤ τ < T < +∞. As a weak solution of evolution inclusion (6.25) on the
interval [τ, T ] we consider an element u(·) of the space ∩i=1 N
L pi (τ, T ; Vi ) such that
for some di (·) ∈ L qi (τ, T ; Vi ∗ ), i = 1, 2, . . . , N , it is fulfilled:

T N 

T

− (ξ (t), y(t))dt + di (t), ξ(t)Vi dt = 0 ∀ξ ∈ C0∞ ([τ, T ]; Vi ), (6.26)
τ i=1 τ

and di (t) ∈ Ai (t, y(t)) for each i = 1, 2, . . . , N and a.e. t ∈ (τ, T ).

6.1 First Order Differential-Operator Equations and Inclusions 135

For fixed nonnegative τ and T , τ < T , let us consider


N
∗
X τ,T = ∩i=1
N
L pi (τ, T ; Vi ), X τ,T = L qi (τ, T ; Vi ∗ ), Wτ,T = {y ∈ X τ,T | y  ∈ X τ,T
∗
},
i=1
∗ ∗
Aτ,T : X τ,T ⇒ X τ,T , Aτ,T (y) = {d ∈ X τ,T | d(t) ∈ A(t, y(t)) for a.e. t ∈ (τ, T )},

N
where y  is a derivative of an element u ∈ X τ,T in the sense of D([τ, T ]; i=1 Vi ∗ )
(see, for example, Gajewski, Gröger, and Zacharias [12, Definition IV.1.10]). Note
that the space Wτ,T is a reflexive Banach space with the graph norm of a deriv-
∗
ative u Wτ,T = u X τ,T + u  X τ,T ∗ , u ∈ W
τ,T . Let ·, · X τ,T : X τ,T × X τ,T → R be
∗
the pairing in X τ,T × X τ,T , that coincides on L 2 (τ, T ; H ) × X τ,T with the inner
T
product in L 2 (τ, T ; H ), i.e., u, v X τ,T = (u(t), v(t))dt for any u ∈ L 2 (τ, T ; H )
τ
and v ∈ X τ,T . Gajewski, Gröger, and Zacharias [12, Theorem IV.1.17] provide that
the embedding Wτ,T ⊂ C([τ, T ]; H ) is continuous and dense one. Moreover,

T
(u(T ), v(T )) − (u(τ ), v(τ )) = u  (t), v(t)Vi + v  (t), u(t)Vi dt, (6.27)
τ

for any u, v ∈ Wτ,T .

Migórski [22, Lemma 7, p. 516] (see paper and references therein) and
Assumptions I–V provide the existence of multi-valued Nemitsky operator Aτ,T :
∗ N
X τ,T ⇒ X τ,T for i=1 Ai that satisfies the following properties:
Property I. The mapping Aτ,T transforms an each bounded set in X τ,T onto
∗
bounded subset of X τ,T ;
Property II. There exist positive constants C1 = C1 (τ, T ) and C2 = C2 (τ, T )
p
such that d, y X τ,T ≥ C1 y X τ,T − C2 for any y ∈ X τ,T and d ∈ Aτ,T (y).
∗
Property III. The multi-valued mapping Aτ,T : X τ,T ⇒ X τ,T is (generalized)
pseudomonotone on Wτ,T , i.e., (a) for every y ∈ X τ,T the set Aτ,T (y) is a
∗
nonempty, convex and weakly compact one in X τ,T ; (b) Aτ,T is upper semi-
∗
continuous from every finite dimensional subspace X τ,T into X τ,T endowed with
∗
the weak topology; (c) if a sequence {yn , dn }n≥1 ⊂ Wτ,T × X τ,T converges weakly
∗ ∗
in Wτ,T × X τ,T towards (y, d) ∈ Wτ,T × X τ,T , dn ∈ Aτ,T (yn ) for any n ≥ 1, and
lim supdn , yn − y X τ,T ≤ 0, then d ∈ Aτ,T (y) and lim dn , yn  X τ,T = d, y X τ,T .
n→+∞ n→+∞

 NFormula (6.26) and definition of the derivative for an element from D([τ, T ];
∗
V
i=1 i ) yield that each weak solution y ∈ X τ,T of Problem (6.25) on [τ, T ] belongs
to the space Wτ,T and y  + Aτ,T (y)  0̄. Vice versa, if y ∈ Wτ,T satisfies the last
inclusion, then y is a weak solution of Problem (6.25) on [τ, T ].
Assumption I, Properties I–III, and Denkowski, Migórski, and Papageorgiou [10,
Theorem 1.3.73] (see also Zgurovsky, Mel’nik, and Kasyanov [30, Chap. 2] and
136 6 Strongest Convergence Results for Weak Solutions …

references therein) provide the existence of a weak solution of Cauchy problem

(6.25) with initial data y(τ ) = y (τ ) on the interval [τ, T ], for any y (τ ) ∈ H .
For fixed τ and T , such that 0 ≤ τ < T < +∞, we denote

Dτ,T (y (τ ) ) = {y(·) | y is a weak solution of (6.25) on [τ, T ], y(τ ) = y (τ ) }, y (τ ) ∈ H.

We remark that Dτ,T (y (τ ) ) = ∅ and Dτ,T (y (τ ) ) ⊂ Wτ,T , if 0 ≤ τ < T < +∞ and

y (τ ) ∈ H . Moreover, the concatenation of Problem (6.25) weak solutions is a
weak solutions too, i.e., if 0 ≤ τ < t < T , y (τ ) ∈ H , y(·) ∈ Dτ,t (y (τ ) ), and v(·) ∈
Dt,T (y(t)), then 
y(s), s ∈ [τ, t],
z(s) =
v(s), s ∈ [t, T ],

belongs to Dτ,T (y (τ ) ); cf. Zgurovsky et al. [31, pp. 55–56].

Gronwall lemma provides that for any finite time interval [τ, T ] ⊂ R+ each weak
solution y of Problem (6.25) on [τ, T ] satisfies estimates
 t N 
 t  s
p
y(t) 2
H −2 β(ξ )dξ + 2α y(ξ ) Vi dξ ≤ y(s) 2
H −2 β(ξ )dξ,
0 i=1 s 0

 (6.28)
t
2 −2αγ (t−s)
y(t) 2
H ≤ y(s) He +2 (β(ξ ) + αγ )e−2αγ (t−ξ ) dξ, (6.29)
s

where t, s ∈ [τ, T ], t ≥ s; γ is a constant that does not depend on y, s, and t; cf.

Zgurovsky et al. [31, p. 56]. In the proof of (6.29) we used the inequality u 2H − 1 ≤
p
u H for any u ∈ H .
Therefore, any weak solution y of Problem (6.25) on a finite time interval
[τ, T ] ⊂ R+ can be extended to a global one, defined on [τ, +∞). For arbitrary
τ ≥ 0 and y (τ ) ∈ H let Dτ (y (τ ) ) be the set of all weak solutions (defined on
[τ, +∞)) of Problem (6.25) with initial data y(τ ) = y (τ ) . Let us consider the fam-
ily Kτ+ = ∪ y (τ ) ∈H Dτ (y (τ ) ) of all weak solutions of Problem (6.25) defined on the
semi-infinite time interval [τ, +∞).
Assumptions (II) and (III) yield that there exist a positive constant α  > 0 and a
t.u.i. function c in L loc
1 (R+ ) such that A(t, u) ⊆ Ac (t) (u) for each u ∈ ∩i=1 Vi and
N

a.e. t > 0, where



N 
N
N

∗  p q 
Ac (t) (u) := pi : pi ∈ Vi ,  pi , uVi ≥ α max u Vi ; p Vi ∗ − c (t) .
i=1
i=1 i=1

Let H (c ) be the hull of t.u.i. function c in L loc 

1,w (R+ ), i.e., H (c ) = cl L loc
1 (R+ )
{c ( · +
h) : h ≥ 0}. This is a weakly compact set in L 1 (R+ ); Gorban et al. [13].
loc

Let us consider the family of problems

6.1 First Order Differential-Operator Equations and Inclusions 137

y  = Aσ (y), σ ∈ Σ := H (c ). (6.30)

To each σ ∈ Σ there corresponds a space of all globally defined on [0, +∞) weak
solutions Kσ+ ⊂ C loc (R+ ; H ) of Problem (6.30). We set KΣ+ = ∪σ ∈Σ Kσ+ .
We remark that any element from KΣ+ satisfies prior estimates.

Lemma 6.5 There exist positive constants c3 and c4 such that for any σ ∈ Σ and
y ∈ Kσ+ the inequalities hold:

t N 

t s
σ (ξ )dξ + 2α 
p
y(t) 2
H −2 y(ξ ) Vi dξ ≤ y(s) 2
H −2 σ (ξ )dξ,
0 i=1 s 0
(6.31)
t
2 −c3 (t−s)
y(t) 2
H ≤ y(s) He + c4 σ (ξ )e−c3 (t−ξ ) dξ, (6.32)
s

for any t ≥ s ≥ 0.

Proof The proof naturally follows from conditions for the parameters of Problem
(6.30) and Gronwall lemma.

Let us provide the result characterizing the compactness properties of solutions

for the family of Problems (6.30).

Theorem 6.2 Let {yn }n≥1 ⊂ KΣ+ be an arbitrary sequence, that is bounded in
L ∞ (R+ ; H ). Then there exist a subsequence {yn k }k≥1 ⊂ {yn }n≥1 and an element
y ∈ KΣ+ such that

max yn k (t) − y(t) H → 0, k → +∞, (6.33)

t∈[τ,T ]

for any finite time interval [τ, T ] ⊂ (0, +∞).

Proof For any n ≥ 1 there exists σn ∈ Σ such that yn ∈ Kσ+ . Furthermore, the
definition of weak solution of evolution inclusion yields that forany n ≥ 1 and
i = 1, 2, . . . , N , there exists dn,i ∈ L qloc
i
(R+ ; Vi ∗ ) such that yn (t) + i=1 N
dn,i (t) = 0̄
for a.e. t > 0. The definition of Aσ and estimates (6.31) and  (6.32) provide that the
∗
sequence {yn , yn , dn,i }n≥1 is bounded in ∩i=1 N
pi (R+ ; Vi ) ×
L loc N
i=1 L qi (R+ ; Vi ) ×
loc
∗
L qi (R+ ; Vi ), i = 1, 2, . . . , N . Since Σ is a weakly compact set in L 1 (R+ ),
loc loc

Banach–Alaoglu theorem (cf. Zgurovsky, Mel’nik, and Kasyanov [30, Chap. 1];
Kasyanov [15]) yields that there exist a subsequence {yn k , dn k ,i }k≥1 ⊂ {yn , dn }n≥1
and elements di ∈ L qloc (R+ ; Vi ∗ ), y ∈ ∩i=1N
pi (R+ ; Vi ), and σ ∈ Σ, such that y ∈
L loc 
N ∗
i

i=1 L q (R+ ; Vi ) and for each i = 1, 2, . . . , N the following convergence hold:

loc
138 6 Strongest Convergence Results for Weak Solutions …

yn k → y weakly in ∩i=1N
L loci (R+ ; Vi ),
 N ploc

yn k → y 
weakly in i=1 L qi (R+ ; Vi ∗ ),
dn k ,i → di weakly in L qloc
i
(R+ ; Vi ∗ ),
yn k → y weakly in C loc (R+ ; H ), (6.34)
yn k → y 2 (R+ ; H ),
in L loc
yn k (t) → y(t) in H for a.e. t > 0,
σn k → σ 1 (R+ ), k → +∞.
weakly in L loc

Formula (6.33) follows from Zgurovsky et al. [31, Steps 1 and 5, p. 58]. We remark
that in the proof we need to consider continuous and nonincreasing (by Lemma 6.5)
functions on R+ :

t t
Jk (t) = yn k (t) 2
H −2 σn k (ξ )dξ, J (t) = y(t) 2
H −2 σ (ξ )dξ, k ≥ 1.
0 0
(6.35)
The two last statements in (6.34) imply Jk (t) → J (t), as k → +∞, for a.e. t > 0.
The definition of a weak solutionof evolution inclusion (cf. Zgurovsky et al. [31,
p. 58]) and (6.34) yield y  (t) = − i=1
N
di (t) for a.e. t > 0. To finish the proof it is
necessary to provide that


N
di (t) ∈ Aσ (t) (y(t)) for a.e. t > 0. (6.36)
i=1

Let ϕ ∈ C0∞ ((0, +∞)), ϕ ≥ 0. Then

 
 N 
ϕ(t) α  max
p q
y(t) Vi ; di (t) Vi ∗ − σ (t) dt ≤
R+ i=1
 
 N 
ϕ(t) α  max
p q
lim inf yn k (t) Vi ; dn k ,i (t) Vi ∗ − σn k (t) dt ≤
k→+∞ R+ i=1
 
N 
1 d
lim ϕ(t) yn k (t) 2H ϕ(t)dt =
dn k ,i (t), yn k (t)V dt = lim
k→+∞ R k→+∞ 2 R dt
+ i=1 +

 N 
1 d
y(t) 2H ϕ(t)dt = ϕ(t)di (t), y(t)Vi dt,
2 R+ dt i=1 R+

where the first inequality holds, because the convex functional

 
 N
ϕ(t) α  max
p q
(y, d) → y(t) Vi ; di (t) Vi ∗ dt
R+ i=1

∗
is weakly lower semi-continuous on ∩i=1 N
pi (R+ ; Vi ) × L q1 (R+ ; V1 ) × L q2
L loc loc loc

(R+ ; V2 ∗ ) × . . . × L qlocN (R+ ; VN ∗ ); the second inequality follows from the

6.1 First Order Differential-Operator Equations and Inclusions 139

definition of Aσ ; the first and the third

 N equalities follow from formula (6.27), because
yn k (t) + i=1
N
dn k ,i (t) = y  (t) + i=1 di (t) = 0̄ for any k ≥ 1 and a.e. t > 0; the
second equality holds, because yn k → y in L loc 2 (R+ ; H ), as k → +∞. As a nonneg-
ative function ϕ ∈ C0∞ ((0, +∞)) is an arbitrary, then, by definition of Aσ , formula
(6.36) holds.
The theorem is proved.

6.2 Second Order Operator Differential Equations

and Inclusions

In this section we consider damped wave equations with possibly nonmonotone dis-
continuous nonlinearities. Then, we generalize the results to the autonomous second
order differential-operator inclusions with possibly nonmonotone potential.
Let β > 0 be a constant, Ω ⊂ Rn be a bounded domain with sufficiently smooth
boundary ∂Ω. Consider the problem

u tt + βu t − u + f (u) = 0,
(6.37)
u|∂Ω = 0,

where u(x, t) is unknown state function defined on Ω × R+ ; f : R → R is an inter-

action function such that
f (u)
lim > −λ1 , (6.38)
|u|→∞ u

where λ1 is the first eigenvalue for − in H01 (Ω);

∃ D≥0: | f (u)| ≤ D(1 + |u|), ∀u ∈ R. (6.39)

Further, we use such denotation

f (s) := lim sup f (t), f (s) := lim f (t), G(s) := [ f (s), f (s)], s ∈ R.
t→s t→s

Let us set V = H01 (Ω) and H = L 2 (Ω). The space X = V × H is a phase space of
Problem (6.37). For the Hilbert space X as (·, ·) X and · X denote the inner product
and the norm in X respectively.

Definition 6.1 Let T > 0, τ < T . The function ϕ(·) = (u(·), u t (·))T ∈
L ∞ (τ, T ; X ) is called a weak solution of Problem (6.37) on (τ, T ) if for a.e. (x, t) ∈
Ω × (τ, T ), there exists l = l(x, t) ∈ L 2 (τ, T ; L 2 (Ω)) l(x, t) ∈ G(u(x, t)), such
that ∀ψ ∈ H01 (Ω), ∀η ∈ C0∞ (τ, T ),
140 6 Strongest Convergence Results for Weak Solutions …

Fig. 6.1 Difference of nondecreasing functions

T T
− (u t , ψ) H ηt dt + (β(u t , ψ) H + (u, ψ)V + (l, ψ) H )ηdt = 0. (6.40)
τ τ

The main goal of the manuscript is to obtain the existence of the global attractor
generated by the weak solutions of Problem (6.37) in the phase space X .

Lemma 6.6 (Zgurovsky et al. [31]) For any ϕτ = (u 0 , u 1 )T ∈ X and τ < T there
exists a weak solution ϕ(·) of Problem (6.37) on (τ, T ) such that ϕ(τ ) = ϕτ .

Further, we assume that

f (s) = f 1 (s) − f 2 (s), s ∈ R,

where f i : R → R, i = 1, 2, are nondecreasing functions (see Fig. 6.1).

We remark that

[ f (s), f (s)] ⊆ [ f 1 (s), f 1 (s)] − [ f 2 (s), f 2 (s)], s ∈ R.

Thus we consider more general evolution inclusion


u tt + βu t − u + [ f 1 (u), f 1 (u)] − [ f 2 (u), f 2 (u)]  0,
(6.41)
u|∂Ω = 0.

Let us set
s 
G i (s) := f i (ξ )dξ, Ji (u) := G i (u(x))d x, J (u) = J1 (u) − J2 (u), u ∈ H, i = 1, 2.
0 Ω

The functionals G i and Ji are locally Lipschitz and regular; Clarke [8, Chap. I]. Thus
the next result holds.

Lemma 6.7 (Kasyanov et al. [31]) Let u ∈ C 1 ([τ, T ]; H ). Then for a.e. t ∈ (τ, T ),
the functions Ji ◦ u are classically differentiable at the point t. Moreover,
6.2 Second Order Operator Differential Equations and Inclusions 141

d
(Ji ◦ u)(t) = ( p, u t (t)) ∀ p ∈ ∂ Ji (u(t)), i = 1, 2,
dt

and d
dt
(Ji ◦ u)(·) ∈ L 1 (τ, T ).

Consider WτT = C([τ, T ]; X ). Lebourgue’s mean value theorem (see Clarke [8,
Chap. 2]) provides the existence of constants c1 , c2 > 0 and μ ∈ (0, λ1 ) such that
μ
|J (u)| ≤ c1 (1 + u H ),
2
J (u) ≥ − u 2
H − c2 ∀u ∈ H. (6.42)
2
The weak solution of Problem (6.37) with initial data

u(τ ) = a, u  (τ ) = b (6.43)

on the interval [τ, T ] exists for any a ∈ V, b ∈ H. It follows from Zadoianchuk and
Kasyanov [29, Theorem 1.4]. Thus the next lemma holds true (see Kasyanov et al.
[16, Lemma 3.2]).

Lemma 6.8 (Kasyanov et al. [16, Lemma 3.2]) For any τ < T, a ∈ V, b ∈ H ,
Cauchy Problem (6.37), (6.43) has the weak solution (u, u t )T ∈ L ∞ (τ, T ; X ). More-
over, each weak solution (u, u t )T of Cauchy Problem (6.37), (6.43) on the interval
[τ, T ] belongs to the space C([τ, T ]; X ) and u tt ∈ L 2 (τ, T ; V ∗ ).

For any ϕτ = (a, b)T ∈ X , denote

  
 (u, u t )T is a weak solution of Problem (6.37) on [τ, T ],
Dτ,T (ϕτ ) = (u(·), u t (·))T  .
 u(τ ) = a, u t (τ ) = b

From Lemma 6.8 it follows that Dτ,T (ϕτ ) ⊂ C([τ, T ]; X ) = WτT . Let us check that
translation and concatenation of weak solutions are weak solutions too.
Lemma 6.9 If τ < T , ϕτ ∈ X , ϕ(·) ∈ Dτ,T (ϕτ ), then ∀s ψ(·) = ϕ(· + s) ∈
Dτ −s,T −s (ϕτ ). If τ < t < T , ϕτ ∈ X , ϕ(·) ∈ Dτ,t (ϕτ ) and ψ(·) ∈ Dt,T (ϕτ ), then

ϕ(s), s ∈ [τ, t],
θ (s) = ∈ Dτ,T (ϕτ ).
ψ(s), s ∈ [t, T ]

Proof The proof is trivial (see Kasyanov et al. [16, Lemma 4.1]).

Let ϕ = (a, b)T ∈ X and

1
V (ϕ) = ϕ 2
X + J1 (a) − J2 (a). (6.44)
2
142 6 Strongest Convergence Results for Weak Solutions …

Lemma 6.10 Let τ < T , ϕτ ∈ X , ϕ(·) = (u(·), u t (·))T ∈ Dτ,T (ϕτ ). Then V ◦ ϕ :
[τ, T ] → R is absolutely continuous and for a.e. t ∈ (τ, T ), dtd V (ϕ(t)) =
−β u t (t) 2H .

Proof Let −∞ < τ < T < +∞, ϕ(·) = (u(·), u t (·))T ∈ WτT be an arbitrary weak
solution of Problem (6.37) on (τ, T ). Since ∂ J (u(·)) ⊂ L 2 (τ, T ; H ), from Temam
[26] and Zgurovsky et al. [31, Chap. 2] we obtain that the function t → u t (t) 2H +
u(t) 2V is absolutely continuous and for a.e. t ∈ (τ, T ),
 
1 d
2 dt
u t (t) 2
H + u(t) 2
V = (u tt (t) − u(t), u t (t))H =
(6.45)
= −β u t (t) 2
H − (d1 (t), u t (t))H + (d2 (t), u t (t))H ,

where di (t) ∈ ∂ Ji (u(t)) for a.e. t ∈ (τ, T ) and di (·) ∈ L 2 (τ, T ; H ), i = 1, 2. As

u(·) ∈ C 1 ([τ, T ]; H ) and Ji : H → R, i = 1, 2 is regular and locally Lipschitz, due
to Lemma 6.7 we obtain that for a.e. t ∈ (τ, T ), ∃ dtd (Ji ◦ u)(t), i = 1, 2. Moreover,
d
dt
(Ji ◦ u)(·) ∈ L 1 (τ, T ), i = 1, 2 and for a.e. t ∈ (τ, T ), ∀ p ∈ ∂ Ji (u(t)),

d
(Ji ◦ u)(t) = ( p, u t (t)) H , i = 1, 2.
dt

In particular for a.e. t ∈ (τ, T ), dtd (Ji ◦ u)(t) = (di (t), u t (t)) H . Taking into account
(6.45) we finally obtain the necessary statement.
This completes the proof.

Lemma 6.11 Let T > 0. Then any weak solution of Problem (6.37) on [0, T ] can
be extended to a global one defined on [0, +∞).

Proof The statement of this lemma follows from Lemmas 6.8–6.10, (6.42) and from
the next estimates

∀τ < T, ∀t ∈ [τ, T ], ∀ϕτ ∈ X, ∀ϕ(·) = (u(·), u t (·))T ∈ Dτ,T (ϕτ ),

 
2c1 + 1 + 2c1
λ1
u(τ ) 2
+ u t (τ )
2
≥ 2V (ϕ(τ )) ≥ 2V (ϕ(t)) =
V
H 
= u(t) 2
V + u t (t) 2
H + 2J (u(t)) ≥ 1 − λμ1 u(t) 2V + u t (t) 2H − 2c2 .

The lemma is proved.

For an arbitrary ϕ0 ∈ X let D(ϕ0 ) be the set of all weak solutions (defined on
[0, +∞)) of Problem (6.37) with initial data ϕ(0) = ϕ0 . We remark that from the
proof of Lemma 6.11 we obtain the next corollary.

Corollary 6.1 For any ϕ0 ∈ X and ϕ ∈ D(ϕ0 ), the next inequality is fulfilled

λ1 + 2c1 2(c1 + c2 )λ1

ϕ(t) 2
≤ ϕ(0) 2
+ ∀t > 0. (6.46)
X
λ1 − μ X
λ1 − μ
6.2 Second Order Operator Differential Equations and Inclusions 143

From Corollary 6.1 in a standard way we obtain such statement.

Theorem 6.3 Let τ < T, {ϕn (·)}n≥1 ⊂ WτT be an arbitrary sequence of weak solu-
tions of Problem (6.37) on [τ, T ] such that ϕn (τ ) → ϕτ weakly in X , n → +∞, and
let {tn }n≥1 ⊂ [τ, T ] be a sequence such that tn → t0 , n → +∞. Then there exists
ϕ ∈ Dτ,T (ϕτ ) such that up to a subsequence ϕn (tn ) → ϕ(t0 ) weakly in X , n → +∞.

Proof We prove this theorem in several steps.

Step 1. Let τ < T, {ϕn (·) = (u n (·), u n (·))}n≥1 ⊂ WτT be an arbitrary sequence of
weak solutions of Problem (6.37) on [τ, T ] and {tn }n≥1 ⊂ [τ, T ] such that

ϕn (τ ) → ϕτ weakly in X, tn → t0 , n → +∞. (6.47)

In virtue of Corollary 6.1 we have that {ϕn (·)}n≥1 is bounded on WτT ⊂

L ∞ (τ, T ; X ). Therefore up to a subsequence {ϕn k (·)}k≥1 ⊂ {ϕn (·)}n≥1 we have

u n k → u weakly star in L ∞ (τ, T ; V ), k → +∞,

u n k → u  weakly star in L ∞ (τ, T ; H ), k → +∞,
u n k → u  weakly star in L ∞ (τ, T ; V ∗ ), k → +∞,
dn k ,i → di weakly star in L ∞ (τ, T ; H ), i = 1, 2, k → +∞,
(6.48)
u n k → u in L 2 (τ, T ; H ), k → +∞,
u n k (t) → u(t) in H for a.e. t ∈ [τ, T ], k → +∞,
u n k (t) → u  (t) in V ∗ for a.e. t ∈ (τ, T ), k → +∞,
u n k → u weakly in L 2 (τ, T ; V ∗ ), k → +∞,

where ∀n ≥ 1 dn,i ∈ L 2 (τ, T ; H ) and

u n (t) + βu n (t) + dn,1 (t) − dn,2 (t) − u n (t) = 0̄,

(6.49)
dn,i (t) ∈ ∂ Ji (u n (t)), i = 1, 2, for a.e. t ∈ (τ, T ).

Step 2. ∂ Ji , i = 1, 2 are demiclosed. So, by a standard way we get that di (·) ∈

∂ Ji (u(·)), i = 1, 2, ϕ := (u, u  ) ∈ Dτ,T (ϕτ ) ⊂ WτT .
Step 3. From (6.48) it follows that for arbitrary fixed h ∈ V the sequences of
real functions (u n k (·), h) H , (u n k (·), h) H : [τ, T ] → R are uniformly bounded and
equipotentionally continuous. Taking into account (6.48), (6.46) and density of the
embedding V ⊂ H we obtain that u n k (tn k ) → u  (t0 ) weakly in H and u n k (tn k ) →
u(t0 ) weakly in V as k → +∞.
The theorem is proved.

Theorem 6.4 Let τ < T, {ϕn (·)}n≥1 ⊂ WτT be an arbitrary sequence of weak solu-
tions of Problem (6.37) on [τ, T ] such that ϕn (τ ) → ϕτ strongly in X , n → +∞,
then up to a subsequence ϕn (·) → ϕ(·) in C([τ, T ]; X ), n → +∞.

Proof Let τ < T, {ϕn (·) = (u n (·), u n (·))T }n≥1 ⊂ WτT be an arbitrary sequence of
weak solutions of Problem (6.37) on [τ, T ] and {tn }n≥1 ⊂ [τ, T ]:
144 6 Strongest Convergence Results for Weak Solutions …

ϕn (τ ) → ϕτ strongly in X, n → +∞. (6.50)

From Theorem 6.3 we have that there exists ϕ ∈ Dτ,T (ϕτ ) such that up to the sub-
sequence {ϕn k (·)}k≥1 ⊂ {ϕn (·)}n≥1 ϕn (·) → ϕ(·) weakly in X , uniformly on [τ, T ],
k → +∞. Let us prove that

ϕn k → ϕ in WτT , k → +∞. (6.51)

By contradiction, suppose the existence of L > 0 and the subsequence {ϕk j } j≥1 ⊂
{ϕn k }k≥1 such that ∀ j ≥ 1,

max ϕk j (t) − ϕ(t) X = ϕk j (t j ) − ϕ(t j ) X ≥ L.

t∈[τ,T ]

Without loss of generality we suggest that t j → t0 ∈ [τ, T ], j → +∞. Therefore

by virtue of a continuity of ϕ : [τ, T ] → X we have

lim ϕk j (t j ) − ϕ(t0 ) X ≥ L. (6.52)

j→+∞

On the other hand, we prove that

ϕk j (t j ) → ϕ(t0 ) in X, j → +∞. (6.53)

First we remark that

ϕk j (t j ) → ϕ(t0 ) weakly in X, j → +∞ (6.54)

(see Theorem 6.3 for details). Secondly let us prove that

lim sup ϕk j (t j ) X ≤ ϕ(t0 ) X. (6.55)

j→+∞

Since J is sequentially weakly continuous, V is sequentially weakly lower semi-

continuous on X . Hence we obtain
V (ϕ(t0 )) ≤ lim V (ϕk j (t j )),
j→+∞
t0 tj (6.56)
u  (s) 2
H ds ≤ lim u k j (s) 2
H ds
τ j→+∞ τ

and
⎛ ⎞
t0 t j
V (ϕ(t0 )) + β u  (s) 2
H ds ≤ lim ⎝V (ϕk j (t j )) + β u k j (s) ⎠
H ds .
2
j→+∞
τ τ
(6.57)
6.2 Second Order Operator Differential Equations and Inclusions 145

Since by the energy equation both sides of (6.57) equal V (ϕ(τ )) (see Lemma 6.10),
it follows from (6.56) that V (ϕk j (t j )) → V (ϕ(t0 )), j → +∞ and (6.55). Conver-
gence (6.53) directly follows from (6.54), (6.55) and Gajewski et al. [12, Chap. I]. To
finish the proof of the theorem we remark that (6.53) contradicts (6.52). Therefore
(6.51) holds.
The theorem is proved.

Now let us consider autonomous second order differential-operator inclusions

with possibly nonmonotone potential. Let V and H be real separable Hilbert spaces
such that V ⊂ H with compact and dense embedding. Let V ∗ be the dual space of
V . We identify H with H ∗ (dual space of H ). For the linear operators A : V → V ∗ ,
B : V → V ∗ and locally Lipschitz functional J : H → R we consider a problem
of investigation of dynamics for all weak solutions defined for t ≥ 0 of non-linear
second order autonomous differential-operator inclusion:

u  (t) + Au  (t) + Bu(t) + ∂ J1 (u(t)) − ∂ J2 (u(t))  0̄. (6.58)

We need the following hypotheses:.

H (A) A : V → V ∗ is a linear symmetric such that ∃c A > 0 : Av, vV ≥ c A v 2V
∀v ∈ V ;
H (B) B : V → V ∗ is linear, symmetric and ∃c B > 0 : Bv, vV ≥ c B v 2V ∀v ∈
V 2;
H (J ) Ji : H → R, i = 1, 2, is a function such that
(i) Ji (·) is locally Lipschitz and regular [8, Chap. II], i.e.,
• for any x, v ∈ H , the usual one-sided directional derivative J  (x; v) =
lim J (x+tv)−J
t
(x)
exists,
t0
J (y+tv)−J (y)
• for all x, v ∈ H , J  (x; v) = J ◦ (x; v), where J ◦ (x; v) = lim t
;
y→x, t0
(ii) ∃c1 > 0: sup d H ≤ c1 (1 + v H) ∀v ∈ H ;
d∈∂ J1 (v)−∂ J2 (v)
(iii) ∃c2 > 0:

inf (d, v) H ≥ −λ v 2
H − c2 ∀v ∈ H,
d∈∂ J1 (v)−∂ J2 (v)

where ∂ Ji (v) = { p ∈ H | ( p, w) H ≤ Ji◦ (v; w) ∀w ∈ H } denotes the Clarke sub-

differential of Ji (·) at a point v ∈ H (see [8] for details), λ ∈ (0, λ1 ), λ1 > 0:
v 2V ≥ λ1 v 2H ∀v ∈ V , i = 1, 2.
We note that in (6.89) we can consider g = 0̄. Indeed, let g ∈ V ∗ , then u ∗ =
B g ∈ V ⊂ H . If u(·) is a weak solution of (6.89), then ū(·) = u(·) − u ∗ is a weak
−1

2 We remark that operators A and B are continuous on V [12, Chap. III].

146 6 Strongest Convergence Results for Weak Solutions …

solution of

ū  (t) + Aū  (t) + B ū(t) + ∂ J3 (ū(t)) − ∂ J4 (ū(t))  0̄ a.e. t > 0,

where Ji+2 (·) = Ji (· + u ∗ ), i = 1, 2, satisfies H (J ) with respective parameters.

Thus, to simplify our conclusions, without loss of generality, further we will consider
problem (6.58).
The phase space for Problem (6.58) is the Hilbert space:

E = {(a, b)T | a ∈ V, b ∈ H },
 
a
where (a, b)T = with (a, b)T E =( a 2
V + b H)
2 1/2
. Let −∞ < τ < T <
b
+∞.
Definition 6.2 The function (u(·), u  (·))T ∈ L ∞ (τ, T ; E) with u  (·) ∈ L 2 (τ, T ; V )
is called a weak solution for (6.58) on [τ, T ], if there exists d(·) ∈ L 2 (τ, T ; H ),
d(t) ∈ ∂ J1 (u(t)) − ∂ J2 (u(t)) for a.e. t ∈ (τ, T ), such that ∀ψ ∈ V , ∀η ∈ C0∞ (τ, T )

T T
 
− (u  (t), ψ) H η (t)dt + Au  (t), ψV + Bu(t), ψV + (d(t), ψ) H η(t)dt = 0.
τ τ

Evidently if (u(·), u  (·))T is a weak solution of (6.58) on [τ, T ], then u  (·) ∈

L 2 (τ, T ; V ∗ ), (u(·), u  (·))T ∈ C([τ, T ]; E) and d(·) ∈ L ∞ (τ, T ; H ).
We consider the class of functions WτT = C([τ, T ]; E). Further c1 , c2 , λ, λ1 ,
c A , c B we recall parameters of Problem (6.58). The main purpose of this work is
to investigate the long-time behavior (as t → +∞) of all weak solutions for the
problem (6.58) on [0, +∞).
To simplify our conclusions, since condition H (B), we suppose that

(u, v)V = Bu, vV , v 2

V = Bv, vV , c B = 1, ∀u, v ∈ V. (6.59)

Lebourg mean value theorem [8, Chap. 2] provides the existence of constants c3 , c4 >
0 and μ ∈ (0, λ1 ):
μ
|J (u)| ≤ c3 (1 + u H ),
2
J (u) ≥ − u 2
H − c4 ∀u ∈ H, (6.60)
2
where J (v) := J1 (v) − J2 (v), v ∈ H.
Lemma 6.12 Let J : H → R be a locally Lipschitz and regular functional, y ∈
C 1 ([τ, T ]; H ). Then for a.e. t ∈ (τ, T ) there exists dtd (J ◦ y)(t) = ( p, y  (t)) for all
p ∈ ∂ J (y(t)). Moreover, dtd (J ◦ y)(·) ∈ L 1 (τ, T ).
Proof Since y ∈ C 1 ([τ, T ]; H ) then y is strictly differentiable at the point t0 for any
t0 ∈ (τ, T ). Hence, taking into account the regularity of J and [8, Theorem 2.3.10],
6.2 Second Order Operator Differential Equations and Inclusions 147

we obtain that the functional J ◦ y is regular one at t0 ∈ (τ, T ) and ∂(J ◦ y)(t0 ) =
{( p, y  (t0 ))| p ∈ ∂ J (y(t0 ))}. On the other hand, since y ∈ C 1 ([τ, T ]; H ), J : H →
R is locally Lipschitz then J ◦ y : [τ, T ] → R is globally Lipschitz and therefore it
◦y)(t) d(J ◦y)(·)
is absolutely continuous. Hence for a.e. t ∈ (τ, T ) ∃ d(J dt , dt
∈ L 1 (τ, T )
t
and d
dξ
(J ◦ y)(ξ )dξ = (J ◦ y)(t) − (J ◦ y)(s) ∀τ ≤ s < t ≤ T . At that taking
s
into account the regularity of J ◦ y, note that (J ◦ y)◦ (t0 , v) = (J ◦ y) (t0 , v) =
d(J ◦y)(t0 )
dt
· v for a.e. t0 ∈ (τ, T ) and all v ∈ R. This implies that for a.e. t0 ∈ (τ, T )
∂(J ◦ y)(t0 ) = { d(J ◦y)(t
dt
0)
}.

