i

CONTRIBUTIONS TO THE CONTROL THEORY OF SOME PARTIAL

FUNCTIONAL INTEGRODIFFERENTIAL EQUATIONS IN BANACH
SPACES

A PHD THESIS PRESENTED TO THE DEPARTMENT OF

PURE AND APPLIED MATHEMATICS (PAM)
AFRICAN UNIVERSITY OF SCIENCE AND TECHNOLOGY, ABUJA,
NIGERIA.

BY
PATRICE NDAMBOMVE

2016
 CONTRIBUTIONS TO THE CONTROL
THEORY OF SOME PARTIAL
FUNCTIONAL INTEGRODIFFERENTIAL
EQUATIONS IN BANACH SPACES

Patrice Ndambomve
B.Sc.(Hons.)(University of Buea, Cameroon),
M.Sc. (AUST, Abuja, Nigeria.)
I.D. 70064
 i

CONTRIBUTIONS TO THE CONTROL THEORY OF SOME PARTIAL

FUNCTIONAL INTEGRODIFFERENTIAL EQUATIONS IN BANACH
SPACES

Dissertation
Presented in Partial Fulfilment of the
Requirements for the Degree of
Doctor of Philosophy
in Pure and Applied Mathematics
of the
African University of Science and Technology
Abuja, Nigeria.

By
Patrice Ndambomve
B.Sc.(Hons.)(University of Buea, Cameroon),
M.Sc. (AUST, Abuja, Nigeria).
I.D. 70064

African University of Science and Technology, Abuja, Nigeria.

February, 2016.
 ii

AFRICAN UNIVERSITY OF SCIENCE AND TECHNOLOGY, ABUJA,

NIGERIA

www.aust.edu.ng
P.M.B 681, Garki, Abuja F.C.T, Nigeria.

CERTIFICATE OF APPROVAL
————————————————–

Ph.D. THESIS
—————————————

This is to certify that the Ph.D. thesis of

PATRICE NDAMBOMVE,
has been approved by the Examining Committee for the thesis requirements
for the award of the degree of Doctor of Philosophy Degree in Mathematics.

Thesis Committee: ——————————————————–

Thesis Supervisor: Prof. Khalil Ezzinbi
Cadi Ayyad University, Marrakech, Morocco.

——————————————————
Member: Prof. Mostafa Adimy
INRIA, Grenoble, France

——————————————————
Member: Prof. C. E. Chidume
African University of Science and Technology, Abuja, Nigeria.
 Dedication

To My Lord and Savior Jesus Christ for His Unmerited Favour without
which I would not have been here and for giving me the wisdom, inspiration
and strength to start and complete this thesis.
To my Family and to my fiancée, Tiako Fani Michele Doriane,
whose love, care, vulnerability, wisdom, strength, unceasing
encouragements, motivation, and support have inspired me to be the best I
can be.

iii
 Acknowledgements

Doing Mathematics may be fun. But, it takes more than just a piece of paper,
a pen, and the ability to recognize patterns, and put together implications and
equivalences to write a logical, sound and correct mathematical proof. Unfor-
tunately, most often, a substantial part of what one may need has neither a
material nor a Mathematical equivalent.
This makes me therefore, very indebted to those who gave me the strength,
the courage and the motivation to withstand the social, emotional and psy-
chological adversities which have been persistent companions over the years.

First and formost, all the glory, thanks, praise, adoration, appreciation
from the bottom of my heart be given to the Almighty God, the creator of
the universe, my strength in weakness, my comfort in sorrow, my Ebenezer,
who lifted me from grass to grace, made ways for me where there seemed to
be no way and brought to perfect completion of this thesis and my Ph.D.
programme using human instrument.

I would like to express my deepest gratitude and appreciation to my the-

sis supervisor Professor Khalil Ezzinbi of Cadi Ayyad University, Marrakech,
Morocco; who inspired me and gave me the motivation and impetus to start
this work, and guided me all the way; for his patience and indeed all his
invaluable support throughout the years of my Ph.D. programme; for not
hiding from me, any of his academic skills and talents while I worked with
him; for finding ways and opportunities for us to meet and work. It was a
great privilege to work with such a great scholar. I am very grateful to him
for taking me through these years of research. May the good Lord reward
him abundantly.

iv
acknowledgements v

My deepest gratitude to Professor Charles Ejike Chidume, FAS, for his

management of the Mathematics Institute in a diligent and patient manner;
for the fatherly disposition and thorough academic guidance; for squizing the
best out of us; for the motivation and for sharing with us his vast experience
and specially for instilling in us a culture of meticulousness and zero com-
promise to mathematical rigour.

My sincere gratitude to Professor Wole Soboyejo, FAS, for his numerous

advices, encouragements and efforts put in place for the success of our Ph.D.
programme in AUST, during his tenure as President of AUST.
I wish to thank the Board, Acting President, faculty and staff of AUST
for the scholarship and all other forms of support and encouragements they
gave me which enabled me undergo the Ph.D. programme. I would also like
to thank specially the African Development Bank (AfDB) for the postgrad-
uate scholarship award, the African Capacity Building Foundation (ACBF)
for paying our stippends and financing the publications of our research pa-
pers. Without their sponsorship this thesis would not have been possible.

I also wish to express gratitude to Prof. Ngalla Djitte for the rich aca-
demic interaction I enjoyed from him in the course of my doctoral study.

I thank Dr. Guy Degla for his numerous contributions towards my Ph.D.
studies; for introducing me to Evolution Equations during my M.Sc pro-
gramme and encouraging me to carry it over to Ph.D. level; for the numerous
hours of tutoring and discussions.

I would like to register my gratitude to Dr. Gokhan Koyunlu and the

entire academic and non academic staff of the Faculties of Engineering and
Natural and Applied Sciences, Nigerian Turkish Nile University, Abuja, Nige-
ria, for the various forms of support during my Ph.D. studies.
My gratitude Dr Ma’aruf S. Minjibir and Dr Ezeora, Jeremiah Nkwegu for
always devoting time for discussions and interactions during this programme.

I would like to mention my thanks to my friends and colleagues Ab-

dulmalik Usman Bello, Mmaduabuchi Ejikeme Okpala, and Opara Micheal
Uchechukwu and the M.Sc students in the Department of Pure and Applied
Mathematics, AUST, for the every day struggle we have been through and
the great stay I enjoyed with them.

Many thanks to my friends and colleagues Adedoyin Adegoke, Arrey-

acknowledgements vi

tambe Tabot, Ignace Djitog, Yoro Diouf, Koffi Ampaw, Joseph Asare, Vitalis
Anye, Aborisade Opeyemi, Azeko Salifu, Emeka Ani, Bruno Dangdobessi, for
the friendship and the great time I have had with you folks while at AUST.
Being with you folks has been a wonderful life-changing experience.

My sincere thanks to Miss Amaka Udigwe, the Administrative Assistant,

Mathematics Institute, AUST, for the various forms of help she rendered
me in the course of my Ph.D. studies. Gratitude to Mrs Nsima Joseph and
Miss Bolade Igbagbo, the students affairs administrative staff, for the various
forms of assistance they rendered me throughout my Ph.D. studies.

My gratitude to Dr. Shu Felix Che, Dr. Nkemzi Boniface, Dr. Nana
Cyrille, and to my lecturers at Universities of Bamenda and Buea, who in
one way or the other inspired me to further studies in Mathematics.

Many thanks to all the staff of Marlima Catering Services, especially to

Gloria and Tina for the great services I benefited from them during my Ph.D.
programme. Many thanks to all the staff of the cleaning services, for helping
me clean my room each time I left it in a mess.

Very special thanks and gratitude to those who close or far said an en-
couraging word to me during my times of despair. Especially my mother
Feteko Margueritte for her love, care, prayers and sacrifices.
My gratitude to my elder brother Dr. Jean-Claude Pedjeu, whose passion for
mathematics inspired and motivated me to pursue a carrier in mathematics
right from Secondary school, for his prayers and encouragements.
Sincere and deep gratitude to my elder brother and his wife Mr and Mrs
Mbemboung who took me under their care right from childhood and taught
me how to love and appreciate learning, for their supports and encourage-
ments over the years.
Many thanks to my sisters, brothers, nieces and nephews for their prayers,
love and support throughout my life. Without them, I would not have been
able to get this far.
Thanks to all those I fail to mention here, who assisted me in any way over
these years.

Patrice Ndambomve
 Abstract

This thesis is a contribution to Control Theory of some Partial Functional

Integrodifferential Equations in Banach spaces. It is made up of two parts:
controllability and existence of optimal controls. In the first part, we consider
the dynamical control systems given by the following models that arise in the
analysis of heat conduction in materials with memory, and viscoelasticity, and
take the form of a:

• Partial functional integrodifferential equation subject to a nonlocal ini-

tial condition in a Banach space (X, k · k) :
 Z t
0
x (t) = Ax(t) + B(t − s)x(s)ds + f (t, x(t)) + Cu(t)




 0
for t ∈ I = [0, b],




x(0) = x0 + g(x),

(0.0.1)
where x0 ∈ X, g : C(I, X) → X and f : I × X → X are functions
satisfying some conditions; A :
D(A) → X is the infinitesimal gen-
erator of a C0 -semigroup T (t) t≥0 on X; for t ≥ 0, B(t) is a closed
linear operator with domain D(B(t)) ⊃ D(A). The control u belongs
to L2 (I, U ) which is a Banach space of admissible controls, where U is
a Banach space.

• Partial functional integrodifferential equation with finite delay in a Ba-

vii
Abstract viii

nach space (X, k · k) :

 Z t
0
x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) + Cu(t) ,




 0
for t ∈ I = [0, b], (0.0.2)




x0 = ϕ ∈ C = C([−r, 0]; X),


where f : I × C → X is a function satisfying some conditions;
A :

D(A) → X is the infinitesimal generator of a C0 -semigroup T (t) t≥0
on X; for t ≥ 0, B(t) is a closed linear operator with domain D(B(t)) ⊃
D(A). The control u belongs to L2 (I, U ) which is a Banach space of
admissible controls, where U is a Banach space, and xt denotes the
history function of C of the state from the time t − r up to the present
time t, and is defined by xt (θ) = x(t + θ) for −r ≤ θ ≤ 0.

• Partial functional integrodifferential equation with infinite delay in a

Banach space (X, k · k) :
 Z t
0
x (t) = Ax(t) + γ(t − s)x(s)ds + f (t, xt ) + Cu(t),




 0
for t ∈ I = [0, b] (0.0.3)




x0 = ϕ ∈ B,


where
A : D(A) → X is the infinitesimal generator of a C0 -semigroup

T (t) t≥0 on a Banach space X; for t ≥ 0, γ(t) is a closed linear op-
erator with domain D(γ(t)) ⊃ D(A). The control u takes values from
another Banach space U . The operator C(t) belongs to L(U, X) which
is the Banach space of bounded linear operators from U into X, and the
phase space B is a linear space of functions mapping ]−∞, 0] into X sat-
isfying axioms which will be described later, for every t ≥ 0, xt denotes
the history function of B defined by xt (θ) = x(t + θ) for − ∞ ≤ θ ≤ 0,
f : I × B → X is a continuous function satisfying some conditions.

We give sufficient conditions that ensure the controllability of the systems

without assuming the compactness of the semigroup, by supposing that their
linear homogeneous and undelayed parts admit a resolvent operator in the
sense of Grimmer and by making use of the Hausdorff measure of noncom-
pactness.
In the second part, we consider equations (0.0.1), (0.0.2) and (0.0.3), in the
Abstract ix

case where the operator C = C(t) (time dependent), the function g ≡= 0,

the Banach spaces X and U are separable and reflexive. Using techniques of
convex optimization, a priori estimation, and applying Balder’s Theorem, we
establish the existence of optimal controls for the following Lagrange optimal
control problem associated to each of the equations:

 Find a control u0 ∈ Uad such that



(LP)
J (u0 ) ≤ J (u) for all u ∈ Uad ,


where Z T  
J (u) := L t, xut , u
x (t), u(t) dt,
0
L is some functional, xu denotes the mild solution corresponding to the con-
trol u ∈ Uad , and Uad denotes the set of admissible controls.
Publications arising from the thesis and other peer-review
publications

[A] Published/Accepted Papers from the thesis

1) Khalil Ezzinbi , Guy Degla and Patrice Ndambomve, Controllability
for some Partial Functional Integrodifferential Equations with Nonlocal
Conditions in Banach Spaces, Discussiones Mathematicae, Differ-
ential Inclusions, Control and Optimization, Vol. 35., 2015,
no. 1, 1-22.
2) Khalil Ezzinbi and Patrice Ndambomve, Solvability and optimal con-
trols for some Partial Functional Integrodifferential Equations with fi-
nite delay, (Accepted to Appear: American Journal of Model-
ing and Optimization).
3) Khalil Ezzinbi and Patrice Ndambomve, Controllability for some
Partial Functional Integrodifferential Equations in Banach Spaces, (Ac-
cepted to Appear: American Journal of Mathematical Anal-
ysis).

4) Khalil Ezzinbi and Patrice Ndambomve, Partial Functional Inte-

grodifferential Equations and Optimal Controls in Banach Spaces, (Ac-
cepted to Appear: Communications in Optimization Theory).
5) Khalil Ezzinbi and Patrice Ndambomve, Solvability and optimal con-
trols for some Partial Functional Integrodifferential Equations with in-
finite delay in Banach Spaces, (Submitted: Optimal Control: Ap-
plications and Methods,(John Wiley and Sons)).

x
Abstract xi

6) Khalil Ezzinbi and Patrice Ndambomve, On the Controllability of

some Partial Functional Integrodifferential Equations with infinite de-
lay in Banach Spaces, (Accepted to Appear: Advances in Fixed
Point Theory ).

[B] Other Published/Accepted peer-review publications

1) C. E. Chidume, P. Ndambomve, A. U. Bello, M. E. Okpala, The

multiple-sets split equality fixed point problem for countable families of
multi-valued demi-contractive mappings, International Journal of
Mathematical Analysis, Vol. 9, 2015, no. 10, 453 - 469.

2) C. E. Chidume, P. Ndambomve, A. U. Bello, The split equality fixed

point problem for demi-contractive mappings, Journal of Nonlinear
Analysis and Optimization, Theory & Applications, Vol. 6,
No. 1, (2015), 61-69.

3) C. E. Chidume, A. U. Bello, P. Ndambomve, Strong and ∆-convergence

theorems for common fixed points of a finite family of multi-valued
demicontractive mappings in CAT(0) spaces, Abstract and Applied
Analysis, Volume 2014, Article ID 805168.

4) C.E. Chidume, A.U. Bello, M.E. Okpala, P. Ndambomve, Strong

Convergence Theorem for Fixed Poins of Nearly Uniformly L-Lipschitzian
Asymptotically generalized Hemicontractive Mappings, International
Journal of Mathematical Analysis, Vol. 9, 2015, no. 52, 2555-
2569.

5) C.E. Chidume, M.E. Okpala, A.U. Bello and P. Ndambomve, Con-

vergence theorems for finite family of a general class of multi-valued
strictly pseudo-contractive mappings, Fixed Point Theory and Ap-
plications, (2015) 2015:119, DOI 10.1186/s13663-015-0365-7.
 Contents

Dedication iii

Acknowledgements iv

Abstract vii

1 Introduction 1
1.1 General Introduction . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Nonlocal Differential Equations . . . . . . . . . . . . . 2
1.1.2 Delay Differential Equations . . . . . . . . . . . . . . . 3
1.2 Partial Functional Integrodifferential Equations . . . . . . . . 6
1.2.1 A Model in Viscoelasticity . . . . . . . . . . . . . . . . 6
1.2.2 A Model in Heat Conduction in Materials with Memory 8
1.3 Controllability of Dynamical Systems . . . . . . . . . . . . . . 10
1.4 Optimal Control of Dynamical Systems . . . . . . . . . . . . . 16
1.5 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.1 Measures of Noncompactness . . . . . . . . . . . . . . 20
1.5.2 Fixed Point Theory . . . . . . . . . . . . . . . . . . . . 22
1.5.3 Semigroup Theory . . . . . . . . . . . . . . . . . . . . 22
1.5.4 Resolvent Operator for Integral Equations . . . . . . . 23
1.6 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . 23

2 Preliminaries 25
2.1 Semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Resolvent Operator for Integral Equations . . . . . . . . . . . 27
2.3 Measure of Noncompactness . . . . . . . . . . . . . . . . . . . 32

xii
Abstract xiii

2.4 The Mönch Fixed Point Theorem and Balder’s Theorem . . . 35

I Controllability of some Partial Functional Integrod-

ifferential Equations in Banach Spaces 37
3 Controllability for some Partial Functional Integrodifferen-
tial Equations with Nonlocal Conditions in Banach Spaces 38
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Example of Application . . . . . . . . . . . . . . . . . . . . . . 47

4 Controllability for some Partial Functional Integrodifferen-

tial Equations with Finite Delay in Banach Spaces 50
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.2 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3 Controllability result . . . . . . . . . . . . . . . . . . . . . . . 53
4.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

5 Controllability Results for some Partial Functional Integrod-

ifferential Equations with Infinite Delay in Banach Spaces 62
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3 Controllability result . . . . . . . . . . . . . . . . . . . . . . . 66
5.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

II Optimal Controls of some Partial Functional In-

tegrodifferential Equations in Banach Spaces 76
6 Solvability and Optimal Control for some Partial Functional
Integrodifferential Equations with Finite Delay 77
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.2 Existence of mild solutions for equation (6.1.1) . . . . . . . . . 79
6.3 Continuous Dependence . . . . . . . . . . . . . . . . . . . . . 82
6.4 Existence of the Optimal Controls . . . . . . . . . . . . . . . . 86
6.5 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

7 On the Solvability and Optimal Control of some Partial Func-

tional Integrodifferential Equations with Infinite Delay in Ba-
nach Spaces 92
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Table of Contents xiv

7.2 Existence of mild solutions for equation (7.1.1) . . . . . . . . . 94

7.3 Continuous Dependence and Existence of the Optimal Control 99
7.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

8 Solvability and Optimal Controls for some Partial Functional

Integrodifferential Equations with Classical Initial Conditions
in Banach Spaces 109
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.2 Existence of mild solutions for equation (8.1.1) . . . . . . . . . 111
8.3 Continuous Dependence and Existence of the Optimal Control 114
8.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Conclusion and Perspectives 123

Bibliography 124
 CHAPTER 1

Introduction

1.1 General Introduction

In various fields of science and engineering such as Electronics, Fluid Dy-
namics, Physical Sciences, etc..., many problems that are related to linear
viscoelasticity, nonlinear elasticity and Newtonian or non- Newtonian fluid
mechanics have mathematical models which are described by differential or
integral equations or integrodifferential equations which have received con-
siderable attention during the last decades. Control Theory arises in many
modern applications in engineering and environmental sciences [2]. It is one
of the most interdisciplinary research areas [22][63] and its empirical concept
for technology goes back to antiquity with the works of Archimede, Philon,
etc..., [73]. A control system is a dynamical system on which one can act
by the use of suitable parameters (i.e., the controls) in order to achieve a
desired behavior or state of the system. Control systems are usually mod-
eled by mathematical formalism involving mainly ordinary differential equa-
tions, partial differential equations or functional differential equations. In
condensed expression, they often take the form of differential equation :

x0 (t) = F (t, x(t), u(t)) for t ≥ 0,

where x is the state and u is the control. While studying a control system, two
most common problems that appear are the controllability and the optimal
controls problems. The controllability problem consists in checking the pos-
sibility of steering the control system from an initial state (initial condition)

1
Introduction 2

to a desired terminal one (boundary condition), by an appropriate choice of

the control u, while the optimal control problem consists in finding the input
function ( the control or the command) so as to optimize (maximize or mini-
mize) the objective function. Control Theory of integrodifferential equations
with classical initial conditions and with delays, have received considerable
attention by researchers during the last decades.
This thesis is a contribution to Control Theory of Partial Functional Inte-
grodifferential equations in Banach spaces. It is made up of two parts:

• Part I: Controllability results for some partial functional integrodif-

ferential equations in Banach spaces.

• Part II: Optimal controls of some partial functional integrodifferential

equations in Banach spaces.

It lies at the interface between Nonlinear Functional Analysis, Optimization

Theory and Dynamical Systems. In the first part, we establish the controlla-
bility for some partial functional integrodifferential equations, with nonlocal
initial conditions, with finite delay and then with infinite delay. The sec-
ond part deals with the solvability and the existence of optimal controls
for these partial functional integrodifferential equations, with Cauchy ini-
tial conditions, with finite delay and then with infinite delay. We use fixed
point techniques to solve the controllability problem and convex optimization
techniques to solve the optimal control problems.

1.1.1 Nonlocal Differential Equations

Many problems arising in engineering and life sciences are modeled mathe-
matically by differential equations. Differential equations are one of the most
powerful and frequently used tools in mathematical modeling. Depending on
the nature of the problem, these equations may take various forms like or-
dinary differential equations, partial differential equations or functional dif-
ferential equations. In condensed expression, they often take the following
form:
x0 (t) = F (t, x(t)) for t ≥ 0,
where x is the state. Most often, these problems are subject to some initial
conditions. The classical initial condition is that referred to as the Cauchy
initial condition, given by x(0) = x0 , where x0 is some initial state of the
system at time t = 0. However, in many real world contexts such as En-
gineering, Environmental sciences, Demography, etc..., nonlocal constraints
(such as isoperimetric or energy condition, multipoint boundary condition
Introduction 3

and flux boundary condition) appear and have received considerable atten-
tion during the last decades, cf. [14] and [15]. They usually take the following
form: x(0) = x0 + g(x), where g is a function satisfying some conditions, and
x is the state or a solution of the differential equation in question. Observe
that the initial condition in this case depends on the solution of the system.
So, the concept of nonlocal initial condition not only extends that of Cauchy
initial condition, but also turns out to have better effects in applications as
it may take into account future measurements over a certain period after the
initial time t equals 0.

1.1.2 Delay Differential Equations

In the mathematical description of a great number of physical phenomena,
one usually suppose that the evolution of the system depends only on its
current state. However, there are situations where the evolution of a process
depends not only on its current state, but also on past states of the system.
Such phenomena arise in many areas, in particular, in population dynamics.
Amongst the mathematical models that can describe such situations are de-
lay differential equations whose delays can be of neutral type. One then has
equations whose temporal terms are, in a general way, nonlocal terms, involv-
ing values of the state at past, discrete or distributed times. These nonlocal
terms are either of order 0 (delay equations) or of order 1 (neutral equations).
In genera1, the delays appear because of the necessary time for the system
to respond to certain evolution, or because a certain threshold (limit value)
must be attained before the system can be activated. Delay differential equa-
tions lie at the interface between ordinary and partial differential equations.
The difference with ordinary differential equations is that the initial data are
themselves functions. This requires more elaborate mathematical study than
for ordinary differential equations, the nature of the delay (discrete, contin-
uous, infinite, state dependent, · · · ) potentially complicating it.

Example.([5]) Let N (t) denote the density of adults at time t, in a biological

population composed of adult and juvenile individuals. Assume that the length
of the juvenile period is exactly r units of time for each individual. Assume
that adults produce offspring at a per capita rate λ and that their probability
per unit of time of dying is µ. Assume that a newborn survives the juvenile
period with probability ν and put t = λν. Then the dynamics of N can be
described by the differential equation
dN
(t) = −µN (t) + kN (t − r)
dt
Introduction 4

which involves a nonlocal term, kN (t − r) meaning that newborns become

adults with some delay. So the time variation of the population density N
involves the current as well as the past values of N . Such equations are called
delay equations.