At the inclusion (6.58) on [τ, T ] we associate the conditions

u(τ ) = a, u  (τ ) = b (6.61)

where a ∈ V and b ∈ H. From [29] we get the following lemma.

Lemma 6.13 For any τ < T, a ∈ V, b ∈ H the Cauchy problem (6.58), (6.61)
has a weak solution (y, y  )T ∈ L ∞ (τ, T ; E). Moreover, each weak solution (y, y  )T
of the Cauchy problem (6.58), (6.61) on the interval [τ, T ] belongs to the space
C([τ, T ]; E) and y  ∈ L 2 (τ, T ; V ), y  ∈ L 2 (τ, T ; V ∗ ).

Let us consider the next denotations: ∀ϕτ = (a, b)T ∈ E we set Dτ,T (ϕτ ) = {
(u(·), u  (·))T | (u, u  )T is a weak solution of (6.58) on [τ, T ], u(τ ) = a, u  (τ )
= b }. From Lemma 6.13 it follows that Dτ,T (ϕτ ) ⊂ C([τ, T ]; E) = WτT .
Let us check that translation and concatenation of weak solutions are weak solu-
tions too.

Lemma 6.14 If τ < T , ϕτ ∈ E, ϕ(·) ∈ Dτ,T (ϕτ ), then ψ(·) = ϕ(· + s) ∈

Dτ −s,T −s(ϕτ ) ∀s. If τ < t < T , ϕτ ∈ E, ϕ(·) ∈ Dτ,t (ϕτ ) and ψ(·) ∈ Dt,T (ϕ(t)), then
ϕ(s), s ∈ [τ, t],
θ (s) = belongs to Dτ,T (ϕτ ).
ψ(s), s ∈ [t, T ]

Proof The first part of the statement of this lemma follows from the autonomy of
the inclusion (6.58). The proof of the second part follows from the definition of the
solution of (6.58) and from that fact that z ∈ Wτ,T as soon as v ∈ Wτ,t , u ∈ Wt,T and
v(t) = u(t), where 
v(s), s ∈ [τ, t],
z(s) =
u(s), s ∈ [t, T ]

Let ϕ = (a, b)T ∈ E and

1
V (ϕ) = ϕ 2
E + J (a). (6.62)
2
148 6 Strongest Convergence Results for Weak Solutions …

Lemma 6.15 Let τ < T , ϕτ ∈ E, ϕ(·) = (y(·), y  (·))T ∈ Dτ,T (ϕτ ). Then V ◦ ϕ :
[τ, T ] → R is absolutely continuous and for a.e. t ∈ (τ, T ) dtd V (ϕ(t)) = −Ay  (t),
y  (t)V ≤ −γ y  (t) 2V , where γ > 0 depends only on parameters of Problem (6.58).3
Proof Let −∞ < τ < T < +∞, ϕ(·) = (y(·), y  (·))T ∈ WτT be an arbitrary weak
solution of (6.58) on (τ, T ). From [12, Chap. IV] we get that the function t →
y  (t) 2H + y(t) 2V is absolutely continuous and for a.e. t ∈ (τ, T )

1 d   
y (t) 2
H + y(t) 2
V = y  (t) + By(t), y  (t)V =
2 dt

= −Ay  (t), y  (t)V − (d(t), y  (t)) H ≤ −γ y  (t) 2

V − (d(t), y  (t)) H , (6.63)

where d(t) ∈ ∂ J1 (y(t)) − ∂ J2 (y(t)) for a.e. t ∈ (τ, T ), d(·) ∈ L 2 (τ, T√ ; H ) and γ >
0 depends only on parameters of Problem (6.58), in virtue of u → Au, uV is
equivalent norm on V . Since y(·) ∈ C 1 ([τ, T ]; H ) and Ji : H → R, i = 1, 2, is
regular and locally Lipschitz, then Lemma 6.12 yields that for a.e. t ∈ (τ, T ) there
exists dtd (Ji ◦ y)(t). Moreover, dtd (Ji ◦ y)(·) ∈ L 1 (τ, T ) and for a.e. t ∈ (τ, T ) and
all p ∈ ∂ J1 (y(t)) − ∂ J2 (y(t)) we have dtd (J ◦ y)(t) = ( p, y  (t)) H . In particular, for
a.e. t ∈ (τ, T ) dtd (J ◦ y)(t) = (d(t), y  (t)) H . Taking into account (6.63) we finally
obtain the necessary statement.
The lemma is proved.
Lemma 6.16 Let T0 > 0. If (u(·), u  (·))T is a weak solution of (6.58) on [0, T0 ], then
there exists an its extension on [0, +∞) (ū(·), ū  (·))T which is weak solution of (6.58)
on [0, +∞), i.e., (ū(·), ū  (·))T ∈ C(R+ ; E) ∩ L ∞ (R+ ; E) with ū  (·) ∈ L 2 (0, T ; V )
∀T > 0 and there exists d(·) ∈ L ∞ (0, +∞; H ), d(t) ∈ ∂ J1 (ū(t)) − ∂ J2 (ū(t)) for
a.e. t ∈ (0, +∞), such that ∀ψ ∈ V, ∀η ∈ C0∞ (0, +∞)

+∞
 +∞

− (ū  (t), ψ) H η (t)dt + [Aū  (t), ψV + B ū(t), ψV + (d(t), ψ) H ]η(t)dt = 0.
0 0

Proof The statement of this lemma follows from Lemmas 6.13–6.15, Conditions

 ∀τ < T , ∀ϕτ ∈ E, ∀ϕ(·) = (y(·), y (·)) ∈
T
(6.59), (6.60) and from the next 
estimates:

Dτ,T (ϕτ ), ∀t ∈ [τ, T ] 2c3 + 1 + λ1 2c3
y(τ ) V + y (τ ) H ≥ 2V (ϕ(τ )) ≥ 2V
2 2
 
(ϕ(t)) = y(t) 2V + y  (t) 2H + 2J (y(t)) ≥ 1 − λμ1 y(t) 2V + y  (t) 2H − 2c4 .
The lemma is proved.
For an arbitrary ϕ0 ∈ E let D(ϕ0 ) be the set of all weak solutions (defined on
[0, +∞)) of problem (6.58) with initial data ϕ(0) = ϕ0 . We remark that from the
proof of Lemma 6.16 we obtain the next corollary.
Corollary 6.2 For any ϕ0 ∈ E and ϕ ∈ D(ϕ0 ) the next inequality is fulfilled:

√
3 We remark that Au, uV is equivalent norm on V , generated by inner product Au, vV .
6.2 Second Order Operator Differential Equations and Inclusions 149

λ1 + 2c3 2(c3 + c4 )λ1

ϕ(t) 2
≤ ϕ(0) 2
+ ∀t > 0. (6.64)
E
λ1 − μ E
λ1 − μ

From Corollary 6.2 and Conditions H (A), H (B), H (J ) in a standard way we

obtain such proposition.

Theorem 6.5 Let τ < T, {ϕn (·)}n≥1 ⊂ WτT be an arbitrary sequence of weak solu-
tions of (6.58) on [τ, T ] such that ϕn (τ ) → ϕτ weakly in E, n → +∞. Then there
exist ϕ ∈ Dτ,T (ϕτ ) and {ϕn k (·)}k≥1 ⊂ {ϕn (·)}n≥1 such that ϕn k (·) → ϕ(·) weakly in
E uniformly on [τ, T ], k → +∞, i.e., ϕn k (tk ) → ϕ(t0 ) weakly in E, k → +∞, for
any {tk }k≥1 ⊂ [τ, T ] with tk → t0 , k → +∞.

Theorem 6.6 Let τ < T, {ϕn (·)}n≥1 ⊂ WτT be an arbitrary sequence of weak solu-
tions of (6.58) on [τ, T ] such that ϕn (τ ) → ϕτ strongly in E, n → +∞. Then there
exist ϕ ∈ Dτ,T (ϕτ ) such that up to a subsequence ϕn (·) → ϕ(·) in C([τ, T ]; E),
n → +∞.

Proof Let {ϕn (·) = (u n (·), u n (·))T }n≥1 ⊂ WτT be an arbitrary sequence of weak solu-
tions of (6.58) on [τ, T ] such that

ϕn (τ ) → ϕτ strongly in E, n → +∞. (6.65)

Let ϕ = (u(·), u  (·))T ∈ Dτ,T (ϕτ ) and {ϕn k (·)}k≥1 ⊆ {ϕn (·)}n≥1 as in Theorem 6.5.
It is important to remark that in the proof of Theorem 6.5, by using the inequality
(Lemma 6.15, Corollary 6.2, (6.60))
μ
γ u n (·) L 2 (τ,T ;V ) ≤ V (ϕn (τ )) − V (ϕn (T )) ≤ sup V (ϕn (τ )) + u n (T ) 2
H + c4 < +∞,
n≥1 2

we establish that

u n k (·) → u  (·) weakly in L 2 (τ, T ; V ), k → +∞.

Let us prove that

ϕn k → ϕ in WτT , k → +∞. (6.66)

By contradiction suppose the existence of L > 0 and subsequence {ϕk j } j≥1 ⊂

{ϕn k }k≥1 such that ∀ j ≥ 1 max ϕk j (t) − ϕ(t) E = ϕk j (t j ) − ϕ(t j ) E ≥ L. With-
t∈[τ,T ]
out loss of generality we suppose that t j → t0 ∈ [τ, T ], j → +∞. Therefore, by
virtue of the continuity of ϕ : [τ, T ] → E, we have

lim ϕk j (t j ) − ϕ(t0 ) E ≥ L. (6.67)

j→+∞

On the other hand we prove that

ϕk j (t j ) → ϕ(t0 ) in E, j → +∞. (6.68)

150 6 Strongest Convergence Results for Weak Solutions …

Firstly we remark that (Theorem 6.5)

ϕk j (t j ) → ϕ(t0 ) weakly in E, j → +∞. (6.69)

Secondly let us prove that

lim ϕk j (t j ) E = ϕ(t0 ) E. (6.70)

j→+∞

Since J is sequentially weakly continuous on V , V is sequentially weakly lower

semicontinuous on E. Hence, we obtain

t0 t j
V (ϕ(t0 )) ≤ lim V (ϕk j (t j )), Au  (s), u  (s)V ds ≤ lim Au k j (s), u k j (s)V ds,
j→+∞ j→+∞
τ τ
(6.71)
and hence
t0 t0
V (ϕ(t0 )) + Au  (s), u  (s)V ds ≤ lim V (ϕk j (t j )) + Au  (s), u  (s)V ds ≤
τ j→+∞ τ
t0
≤ lim V (ϕk j (t j )) + Au  (s), u  (s)V ds
j→+∞ τ
tj
≤ lim V (ϕk j (t j )) + lim Au k j (s), u k j (s)V ds ≤
j→+∞ j→+∞ τ
 
tj
≤ lim V (ϕk j (t j )) + Au k j (s), u k j (s)V ds .
j→+∞ τ
(6.72)
We remark that the last inequality in (6.71) follows from weak convergence of
T
u n k (·) to u  (·) in L 2 (τ, T ; V ) and because of the functional v → Av(s), v(s)V ds
τ
is sequentially weakly lower semi-continuous on L 2 (τ, T ; V ).
Since the energy equation and (6.65) both sides of (6.72) are equal to V (ϕ(τ ))
(see Lemma 6.15), it follows that V (ϕk j (t j )) → V (ϕ(t0 )), j → +∞ and then (6.70).
Convergence (6.68) directly follows from (6.69), (6.70). Finally we remark that (6.68)
contradicts (6.67). Therefore, (6.66) is true.
The theorem is proved.

6.3 Examples of Applications

In this section we consider the following examples of applications: nonlinear

parabolic equations of divergent form, nonlinear problems on manifolds with
and without boundary: a climate energy balance model; a model of conduction
6.3 Examples of Applications 151

of electrical impulses in nerve axons; and viscoelastic problems with nonlinear

“reaction-displacement” law.

6.3.1 Nonlinear Parabolic Equations of Divergent Form

Consider an example of the class of nonlinear boundary value problems for which
we can investigate the dynamics of solutions as t → +∞. Note that in discussion
we do not claim generality.
Let n ≥ 2, m ≥ 1, p ≥ 2, 1 < q ≤ 2, 1p + q1 = 1, Ω ⊂ Rn be a bounded domain
with sufficiently smooth boundary Γ = ∂Ω. We denote a number of differentiations
by x of order ≤ m − 1 (correspondingly of order = m) by N1 (correspondingly by
N2 ). Let Aα (x, η; ξ ) be a family of real functions (|α| ≤ m), defined in Ω × R N1 ×
R N2 and satisfying the next properties:
(C1 ) for a.e. (x, t) ∈ Ω × (0, ∞) the function (η, ξ ) → Aα (x, t, η, ξ ) is contin-
uous one in R N1 × R N2 ;
(C2 ) for each (η, ξ ) ∈ R N1 × R N2 the function x → Aα (x, t, η, ξ ) is measurable
on Ω × (0, ∞);
(C3 ) there exist c1 ≥ 0 and k1 ∈ L q (Ω) such that

|Aα (x, t, η, ξ )| ≤ c1 [|η| p−1 + |ξ | p−1 + k1 (x)]

for a.e. (x, t) ∈ Ω × (0, ∞) and for each (η, ξ ) ∈ R N1 × R N2 ;

(C4 ) there exist c2 > 0 and k2 ∈ L 1 (Ω) such that

Aα (x, t, η, ξ )ξα ≥ c2 |ξ | p − k2 (x)
|α|=m

for a.e. (x, t) ∈ Ω × (0, ∞) and for each (η, ξ ) ∈ R N1 × R N2 ;

(C5 ) there exists an increasing function ϕ : R+ → R such that the following
inequality holds:

(Aα (x, t, η, ξ ) − Aα (x, t, η, ξ ∗ ))(ξα − ξα∗ ) ≥ (ϕ(|ξα |) − ϕ(|ξα∗ |))(|ξα | − |ξα∗ |)
|α|=m

for a.e. (x, t) ∈ Ω × (0, ∞) and each η ∈ R N1 and ξ, ξ ∗ ∈ R N2 , ξ = ξ ∗ .

Consider the following notations:

D k u = {D β u, |β| = k}, δu = {u, Du, . . . , D m−1 u}.

Let us examine the dynamics of all weak (generalized) solutions defined on

[0, +∞) for the following problem:
152 6 Strongest Convergence Results for Weak Solutions …

∂ y(x, t)   
+ (−1)|α| D α Aα (x, δy(x, t), D m y(x, t)) = 0onΩ × (0, +∞),
∂t |α|≤m
(6.73)
D α y(x, t) = 0on Γ × (0, +∞), |α| ≤ m − 1. (6.74)

as t → +∞.
m, p
Consider such denotations: H = L 2 (Ω), V = W0 (Ω) is a real Sobolev space,
 
a(t, u, ω) = Aα (x, t, δu(x), D m u(x))D α ω(x)d x, u, ω ∈ V.
|α|≤m Ω

Note that the operator A(t) : V → V ∗ , t ≥ 0, defined by the formula A(t, u), ωV =
a(t, u, ω), t ≥ 0, u, ω ∈ V , satisfies Assumptions (C1)–(C5). Therefore, we pass
from Problem (6.73) and (6.74) to the respective problem in the “generalized” setting
(6.1). Here we note that
  
A(t, u) = (−1)|α| D α Aα (x, t, δu, D m u) , u ∈ C0∞ (Ω), t ≥ 0.
|α|≤m

Therefore, all statements from Sect. 6.1 hold for all weak (generalized) solutions of
Problem (6.73) and (6.74).

6.3.2 Nonlinear Non-autonomous Problems on Manifolds

with and Without Boundary: A Climate Energy
Balance Model

Let (M , g) be a C ∞ compact connected oriented two-dimensional Riemannian man-

ifold without boundary (as, e.g. M = S 2 the unit sphere of R3 ). Consider the problem
(see Sect. 2.4.3 for autonomous setting):
∂u
∂t
− Δu + Re (x, t, u) ∈ Q S(x, t)β(u), (x, t) ∈ M × R, (6.75)

where Δu = divM (∇M u) ; ∇M is understood in the sense of the Riemannian

metric g. Note that (6.75) is the so-called climate energy balance model (see
Sect. 2.4.3). The unknown u(x, t) represents the average temperature of the Earth’s
surface. The energy balance is expressed as

heat variation = Ra − Re + D.

Here Ra = Q S(x, t)β(u). It represents the solar energy absorbed by the Earth, Q > 0
is a solar constant, S(x, t) is an insolation function, given the distribution of solar
radiation falling on upper atmosphere, β represents the ratio between absorbed and
6.3 Examples of Applications 153

incident solar energy at the point x of the Earth’s surface (so-called co-albedo func-
tion). The term Re represents the energy emitted by the Earth into space, as usual,
it is assumed to be an increasing function on u. The term D is heat diffusion, we
assume (for simplicity) that it is constant. We consider Re = Bu as in Budyko; see
[31] and references therein.
Let S : M → R be a function such that S ∈ L ∞ (M ) and there exist S0 , S1 > 0
such that
0 < S0 ≤ S(x, t) ≤ S1 .

Suppose also that β is a bounded maximal monotone graph of R2 , that is there exist
m, M ∈ R such that for all s ∈ R and z ∈ β(s)

m ≤ z ≤ M.

Let us consider real Hilbert spaces

H := L 2 (M ), V := {u ∈ L 2 (M ) : ∇M u ∈ L 2 (T M )}

with respective standard norms · H , · V , and inner products ( · , · ) H , ( · , · )V ,

where T M represents the tangent bundle and the functional spaces L 2 (M ) and
L 2 (T M ) are defined in a standard way. Therefore, all statements from Sect. 6.1 hold
for weak solutions of Problem (6.75).

6.3.3 A Model of Conduction of Electrical Impulses

in Nerve Axons

Consider the problem (see Sect. 2.4.2 and Fig. 6.2):

 ∂u
− ∂∂ xu2 + u ∈ λH (u − a), (x, t) ∈ (0, π ) × R,
2

∂t (6.76)
u(0, t) = u(π, t) = 0, t ∈ R,
 
where a ∈ 0, 21 (see Sect. 2.4.2) Therefore, all statements from Sect. 6.1 hold for
weak solutions of Problem (6.76).

Fig. 6.2 Nerve impulse propagation

154 6 Strongest Convergence Results for Weak Solutions …

6.3.4 Viscoelastic Problems with Nonlinear

“Reaction-Displacement” Law

Let a viscoelastic body occupy a bounded domain Ω ⊂ Rd , d = 2, 3 in applications,

and it is acted upon by volume forces and surface tractions.4 The boundary Γ of Ω is
supposed to be Lipschitz continuous and it is partitioned into two disjoint measurable
parts Γ D and Γ N such that meas(Γ D ) > 0. We consider the process of evolution of
the mechanical state on the interval (0, +∞). The body is clamped on Γ D and
thus the displacement vanishes there. The forces field of density f 0 acts in Ω, the
surface tractions of density g0 are applied on Γ N . We denote by u = (u 1 , . . . , u d )
the displacement vector, by σ = (σi j ) the stress tensor and by ε(u) = (εi j (u)) the
linearized (small) strain tensor (εi j (u) = 21 (∂ j u i + ∂i u j )), where i, j = 1, . . . , d.
The mechanical problem consists in finding the displacement field u : Ω ×
(0, +∞) → Rd such that

u  (t) − divσ (t) = f 0 in Ω × (0, +∞), (6.77)

σ (t) = C ε(u  (t)) + E ε(u(t)) in Ω × (0, +∞), (6.78)

u(t) = 0 on Γ D × (0, +∞), (6.79)

σ n(t) = g0 on Γ N × (0, +∞), (6.80)

u(0) = u 0 , u  (0) = u 1 in Ω, (6.81)

where C and E are given linear constitutive functions, n being the outward unit
normal vector to Γ .
In the above model dynamic equation (6.77) is considered with the viscoelastic
constitutive relationship of the Kelvin–Voigt type (6.78) while (6.79) and (6.80) rep-
resent the displacement and traction boundary conditions, respectively. The functions
u 0 and u 1 are the initial displacement and the initial velocity, respectively. In order
to formulate the skin effects, we suppose that the body forces of density f 0 consists
of two parts: f 1 which is prescribed external loading and f 2 which is the reaction
of constrains introducing the skin effects, i.e., f 0 = f 1 + f 2 . Here f 2 is a possibly
multi-valued function of the displacement u. We consider the reaction-displacement
law of the form

− f 2 (x, t) ∈ ∂ j (x, u(x, t)) in Ω × (0, +∞), (6.82)

where j : Ω × Rd → R is locally Lipschitz function in u and ∂ j represents the

Clarke subdifferential with respect to u. Let Yd be the space of second-order sym-
metric tensors on Rd .

4 This section is based on results of [23] and references therein.

6.3 Examples of Applications 155

We consider the following problem:

examine the long-time (as t → +∞) behavior of all (weak, generalized) solutions
for (6.77)–(6.81) and (6.82).
In [23] for finite time interval it was presented the hemivariational formulation of
problems similar to (6.77)–(6.82) and an existence theorem for evolution inclusions
with pseudomonotone operators. We give now variational formulation of the above
problem. To this aim let H = L 2 (Ω, Rd ), H1 = H 1 (Ω, Rd ), H = L 2 (Ω, Yd ) and
V be the closed subspace of H1 defined by

V = {v ∈ H1 : v = 0 on Γ D }.

On V we consider the inner product and the corresponding norm given by

(u, v)V = ε(u), ε(v)H , v V = ε(v) H for u, v ∈ V.

From the Korn inequality v H1 ≤ C1 ε(v) H for v ∈ V with C1 > 0, it follows

that · H1 and · V are the equivalent norms on V. Identifying H with its dual,
we have an evolution triple V ⊂ H ⊂ V ∗ (see e.g. [12]) with dense and compact
embeddings. We denote by ·, ·V the duality of V and its dual V ∗ , by · V ∗ the
norm in V ∗ . We have u, vV = (u, v) H for all u ∈ H and v ∈ V.
We admit the following hypotheses:
H(C ). The linear symmetric viscosity operator C : Ω × Yd → Yd satisfies the
Carathéodory condition (i.e., C (·, ε) is measurable on Ω for all ε ∈ Yd and C (x, ·)
is continuous on Yd for a.e. x ∈ Ω) and

C (x, ε) : ε ≥ C2 ε 2
Yd for all ε ∈ Yd and a.e. x ∈ Ω with C2 > 0. (6.83)

H(E ). The elasticity operator E : Ω × Yd → Yd is of the form E (x, ε) = E(x)ε

(Hooke’s law) with a symmetric elasticity tensor E ∈ L ∞ (Ω), i.e., E = (gi jkl ),
i, j, k, l = 1, . . . , d with gi jkl = g jikl = glki j ∈ L ∞ (Ω). Moreover,

E (x, ε) : ε ≥ C3 ε 2
Yd for all ε ∈ Yd and a.e. x ∈ Ω with C3 > 0.

H(j). j : Ω × Rd → R is a function such that

(i) j (·, ξ ) is measurable for all ξ ∈ Rd and j (·, 0) ∈ L 1 (Ω);
(ii) j (x, ·) is locally Lipschitz and it admits the representation via the difference
of regular functions [8] for all x ∈ Ω;
(iii) η ≤ C4 (1 + ξ ) for all η ∈ ∂ j (x, ξ ), x ∈ Ω with C4 > 0;
(iv) j 0 (x, ξ ; −ξ ) ≤ C5 (1 + ξ ) for all ξ ∈ Rd , x ∈ Ω, with C5 ≥ 0, where
j 0 (x, ξ ; η) is the directional derivative of j (x, ·) at the point ξ ∈ Rd in the direction
η ∈ Rd .
H(f). f 1 ∈ V ∗ , g0 ∈ L 2 (Γ N ; Rd ), u 0 ∈ V and u 1 ∈ H.
Next we need the spaces V = L 2 (τ, T ; V ), Hˆ = L 2 (τ, T ; H ) and W = {w ∈
V : w  ∈ V ∗ }, where the time derivative involved in the definition of W is
156 6 Strongest Convergence Results for Weak Solutions …

understood in the sense of vector-valued distributions, −∞ < τ < T < +∞.

Endowed with the norm v W = v V + v  V ∗ , the space W becomes a separa-
ble reflexive Banach space. We also have W ⊂ V ⊂ Hˆ ⊂ V ∗ . The duality for the
T
pair (V , V ∗ ) is denoted by z, wV = z(s), w(s)V ds. It is well known (cf. [12])
τ
that the embedding W ⊂ C([τ, T ]; H ) and {w ∈ V : w  ∈ W } ⊂ C([τ, T ]; V ) are
continuous. Next we define g ∈ V ∗ by

g, vV =  f 1 , vV + g0 , v L 2 (Γ N ;Rd ) for v ∈ V. (6.84)

According to condition (6.82), we obtain the following variational formulation of

our problem:
⎧ 
⎪
⎪ u (t), vV + (σ (t), ε(v))H + j 0 (x, u(t); v)d x ≥
⎪
⎨ Ω
≥ g, vV for all v ∈ V and a.e. t ∈ (0, +∞), (6.85)
⎪
⎪ σ (t) = C (ε(u  (t))) + E (ε(u(t))) for a.e. t ∈ (0, +∞),
⎪
⎩
u(0) = u 0 , u  (0) = u 1 .

We define the operators A : V → V ∗ and B : V → V ∗ by

A(u), vV = (C (x, ε(u)), ε(v))H for u, v ∈ V, (6.86)

Bu, vV = (E (x, ε(u)), ε(v))H for u, v ∈ V. (6.87)

Note that bilinear forms (6.86) and (6.87) are symmetric, continuous, and coercive.
Let us introduce the functional J : L 2 (Ω; Rd ) → R defined by

J (v) = j (x, v(x))d x for v ∈ L 2 (Ω; Rd ). (6.88)
Ω

From [8, Chap. II] under Assumptions H(j), the functional J defined by (6.88)
satisfies
H(J). J : L 2 (Ω; Rd ) → R is a functional such that:
(i) J (·) is well defined, locally Lipschitz (in fact, Lipschitz on bounded subsets of
L 2 (Ω; Rd )) and it admits the representation via the difference of regular functions
J1 and J2 on H ;
(ii) ζ ∈ ∂ J1 (v) − ∂ J2 (v) implies ζ L 2 (Ω;Rd ) ≤ C6 (1 + v L 2 (Ω;Rd ) ) for v ∈ L 2
(Ω; Rd ) with C6 > 0;
(iii) J 0 (v; −v) ≤ C7 (1 + v L 2 (Ω;Rd ) ) for v ∈ L 2 (Ω; Rd ) with C7 ≥ 0, where
J (u; v) denotes the directional derivative of J (·) at a point u ∈ L 2 (Ω; Rd ) in the
0

direction v ∈ L 2 (Ω; Rd ).
We can now formulate the second-order evolution inclusions associated with the
variational form of our problem
6.3 Examples of Applications 157
⎧
⎪
⎪ Find u ∈ C([0, +∞); V ) with u  ∈ C([0, +∞); H ) ∩ L loc 2 (0, +∞; V )
⎨
and u  ∈ L loc
2 (0, +∞; V ∗
) such that
⎪
⎪ u  (t) + Au  (t) + Bu(t) + ∂ J1 (u(t)) − ∂ J2 (u(t))  g a.e. t ∈ (0, +∞),
⎩
u(0) = u 0 , u  (0) = u 1 .
(6.89)

Theorem 6.6 yields that, if τ < T, {ϕn (·)}n≥1 ⊂ WτT is an arbitrary sequence of
weak solutions of (6.89) on [τ, T ] such that ϕn (τ ) → ϕτ strongly in E, n → +∞,
then there exists ϕ ∈ Dτ,T (ϕτ ) such that up to a subsequence ϕn (·) → ϕ(·) in
C([τ, T ]; E), n → +∞ (see Sect. 6.2 for details).

References

1. Aubin, J.-P., Cellina, A.: Differential Inclusions. Set-Valued Maps and Viability Theory.
Grundlehren der Mathematischen Wissenschaften, vol. 264. Springer, Berlin (1984)
2. Aubin, J.-P., Frankowska, H.: Set-Valued Analysis. Systems Control Foundation Application.
Birkhauser, Boston (1990)
3. Babin, A.V., Vishik, M.I.: Maximal attractors of semigroups corresponding to evolution dif-
ferential equations. Math. USSR-Sb. 54(2), 387–408 (1986)
4. Babin, A.V., Vishik, M.I.: Attractors of Evolution Equations [in Russian]. Nauka, Moscow
(1989)
5. Chepyzhov, V.V., Vishik, M.I.: Trajectory attractors for evolution equations. C. R. Acad. Sci.
Paris Ser. I Math. 321(10), 1309–1314 (1995)
6. Chepyzhov, V.V., Vishik, M.I.: Evolution equations and their trajectory attractors. J. Math.
Pures Appl. 76(10), 913–964 (1997)
7. Chepyzhov, V.V., Vishik, M.I.: Trajectory attractor for reaction-diffusion system with diffusion
coefficient vanishing in time. Discrete Contin. Dyn. Syst. 27(4), 1493–1509 (2010)
8. Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley, New York (1983)
9. Dubinskii, YuA: Higher-order nonlinear parabolic equations. J. Soviet Math. 56(4), 2557–2607
(1991)
10. Denkowski, Z., Migórski, S., Papageorgiou, N.S.: An Introduction to Nonlinear Analysis:
Applications. Kluwer Academic/Plenum Publishers, Boston (2003)
11. Eden, A., Foias, C., Nicolaenko, B., Temam, R.: Exponential Attractors for Dissipative Evo-
lution Equations. RAM Research in Applied Mathematics, vol. 37. Wiley, Chichester (1994)
12. Gajewski, H., Groger, K., Zacharias, K.: Nichtlineare Operatorgleichungen und
Operatordifferential-gleichungen. Akademie-Verlag, Berlin (1974)
13. Gorban, N.V., Kapustyan, O.V., Kasyanov, P.O.: Uniform trajectory attractor for non-
autonomous reaction-diffusion equations with Caratheodory’s nonlinearity. Nonlinear Anal.
Theory Methods Appl. 98, 13–26 (2014). doi:10.1016/j.na.2013.12.004
14. Hale, J.K.: Asymptotic Behavior of Dissipative Systems. Mathematical Surveys and Mono-
graphs, vol. 25. American Mathematical Society, Providence (1988)
15. Kasyanov, P.O.: Multi-valued dynamics of solutions of autonomous operator differential equa-
tions with pseudomonotone nonlinearity. Math. Notes 92(2), 57–70 (2012)
16. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Long-time behavior of solutions
for autonomous evolution hemivariational inequality with multidimensional “reaction-
displacement” law. Abstr. Appl. Anal. (2012). doi:10.1155/2012/450984
17. Kuttler, K.: Non-degenerate implicit evolution inclusions. Electron. J. Differ. Equ. 34, 1–20
(2000)
158 6 Strongest Convergence Results for Weak Solutions …

18. Ladyzhenskaya, O.A.: The Dynamical System That is Generated by the Navier–Stokes Equa-
tions [in Russian]. Zap. Naucn. Sem. Leningrad. Otdel. Mat. Inst. Steklov, vol. 27(6), pp.
91–115. Nauka, Leningrad (1972)
19. Ladyzhenskaya, O.A.: The Infinite-dimensionality of Bounded Invariant Sets for the Navier–
Stokes System and Other Dissipative Systems [in Russian]. Zap. Naucn. Sem. Leningrad.
Otdel. Mat. Inst. Steklov, vol. 115(6), pp. 137–155. Nauka, Leningrad (1982)
20. Ladyzhenskaya, O.: Attractors for Semigroups and Evolution Equations. Lezioni Lincee. Cam-
bridge University Press, Cambridge (1991)
21. Lions, J.-L.: Quelques Methodes de Resolution des Problemes aux Limites Nonlineaires.
Dunod, Paris (1969)
22. Migorski, S.: Boundary hemivariational inequalities of hyperbolic type and applications. J.
Global Optim. 31(3), 505–533 (2005)
23. Migórski, S.: Dynamic hemivariational inequalities in contact mechanics. Nonlinear Anal. 63,
77–86 (2005). doi:10.1016/j.na.2005.01.012
24. Melnik, V.S., Valero, J.: On attractors of multi-valued semi-flows and generalized differential
equations. Set-Valued Anal. 6(1), 83–111 (1998)
25. Sell, G.R.: Global attractors for the three-dimensional Navier-Stokes equations. J. Dyn. Differ.
Equ. 8(1), 1–33 (1996)
26. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Applied
Mathematical Sciences, vol. 68. Springer, New York (1988)
27. Vishik, M.I., Chepyzhov, V.V.: Trajectory and global attractors of the three-dimensional Navier-
Stokes system. Mat. Zametki. 71(2), 194–213 (2002)
28. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong trajectory attractor for a dissipative reaction-
diffusion system. Dokl. Ross. Akad. Nauk. 435(2), 155–159 (2010)
29. Zadoyanchuk, N.V., Kas’yanov, P.O.: Faedo-Galerkin method for second-order evolution inclu-
sions with Wλ -pseudomonotone mappings. Ukr. Math. J. 61(2), 236–258 (2009)
30. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation Inequali-
ties for Earth Data Processing II. Springer, Berlin (2011)
31. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
Inclusions and Variation Inequalities for Earth Data Processing III. Springer, Berlin (2012)
 Part III
Uniform Global Behavior of Solutions:
Uniform Attractors, Flattening and
Entropy
Chapter 7
Uniform Global Attractors for
Non-autonomous Dissipative Dynamical
Systems

Abstract In this chapter we consider sufficient conditions for the existence of

uniform compact global attractor for non-autonomous dynamical systems in spe-
cial classes of infinite-dimensional phase spaces. The obtained generalizations allow
us to avoid the restrictive compactness assumptions on the space of shifts of non-
autonomous terms in particular evolution problems. The results are applied to several
evolution inclusions.