In finite dimension, a complete theory has been developed for delay differ-
ential equations, while the theory in infinite dimension is far from being
complete. Such equations in infinite dimension are represented by differen-
tial equations which are nonlocal in time: knowing the solution at a given
time requires knowing it on a time interval whose lenght is equal to the de-
lay. These types of equations have lots of applications in physics, chemistry,
biology, population dynamics, · · · .
Equations with finite delay genera1ly take the following abstract form:

for t ≥ 0
 0
 x (t) = Ax(t) + F (t, xt )
(1.1.1)
x0 = ϕ ∈ C = C([−r, 0]; X),


where F : R+ × C → X is a function satisfying some conditions; A
: D(A) ⊆

X → X is the infinitesimal generator of a C0 -semigroup T (t) t≥0 on a
Banach space X; for t ≥ 0; C([−r, 0], X) (the phase space), denotes the
Banach space of continuous functions x : [−r, 0] → X with supremum norm
kxk∞ = supt∈I kx(t)kX , xt denotes the history function of C of the state from
the time t − r up to the present time t, and is defined by xt (θ) = x(t + θ)
for −r ≤ θ ≤ 0. When r = ∞, we say that the delay is infinite, and this
class of equations englobes equations known in mechanics, namely Volterra
equations. When A = 0, and X = Rn , we have delay differential equations in
finite dimension. For this type of equations, the solution operator is compact,
and this property has made it possible to develop a complete theory for delay
differential equations in finite dimension. In infinite dimension, this property
is no longer satisfied, and this lack of regularity requires the development of
new functional analysis tools to tackle problems related to the qualitative
aspect of the solutions.
In the literature devoted to equations with finite delay, the phase space is
the space of continuous functions on [−r, 0], for some r > 0, endowed with
the uniform norm topology. But when the delay is unbounded (i.e., infinite),
the phase space denoted by B is a linear space of functions mapping ] − ∞, 0]
into X satisfying some axioms. The selection of the phase space B plays
an important role in both qualitative and quantitative theories. A usual
choice is a normed space satisfying the following suitable axioms, which was
introduced by Hale and Kato [82]:
Introduction 5

(B, k · kB ) will be a normed linear space of functions mapping ] − ∞, 0] into

X and satisfying the following axioms:
(A1 ) There exist a positive constant H and functions K : R+ → R+ con-
tinuous and M : R+ → R+ locally bounded, such that for a > 0, if
x : ] − ∞, a] → X is continuous on [0, a] and x0 ∈ B, then for every
t ∈ [0, a], the following conditions hold:
(i) xt ∈ B,
(ii) kx(t)k ≤ Hkxt kB , which is equivalent to kϕ(0)k ≤ HkϕkB for every
ϕ ∈ B,
(iii) kxt kB ≤ K(t) sup kx(s)k + M (t)kx0 kB .
0≤s≤t

(A2 ) For the function x in (A1 ), t → xt is a B-valued continuous function

for t ∈ [0, a].
(A3 ) The space B is complete.
Example [83] Let the spaces
BC the space of bounded continuous functions defined from (−∞, 0] to X;
BU C the space of bounded uniformly continuous functions defined from
(−∞, 0]nto X; o
C := φ ∈ BC : limθ→−∞ φ(θ) exists ;
∞
n o
C 0 := φ ∈ BC : limθ→−∞ φ(θ) = 0 , be endowed with the uniform norm

kφk = sup kφ(θ)k.

θ≤0

We have that the spaces BU C, C ∞ and C 0 satisfy conditions (A1 ) − (A3 ).

Equations with infinite delay genera1ly take the following abstract form:

for t ≥ 0
 0
 x (t) = Ax(t) + F (t, xt )
(1.1.2)
x0 = ϕ ∈ B,


where F : R+ × B → X is a function satisfying some conditions; A
: D(A) ⊆

X → X is the infinitesimal generator of a C0 -semigroup T (t) t≥0 on a
Banach space X; for t ≥ 0; and the phase space B is a linear space of
functions mapping ] − ∞, 0] into X satisfying axioms (A1 ) − (A3 ), for every
t ≥ 0, xt denotes the history function of B defined by

xt (θ) = x(t + θ) for − ∞ ≤ θ ≤ 0,

Introduction 6

1.2 Partial Functional Integrodifferential Equa-

tions
In many areas of applications such as Engineering, Electronics, Fluid Dy-
namics, Physical Sciences, etc..., integrodifferential equations appear and
have received considerable attention during the last decades. Consider the
following system:

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t) for t ∈ I = [0, b]


0


x(0) = x0 ,

(1.2.1)
which is a linear Volterra integrodifferential equation that arises in the analy-
sis of heat conduction in materials with memory and viscoelasticity. Equation
(1.2.1) has been studied by many authors under various hypotheses concern-
ing the operators A and B, see for instance the works by Chen and Grimmer
[17], Hannsgen [32, 33], Miller [58, 59], Miller and Wheeler [60, 61], and the
references contained in them.
It was not until 1982 that Grimmer [85] proved the existence and uniqueness
of resolvent operators for this integrodifferential equation that give the vari-
ation of parameter formula for the solutions.
In recent years, much work on the existence and regularity of solutions of the
nonlinear Volterra integrodifferential equations with delays and with nonlo-
cal conditions have been done by many authors by applying the resolvent
operator theory giving by Grimmer in [85], see e.g., [43, 106] and the ref-
erences therein. The objective of this thesis is to study the controllability
and the existence of optimal controls for some class of partial functional in-
tegrodifferential equations in Banach spaces.
We motivate the study by giving the occurence of these partial functional
integrodifferential equations in different areas of applications in science.

1.2.1 A Model in Viscoelasticity

Integrodifferential equations have applications in many problems arising in
physical systems. The following one-dimensional model in viscoelasticity is
Introduction 7

one of the applications of that theory


∂ 2ω ∂ω ∂ϕ
α (t, ξ) + β (t, ξ) = (t, ξ) + h(t, ξ),


2



 ∂t ∂t ∂ξ



 Z t
 ∂ω ∂ω


γ (t, ξ) + a(t − s) (s, ξ)ds = ϕ(t, ξ), (t, ξ) ∈ R+ × [0, 1],
∂ξ 0 ∂ξ




ω(t, 0) = ω(t, 1) = 0, t ∈ R+ ,








ω(0, ξ) = ω0 (ξ), ξ ∈ [0, 1],


where, ω is the displacement, ϕ is the stress, h is the external force, α, γ > 0

and β are constants. In this model, the first equation describes the linear
momentum while the second equation describes the constitutive relation be-
tween stress and strain. Setting γ = 1, v = ∂ω ∂t
, and u = ∂ω
∂ξ
, the above
equations can be rewritten as follows

Z t
u0 (t)
        
0 ∂ξ u(t) a(t − s) 0 u(s)
= ∂ξ + ds
v 0 (t) α
0 v(t) 0 0 0 v(s)

    
0 0 u(t) 0
+ + , t ≥ 0.
0 − αβ v(t) h(t)
α
Setting
     
u(t) 0 ∂ξ a(t − s) 0
x(t) = , A= ∂ξ , B(t) =
v(t) α
0 0 0

   
0 0 0
K= , and p(t) = ,
0 − αβ h(t)
α
we can rewrite the above equation into the following abstract form:
 h Z t i
0
 x (t) = A x(t) + B(t − s)x(s) ds + Kx(t) + p(t) for t ≥ 0


0


x(0) = x0 .


The operator A here is unbounded, while K and B(t) are bounded operators
for t ≥ 0 on a Banach space X. When AB(t) = B(t)A, we obtain the
Introduction 8

following equation:
 Z t
0
 x (t) = Ax(t) + B(t − s)Ax(s)ds + Kx(t) + p(t) for t ≥ 0


0


x(0) = x0 .

(1.2.2)
which has been studied in [23]. We note that in general, the equality AB(t) =
B(t)A does not hold.
Setting f (t, x(t)) = Kx(t) + p(t), equation (1.2.2) becomes
 Z t
0
 x (t) = Ax(t) + B(t − s)Ax(s)ds + f (t, x(t)) for t ≥ 0


0 (1.2.3)


x(0) = x0 .


which to the best of our knowledge has not been investigated for controlla-
bility problem. This problem is addressed in the first part of this thesis, in
the case where equation (1.2.3) admits a nonlocal condition.

1.2.2 A Model in Heat Conduction in Materials with

Memory
Consider a heat flow in a rigid body Ω of a material with memory. Let
w(t, ξ), e(t, ξ), q(t, ξ) and s(t, ξ) denote, respectively, the temperature, the
internal energy, the heat flux, and the external heat supply at time t and
position ξ. The balance law for the heat transfer is given by: (see e.g., [53]

et (t, ξ) + div q(t, ξ) = s(t, ξ) (1.2.4)

and the physical properties of the body suggest the dependence of e and q
on w and ∇w, respectively. For instance, assuming the Fourier Law, i.e.,

e(t, ξ) = c1 w(t, ξ) (1.2.5)

q(t, ξ) = −c2 ∇w(t, ξ), (1.2.6)
where c1 , c2 are positive constants, one deduces from (1.2.4) the classical
heat equation
wt (t, ξ) = c∆w(t, ξ) + f (t, ξ) (1.2.7)
with c = c−1
1 c2 and f (t, ξ) = c1 s(t, ξ).
−1

In many materials, the assumptions (1.2.5) and (1.2.6) are not justified be-
cause they take no account of the memory effects: several models have been
Introduction 9

proposed to overcome this difficulty (see e.g. [100, 101, 102]): one of them
consists in substituting (1.2.6) with
Z t
q(t, ξ) = −c2 ∇w(t, ξ) − h(t − s)∇w(s, ξ)ds. (1.2.8)
−∞

Taking for simplicity c1 = c2 = 1, we get from (1.2.4), (1.2.5) and (1.2.8)

that
Z t
wt (t, ξ) = ∆w(t, ξ) + h(t − s)∆w(s, ξ)ds + s(t, ξ). (1.2.9)
−∞

If we assume that the thermal history w of the body Ω is known up to t = 0

and the temperature of the boundary Γ of Ω is constant (=0) for all t, we
are led to the following system:
 Z t
wt (t, ξ) = ∆w(t, ξ) + h(t − s)∆w(s, ξ)ds + f (t, ξ), (t, ξ) ∈ [0, b] × Ω






 0


 w(0, ξ) = w0 (ξ), ξ ∈ Ω




w(t, ξ) = 0, (t, ξ) ∈ [0, b] × Γ,

(1.2.10)
where b > 0 is arbitrarily fixed. If we prescribe h (in addition to f ) then
(1.2.10) is a Cauchy-Dirichlet problem for an integrodifferential equation in
the unknown w, which has been studied by several authors in the last decades
(see e.g., [103, 104, 105] and references therein).
Now, if we consider that the thermal history of the body Ω is known from
the time t − r (for some r > 0) up to the present time t, the temperature of
the boundary ∂Ω of Ω is constant (= 0) for all t, and the external heat flux
depends on the thermal history of the body, then, system (1.2.10) becomes
the following integrodifferential equation with finite delay:
 Z t
wt (t, ξ) = ∆w(t, ξ) + h(t − s)∆w(s, ξ)ds + f (t, w(t − r, ξ))





 0



for (t, ξ) ∈ [0, b] × Ω (1.2.11)



w(t, ξ) = ψ(t, ξ) for t ∈ [−r, 0] and x ∈ Ω







where ψ is a given initial function and r is a positive number. In the one

dimensional setting, Travis and Webb [75] were the first to consider equation
Introduction 10

(1.2.11) with h = 0, and studied its existence and stability properties.

Now define
x(t)(ξ) = w(t, ξ)
Ax = ∆x
ϕ(θ)(ξ) = ψ(θ, ξ) for θ ∈ [−r, 0] and ξ ∈ Ω.
f (t, ϕ)(ξ) = f (t, ϕ(−r)(ξ)) for t ∈ [0, b] and ξ ∈ Ω
(B(t)x)(ξ) = h(t)∆x(t)(ξ) for t ∈ [0, b] and ξ ∈ Ω.
Then, equation (1.2.11) can be transformed into the following abstract form:

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) for t ∈ I = [0, b],


0


x0 = ϕ ∈ C([−r, 0], X),

(1.2.12)
where X is a Banach space. Equation (1.2.12) has been studied by many
authors (see e.g., [83] and the references contained in it). But to the best
of our knowledge, this equation has never been considered for controllability
and existence of optimal controls and this motivates the work in this thesis.
An example of a material with memory is Shape-memory polymers (SMPs),
which are polymeric smart materials that have the ability to return from a de-
formed state (temporary shape) to their original (permanent) shape induced
by an external stimulus (trigger), such as temperature change.

1.3 Controllability of Dynamical Systems

A dynamical system is a system that evolves in time through the iterated
application of an underlying dynamical rule. It is a mathematical model
that one usually constructs in order to investigate some physical phenomenon
that evolves in time. This model usually involves mainly ordinary differential
equations, partial differential equations or functional differential equations,
which describe the evolution of the process under study in mathematical
terms.
Controllability plays an essential role in the development of modern math-
ematical control systems. It has many important applications not only in
control theory and systems theory, but also in such areas as industrial and
chemical process control, reactor control, control of electric bulk power sys-
tems, aerospace engineering and recently in quantum systems theory. It is
Introduction 11

one of the fundamental concepts in the mathematical control theory. This is

a qualitative property of dynamical control systems and is of particular im-
portance in control theory. A systematic study of controllability started at
the beginning of sixties in the last century, when the theory of controllability
based on the description in the form of state space for both time-invariant
and time-varying linear control systems was worked out [46]. Roughly speak-
ing, controllability generally means, that it is possible to steer a dynamical
control system from an arbitrary initial state to an arbitrary final state using
the set of admissible controls. The notion of controllability was identified by
Kalman [41], as one of the central properties determining system behavior.
The basic controllability problem in continuous time is formulated as follows:
one is interested in steering a control system, whose state x(t) (at time t)
defined on a fixed time interval 0 ≤ t ≤ b, is modelled by the solution of a
differential equation:
dx
= F (t, x, u) on [0, b], (1.3.1)
dt
from an initial state x(0) = x0 , to a desired state x1 , using a control u from
the set of admissible controls, in time b.
For finite dimensional autonomous linear systems, when in (1.3.1), F (t, x, u) =
Ax + Bu, where A is an n × n matrix and B is an n × m matrix, the
Kalman rank condition [41] gives a necessary and sufficient condition for
controllability of the system. It says that system (1.3.1) with F (t, x, u) =
Ax + Bu is controllable if and only if the controllability matrix [A|B] =
[B|AB|A2 B| · · · |An−1 B] of size n × nm has rank n. This is a useful and sim-
ple test, and much effort has been spent on trying to generalize it to nonlinear
systems in various forms. The systematic study of controllability questions
for continuous time nonlinear systems was begun in the early 70’s. At that
time, the papers [52], [74], and [47], building on previous work ([18], [34]) on
partial differential equations, gave a nonlinear analogue of the above Kalman
controllability rank condition. This analogue provides only a necessary test,
not sufficient. Also in infinite dimensional spaces, even for linear system,
one can hardly get a necessary and sufficient condition for controllability, for
more details see [37], and the references contained in it.
Consider the following infinite dimensional linear system:

x (t) = Ax(t) + Bu(t) on [0, b],

 0
(1.3.2)
x(0) = x0

where A generates a strongly continuous semigroup of bounded linear opera-

tors T (t) t≥0 on a Banach space X and B is a bounded linear operator from

a Banach space U into X.

Introduction 12

Now if x is a classical solution of (1.3.2), then x(t) ∈ D(A), for all t ∈ [0, b].
In the genera1 case, when A is unbounded, D(A) 6= X, which means that
the system cannot be steered to all of X. Therefore, only a mild solution of
(1.3.2) given by
Z t
x(t) = T (t)x0 + T (t − s)Bu(s)ds, (1.3.3)
0

will be considered with the following definition of controllability (exact con-

trollability).
Definition 1.3.1 The system (1.3.2) is said to be controllable (exactly con-
trollable) on the interval I = [0, b] if for any two states x0 , x1 ∈ X, there
exists a control u ∈ L2 (I, U ) such that the mild solution x of (1.3.2) satisfies
x(b) = x1 .
Solving the controllability problem in infinite dimension boils down to show-
ing the existence of mild solutions, using fixed point theorems, and finding
the appropriate control for which the mild solution satifies the equality con-
dition in the above definition. The appropriate control varies from system to
system; it is defined or constructed using the following assumption:
(H) The linear operator W from L2 (I, U ) into X, defined by
Z b
Wu = T (b − s)Bu(s) ds,
0

induces a bounded inverse operator W f −1 defined on L2 (I, U )/Ker(W ).

 
Then by defining the control u by u(t) = W f −1
x1 − T (b)x0 (t), and using
it in equation (1.3.3), Hypothesis (H) yields x(b) = x1 , showing that system
(1.3.2) is controllable on I = [0, b].
The construction of W f −1 is outlined as follows ( see [7, 68]):
Let Y = L (I, U )/Ker(W ). Since Ker(W ) is closed, Y is a Banach space
2

under the norm

k[u]kY = inf kukL2 (I,U ) = inf ku + ûk

u∈[u] W û=0

where [u] are the equivalence classes of u.

Define W
f : Y → X by

W
f [u] = W u, u ∈ [u].

Then, Wf is one-to-one and kW

f [u]k ≤ kW kk[u]kY .
Also, V = Range(W ) is a Banach space with the norm kvkV = kW
f −1 vkY .
Introduction 13

To see this, note that this norm is equivalent to the graph norm on D(Wf −1 ) =
Range(W f ). Wf is bounded, and since D(W f ) = Y is closed, W
f −1 is closed.
So, the above norm makes Range(W ) = V , a Banach space.
Moreover,
f −1 W ukY = kW
kW ukV = kW f −1 W
f [u]k = k[u]k = inf kuk ≤ kuk,
u∈[u]

So, W ∈ B(L2 (I, U ), X).

Since, L2 (I, U ) is reflexive, and Ker(W ) is weakly closed, the infimum is
actually achieved. Therefore, for any v ∈ V , a control u ∈ L2 (I, U ) can be
chosen so that u = W f −1 v.

Several authors have studied the controllability of linear and nonlin-

ear systems with various initial conditions, and linear and nonlinear delay
systems in infinite dimensional Banach spaces (see [7] and the references
therein).
The controllability problem of nonlinear systems described by functional in-
tegrodifferential equations with nonlocal conditions in infinite dimensional
Banach spaces, has been studied extensively by many authors, see for in-
stance [8]-[16], [40, 55, 56, 69, 70, 83, 71], [77], and the references therein.
For example in [77], the authors proved the controllability of an integrodif-
ferential system with nonlocal conditions basing on the measure of noncom-
pactness and the Sadovskii fixed-point theorem, and in [69], R. Atmania and
S. Mazouzi have proved the controllability of a semilinear integrodifferential
system using Schaefer fixed-point theorem and requiring the compactness of
the semigroup.
In [69], the authors assumed the compactness of the operator semigroup and
in [9], the authors assumed the compactness of the resolvent operator for
integral equations, whereas in [16, 40, 77], the authors managed to drop this
condition, in the same way as J. Wang, Z. Fan and Y. Zhou [40] have done for
the nonlocal controllability of some semilinear dynamic systems with frac-
tional derivative.
Many authors have also studied the controllability problem of nonlinear sys-
tems with delay in inifinite dimensional Banach spaces: see for instance [8],
[55], [70], [83], [71], etc and the references therein. For example in [83], the au-
thors proved the controllability of semilinear functional evolution equations
with infinite delay using the nonlinear alternative of Leray-Schauder type.
In [56], Meili Li, Miansen Wang and Fengqin Zhang proved the controlla-
bility of an impulsive functional differential system with finite delay using
the Schaefer fixed-point. In [71], S. Selvi and M. Mallika Arjunan proved
the controllability for impulsive differential systems with finite delay using
Introduction 14

the Mönch fixed-point theorem, and in [8], K. Balachandran and R. Sak-

thivel studied the controllability of functional semilinear integrodifferential
systems in Banach spaces using Schaefer fixed point theorem and assuming
the compactness of the operator semigroup.
In this part of the thesis, motivated by the above works, we give sufficient
conditions that guarantee the controllability of the following dynamical sys-
tems described by the following partial functional integrodifferential models:

• Partial functional integrodifferential equation subject to a nonlocal ini-

tial condition in a Banach space (X, k · k) :
 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, x(t)) + Cu(t),



 0
for t ∈ I = [0, b]




x(0) = x0 + g(x),

(1.3.4)
where x0 ∈ X, g : C(I, X) → X and f : I × X → X are functions
satisfying some conditions; A :
D(A) → X is the infinitesimal gen-
erator of a C0 -semigroup T (t) t≥0 on X; for t ≥ 0, B(t) is a closed
linear operator with domain D(B(t)) ⊃ D(A). The control u belongs
to L2 (I, U ) which is a Banach space of admissible controls, where U is
a Banach space.

• Partial functional integrodifferential equation with finite delay in a Ba-

nach space (X, k · k) :
 Z t
0
x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) + Cu(t),




 0
for t ∈ I = [0, b] (1.3.5)




x0 = ϕ ∈ C = C([−r, 0]; X),


where f : I × C → X is a function satisfying some conditions;
A :

D(A) → X is the infinitesimal generator of a C0 -semigroup T (t) t≥0
on X; for t ≥ 0, B(t) is a closed linear operator with domain D(B(t)) ⊃
D(A). The control u belongs to L2 (I, U ) which is a Banach space of
admissible controls, where U is a Banach space, and xt denotes the
history function of C of the state from the time t − r up to the present
time t, and is defined by xt (θ) = x(t + θ) for −r ≤ θ ≤ 0.
Introduction 15

• Partial functional integrodifferential equation with infinite delay in a

Banach space (X, k · k) :
 Z t
0
x (t) = Ax(t) + γ(t − s)x(s)ds + f (t, xt ) + Cu(t),




 0
for t ∈ I = [0, b] (1.3.6)




x0 = ϕ ∈ B,


where
A : D(A) → X is the infinitesimal generator of a C0 -semigroup

T (t) t≥0 on a Banach space X; for t ≥ 0, γ(t) is a closed linear operator
with domain D(γ(t)) ⊃ D(A). The control u takes values from another
Banach space U . The operator C belongs to L(U, X) which is the
Banach space of bounded linear operators from U into X, and the
phase space B is a linear space of functions mapping ] − ∞, 0] into
X satisfying axioms which will be described later, for every t ≥ 0, xt
denotes the history function of B defined by

xt (θ) = x(t + θ) for − ∞ ≤ θ ≤ 0,

f : I × B → X is a continuous function satisfying some conditions.

To the best of our knowledge, up to now no work has reported neither on the
controllability of partial functional integrodifferential equation with nonlocal
initial condition (equation (1.3.4)), with finite delay (equation (1.3.5)) and
with inifinite delay (equation (1.3.6)) in Banach spaces, using the resolvent
operator approach. It has been an untreated topic in the literature, and this
fact is the main aim and motivation of the present work.
Our approach constists in transforming the problem into a fixed-point prob-
lem of an appropriate operator and to apply the Mönch fixed-point theorem,
making use of the Hausdorff measure of noncompactness, and without as-
suming the compactness of the resolvent operator for the associated linear
integral part. The results obtained in this part improve, extend and comple-
ment many other important results in the litterature. They are summarized
in chapters 3, 4 and 5.
Equations (1.3.4), (1.3.5) and (1.3.6) are models that arise in the anal-
ysis of heat conduction in materials with memory [85], and viscoelastic-
ity.
Z The control term Cu(t) is the heating intensity and the integral part
t
B(t − s)x(s)ds is the memory of the system. Materials with memory are
0
interesting because they act adaptively to their environment. They can be
shaped easily into different forms at low temperature, but return to their
Introduction 16

original forms on heating. Steering such systems from an initial state (initial
condition) to a desired terminal one (boundary condition), by an appropriate
choice of a control (which could be the heating intensity), is of interest to
many scientists and engineers. The questions, whether one can heat the ma-
terial in such a way that the initial state is transferred onto a desired state in
time b and under which constraints on the control parameter u, are of interest.