7.1 General Methodology

The standard scheme of investigation of uniform the long-time behavior for all solu-
tions of non-autonomous problems covers non-autonomous problems of the form

∂t u(t) = Aσ (t) (u(t)), (7.1)

where σ (s), s ≥ 0, is a functional parameter called the time symbol of Eq. (7.1)
(t is replaced by s). In applications to mathematical physics equations, a function
σ (s) consists of all time-dependent terms of the equation under consideration: exter-
nal forces, parameters of mediums, interaction functions, control functions, etc.;
Chepyzhov and Vishik [4, 5, 8]; Sell [36]; Zgurovsky et al. [48] and references
therein; see also Hale [16]; Ladyzhenskaya [30]; Mel’nik and Valero [32]; Iovane,
Kapustyan and Valero [17]. In the mentioned above papers and books it is assumed
that the symbol σ of Eq. (7.1) belongs to a Hausdorff topological space Ξ+ of func-
tions defined on R+ with values in some complete metric space. Usually, in applica-
tions, the topology in the space Ξ+ is a local convergence topology on any segment
[t1 , t2 ] ⊂ R+ . Further, they consider the family of Eq. (7.1) with various symbols
σ (s) belonging to a set Σ ⊆ Ξ+ . The set Σ is called the symbol space of the fam-
ily of Eq. (7.1). It is assumed that the set Σ, together with any symbol σ (s) ∈ Σ,
contains all positive translations of σ (s): σ (t + s) = T (t)σ (s) ∈ Σ for any t, s ≥ 0.
The symbol space Σ is invariant with respect to the translation semigroup {T (t)}t≥0 :
T (t)Σ ⊆ Σ for any t ≥ 0. To prove the existence of uniform trajectory attractors
they suppose that the symbol space Σ with the topology induced from Ξ+ is a
© Springer International Publishing AG 2018 161
M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_7
162 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems

compact metric space. Mostly in applications, as a symbol space Σ it is naturally to

consider the hull of translation-compact function σ0 (s) in an appropriate Hausdorff
topological space Ξ+ . The direct realization of this approach to differential-operator
inclusions, PDEs with Caratheodory’s nonlinearities, optimization problems, etc., is
problematic without any additional assumptions for parameters of Problem (7.1) and
requires the translation-compactness of the symbol σ (s) in some compact Hausdorff
topological space of measurable multi-valued mappings acts from R+ to some metric
∗
space of operators from (V → 2V ), where V is a Banach space and V ∗ is its dual
space, satisfying (possibly) only growth and sign assumptions. To avoid this techni-
cal difficulties we present an alternative approach for the existence and construction
of the uniform global attractor for classes of non-autonomous dynamical systems in
special classes of infinite-dimensional phase spaces; see also [1, 6, 12–15, 18, 21,
22, 24–29, 37–44, 46].

7.2 Main Constructions and Results

Let p ≥ 2 and q > 1 be such that 1p + q1 = 1, (V ; H ; V ∗ ) to be evolution triple

such that V ⊂ H with compact embedding. For each t1 , t2 ∈ R, 0 ≤ t1 < t2 < +∞,
consider the space

Wt1 ,t2 := {y(·) ∈ L p (t1 , t2 ; V ) : y (·) ∈ L q (t1 , t2 ; V ∗ )},

where y (·) is a derivative of an element y(·) ∈ L p (t1 , t2 ; V ) in the sense of distrib-

utions D ∗ ([t1 , t2 ]; V ∗ ). The space Wt1 ,t2 endowed with the norm

y Wt1 ,t2 := y L p (t1 ,t2 ;V ) + y L q (t1 ,t2 ;V ∗ ) , y ∈ Wt1 ,t2 ,

is a reflexive Banach space. Note that Wt1 ,t2 ⊂ C([t1 , t2 ]; H ) with continuous and
dense embedding; Gajewsky et al. [11, Chap. IV]. For each τ ≥ 0, consider the
Fréchet space

W loc ([τ, +∞)) := {y : [τ, +∞) → H : Πt1 ,t2 y ∈ Wt1 ,t2 for each [t1 , t2 ] ⊂ [τ, +∞)},

where Πt1 ,t2 is the restriction operator to the finite time interval [t1 , t2 ]. We recall that
the sequence { f n }n≥1 converges in W loc ([τ, +∞)) (in C loc ([τ, +∞); H ) respec-
tively) to f ∈ W loc ([τ, +∞)) (to f ∈ C loc ([τ, +∞); H ) respectively) as n → +∞
if and only if the sequence {Πt1 ,t2 f n }n≥1 converges in Wt1 ,t2 (in C([t1 , t2 ]; H ) respec-
tively) to Πt1 ,t2 f as n → +∞ for each finite time interval [t1 , t2 ] ⊂ [τ, +∞). Further
we denote that

T (h)y(·) = Π0,+∞ y( · + h), y ∈ W loc (R+ ), h ≥ 0,

where R+ = [0, +∞) and Π0,+∞ is the restriction operator to the time interval
[0, +∞).
7.2 Main Constructions and Results 163

Throughout the chapter we consider the family of solution sets {Kτ+ }τ ≥0 such that
Kτ+ ⊂ W loc ([τ, +∞)) for each τ ≥ 0 and Kτ0+ = ∅ for some τ0 ≥ 0. In the most
of applications as Kτ+ can be considered the family of globally defined on [τ, +∞)
weak solutions for particular non-autonomous evolution problem (see Sect. 7.4).
To state the main assumptions on the family of solution sets {Kτ+ }τ ≥0 it is nec-
essary to formulate two auxiliary definitions.
γ (R+ ), γ > 1, is called translation bounded function in
A function ϕ ∈ L loc
L loc
γ (R+ ) if
 t+1
sup |ϕ(s)|γ ds < +∞;
t≥0 t

Chepyzhov and Vishik [7, p. 105]. A function ϕ ∈ L loc

1 (R+ ) is called a translation
1 (R+ ) if
uniform integrable (t.u.i.) function in L loc
 t+1
lim sup |ϕ(s)|I{|ϕ(s)| ≥ K }ds = 0;
K →+∞ t≥0 t

Gorban et al. [14]. Note that Dunford–Pettis compactness criterion provides that
ϕ ∈ L loc
1 (R+ ) is a t.u.i. function in L 1 (R+ ) if and only if for every sequence of
loc

elements {τn }n≥1 ⊂ R+ , the sequence {ϕ( · + τn )}n≥1 contains a subsequence con-
verging weakly in L loc 1 (R+ ). Note that for each γ > 1, every translation bounded in
L loc
γ (R + ) function is 1 (R+ ); Gorban et al. [14].
t.u.i. in L loc
Main assumptions. Let the following two assumptions hold:

1 (R+ ) function c1 : R+ → R+ and a constant α1 > 0

(A1) there exist a t.u.i. in L loc
such that for each τ ≥ 0, y ∈ Kτ+ , and t2 ≥ t1 ≥ τ , the following inequality
holds:
 t2  t2
p
y(t2 ) H − y(t1 ) H + α1
2 2
y(t) V dt ≤ c1 (t)dt; (7.2)
t1 t1

1 (R+ ) function c2 : R+ → R+ and a constant α2 > 0

(A2) there exist a t.u.i. in L loc
such that for each τ ≥ 0, y ∈ Kτ+ , and t2 ≥ t1 ≥ τ , the following inequality
holds:  t2  t2  t2
q p
y (t) V ∗ dt ≤ α2 y(t) V dt + c2 (t)dt. (7.3)
t1 t1 t1

To characterize the uniform long-time behavior of solutions for non-autonomous

dissipative dynamical system consider the united trajectory space K∪+ for the family
of solutions {Kτ+ }τ ≥0 shifted to zero:
 
K∪+ := T (h)y( · + τ ) : y( · ) ∈ Kτ+ , h ≥ 0 ⊂ W loc (R+ ), (7.4)
τ ≥0
164 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems

and the extended united trajectory space for the family {Kτ+ }τ ≥0 :
+
 
K := clC loc (R+ ;H ) K∪+ , (7.5)

where clC loc (R+ ;H ) [ · ] is the closure in C loc (R+ ; H ). Since T (h)K∪+ ⊆ K∪+ for each
h ≥ 0, then
T (h)K + ⊆ K + for each h ≥ 0, (7.6)

due to

ρC loc (R+ ;H ) (T (h)u, T (h)v) ≤ ρC loc (R+ ;H ) (u, v) for each u, v ∈ C loc (R+ ; H ),

where ρC loc (R+ ;H ) is the standard metric on Fréchet space C loc (R+ ; H ). Therefore
the set
X := {y(0) : y ∈ K + } (7.7)

is closed in H (it follows from Theorem 7.2). We endow this set X with metric

ρX (x1 , x2 ) = x1 − x2 H, x1 , x2 ∈ X.

Then we obtain that (X, ρ) is a Polish space (complete separable metric space).
Let us define the multi-valued semiflow (m-semiflow) G : R+ × X → 2X :
+
G(t, y0 ) := {y(t) : y(·) ∈ K and y(0) = y0 }, t ≥ 0, y0 ∈ X. (7.8)

According to (7.6), (7.7), and (7.8) for each t ≥ 0 and y0 ∈ X the set G(t, y0 ) is
nonempty. Moreover, the following two conditions hold:
(i) G (0, ·) = I is the identity map;
(ii) G (t1 + t2 , y0 ) ⊆ G (t1 , G (t2 , y0 )) , ∀t1 , t2 ∈ R+ , ∀y0 ∈ X,
where G (t, D) = ∪ G (t, y) , D ⊆ X.
y∈D
We denote by distX (C, D) = supc∈C inf d∈D ρ(c, d) the Hausdorff semidistance
between nonempty subsets C and D of the Polish space X. Recall that the set  ⊂ X
is a global attractor of the m-semiflow G if it satisfies the following conditions:
(i)  attracts each bounded subset B ⊂ X, i.e.

distX (G(t, B), ) → 0, t → +∞; (7.9)

(ii)  is negatively semi-invariant set, i.e.  ⊆ G (t, ) for each t ≥ 0;

(iii)  is the minimal set among all nonempty closed subsets C ⊆ X that satisfy
(7.9).
In this chapter we examine the uniform long-time behavior of solution sets
{Kτ+ }τ ≥0 in the strong topology of the natural phase space H (as time t → +∞) in
7.2 Main Constructions and Results 165

the sense of the existence of a compact global attractor for m-semiflow G generated
by the family of solution sets {Kτ+ }τ ≥0 and their shifts. The following theorem is
the main result of the chapter.
Theorem 7.1 Let assumptions (A1)–(A2) hold. Then the m-semiflow G, defined in
(7.8), has a compact global attractor  in the phase space X.

7.3 Proof of Theorem 7.1

Before the proof of Theorem 7.1 we provide the following statement characteriz-
ing the compactness properties of the family K + in the topology induced from
C loc (R+ ; H ).
Theorem 7.2 Let assumptions (A1)–(A2) hold. Then the following two statements
hold:
(a) for each y ∈ K + , the following estimate holds
2 −c3 t
y(t) 2
H ≤ y(0) He + c4 , t ≥ 0, (7.10)

where the positive constants c3 and c4 do not depend on y ∈ K + and t ≥ 0;

(b) for any bounded in L ∞ (R+ ; H ) sequence {yn }n≥1 ⊂ K + , there exist an increas-
ing sequence {n k }k≥1 ⊆ N and an element y ∈ K + such that

Πτ,T yn k − Πτ,T y C([τ,T ];H ) → 0, k → +∞, (7.11)

for each finite time interval [τ, T ] ⊂ (0, +∞). If, additionally, there exists y0 ∈
H such that yn k (0) → y0 in H , then y(0) = y0 .
Proof Let us prove statement (a). If statement (a) holds for each y ∈ K∪+ , then
inequality (7.10) holds for each y ∈ K + , due to (7.5). The rest of the proof of
statement (a) establishes inequality (7.10) for each y ∈ K∪+ .
For an arbitrary y ∈ K∪+ , there exist τ, h ≥ 0 and z( · ) ∈ Kτ+ such that y( · ) =
T (τ + h)z( · ). Assumption (A1) implies the following inequality:
 t2  t2
p
y(t2 ) 2
H − y(t1 ) 2
H + α1 y(t) V dt ≤ c1 (t + τ + h)dt, (7.12)
t1 t1

for each t2 ≥ t1 ≥ 0, where c1 (·) is t.u.i. in L loc

1 (R+ ). Since the embedding V ⊂ H
is compact, then this embedding is continuous. So, there exists a constant β > 0
such that b H ≤ β b V for each b ∈ V. According to (7.12), since the inequality
a 2 ≤ 1 + a p holds for each a ≥ 0, then the following inequality holds:
 t2  t2
y(t2 ) 2
H − y(t1 ) 2
H + α3 y(t) 2
H dt ≤ [c1 (t + τ + h) + α3 ] dt, (7.13)
t1 t1
166 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems

α1
for each t2 ≥ t1 ≥ 0, where α3 = βp
. Let us set
 t  t
ρ(t) := y(t) 2
H + α3 y(s) 2
H ds − [c1 (s + τ + h) + α3 ] ds, t ≥ 0.
0 0

Inequality (7.13) and Ball [3, Lemma 7.1] yield that dtd ρ ≤ 0 in D ∗ ((0, +∞)), where
d
dt
is the derivative operation in the sense of D ∗ ((0, +∞)). Thus,

d
y(t) 2
H + α3 y(t) 2
H − [c1 (t + τ + h) + α3 ] ≤ 0 in D ∗ ((0, +∞)).
dt
Therefore,

d  2 α3 t

y(t) He − eα3 t [c1 (t + τ + h) + α3 ] ≤ 0 in D ∗ ((0, +∞)). (7.14)
dt
Ball [3, Lemma 7.1] and inequality (7.14) imply
 t2
2 −α3 (t2 −t1 )
y(t2 ) 2
H ≤ y(t1 ) He + e−α3 (t2 −t) [c1 (t + τ + h) + α3 ] dt, (7.15)
t1

for each t2 ≥ t1 ≥ 0. Therefore,

 t2
2 −α3 (t2 −t1 )
y(t2 ) 2
H ≤ y(t1 ) He + e−α3 (t2 −t) [c1 (t + τ + h) + α3 ] dt ≤
t1
 t2 +τ +h
2 −α3 (t2 −t1 )
y(t1 ) He +1+ e−α3 (t2 −t+τ +h) c1 (t)dt ≤
t1 +τ +h
K
y(t1 ) 2H e−α3 (t2 −t1 ) + 1 + +
α3
 t2 +τ +h
e−α3 (t2 −t+τ +h) |c1 (t)|I{|c1 (t)| ≥ K }dt,
t1 +τ +h

for each K > 0, t2 ≥ t1 ≥ 0. Since the function c1 : R+ → R+ is t.u.i. in L loc

1 (R+ )
(see assumption (A1)), then there exists K 0 > 0 such that
 t+1
sup |c1 (s)|I{|c1 (s)| ≥ K 0 }ds ≤ 1.
t≥0 t

Thus,
2 −α3 (t2 −t1 ) K0
y(t2 ) 2
≤ y(t1 ) He +1+ + eα3 + 1,
H
α3

that yields estimate (7.10) with c3 := α3 and c4 := 1 + Kα30 + eα3 + 1, where the
positive constants c3 and c4 do not depend on y ∈ K + and t ≥ 0.
7.3 Proof of Theorem 7.1 167

Let us prove statement (b). Let {yn }n≥1 ⊂ K + be an arbitrary sequence that is
bounded in L ∞ (R+ ; H ). Since K∪+ is the dense set in a Polish space K + endowed
with the topology induced from C loc (R+ ; H ), then for each n ≥ 1 there exists u n ∈
K∪+ such that
1
ρC loc (R+ ;H ) (yn , u n ) ≤ , for each n ≥ 1. (7.16)
n
Note that a priori estimate (7.10) provides that the sequence {u n }n≥1 is bounded
in L ∞ (R+ ; H ). Therefore, the rest of the proof establishes statement (b) for the
sequence {u n }n≥1 .
Let us fix n ≥ 1. Formula (7.4) provides the existence of τn , h n ≥ 0 and z n ( · ) ∈
Kτn+ such that u n ( · ) = z n ( · + τn + h n ). Then, assumptions (A1) and (A2) yield
 t2  t2
p
u n (t2 ) 2
H − u n (t1 ) 2
H + α1 u n (t) V dt ≤ c1 (t + τn + h n )dt,
t1 t1
 t2  t2  t2 (7.17)
q p
u n (t) V ∗ dt ≤ α2 u n (t) V dt + c2 (t + τn + h n )dt,
t1 t1 t1

for each t2 ≥ t1 ≥ 0 and n ≥ 1.

We remark that
 t2  t2
sup |c1 (t + τn + h n )|dt < ∞ and sup |c2 (t + τn + h n )|dt < ∞, (7.18)
n≥1 t1 n≥1 t1

for each t2 ≥ t1 ≥ 0, since the functions c1 , c2 : R+ → R+ are t.u.i. in L loc

1 (R+ ).
Formulae (7.17) and (7.18) imply that the sequence {u n }n≥1 is bounded in
W loc (R+ ). Thus, Banach–Alaoglu theorem and Zgurovsky et al. [47, Theorems 1.16
and 1.21] yield that there exist an increasing sequence {n k }k≥1 ⊆ N and elements
y ∈ W loc (R+ ) ⊂ C loc (R+ ; H ) and c̄1 ∈ L loc
1 (R+ ) such that

u nk → y p (R+ ; V ),
weakly in L loc
u nk → y weakly in L q (R+ ; V ∗ ),
loc

u nk → y weakly in C loc (R+ ; H ), (7.19)

u n k (t) → y(t) in H for a.e. t > 0,
c1 ( · + τn k + h n k ) → c̄1 1 (R+ ), k → ∞,
weakly in L loc

where the last convergence holds due to the fact that c1 ∈ L loc
1 (R+ ) is t.u.i. in
1 (R+ ). According to (7.19), we can pass to the limit in (7.2). So, we obtain
L loc
that y satisfies (7.2).
We consider the continuous and nonincreasing (by assumption (A1)) functions
on R+ :
168 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems
 t
Jk (t) = u n k (t) 2
H − c1 (s + τn k + h n k )ds,
0
 t (7.20)
J (t) = y(t) 2
H − c̄1 (s)ds, k ≥ 1;
0

cf. Kapustyan and Valero [19]. The last two statements in (7.19) imply

Jk (t) → J (t), as k → +∞, for a.e. t > 0. (7.21)

Similarly to Zgurovsky et al. [48, p. 57] (see the book and references therein) we
show that (7.11) holds. By contradiction suppose the existence of a positive constant
L > 0, a finite interval [τ, T ] ⊂ (0, +∞), and a subsequence {u k j } j≥1 ⊆ {u n k }k≥1
such that
max u k j (t) − y(t) H = u k j (t j ) − y(t j ) H ≥ L ,
t∈[τ,T ]

for each j ≥ 1. Suppose also that t j → t0 ∈ [τ, T ], as j → +∞. Continuity of

Πτ,T y : [τ, T ] → H implies

lim inf u k j (t j ) − y(t0 ) H ≥ L. (7.22)

j→+∞

On the other hand, we prove that

u k j (t j ) → y(t0 ) in H, j → +∞. (7.23)

For this purpose we firstly note that from (7.19) we have

u k j (t j ) → y(t0 ) weakly in H, j → +∞. (7.24)

Secondly we prove that

lim sup u k j (t j ) H ≤ y(t0 ) H. (7.25)

j→+∞

We consider the continuous nonincreasing functions J and Jk j , j ≥ 1, defined in

(7.20). Let us fix an arbitrary ε > 0. The continuity of J and (7.21) provide the exis-
tence of t¯ ∈ (τ, t0 ) such that lim j→∞ Jk j (t¯) = J (t¯) and |J (t¯) − J (t0 )| < ε. Then,

Jk j (t j ) − J (t0 ) ≤ |Jk j (t¯) − J (t¯)| + |J (t¯) − J (t0 )| ≤ |Jk j (t¯) − J (t¯)| + ε,

for rather large j ≥ 1. Thus, lim sup j→+∞ Jk j (t j ) ≤ J (t0 ) and inequality (7.25)
holds.
Thirdly note that the convergence (7.23) holds due to (7.24) and (7.25); cf. Gajew-
ski et al. [11, Chap. I]. Finally, we remark that statement (7.23) contradicts assumption
7.3 Proof of Theorem 7.1 169

(7.22). Therefore, according to (7.16), the first statement of the theorem holds for
each sequence {yn }n≥1 ⊂ K + .
To finish the proof of statement (b) we note that if, additionally, there exists
y0 ∈ H such that yn k (0) → y0 in H , then, according to the third convergence in
(7.19), y(0) = y0 .

Let us provide the proof of the main result.

Proof (Proof of Theorem 7.1) Theorem 7.2 implies the following properties for the
m-semiflow G, defined in (7.8):
(a) for each t ≥ 0 the mapping G(t, · ) : X → 2X \ {∅} has a closed graph;
(b) for each t ≥ 0 and y0 ∈ X the set G(t, y0 ) is compact in X;
(c) the set G(1, C̃), where C̃ := {z ∈ X : z 2H < c4 + 1}, is precompact and
attracts each bounded subset C ⊂ X.
Indeed, property (a) follows from Theorem 7.2 (see formulae (7.5) and (7.8)); prop-
erty (b) directly follows from (a) and Theorem 7.2(b); property (c) holds, since
G(1, C̃) is precompact in X (Theorem 7.2(b) and formula (7.8)) and the following
inequalities and equality hold:

distX (G(t, C), G(1, C̃)) ≤ distX (G(1, G(t − 1, C)), G(1, C̃)) ≤

distX (G(1, C̃), G(1, C̃)) = 0,

for sufficiently large t.

According to properties (a)–(c), Mel’nik and Valero [31, Theorems 1, 2, Remark 2,
Proposition 1] yields that the m-semiflow G has a compact global attractor  in the
phase space X.

7.4 Example of Applications

In the following three examples we examine the uniform global attractor for the
family of solution sets {Kτ+ } generated by particular evolution problems. In all the
cases we assume that

∀z ∈ H ∀τ ≥ 0 ∃y ∈ Kτ+ such that y(τ ) = z.

This assumption guarantees the equality X = H .

170 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems

7.4.1 Autonomous Evolution Problem

Let {Kτ+ } be a family of solutions for an autonomous problem on [τ, +∞), τ ≥ 0.

Then we have:
∀h ≥ 0 T (h)K0+ ⊂ K0+ ; (7.26)

∀τ ≥ 0 ∀y ∈ Kτ+ y(· + τ ) ∈ K0+ . (7.27)

So, K∪+ = K0+ . If additionally we have that

K0+ is closed in C loc (R+ ; H ), (7.28)

then
K +
= K0+ .

It implies that the m-semiflow G (defined by (7.8)) is a classical multi-valued semi-

group generated by an autonomous evolution problem.

7.4.2 Non-autonomous Evolution Problem

Let {Kτ+ } be a family of solutions for non-autonomous problem on [τ, +∞), τ ≥ 0,

and the following condition holds:

∀s ≥ τ ≥ 0 ∀y ∈ Kτ+ Πs,+∞ y(·) ∈ Ks + . (7.29)

Then, according to Kapustyan et al. [23], formula

U (t, τ, z) = {y(t) : y(·) ∈ Kτ+ , y(τ ) = z} (7.30)

defines a m-semiprocess, that is

∀t ≥ s ≥ τ U (t, τ, z) ⊂ U (t, s, U (s, τ, z)).

One of the most important objects for m-semiprocess (7.30) is uniform global attrac-
tor; Chepyzhov and Vishik [7], Kapustyan et al. [20], Zgurovsky et al. [48]. It is a
set  such that for every bounded subset C ⊂ H

sup dist H (U (t + τ, τ, C), ) → 0, t → ∞, (7.31)

τ ≥0

and  is minimal among all closed sets satisfying this property. Then under assump-
tions (A1), (A2) and from (7.29) it follows that the m-semiprocess (7.30) has the
7.4 Example of Applications 171

compact uniform global attractor  ⊆ , where  is the global attractor for the
m-semiflow (7.8).
Indeed,
∀t ≥ τ ≥ 0 ∀z ∈ H U (t + τ, τ, z) ⊂ G(t, z). (7.32)

So, if  is a compact global attractor for the m-semiflow G then, according

to Kapustyan et al. [20], there exists a compact uniform global attractor  for
m-semiprocess U and, moreover,  ⊂ .
In the following example we examine the existence of uniform global attractor for
non-autonomous differential-operator inclusion. The uniform trajectory attractors
for classes of non-autonomous inclusions and equations were proved to exist in
Zgurovsky and Kasyanov [45] (see also Gorban et al. [14]).

7.4.3 Non-autonomous Differential-Operator Inclusion

∗
For the multi-valued map A : R+ × V → 2V \{∅} we consider the problem of long-
time behavior of all globally defined weak solutions for non-autonomous evolution
inclusion
y (t) + A(t, y(t))  0̄, (7.33)

as t → +∞. Let ·, ·V : V ∗ × V → R be the pairing in V ∗ × V , that coincides on

H × V with the inner product (·, ·) in the Hilbert space H .
We note that Problem (7.33) arises in many important models for distributed
parameter control problems and that large class of identification problems enter this
formulation. Let us indicate a problem which is one of the motivations for the study
of the non-autonomous evolution inclusion (7.33) (see, for example, Migórski and
Ochal [34]; Zgurovsky et al. [48] and references therein). In a subset Ω of R3 , we
consider the nonstationary heat conduction equation

∂y
− y = f in Ω × (0, +∞)
∂t

with initial conditions and suitable boundary ones. Here y = y(x, t) represents the
temperature at the point x ∈ Ω and time t > 0. It is supposed that f = f 1 + f 2 ,
where f 2 is given and f 1 is a known function of the temperature of the form (see
Fig. 7.1)
− f 1 (x, t) ∈ ∂ j (x, t, y(x, t)) a.e. (x, t) ∈ Ω × (0, +∞).

Here ∂ j (x, t, ξ ) denotes generalized gradient of Clarke (see Clarke [9]) with
respect to the last variable of a function j : Ω × R+ × R → R which is assumed to
be locally Lipschitz in ξ (cf. Migórski and Ochal [34] and references therein). The
multi-valued function ∂ j (x, t, ·) : R → 2R is generally nonmonotone and it includes
the vertical jumps. In a physicist’s language it means that the law is characterized
172 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems

Fig. 7.1 Control laws

by the generalized gradient of a nonsmooth potential j (cf. Panagiotopoulos [35]).

Models of physical interest includes also the next (see, for example, Balibrea et al.
[2] and references therein): a model of combustion in porous media; a model of
conduction of electrical impulses in nerve axons; a climate energy balance model;
etc.
Let the following assumptions hold:

1 (R+ ) function c1 : R+ → R+
(H1) (Growth condition) There exist a t.u.i. in L loc
q p
and a constant c2 > 0 such that d V ∗ ≤ c1 (t) + c2 u V for any u ∈ V , d ∈
A(t, u), and a.e. t > 0;
1 (R+ ) function
(H2) (Sign condition) There exist a constant α > 0 and a t.u.i. in L loc
p
β : R+ → R+ such that d, uV ≥ α u V − β(t) for any u ∈ V , d ∈ A(t, u),
and a.e. t > 0;
(H3) (Strong measurability) If C ⊆ V ∗ is a closed set, then the set {(t, u) ∈
(0, +∞) × V : A(t, u) ∩ C = ∅} is a Borel subset in (0, +∞) × V ;
(H4) (Pointwise pseudomonotonicity) Let for a.e. t > 0 the following two assump-
tions hold:
a) for every u ∈ V the set A(t, u) is nonempty, convex, and weakly compact
one in V ∗ ;
b) if a sequence {u n }n≥1 converges weakly in V towards u ∈ V as n → +∞,
dn ∈ A(t, u n ) for any n ≥ 1, and lim supn→+∞ dn , u n − uV ≤ 0, then for
any ω ∈ V there exists d(ω) ∈ A(t, u) such that

lim inf dn , u n − ωV ≥ d(ω), u − ωV .

n→+∞

Let 0 ≤ τ < T < +∞. As a weak solution of evolution inclusion (7.33) on the
interval [τ, T ] we consider an element u(·) of the space L p (τ, T ; V ) such that for
some d(·) ∈ L q (τ, T ; V ∗ ) it is fulfilled:
7.4 Example of Applications 173
 T  T
− (ξ (t), y(t))dt + d(t), ξ(t)V dt = 0 ∀ξ ∈ C0∞ ([τ, T ]; V ), (7.34)
τ τ

and d(t) ∈ A(t, y(t)) for a.e. t ∈ (τ, T ). For fixed nonnegative τ and T , τ < T , let
us consider
∗
X τ,T = L p (τ, T ; V ), X τ,T = L q (τ, T ; V ∗ ),
∗
∗
Wτ,T = {y ∈ X τ,T | y ∈ X τ,T }, Aτ,T : X τ,T → 2 X τ,T \ {∅},
∗
Aτ,T (y) = {d ∈ X τ,T | d(t) ∈ A(t, y(t)) for a.e. t ∈ (τ, T )},

where y is a derivative of an element u ∈ X τ,T in the sense of D([τ, T ]; V ∗ ) (see,

for example, Gajewski, Gröger, and Zacharias [11, Definition IV.1.10]). Gajewski,
Gröger, and Zacharias [11, Theorem IV.1.17] provide that the embedding Wτ,T ⊂
C([τ, T ]; H ) is continuous and dense. Moreover,
 T
(u(T ), v(T )) − (u(τ ), v(τ )) = u (t), v(t)V + v (t), u(t)V dt, (7.35)
τ

for any u, v ∈ Wτ,T .

Migórski [33, Lemma 7, p. 516] (see the paper and references therein) and the
∗
assumptions above provide that the multi-valued mapping Aτ,T : X τ,T → 2 X τ,T \ {∅}
satisfies the listed below properties:
(P1) There exists a positive constant C1 = C1 (τ, T ) such that d X τ,T ∗ ≤ C1 (1 +
p−1
y X τ,T ) for any y ∈ X τ,T and d ∈ Aτ,T (y);
(P2) There exist positive constants C2 = C2 (τ, T ) and C3 = C3 (τ, T ) such that
p
d, y X τ,T ≥ C2 y X τ,T − C3 for any y ∈ X τ,T and d ∈ Aτ,T (y);
∗
(P3) The multi-valued mapping Aτ,T : X τ,T → 2 X τ,T \ {∅} is (generalized) pseudo-
monotone on Wτ,T , i.e.
(a) for every y ∈ X τ,T the set Aτ,T (y) is a nonempty, convex and weakly com-
∗
pact one in X τ,T ;
(b) Aτ,T is upper semi-continuous from every finite dimensional subspace X τ,T
∗
into X τ,T endowed with the weak topology;
∗
(c) if a sequence {yn , dn }n≥1 ⊂ Wτ,T × X τ,T converges weakly in Wτ,T ×
∗ ∗
X τ,T towards (y, d) ∈ Wτ,T × X τ,T , dn ∈ Aτ,T (yn ) for any n ≥ 1, and
lim supn→+∞ dn , yn − y X τ,T ≤ 0, then d ∈ Aτ,T (y) and limn→+∞
dn , yn  X τ,T = d, y X τ,T .
Formula (7.34) and the definition of the derivative for an element from D([τ, T ];
V ∗ ) yield that each weak solution y ∈ X τ,T of Problem (7.33) on [τ, T ] belongs
to the space Wτ,T and y + Aτ,T (y)  0̄. On the contrary, suppose that y ∈ Wτ,T
satisfies the last inclusion, then y is a weak solution of Problem (7.33) on [τ, T ].
Assumption (H1), properties (P1)–(P3), and Denkowski, Migórski, and
Papageorgiou [10, Theorem 1.3.73] (see also Zgurovsky, Mel’nik, and Kasyanov
174 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems

[47, Chap. 2] and references therein) provide the existence of a weak solution of
Cauchy problem (7.33) with initial data y(τ ) = y (τ ) on the interval [τ, T ], for any
y (τ ) ∈ H .
For fixed τ and T , such that 0 ≤ τ < T < +∞, we denote

Dτ,T (y (τ ) ) = {y(·) | y is a weak solution of (7.33) on [τ, T ], y(τ ) = y (τ ) }, y (τ ) ∈ H.

We remark that Dτ,T (y (τ ) ) = ∅ and Dτ,T (y (τ ) ) ⊂ Wτ,T , if 0 ≤ τ < T < +∞ and

y (τ ) ∈ H . Moreover, the concatenation of weak solutions of Problem (7.33) is a
weak solutions too, i.e. if 0 ≤ τ < t < T , y (τ ) ∈ H , y(·) ∈ Dτ,t (y (τ ) ), and v(·) ∈
Dt,T (y(t)), then
y(s), s ∈ [τ, t],
z(s) =
v(s), s ∈ [t, T ],

belongs to Dτ,T (y (τ ) ); cf. Zgurovsky et al. [48, pp. 55–56].

Gronwall’s lemma provides that for any finite time interval [τ, T ] ⊂ R+ each
weak solution y of Problem (7.33) on [τ, T ] satisfies the estimates
 t  t  s
p
y(t) 2
H −2 β(ξ )dξ + 2α y(ξ ) V dξ ≤ y(s) 2
H −2 β(ξ )dξ, (7.36)
0 s 0

 t
2 −2αγ (t−s)
y(t) 2
H ≤ y(s) He +2 (β(ξ ) + αγ )e−2αγ (t−ξ ) dξ, (7.37)
s

p p
where t, s ∈ [τ, T ], t ≥ s; γ > 0 is a constant such that γ u H ≤ u V for any
u ∈ V ; cf. Zgurovsky et al. [48, p. 56]. In the proof of (7.37) we used the inequality
p
u 2H − 1 ≤ u H for any u ∈ H .
Therefore, any weak solution y of Problem (7.33) on a finite time interval
[τ, T ] ⊂ R+ can be extended to a global one, defined on [τ, +∞). For arbitrary
τ ≥ 0 and y (τ ) ∈ H let Dτ (y (τ ) ) be the set of all weak solutions (defined on
[τ, +∞)) of Problem (7.33) with initial data y(τ ) = y (τ ) . Let us consider the fam-
ily Kτ+ = ∪ y (τ ) ∈H Dτ (y (τ ) ) of all weak solutions of Problem (7.33) defined on the
semi-infinite time interval [τ, +∞).
Properties (P1)–(P2) imply assumptions (A1) and (A2). Therefore, Theorem 7.1
yields that the m-semiflow G, defined in (7.8), has a compact global attractor  in
the phase space H.

References

1. Babin, A.V., Vishik, M.I.: Attractors of Evolution Equations (Russian). Nauka, Moscow (1989)
2. Balibrea, F., Caraballo, T., Kloeden, P.E., Valero, J.: Recent developments in dynam-
ical systems: three perspectives. Int. J. Bifurc. Chaos 20, 2591 (2010). doi:10.1142/
S0218127410027246
References 175

3. Ball, J.M.: Continuity properties and global attractors of generalized semiflows and the Navier-
Stokes equations. Nonlinear Sci. 7(1997), 475–502. Erratum, ibid 8:233: Corrected version
appears in Mechanics: from Theory to Computation. Springer 2000, 447–474 (1998)
4. Chepyzhov, V.V., Vishik, M.I.: Trajectory attractors for evolution equations. C. R. Acad. Sci.
Paris. Serie I 321, 1309–1314 (1995)
5. Chepyzhov, V.V., Vishik, M.I.: Evolution equations and their trajectory attractors. J. Math.
Pures Appl. 76, 913–964 (1997)
6. Chepyzhov, V.V., Vishik, M.I.: Trajectory and global attractors for 3D Navier–Stokes system.
Mat. Zametki. 71, 177 (2002). doi:10.1023/A:1014190629738
7. Chepyzhov, V.V., Vishik, M.I.: Attractors for Equations of Mathematical Physics. American
Mathematical Society, Providence RI (2002)
8. Chepyzhov, V.V., Vishik, M.I.: Trajectory attractor for reaction-diffusion system with diffusion
coefficient vanishing in time. Discret. Contin. Dyn. Syst. 27(4), 1498–1509 (2010)
9. Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley, New York (1983)
10. Denkowski, Z., Migórski, S.: An Introduction to Nonlinear Analysis: Applications. Kluwer
Academic/Plenum Publishers, Boston (2003)
11. Gajewski, H., Gröger, K., Zacharias, K.: Nichtlineare Operatorgleichungen und Operatordif-
ferentialgleichungen. Akademie-Verlag, Berlin (1978)
12. Gluzman, M.O., Gorban, N.V., Kasyanov, P.O.: Lyapunov type functions for classes of
autonomous parabolic feedback control problems and applications. Appl. Math. Lett. (2014).
doi:10.1016/j.aml.2014.08.006
13. Gorban, N.V., Kasyanov, P.O.: On regularity of all weak solutions and their attractors for
reaction-diffusion inclusion in unbounded domain. Continuous and Distributed Systems: The-
ory and Applications. Solid Mechanics and its Applications. Springer, Berlin (2014). doi:10.
1007/978-3-319-03146-0_15
14. Gorban, N.V., Kapustyan, O.V., Kasyanov, P.O.: Uniform trajectory attractor for non-
autonomous reaction-diffusion equations with Caratheodory’s nonlinearity. Nonlinear Anal.
Theory, Methods Appli. 98, 13–26 (2014). doi:10.1016/j.na.2013.12.004
15. Gorban, N.V., Kapustyan, O.V., Kasyanov, P.O., Paliichuk, L.S.: On global attractors for
autonomous damped wave equation with discontinuous nonlinearity. Continuous and Dis-
tributed Systems. Theory and Applications. Solid Mechanics and its Applications. Springer,
Berlin (2014). doi:10.1007/978-3-319-03146-0_16
16. Hale, J.K.: Asymptotic Behavior of Dissipative Systems. AMS, Providence, RI (1988)
17. Iovane, G., Kapustyan, A.V., Valero, J.: Asymptotic behavior of reaction-diffusion equations
with non-damped impulsive effects. Nonlinear Anal. 68, 2516–2530 (2008)
18. Kapustyan, A.V., Valero, J.: Weak and strong attractors for the 3D Navier–Stokes system. J.
Differ. Equ. 240(2), 249–278 (2007). doi:10.1016/j.jde.2007.06.008
19. Kapustyan, A.V., Valero, J.: On the Kneser property for the complex Ginzburg–Landau equation
and the Lotka–Volterra system with diffusion. J. Math. Anal. Appl. 357, 254–272 (2009)
20. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Pullback attractors for a class of extremal solutions
of the 3D Navier.-Stokes equations. J. Math. Anal. Appl. 373(2), 535–547 (2011). doi:10.1016/
j.jmaa.2010.07.040
21. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Structure and regularity of the global attractor of
a reaction-diffusion equation with non-smooth nonlinear term. Discret. Contin. Dyn. Syst. Ser.
A 34(10), 4155–4182 (2014). doi:10.3934/dcds.2014.34.4155
22. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Regular solutions and global attractors for reaction-
diffusion systems without uniqueness. Commun. Pure Appl. Anal. 13(5), 1891–1906 (2014).
doi:10.3934/cpaa.2014.13.1891
23. Kapustyan, A.V., Kasyanov, P.O., Valero, J., Zgurovsky, M.Z.: Sructure of uniform global
attractor for general non-autonomous reaction-diffusion system. Continuous and Distrib-
uted Systems: Theory and Applications. Solid Mechanics and Its Applications, pp. 163–180.
Springer, Berlin (2014)
24. Kasyanov, P.O.: Multi-valued dynamics of solutions of an autonomous differential-operator
inclusion with pseudomonotone nonlinearity. Cybern. Syst. Anal. 47, 800–811 (2011)
176 7 Uniform Global Attractors for Non-autonomous Dissipative Dynamical Systems

25. Kasyanov, P.O.: Multi-valued dynamics of solutions of autonomous operator differential equa-
tions with pseudomonotone nonlinearity. Math. Notes 92, 205–218 (2012)
26. Kasyanov, P.O., Mel’nik, V.S., Toscano, S.: Solutions of Cauchy and periodic problems for
evolution inclusions with multi-valued wλ0 -pseudomonotone maps. J. Differ. Equ. 249(6),
1258–1287 (2010). doi:10.1016/j.jde.2010.05.008
27. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Regularity of weak solutions and their attrac-
tors for a parabolic feedback control problem. Set-Valued Var. Anal. 21, 271 (2013). doi:10.
1007/s11228-013-0233-8
28. Kloeden, P.E., Marin-Rubio, P., Valero, J.: The Envelope attractor of non-strict multi-valued
dynamical systems with application to the 3D Navier–Stokes and reaction-diffusion equations.
Set-Valued Var. Anal. 21, 517–540 (2013). doi:10.1007/s11228-012-0228-x
29. Krasnosel’skii, M.A.: Topological Methods in the Theory of Nonlinear Integral Equations.
International Series of Monographs on Pure and Applied Mathematics. Oxford, London (1964)
30. Ladyzhenskaya, O.A.: Attractors for Semigroups and Evolution Equations. Cambridge Uni-
versity Press, Cambridge (1991)
31. Melnik, V.S., Valero, J.: On attractors of multi-valued semi-flows and generalized differential
equations. Set-Valued Anal. 6(1), 83–111 (1998)
32. Mel’nik, V.S., Valero, J.: On global attractors of multi-valued semiprocesses and
non-autonomous evolution inclusions. Set-Valued Anal. 8, 375 (2000). doi:10.1023/A:
1026514727329
33. Migórski, S.: Boundary hemivariational inequalities of hyperbolic type and applications. J.
Glob. Optim. 31(3), 505–533 (2005)
34. Migórski, S., Ochal, A.: Optimal control of parabolic hemivariational inequalities. J. Glob.
Optim. 17, 285–300 (2000)
35. Panagiotopoulos, P.D.: Inequality Problems in Mechanics and Applications: Convex and Non-
convex Energy Functions. Birkhauser, Basel (1985)
36. Sell, G.R.: Global attractors for the three-dimensional Navier–Stokes equations. J. Dyn. Differ.
Equ. 8(12), 1–33 (1996)
37. Smoller, J.: Shock Waves and Reaction-Diffusion Equations (Grundlehren der mathematischen
Wissenschaften). Springer, New York (1983)
38. Sohr, H.: The Navier–Stokes Equations: An Elementary Functional Analytic Approach.
Birkhauser, Basel (2001)
39. Temam, R.: Navier–Stokes Equations. North-Holland, Amsterdam (1979)
40. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Applied
Mathematical Sciences. Springer, New York (1988)
41. Valero, J., Kapustyan, A.V.: On the connectedness and asymptotic behaviour of solutions of
reaction-diffusion systems. J. Math. Anal. Appl. 323(1), 614–633 (2006). doi:10.1016/j.jmaa.
2005.10.042
42. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong trajectory attractor for a dissipative reaction-
diffusion system. Dokl. Math. 82(3), 869–873 (2010)
43. Warga, J.: Optimal Control of Differential and Functional Equations. Academic Press, Dublin
(1972)
44. Zgurovsky, M.Z., Kasyanov, P.O.: Multi-valued dynamics of solutions for autonomous operator
differential equations in strongest topologies. Continuous and Distributed Systems: Theory and
Applications. Solid Mechanics and its Applications. Springer, Berlin (2014). doi:10.1007/978-
3-319-03146-0_11
45. Zgurovsky, M.Z., Kasyanov, P.O.: Evolution inclusions in nonsmooth systems with applications
for earth data processing: uniform trajectory attractors for non-autonomous evolution inclu-
sions solutions with pointwise pseudomonotone mappings. Advances in Global Optimization.
Springer Proceedings in Mathematics and Statistics. Springer, Berlin (2015). doi:10.1007/978-
3-319-08377-3_28
46. Zgurovsky, M.Z., Kasyanov, P.O., Zadoianchuk, N.V.: Long-time behavior of solutions for qua-
silinear hyperbolic hemivariational inequalities with application to piezoelectricity problem.
Appl. Math. Lett. 25(10), 1569–1574 (2012)
References 177

47. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation Inequali-
ties for Earth Data Processing II. Springer, Berlin (2011)
48. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
inclusions and variation Inequalities for Earth data processing III. Springer, Berlin (2012)
Chapter 8
Uniform Trajectory Attractors
for Non-autonomous Nonlinear Systems

Abstract In this chapter we study uniform trajectory attractors for non-autonomous

nonlinear systems. In Sect. 8.1 we establish the existence of uniform trajectory attrac-
tor for non-autonomous reaction-diffusion equations with Carathéodory’s nonlinear-
ity. Section 8.2 devoted to structural properties of the uniform global attractor for
non-autonomous reaction-diffusion system in which uniqueness of Cauchy problem
is not guarantied. In the case of translation compact time-depended coefficients it is
established that the uniform global attractor consists of bounded complete trajectories
of corresponding multi-valued processes. Under additional sign conditions on non-
linear term we also prove (and essentially use previous result) that the uniform global
attractor is, in fact, bounded set in L ∞ (Ω) ∩ H01 (Ω). Section 8.3 devoted to uniform
trajectory attractors for nonautonomous dissipative dynamical systems. As applica-
tions we may consider FitzHugh–Nagumo system (signal transmission across axons),
complex Ginzburg–Landau equation (theory of superconductivity), Lotka–Volterra
system with diffusion (ecology models), Belousov–Zhabotinsky system (chemical
dynamics) and many other reaction-diffusion type systems from Sect. 2.4.

8.1 Uniform Trajectory Attractor for Non-autonomous

Reaction-Diffusion Equations with Carathéodory’s
Nonlinearity

Let N , M = 1, 2, . . ., Ω ⊂ R N be a bounded domain with sufficiently smooth

boundary ∂Ω. We consider a problem of long-time behavior of all globally defined
weak solutions for the non-autonomous parabolic problem (named reaction-diffusion
or RD-system; see Chap. 4 and [1–23]).

yt = aΔy − f (x, t, y), x ∈ Ω, t > 0,
(8.1)
y|∂Ω = 0,

as t → +∞, where y = y(x, t) = (y (1) (x, t), ..., y (M) (x, t)) is unknown vector-
function, f = f (x, t, y) = ( f (1) (x, t, y), ..., f (M) (x, t, y)) is given function, a is
real M × M matrix with positive symmetric part.
© Springer International Publishing AG 2018 179
M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_8
180 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

Throughout this section we suppose that the listed below assumptions hold (see
Chap. 5).
Assumption I Let pi ≥ 2 and qi > 1 are such that p1i + q1i = 1, for any i =
1, 2, . . . , M. Moreover, there exists a positive constant d such that 21 (a+a ∗ ) ≥ d I ,
where I is unit M × M matrix, a ∗ is a transposed matrix for a.
Assumption II The interaction function f = ( f (1) , ..., f (M) ) : Ω × R+ × R M →
R M satisfies the standard Carathéodory’s conditions, i.e. the mapping (x, t, u) →
f (x, t, u) is continuous in u ∈ R M for a.e. (x, t) ∈ Ω × R+ , and it is measurable
in (x, t) ∈ Ω × R+ for any u ∈ R M .
Assumption III (Growth Condition) There exist a t.u.i. in L loc 1 (R+ ; L 1 (Ω)) func-
tion c1 : Ω × R+ → R+ and a constant c2 > 0 such that


M
 (i)  
M
 (i)  pi
 f (x, t, u)qi ≤ c1 (x, t) + c2 u 
i=1 i=1

for any u = (u (1) , ..., u (M) ) ∈ R M , and a.e. (x, t) ∈ Ω × R+ .

Assumption IV (Sign Condition). There exists a constant α > 0 and a t.u.i. in
1 (R+ ; L 1 (Ω)) function β : Ω × R+ → R+ such that
L loc


M 
M
 (i)  pi
f (i) (x, t, u)u (i) ≥ α u  − β(x, t)
i=1 i=1

for any u = (u (1) , ..., u (M) ) ∈ R M , and a.e. (x, t) ∈ Ω × R+ .

In further arguments we will use standard functional Hilbert spaces H =
(L 2 (Ω)) M , V = (H01 (Ω)) M , and V ∗ = (H −1 (Ω)) M with standard respective inner
products and norms (·, ·) H and · H , (·, ·)V and · V , and (·, ·)V ∗ and · V ∗ , vector
notations p = ( p1 , p2 , ..., p M ) and q = (q1 , q2 , ..., q M ), and the spaces

Lp (Ω) := L p1 (Ω) × ... × L p M (Ω), Lq (Ω) := L q1 (Ω) × ... × L q M (Ω),

Lp (τ, T ; Lp (Ω)) := L p1 (τ, T ; L p1 (Ω)) × ... × L p M (τ, T ; L p M (Ω)),
Lq (τ, T ; Lq (Ω)) := L q1 (τ, T ; L q1 (Ω)) × ... × L q M (τ, T ; L q M (Ω)), 0 ≤ τ < T < +∞.

Let 0 ≤ τ < T < +∞. We recall that a function y = y(x, t) ∈ L2 (τ, T ; V ) ∩

Lp (τ, T ; Lp (Ω)) is a weak solution of Problem (8.1) on [τ, T ], if for any function
ϕ = ϕ(x) ∈ (C0∞ (Ω)) M , the following identity holds
 
d
y(x, t) · ϕ(x)d x + {a∇ y(x, t) · ∇ϕ(x) + f (x, t, y(x, t)) · ϕ(x)}d x = 0
dt Ω Ω

in the sense of scalar distributions on (τ, T ).

In the general case Problem (8.1) on [τ, T ] with initial condition y(x, τ ) = yτ (x)
in Ω has more than one weak solution with yτ ∈ H (cf. Zgurovsky et al. [23] and
8.1 Uniform Trajectory Attractor for Non-autonomous Reaction-Diffusion … 181

references therein). Thus, for investigation of the long-time behavior as t → +∞

of all weak solutions of Problem (8.1) with initial data from H , the results for
uniform global and trajectory attractors of multi-valued semi-processes in infinite-
dimensional spaces were applied; Babin and Vishik [2], Chepyzhov and Vishik [6],
Mel’nik and Valero [14, 15] and references therein. Theses approaches were applied
to various non-autonomous problems of the form

∂t y(t) = Aσ (t) (y(t)), (8.2)

where σ (s), s ≥ 0, is a functional parameter called the time symbol of Eq. (8.2) (t
is replaced by s). In applications to mathematical physics equations, a function σ (s)
consists of all time-dependent terms of the equation under consideration: external
forces, parameters of mediums, interaction functions, control functions, etc. It is
assumed that the symbol σ of Eq. (8.2) belongs to a Hausdorff topological space Ξ+
of functions defined on R+ with values in some complete metric space. Usually, in
applications, the topology in the space Ξ+ is a local convergence topology on any
segment [t1 , t2 ] ⊂ R+ . Further, they consider the family of Eq. (8.2) with various
symbols σ (s) belonging to a set Σ ⊆ Ξ+ . The set Σ is called the symbol space
of the family of Eq. (8.2). It is assumed that the set Σ, together with any symbol
σ (s) ∈ Σ, contains all positive translations of σ (s): σ (t + s) = T (t)σ (s) ∈ Σ
for any t, s ≥ 0. The symbol space Σ is invariant with respect to the translation
semigroup {T (t)}t≥0 : T (t)Σ ⊆ Σ for any t ≥ 0. To prove the existence of uniform
trajectory attractor they supposed that the symbol space Σ with the topology induced
from Ξ+ is a compact metric space. Mostly in applications, as a symbol space Σ it is
natural to consider the hull of translation-compact function σ0 (s) in an appropriate
Hausdorff topological space Ξ+ . The direct realization of this approach for Problem
(8.1) is problematic without any additional assumptions for parameters of Problem
(8.1) and requires the translation-compactness of the symbol σ (s) = f (·, s, ·) in
some compact Hausdorff topological space of mappings act from R+ to some metric
space of Carathéodory’s vector-functions satisfying growth and signed assumptions.
To avoid this technical difficulties we present the alternative direct approach for the
existence and construction of the uniform trajectory attractor for all weak solutions
for Problem (8.1).
The main purpose of this section is to investigate uniform long-time behavior of
all globally defined weak solutions for Problem (8.1) with initial data u τ ∈ H under
listed above assumptions. The main results of this paper are: (i) the existence of uni-
form trajectory attractor for all globally defined weak solutions of non-autonomous
reaction-diffusion equations with Carathéodory’s nonlinearity (Theorem 8.1), and
(ii) sufficient conditions for the existence of uniform trajectory attractor in strongest
topologies (Theorem 8.2).
In further arguments as a Banach space Ft1 ,t2 we consider either C([t1 , t2 ]; H ) or
Wt1 ,t2 with respective topologies of strong convergence, where 0 ≤ t1 < t2 < +∞.
Consider the Fréchet space

F loc (R+ ) := {y : R+ → H : Πt1 ,t2 y ∈ Ft1 ,t2 for any [t1 , t2 ] ⊂ R+ },

182 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

Fig. 8.1 Translation

operation

where Πt1 ,t2 is the restriction operator to the interval [t1 , t2 ]; Chepyzhov and Vishik [6,
p. 918]. We remark that the sequence { f n }n≥1 converges (converges weakly respec-
tively) in F loc (R+ ) towards f ∈ F loc (R+ ) as n → +∞ if and only if the sequence
{Πt1 ,t2 f n }n≥1 converges (converges weakly respectively) in Ft1 ,t2 towards Πt1 ,t2 f as
n → +∞ for any finite interval [t1 , t2 ] ⊂ R+ .
We denote T (h)y(·) = yh (·), where yh (t) = y(t + h) for any y ∈ F loc (R+ ) and
t, h ≥ 0 (see Fig. 8.1).
In the autonomous case, when f (x, t, y) does not depend on t, the long-time
behavior of all globally defined weak solutions for Problem (8.1) is described by
using trajectory and global attractors theory. In this situation the set K + := K0+ is
translation semi-invariant, i.e. T (h)K + ⊆ K + for any h ≥ 0. As trajectory attrac-
tor it is considered a classical global attractor for translation semigroup {T (h)}h≥0 ,
that acts on K + .
In the non-autonomous case we notice that T (h)K0+  K0+ . Therefore, we
need to consider united trajectory space that includes all globally defined on any
[τ, +∞) ⊆ R+ weak solutions of Problem (8.1) shifted to τ = 0:
 
K∪+ := y( · + τ ) ∈ W loc (R+ ) : y( · ) ∈ Kτ+ , (8.3)
τ ≥0

Note that T (h){y( · + τ ) : y ∈ Kτ+ } ⊆ {y( · + τ + h) : y ∈ Kτ++h } for any

τ, h ≥ 0. Therefore, T (h)K∪+ ⊆ K∪+ for any h ≥ 0.
8.1 Uniform Trajectory Attractor for Non-autonomous Reaction-Diffusion … 183

To define an uniform trajectory attractor, the united trajectory space need to be a

closed subset of a Polish space. Further we consider extended united trajectory space
for Problem (8.1):
KF+loc (R+ ) = clF loc (R+ ) K∪+ , (8.4)

where clF loc (R+ ) [ · ] is the closure in F loc (R+ ). We note that

T (h)KF+loc (R+ ) ⊆ KF+loc (R+ ) for any h ≥ 0,

because

ρF loc (R+ ) (T (h)u, T (h)v) ≤ ρF loc (R+ ) (u, v) for any u, v ∈ F loc (R+ ),

where ρF loc (R+ ) is a standard metric on Fréchet space F loc (R+ ); cf. Chepyzhov and
Vishik [6]; Vishik et al. [21].
A set P ⊂ F loc (R+ ) ∩ L ∞ (R+ ; H ) is said to be a uniformly attracting set (cf.
Chepyzhov and Vishik [6, p. 921]) for the extended united trajectory space KF+loc (R+ )
of Problem (8.1) in the topology of F loc (R+ ), if for any bounded in L ∞ (R+ ; H ) set
B ⊆ KF+loc (R+ ) and any segment [t1 , t2 ] ⊂ R+ the following relation holds:

distFt1 ,t2 (Πt1 ,t2 T (t)B, Πt1 ,t2 P) → 0, t → +∞, (8.5)

where distFt1 ,t2 is the Hausdorff semi-metric.

A set U ⊂ KF+loc (R+ ) is said to be a uniform trajectory attractor (cf. Chepyzhov
and Vishik [6, p. 921] and references therein) of the translation semigroup {T (t)}t≥0
on KF+loc (R+ ) in the induced topology from F loc (R+ ), if

(i) U is a compact set in F loc (R+ ) and bounded in L ∞ (R+ ; H );

(ii) U is strictly invariant with respect to {T (h)}h≥0 , i.e. T (h)U = U ∀h ≥ 0;
(iii) U is a minimal uniformly attracting set for KF+loc (R+ ) in the topology of
F loc (R+ ), i.e. U belongs to any compact uniformly attracting set P of
KF+loc (R+ ) : U ⊆ P.
Note that uniform trajectory attractor of the translation semigroup {T (t)}t≥0 on
KF+loc (R+ ) in the induced topology from F loc (R+ ) coincides with the classical global
attractor for the continuous semi-group {T (t)}t≥0 defined on KF+loc (R+ ) .
Assumptions I–IV are sufficient conditions for the existence of uniform trajectory
attractor for weak solutions of Problem (8.1) in the topology of C loc (R+ ; H ).

Theorem 8.1 Let Assumptions I–IV hold. Then there exists an uniform trajectory
attractor U ⊂ KC+loc (R+ ;H ) of the translation semigroup {T (t)}t≥0 on KC+loc (R+ ;H )
in the induced topology from C loc (R+ ; H ). Moreover, there exists a compact in
C loc (R+ ; H ) uniformly attracting set P ⊂ C loc (R+ ; H ) ∩ L ∞ (R+ ; H ) for the
184 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

extended united trajectory space KC+loc (R+ ;H ) of Problem (8.1) in the topology of
C loc (R+ ; H ) such that U coincides with ω-limit set of P:


U = clC loc (R+ ;H ) T (h)P . (8.6)
t≥0 h≥t

For the existence of uniform trajectory attractor in the strong topology of the nat-
ural extended phase space W loc (R+ ) it is necessary to claim that additional assump-
tion holds (see Example 8.1). To formulate this additional assumption we provide
some auxiliary constructions. A function ϕ ∈ L loc 1 (R+ ; L 1 (Ω)) is called translation-
compact (tr.-c.) in L loc
1 (R + ; L 1 (Ω)), if the set {ϕ( · + h) : h ≥ 0} is precompact
in L loc
1 (R+ ; L 1 (Ω)); cf. Chepyzhov and Vishik [6, p. 917]. Note that a function
ϕ ∈ L loc
1 (R ;
+ 1L (Ω)) is tr.-c. in L
1
loc
(R + ; L 1 (Ω)) if and only if two conditions hold:
t+h
(a) the set t ϕ(s)ds : t ≥ 0 is precompact in L 1 (Ω) for any h > 0; (b) there
exists a function ψ(s), ψ(s) → 0+ as s → 0+ such that
 t+1 
|ϕ(x, s) − ϕ(x, s + h)|d xds ≤ ψ(|h|) for any t ≥ 0 and h ≥ −t;
t Ω

Chepyzhov and Vishik [6, Proposition 6.5].

Assumption V Let the conditions hold:
(i) the functions c1 and β from Assumptions (III) and (IV) respectively are tr.-c.
1 (R+
in L loc  ; L1 (Ω)); 
t+h
(ii) the set h1 t f ( · , s, u)ds : t ≥ 0, h ∈ (0, h 0 ), u R M ≤ R is precom-
pact in (L 1 (Ω)) M for any R > 0 and some h 0 = h 0 (R) > 0;
(iii) for any r > 0 there exist a nondecreasing function ψ(s, r ) : R2+ → R+ ,
ψ(s, r ) → 0+ as s → 0+, and h 0 = h 0 (r ) > 0 such that

M    
1  t+h 1  (i) 
 f (x, s, u) − f (i) (x, s + h 2 , v) d xds ≤ ψ(|h 2 | + u − v R M , r )
h1 t Ω
i=1

for each t ≥ 0, h 1 ∈ (0, h 0 ), h 2 ≥ −t, and u, v ∈ R M such that

u RM , v RM ≤ r .
Remark 8.1 Let us discuss sufficient conditions for Assumption V.
(i) The autonomous case. Let f does not depend on the time variable t and it
satisfies Assumptions I–IV with c1 , β ∈ L 1 (Ω) (in particular, assumptions from
Vishik et al. [21] hold). Then Assumption V hold; see Remark 4.1.
(ii) The non-autonomous case. Let f = f (t, u) is jointly continuous mapping,
it satisfies Assumptions I–IV with positive constants c1 and β, and f being tr.-c. in
C loc (R+ ; C(R M )), that is,

f (t, u) − f (s, v) RM ≤ ω (|t − s| + u − v RM , K)

8.1 Uniform Trajectory Attractor for Non-autonomous Reaction-Diffusion … 185

for all t, s ∈ R+ , u R M , v R M ≤ K , K > 0, where ω (l, K ) → 0, as l → 0+;

see, for example, Chepyzhov and Vishik [6, p. 105], Kapustyan and Valero [8–10],
where uniform global in H and uniform trajectory in C loc (R+ ; H ) attractors were
investigated. Then Assumption V holds.
(iii) The sufficient condition for Assumption V(iii) is: for any r > 0 there exist a
nondecreasing function ψ(s, r ) : R2+ → R+ , ψ(s, r ) → 0+ as s → 0+, such that

M 
  (i) 
 f (x, t, u) − f (i) (x, t + h, v) d xds ≤ ψ(|h| + u − v RM , r )
i=1 Ω

for each t ≥ 0, h ≥ −t, and u, v ∈ R M such that u R M , v R M ≤ r .

Note that Assumption V is a generalization of the above assumptions to the case
when f depends on the space, time and state variables simultaneously and it is
not necessarily continuous by t. Meanwhile, Example 8.1 below provide piecewise
continuous function f that satisfies Assumptions I-IV, but it does not satisfy Assump-
tion V. Moreover, the statement of Theorem 8.2 below does not hold for Problem
(8.1) with such interaction function.
The main result on the existence of uniform trajectory attractor for weak solutions
of Problem (8.1) in the topology of W loc (R+ ) has the following form:
Theorem 8.2 Let Assumptions I–V hold. Then there exists an uniform trajectory
attractor U ⊂ KW+loc (R+ ) of the translation semigroup {T (t)}t≥0 on KW+loc (R+ ) in
the induced topology from W loc (R+ ). Moreover, there exists a compact in W loc (R+ )
uniformly attracting set P ⊂ W loc (R+ ) ∩ L ∞ (R+ ; H ) for the extended united
trajectory space KW+loc (R+ ) of Problem (8.1) in the topology of W loc (R+ ) such that
U coincides with ω-limit set of P:


U = clW loc (R+ ) T (h)P . (8.7)
t≥0 h≥t

Remark 8.2 All statements of Theorems 8.1 and 8.2 hold for the function f (x, t, y)
equals to the sum of an interaction function f 1 (x, t, y), satisfying Assumptions I–
∗
IV (Assumptions I–V respectively), and an external force g ∈ L loc 2 (R+ ; V ). In
∗
Theorem 8.1 g is need to be translation bounded in L 2 (R+ ; V ) and g is transla-
loc
∗
2 (R+ ; V ) in Theorem 8.2 respectively. The proofs are similar
tion compact in L loc
with some standard technical modifications. To simplify the conclusions, further we
consider the case g ≡ 0.
Proof of Theorems 8.1 and 8.2 The proofs of both two theorems are similar and
based on the respective statements of Theorems 4.1 and 4.2. To avoid reduplication
we set F loc (R+ ) := C loc (R+ ; H ) for the proof of Theorem 8.1 and F loc (R+ ) :=
W loc (R+ ) for the proof of Theorem 8.2 respectively.
We provide the proof in several steps. First, let us show that there exists a uni-
form trajectory attractor U ⊂ KF+loc (R+ ) of the translation semigroup {T (t)}t≥0 on
186 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

KF+loc (R+ ) in the induced topology from F loc (R+ ). Theorem 4.1, if F loc (R+ ) =
C loc (R+ ; H ), and Theorem 4.2, if F loc (R+ ) = W loc (R+ ), yields that the translation
semigroup {T (t)}t≥0 has a compact absorbing (and, therefore, an uniformly attract-
ing) set in the space of trajectories KF+loc (R+ ) ; Zgurovsky et al. [23] and references
therein. This set can be constructed as follows: 1) consider PF loc (R+ ) , the intersection
of KF+loc (R+ ) with a ball in the space of bounded continuous functions on R+ with
values in H , Cb (R+ ; H ), of sufficiently large radius; 2) shift the resulting set by any
fixed distance h > 0. Thus, we obtain T (h)PF loc (R+ ) , a set with the required proper-
ties. Recall that the semigroup {T (t)}t≥0 is continuous. Therefore, the set PF loc (R+ )
is a compact absorbing (and, therefore, an uniformly attracting) for KF+loc (R+ ) with
the induced topology of F loc (R+ ). Then we can apply, for example, Theorem 2.2
from Chepyzhov and Vishik [6, Chap. XI] and finish the proof. In particular, formula
(8.6) holds; cf. Babin and Vishik [2]; Melnik and Valero [14], Temam [20] etc.
The example provided below implies that additional Assumption V in Theorem 4.2
is essential for the existence of uniform trajectory attractor in strongest topology of
an extended united phase space of weak solutions for Problem (4.1).

Example 8.1 Interaction function f : Ω × R+ × R M → R M satisfies Assumptions

I–IV, Assumption V does not hold, and the statement of Theorem 4.2 does not
hold. Let N , M = 1, Ω = (0, π ), a = 1, f (x, t, u) := u − sin(x) · sin(π [t]t),
(x, t, u) ∈ Ω × R+ × R, where [t] is a largest integer, that does not exceed t. The
verifying of Assumptions I–IV and V(i,ii) is trivial. Assumption V(iii) does not hold,
because | sin(π k 2 ) − sin(π(k + 2k1 )k)| = 1 → 0, as k → +∞.
The statement of Theorem 4.2 does not hold. On the contrary assume that there
exists an uniform trajectory attractor U ⊂ KW+loc (R+ ) of the translation semigroup
{T (t)}t≥0 on KW+loc (R+ ) in the induced topology from W loc (R+ ). By definition of
an uniform trajectory attractor, since U is a compact set in W loc (R+ ), for each
y(·) ∈ KW+loc (R+ ) (we note that the set {T (h)y(·) : h ≥ 0} is bounded in L ∞ (R+ ; H ))
any monotone increasing unbounded sequence {h n }n≥1 ⊂ R+ has a subsequence
{h n k }k≥1 ⊆ {h n }n≥1 such that {T (h n k )y(·)}k≥1 is precompact in W loc (R+ ). On the
other hand, let
 t
y(x, t) := sin(x) sin(π [s]s)e−2(t−s) ds, (x, t) ∈ Ω × R+ ,
0

and h n := n, n = 1, 2, . . .. Note that y ∈ K0+ ⊆ KW+loc (R+ ) , and


∂ t+n
y(x, t +n) = −2 sin(x) sin(π [s](s))e−2(t+n−s) ds +(−1)n sin(x)·sin(π nt),
∂t 0

(x, t) ∈ Ω × (0, 1), n = 1, 2, . . .. Therefore,

lim inf Π0,1 T (n)y(·) − Π0,1 T (m)y(·) W0,1 > 0,

n,m→+∞
8.1 Uniform Trajectory Attractor for Non-autonomous Reaction-Diffusion … 187
 t+n
because the sequence of functions {(x, t) → sin(x) 0 sin(π [s]s)e−2(t+n−s) ds}n≥1 ,
∗
restricted on Ω × (0, 1), converges strongly in L 2 (0, 1; V ) ⊂ X 0,1 as n → +∞,
and the sequence of functions {(x, t) → π sin(x) · sin(π nt)}n≥1 , restricted on
√2
∗
Ω ×(0, 1), is orthonormal in X 0,1 . This is a contradiction with the existence of a sub-
sequence {h n k }k≥1 ⊆ {h n }n≥1 such that {T (h n k )y(·)}k≥1 is precompact in W loc (R+ ).
Therefore, the statement of Theorem 4.2 does not hold.

8.2 Structure of Uniform Global Attractor

for Non-autonomous Reaction-Diffusion Equations

In this section we study the structural properties of the uniform global attractor
of non-autonomous reaction-diffusion system in which the nonlinear term satisfy
suitable growth and dissipative conditions on the phase variable, suitable translation
compact conditions on time variable, but there is no condition ensuring uniqueness
of Cauchy problem. In autonomous case such system generates in the general case a
multi-valued semiflow having a global compact attractor (see [8, 12, 23]). Also, it is
known [12], that the attractor is the union of all bounded complete trajectories of the
semiflow. Here we prove the same result for non-autonomous system. More precisely,
we prove that the family of multi-valued processes, generated by weak solutions of
reaction-diffusion system, has uniform global attractor which is union of bounded
complete trajectories of corresponding processes. Using this result, we can prove
that under additional restrictions on nonlinear term obtained uniform global attractor
is bounded set in the space L ∞ (Ω) ∩ H01 (Ω).
In a bounded domain Ω ⊂ Rn with sufficiently smooth boundary ∂Ω we consider
the following non-autonomous parabolic problem (named RD-system)

u t = aΔu − f (t, u) + h(t, x), x ∈ Ω, t > τ,
(8.8)
u|∂Ω = 0,

where τ ∈ R is initial moment of time, u = u(t, x) = (u 1 (t, x), ..., u N (t, x)) is
unknown vector-function, f = ( f 1 , ..., f N ), h = (h 1 , ..., h N ) are given functions,
a is real N × N matrix with positive symmetric part 21 (a + a ∗ ) ≥ β I , β > 0,

h ∈ L loc
2
(R; (L 2 (Ω)) N ), f ∈ C(R × R N ; R N ), (8.9)

∃ C1 , C2 > 0, γi > 0, pi ≥ 2, i = 1, N such that ∀ t ∈ R, ∀ v ∈ R N


N
pi 
N
| f i (t, v)| pi −1 ≤ C1 (1 + |vi | pi ), (8.10)
i=1 i=1
188 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems


N 
N
f i (t, v)vi ≥ γi |vi | pi − C2 . (8.11)
i=1 i=1

In further arguments we will use standard functional spaces

 
N
H = (L (Ω)) with the norm |v| =
2 N 2
|vi (x)|2 d x,
Ω i=1
 
N
V = (H01 (Ω)) N with the norm v 2
= |∇vi (x)|2 d x.
Ω i=1

Let us denote V  = (H −1 (Ω)) N , qi = pi

pi −1
, p = ( p1 , ..., p N ), q = (q1 , ..., q N ),
L p (Ω) = L p1 (Ω) × ... × L p N (Ω).
 p
Definition 8.1 The function u = u(t, x) ∈ L loc 2
(τ, +∞; V ) L loc (τ, +∞;
L p (Ω)) is called a (weak) solution of the problem (8.8) on (τ, +∞) if for all
T > τ, v ∈ V ∩ L p (Ω)
   
d
u(t, x)v(x)d x + a∇u(t, x)∇v(x) + f (t, u(t, x))v(x) − h(t, x)v(x) d x = 0
dt
Ω Ω
(8.12)
in the sense of scalar distributions on (τ, T ).