1.4 Optimal Control of Dynamical Systems

In studying dynamical systems in order to improve the system behavior,
one problem that usually surfaces is that of optimal control. In the sim-
plest form, there is a given dynamical system (linear or nonlinear, discrete or
continuous), described by an ordinary differential equation, a partial differ-
ential equation or a functional differential equation, for which input functions
(controls or commands) can be specified. There is also an objective or a cost
function whose value is determined by system behavior, and is in some sense
a measure of the quality of that behavior. The optimal control problem is
that of finding the input function ( the control or the command) so as to
optimize (maximize or minimize) the objective function. The problem seeks
to optimize the objective function subject to the constraints construed by
the model describing the evolution of the underlying system.
Before even precising mathematically this vocabulary, it is important to note
that we all do (more or less) optimal control without even paying attention
to it: to go from one place to another as fast as possible, or using the shortest
path, to maximize the income of investments or shares or minimize debts are
examples of problems of this type.
For example, the dynamical system might be a space vehicule with inputs
corresponding to rocket thrust. The objective might then be to reach the
moon with minimum expenditure of fuel. As another example, the system
might be the nation’s economy, with controls corresponding to governement
monetary and fiscal policy. The objective might be to maximize the aggre-
gate deviations of unemployment and interest rates from fixed target values.
Finally, as a third example, which is of interest in this thesis, the system
might represent the dynamics of heat flow in materials with memory, with
controls corresponding to the heating intensities. The objective might be to
minimize or maximize the heating intensity so as to obtain a particular form
for the material.
The basic optimal control problem in continuous time is formulated as fol-
lows: one is interested in a system (a space vehicule, a nation’s economy,
Introduction 17

heat flow in materials with memory, etc. . . ) whose state x(t) (at time t)
defined on a fixed time interval 0 ≤ t ≤ T , is modelled by the solution of a
differential equation:
dx
= ẋ = F (t, x, u) on [0, T ], x(0) = x0 , (1.4.1)
dt
where x0 ∈ Rn , and T ∈ R are fixed with 0 < T ; F is a given function
from R × Rn × U with values in Rn . As we can see, one of the arguments
of F is a function u defined on the interval [0, T ] with values in a given set
U , which is the set of admissible controls. This function u (the control, or
the command) translates mathematically the actions (or decisions) that one
can exercise on the evolution of the system; the set U corresponds to the
restrictions or the constraints that must be respected by the controls ( for
example: limited ressources, bounds on the acceleration or the speed for the
driving of a vehicule or the heat for the heat flow in a material, etc. . . ).
To formulate an optimal control problem, is to define the state of the system
x(t) and the differential system that describes its evolution, the class of
admissible controls u(t) and finally an evolution criterion or a cost function;
most often, it is a criterion cumulated in time added to a final cost, whose
typical form is:
Z T
J(u) = g(t, x(t), u(t)) dt + h(x(T )) (1.4.2)
0

where g and h are given functions, defined respectively in R × Rn × U and Rn

with values in R. The problem to solve is then to determine the optimal cost
and an optimal control, that is to solve the following optimization problem:
 Z T 
 inf g(t, x(t), u(t))dt + h(x(T ))



u(t) 0
(1.4.3)
 dx = F (t, x, u) on [0, T ], x(0) = x ,



0
dt
Problem (1.4.3) is called Bolza P roblem, and when h ≡ 0, we have a
Lagrange P roblem, which we are interested in in this thesis. The two popu-
lar solution techniques of an optimal control problem are Pontryagin’s max-
imum principle and the Hamilton- Jacobi-Bellman equation [50]. In infinite
dimension, methods and techniques of convex optimization are used. Optimal
control has become a highly established research front in recent years with
numerous contributions to the theory, in both deterministic and stochastic
contexts. Its application to diverse fields such as biology, economics, ecology,
Introduction 18

engineering, finance, management, and medicine cannot be overlooked (see,

e.g., [49] and the references contained in it). The associated mathematical
models are formulated, for example, as systems of ordinary, partial, delay, or
stochastic differential and integrodifferential equations or discrete dynamical
systems, for both scalar and multicriteria decision-making contexts.
Problems of existence of optimal controls for nonlinear differential systems
have been studied extensively by many authors under various hypotheses see
e.g.,[96], [97], [98], [87], [99], [95],[94], [88], [90], [89], and the references con-
tained in them.
In [96], Wang et al. studied the existence and continuous dependence of mild
solutions and the optimal controls of a Lagrange problem for some fractional
integrodifferential equation with infinite delay in Banach spaces using the
using the techniques of a priori estimation and extension of step by steps.
Wand and Zhou [97] discussed the optimal controls of a Lagrange problem
for fractional evolution equations. In [98], Wei et al. studied the optimal
controls for nonlinear impulsive integrodifferential equations of mixed type
on Banach spaces. In [99], the authors studied the existence of mild solutions
and the optimal controls of a Lagrange problem for some impulsive fractional
semilinear differential equations, using the techniques of a priori estimation.
In this part of the thesis, motivated by the above works, we establish the solv-
ability and existence of optimal controls of the following dynamical systems
described by the following partial functional integrodifferential models:
• Partial functional integrodifferential equation with finite delay in a Ba-
nach space (X, k · k) :
 Z t
0
x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) + C(t)u(t),




 0
for t ∈ I = [0, T ],




x0 = ϕ ∈ C = C([−r, 0]; X),

(1.4.4)
where f : I × C → X is a function satisfying some conditions;
A :
D(A) → X is the infinitesimal generator of a C0 -semigroup T (t) t≥0
on a separable reflexive Banach space X; for t ≥ 0, B(t) is a closed
linear operator with domain D(B(t)) ⊃ D(A). The control u takes
values from another separable reflexive Banach space U . The oper-
ator C(t) belongs to B(U, X) which is the Banach space of bounded
linear operators from U into X, and C([−r, 0], X) denotes the Banach
space of continuous functions x : [−r, 0] → X with supremum norm
kxk∞ = supt∈I kx(t)kX , xt denotes the history function of C of the
state from the time t − r up to the present time t, and is defined by
Introduction 19

xt (θ) = x(t + θ) for −r ≤ θ ≤ 0.

• Partial functional integrodifferential equation with infinite delay in a

Banach space (X, k · k) :
 Z t
0
x (t) = Ax(t) + γ(t − s)x(s)ds + f (t, xt ) + C(t)u(t),




 0
for t ∈ I = [0, b]




x0 = ϕ ∈ B,

(1.4.5)
where
A : D(A) → X is the infinitesimal generator of a C0 -semigroup
T (t) t≥0 on a separable reflexive Banach space X; for t ≥ 0, γ(t) is
a closed linear operator with domain D(γ(t)) ⊃ D(A). The control
u takes values from another separable reflexive Banach space U . The
operator C(t) belongs to L(U, X) which is the Banach space of bounded
linear operators from U into X, and the phase space B is a linear space
of functions mapping ] − ∞, 0] into X satisfying axioms which will be
described later, for every t ≥ 0, xt denotes the history function of B
defined by
xt (θ) = x(t + θ) for − ∞ ≤ θ ≤ 0,
f : I × B → X is a continuous function satisfying some conditions.

• Partial functional integrodifferential equation subject to Cauchy initial

condition in a Banach space (X, k · k) :
 Z t
0
x (t) = Ax(t) + B(t − s)x(s)ds + f (t, x(t)) + C(t)u(t),




 0
for t ∈ I = [0, b]




x(0) = x0 ∈ X

(1.4.6)
where x0 ∈ X and f : I × X → X is a function satisfying some condi-
tions;
A : D(A) → X is the infinitesimal generator of a C0 -semigroup
T (t) t≥0 on a separable reflexive Banach space X; for t ≥ 0, B(t) is
a closed linear operator with domain D(B(t)) ⊃ D(A). The control u
takes values from another separable reflexive Banach space U . The op-
erator C(t) belongs to L(U, X) which is the Banach space of bounded
linear operators from U into X.
Introduction 20

Little is known and done about the existence of optimal controls for inte-
grodifferential equations with delay using the resolvent operator for integral
equations approach, especially the problem of existence of optimal controls
of a Lagrange problem for equations (1.4.4), (1.4.5) and (1.4.6) have been an
untreated topic in the literature, and this motivates the present work. We
tackle this problem by using the techniques of a priori estimation of mild
solutions, giving by the resolvent operator for the associated linear integral
part, and techniques of convex optimization. The results obtained in this
part complement many other important results in the literature. They are
summarized in chapters 6, 7 and 8.

1.5 Methods
In [68], Quinn and Carmichael proved that a controllability problem can be
converted into a fixed point problem. So our approach constists in trans-
forming the problems (1.3.4), (1.3.5) and (1.3.6) into fixed-point problems of
appropriate operators and to apply the Hausdorff measure of noncompact-
ness and the Mönch fixed-point theorem, without requiring the compactness
of the operator semigroup, in the same way as J. Wang, Z. Fan and Y. Zhou
[40] have done for the nonlocal controllability of some semilinear dynamic
systems with fractional derivative. This method enables us overcome the
resolvent operator case considered in this thesis. In contrary to the evolu-
tion semigroup case considered in [55, 71], here the semigroup property can
not be used because resolvent operators in general do not form semigroups.
As for the existence of optimal controls of the associated Lagrange prob-
lems to equations (1.4.4), (1.4.5) and (1.4.6), we apply techniques of convex
optimization together with Balder’s Theorem to obtain our results.

1.5.1 Measures of Noncompactness

Measures of noncompactness are very useful tools in nonlinear analysis. For
instance, in metric fixed point theory and in the theory of operator equations
in Banach spaces. They are also used in the studies of functional equations,
ordinary and partial differential equations, fractional partial differential equa-
tions, integral and integrodifferential equations, optimal control theory, and
in the characterizations of compact operators between Banach spaces (see
e.g., [6, 11, 12, 67]). The concept of measure of noncompactness was first
defined and studied by Kuratowski [48] in 1930. In 1955, G. Darbo [21] used
it to prove his fixed point theorem. We have the following definition:
Introduction 21

Definition 1.5.1 Let E + be the positive cone of an order Banach space

(E, ). That is
E+ = x ∈ E : 0  x .


And let X be an arbitrary Banach space. A function Φ defined on the set of all
bounded subsets of X with values in E + is called a measure of noncompactness
(M N C) on X if Φ(co(D)) = Φ(D) for every bounded subset D ⊂ X, where
co(D) stands for the closed convex hull of D.

Example 1.5.2 ( [6], Example 1, p. 19]) Let X be an arbitrary metric

space and MX denote the set of all bounded subsets of X. Then the map
Φ1 : MX → [0, ∞) defined by

 0, if D is relatively compact
Φ1 (D) =
1, otherwise.


is a measure of noncompactness, the so-called discrete measure of noncom-

pactness.

Two measures of noncompactness which are frequently used in many branches

of nonlinear analysis and its applications, are Kuratowski measure of non-
compactness and the Hausdorff measure of noncompactness, which we shall
use in this thesis. The frequent use of this latter one is due to the fact that
it is defined in a natural way and has several useful properties.

Definition 1.5.3 Let D be a bounded subset of a normed space Z. The

Kuratowski measure of noncompactness of D (shortly MNC) is defined by
n o
α(D) = inf  > 0 : D has a f inite cover by sets of diameter less than  ,

n n
[ o
= inf  > 0 : D ⊂ Sk , Sk ⊂ Z, diam(Sk ) <  (k = 1, 2, · · · , n ∈ N) .
k=1

Definition 1.5.4 Let D be a bounded subset of a normed space Z. The

Hausdorff measure of noncompactness of D (shortly MNC) is defined by
n o
β(D) = inf  > 0 : D has a f inite cover by balls of radius less than 

n n
[ o
= inf  > 0 : D ⊂ B(xk , rk ), xk ∈ Z, rk <  (k = 1, 2, · · · , n ∈ N) .
k=1
Introduction 22

The compactness conditions described by means of measures of noncom-

pactness are useful in showing the existence of solutions for differential and
integral equations in Banach spaces. Measures of noncompactness have ap-
plications in many fields where loss of compactness arises. For example,
integral equations with strongly singular kernels, differential equations over
unbounded domains, functional differential equations of neutral type or with
deviating argument, linear differential operators with nonempty essential
spectrum, nonlinear superposition operators between various function spaces,
initial value problems in Banach spaces etc, see e.g.,[1, 11, 4, 25] and the ref-
erences therein.