From (8.10) and Sobolev embedding theorem we see that every solution of
(8.8) satisfies inclusion u t ∈ L loc (τ, +∞; H −r (Ω)), where r = (r1 , ..., r N ),
q

rk = max{1, n( 2 − pk )}. The following theorem is well-known result about global

1 1

resolvability of (8.8) for initial conditions from the phase space H .

Theorem 8.3 ([1, Theorem 2] or [6, p.284]) Under conditions (8.10), (8.11) for
every τ ∈ R, u τ ∈ H there exists at least one weak solution of (8.8) on (τ, +∞)
with u(τ ) = u τ (and it may be non unique) and any weak solution of (8.8) belongs to
C ([τ, +∞); H ). Moreover, the function t → |u(t)|2 is absolutely continuous and
for a.a. t ≥ τ the following energy equality holds

1 d
|u(t)|2 + (a∇u(t), ∇u(t)) + ( f (t, u(t)), u(t)) = (h(t), u(t)). (8.13)
2 dt
Under additional not-restrictive conditions on function f and h it is known that
solution of (8.8) generate non-autonomous dynamical system (two-parametric family
of m-processes), which has uniform global attractor. The aim of this paper is to give
description of the attractor in terms of bounded complete trajectories and show some
regularity property of this set.
Let (X, ρ) be a complete metric space. The Hausdorff semidistance from A to B
is given by
8.2 Structure of Uniform Global Attractor for Non-autonomous … 189

dist (A, B) = sup inf ρ(x, y),

x∈A y∈B

By A and Oε (A) = {x ∈ X | inf ρ(x, y) < ε} we denote closure and ε-

y∈A
neighborhood of the set A. Denote by P(X ) (β(X ), C(X ), K (X )) the set of all
non-empty (not-empty bounded, not-empty closed, not-empty compact) subsets of
X,
Rd = {(t, τ ) ∈ R2 |t ≥ τ }.

Let Σ be some complete metric space, {T (h) : Σ → Σ}h≥0 be a continuous

semigroup, acting on Σ. Note, that in most applications T (h) is shift semigroup.
Definition 8.2 Two-parameter family of multi-valued mappings {Uσ : Rd × X →
P(X )}σ ∈Σ is said to be the family of m-processes (family of MP), if ∀ σ ∈ Σ, τ ∈ R:
(1) Uσ (τ, τ, x) = x ∀x ∈ X,
(2) Uσ (t, τ, x) ⊆ Uσ (t, s, Uσ (s, τ, x)), ∀t ≥ s ≥ τ ∀x ∈ X,
(3) Uσ (t + h, τ + h, x) ⊆ UT (h)σ (t, τ, x)∀t≥ τ ∀h ≥ 0, ∀ x ∈ X,
where for A ⊂ X , B ⊂ Σ U B (t, s, A) = Uσ (t, s, x), in particular
σ ∈B x∈A


UΣ (t, τ, x) = Uσ (t, τ, x);
σ ∈Σ

see also Fig. 8.2.

Family of MP {Uσ |σ ∈ Σ} is called strict, if in conditions (2), (3) equality take
place.
Definition 8.3 A set A ⊂ X is called uniformly attracting for the family of MP
{Uσ |σ ∈ Σ}, if for arbitrary τ ∈ R, B ∈ β(X )

dist (UΣ (t, τ, B), A) → 0, t → +∞, (8.14)

Fig. 8.2 Concatenation

190 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

that is ∀ ε > 0, τ ∈ R and B ∈ β(X ) there exists T = T (τ, ε, B) such that

UΣ (t, τ, B) ⊂ Oε (A) ∀ t ≥ T.

For fixed B ⊂ X and (s, τ ) ∈ Rd let us define the following sets

 
τ τ
γs,σ (B) = Uσ (t, τ, B), γs,Σ (B) = UΣ (t, τ, B),
t≥s t≥s

τ
ωΣ (τ, B) = cl X (γs,Σ (B)).
s≥τ

 τ
It is clear that ωΣ (τ, B) = cl X (γs,Σ (B)) ∀ p ≥ τ.
s≥ p

Definition 8.4 The family of MP {Uσ |σ ∈ Σ} is called uniformly asymptotically

compact, if for arbitrary τ ∈ R and B ∈ β(X ) there exists A(τ, B) ∈ K (X ) such
that
UΣ (t, τ, B) → A(τ, B), t → +∞ in X.

τ
It is known [9] that if ∀τ ∈ R, ∀B ∈ β(X ) ∃T = T (τ, B) γT,Σ (B) ∈ β(X ), then
the condition of uniformly asymptotically compactness is equivalent to the following
one:
∀τ ∈ R ∀B ∈ β(X ) ∀tn  ∞

every sequence ξn ∈ UΣ (tn , τ, B) is precompact.

Definition 8.5 A set ΘΣ ⊂ X is called uniform global attractor of the family of MP

{Uσ |σ ∈ Σ}, if:
(1) ΘΣ is uniformly attracting set;
(2) for every uniformly attracting set Y we have ΘΣ ⊂ cl X Y .

Uniform global attractor ΘΣ ⊂ X is called invariant (semiinvariant), if ∀ (t, τ ) ∈

Rd
ΘΣ = UΣ (t, τ, ΘΣ ) (ΘΣ ⊂ UΣ (t, τ, ΘΣ )).

If ΘΣ is compact, invariant uniform global attractor, then it is called stable if

∀ ε > 0 ∃ δ > 0 ∀ (t, τ ) ∈ Rd

UΣ (t, τ, Oδ (ΘΣ )) ⊂ Oε (ΘΣ ).

The following sufficient conditions we can obtain with slight modifications from [9].

Theorem 8.4 (I) Let us assume that the family of MP {Uσ |σ ∈ Σ} satisfies the
following conditions:
8.2 Structure of Uniform Global Attractor for Non-autonomous … 191

(1) ∃ B0 ∈ β(X ) ∀ B ∈ β(X ) ∀τ ∈ R ∃ T = T (τ, B)

∀ t ≥ T UΣ (t, τ, B) ⊂ B0 ;

(2) {Uσ |σ ∈ Σ} is uniformly asymptotically compact.

Then {Uσ }σ ∈Σ has compact uniform global attractor
 
ΘΣ = ωΣ (τ, B) = ωΣ (0, B0 ) = ωΣ (τ, B0 ) ∀τ ∈ R. (8.15)
τ ∈R B∈β(X )

(II) If {Uσ }σ ∈Σ satisfy 1),2), Σ is compact and ∀t ≥ τ the mapping

(x, σ ) → Uσ (t, τ, x) (8.16)

has closed graph, then ΘΣ is semiinvariant.

If, moreover, ∀h ≥ 0 T (h)Σ = Σ and the family MP {Uσ |σ ∈ Σ} is strict, then
ΘΣ is invariant.
(III) If {Uσ }σ ∈Σ satisfy (1), (2), Σ is connected and compact, ∀t ≥ τ the mapping
(8.16) is upper semicontinuous and has connected values, B0 is connected set, then
ΘΣ is connected set.
(IV) If {Uσ |σ ∈ Σ} has compact, invariant uniform global attractor ΘΣ and the
following condition hold:

i f yn ∈ UΣ (tn , τ, xn ), tn → t0 , xn → x0 ,

then up to subsequence yn → y0 ∈ UΣ (t0 , τ, x0 ), (8.17)

then ΘΣ is stable.

Proof (I) From conditions (1), (2) due to [9] we have that ∀τ ∈ R ∀B ∈ β(X )
ωΣ (τ, B) = ∅, is compact, ωΣ (τ, B) ⊂ B0 and the set
 
ΘΣ = ωΣ (τ, B)
τ ∈R B∈β(X )

is uniform global attractor. Let us prove that ωΣ (τ, B) ⊂ ωΣ (τ0 , B0 ) ∀τ, τ0 ∈ R.

t t t t
Uσ (t, τ, B) ⊂ Uσ (t, , Uσ ( , τ, B)) ⊂ UT ( 2t −τ0 )σ ( + τ0 , τ0 , Uσ ( , τ, B)) ⊂
2 2 2 2
t t
⊂ UΣ ( + τ0 , τ0 , B0 ), if ≥ T (τ, B) + |τ0 | + |τ | := T.
2 2
So, for t ≥ 2T
t
UΣ (t, τ, B) ⊂ UΣ ( + τ0 , τ0 , B0 ).
2
192 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

Then for s ≥ 2T
  t 
UΣ (t, τ, B) ⊂ UΣ ( + τ0 , τ0 , B0 ) = UΣ ( p, τ0 , B0 ),
2
t≥s t≥s s
p≥ 2 +τ0

 
UΣ (t, τ, B) = ωΣ (τ, B) ⊂ UΣ ( p, τ0 , B0 ) =
s≥2T t≥s s≥2T p≥ 2s +τ0


UΣ ( p, τ0 , B0 ) = ωΣ (τ0 , B0 ).
s  ≥T +τ0 p≥s 

So we deduce equality (8.15).

(II) Due to (8.15) ∀ξ ∈ ΘΣ = ωΣ (τ, B0 ) ∃tn  +∞, ∃σn ∈ Σ ∃ξn ∈
UΣn (tn , τ, B0 ) such that ξ = lim ξn . Then
n→∞

ξn ∈ Uσn (tn − t − τ + t + τ, τ, B0 ) ⊂

⊂ Uσn (tn − t − τ + t + τ, tn − t + τ, Uσn (tn − t + τ, τ, B0 )) ⊂ UT (tn −t)σn (t, τ, ηn ),

where ηn ∈ Uσn (tn − t + τ, τ, B0 ), t ≥ τ and for sufficiently large n ≥ 1.

From uniform asymptotically compactness we have that on some subsequence
ηn → η ∈ ωΣ (τ, B0 ) = ΘΣ ,

T (tn − t)σn → σ ∈ Σ.

Then from (8.16) we deduce:

ξ ∈ UΣ (t, τ, ΘΣ ),

and therefore ΘΣ ⊂ UΣ (t, τ, ΘΣ ).

Other statements of the theorem are proved similarly to [9].
Theorem is proved.
Corollary 8.1 If for the family of MP {Uσ }σ ∈Σ we have:
(1) ∀h ≥ 0 T (h)Σ = Σ;
(2) ∀(t, τ ) ∈ Rd ∀h ≥ 0 ∀σ ∈ Σ ∀x ∈ X

Uσ (t + h, τ + h, x) = UT (h)σ (t, τ, x),

then all conditions of previous theorem can be verified only for τ = 0.

Proof Under conditions (1), (2) ∀t ≥ τ if τ ≥ 0 then

Uσ (t, τ, x) = UT (τ )σ (t − τ, 0, x),
8.2 Structure of Uniform Global Attractor for Non-autonomous … 193

and if τ ≤ 0 then ∃σ  ∈ Σ : σ = T (−τ )σ  , so

Uσ (t, τ, x) = UT (−τ )σ  (t, τ, x) = Uσ  (t − τ, 0, x).

In the single-valued case it is known [6], that the uniform global attractor consists
of bounded complete trajectories of processes {Uσ }σ ∈Σ .
Definition 8.6 The mapping ϕ : [τ, +∞) → X is called trajectory of MP Uσ , if
∀t ≥ s ≥ τ
ϕ(t) ∈ Uσ (t, s, ϕ(s)). (8.18)

If for ϕ : R → X the equality (8.18) takes place ∀t ≥ s, then ϕ is called complete

trajectory.
Now we assume that for arbitrary σ ∈ Σ and τ ∈ R we have the set K στ of
mappings ϕ : [τ, +∞) → X such that:
(a) ∀x ∈ X ∃ϕ(·) ∈ K στ such, that ϕ(τ ) = x;
(b) ∀ϕ(·) ∈ K στ ∀s ≥ τ ϕ(·)|[s,+∞) ∈ K σs ;
(c) ∀h ≥ 0 ∀ϕ(·) ∈ K στ +h ϕ(· + h) ∈ K Tτ (h)σ .
Let us put
Uσ (t, τ, x) = {ϕ(t)|ϕ(·) ∈ K στ , ϕ(τ ) = x}. (8.19)

Lemma 8.1 Formula (8.19) defines the family of MP {Uσ }σ ∈Σ , and ∀ϕ(·) ∈ K στ

∀t ≥ s ≥ τ ϕ(t) ∈ Uσ (t, s, ϕ(s)). (8.20)

Proof Let us check conditions of the Definition 8.2.

(1) Uσ (τ, τ, x) = ϕ(τ ) = x;
(2) ∀ξ ∈ Uσ (t, τ, x) ξ = ϕ(t), where ϕ ∈ K στ , ϕ(τ ) = x. Then for s ∈ [τ, t]
ϕ(s) ∈ Uσ (s, τ, x) and from ϕ|[s,+∞) ∈ K σs we have ϕ(t) ∈ Uσ (t, s, ϕ(s)). So

ξ ∈ Uσ (t, s, Uσ (s, τ, x)).

(3) ∀ξ ∈ Uσ (t + h, τ + h, x) ξ = ϕ(t + h), where ϕ ∈ K στ +h , ϕ(τ + h) = x. Then

ψ(·) = ϕ(· + h) ∈ K Tτ (h)σ , ψ(τ ) = x, so ξ = ψ(t) ∈ UT (h)σ (t, τ, x). Lemma is
proved.
It is easy to show that under conditions a)-c), if ∀s ≥ τ ∀ψ ∈ K στ , ∀ϕ ∈ K σs such
that ψ(s) = ϕ(s), we have

ψ( p), p ∈ [τ, s]
θ ( p) = ∈ K στ , (8.21)
ϕ( p), p > s,

then in the condition (2) of Definition 8.2 equality takes place.

If ∀h ≥ 0 ∀ϕ ∈ K Tτ (h)σ ϕ(· − h) ∈ K στ +h , then in the condition 3) of Definition 8.2
equality takes place.
194 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

From (8.20) we immediately obtain that if for mapping ϕ(·) : R → X for arbitrary
τ ∈ R we have ϕ(·)|[τ,+∞) ∈ K στ , then ϕ(·) is complete trajectory of Uσ .
The next result is generalization on non-autonomous case results from [5].

Lemma 8.2 Let the family of MP {Uσ }σ ∈Σ be constructed by the formula (8.19),
∀ϕ(·) ∈ K στ is continuous on [τ, +∞), the condition (8.21) takes place and the
following one:
if ϕn (·) ∈ K στ , ϕn (τ ) = x, then ∃ϕ(·) ∈ K στ , ϕ(τ ) = x such that on some
subsequence
ϕn (t) → ϕ(t) ∀t ≥ τ.

Then every continuous on [τ, +∞) trajectory of MP Uσ belongs to K στ .

Proof Let ψ : [τ, +∞) → X be continuous trajectory. Let us construct sequence

{ϕn (·)}∞ τ
n=1 ⊂ K σ such that

ϕn (τ + j2−n ) = ψ(τ + j2−n ), j = 0, 1, ..., n2n .

For ϕ1 (·) we have

1 1
ψ(τ + ) ∈ Uσ (τ + , τ, ψ(τ )),
2 2
1 1
ψ(τ + 1) ∈ Uσ (τ + 1, τ + , ψ(τ + ).
2 2
τ + 21
ϕ (·) ∈ K στ , there exists 
So there exists  ϕ (·) ∈ K σ such that

1 1
ψ(τ + ) = 
ϕ (τ + ), 
ϕ (τ ) = ψ(τ ),
2 2
1 1
ψ(τ + 1) = 
ϕ (τ + 1), 
ϕ (τ + ) = ψ(τ + ).
2 2
Therefore due to (8.21) for function

ϕ ( p), τ ≤ p ≤ τ + 21 ,

ϕ1 ( p) =  we have:
ϕ ( p), p > τ + 21

1 1
ϕ1 (·) ∈ K στ , ϕ1 (τ ) = ψ(τ ), ϕ1 (τ + ) = ψ(τ + ), ϕ1 (τ + 1) = ψ(τ + 1).
2 2
Further, using (8.21), we obtain required property for every n ≥ 1. As ϕn (τ ) = ψ(τ ),
so ∃ϕ(·) ∈ K στ , ϕ(τ ) = ψ(τ ) such that on subsequence ∀t ≥ τ ϕn (t) → ϕ(t). As
∀t = τ + j2−n ϕ(t) = ψ(t), so from continuity ϕ(t) = ψ(t) ∀t ≥ τ . Lemma is
proved.
8.2 Structure of Uniform Global Attractor for Non-autonomous … 195

The following theorem declare structure of uniform global attractor in terms of

bounded complete trajectories of corresponding m-processes. It should be noted that
this result is known for single-valued case [6] and in multi-valued case for very
special class of strict processes, generated by strict compact semiprocesses, which
act in Banach spaces [22].

Theorem 8.5 Let Σ is compact, T (h)Σ = Σ ∀ h ≥ 0, the family of MP {Uσ }σ ∈Σ

satisfies (8.19), in condition (3) of Definition 8.2 equality takes place, the mapping
(x, σ ) → Uσ (t, 0, x) has closed graph. Let us assume that there exists ΘΣ - compact
uniform global attractor of the family {Uσ }σ ∈Σ , and one of two conditions hold: either
the family of MP {Uσ }σ ∈Σ is strict, or

f or ever y σn → σ0 , xn → x0 i f ϕn (·) ∈ K σ0n , ϕn (0) = xn ,

so ∃ ϕ(·) ∈ K σ00 , ϕ(0) = x0 such that on subsequnce ∀t ≥ 0 ϕn (t) → ϕ(t).

(8.22)
Then the following structural formula holds

ΘΣ = Kσ (0), (8.23)
σ ∈Σ

where Kσ is the set of all bounded complete trajectories of MP Uσ .

Proof First let us consider situation when the family of MP {Uσ }σ ∈Σ is strict. In this
case one can consider multi-valued semigroup (m-semiflow) on the extended phase
space X × Σ by the rule

G(t, {x, σ }) = {Uσ (t, 0, x), T (t)σ }. (8.24)

Then G is strict, has closed graph and compact attracting set ΘΣ × Σ. So G has
compact invariant global attractor

A = G(t, ΘΣ × Σ) = {γ (0)|γ is bounded complete trajectories of G}.
s≥0 t≥s

Here under complete trajectory of m-semiflow G we mean the mapping R  t →

γ (t) such that
∀ t ∈ R ∀ s ≥ 0 γ (t + s) ∈ G(s, γ (t)).

Let us consider two projectors Π1 and Π2 , Π1 (u, σ ) = u, Π2 (u, σ ) = σ . As

T (t)Σ = Σ, so Π2 A = Σ. Let us prove that Π1 A = ΘΣ .
As ∀ B ∈ β(X ) G(t, B × Σ) → A , t → +∞, so

UΣ (t, τ, B) → Π1 A ,
196 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems


so ΘΣ ⊂ Π1 A . Let us prove that Π1 A = Kσ (0). For this purpose we take
σ ∈Σ
(u 0 , σ0 ) ∈ A . Then there exists γ (·) = {u(·), σ (·)}, which is bounded complete
trajectory of G and such that γ (0) = (u 0 , σ0 ). Then ∀t ≥ τ

u(t) ∈ Uσ (τ ) (t − τ, 0, u(τ )), σ (t) = T (t − τ )σ (τ ).

If τ ≥ 0, then σ (τ ) = T (τ )σ0 , that is

u(t) ∈ UT (τ )σ0 (t − τ, 0, u(τ )) = Uσ0 (t, τ, u(τ )).

If τ < 0, then σ0 = T (−τ )σ (τ ), so

u(t) ∈ Uσ (τ ) (t − τ, τ − τ, u(τ )) = UT (−τ )σ (τ ) (t, τ, u(τ )) = Uσ0 (t, τ, u(τ )).


Therefore u 0 = u(0) ∈ Kσ0 (0) ⊂ Kσ (0).
σ ∈Σ
Now let u 0 = u(0) ∈ K σ0 (0), u(t) ∈ Uσ0 (t, τ, u(τ )) ∀t ≥ τ. As T (t)Σ = Σ,
so there exists σ (s), s ∈ R, such that σ (t) = T (t − τ )σ (τ ), ∀t ≥ τ, σ (0) = σ0 .
Then for s ≥ 0 we have

G(t, {u(s), σ (s)}) = (Uσ (s) (t, 0, u(s)), T (t)σ (s)) =

= (UT (s)σ0 (t, 0, u(s)), σ (t + s)) = (Uσ0 (t + s, s, u(s)), σ (t + s)),

{u(t + s), σ (t + s)} ∈ (Uσ0 (t + s, s, u(s)), σ (t + s)).

If s < 0, then σ0 = T (−s)σ (s), and

u(t + s) ∈ Uσ0 (t + s, s, u(s)) = UT (−s)σ (s) (t + s, s, u(s)) = Uσ (s) (t, 0, u(s)).


Then u 0 ∈ Π1 A and Π1 A = σ ∈Σ Kσ (0).
Since for arbitrary attracting set P and for arbitrary bounded complete trajectory
Γ = {u(s)}s∈R of the process Uσ we have

u(0) ∈ Uσ (0, −n, u(−n)) = UT (n)σ (−n) (0, −n, u(−n)) ⊂

⊂ UΣ (n, 0, Γ ) → P, n → +∞,

so u(0) ∈ P, and we obtain (8.23).

Now let us consider another case, when family of m-processes is not strict, but
the condition (8.22) holds. Let us show that Kσ (0) ⊂ ΘΣ . If z ∈ Kσ (0), then there
exists bounded complete
 trajectory ϕ(·) of m-process Uσ , such that ϕ(0) = z. Let
us denote Γ = ϕ(t) ∈ β(X ). Then for z = ϕ(0) we have
t∈R
8.2 Structure of Uniform Global Attractor for Non-autonomous … 197

ϕ(0) ∈ Uσ (0, −n, ϕ(−n)) = UT (n)σn (0, −n, ϕ(−n)) ⊂ UΣ (n, 0, Γ ).

Since ∀ε > 0 ∃n 0 ∀n ≥ n 0 UΣ (n, 0, Γ ) ⊂ Oε (ΘΣ ), then z ∈ ΘΣ and we obtain

required embedding.
Now let z ∈ ΘΣ = ωΣ (0, B0 ). Then z = lim ξn , ξn ∈ UΣ (tn , 0, B0 ). Therefore
n→+∞
on some subsequence

z = lim ϕn (tn ), ϕn (·) ∈ K σ0n , ϕn (0) ∈ B0 , σn → σ.

n→+∞

For ∀n ≥ 1 let us consider

ψn (·) := ϕn (· + tn ) ∈ K T−t(tnn )σn ,

that is ψn (·) ∈ K σ−t

˜n , where σ˜n = T (tn )σn . Then ψn (·) ∈ K σ˜n , σ˜n → σ̃ , ψn (0) =
n 0

ϕn (tn ) → z, so there exists ψ (0) (·) ∈ K σ̃0 , ψ (0) (0) = z, such that

∀t ≥ 0 ψn (t) = ϕn (t + tn ) → ψ (0) (t).

For τ = −1 ∀n ≥ n 1 − tn < −1, therefore ψn (·) ∈ K σ−1

˜n and on some subsequence

ψn (−1) = ϕn (tn − 1) → z 1 .

Herewith there exists ψ (−1) (·) ∈ K σ̃−1 such that on subsequence

ψn (t) = ϕn (t + tn ) → ψ (−1) (t) ∀t ≥ −1,

and ∀t ≥ 0 ψ (0) (t) = ψ (−1) (t). By standard diagonal procedure we construct

sequence of functions
ψ (−k) (·) ∈ K σ̃−k , k ≥ 0,

with ψ (−k+1) (t) = ψ (−k) (t) ∀t ≥ −k + 1. Let us put

ψ(t) := ψ (−k) (t), i f t ≥ −k.

Then the function ψ(·) is correctly defined, ψ : R → X.

Moreover ∀τ < 0 ∃k such that [τ, +∞) ⊂ [−k, +∞), on [−k, +∞) ψ(·) ≡
ψ (−k) , so ψ(·) ∈ K σ̃−k , and from this

ψ(·) ∈ K σ̃τ , ψ(0) = ψ (0) (0) = z.

Since on subsequence

∀t ∈ R ψ(t) = lim ϕn (t + tn ) ∈ ωΣ (0, B0 ) ∈ β(X ),

n→+∞
198 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

then z = ψ(0) ∈ Kσ̃ and theorem is proved.

Definition 8.7 Let Θ be some topological space of functions from R to topological

space E. The function ξ ∈ Θ is called translation compact in Θ, if the set

H (ξ ) = clΘ {ξ(· + s) | s ∈ R}

is compact in Θ.

To construct family of m-processes for the problem (8.8) we suppose that time-
depended functions f and h are translation compact in natural spaces [6]. More
precisely, we will assume that
2,w
h is translation compact in L loc (R; H ), (8.25)

2,w
where L loc (R; H ) is the space L loc
2
(R; H ) with the local weak convergence topology,
and
f is translation compact in C(R; C(R N , R N )), (8.26)

where C(R; C(R N , R N )) equipped with local uniform convergence topology.

It is known that condition (8.25) is equivalent to

t+1
|h|2+ := sup |h(s)|2 ds < ∞ (8.27)
t∈R
t

It is also known that condition (8.26) is equivalent to

∀ R > 0 f is bounded and uni f or mly continuous on

(8.28)
Q(R) = {(t, v) ∈ R × R N | |v|R N ≤ R}.

If conditions (8.25), (8.26) take place, then the symbol space

Σ = clC(R;C(R N ,R N ))×L loc

2,w
(R;H ) {( f (· + s), h(· + s)) | s ∈ R} (8.29)

is compact, and ∀ s ≥ 0 T (s)Σ = Σ, where T (s) is translation semigroup, which

is continuous on Σ.
For every σ = ( f σ , h σ ) ∈ Σ we consider the problem

u t = aΔu − f σ (t, u) + h σ (t, x), x ∈ Ω, t > τ,
(8.30)
u|∂Ω = 0.

It is proved in [9] that ∀ σ ∈ Σ f σ satisfies (8.10), (8.11) with the same constants C1 ,
C2 , γi , |h σ |+ ≤ |h|+ . So we can apply Theorem 2 and obtain that ∀ τ ∈ R, u τ ∈ H
the problem (8.30) has at least one solution on (τ, +∞), each solution of (8.30)
8.2 Structure of Uniform Global Attractor for Non-autonomous … 199

belongs to C([τ, +∞); H ) and satisfies energy equality (8.13). For every σ ∈ Σ,
τ ∈ R we define

K στ = {u(·) | u(·) is solution o f (8.30) on (τ, +∞)} (8.31)

and according to (8.19) we put ∀ σ ∈ Σ, ∀ t ≥ τ , ∀ u τ ∈ H

Uσ (t, τ, u τ ) = {u(t)|u(·) ∈ K στ , u(τ ) = u τ }. (8.32)

From [9] and Theorem 8.5 we obtain the following result. The following theorem
was proved in [11].

Theorem 8.6 Under conditions (8.10), (8.11), (8.25), (8.26) formula (8.32) defines
a strict family of MP {Uσ }σ ∈Σ which has compact, invariant, stable and connected
uniform global attractor ΘΣ , which consists of bounded complete trajectories, that
is 
ΘΣ = Kσ (0), (8.33)
σ ∈Σ

where Kσ is the set of all bounded complete trajectories of MP Uσ .

Now we want to use formula (8.33) for proving that the uniform global attractor
of RD-system is bounded set in the space (L ∞ (Ω)) N ∩ V .
First let us consider the following conditions:

∃ Mi > 0, i = 1, N such that f or all v = (v1 , ..., v N ) ∈ R N f or a.a. x ∈ Ω ∀ t ∈ R


N
( f i (t, v) − h i (t, x))(vi − Mi )+ ≥ 0 (8.34)
i=1


N
( f i (t, v) − h i (t, x))(vi + Mi )− ≤ 0 (8.35)
i=1

where ϕ + = max{0, ϕ}, ϕ − = max{0, −ϕ}, ϕ = ϕ + − ϕ − .

Let us consider some example, which allow to verify conditions (8.34), (8.35).

Lemma 8.3 If N = 1 (scalar equation), then from (8.10), (8.11) and h ∈ L ∞ (R ×

Ω) we have (8.34), (8.35).

Proof From (8.10) and h ∈ L ∞ (Ω) for a.a. x ∈ Ω and u ∈ R,

γ̃ |u| p − C̃2 ≤ g(t, x, u)u ≤ C̃1 |u| p + C̃1 ,

where g(t, x, u) = f (t, u) − h(t, x), γ̃ does not depend on t, u, x.

200 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

If u ≤ M, then g(t, x, u)(u − M)+ = 0.

If u > M, then

(u − M)+ M
g(t, x, u)(u − M)+ = g(t, x, u)u = g(t, x, u)u(1 − )
u u
M M
≥ (γ̃ u p − C̃2 )(1 − ) ≥ (γ̃ M p − C̃2 )(1 − )
u u
  1p
and if we choose M = C̃2
γ̃
, then g(t, x, u)(u − M)+ ≥ 0 a.e.

Lemma 8.4 If for arbitrary N ≥ 1 h ≡ 0, f (t, u) = ( f 1 (t, u), ... f N (t, u)), where

N
f i (t, u) = ( |u i |2 − R 2 )u i , R > 0 is positive constant, then conditions (8.34),
i=1
(8.35) hold for Mi = R.

N
Proof If |u i |2 < R 2 , so ∀ i = 1, N |u i | < R and
i=1


N
f i (t, u)(u i − R)+ = 0,
i=1


N
f i (t, u)(u i + R)− = 0.
i=1


N
If |u i |2 ≥ R 2 , then
i=1

 N 

N  
N
+
f (t, u)(u − R) =
i i
|u | − R
i 2 2
u i (u i − R)+ ≥ 0,
i=1 i=1 i=1

 N 

N  
N
−
f (t, u)(u + R) =
i i
|u | − R
i 2 2
u i (u i + R)− ≤ 0.
i=1 i=1 i=1

Theorem 8.7 If conditions (8.10), (8.11), (8.25), (8.26), (8.34), (8.35) hold and
matrix a is diagonal, then the uniform global attractor ΘΣ is bounded set in the
space (L ∞ (Ω)) N ∩ V .
Proof First let us prove that ∀ σ ∈ Σ functions f σ , h σ satisfy (8.34), (8.35). Indeed,
there exists sequence tn  ∞ such that ∀ T > 0, R > 0, η ∈ L 2 ((−T, T ) × Ω)


N
sup sup | f i (t + tn , v) − f σi (t, v)|2 → 0, n → ∞,
|t|≤T |v|≤R i=1
8.2 Structure of Uniform Global Attractor for Non-autonomous … 201

N T 

(h i (t + tn , x) − h iσ (t, x))η(t, x)d xdt → 0, n → ∞.
i=1 −T Ω

From (8.34)

N
( f i (t + tn , v) − h i (t + tn , x))(vi − Mi )+ ≥ 0. (8.36)
i=1

Therefore for fixed v and for arbitrary ε > 0 there exists N ≥ 1 such that ∀ n ≥ N


N 
N 
N
h i (t + tn , x)(vi − Mi )+ ≤ f i (t + tn , v)(vi − Mi )+ < f σi (t, v)(vi − Mi )+ + ε.
i=1 i=1 i=1

Because


N 
N
+
h (t +tn , x)(v − Mi ) →
i i
h iσ (t, x)(vi − Mi )+ weakly in L 2 ((−T, T )×Ω),
i=1 i=1

from Mazur Theorem we deduce that


N 
N
+
h iσ (t, x)(vi − Mi ) ≤ f σi (t, v)(vi − Mi )+ + ε f or a.a. x ∈ Ω.
i=1 i=1

From arbitrary choice of ε we can obtain required result.

It is easy to obtain that for arbitrary weak solution of (8.8) and for every η ∈
C0∞ (τ, T )
T T
+ 1
(u t , u )ηdt = − |u + |2 ηt dt. (8.37)
2
τ τ

Then putting gσ = f σ − h σ and for numbers M1 , ..., M N from condition (8.34) we

have
 
1 d  i 
N N N
+ 2 + 2
|(u − Mi ) | +β (u − Mi )
i
+ gσi (t, x, u)(u i − Mi )+ d x = 0.
2 dt i=1 i=1 i=1 Ω

Then from (8.34)

d  i 
N N
|(u − Mi )+ |2 + 2β |(u i − Mi )+ |2 ≤ 0
dt i=1 i=1

and for all t > τ

202 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems


N 
N
|(u i − Mi )+ (t)|2 ≤ |(u i − Mi )+ (τ )|2 e−2λ1 β(t−τ ) . (8.38)
i=1 i=1

If u(·) ∈ Kσ then from (8.38) taking τ → −∞ we obtain u i (x, t) ≤ Mi , i =

1, N , ∀t ∈ R, for a.a. x ∈ Ω.
In the same way we will have u i (x, t) ≥ Mi (using (u i + Mi )− ).
Then
ess sup |z i (x)| ≤ Mi ∀z = (z 1 , ..., z N ) ∈ ΘΣ .
x∈Ω

So we obtain that ΘΣ is bounded set in the space (L ∞ (Ω)) N . From the equality
ΘΣ = UΣ (t, τ, ΘΣ ) ∀ t ≥ τ we deduce that ∀ σ ∈ Σ Uσ (t, τ, ΘΣ ) ⊂ ΘΣ . Now
let us consider arbitrary complete trajectory u(·) ∈ Kσ . Due to definition of weak
solution for a.a. t ∈ R u(t) ∈ V . We take such τ ∈ R that u(τ ) ∈ V and consider
the following Cauchy problem
⎧
⎨ vt = aΔv − f σ (t, u) + h σ (t, x), x ∈ Ω, t > τ,
v|∂Ω = 0, (8.39)
⎩
v|t=τ = u(τ ).