1.5.2 Fixed Point Theory

The fixed point technique is one of the useful methods mainly applied in the
existence and uniqueness of solutions of differential equations and the con-
trollability of dynamical systems. One of the main branches of fixed point
theory deals with the topological properties of the operators involved. With
respect to the topological aspect, the two main theorems are Brouwer’s the-
orem and its infinite dimensional version, Schauder’s fixed point theorem.
In both theorems, compactness plays an essential role. In 1955, Darbo [21]
extended Schauder’s theorem to the setting of noncompact operators, intro-
ducing the notion of k-set contraction which is closely related to the idea
of measures of noncompactness. In 1967, Sadovskii [26] gave a fixed point
result more general than Darbo’s theorem using the concept of condensing
map. Thus, the fixed point theory for condensing mappings allows us to
obtain a relationship between the two theories. In 1980, Mönch [62] gave
a fixed point theorem for maps between Banach spaces, which extends the
Schauder and more generally, Sadovskii fixed point theorems.
In this thesis, we use the Mönch fixed point theorem and the Hausdorff mea-
sure of noncompactness to prove the controllability results for partial func-
tional integrodifferential equations. The advantage of using this fixed point
theorem is to weaken the compactness assumption of the operator semigroup
and the resolvent operator for integral equations.

1.5.3 Semigroup Theory

The theory of semigroups of bounded linear operators is part of functional
analysis. It is an extensive mathematical subject with substantial applica-
tions to many fields of analysis, and has developed quite rapidly since the
discovery of the generation theorem by Hille and Yosida in 1948. Semigroups
are a powerful tool in solving evolution equations. They give the variation of
Introduction 23

parameter formula for mild solutions. In [66], Pazy discussed the existence
and uniqueness of mild, strong and classical solutions of evolution equations
using semigroup theory and fixed point theorems.

1.5.4 Resolvent Operator for Integral Equations

This concept was first introduced in 1982 by Ronald Grimmer [85]. He
proved the existence and uniqueness of resolvent operators for integrodiffer-
ential equations that give the variation of parameter formula for the solution.
Resolvent operators in general do not form semigroup, and are a better tool
in solving integrodifferential equations than semigroups. In [24], W. Desch,
R. Grimmer and W. Schappacher proved the equivalence of the compactness
of the resolvent operator for integral equations and that of the operator semi-
group. In this thesis, we show a similar result on the equivalence between
the operator-norm continuity of the resolvent operator for integral equations
and the C0 -semigroup. This result is important because it allows to drop the
compactness assumption on the resolvent operator and assume its operator-
norm continuity, in proving our controllability results. In [69], the authors
assumed the compactness of the operator semigroup and in [9], the authors
assumed the compactness of the resolvent operator whereas in [16, 40, 77],
the authors managed to drop this condition which motivates our current
work. Our contributions in this direction are summerized in Chapters 3 and
4.

1.6 Organization of the Thesis

The present study in this thesis deals firstly with the fixed point approach
via measures of noncompactness for proving controllability results for partial
functional integrodifferential systems in Banach spaces.
We note that the partial functional integrodifferential equations (1.3.4), (1.3.5)
and (1.3.6) have not been investigated (to the best of our knowledge) for con-
trollability.
The second part of the thesis deals with the existence of optimal controls of
the associated Lagrange problems to equations (1.4.4), (1.4.5) and (1.4.6),
we apply techniques of convex optimization together with Balder’s Theorem.
In chapter 2, we give preliminary results that will be used in proving our
main results. In particular, we establish
the equivalence between operator-
norm continuity of the semigroup T (t) t≥0 generated by A and the resolvent
operator R(t) t≥0 corresponding to the associated linear equation.


Introduction 24

In chapter 3 , we prove the controllability result for some partial functional

integrodifferential equation with nonlocal initial conditions in Banach spaces
(equation (1.3.4)), by using resolvent operators for integral equations and the
Mönch fixed point theorem.

In chapter 4, we establish a controllability result for some partial functional

integrodifferential equation with finite delay in Banach spaces (equation
(1.3.5)), by using resolvent operators for integral equations and the Mönch
fixed point theorem.

In chapter 5, we establish a controllability result for some partial functional

integrodifferential equation with infinite delay in Banach spaces (equation
(1.3.6)), by using resolvent operators for integral equations and the Mönch
fixed point theorem.

In chapter 6, we prove the solvability and the existence of optimal controls

for some partial functional integrodifferential equation with finite delay in
Banach spaces (equation (1.4.4)), by using the techniques of convex opti-
mization, a-priori estimation and contraction mapping principle.

In Chapter 7 , we prove the solvability and the existence of optimal controls

for some partial functional integrodifferential equation with infinite delay in
Banach spaces (equation (1.4.5)), by using the techniques of convex opti-
mization, a-priori estimation and contraction mapping principle.

In Chapter 8, we prove the solvability and the existence of optimal controls

for some partial functional integrodifferential equation with classical initial
conditions in Banach spaces (equation (1.4.6)), by using the techniques of
convex optimization, a-priori estimation and contraction mapping principle.
 CHAPTER 2

Preliminaries

In this Chapter, we give some definitions and preliminary results that will
be needed in the subsequent chapters.

2.1 Semigroups


Definition 2.1.1 Let X be a Banach space. A family T (t) t≥0 of bounded
linear operators from X to X is called a strongly continuous semigroup of
bounded linear operators if the following three conditions are satisfied :
i) T (0) = IdX

ii) T (t + s) = T (t)T (s), ∀t, s ≥ 0

iii) ∀x ∈ X, the map t 7→ T (t)x ∈ X defined from [0, +∞) into X is

continuous at the right of 0.

A strongly continuous semigroup of bounded linear operators on X will be

called a C0 -semigroup.

If limt→0+ kT (t) − Ik = 0, then we say that the semigroup is uniformly

continuous.
Example of Semigroup:
Let’s define

T : [0, ∞) −→ B C0 (R)
t 7→ T (t)

25
Preliminaries 26

which at each f ∈ C0 (R) assigns T (t)f ∈ C0 (R) defined by

T (t)f : x 7→ f (t + x) .
Then the family of operators (T (t))t≥0 is a strongly continuous semigroup.
Justification:
Clearly T is well-defined and (i) For all f ∈ C0 (R) and for all x ∈ R, we have
(T (0)f )(x) = f (0 + x) = f (x).
Thus T (0)f = f and so T (0) = I
(ii) Let s, t ∈ [0, ∞), f ∈ C0 (R) and x ∈ R. Then we have that

h
i
(T (s)T (t))f (x) = T (s) T (t)f (x)


= T (t)f (s + x)
= f (t + s + x)
= f (s + t + x)


= T (s + t)f (x)
which shows that T (s)T (t) = T (s + t).
(iii) We now show the continuity of t 7→ T (t)f at the right of 0. Let f ∈ C0 (R),
then


kT (t)f − f kC0 (R) = sup| T (t)f (x) − f (x)|
x∈R
= sup|f (t + x) − f (x)| −→ 0 as t −→ 0+
x∈R

by uniform continuity of f .
Thus
lim+ kT (s)f − f kC0 (R) = 0.
t−→0
And therefore (T (s))s≥0 defined above is a strongly continuous semigroup.

Infinitesimal Generator of a C0 -semigroup

Definition 2.1.2 The infinitesimal generator of a C0 -semigroup (T (t))t≥0
of bounded linear operators of a Banach space X, is the linear operator A :
D(A) −→ X with domain
 
T (t)x − x
D(A) = x ∈ X : lim+ exists
t→0 t
and defined by
T (t)x − x d+ T (t)x
Ax := lim+ = |t=0 f or x ∈ D(A) .
t→0 t dt
Preliminaries 27

2.2 Resolvent Operator for Integral Equations

Let I = [0, b], b > 0 and let X be a Banach space. A measurable function
x : I → X is Bochner integrable if and only if kxk is Lebesgue integrable.
We denote by L1B (I, X) the Banach space of functions x : I → X which are
Bochner integrable normed by
Z b
kxkL1 = kx(t)kdt.
0

Consider the following linear homogeneous equation:

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds for t ≥ 0


0 (2.2.1)


x(0) = x0 ∈ X.


where A and B(t) are closed linear operators on a Banach space X.

In the sequel, we assume A and B(t) t≥0 satisfy the following conditions:

(H1 ) A is a densely defined closed linear operator in X. Hence D(A) is

a Banach space equipped with the graph norm defined by, |y| = kAyk + kyk
which will be denoted by (X1 , | · |).

(H2 ) B(t) t≥0 is a family of linear operators on X such that B(t) is

continuous when regarded as a linear map from (X1 , | · |) into (X, k · k) for
almost all t ≥ 0 and the map t 7→ B(t)y is measurable for all y ∈ X1 and
t ≥ 0, and belongs to W 1,1 (R+ , X). Moreover there is a locally integrable
function ρ : R+ → R+ such that

d
kB(t)yk ≤ ρ(t)|y| and dt B(t)y ≤ ρ(t)|y| .

Remark 2.2.1 Note that (H2 ) is satisfied in the modelling of heat conduc-
tion in materials with memory and viscosity. More details can be found in
[106].

(see e.g., [83]) A resolvent operator for equation (2.2.1) is

Definition 2.2.2
a family R(t) t≥0 of bounded linear operators valued function

R : [0, +∞) −→ B(X)

such that
Preliminaries 28

(i) R(0) = IdX and kR(t)k ≤ N eβt for some constants N and β.

(ii) For all x ∈ X, the map t 7→ R(t)x is continuous for t ≥ 0.

(iii) Moreover for x ∈ X1 , R(·)x ∈ C 1 (R+ ; X) ∩ C(R+ ; X1 ) and

Z t
0
R (t)x = AR(t)x + B(t − s)R(s)xds
0
Z t
= R(t)Ax + R(t − s)B(s)xds.
0

Observe that the map defined on R+ by t 7→ R(t)x0 solves equation (2.2.1)

for x0 ∈ D(A).

We have the following example of a resolvent operator for equation (2.2.1) in

R.

Example([24]) Let X = R, Ay = 2y, and B(t)y = −2y in (2.2.1). Then

in that case, we have

R(t)x0 = et (cos t + sin t)x0 and T (t)x0 = e2t x0 .

Remark 2.2.3
The above example also shows that, in general, the resolvent
operator R(t) t≥0 for equation (2.2.1) does not satisfy the semigroup law,
namely,
R(t + s) 6= R(t)R(s) for some t, s > 0 .
In this chapter, we establish
the equivalence between operator-norm conti-
nuity of the semigroup T (t) t≥0 generated by A and the resolvent operator
R(t) t≥0 corresponding to the linear equation (2.2.1). We shall then as-

sume the operator-norm continuity of R(t) t≥0 , to prove the controllability

of equations (1.3.4), (1.3.5) and (1.3.6).

Now consider the following system:

 Z t
0
 x (t) = Ax(t) + [B1 (t − s) + B2 (t − s)] x(s)ds for t ≥ 0


0 (2.2.2)


x(0) = x0 ∈ X,

Preliminaries 29

where B1 (t) and B2 (t) are closed linear operators in X and satisfy (H2 ).
Then we have the next Lemma coming from [24].


Lemma 2.2.4 (Perturbation result)([24]) Suppose A satisfies (H1 ) and B1 (t) t≥0



and B2 (t) t≥0 satisfy (H2 ). Let RB1 (t) t≥0 be a resolvent operator of equa-


tion (2.2.1) and RB1 +B2 (t) t≥0 be a resolvent operator of equation (2.2.2).
Then Z t
RB1 +B2 (t)x − RB1 (t)x = RB1 (t − s)Q(s)x ds
0
where the operator Q is defined by
Z t Z s Z t
0
Q(t)x = B2 (t − s) RB1 +B2 (τ )x dτ ds + B2 (0) RB1 +B2 (s)x ds,
0 0 0

is uniformly bounded on bounded intervals, and for each x ∈ X, Q(·)x belongs

to C([0, ∞), X).
Based on this and the following corollary from ([24], p.224), we prove the
operator-norm continuity of the resolvent operator R(t) t≥0 for t > 0.

Corollary 2.2.5 ([24]) Let A be a closed,

densely defined linear operator in
X, B(t) = 0 for all t ≥ 0, and R(t) t≥0 be a resolvent operator for equation


(2.2.1). Then R(t) t≥0 is a C0 -semigroup with infinitesimal generator A.


Theorem 2.2.6 Let A be the infinitesimal generator of a C0 -semigroup T (t) t≥0



and let B(t) t≥0 satisfy (H2 ). Then the resolvent operator R(t) t≥0 for
equation (2.2.1) is operator-norm continuous (or
continuous in the uniform
operator topology) for t > 0 if and only if T (t) t≥0 is operator-norm contin-
uous for t > 0.

Proof. Since T (t) t≥0 is a C0 -semigroup, there exist M and ω such that

||T (t)|| ≤ M eωt for all t ≥ 0.

Let
Z t Z t Z s
0
Q(t)x = B(0) R(s)x ds + B (t − s) R(τ )x dτ ds.
0 0 0

By Lemma 2.2.4 (applied to equation (2.2.1) ) we have that the operators

Q(t) t≥0 are uniformly bounded for t in a bounded interval and that

Z t
R(t)x = T (t)x + T (t − s)Q(s)x ds.
0
Preliminaries 30

Now set α = sup kQ(t)k and let x ∈ X be arbitrary and such that kxk ≤ 1.
0≤t≤b
Suppose first that T (t) t≥0 is operator-norm continuous for t > 0. Then on

the one hand, we have for every t ∈ (0, b) and h ∈ (0, b − t) that :
h i
R(t + h)x − R(t)x ≤ T (t + h) − T (t) x

Z t h i
+ α T ((t − s) + h) − T (t − s) x ds

0
Z t+h
+ α M eω(t−s+h) ds.
t

It follows that
Z t
R(t + h) − R(t) ≤ T (t + h) − T (t) + α T ((t − s) + h) − T (t − s) ds

0
Z t+h
+ α M eω(t−s+h) ds.
t

Therefore by using the operator-norm continuity of T (t) t≥0 on (0, +∞) and

the Lebesgue dominated convergence theorem, we deduce that

R(t + h) − R(t) → 0 as h → 0+ .

On the other hand, for every t ∈ (0, b] and h ∈ (−t, 0), we have that
Z t
R(t + h) − R(t) ≤ T (t + h) − T (t) + α T ((t − s) + h) − T (t − s) ds

0
Z t
+ α M eω(t−s) ds.
t+h

And using the operator-norm continuity of T (t) t≥0 and the Lebesgue dom-

inated convergence theorem, we have that

R(t + h) − R(t) → 0 as h → 0− .

Hence

R(t + h) − R(t) → 0 when h → 0 with t + h ∈ [0, b].

Preliminaries 31

Thus R(t) t≥0 is operator-norm continuous when T (t) t≥0 is operator-norm

continuous.

The converse is proved similarly by using the identities

Z t Z t
T (t)x = −R(t)x + T (t − s)Q(s)x ds = −R(t)x + T (s)Q(t − s)x ds
0 0

and the continuity of Q(t) t≥0 that will follow from the property (H2 ) of

B(t) t≥0 and the continuity of R(t) t≥0 .

In fact suppose that R(t) t≥0 is operator-norm continuous for t > 0. Then

on the one hand, we have for every t ∈ (0, b) and h ∈ (0, b − t) that :
h i
T (t + h)x − T (t)x ≤ R(t + h) − R(t) x


Z t h i
+ M eωs Q((t − s) + h) − Q(t − s) x ds

0
Z t+h
+ α M eωs ds.
t
It follows that

T (t + h) − T (t) ≤ R(t + h) − R(t)

Z t
+ M eωs Q((t − s) + h) − Q(t − s) ds

0
Z t+h
+ α M eωs ds.
t

Therefore by using the operator-norm continuity of R(t) t≥0 on (0, +∞), the

continuity of Q(t) t≥0 and the Lebesgue dominated convergence theorem,

we have that
T (t + h) − T (t) → 0 as h → 0+ .

On the other hand, for every t ∈ (0, b] and h ∈ (−t, 0), we have that
h i
T (t + h)x − T (t)x ≤ R(t + h) − R(t) x

Z t h i
ωs
+ M e Q((t − s) + h) − Q(t − s) x ds

0
Z t
+ α M eωs ds.
t+h
Preliminaries 32

It follows that

T (t + h) − T (t) ≤ R(t + h) − R(t)

Z t
+ M eωs Q((t − s) + h) − Q(t − s) ds

0
Z t
+ α M eωs ds.
t+h

Therefore by using the operator-norm continuity of R(t) t≥0 on (0, +∞), the

continuity of Q(t) t≥0 and the Lebesgue dominated convergence theorem,

we have that
T (t + h) − T (t) → 0 as h → 0− .

Hence,

T (t + h) − T (t) → 0 as h → 0 with t + h ∈ [0, b].

Thus T (t) is operator-norm continuous when R(t) t≥0 is.

t≥0

2.3 Measure of Noncompactness

For proving the main results in the first part of the thesis, we recall the
notion of measure of noncompactness and the Mönch fixed-point theorem.
For more information on this, the reader can see [12]. In order to use the
Hausdorff measure of noncompactness, we recall some properties related to
this concept.

Definition 2.3.1 Let D be a bounded subset of a normed space Z. The

Hausdorff measure of noncompactness ( shortly MNC) is defined by
n o
β(D) = inf  > 0 : D has a f inite cover by balls of radius less than  .

Theorem 2.3.2 ([12])

Let D, D1 , D2 be bounded subsets of a Banach space Z. The Hausdorff
MNC has the following properties:

(i) If D1 ⊂ D2 , then β(D1 ) ≤ β(D2 ), (monotonicity).

Preliminaries 33

(ii) β(D) = β(D).

(iii) β(D) = 0 if and only if D is relatively compact.

(iv) β(λD) = |λ|β(D) for any λ ∈ R, (Homogeneity)

(v) β(D1 + D2 ) ≤ β(D1 ) + β(D2 ), where D1 + D2 = {d1 + d2 : d1 ∈ D1 , d2 ∈

D2 }, (subadditivity)

(vi) β({a} ∪ D) = β(D) for every a ∈ Z.

(vii) β(D) = β(co(D)), where co(D) is the closed convex hull of D.

(viii) For any map G : D(G) ⊆ X → Z which is Lipschitz continuous with

a Lipschitz constant k, we have

β(G(D)) ≤ kβ(D),

for any subset D ⊆ D(G).

(xi) β(D) = inf{dZ (D, E); E ⊆ Z is relatively compact} = inf{dZ (D, E); E ⊆
Z is finite valued}, where dZ (D, E) means the nonsymmetric (or sym-
metric) Hausdorff distance between D and E in Z.

(x) If {Dn }∞
n=1 is a decreasing sequence of bounded closed nonempty subsets
\∞
of Z and limn→∞ β(Dn ) = 0, then, Dn is nonempty and compact in
n=1
Z.

Lemma 2.3.3 ([40, 67]) If (un )n≥1 is a sequence of Bochner integrable func-
tions from I = [a, b] into a Banach space Z with the estimation kun (t)k ≤
µ(t) for almost all t ∈ I and every n ≥ 1, where µ ∈ L1 (I, R), then the
function
ψ(t) = β({un (t) : n ≥ 1})
belongs to L1 (I, R) and satisfies the following estimation
Z t  Z t
β un (s)ds : n ≥ 1 ≤ 2 ψ(s)ds,
a a

for all t ∈ [a, b]

Preliminaries 34

Lemma 2.3.4 ([12]) If W ⊆ C([a, b], X) is bounded and equicontinuous,

then β(W (t)) is continuous and
Z t  Z t
β W (s) ds ≤ β(W (s)) ds,
a a
Z t Z t 
for all t ∈ [a, b], where W (s) ds = w(s) ds : w ∈ W
a a

Lemma 2.3.5 Let Z be a Banach space and (Tn )n≥1 be a sequence of bounded
linear maps on Z converging pointwise to T ∈ B(Z) . Then for any compact
set K in Z, Tn converges to T uniformly in K, namely,

sup kTn (x) − T (x)k −→ 0, as n → +∞.

x∈K

Proof. By the uniform boundedness principle, we have that sup kTn k <
n≥1
∞. Let M = sup kTn k and  > 0 be arbitrarily given. Then there exist
n≥1
m
[   
a1 , a2 , · · · , am such that K ⊂ B ai , .
i=1
2(M + 1)
 
Also, for any x ∈ K, there exists i ∈ {1, · · · , m} such that x ∈ B ai , 2(M+1) .
Since for i = 1, 2, · · · , m,
Tn (ai ) → T (ai ), there exists N > 0 such that

kTn (ai ) − T (ai )k ≤ , for n ≥ N and for any i = 1, · · · , m.
M +1
We have that

kTn (x) − T (x)k ≤ kTn (x) − Tn (ai )k + kTn (ai ) − T (ai )k + kT (ai ) − T (x)k
≤ kTn kkx − ai k + kT kkai − xk + kTn (ai ) − T (ai )k
M
≤ + kTn (ai ) − T (ai )k
M +1
M 
≤ + =
M +1 M +1
Therefore sup kTn (x) − T (x)k → 0 as n → +∞.
x∈K

2
Preliminaries 35

2.4 The Mönch Fixed Point Theorem and Balder’s

Theorem
We recall a nonlinear alternative of Mönch’s type for non selfmaps.

Theorem 2.4.1 [62](Mönch, 1980) Let G be an open neighborhood of the

origin in a Banach space Z. Suppose that F : G → Z is a continuous map
which satisfies the following conditions:

(i) D ⊂ G countable and D ⊆ co({0} ∪ F (D)) =⇒ D is compact.

(ii) F (x) 6= λx for all x ∈ ∂G and λ > 1, (Leray-Schauder condition).

Then F has a fixed point.

We now state the following nonlinear alternative of Mönch’s type for self-
maps, which we shall use in the proof of the controllability results.

Theorem 2.4.2 [62](Mönch, 1980) Let K be a closed and convex subset of

a Banach space Z and 0 ∈ K. Assume that F : K → K is a continuous map
which satisfies Mönch’s condition, namely,
let D ⊆ K be countable and D ⊆ co({0} ∪ F (D)), then D is compact.
Then F has a fixed point.

The following Theorem is needed in the proof of the existence of optimal

controls.

Theorem 2.4.3 (Balder’s Theorem, [107]) Let (Σ, F, µ) be a finite nonatomic

measure space, (X, k · k) a separable Banach space, and (V, | · |) a separable
reflexive Banach space, and V 0 its dual. Let θ : Σ ×X ×V → (−∞, +∞] be a
given F × L(X × V )- measurable function. The associated integral functional
Iθ : L1X × L1V → [−∞, +∞] is defined by:
Z
Iθ (x, v) = θ(t, x(t), v(t))µ(dt),
Σ

where L1X denotes the set of all absolutely summable functions from Σ to X.
The following three conditions

(i) θ(t, ·, ·) is sequencially lower semicontinuous on X × V, µ-a.e.,

(ii) θ(t, x, ·) is convex on V for x ∈ X, µ-a.e.,

Preliminaries 36

(iii) there exist σ > 0 and ϕ ∈ L1R such that

θ(t, x, v) ≥ ϕ(t) − σ(kxk + |v|), for all x ∈ X, v ∈ V, µ-a.e.,

are sufficient for sequential strong-weak lower semicontinuity of Iθ on L1X ×

L1V . Moreover, they are also necessary, provided that Iθ (x, v) < +∞ for some
x ∈ L1X , v ∈ L1V .