Because ∀ t ≥ τ u(t) ∈ ΘΣ , which is bounded in (L ∞ (Ω)) N , we have that

f σ (t, u(t, x)) ∈ (L ∞ (Ω)) N . Thus for linear problem (8.39) from well-known results
one can deduce that ∀ T > τ v ∈ C([τ, T ]; V ). So from uniqueness of the solution
of Cauchy problem (8.39) v ≡ u on [τ, +∞) and, therefore, ∀ t ≥ τ u(t) ∈ V . It
means that ∀ t ∈ R u(t) ∈ V and from the formula (8.33) ΘΣ ⊂ V .
From the energy equality, applying to function u, and boundness of ΘΣ in the
space H we deduce, that ∃ C > 0, which does not depend on σ , such that ∀ t ∈ R

t+1 t+1
u(s) ds ≤ C(1 +
2
|h σ (s)|2 ds).
t t

From translation compactness of h we have

t+1
u(s) 2 ds ≤ C(1 + |h|2+ ).
t

So for arbitrary t ∈ R we find τ ∈ [t, t + 1] such that u(τ ) 2

≤ C(1 + |h|2+ ). Then
for the problem (8.39) we obtain inequality

∀t ≥τ v(t) 2
≤ e−δ(t−τ ) u(τ ) 2
+ D,
8.2 Structure of Uniform Global Attractor for Non-autonomous … 203

where positive constants δ, D do not depend on σ . Thus

∀ t ∈ R u(t) 2
≤ C(1 + |h|2+ ) + D

and theorem is proved.

8.3 Uniform Trajectory Attractors for Nonautonomous

Dissipative Dynamical Systems

For evolution triple (Vi ; H ; Vi∗ )1 and multi-valued map Ai : R+ × V ⇒ V ∗ , i =

1, 2, . . . , N , N = 1, 2, . . . we consider a problem of long-time behavior of all
globally defined weak solutions for nonautonomous evolution inclusion


N
y  (t) + Ai (t, y(t))  0̄, (8.40)
i=1

as t → +∞. Let ·, ·Vi : Vi ∗ × Vi → R be the pairing in Vi ∗ × Vi , that coincides

on H × Vi with the inner product (·, ·) in the Hilbert space H .
Note that Problem (8.40) arises in many important models for distributed para-
meter control problems and that large class of identification problems enter this
formulation.
Throughout this subsection we suppose that the listed below assumptions hold:
Assumption I Let pi ≥ 2, qi > 1 are such that p1i + q1i = 1, for each for
i = 1, 2, . . . , N , and the embedding Vi ⊂ H is compact one, for some for i =
1, 2, . . . , N .
Assumption II (Grows Condition) There exist a t.u.i. in L loc 1 (R+ ) function c1 :
R+ → R+ and a constant c2 > 0 such that

N q

N
p
max di Vi ∗ ≤ c1 (t) + c2 u Vi
i=1
i=1

for any u ∈ Vi , di ∈ Ai (t, u), i = 1, 2, . . . , N , and a.e. t > 0.

Assumption III (Signed Assumption) There exist a constant α > 0 and a t.u.i. in
1 (R+ ) function β : R+ → R+ such that
L loc


N 
N
p
di , uVi ≥ α u Vi − β(t)
i=1 i=1

1 i.e.Vi is a real reflexive separable Banach space continuously and densely embedded into a real
Hilbert space H , H is identified with its topologically conjugated space H ∗ , Vi∗ is a dual space to
Vi . So, there is a chain of continuous and dense embeddings: Vi ⊂ H ≡ H ∗ ⊂ Vi∗ .
204 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

for any u ∈ Vi , di ∈ Ai (t, u), i = 1, 2, . . . , N , and a.e. t > 0.

Assumption IV (Strong Measurability) If C ⊆ Vi ∗ is a closed set, then the set
{(t, u) ∈ (0, +∞) × Vi : Ai (t, u) ∩ C = ∅} is a Borel subset in (0, +∞) × Vi .
Assumption V (Pointwise Pseudomonotonicity) Let for each i = 1, 2, . . . , N
and a.e. t > 0 two assumptions hold:
(a) for every u ∈ Vi the set Ai (t, u) is nonempty, convex, and weakly compact one
in Vi ∗ ;
(b) if a sequence {u n }n≥1 converges weakly in Vi towards u ∈ Vi as n → +∞,
dn ∈ Ai (t, u n ) for any n ≥ 1, and lim supdn , u n − uVi ≤ 0, then for any
n→+∞
ω ∈ Vi there exists d(ω) ∈ Ai (t, u) such that

lim inf dn , u n − ωVi ≥ d(ω), u − ωVi .

n→+∞

Let 0 ≤ τ < T < +∞. As a weak solution of evolution inclusion (8.40) on the
interval [τ, T ] we consider an element u(·) of the space ∩i=1 N
L pi (τ, T ; Vi ) such that
∗
for some di (·) ∈ L qi (τ, T ; Vi ), i = 1, 2, . . . , N , it is fulfilled:

T N 

T

− (ξ (t), y(t))dt + di (t), ξ(t)Vi dt = 0 ∀ξ ∈ C0∞ ([τ, T ]; Vi ), (8.41)
τ i=1 τ

and di (t) ∈ Ai (t, y(t)) for each i = 1, 2, . . . , N and a.e. t ∈ (τ, T ).

For fixed nonnegative τ and T , τ < T , let us consider


N
∗
X τ,T = ∩i=1
N
L pi (τ, T ; Vi ), X τ,T = L qi (τ, T ; Vi ∗ ), Wτ,T = {y ∈ X τ,T | y  ∈ X τ,T
∗
},
i=1
∗ ∗
Aτ,T : X τ,T ⇒ X τ,T , Aτ,T (y) = {d ∈ X τ,T | d(t) ∈ A(t, y(t)) for a.e. t ∈ (τ, T )},

N
where y  is a derivative of an element u ∈ X τ,T in the sense of D([τ, T ]; i=1 Vi ∗ )
(see, for example, Sect. 6.1). Note that the space Wτ,T is a reflexive Banach space with
the graph norm of a derivative u Wτ,T = u X τ,T + u  X τ,T ∗ ,u ∈ W
τ,T . Let ·, · X τ,T :
∗ ∗
X τ,T × X τ,T → R be the pairing in X τ,T × X τ,T , that coincides on L 2 (τ, T ; H ) ×
T
X τ,T with the inner product in L 2 (τ, T ; H ), i.e. u, v X τ,T = (u(t), v(t))dt for any
τ
u ∈ L 2 (τ, T ; H ) and v ∈ X τ,T . The embedding Wτ,T ⊂ C([τ, T ]; H ) is continuous
and dense one. Moreover,

T  
(u(T ), v(T )) − (u(τ ), v(τ )) = u  (t), v(t)Vi + v (t), u(t)Vi dt, (8.42)
τ

for any u, v ∈ Wτ,T .

8.3 Uniform Trajectory Attractors for Nonautonomous Dissipative … 205

For fixed τ and T , such that 0 ≤ τ < T < +∞, we denote

Dτ,T (y (τ ) ) = {y(·) | y is a weak solution of (8.40) on [τ, T ], y(τ ) = y (τ ) }, y (τ ) ∈ H.

We remark that Dτ,T (y (τ ) ) = ∅ and Dτ,T (y (τ ) ) ⊂ Wτ,T , if 0 ≤ τ < T < +∞

and y (τ ) ∈ H . Moreover, the concatenation of Problem (8.40) weak solutions is
a weak solutions too, i.e. if 0 ≤ τ < t < T , y (τ ) ∈ H , y(·) ∈ Dτ,t (y (τ ) ), and
v(·) ∈ Dt,T (y(t)), then 
y(s), s ∈ [τ, t],
z(s) =
v(s), s ∈ [t, T ],

belongs to Dτ,T (y (τ ) ); see Sect. 6.1.

Gronwall lemma provides that for any finite time interval [τ, T ] ⊂ R+ each weak
solution y of Problem (8.40) on [τ, T ] satisfies estimates
 t N 
 t  s
p
y(t) 2
H −2 β(ξ )dξ + 2α y(ξ ) Vi dξ ≤ y(s) 2
H −2 β(ξ )dξ,
0 i=1 s 0

 (8.43)
t
2 −2αγ (t−s) −2αγ (t−ξ )
y(t) 2
H ≤ y(s) He +2 (β(ξ ) + αγ )e dξ, (8.44)
s

where t, s ∈ [τ, T ], t ≥ s; γ is a constant that does not depend on y, s, and t; cf.

p
Sect. 6.1. In the proof of (8.44) we used the inequality u 2H − 1 ≤ u H for any
u ∈ H.
Therefore, any weak solution y of Problem (8.40) on a finite time interval [τ, T ] ⊂
R+ can be extended to a global one, defined on [τ, +∞). For arbitrary τ ≥ 0 and
y (τ ) ∈ H let Dτ (y (τ ) ) be the set of all weak solutions (defined on [τ, +∞)) of
Problem (8.40) with initial data y(τ ) = y (τ ) . Let us consider the family Kτ+ =
∪ y (τ ) ∈H Dτ (y (τ ) ) of all weak solutions of Problem (8.40) defined on the semi-infinite
time interval [τ, +∞). Consider the Fréchet space C loc (R+ ; H ). We remark that the
sequence { f n }n≥1 converges in C loc (R+ ; H ) towards f ∈ C loc (R+ ; H ) as n → +∞
iff the sequence {Πt1 ,t2 f n }n≥1 converges in C([t1 , t2 ]; H ) towards Πt1 ,t2 f as n → +∞
for any finite interval [t1 , t2 ] ⊂ R+ , where Πt1 ,t2 is the restriction operator to the
interval [t1 , t2 ]; Chepyzhov and Vishik [6, p. 918]. We denote T (h)y(·) = yh (·),
where yh (t) = y(t + h) for any y ∈ C loc (R+ ; H ) and t, h ≥ 0.
In the autonomous case, when A(t, y) does not depend on t, the long-time behav-
ior of all globally defined weak solutions for Problem (8.40) is described by using
trajectory and global attractors theory. In this situation the set K + := K0+ is trans-
lation invariant, i.e. T (h)K + ⊆ K + for any h ≥ 0. As trajectory attractor it is
considered a classical global attractor for translation semigroup {T (h)}h≥0 , that acts
on K + .
In the nonautonomous case we notice that T (h)K0+  K0+ . Therefore, we
need to consider united trajectory space that includes all globally defined on any
[τ, +∞) ⊆ R+ weak solutions of Problem (8.40) shifted to τ = 0:
206 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

 
+
K = clC loc (R+ ;H ) y( · + τ ) : y ∈ Kτ+ ,
τ ≥0

where clC loc (R+ ;H ) [ · ] is the closure in C loc (R+ ; H ). Note that T (h){y( · + τ ) : y ∈
Kτ+ } ⊆ {y( · + τ + h) : y ∈ Kτ++h } for any τ, h ≥ 0. Moreover,
+ +
T (h)K ⊆K for any h ≥ 0,

because

ρC loc (R+ ;H ) (T (h)u, T (h)v) ≤ ρC loc (R+ ;H ) (u, v) for any u, v ∈ C loc (R+ ; H ),

where ρC loc (R+ ;H ) is a standard metric on Fréchet space C loc (R+ ; H ).

A set P ⊂ C loc (R+ ; H ) ∩ L ∞ (R+ ; H ) is said to be a uniformly attracting
set (cf. Chepyzhov and Vishik [6, p. 921]) for the united trajectory space K + of
Problem (8.40) in the topology of C loc (R+ ; H ), if for any bounded in L ∞ (R+ ; H )
set B ⊆ K + and any segment [t1 , t2 ] ⊂ R+ the following relation holds:

distC([t1 ,t2 ];H ) (Πt1 ,t2 T (t)B, Πt1 ,t2 P) → 0, t → +∞, (8.45)

where distC([t1 ,t2 ];H ) is the Hausdorff semi-metric.

A set U ⊂ K + is said to be a uniform trajectory attractor (cf. Chepyzhov and
Vishik [6, p. 921]) of the translation semigroup {T (t)}t≥0 on K + in the induced
topology from C loc (R+ ; H ), if
(i) U is a compact set in C loc (R+ ; H ) and bounded in L ∞ (R+ ; H );
(ii) U is strictly invariant with respect to {T (h)}h≥0 , i.e. T (h)U = U ∀h ≥ 0;
(iii) U is a minimal uniformly attracting set for K + in the topology of C loc (R+ ; H ),
i.e. U belongs to any compact uniformly attracting set P of K + : U ⊆ P.
Note that uniform trajectory attractor of the translation semigroup {T (t)}t≥0 on
K + in the induced topology from C loc (R+ ; H ) coincides with the classical trajectory
attractor for the continuous semi-group {T (t)}t≥0 defined on K + (see, for example,
Chepyzhov and Vishik [6, Definition 1.1]).
Presented construction is coordinated with the theory of uniform trajectory attrac-
tors for nonautonomous problems of the form

∂t u(t) = Aσ (t) (u(t)), (8.46)

where σ (s), s ≥ 0, is a functional parameter called the time symbol of Eq. (8.46)
(t is replaced by s). In applications to mathematical physics equations, a function
σ (s) consists of all time-dependent terms of the equation under consideration: exter-
nal forces, parameters of mediums, interaction functions, control functions, etc. In
mentioned above papers and books it is assumed that the symbol σ of Eq. (8.46)
belongs to a Hausdorff topological space Ξ+ of functions defined on R+ with val-
8.3 Uniform Trajectory Attractors for Nonautonomous Dissipative … 207

ues in some complete metric space. Usually, in applications, the topology in the
space Ξ+ is a local convergence topology on any segment [t1 , t2 ] ⊂ R+ . Further,
they consider the family of Eq. (8.46) with various symbols σ (s) belonging to a set
Σ ⊆ Ξ+ . The set Σ is called the symbol space of the family of Eq. (8.46). It is
assumed that the set Σ, together with any symbol σ (s) ∈ Σ, contains all positive
translations of σ (s): σ (t + s) = T (t)σ (s) ∈ Σ for any t, s ≥ 0. The symbol space
Σ is invariant with respect to the translation semigroup {T (t)}t≥0 : T (t)Σ ⊆ Σ
for any t ≥ 0. To prove the existence of uniform trajectory attractor they suppose
that the symbol space Σ with the topology induced from Ξ+ is a compact metric
space. Mostly in applications, as a symbol space Σ it is naturally to consider the
hull of translation-compact function σ0 (s) in an appropriate Hausdorff topological
space Ξ+ . The direct realization of this approach for Problem (8.40) is problematic
without any additional assumptions for parameters of Problem (8.40) and requires
the translation-compactness of the symbol σ (s) = A(s, ·) in some compact Haus-
dorff topological space of measurable multi-valued mappings acts from R+ to some
∗
metric space of pseudomonotone operators from (Vi → 2Vi ) satisfying grows and
signed assumptions. To avoid this technical difficulties we present the alternative
approach for the existence and construction of the uniform trajectory attractor for all
weak solutions for Problem (8.40). Note that Assumptions (I)–(V) are natural and
guaranty, in the general case, only existence of weak solution for Cauchy problem
on any finite time interval [τ, T ] ⊂ R+ and for any initial data form H .
The main result of this section has the following form.

Theorem 8.8 Let Assumptions (I)–(V) hold. Then there exists an uniform trajectory
attractor U ⊂ K + of the translation semigroup {T (t)}t≥0 on K + in the induced
topology from C loc (R+ ; H ). Moreover, there exists a compact in C loc (R+ ; H ) uni-
formly attracting set P ⊂ C loc (R+ ; H ) ∩ L ∞ (R+ ; H ) for the united trajectory
space K + of Problem (8.40) in the topology of C loc (R+ ; H ) such that U coincides
with ω-limit set of P:


U = clC loc (R+ ;H ) T (h)P . (8.47)
t≥0 h≥t

Before the proof of Theorem 8.8 we provide some auxiliary constructions (see
Sect. 6.1).
Assumptions (II) and (III) yield that there exist a positive constant α  > 0 and a
t.u.i. function c in L loc
1 (R+ ) such that A(t, u) ⊆ Ac (t) (u) for each u ∈ ∩i=1 Vi and
N

a.e. t > 0, where

 

N 
N
N
 
pi : pi ∈ Vi ∗ ,  pi , uVi ≥ α  max − c (t) .
p q
Ac (t) (u) := u Vi ; p Vi ∗
i=1
i=1 i=1

Let H (c ) be the hull of t.u.i. function c in L loc 

1,w (R+ ), i.e. H (c ) = cl L loc
1 (R+ )
{c ( · +
h) : h ≥ 0}. This is a weakly compact set in L loc 1 (R+ ).
208 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

Let us consider the family of problems

y  = Aσ (y), σ ∈ Σ := H (c ). (8.48)

To each σ ∈ Σ there corresponds a space of all globally defined on [0, +∞) weak
solutions Kσ+ ⊂ C loc (R+ ; H ) of Problem (8.48). We set KΣ+ = ∪σ ∈Σ Kσ+ .
We remark that (see Sect. 6.1) any element from KΣ+ satisfies prior estimates.

Lemma 8.5 There exist positive constants c3 and c4 such that for any σ ∈ Σ and
y ∈ Kσ+ the inequalities hold:

t N 

t s
 p
H −2 σ (ξ )dξ +2α y(ξ ) ≤ y(s) H −2 σ (ξ )dξ, (8.49)
2 2
y(t) Vi dξ
0 i=1 s 0
t
2 −c3 (t−s)
y(t) 2
H ≤ y(s) He + c4 σ (ξ )e−c3 (t−ξ ) dξ, (8.50)
s

for any t ≥ s ≥ 0.

Moreover, the following result characterizing the compactness properties of solu-

tions for the family of Problems (8.48) holds:

Theorem 8.9 Let {yn }n≥1 ⊂ KΣ+ be an arbitrary sequence, that is bounded in
L ∞ (R+ ; H ). Then there exist a subsequence {yn k }k≥1 ⊂ {yn }n≥1 and an element
y ∈ KΣ+ such that

max yn k (t) − y(t) H → 0, k → +∞, (8.51)

t∈[τ,T ]

for any finite time interval [τ, T ] ⊂ (0, +∞).

Proof of Theorem 8.8 First, let us show that there exists a uniform trajectory
attractor U ⊂ K + of the translation semigroup {T (t)}t≥0 on K + in the induced
topology from C loc (R+ ; H ). Lemma 8.5 and Theorem 8.9 yields that the translation
semigroup {T (t)}t≥0 has a compact absorbing (and, therefore, an uniformly attract-
ing) set in the space of trajectories KΣ+ . This set can be constructed as follows: 1)
consider P, the intersection of KΣ+ with a ball in the space of bounded continuous
functions on R+ with values in H , Cb (R+ ; H ), of sufficiently large radius; 2) shift
the resulting set by any fixed distance h > 0. Thus, we obtain T (h)P, a set with
the required properties. Recall that the semigroup {T (t)}t≥0 is continuous. There-
fore, the set P1 := P ∩ K + is a compact absorbing (and, therefore, an uniformly
attracting) in the space K + with the induced topology of C loc (R+ ; H ). In fact, here
one can apply the classical theorem on the global attractor of a (unique) continuous
semigroup in a complete metric space, the semigroup in question having a compact
attracting.
8.4 Notes on Applications 209

8.4 Notes on Applications

As applications we may consider FitzHugh–Nagumo system (signal transmission

across axons), complex Ginzburg–Landau equation (theory of superconductivity),
Lotka–Volterra system with diffusion (ecology models), Belousov–Zhabotinsky sys-
tem (chemical dynamics) and many other nonlinear systems (see Sects. 2.4 and 4.6).

References

1. Anguiano, M., Caraballo, T., Real, J., Valero, J.: Pullback attractors for reaction-diffusion
equations in some unbounded domains with an H −1 -valued non-autonomous forcing term and
without uniqueness of solutions. Discret. Contin. Dyn. Syst. B 14, 307–326 (2010)
2. Babin, A.V., Vishik, M.I.: Attractors of Evolution Equations. Nauka, Moscow (1989)
3. Ball, J.M.: Global attractors for damped semilinear wave equations. Discret. Contin. Dyn. Syst.
10, 31–52 (2004)
4. Brunovsky, P., Fiedler, B.: Connecting orbits in scalar reaction diffusion equations. Dyn. Rep.
1, 57–89 (1988)
5. Caraballo, T., Marin-Rubio, P., Robinson, J.: A comparison between two theories for multi-
valued semiflows and their asymptotic behavior. Set-valued anal. 11, 297–322 (2003)
6. Chepyzhov, V.V., Vishik, M.I.: Attractors for Equations of Mathematical Physics. American
Mathematical Society, Providence (2002)
7. Hale, J.K.: Asymptotic Behavior of Dissipative Systems. American Mathematical Society,
Providence (1988)
8. Kapustyan, O.V., Valero, J.: On the connectedness and asymptotic behaviour of solutions of
reaction-diffusion systems. J. Math. Anal. Appl. 323, 614–633 (2006)
9. Kapustyan, O.V., Valero, J.: On the Kneser property for the complex Ginzburg–Landau equation
and the Lotka–Volterra system with diffusion. J. Math. Anal. Appl. 357, 254–272 (2009)
10. Kapustyan, O.V., Valero, J.: Comparison between trajectory and global attractors for evolution
systems without uniqueness of solutions. Int. J. Bifurc. Chaos 20, 2723–2734 (2010)
11. Kapustyan, O.V., Kasyanov, P.O., Valero, J., Zgurovsky, M.Z.: Structure of uniform global
attractor for general non-autonomous reaction-diffusion system. Continuous and Distributed
Systems: Theory and Applications. Solid Mechanics and Its Applications. 211, 163–180 (2014).
doi:10.1007/978-3-319-03146-0_12
12. Kapustyan, O.V., Melnik, V.S., Valero, J., Yasinsky, V.V.: Global Attractors of Multi-valued
Dynamical Systems and Evolution Equations Without Uniqueness. Naukova Dumka, Kyiv
(2008)
13. Lions, J.L.: Quelques Méthodes de Résolution des Problèmes aux Limites non Linéaires.
Gauthier-Villar, Paris (1969)
14. Melnik, V.S., Valero, J.: On attractors of multi-valued semi-flows and differential inclusions.
Set-valued Anal. 6, 83–111 (1998)
15. Melnik, V.S., Valero, J.: On global attractors of multi-valued semiprocesses and nonautonomous
evolution inclusions. Set-valued Anal. 8, 375–403 (2000)
16. Rocha, C.: Examples of attractors in scalar reaction-diffusion equations. J. Differ. Equ. 73,
178–195 (1988)
17. Rocha, C.: Properties of the attractor of a scalar parabolic PDE. J. Dyn. Differ. Equ. 3, 575–591
(1991)
18. Rocha, C., Fiedler, B.: Heteroclinic orbits of semilinear parabolic equations. J. Differ. Equ.
125, 239–281 (1996)
19. Sell, G.R., You, Y.: Dynamics of Evolutionary Equations. Springer, Berlin (2002)
210 8 Uniform Trajectory Attractors for Non-autonomous Nonlinear Systems

20. Temam, R.: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Springer,
New York (1997)
21. Vishik, M.I., Zelik, S.V., Chepyzhov, V.V.: Strong trajectory attractor of dissipative reaction-
diffusion system. Doklady RAN 435(2), 155–159 (2010)
22. Wang, Y., Zhou, S.: Kernel sections and uniform attactors of multi-valued semiprocesses. J.
Differ. Equ. 232, 573–622 (2007)
23. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
Inclusions and Variation Inequalities for Earth Data Processing III. Long-time Behavior of
Evolution Inclusions Solutions in Earth Data Analysis. Springer, Berlin (2012)
Chapter 9
Indirect Lyapunov Method for Autonomous
Dynamical Systems

Abstract In this chapter we establish indirect Lyapunov method for autonomous

dynamical systems. Section 9.1 devoted to the first order autonomous differential-
operator equations and inclusions. In Sect. 9.2 we consider the second order
autonomous operator differential equations and inclusions. In Sect. 9.3 we exam-
ine examples of applications. In particular, a model of combustion in porous media;
a model of conduction of electrical impulses in nerve axons; viscoelastic problems
with nonlinear “reaction-displacement” law etc.

9.1 First Order Autonomous Differential-Operator

Equations and Inclusions

Let (M , g) be a C ∞ compact connected oriented two-dimensional Riemannian man-

ifold without boundary (as, e.g. M = S 2 the unit sphere of R3 ). Consider the Budyko
model:
∂u
∂t
− Δu + Bu ∈ Q S(x)β(u), (x, t) ∈ M × R, (9.1)

where Δu = divM (∇M u) ; ∇M is understood in the sense of the Riemannian metric

g (see Sect. 2.4.3).
Let S : M → R be a function such that S ∈ L ∞ (M ) and there exist S0 , S1 > 0
such that
0 < S0 ≤ S(x) ≤ S1 .

Suppose also that β is a bounded maximal monotone graph of R2 , that is there exist
m, M ∈ R such that for all s ∈ R and z ∈ β(s)

m ≤ z ≤ M.

Let us consider real Hilbert spaces

H := L 2 (M ), V := {u ∈ L 2 (M ) : ∇M u ∈ L 2 (T M )}

© Springer International Publishing AG 2018 211

M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6_9
212 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

with respective standard norms  ·  H ,  · V , and inner products ( · , · ) H , ( · , · )V ,

where T M represents the tangent bundle and the functional spaces L 2 (M ) and
L 2 (T M ) are defined in a standard way. According to Theorem 2.2, for any −∞ <
τ < T < +∞ each weak solution u τ ∈ L 2 (Ω) of Problem (9.1) on [τ, T ] belongs to
C([τ + ε, T ]; H01 (Ω)) ∩ L 2 (τ + ε, T ; H 2 (Ω) ∩ H01 ((0, π ))) and dudt
(·) ∈ L 2 (τ +
ε, T ; L (Ω)) for each ε ∈ (0, T − τ ).
2

Consider the generalized setting of Problem (9.1):

du
+ Au(t) + ∂ J1 (u(t)) − ∂ J2 (u(t)) 0̄ on (−∞ < τ < T < +∞), (9.2)
dt

where A : V → V be a linear symmetric operator such that ∃c > 0 : Av, v V ≥

cv2V , for each v ∈ V and Ji : H → R be a convex, lower semi-continuous function
such, that the following assumptions hold: (i) (growth condition) there exists c1 > 0
such that y H ≤ c1 (1 + u H ), for each u ∈ H and y ∈ ∂ Ji (u) and i = 1, 2; (ii)
(sign condition) there exist c2 > 0, λ ∈ (0, c) such that (y1 − y2 , u) H ≥ −λu2H −
c2 , for each yi ∈ ∂ Ji (u), u ∈ H , where ∂ Ji (u) the subdifferential of Ji (·) at a point
u. Note that u ∗ ∈ ∂ Ji (u) if and only if u ∗ (v − u) ≤ Ji (v) − Ji (u) ∀v ∈ H ; i = 1, 2.
Let D(A) = {u ∈ V : Au ∈ H }. We note that the mapping v → Av H defines the
equivalent norm on D(A).
We recall that the function u(·) ∈ L 2 (τ, T ; V ) is called a weak solution of Problem
(9.2) on [τ, T ], if there exist Bochner measurable functions di : (τ, T ) → H ; i =
1, 2, such that

di (t) ∈ ∂ Ji (u(t)) for a.e. t ∈ (τ, T ), i = 1, 2; and (9.3)

 T  
− u, v ξ (t) + Au, v ξ(t) + d1 , v ξ(t) − d2 , v ξ(t) dt = 0, (9.4)
τ

for all ξ ∈ C0∞ (τ, T ) and for all v ∈ V .

We note that for any u τ ∈ H there exists at least one weak solution of Problem
(9.2) on [τ, T ] with initial condition u(τ ) = u τ . The regularity of each weak solution
follows from Theorem 2.3.
Denote by K+ the family of all, globally defined on [0, +∞), weak solutions of
Problem (9.2). Let us set

1
E(u) = Au, u + J1 (u) − J2 (u), u ∈ V. (9.5)
2

For each u ∈ K+ and all τ and T , 0 < τ < T < ∞, the energy equality holds
 T
du
E(u(T )) − E(u(τ )) = −  (s)2H ds. (9.6)
τ ds
9.1 First Order Autonomous Differential-Operator Equations and Inclusions 213

Fig. 9.1 Multi-valued semiflow

Thus, the function E(u(·)) is absolutely continuous on [τ, T ] as the linear combina-
tion of absolutely continuous on [τ, T ] functions.
Let

Dτ,T (u τ ) = {u(·) ∈ L 2 (τ, T ; V ) u(·) is a weak solution of Problem (9.2) and u(τ ) = u τ },

for any u τ ∈ H .
Define real Banach space

du
W (M1 , M2 ) = {u(·) ∈ C([M1 , M2 ]; V ) : (·) ∈ L 2 (M1 , M2 ; H )}
dt

with the norm uW (M1 ,M2 ) = uC([M1 ,M2 ];V ) +  du  2

dt L (M1 ,M2 ;H )
, u ∈ W (M1 , M2 ),
−∞ < M1 < M2 < +∞.
We denote the set of all nonempty (nonempty bounded) subsets of H by P(H )
(B(H )). Let us define the strict m-semiflow G : R+ × H → P(H ) in the following
way: G(t, u 0 ) = {u(t) : u(·) ∈ K+ , u(0) = u 0 }. We recall that the multi-valued map
G : R+ × H → P(H ) is said to be a strict multi-valued semiflow (strict m-semiflow)
if (see also Fig. 9.1):
(a) G(0, ·) = Id (the identity map);
(b) G(t + s, x) = G(t, G(s, x)) ∀x ∈ H, t, s ∈ R+ .
We recall that the set A ⊆ H is said to be an invariant global attractor of G if:
(1) A is invariant (that is A = G(t, A ) ∀t ≥ 0);
(2) A is attracting set, that is,

dist H (G(t, B), A ) → 0, t → +∞ ∀B ∈ B(H ), (9.7)

where dist H (C, D) = sup inf c − d H is the Hausdorff semidistance;

c∈C d∈D
(3) for any closed set Y ⊆ H satisfying (9.7), we have A ⊆ Y (minimality).
Let {T (h)}h≥0 be the translation semigroup acting on K+ , that is T (h)u(·) =
u(· + h), h ≥ 0, u(·) ∈ K+ . On K+ we consider the topology induced from the
214 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

Fréchet space C loc (R+ ; H ). Note that f n (·) → f (·) in C loc (R+ ; H ) if and only if
∀M > 0 Π M f n (·) → Π M f (·) in C([0, M]; H ), where Π M is the restriction operator
to the interval [0, M].
A set U ⊂ K+ is said to be trajectory attractor in the trajectory space K+ with
respect to the topology of C loc (R+ ; H ), if U ⊂ K+ is a global attractor for the
translation semigroup {T (h)}h≥0 acting on K+ .
The following theorem completely describes the long-time behavior of all weak
solutions, as time t → +∞, for the problem in hands. The structure properties of
global and trajectory attractors and the strongest convergence results of solutions are
provided.

Theorem 9.1 The following statements hold:

(i) the strict m-semiflow G : R+ × H → P(H ) has the invariant global attrac-
tor A ;
(ii) there exists the trajectory attractor U ⊂ K+ in the space K+ ;
(iii) the following equalities hold: U = Π+ K = {y ∈ K+ | y(t) ∈ A ∀t ∈
R+ };
(iv) A is a compact subset of V ;
(v) for each B ∈ B(H ) dist V (G(t, B), A ) → 0, as t → ∞;
(vi) U is a bounded subset of L ∞ (R+ ; V ) and compact subset of W loc (R+ ), that
is Π M U is compact in W (0, M) for each M > 0;
(vii) for any bounded in L ∞ (R+ ; H ) set B ⊂ K+ and any M ≥ 0 the following
relation holds: dist W (0,M) (Π M T (t)B, Π M U ) → 0, t → +∞;
(viii) K is a bounded subset of L ∞ (R; V ) and compact subset of W loc (R),
that is Π M1 ,M2 U is compact in W (M1 , M2 ) for each M1 , M2 , −∞ < M1 <
M2 < +∞;
(ix) for each u ∈ K the limit sets

α(u) = {z ∈ V | u(t j ) → z in V for some sequence t j → −∞},

ω(u) = {z ∈ V | u(t j ) → z in V for some sequence t j → +∞}

are connected subsets of Z on which E is constant. If Z is totally disconnected

(in particular, if Z is countable) the limits in V

z − = lim u(t), z + = lim u(t) (9.8)

t→−∞ t→+∞

exist and z − , z + are rest points; furthermore, u(t) tends in V to a rest point as
t → +∞ for every u ∈ K+ .

Proof Statements (i)–(v) of Theorem 9.1 follow from Kasyanov et al. [2, Theorems
4–6]. Statements (vi)–(viii) of Theorem 9.1 follow from Theorem 5.5 and Kasyanov
et al. [2, Theorem 6]. Statement (ix) of Theorem 9.1 follows from Theorem 5.4 and
Ball [1, Theorem 2.7].
9.2 Second Order Autonomous Operator Differential Equations and Inclusions 215

9.2 Second Order Autonomous Operator Differential

Equations and Inclusions

Let β > 0 be a constant, Ω ⊂ Rn be a bounded domain with sufficiently smooth

boundary ∂Ω. Consider the problem

u tt + βu t − u + f (u) = 0,
(9.9)
u|∂Ω = 0,

where u(x, t) is unknown state function defined on Ω × R+ ; f : R → R is an inter-

action function such that
f (u)
lim > −λ1 , (9.10)
|u|→∞ u

where λ1 is the first eigenvalue for − in H01 (Ω);

∃ D≥0: | f (u)| ≤ D(1 + |u|), ∀u ∈ R. (9.11)

Further, we use such denotation

f (s) := lim sup f (t), f (s) := lim f (t), G(s) := [ f (s), f (s)], s ∈ R.
t→s t→s

Let us set V = H01 (Ω) and H = L 2 (Ω). The space X = V × H is a phase space of
Problem (9.9). For the Hilbert space X as (·, ·) X and  ·  X denote the inner product
and the norm in X respectively.

Definition 9.1 Let T > 0, τ < T . The function ϕ(·) = (u(·), u t (·))T ∈ L ∞ (τ, T ; X )
is called a weak solution of Problem (9.9) on (τ, T ) if for a.e. (x, t) ∈ Ω × (τ, T ),
there exists l = l(x, t) ∈ L 2 (τ, T ; L 2 (Ω)) l(x, t) ∈ G(u(x, t)), such that ∀ψ ∈
H01 (Ω), ∀η ∈ C0∞ (τ, T ),

T T
− (u t , ψ) H ηt dt + (β(u t , ψ) H + (u, ψ)V + (l, ψ) H )ηdt = 0. (9.12)
τ τ

The main goal of the manuscript is to obtain the existence of the global attractor
generated by the weak solutions of Problem (9.9) in the phase space X .
Thus we consider more general evolution inclusion

u tt + βu t − u + [ f 1 (u), f 1 (u)] − [ f 2 (u), f 2 (u)] 0,
(9.13)
u|∂Ω = 0.
216 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

Let us set
s 
G i (s) := f i (ξ )dξ, Ji (u) := G i (u(x))d x, J (u) = J1 (u) − J2 (u), u ∈ H, i = 1, 2.
0 Ω

The functionals G i and Ji are locally Lipschitz and regular. Consider WτT =
C([τ, T ]; X ). For any ϕτ = (a, b)T ∈ X , denote
  
 (u, u t )T is a weak solution of Problem (9.9) on [τ, T ],
Dτ,T (ϕτ ) = (u(·), u t (·))T  ;
 u(τ ) = a, u t (τ ) = b

see Sect. 7.2.

Define the m-semiflow G as

G (t, ξ0 ) = {ξ(t) | ξ(·) ∈ D(ξ0 )}, t ≥ 0.

Denote the set of all nonempty (nonempty bounded) subsets of X by P(X )(β(X )).
Note that the multi-valued map G : R+ × X → P(X ) is a strict m-semiflow,
i.e., (see Lemma 6.9)
1. G (t, ·) = Id (the identity map);
2. G (t + s, x) = G (t, G (s, x)) ∀x ∈ X, t, s ∈ R+ .
Further, ϕ ∈ G means that ϕ ∈ D(ξ0 ) for some ξ0 ∈ X .

Definition 9.2 G is called an asymptotically compact m-semiflow if for any sequence

{ϕn }n≥1 ⊂ G with {ϕn (0)}n≥1 bounded, and for any sequence {tn }n≥1 : tn → +∞,
n → ∞, the sequence {ϕn (tn )}n≥1 has a convergent subsequence Ball [1, p. 35].

Theorem 9.2 G is an asymptotically compact m-semiflow.

Proof Let ξn ∈ G (tn , vn ), vn ∈ B, B ∈ β(X ), n ≥ 1, tn → +∞, n → +∞. Let us

check a precompactness of {ξn }n≥1 in X . Without loss of the generality, we extract a
convergent in X subsequence from {ξn }n≥1 . From Corollary 6.1 we obtain that there
exists {ξn k }k≥1 and ξ ∈ X such that ξn k → ξ weakly in X , ξn k  X → a ≥ ξ  X ,
k → +∞. Show that a ≤ ξ  X .
Let us fix an arbitrary T0 > 0. Then for rather big k ≥ 1, G (tn k , vn k ) ⊂ G (T0 ,
G (tn k − T0 , vn k )). Hence ξn k ∈ G (T0 , βn k ), where βn k ∈ G (tn k − T0 , vn k ) and
sup βn k  X < +∞ (see Corollary 6.1). From Theorem 6.3 for some {ξk j , βk j } j≥1 ⊂
k≥1
{ξn k , βn k }k≥1 , βT0 ∈ X , we obtain

ξ ∈ G (T0 , βT0 ), βk j → βT0 weakly in X, j → +∞. (9.14)

From the definition of G we set ∀ j ≥ 1, ξk j = (u j (T0 ), u j (T0 ))T , βk j = (u j (0),

u j (0))T , ξ = (u 0 (T0 ), u 0 (T0 ))T , βT0 = (u 0 (0), u 0 (0))T , where ϕ j = (u j , u j )T ∈
C([0, T0 ]; X ), u j ∈ L 2 (0, T0 ; V ∗ ), d j ∈ L ∞ (0, T0 ; H ),
9.2 Second Order Autonomous Operator Differential Equations and Inclusions 217

u j (t) + βu j (t) − u j (t) + d j,1 (t) − d j,2 (t) = 0̄,

d j,i (t) ∈ ∂ Ji (u j (t)), i = 1, 2 for a.e. t ∈ (0, T0 ).

Let for every t ∈ [0, T0 ],

1 β
I (ϕ j (t)) := ϕ j (t)2X + J1 (u j (t)) − J2 (u j (t)) + (u j (t), u j (t)) H .
2 2
Then in virtue of Lemma 6.7
d I (ϕ j (t))
= −β I (ϕ j (t)) + βH (ϕ j (t)), for a.e. t ∈ (0, T0 ),
dt
where
1 1
H (ϕ j (t)) = J1 (u j (t)) − (d j,1 (t), u j (t)) − J2 (u j (t)) + (d j,2 (t), u j (t)) H .
2 2

From (9.14) we have ∃ R̄ > 0 : ∀ j ≥ 0, ∀t ∈ [0, T0 ],

u j (t)2H + u j (t)2V ≤ R̄ 2 .

Moreover,

u j → u 0 weakly in L 2 (0, T0 ; V ), j → +∞,

u j → u 0 weakly in L 2 (0, T0 ; H ), j → +∞,
u j → u 0 in L 2 (0, T0 ; H ), j → +∞,
(9.15)
d j,i → di weakly in L 2 (0, T0 ; H ), i = 1, 2, j → +∞,
u j → u 0 weakly in L 2 (0, T0 ; V ∗ ), j → +∞,
∀t ∈ [0, T0 ] u j (t) → u 0 (t) in H, j → +∞.

For every j ≥ 0 and t ∈ [0, T0 ],

t
−βt
I (ϕ j (t)) = I (ϕ j (0))e + H (ϕ j (s))e−β(t−s) ds.
0

T0
In particular I (ϕ j (T0 )) = I (ϕ j (0))e−βT0 + H (ϕ j (s))e−β(T0 −s) ds.
0
From (9.15) and Lemma 6.7 we have

T0 T0
−β(T0 −s)
H (ϕ j (s))e ds → H (ϕ0 (s))e−β(T0 −s) ds, j → +∞.
0 0
218 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

Therefore

T0
lim sup I (ϕ j (T0 )) ≤ lim sup I (ϕ j (0))e−βT0 + H (ϕ0 (s))e−β(T0 −s) ds =
j→+∞ j→+∞ 0

= I (ϕ0 (T0 )) + lim sup I (ϕ j (0)) − I (ϕ0 (0)) e−βT0 ≤ I (ϕ0 (T0 )) + c3 e−βT0 ,
j→+∞

where c3 does not depend on T0 > 0.

On the other hand, from (9.15) we have

1 β
lim sup I (ϕ j (T0 )) ≥ lim ϕ j (T0 )2X + J (u 0 (T0 )) + (u 0 (T0 ), u 0 (T0 )).
j→+∞ 2 j→+∞ 2

Therefore we obtain 21 a 2 ≤ 21 ξ 2X + c3 e−βT0 ∀T0 > 0.

Thus, a ≤ ξ  X .
The Theorem is proved.

Let us consider the family K+ = ∪u 0 ∈X D(u 0 ) of all weak solutions of Problem

(9.9) defined on [0, +∞). Note that K+ is translation invariant one, i.e., ∀u(·) ∈ K+ ,
∀h ≥ 0, u h (·) ∈ K+ , where u h (s) = u(h + s), s ≥ 0. On K+ we set the translation
semigroup {T (h)}h≥0 , T (h)u(·) = u h (·), h ≥ 0, u ∈ K+ . In view of the translation
invariance of K+ we conclude that T (h)K+ ⊂ K+ as h ≥ 0.
On K+ we consider a topology induced from the Fréchet space C loc (R+ ; X ).
Note that

f n (·) → f (·) in C loc (R+ ; X ) ⇐⇒ ∀M > 0, Π M f n (·) → Π M f (·) in C([0, M]; X ),

where Π M is the restriction operator to the interval [0, M]. We denote the restriction
operator to [0, +∞) by Π+ .
Let us consider Problem (9.9) on the entire time axis. Similarly to the space
C loc (R+ ; X ) the space C loc (R; X ) is endowed with the topology of a local uni-
form convergence on each interval [−M, M] ⊂ R. A function u ∈ C loc (R; X ) ∩
L ∞ (R; X ) is said to be a complete trajectory of Problem (9.9) if ∀h ∈ R, Π+ u h (·) ∈
K+ .
Let K be a family of all complete trajectories of Problem (9.9). Note that ∀h ∈ R,
∀u(·) ∈ K u h (·) ∈ K . We say that the complete trajectory ϕ ∈ K is stationary if
ϕ(t) = z for all t ∈ R for some z ∈ X. Following Ball [1, p. 486] we denote by Z (G )
the set of all rest points of G . Note that

Z (G ) = {(0̄, u) | u ∈ V, −(u) + ∂ J (u) 0̄}.

Lemma 9.1 Z (G ) is an bounded set in X .

The existence of a Lyapunov function for G follows from Lemma 6.10 (see Ball
[1, p. 486]).
9.2 Second Order Autonomous Operator Differential Equations and Inclusions 219

Lemma 9.2 A functional V : X → R defined by (6.44) is a Lyapunov function

for G .

We recall that the set A is said to be a global attractor G if

(1) A is negatively semiinvariant (i.e., A ⊂ G (t, A ) ∀t ≥ 0);
(2) A is attracting set, i.e.,

dist(G (t, B), A ) → 0, t → +∞, ∀B ∈ β(X ), (9.16)

where dist(C, D) = sup inf c − d X is the Hausdorff semidistance;

c∈C d∈D
(3) for any closed set Y ⊂ H satisfying (9.16), we have A ⊂ Y (minimality).
The global attractor is said to be invariant if A = G (t, A ), ∀t ≥ 0.
Note that by definition a global attractor is unique.
We prove the existence of an invariant compact global attractor.

Theorem 9.3 The m-semiflow G has an invariant compact in the phase space X
global attractor A . For each ψ ∈ K the limit sets

α(ψ) = {z ∈ X| ψ(t j ) → z for some sequence t j → −∞},

ω(ψ) = {z ∈ X| ψ(t j ) → z for some sequence t j → +∞}

are connected subsets of Z (G ) on which V is constant. If Z (G ) is totally disconnected

(in particular if Z (G ) is countable) the limits

z − = lim ψ(t), z + = lim ψ(t)

t→−∞ t→+∞

exist and z − , z + are rest points; furthermore, ϕ(t) tends to a rest point as t → +∞
for every solution ϕ ∈ K+ .

Proof The existence of a global attractor for Second Order Evolution Inclusions
directly follows from Lemmas 6.8, 6.9, 9.1 and 9.2, Theorems 6.3, 6.4, 9.2 and Ball
[1, Theorem 2.7].

9.3 Examples of Applications

In this section we provide examples of applications to theorems established in pre-

vious sections. We consider a model of combustion in porous media, a model of
conduction of electrical impulses in nerve axons, a climate energy balance model
etc. (see also [4–21, 23–46, 49–96]).
220 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

Fig. 9.2 Approximations for trajectories to model of combustion in porous media in a moment
a t = 0; b t = 0.8; c t = 1.6; d t = 2.4; e t = 3.2; f t = 4

9.3.1 A Model of Combustion in Porous Media

Let us consider the following problem:

 ∂u
− ∂∂ xu2 − f (u) ∈ λH (u − 1), (x, t) ∈ (0, π ) × R,
2

∂t (9.17)
u(0, t) = u(π, t) = 0, t ∈ R,

where f : R → R is a continuous and nondecreasing function satisfying growth

and sign assumptions, λ > 0, and H (0) = [0, 1], H (s) = I{s > 0}, s = 0. For each
u τ ∈ L 2 ((0, π )) we set Dτ,T (u τ ) = {u(·) ∈ L 2 (τ, T ; H01 ((0, π ))) u(·) is a weak
solution of Problem (9.17) and u(τ ) = u τ }. Since Problem (9.17) is a particular case
of Problem (9.2), then all statements from Sect. 10.1 hold; Fig. 9.2.

9.3.2 A Model of Conduction of Electrical Impulses in Nerve

Axons

Consider the problem:

 ∂u
− ∂∂ xu2 + u ∈ λH (u − a), (x, t) ∈ (0, π ) × R,
2

∂t (9.18)
u(0, t) = u(π, t) = 0, t ∈ R,

where a ∈ 0, 21 ; Terman [47, 48]. Since Problem (9.18) is a particular case of

Problem (9.2), then all statements from Sect. 9.1 hold; Fig. 9.3.
9.3 Examples of Applications 221

Fig. 9.3 Modeling of

solutions for a model of
conduction of electrical
impulses in nerve axons

9.3.3 Viscoelastic Problems with Nonlinear

“Reaction-Displacement” Law

Let a viscoelastic body occupy a bounded domain Ω ⊂ Rd , d = 2, 3 in applications,

and it is acted upon by volume forces and surface tractions.1 The boundary Γ of Ω is
supposed to be Lipschitz continuous and it is partitioned into two disjoint measurable
parts Γ D and Γ N such that meas(Γ D ) > 0. We consider the process of evolution of
the mechanical state on the interval (0, +∞). The body is clamped on Γ D and
thus the displacement vanishes there. The forces field of density f 0 act in Ω, the
surface tractions of density g0 are applied on Γ N . We denote by u = (u 1 , . . . , u d )
the displacement vector, by σ = (σi j ) the stress tensor and by ε(u) = (εi j (u)) the
linearized (small) strain tensor (εi j (u) = 21 (∂ j u i + ∂i u j )), where i, j = 1, . . . , d.
The mechanical problem consists in finding the displacement field u : Ω ×
(0, +∞) → Rd such that

u (t) − divσ (t) = f 0 in Ω × (0, +∞), (9.19)

σ (t) = C ε(u (t)) + E ε(u(t)) in Ω × (0, +∞), (9.20)

u(t) = 0 on Γ D × (0, +∞), (9.21)

σ n(t) = g0 on Γ N × (0, +∞), (9.22)

u(0) = u 0 , u (0) = u 1 in Ω, (9.23)

where C and E are given linear constitutive functions, n being the outward unit
normal vector to Γ.

1 This section is based on results of [22] and references therein.

222 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

Fig. 9.4 Foundation, body,

and main forces

In the above model the dynamic equation (9.19) is considered with the viscoelas-
tic constitutive relationship of the Kelvin-Voigt type (9.20) while (9.21) and (9.22)
represent the displacement and traction boundary conditions (Fig. 9.4), respectively.
The functions u 0 and u 1 are the initial displacement and the initial velocity, respec-
tively. In order to formulate the skin effects, we suppose that the body forces of
density f 0 consists of two parts: f 1 which is prescribed external loading and f 2
which is the reaction of constrains introducing the skin effects, i.e. f 0 = f 1 + f 2 .
Here f 2 is a possibly multi-valued function of the displacement u. We consider the
reaction-displacement law of the form

− f 2 (x, t) ∈ ∂ j (x, u(x, t)) in Ω × (0, +∞), (9.24)

where j : Ω × Rd → R is locally Lipschitz function in u and ∂ j represents the

Clarke subdifferential with respect to u. Let Yd be the space of second-order sym-
metric tensors on Rd .
We consider the following problem:
examine the long-time (as t → +∞) behavior of all (weak, generalized) solutions
for (9.19)–(9.23) and (9.24).
In [22] for finite time interval it was presented the hemivariational formulation of
problems similar to (9.19)–(9.24) and an existence theorem for evolution inclusions
with pseudomonotone operators. We give now variational formulation of the above
problem. To this aim let H = L 2 (Ω, Rd ), H1 = H 1 (Ω, Rd ), H = L 2 (Ω, Yd ) and
V be the closed subspace of H1 defined by

V = {v ∈ H1 : v = 0 on Γ D }.

On V we consider the inner product and the corresponding norm given by

(u, v)V = ε(u), ε(v) H , vV = ε(v)H for u, v ∈ V.

From the Korn inequality v H1 ≤ C1 ε(v)H for v ∈ V with C1 > 0, it follows
that  ·  H1 and  · V are the equivalent norms on V. Identifying H with its dual,
9.3 Examples of Applications 223

we have an evolution triple V ⊂ H ⊂ V ∗ (see e.g. [3]) with dense and compact
embeddings. We denote by ·, · V the duality of V and its dual V ∗ , by  · V ∗ the
norm in V ∗ . We have u, v V = (u, v) H for all u ∈ H and v ∈ V.
We admit the following hypotheses:
H(C ). The linear symmetric viscosity operator C : Ω × Yd → Yd satisfies the
Carathéodory condition (i.e. C (·, ε) is measurable on Ω for all ε ∈ Yd and C (x, ·)
is continuous on Yd for a.e. x ∈ Ω) and

C (x, ε) : ε ≥ C2 ε2Yd for all ε ∈ Yd and a.e. x ∈ Ω with C2 > 0. (9.25)

H(E ). The elasticity operator E : Ω × Yd → Yd is of the form E (x, ε) = E(x)ε

(Hooke’s law) with a symmetric elasticity tensor E ∈ L ∞ (Ω), i.e. E = (gi jkl ),
i, j, k, l = 1, . . . , d with gi jkl = g jikl = glki j ∈ L ∞ (Ω). Moreover,

E (x, ε) : ε ≥ C3 ε2Yd for all ε ∈ Yd and a.e. x ∈ Ω with C3 > 0.

H(j). j : Ω × Rd → R is a function such that

(i) j (·, ξ ) is measurable for all ξ ∈ Rd and j (·, 0) ∈ L 1 (Ω);
(ii) j (x, ·) is locally Lipschitz and regular [8] for all x ∈ Ω;
(iii) η ≤ C4 (1 + ξ ) for all η ∈ ∂ j (x, ξ ), x ∈ Ω with C4 > 0;
(iv) j 0 (x, ξ ; −ξ ) ≤ C5 (1 + ξ ) for all ξ ∈ Rd , x ∈ Ω, with C5 ≥ 0, where
j (x, ξ ; η) is the directional derivative of j (x, ·) at the point ξ ∈ Rd in the direction
0

η ∈ Rd .
H(f). f 1 ∈ V ∗ , g0 ∈ L 2 (Γ N ; Rd ), u 0 ∈ V and u 1 ∈ H.
Next we need the spaces V = L 2 (τ, T ; V ), Hˆ = L 2 (τ, T ; H ) and W = {w ∈
V : w ∈ V ∗ }, where the time derivative involved in the definition of W is under-
stood in the sense of vector-valued distributions, −∞ < τ < T < +∞. Endowed
with the norm vW = vV + v V ∗ , the space W becomes a separable reflexive
Banach space. We also have W ⊂ V ⊂ Hˆ ⊂ V ∗ . The duality for the pair (V , V ∗ )
T
is denoted by z, w V = z(s), w(s) V ds. It is well known (cf. [13]) that the embed-
τ
ding W ⊂ C([τ, T ]; H ) and {w ∈ V : w ∈ W } ⊂ C([τ, T ]; V ) are continuous.
Next we define g ∈ V ∗ by

g, v V = f1 , v V + g0 , v L 2 (Γ N ;Rd ) for v ∈ V. (9.26)

Taking into account the condition (9.24), we obtain the following variational formu-
lation of our problem:
⎧ 
⎪
⎪ u (t), v V + (σ (t), ε(v))H + j 0 (x, u(t); v)d x ≥
⎪
⎨ Ω
≥ g, v V for all v ∈ V and a.e. t ∈ (0, +∞), (9.27)
⎪
⎪ σ (t) = C (ε(u (t))) + E (ε(u(t))) for a.e. t ∈ (0, +∞),
⎪
⎩
u(0) = u 0 , u (0) = u 1 .
224 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

We define the operators A : V → V ∗ and B : V → V ∗ by

A(u), v V = (C (x, ε(u)), ε(v))H for u, v ∈ V, (9.28)

Bu, v V = (E (x, ε(u)), ε(v))H for u, v ∈ V. (9.29)

Obviously the bilinear forms (9.28) and (9.29) are symmetric, continuous and
coercive.
Let us introduce the functional J : L 2 (Ω; Rd ) → R defined by

J (v) = j (x, v(x))d x for v ∈ L 2 (Ω; Rd ). (9.30)
Ω

From Sect. 7.3, under Assumptions H(j), the functional J defined by (9.30) satisfies
H(J). J : L 2 (Ω; Rd ) → R is a functional such that:
(i) J (·) is well defined, locally Lipschitz (in fact, Lipschitz on bounded subsets
of L 2 (Ω; Rd )) and admits the representation via the difference of convex functions;
(ii) ζ ∈ ∂ J (v) implies ζ  L 2 (Ω;Rd ) ≤ C6 (1 + v L 2 (Ω;Rd ) ) for v ∈ L 2 (Ω; Rd )
with C6 > 0;
(iii) J 0 (v; −v) ≤ C7 (1 + v L 2 (Ω;Rd ) ) for v ∈ L 2 (Ω; Rd ) with C7 ≥ 0, where
J (u; v) denotes the directional derivative of J (·) at a point u ∈ L 2 (Ω; Rd ) in the
0

direction v ∈ L 2 (Ω; Rd ).
We can now formulate the second-order evolution inclusions associated with the
variational form of our problem
⎧
⎪
⎪ Find u ∈ C([0, +∞); V ) with u ∈ C([0, +∞); H ) ∩ L loc
2 (0, +∞; V )
⎨ ∗
and u ∈ L loc2 (0, +∞; V ) such that
(9.31)
⎪
⎪ u (t) + Au (t) + Bu(t) + ∂ J (u(t)) g a.e. t ∈ (0, +∞),
⎩
u(0) = u 0 , u (0) = u 1 .

Theorem 6.6 yields that, if τ < T, {ϕn (·)}n≥1 ⊂ WτT is an arbitrary sequence of weak
solutions of (9.31) on [τ, T ] such that ϕn (τ ) → ϕτ strongly in E, n → +∞, then
there exist ϕ ∈ Dτ,T (ϕτ ) such that up to a subsequence ϕn (·) → ϕ(·) in C([τ, T ]; E),
n → +∞ (see Sect. 7.3 for details).
We define the m-semiflow G as G (t, ξ0 ) = {ξ(t) | ξ(·) ∈ D(ξ0 )}, t ≥ 0. Denote the
set of all nonempty (nonempty bounded) subsets of E by P(E) (β(E)). We remark
that the multi-valued map G : R+ × E → P(E) is strict m-semiflow, i.e. G (0, ·) =
Id (the identity map), G (t + s, x) = G (t, G (s, x)) ∀x ∈ E, t, s ∈ R+ . Further ϕ ∈ G
will mean that ϕ ∈ D(ξ0 ) for some ξ0 ∈ E.

Definition 9.3 ([1, p. 35]) The m-semiflow G is called asymptotically compact, if

for any sequence ϕ j ∈ G with ϕ j (0) bounded, and for any sequence t j → +∞, the
sequence ϕ j (t j ) has a convergent subsequence.
9.3 Examples of Applications 225

Theorem 9.4 The m-semiflow G is asymptotically compact.

Proof Let ξn ∈ G (tn , vn ), vn ∈ B ∈ β(E), n ≥ 1, tn → +∞, n → +∞. Let us

check the precompactness of {ξn }n≥1 in E. In order to do that without loss of the
generality it is sufficiently to extract a convergent in E subsequence from {ξn }n≥1 .
From Corollary 6.2 we obtain that there exist {ξn k }k≥1 and ξ ∈ E such that ξn k → ξ
weakly in E, ξn k  E → a ≥ ξ  E , k → +∞. We show that a ≤ ξ  E . Let us fix
an arbitrary T0 > 0. Then for rather big k ≥ 1 G (tn k , vn k ) = G (T0 , G (tn k − T0 , vn k )).
Hence ξn k ∈ G (T0 , βn k ), where βn k ∈ G (tn k − T0 , vn k ) and

δ := sup βn k  E < +∞

k≥1

(see Corollary 6.2). From Theorem 6.5 for some {ξk j , βk j } j≥1 ⊂ {ξn k , βn k }k≥1 , βT0 ∈
E we obtain:

ξ ∈ G (T0 , βT0 ), βk j → βT0 weakly in E, j → +∞. (9.32)

From the definition of G we set: ∀ j ≥ 1 ξk j = (y j (T0 ), y j (T0 ))T , βk j = (y j (0),

y j (0))T , ξ = (y0 (T0 ), y0 (T0 ))T , βT0 = (y0 (0), y0 (0))T , where ϕ j = (y j , y j )T ∈ C
([0, T0 ]; E), y j ∈ L 2 (0, T0 ; V ), y j ∈ L 2 (0, T0 ; V ∗ ), d j ∈ L ∞ (0, T0 ; H ),

y j (t) + Ay j (t) + By j (t) + d j (t) = 0̄, d j (t) ∈ ∂ J (y j (t)) for a.e. t ∈ (0, T0 ), ∀ j ≥ 0.

Now we fix an arbitrary ε > 0. Let for each t ∈ [0, T0 ], j ≥ 0

1
Iε (ϕ j (t)) := ϕ j (t)2E + J (y j (t)) + ε(y j (t), y j (t)) H .
2
Then,

d Iε (ϕ j (t))
= −2ε Iε (ϕ j (t)) + 2εHε (ϕ j (t)) − ε Ay j (t), y j (t) V − Ay j (t), y j (t) V ,
dt

for a.e. t ∈ (0, T0 ), where

1
Hε (ϕ j (t)) = J (y j (t)) − (d j (t), y j (t)) H + y j (t)2H + ε(y j (t), y j (t)) H , for a.e. t ∈ (0, T0 ).
2

Thus, for any j ≥ 0 and t ∈ [0, T0 ]

T0
−2εT0
Iε (ϕ j (T0 )) = Iε (ϕ j (0))e + 2ε Hε (ϕ j (t))e−2ε(T0 −t) dt−
0
226 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

T0 T0
−2ε(T0 −t)
−ε Ay j (t), y j (t) V e dt − Ay j (t), y j (t) V e−2ε(T0 −t) dt.
0 0

From (6.64), (9.32) and Lemma 6.15 we have

t
y j (t)2H + y j (t)2V +γ y j (t)2V dt ≤ R̄ ∀ j ≥ 0, ∀t ∈ [0, T0 ],
0

where R̄ > 0 is a constant. Moreover,

yj → y0 weakly in L 2 (0, T0 ; V ), y j → y0 weakly in L 2 (0, T0 ; V ),

yj → y0 strongly in L 2 (0, T0 ; H ), y j → y0 strongly in L 2 (0, T0 ; H ),
yj → y0 weakly in L 2 (0, T0 ; V ∗ ), d j → d0 weakly in L 2 (0, T0 ; H ),
∀t ∈ [0, T0 ] y j (t) → y0 (t) in H, y j (t) → y0 (t) weakly in H, j → +∞.
(9.33)
Therefore,

T0 T0
−2ε(T0 −t)
Hε (ϕ j (t))e dt → Hε (ϕ0 (t))e−2ε(T0 −t) dt, j → +∞,
0 0

T0 T0
−2ε(T0 −t)
lim Ay j (t), y j (t) V e dt ≥ Ay0 (t), y0 (t) V e−2ε(T0 −t) dt.
j→+∞
0 0

T0
The last inequality holds, because of the functional v(·)→ Av(t), v(t) V e−2ε(T0 −t) dt
0
is sequentially weakly lower semi-continuous on L 2 (τ, T0 ; V ). Furthermore,

T0
ε ε
ε Ay j (t), y j (t) V e−2ε(T0 −t) dt = Ay j (T0 ), y j (T0 ) V − Ay j (0), y j (0) V e−2εT0 −
2 2
0

T0
−ε 2
Ay j (t), y j (t) V e−2ε(T0 −t) dt ∀ j ≥ 0,
0

from which, by Corollary 6.2, we have

 
 T0   
  λ1 + 2c3 2 2(c3 + c4 )λ1
 −2ε(T −t) 
ε Ay j (t), y j (t) V e 0 dt  ≤ 2εγ δ + , ∀ j ≥ 0.
  λ1 − μ λ1 − μ
 0 
9.3 Examples of Applications 227

Thus,
 
lim Iε (ϕ j (T0 )) ≤ Iε (ϕ0 (0))e−2εT0 + lim Iε (ϕ j (0)) − Iε (ϕ0 (0)) e−2εT0 +
j→+∞ j→+∞

T0 T0
−2ε(T0 −t)
+2ε Hε (ϕ0 (t))e dt − ε Ay0 (t), y0 (t) V e−2ε(T0 −t) dt−
0 0

T0  
λ1 + 2c3 2 2(c3 + c4 )λ1
− Ay0 (t), y0 (t) V e−2ε(T0 −t) dt + 4εγ δ + ≤
λ1 − μ λ1 − μ
0

 
2 −2εT0 λ1 + 2c3 2 2(c3 + c4 )λ1
≤ Iε (ϕ0 (T0 )) + δ e + 4εγ δ +
λ1 − μ λ1 − μ

and, due to (9.33), for any T0 > 0 and ε > 0

 
1 1 1 1 λ1 + 2c3 2 2(c3 + c4 )λ1
ξ 2E ≤ a 2 = lim ξk j 2E ≤ ξ 2E + δ 2 e−2εT0 + 4εγ δ + .
2 2 2 j→+∞ 2 λ1 − μ λ1 − μ

Hence, for all ε > 0 we have

 
1 1 1 λ1 + 2c3 2 2(c3 + c4 )λ1
ξ 2E ≤ a 2 ≤ ξ 2E + 4εγ δ + .
2 2 2 λ1 − μ λ1 − μ

Thus, a = ξ  E .
The theorem is proved.

Let us consider the family K+ = ∪ y0 ∈E D(y0 ) of all weak solutions of the inclu-
sion (9.31), defined on [0, +∞). Note that K+ is translation invariant one, i.e. for all
u(·) ∈ K+ and all h ≥ 0 we have u h (·) ∈ K+ , where u h (s) = u(h + s), s ≥ 0. On
K+ we set the translation semigroup {T (h)}h≥0 , T (h)u(·) = u h (·), h ≥ 0, u ∈ K+ .
In view of the translation invariance of K+ we conclude that T (h)K+ ⊂ K+ as
h ≥ 0.
On K+ we consider the topology induced from the Fréchet space C loc (R+ ; E).
Note that

f n (·) → f (·) in C loc (R+ ; E) ⇐⇒ ∀M > 0 Π M f n (·) → Π M f (·) in C([0, M]; E),

where Π M is the restriction operator to the interval [0, M]. We denote the restriction
operator to [0, +∞) by Π+ .
Let us consider the autonomous inclusion (9.31) on the entire time axis. Similarly
to the space C loc (R+ ; E) the space C loc (R; E) is endowed with the topology of local
uniform convergence on each interval [−M, M] ⊂ R. A function u ∈ C loc (R; E) ∩
228 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

L ∞ (R; E) is said to be a complete trajectory of the inclusion (9.31), if ∀h ∈ R

Π+ u h (·) ∈ K+ . Let K be a family of all complete trajectories of the inclusion
(9.31). Note that ∀h ∈ R, ∀u(·) ∈ K u h (·) ∈ K . We say that the complete trajectory
ϕ ∈ K is stationary if ϕ(t) = z for all t ∈ R for some z = (u, 0̄)T ∈ E (rest point).
We denote the set of rest points of G by Z (G ). We remark that Z (G ) = {(u, 0̄)T | u ∈
V, B(u) + ∂ J (u) 0̄}.
From Conditions H (B) and H (J ) it follows that
Lemma 9.3 The set Z (G ) is nonempty and bounded in E.
From Lemma 6.10 the existence of Lyapunov function for G follows.

Lemma 9.4 The functional V : E → R, defined by

1
V (ϕ) = ϕ2E + J (a). (9.34)
2

, is a Lyapunov function for G .

We recall that the set A ⊂ E is said to be a global attractor for G , if

(1) A is negatively semiinvariant (i.e. A ⊂ G (t, A ) ∀t ≥ 0);
(2) A is attracting set i.e.

dist(G (t, B), A ) → 0, t → +∞ ∀B ∈ β(E), (9.35)

where dist(C, D) = sup inf c − d E is the Hausdorff semidistance;

c∈C d∈D
(3) for any closed set Y ⊂ E, satisfying (9.35), we have A ⊂ Y (minimality).
The global attractor is said to be invariant, if A = G (t, A ) ∀t ≥ 0.
Note that from the definition of the global attractor it follows that it is unique.
We prove the existence of the invariant compact global attractor.
Theorem 9.5 The m-semiflow G has the invariant compact in the phase space E
global attractor A . For each ψ ∈ K the limit sets

α(ψ) = {z ∈ E| ψ(t j ) → z f or some sequence t j → −∞},

ω(ψ) = {z ∈ E| ψ(t j ) → z f or some sequence t j → +∞}

are connected subsets of Z (G ) on which V is constant. If Z (G ) is totally disconnected

(in particular, if Z (G ) is countable) the limits

z − = lim ψ(t), z + = lim ψ(t)

t→−∞ t→+∞

exists and z − , z + are rest points; furthermore, ϕ(t) tends to a rest point as t → +∞
for every ϕ ∈ K+ .
9.3 Examples of Applications 229

Proof The existence of the global attractor with required properties directly follows
from previous theorems and [1, Theorem 2.7].

We remark in advance

∀h ∈ R, ∀u(·) ∈ K u h (·) ∈ K . (9.36)

Lemma 9.5 The set K is nonempty and

∀ξ(·) ∈ K , ∀t ∈ R ξ(t) ∈ A , (9.37)

where A is the global attractor from Theorem 9.5.

Proof Let us show that K = ∅. Note that in virtue of Lemma 9.3, the set Z (G )
is nonempty and bounded in E. Let (v, 0̄)T ∈ Z (G ). We set u(t) = v ∀t ∈ R. Then
(u, u )T ∈ K = ∅.
Let us prove (9.37). For any y ∈ K ∃d > 0: y(t) E ≤ d ∀t ∈ R. We set B =
∪t∈R {y(t)} ∈ β(E). Note that ∀τ ∈ R, ∀t ∈ R+ y(τ ) = yτ −t (t) ∈ G (t, yτ −t (0)) ⊂
G (t, B). From Theorem 9.5 and from (9.35) it follows that ∀ε > 0 ∃T > 0: ∀τ ∈ R
dist(y(τ ), A ) ≤ dist(G (T, B), A ) < ε. Hence taking into account the compactness
of A in E, it follows that y(τ ) ∈ A for any τ ∈ R.

Lemma 9.6 The set K is compact in C loc (R; E) and bounded in L ∞ (R; E).

Proof The boundedness of K in L ∞ (R+ ; E) follows from (9.37) and from the
boundedness of A in E.
Let us check the compactness of K in C loc (R; E). In order to do that it is sufficient
to check the precompactness and completeness.
Step 1. Let us check the precompactness of K in C loc (R; E). If it is not true then
in view of (9.36), ∃M > 0: Π M K is not precompact set in C([0, M]; E). Hence
there exists a sequence {vn }n≥1 ⊂ Π M K , that has not a convergent subsequence in
C([0, M]; E). On the other hand vn = Π M u n , where u n ∈ K , vn (0) = u n (0) ∈ A ,
n ≥ 1. Since A is compact set in E (see Theorem 9.5), then in view of Theo-
rem 6.4, ∃{vn k }k≥1 ⊂ {vn }n≥1 , ∃η ∈ E, ∃v(·) ∈ D0,M (η): vn k (0) → η in E, vn k → v
in C([0, T ]; E), k → +∞. We obtained a contradiction.
Step 2. Let us check the completeness of K in C loc (R; E). Let {vn }n≥1 ⊂ K ,
v ∈ C loc (R; E): vn → v in C loc (R; E), n → +∞. From the boundedness of K in
L ∞ (R; E) it follows that v ∈ L ∞ (R; E). From Theorem 6.4 we have that ∀M > 0
the restriction v(·) to the interval [−M, M] belongs to D−M,M (v(−M)). Therefore
v(·) is a complete trajectory of the inclusion (9.31). Thus, v ∈ K .

Lemma 9.7 Let A be the global attractor from Theorem 9.5. Then

∀y0 ∈ A ∃y(·) ∈ K : y(0) = y0 . (9.38)

230 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

Proof Let y0 ∈ A , u(·) ∈ D(y0 ). From Theorem 9.5 G (t, A ) = A ∀t ∈ R+ . There-

fore,
u(t) ∈ A ∀t ∈ R+ ,

∀η ∈ A ∃ξ ∈ A , ∃ϕη (·) ∈ D0,1 (ξ ) : ϕη (1) = η.

For any t ∈ R we set


u(t), t ∈ R+ ,
y(t) =
ϕ y(−k+1) (t + k), t ∈ [−k, −k + 1), k ∈ N.

Note that y ∈ C loc (R; E), y(t) ∈ A ∀t ∈ R (hence y ∈ L ∞ (R; E)) y ∈ K . More-
over y(0) = y0 .

Now we shall construct the attractor of the translation semigroup {T (h)}h≥0 ,

acting on K+ . We recall that the set P ⊂ C loc (R+ ; E) ∩ L ∞ (R+ ; E) is said to be
an attracting one for the trajectory space K+ of the inclusion (9.31) in the topology
of C loc (R+ ; E), if for any bounded (in L ∞ (R+ ; E)) set B ⊂ K+ and an arbitrary
number M ≥ 0 the next relation

distC([0,M];E) (Π M T (t)B, Π M P) → 0, t → +∞ (9.39)

holds.
A set U ⊂ K+ is said to be trajectory attractor in the trajectory space K+ with
respect to the topology of C loc (R+ ; E) (cf. [5, Definition 1.2, p. 179]), if
(i) U is a compact set in C loc (R+ ; E) and bounded in L ∞ (R+ ; E);
(ii) U is strictly invariant with respect to {T (h)}h≥0 , i.e. T (h)U = U ∀h ≥ 0;
(iii) U is an attracting set in the trajectory space K+ in the topology C loc (R+ ; E).
Note that from the definition of the trajectory attractor it follows that it is unique.
The existence of the trajectory attractor and its structure properties follow from
such theorem:
Theorem 9.6 Let A be the global attractor from Theorem 9.5. Then there exists the
trajectory attractor P ⊂ K+ in the space K+ and we have

P = Π+ K = {y ∈ K+ | y(t) ∈ A ∀t ∈ R+ }. (9.40)

Proof The proof intersects with proofs of previous sections results.

From Lemmas 9.5, 9.6 and the continuity of the operator Π+ : C loc (R; E) →
C (R+ ; E) it follows that the set Π+ K is nonempty, compact in C loc (R+ ; E) and
loc

bounded in L ∞ (R+ ; E). Moreover, the second equality in (9.40) holds (Lemma 9.5
and the proof of Lemma 9.7). The strict invariance of Π+ K follows from the auton-
omy of the inclusion (9.31).
Let us prove that Π+ K is the attracting set for the trajectory space K+ in the
topology of C loc (R+ ; E). Let B ⊂ K+ be a bounded set in L ∞ (R+ ; E), M ≥ 0.
9.3 Examples of Applications 231

Let us suppose M > 0. Let us check (9.39). If it is not true, then there exist ε > 0,
the sequences tn → +∞, vn (·) ∈ B such that

∀n ≥ 1 distC([0,T ];E) (Π M vn (tn + ·), Π M K ) ≥ ε. (9.41)

On the other hand, from the boundedness of B in L ∞ (R+ ; E) it follows that ∃R > 0:
∀v(·) ∈ B, ∀t ∈ R+ v(t) E ≤ R. Hence, taking into account (9.35) and the asymp-
totic compactness of m-semiflow G (Theorem 9.4) we obtain that ∃{vn k (tn k )}k≥1 ⊂
{vn (tn )}n≥1 , ∃z ∈ A : vn k (tn k ) → z in E, k → +∞. Further, ∀k ≥ 1 we set ϕk (t) =
vn k (tn k + t), t ∈ [0, M]. Note that ∀k ≥ 1 ϕk (·) ∈ D0,M (vn k (tn k )). Then from The-
orem 6.4 there exists a subsequence {ϕk j } j≥1 ⊂ {ϕk }k≥1 and an element ϕ(·) ∈
D0,M (z):
ϕk j → ϕ in C([0, M]; E), j → +∞. (9.42)

Moreover, taking into account the invariance of A (see Theorem 9.5), for all t ∈
[0, M] ϕ(t) ∈ A . From Lemma 9.7 there exist y(·), v(·) ∈ K : y(0) = z, v(0) =
ϕ(M). For any t ∈ R we set
⎧
⎨ y(t), t ≤ 0,
ψ(t) = ϕ(t), t ∈ [0, M],
⎩
v(t − M), t ≥ M.

Therefore, from (9.41) we obtain:

∀k ≥ 1 Π M vn k (tn k + ·) − Π M ψ(·)C([0,M];E) = ϕk − ϕC([0,M];E) ≥ ε,

that contradicts with (9.42). We reason in the same way when M = 0.

Thus, the set P constructed in (9.40) is the trajectory attractor in the trajectory
space K+ with respect to the topology of C loc (R+ ; E).

Let A be the global attractor from Theorem 9.5, P be the trajectory attractor
from Theorem 9.6. From previous sections results we have:

A is a compact set in the space E; (9.43)

P is a compact set in the space C loc (R+ ; E); (9.44)

P = Π+ K = {y ∈ K+ | y(t) ∈ A ∀t ∈ R+ } = {y ∈ K+ , | y(0) ∈ A }, (9.45)

where K is the family of all complete trajectories of the inclusion (9.31), Π+ is the
restriction operator on R+ . Moreover,

K is a compact in the space C loc (R; E); (9.46)

232 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

∀ξ(·) ∈ K and ∀t ∈ R ξ(t) ∈ A ; (9.47)

∀y0 ∈ A and ∀t0 ∈ R ∃y(·) ∈ K : y(t0 ) = y0 . (9.48)

For any y ∈ K let us set

H (y) = clC loc (R;E) {y(· + s)| s ∈ R} ⊂ C loc (R; E) ∩ L ∞ (R; E).

Such family is said to be the hull of function y(·) in Ξ = C loc (R; E).
Definition 9.4 The function y(·) ∈ Ξ is said to be translation-compact (tr.-c.) in Ξ
if the hull H (y) is compact in Ξ.

Definition 9.5 The family U ⊂ Ξ is said to be translation-compact, if H (U ) =

clΞ {y(· + s)| y(·) ∈ U , s ∈ R} is a compact in Ξ.

From the autonomy of problem (9.31) and (9.46) it follows that

Corollary 9.1 K is translation-compact set in Ξ.

From autonomy of system (9.31), applying the Arzelá-Ascoli compactness criterion,

we obtain the translation compactness criterion for the family K :
a) the set {y(t)| y ∈ K } is a compact in E ∀t ∈ R;
b) there exists a positive function α(s) → 0+ (s → 0+) such that

y(t1 ) − y(t2 ) E ≤ α(|t1 − t2 |) ∀y ∈ K and ∀t1 , t2 ∈ R.

Similarly if we set Ξ+ = C loc (R+ ; E) we obtain:

Corollary 9.2 P is translation-compact set in Ξ+ .

References

1. Ball, J.M.: Global attractors for damped semilinear wave equations. DCDS 10, 31–52 (2004)
2. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Regularity of weak solutions and their attrac-
tors for a parabolic feedback control problem. Set-Valued Var. Anal. 21, 271–282 (2013).
doi:10.1007/s11228-013-0233-8
3. Kasyanov, P.O., Mel’nik, V.S., Yasinskiy, V.V.: Evolution Inclusions and Inequalities in Banach
Spaces with wλ -pseudomonotone Maps. Naukova dumka, Kyiv (2007)
4. Zgurovsky, M.Z., Kasyanov, P.O., Mel’nik, V.S.: Differential-Operator Inclusions and Variation
Inequalities in Infinite-Dimensional Spaces. Naukova dumka, Kyiv (2008). [in Russian]
5. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation Inequali-
ties for Earth Data Processing. I. Operator Inclusions and Variation Inequalities for Earth Data
Processing. Advances in Mechanics and Mathematics, vol. 24. Springer, Berlin (2011). doi:10.
1007/978-3-642-13837-9
References 233

6. Zgurovsky, M.Z., Mel’nik, V.S., Kasyanov, P.O.: Evolution Inclusions and Variation Inequal-
ities for Earth Data Processing. II. Differential-Operator Inclusions and Evolution Variation
Inequalities for Earth Data Processing. Advances in Mechanics and Mathematics, vol. 25.
Springer, Berlin (2011). doi:10.1007/978-3-642-13878-2
7. Zgurovsky, M.Z., Kasyanov, P.O., Kapustyan, O.V., Valero, J., Zadoianchuk, N.V.: Evolution
Inclusions and Variation Inequalities for Earth Data Processing III. Long-Time Behavior of
Evolution Inclusions Solutions in Earth Data Analysis. Advances in Mechanics and Mathe-
matics, vol. 27. Springer, Berlin (2012). doi:10.1007/978-3-642-28512-7
8. Kapustyan, O.V., Kasyanov, P.O.: Multi-valued dynamics of non-autonomous inclusion with
upper semi-continuous right-hand side [in Ukrainian]. Rep. Natl. Acad. Sci. Ukr. 5, 13–16
(2003)
9. Kapustyan, O.V., Kasyanov, P.O.: Global attractor inclusion of non-autonomous discontinuous
right-hand side. Ukr. Math. J. 55(11), 1467–1475 (2003). doi:10.1023/B:UKMA.0000027041.
12041.e8
10. Kasyanov, P.O.: Galerkin method for a class of differential-operator inclusions with multi-
valued maps of psevdomonotone type [in Ukrainian]. Res. Bull. NTUU “KP” 2, 139–151
(2005)
11. Kasyanov, P.O.: Faedo–Galerkin method for a class of differential-operator inclusions [in
Ukrainian]. Rep. Natl. Acad. Sc. Ukr. 9, 20–24 (2005)
12. Kasyanov, P.O., Mel’nik, V.S.: Faedo–Galerkin method for differential-operator inclusions in
Banach spaces with maps of wλ -psevdomonotone type [in Ukrainian]. Proc. Inst. Math. NAS
Ukr. 2(1), 103–126 (2005)
13. Kasyanov, P.O., Mel’nik, V.S.: On properties of subdifferential mappings in Fréchet spaces.
Ukr. Math. J. 57(10), 1385–1394 (2005). doi:10.1007/s11253-006-0017-5
14. Kasyanov, P.O., Mel’nik, V.S., Toscano, L.: Method of approximation of evolutionary inclu-
sions and variational inequalities by stationary. Syst. Res. Inf. Technol. 4, 106–119 (2005)
15. Kasyanov, P.O., Mel’nik, V.S.: On the subdifferential maps in Banach and Fréchet spaces [in
Russian]. Rep. Natl. Acad. Sci. Ukr. 12, 23–27 (2005)
16. Kasyanov, P.O., Mel’nik, V.S., Piccirillo, A.M.: On Schauder basis in some Banach spaces.
Rep. Natl. Acad. Sci. Ukr. 4, 31–35 (2006)
17. Kasyanov, P.O., Mel’nik, V.S., Piccirillo, A.M.: On some theorems of compact embedding for
Banach and locally convex spaces. Rep. Natl. Acad. Sci. Ukr. 5, 34–40 (2006)
18. Kasyanov, P.O., Mel’nik, V.S., Toscano, S.: Periodic solutions for nonlinear evolution equations
with wλ0 -pseudomonotone maps. Nonlinear Oscil. 9(2), 181–206 (2006). doi:10.1007/s11072-
006-0037-y
19. Zadoyanchuk, N.V., Kasyanov, P.O.: On the solvability of differential-operator equations of II
order with non-coercive operators of wλ0 -psevdomonotonnoho type [in Ukrainian]. Rep. Natl.
Acad. Sci. Ukr. 12, 15–19 (2006)
20. Zadoyanchuk, N.V., Kasyanov, P.O.: Faedo–Galerkin method for nonlinear second-order evo-
lution equations with Volterra operators. Nonlinear Oscil. 10(2), 203–228 (2007). doi:10.1007/
s11072-007-0016-y
21. Kasyanov P.O., Mel’nik V.S.: Evolution inequalities with non-coercive maps of wλ0 -
pseudomonotone type. In: Proceedings of the XVth International Hutsulian Workshop Category
Theory and Applied Systems Analysis in Decision-Making Problems, 73–110 (2005)
22. Kasyanov, P.O., Mel’nik, V.S., Toscano, L.: The classes and the main properties of the multi-
valued wλ0 -pseudomonotone maps. Syst. Res. Inf. Technol. 3, 122–144 (2007)
23. Kasyanov, P.O.: On the solvability of a class of evolution variational inequalities with wλ0 -
pseudomonotone maps. Res. Bull. NTUU “KP” 5, 142–147 (2007)
24. Kasyanov, P.O., Mel’nik, V.S., Piccirillo, A.M.: On some approximations and main topological
descriptions for special classes of Frechet spaces with integrable derivatives. Syst. Res. Inf.
Technol. 4, 93–110 (2007)
25. Kasyanov, P.O., Mel’nik, V.S.: On the solvability of operator-differential inclusions and evo-
lution variational inequalities generated by maps of wλ0 -pseudomonotone type [in Russian].
Ukr. Math. Bull. 4(4), 535–581 (2007)
234 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

26. Kasyanov, P.O., Mel’nik, V.S., Toscano, L.: The multi-valued penalty method for evolution
variational inequalities with wλ0 -pseudomonotone multi-valued maps. Nonlinear Oscil. 10(4),
481–509 (2007). doi:10.1007/s11072-008-0006-8
27. Zadoyanchuk, N.V., Kas’yanov, P.O.: On the solvability of nonlinear evolution equations with
M-pseudomonotone noncoercive maps [in Ukrainian]. Bull. Taras Shevchenko Natl. Univ.
Kyiv. Math. Mech. 19–20, 7–12 (2008)
28. Kasyanov, P.O., Mel’nik, V.S., Valero, J.: On the method of approximation for evolutionary
inclusions of pseudomonotone type. Bull. Aust. Math. Soc. 77(1), 115–143 (2008). doi:10.
1017/S0004972708000130
29. Zadoyanchuk, N.V., Kasyanov, P.O.: On the solvability of differential-operator inclusions of
second order with non-coercive operators of wλ0 -pseudomonotone type [in Ukrainian]. Rep.
Natl. Acad. Sci. Ukr. 4, 19–24 (2008)
30. Kasyanov, P.O.: Dubinsky scheme for a class of evolution equations with maps of
pseudomonotone type in non-reflexive Banach spaces [in Ukrainian]. Rep. Natl. Acad. Sci.
Ukr. 5, 20–24 (2008)
31. Kasyanov, P.O.: On periodic solutions of evolution inclusions of first order with wλ0 -
pseudomonotone maps. Rep. Natl. Acad. Sci. Ukr. 6, 23–28 (2008)
32. Kasyanov, P.O., Mel’nik, V.S., Piccirillo, A.M.: Local subdifferentials and multivariational
inequalities in Banach and Frechet spaces. Opuscula Math. 3, 295–311 (2008)
33. Zadoyanchuk, N.V., Kasyanov, P.O.: On solvability of the nonlinear evolution equations of
second order with non-coercive wλ0 -pseudomonotone maps [in Ukrainian]. Res. Bull. NTUU
“KP”. 3, 142–149 (2008)
34. Kasyanov, P.O., Mel’nik, V.S., Piccirillo, A.M.: On some approximations and main topological
descriptions for special classes of Banach spaces with integrable derivatives. Methods Funct.
Anal. Topol. 14(3), 255–270 (2008)
35. Perestyuk, M.O., Kasyanov, P.O., Zadoyanchuk, N.V.: On Faedo–Galerkin method for evolu-
tion inclusions with wλ0 -pseudomonotone maps. Memoirs Differ. Equ. Math. Phys. 44, 105–
132 (2008)
36. Kasyanov, P.O., Mel’nik, V.S., Toscano, S., Zadoyanchuk, N.V.: Periodic solutions of evolu-
tionary equations in the class of nonreflexive Banach spaces [in Russian]. J. Autom. Inf. Sci.
40(9), 1–19 (2008)
37. Kasyanov, P.O., Mel’nik, V.S.: On some topological properties for special classes of Banach
spaces II. Syst. Res. Inf. Technol. 3, 88–100 (2008)
38. Kasyanov, P.O., Mel’nik, V.S.: Evolution inequalities with noncoercive wλ0 -pseudomonotone
Volterra-type mappings. Ukr. Math. J. 60(11), 1752–1777 (2008). doi:10.1007/s11253-009-
0167-3
39. Kapustyan, V.O., Kasyanov, P.O., Kohut, P.I.: On the solvability of one class of parameterized
operator inclusions. Ukr. Math. J. 60(12), 1901–1914 (2008). doi:10.1007/s11253-009-0179-
z
40. Perestyuk, M.O., Kasyanov, P.O., Zadoyanchuk, N.V.: On solvability of second order evolution
inclusions with Volterra type operators. Miskolc Math. Notes 9(2), 119–135 (2008)
41. Kapustyan, V.O., Kasyanov, P.O., Kogut, O.P.: Properties of solutions for class of parameterized
operator inclusions [in Ukrainian]. Res. Bull. NTUU “KP” 5, 129–136 (2008)
42. Kasyanov, P.O.: Periodic solutions for a class of differential-operator inclusions with the Sk-
type maps [in Ukrainian]. Res. Bull. NTUU “KP” 6, 144–148 (2008)
43. Kasyanov, P.O.: Faedo–Galerkin method for evolution inclusions with non-coercive wλ0 -
pseudomonotone maps [in Ukrainian]. Rep. Natl. Acad. Sci. Ukr. 1, 14–20 (2009)
44. Kasyanov, P.O.: On the solvability of one class of parameterized multivariational inequalities
[in Ukrainian]. Rep. Natl. Acad. Sci. Ukr. 2, 20–25 (2009)
45. Zadoyanchuk, N.V., Kasyanov, P.O.: Faedo-Galerkin method for second-order evolution inclu-
sions with wλ -pseudomonotone mappings. Ukr. Math. J. 61(2), 236–258 (2009). doi:10.1007/
s11253-009-0207-z
46. Kasyanov, P.O.: On weak solvability for class of evolution variational inequalities in infinite-
dimensional spaces [in Ukrainian]. Rep. Natl. Acad. Sci. Ukr. 3, 19–24 (2009)
References 235

47. Kasyanov, P.O., Mel’nik, V.S., Toscano, S.: Initial time value problem solutions for evolution
inclusions with type operators. Syst. Res. Inf. Technol. 1, 116–130 (2009)
48. Kasyanov, P.O.: Nonmonotone penalty method for class of multivariational inequalities with
maps of pseudomonotone type [in Ukrainian]. Res. Bull. NTUU “KP” 1, 139–142 (2009)
49. Zadoyanchuk, N.V., Kasyanov, P.O.: Singular-perturbation method for nonlinear second-order
evolution inclusions with Volterra operators. Nonlinear Oscil. 12(1), 27–43 (2009). doi:10.
1007/s11072-009-0057-5
50. Kasyanov, P.O., Zadoyanchuk, N.V., Yasinskiy, V.V.: Periodic solutions for a class of nonlinear
hyperbolic evolution equations. Cybern. Syst. Anal. 5, 118–128 (2009). doi:10.1007/s10559-
009-9149-6
51. Kasyanov, P.O.: Differential-operator inclusions and multivariational inequalities with
pseudomonotone mappings. Cybern. Syst. Anal. 2, 126–134 (2010). doi:10.1007/s10559-010-
9205-2
52. Zadoyanchuk, N.V., Kasyanov, P.O.: Analysis and control of second-order differential-operator
inclusions with +-coercive damping. Cybern. Syst. Anal. 2, 152–160 (2010). doi:10.1007/
s10559-010-9208-z
53. Zadoyanchuk, N.V., Kasyanov, P.O.: The research method for dynamic contact problems with
nonlinear damping [in Ukrainian]. Rep. Natl. Acad. Sci. Ukr. 5, 18–22 (2010)
54. Kasyanov, P.O., Mel’nik, V.S., Toscano, S.: Solutions of Cauchy and periodic problems for
evolution inclusions with multi-valued wλ0 -pseudomonotone maps. J. Differ. Equ. 249(6),
1258–1287 (2010). doi:10.1016/j.jde.2010.05.008
55. Zgurovsky, M.Z., Kasyanov, P.O., Valero, J.: Noncoercive evolution inclusions for Sk type
operators. Int. J. Bifurc. Chaos 20(9), 2823–2834 (2010). doi:10.1142/S0218127410027386
56. Zadoyanchuk, N.V., Kasyanov, P.O.: Properties for solutions of evolution inclusions of second
order with maps of pseudomonotone type. J. Comput. Appl. Math. 3(102), 63–78 (2010)
57. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Pullback attractors for a class of extremal solutions
of the 3D Navier–Stokes equations. J. Math. Anal. Appl. 373(2), 535–547 (2011). doi:10.1016/
j.jmaa.2010.07.040
58. Kasyanov, P.O.: Multi-valued dynamics of solutions of an autonomous differential-operator
inclusion with pseudomonotone nonlinearity. Cybern. Syst. Anal. 5, 150–163 (2011). doi:10.
1007/s10559-011-9359-6
59. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Long-time behaviour of solutions
for autonomous evolution hemivariational inequality with multidimensional “reaction-
displacemen” law. Abstr. Appl. Anal. 2012, Article ID 450984, 21 p (2012). doi:10.1155/
2012/450984
60. Zadoianchuk, N.V., Kasyanov, P.O.: Dynamics of solutions of a class of second-order
autonomous evolution inclusions. Cybern. Syst. Anal. 3, 111–126 (2012). doi:10.1007/s10559-
012-9421-z
61. Zgurovsky, M.Z., Kasyanov, P.O., Zadoianchuk, N.V.: Long-time behavior of solutions for qua-
silinear hyperbolic hemivariational inequalities with application to piezoelectricity problem.
Appl. Math. Lett. 25(10), 1569–1574 (2012). doi:10.1016/j.aml.2012.01.016
62. Kasyanov, P.O.: Multi-valued dynamics of solutions of autonomous operator differential equa-
tions with pseudomonotone nonlinearity. Math. Notes 92(1–2), 205–218 (2012). doi:10.1134/
S0001434612070231
63. Feinberg, E.A., Kasyanov, P.O., Zadoianchuk, N.V.: Average cost Markov decision processes
with weakly continuous transition probabilities. Math. Op. Res. 37(4), 591–607 (2012). doi:10.
1287/moor.1120.0555
64. Feinberg, E.A., Kasyanov, P.O., Zadoianchuk, N.V.: Berges theorem for noncompact image
sets. J. Math. Anal. Appl. 397(1), 255–259 (2013). doi:10.1016/j.jmaa.2013.12.011
65. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: A criterion for the existence of strong solutions
for the 3D NavierStokes equations. Appl. Math. Lett. 26(1), 15–17 (2013). doi:10.1016/j.aml.
2012.08.007
66. Feinberg, E.A., Kasyanov, P.O., Zgurovsky, M.Z.: Optimality conditions for total-cost partially
observable Markov decision processes. In: 52nd IEEE Conference on Decision and Control.
(2013). doi:10.1109/CDC.2013.6760790
236 9 Indirect Lyapunov Method for Autonomous Dynamical Systems

67. Gorban, N.V., Kapustyan, O.V., Kasyanov, P.O.: Uniform trajectory attractor for non-
autonomous reactiondiffusion equations with Carathéodorys nonlinearity. Nonlinear Anal. 98,
13–26 (2014). doi:10.1016/j.na.2013.12.004
68. Zgurovsky, M.Z., Kasyanov, P.O.: Multi-valued dynamics of solutions for autonomous operator
differential equations in strongest topologies. Contin. Distrib. Syst.: Theory Appl. Solid Mech.
Appl. 211, 149–162 (2014). doi:10.1007/978-3-319-03146-0_11
69. Feinberg, E.A., Kasyanov, P.O., Zgurovsky, M.Z.: Optimality conditions for partially observ-
able Markov decision processes. Contin. Distrib. Syst.: Theory Appl. Solid Mech. Appl. 211,
251–264 (2014). doi:10.1007/978-3-319-03146-0_18
70. Gorban, N.V., Kasyanov, P.O., Kapustyan, O.V., Palichuk, L.S.: On global attractors for
autonomous wave equation with discontinuous nonlinearity. Contin. Distrib. Syst.: Theory
Appl. Solid Mech. Appl. 211, 221–237 (2014). doi:10.1007/978-3-319-03146-0_16
71. Kasyanov, P.O., Toscano, L., Zadoianchuk, N.V.: Topological properties of strong solutions for
the 3D Navier-Stokes equations. Contin. Distrib. Syst.: Theory Appl. Solid Mech. Appl. 211,
181–187 (2014). doi:10.1007/978-3-319-03146-0_13
72. Gorban, N.V., Kasyanov, P.O.: On regularity of all weak solutions and their attractors for
reaction-diffusion inclusion in unbounded domain. Contin. Distrib. Syst.: Theory Appl. Solid
Mech. Appls. 211, 205–220 (2014). doi:10.1007/978-3-319-03146-0_15
73. Feinberg, E.A., Kasyanov, P.O., Voorneveld, M.: Berges maximum theorem for noncompact
image sets. J. Math. Anal. Appl. 413, 1040–1046 (2014). doi:10.1016/j.jmaa.2013.12.011
74. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Structure and regularity of the global attractor
of a reaction-diffusion equation with non-smooth nonlinear term. Discret. Contin. Dyn. Syst.
34(10), 4155–4182 (2014). doi:10.3934/dcds.2014.34.4155
75. Kasyanov, P.O., Paliichuk, L.S., Tkachuk, A.M.: Method of multi-valued operator semigroup
to investigate the long-term forecasts foe controlled piezoelectric fields [in Russian]. Cheby-
shevskii Sb. 15(2), 21–32 (2014)
76. Zgurovsky, M.Z., Kasyanov, P.O., Paliichuk, L.S.: Automatic feedback control for one class
of contact piezoelectric problems. Syst. Res. Inf. Technol. 1, 56–68 (2014)
77. Feinberg, E.A., Kasyanov, P.O., Zadoianchuk, N.V.: Fatou’s lemma for weakly converging
probabilities. Theory Probab. Appl. 58(4), 683–689 (2014). doi:10.1137/S0040585X97986850
78. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Regular solutions and global attractors for reaction-
diffusion systems without uniqueness. Commun. Pure Appl. Anal. 13(5), 1891–1906 (2014).
doi:10.3934/cpaa.2014.13.1891
79. Feinberg, E.A., Kasyanov, P.O., Zgurovsky, M.Z.: Convergence of probability measures and
Markov decision models with incomplete information. Proc. Steklov Inst. Math. 287, 103–124
(2014). doi:10.1134/S0081543814080069
80. Feinberg, E.A., Kasyanov, P.O., Zgurovsky, M.Z.: Convergence of value iterations for total-cost
MDPs and POMDPs with general state and action sets. In: Proceedings of IEEE Symposium on
Adaptive Dynamic Programming and Reinforcement Learning, pp. 1–8 (2014). doi:10.1109/
ADPRL.2014.7010613
81. Zgurovsky, M.Z., Kasyanov, P.O.: Evolution inclusions in nonsmooth systems with applications
for Earth data processing: Uniform trajectory attractors for nonautonomous evolution inclusions
solutions with pointwise pseudomonotone mappings. Adv. Global Optim. Springer Proc. Math.
Stat. 95, 283–294 (2015). doi:10.1007/978-3-319-08377-3_28
82. Kasyanov, P.O., Paliichuk, L.S.: Trajectory behavior for class of controlled piezoelectric fields
with nonmonotone potential [in Ukrainian]. Res. Bull. NTUU “KPI”. 2, 21–27 (2014)
83. Gluzman, M.O., Gorban, N.V., Kasyanov, P.O.: Lyapunov type functions for classes of
autonomous parabolic feedback control problems and applications. Appl. Math. Lett. 39, 19–21
(2015). doi:10.1016/j.aml.2014.08.006
84. Feinberg, E.A., Kasyanov, P.O.: Continuity of minima: local results. Set-Valued Var. Anal. 23,
485–499 (2015). doi:10.1007/s11228-015-0318-7
85. Kapustyan, O.V., Kasyanov, P.O., Valero, J.: Structure of the global attractor for weak solutions
of reaction-diffusion equation. Appl. Math. Inf. Sci. 9(5), 2257–2264 (2015). doi:10.12785/
amis/090506
References 237

86. Kasyanov, P.O.: On the existence of strongly continuous physical solutions for classes of
autonomous evolutionary variational inequalities. Cybern. Syst. Anal. 51(4), 574–582 (2015).
doi:10.1007/s10559-015-9748-3
87. Zgurovsky, M.Z., Kasyanov, P.O.: Uniform trajectory attractors for nonautonomous dissipative
dynamical systems. Contin. Distrib. Syst. II: Theory Appl. Stud. Syst. Decis. Control. 30, 221–
232 (2015). doi:10.1007/978-3-319-19075-4_13
88. Gluzman, M.O., Gorban, N.V., Kasyanov, P.O.: Lyapunov functions for differential inclusions
and applications in physics, biology, and climatology. Contins Distrib. Syst. II: Theory Appl.
Stud. Syst. Decis. Control. 30, 233–243 (2015). doi:10.1007/978-3-319-19075-4_14
89. Zgurovsky, M.Z., Kasyanov, P.O., Paliichuk, L.S., Tkachuk, A.M.: Dynamics of solutions for
controlled piezoelectric fields with multi-valued “reaction-displacement” law. Contin. Distrib.
Syst. II: Theory Appl. Stud. Syst. Decis. Control. 30, 267–276 (2015). doi:10.1007/978-3-
319-19075-4_16
90. Gluzman, M.O., Gorban, N.V., Kasyanov, P.O.: Lyapunov functions for weak solutions of
reaction-diffusion equations with discontinuous interaction functions and its applications.
Nonauton. Dyn. Syst. 2(1), 1–11 (2015). doi:10.1515/msds-2015-0001
91. Kapustyan, O.V., Kasyanov, P.O., Valero, J., Giménez, A., Amigó, J.M.: Convergence of numer-
ical approximations for a non-Newtonian model of suspensions. Int. J. Bifurc. Chaos. 25(14),
1540022-1–1540022-24 (2015). doi:10.1142/S0218127415400222
92. Feinberg, E.A., Kasyanov, P.O., Zgurovsky, M.Z.: Partially observable total-cost Markov deci-
sion processes with weakly continuous transition probabilities. Math. Op. Res. 41(2), 656–681
(2016). doi:10.1287/moor.2015.0746
93. Valero, J., Giménez, A., Kapustyan, O.V., Kasyanov, P.O., Amigó, J.M.: Convergence of equi-
libria for numerical approximations of a suspension model. Computers and Mathematics with
Applications. 72(4), 856–878 (2016). doi:10.1016/j.camwa.2016.05.034
94. Zgurovsky, M.Z., Kasyanov, P.O.: Uniform global attractors for nonautonomous evolution
inclusions. Adv. Dyn. Syst. Control. Stud. Syst. Decis. Control. 69, 35–42 (2016). doi:10.
1007/978-3-319-40673-2_3
95. Gorban, N.V., Gluzman, M.O., Kasyanov, P.O., Tkachuk, A.M.: Long-time behavior of state
functions for Badyko models. Adv. Dyn. Syst. Control. Stud. Syst. Decis. Control. 69, 351–359
(2016). doi:10.1007/978-3-319-40673-2_18
96. Feinberg, E.A., Kasyanov, P.O., Zgurovsky, M.Z.: Uniform Fatous lemma. J. Math. Anal. Appl.
(2016). doi:10.1016/j.jmaa.2016.06.044
Index

A second order, 145

Admissible triple, 72 3D Navier-Stokes equations, 72
Approximate problem, 22
Artificial control method, 50, 72
Attractor E
global, 213, 219, 228 Energy quality, 10, 73, 120, 121, 153
trajectory, 214, 229 Evolutional multivariational inequality, 33
uniform global, 9, 10, 162 Evolution triple (Gelfand triple), 50
uniform trajectory, 3, 5

F
B Family of m-processes, 10
Banach-Alaoglu theorem, 6 strict, 11
Budyko model, 118 uniformly asymptotically compact, 12
FitzHugh-Nagumo system, 116
Function
C Lyapunov, 218
Climate energy balance model, 62, 118, 152 translation bounded, 90
Cm(S;X), xxiv translation-compact, 5, 98, 230
translation uniform integrable, 5, 98

D
Differential-operator equation, 12, 125 G
first order, 125 Gagliardo–Nirenberg inequality, 54
nonlinear parabolic, 9, 150, 151
reaction-diffusion, 1, 9, 89
second order, 31, 139, 215 H
Differential-operator inclusion, 3, 24, 57, Hausdorff semidistance, 164
125, 171 Heat conduction equation, 60, 171
first order, 125
reaction-diffusion, 55, 57, 64, 111
I
Indirect Lyapunov method, 211

© Springer International Publishing AG 2018 239

M.Z. Zgurovsky and P.O. Kasyanov, Qualitative and Quantitative Analysis
of Nonlinear Systems, Studies in Systems, Decision and Control 111,
DOI 10.1007/978-3-319-59840-6
240 Index

Interpolation family, xiv Sk -property, 16, 23

Interpolation pair, xiv Solution
extremal, 79
Leray-Hopf, 69, 71
L physical, 43
Leray-Hopf property, 72 regular, 60
strong, 42
weak, 3, 10, 24, 40, 61, 78, 92, 94, 120,
M 126, 134, 140, 146, 214
Model of combustion in porous media, 116, Special basis, 77
117, 220 (S)-property, 127
Model of conduction of electrical impulses Strict multi-valued map
in nerve, 61, 106, 116, 117 bounded, 7, 16
Multi-valued map, 14, 17 coercive, 7, 14
λ0 -pseudomonotone, 23 demi-closed, 7, 14
weakly closed, 15, 25 of the Volterra type, 15, 16
Multivalued semiflow, 165 pseudomonotone, 23
asymptotically compact, 216, 224
strict, 213, 214
T
Trajectory
P
complete, 218
Parabolic feedback control problem, 60
of m-process, 10
Penalty method, 42
of m-process complete, 10
Phase space, 31
stationary, 218
Pointwise pseudomonotonicity, 134, 172
Translation semigroup, 227

R
Reflexivity criterium, xii U
Regularity of solutions, 60 United trajectory space, 5, 27, 94, 163
extended, 5, 94, 163

S
Set V
translation-compact, 232 Viscoelastic problems with nonlinear
translation semi-invariant, 94 “Reaction-Displacement” law, 154,
uniformly attracting, 5, 11, 30 221