Lemma 2.4.4 (Mazur’s Lemma, [108]) Let Z be a Banach space and G be

a convex and closed set in Z. Then G is weakly closed in Z.
 Part I

Controllability of some Partial

Functional Integrodifferential
Equations in Banach Spaces

37
 CHAPTER 3

Controllability for some Partial Functional

Integrodifferential Equations with Nonlocal Conditions in
Banach Spaces

3.1 Introduction
Several authors have studied the controllability problem of nonlinear systems
described by functional integrodifferential equations with nonlocal conditions
in infinite dimensional Banach spaces: see for instance [69], [9], [16], [40], [77],
and the references therein. For example in [77], the authors proved the con-
trollability of an integrodifferential system with nonlocal conditions basing
on the measure of noncompactness and the Sadovskii fixed-point theorem,
and in [69], R. Atmania and S. Mazouzi have proved the controllability of a
semilinear integrodifferential system using Schaefer fixed-point theorem and
requiring the compactness of the semigroup.
In [69], the authors assumed the compactness of the operator semigroup and
in [9], the authors assumed the compactness of the resolvent operator whereas
in [16, 40, 77], the authors managed to drop this condition which motivates
our current work.
In this chapter, we study the controllability of some systems that arise in the
1
The results of this chapter are contents of the following paper
- Khalil Ezzinbi, Guy Degla, Patrice Ndambomve [42]

38
Controllability for Nonlocal Integrodifferential Equations 39

analysis of heat conduction in materials with memory [85], and viscoelastic-

ity, and take the form of the following model of partial functional integrod-
ifferential equation subject to a nonlocal initial condition in a Banach space
(X, k · k) :
 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, x(t)) + Cu(t) for t ∈ I = [0, b]


0


x(0) = x0 + g(x),

(3.1.1)
where x0 ∈ X, g : C(I, X) → X and f : I × X → X are functions sat-
isfying some conditions; A : D(A) → X is the infinitesimal generator of a
C0 -semigroup T (t) t≥0 on X; for t ≥ 0, B(t) is a closed linear operator with

domain D(B(t)) ⊃ D(A). The control u belongs to L2 (I, U ) which is a Ba-

nach space of admissible controls, where U is a Banach space. The operator
C belongs to B(U, X) which is the Banach space of bounded linear operators
from U into X, and C(I, X) denotes the Banach space of continuous func-
tions x : I → X with supremum norm kxk∞ = supt∈I kx(t)kX .
The particular case B(t) = 0 has been considered by Y.K. Chang, J.J. Nieto
and W.S. Li [16], where the authors used Sadovskii fixed-point theorem and
the operator semigroup to prove their result.
Here our goal is to study equation (3.1.1) where B(t) 6= 0 as a generalization
of the result by Y.K. Chang, J.J. Nieto and W.S. Li without requiring the
compactness of the operator semigroup, in the same way as J. Wang, Z. Fan
and Y. Zhou [40] have done for the nonlocal controllability of some semilin-
ear dynamic systems with fractional derivative. Our approach constists in
transforming the problem (3.1.1) into a fixed-point problem of an appropri-
ate operator and to apply the Mönch fixed-point theorem.

Definition 3.1.1 A mild solution of equation (3.1.1) is a function x ∈

C(I, X) such that
Z t
 
x(t) = R(t)[x0 + g(x)] + R(t − s) f (s, x(s)) + Cu(s) ds for t ∈ I.
0

Definition 3.1.2 Equation (3.1.1) is said to be controllable on the interval

I if for every x0 , x1 ∈ X, there exist a control u ∈ L2 (I, U ) and a mild
solution x of equation (3.1.1) satisfying the condition x(b) = x1 .
Controllability for Nonlocal Integrodifferential Equations 40

3.2 Main Result

In the sequel, we shall denote by β the Hausdorff measure of noncompactness.
Consider the following hypotheses.

(H3 ) The linear operator W : L2 (I, U ) → X satisfies the following condi-

tion:

(i) W defined by
Z b
Wu = R(b − s)Cu(s) ds,
0

is surjective so that it induces an isomorphism between L2 (I, U ) /KerW

and X again denoted by W with inverse W −1 taking values in L2 (I, U ) /KerW ,
(see e.g., [68]).

(ii) There exists a function LW ∈ L1 (I, R+ ) such that for any bounded
set Q ⊂ X we have

β((W −1 Q)(t)) ≤ LW (t)β(Q),

where β is the Hausdorff MNC.

(H4 ) The function f : I × X −→ X satisfies the following conditions:

(i) f (·, x) is measurable for x ∈ X and f (t, ·) is continuous for a.e t ∈ I.

(ii) There exist a function Lf ∈ L1 (I, R+ ) and a nondecreasing continu-

ous function φ : R+ → R+ such that

φ(r)
kf (t, x)k ≤ Lf (t)φ(kxk) for x ∈ X, t ∈ I and lim inf = 0.
r→+∞ r

(iii) There exists a function h ∈ L1 (I, R+ ) such that for any bounded set
D ⊂ X,
β(f (t, D)) ≤ h(t)β(D) for a.e t ∈ I,
where β is the Hausdorff MNC.
gr
(H5 ) g : C(I, X) −→ X is continuous, compact and satisfies lim inf =0
r→+∞ r
where n o
gr = sup kg(x)k : kxk ≤ r .
Controllability for Nonlocal Integrodifferential Equations 41

 
Theorem 3.2.1 Suppose equation (2.2.1) has a resolvent operator R(t)
t≥0
that is continuous in the operator-norm topology for t > 0 and hypotheses
(H3 ) − (H5 ) are satisfied. Then equation (3.1.1) is controllable on I, provided
that   
γ = 1 + 2M1 M2 kLW kL1 2M1 khkL1 < 1,

where M1 = sup kR(t)k and M2 is such that kCk = M2 .

0≤t≤b

Proof. Note at once that M
1 < +∞ according to the exponential growth of
the resolvent operator R(t) t≥0 . Using (H3 ) we define the following control
by
 Z b 
ux (t) = W −1
x1 − R(b)[x0 + g(x)] − R(b − s)f (s, x(s)) ds (t) for t ∈ I,
0

for an arbitrarily given function x ∈ C(I, X).

Using this control, we shall show that the operator K : C(I, X) → C(I, X)
defined by
Z t h i
(Kx)(t) = R(t)[x0 + g(x)] + R(t − s) f (s, x(s)) + Cux (s) ds,
0

has a fixed point x which is just a mild solution of the equation (3.1.1). Ob-
serve that (Kx)(b) = x1 and so the control ux steers the integrodifferential
equation from x0 to x1 in time b. This means that equation (3.1.1) is con-
trollable on I.

For each positive number r, let Br = {x ∈ C(I, X) : kxk∞ ≤ r}. We shall

prove the above theorem through the following steps.

Step 1. We claim that there exists r > 0 such that K(Br ) ⊂ Br .

Suppose on the contrary that this is not true. Then for each positive number
r, there exists a function xr ∈ Br , such that K(xr ) ∈
/ Br , i.e., k(Kxr )(τ )k >
r, for some τ = τ (r) ∈ I. Now

k(Kxr )(τ )k k(Kxr )(τ )k

> 1 that implies that lim inf ≥ 1. (∗)
r r→+∞ r
Controllability for Nonlocal Integrodifferential Equations 42

On the other hand, let M3 = kW −1 k. We have

Z b
(Kxr )(τ ) ≤ M1 kx0 k + M1 kg(xr )k + M1 f (s, xr (s)) ds

0
 
+ bM1 M2 M3 kx1 k + M1 kx0 k
 Z b 
+ bM1 M2 M3 M1 kg(xr )k + M1 f (s, xr (s)) ds

0

≤ M1 kx0 k + M1 gr + M1 φ(r)kLf kL1

 
+ bM1 M2 M3 kx1 k + M1 kx0 k
 
+ bM1 M2 M3 M1 gr + M1 φ(r)kLf kL1
   
≤ wr := 1 + bM1 M2 M3 M1 kx0 k + 1 + bM1 M2 M3 M1 gr
 
+ 1 + bM1 M2 M3 M1 kLf kL1 φ(r) + bM1 M2 M3 kx1 k.

It follows now that

k(Kxr )(τ )k
lim inf = 0
r→+∞ r
since
wr gr
lim inf = 0 = lim inf .
r→+∞ r r→+∞ r
This is clearly a contradiction to (∗). Consequently, there exists r > 0 such
that K(Br ) ⊂ Br .

Step 2. The operator K is continuous on Br .

To this end, let (xn )n≥1 ⊂ Br such that xn → x in Br . Set

(K1 x)(t) := R(t)[x0 + g(x)] and

Z t
R(t − s) f (s, x(s)) + Cux (s) ds for t ∈ I,
 
(K2 x)(t) :=
0
then K = K1 + K2 .
Therefore, since g is continuous, we have that

kK1 xn − K1 xk ≤ M1 kg(xn ) − g(x)k → 0, as n → +∞.

Controllability for Nonlocal Integrodifferential Equations 43

To see the continuity of K2 , we first set

Fn (s) = f (s, xn (s)) for every n and a.e. s , and F (s) = f (s, x(s)) for a.e. s .
Therefore by (H4 ) − (i), Fn (s) → F (s) and by (H4 ) − (ii),

F (s) = f (s, x (s)) ≤ Lf (s)φ(kxn k) ≤ φ(r)Lf (s)

n n

for every n and a.e. s.

It follows from Lebesgue dominated convergence theorem that
Z t
Fn (s) − F (s) ds −→ 0, as n → +∞, t ∈ I.

0

Moreover, we have that

Z t
1
K2 xn − K2 x ≤ M1 Fn (s) − F (s) ds + M1 M2 b 2 uxn − ux ,

0 L2 (I,U )

where
 Z b 
uxn − ux ≤ M3 M1 g(xn ) − g(x) + M1 Fn (s) − F (s) ds .

L2 (I,U ) 0

Thus it follows that K2 xn − K2 x −→ 0 as n → +∞, showing that K2 is

continuous on Br . Hence K is continuous on Br .
Step 3. The Mönch condition holds.
 
Suppose that D ⊆ Br is countable and D ⊆ co {0} ∪ K(D) . We have to
show that D is relatively compact.
To this end, it suffices to show that β(D) = 0, where β is the Hausdorff
MNC. Since D is countable, we can describe it as D = {xn }n≥1 . Therefore
K(D) = {Kxn }n≥1 (t) and its relative compactness implies that D is also
relatively compact. So we have to prove that K(D) is equibounded and
equicontinuous on I in order to use Ascoli-Arzela’s theorem. We show that
K(D) is equicontinuous. To this end, let y ∈ K(D), and 0 ≤ t1 < t2 ≤ b,
there is x ∈ D such that y = Kx and then

y(t2 ) − y(t1 ) ≤ R(t2 )x0 − R(t1 )x0 + R(t2 )g(x) − R(t1 )g(x)

Z t2 Z t1

+
R(t2 − s)f (s, x(s))ds − R(t1 − s)f (s, x(s))ds

0 0
Z t2 Z t1

+
R(t2 − s)Cu x (s)ds − R(t1 − s)Cu x (s)ds

0 0
Controllability for Nonlocal Integrodifferential Equations 44

Firstly for t1 > 0. By (ii) of definition 2.2.2, the first term on the right hand
side tends to 0 as |t2 − t1 | → 0. That is

R(t2 )x0 − R(t1 )x0 → 0 as |t2 − t1 | → 0.

Moreover, we have that

R(t2 )g(x) − R(t1 )g(x) ≤ R(t2 ) − R(t1 ) g(x)


≤ R(t2 ) − R(t1 ) gr ,

where gr = sup{kg(x)k : kxk ≤ r}.

And so R(t2 ) − R(t1 ) → 0 as |t2 − t1 | → 0, by the continuity of R(t) t≥0

for t > 0 in the operator-norm topology.

Now
for t1 = 0, we have that
R(h)g(x) − g(x) ≤ sup R(h)y − y → 0, as h → 0+ , by Lemma 2.3.5

y∈g(Br )
since g(Br ) is compact. Therefore

R(t2 )g(x) − R(t1 )g(x) → 0 as t2 → t1 .

Now we have that

Z t2 Z t1



R(t2 − s)f (s, x(s))ds − R(t1 − s)f (s, x(s))ds

0 0

Z t2
≤ M1 f (s, x(s)) ds

t1

Z t1
+ R(t2 − s) − R(t1 − s) f (s, x(s)) ds

0

Z t2
≤ M1 φ(r) Lf (s)ds
t1

Z t1
+φ(r) R(t2 − s) − R(t1 − s) Lf (s)ds

0

The right hand side tends to 0 as t2 →t1Z by Lebesgue dominated conver-

t 
gence theorem, showing that the family R(t − s)f (s, x(s))ds , x ∈ D
0
is equicontinuous. Also we have that
Controllability for Nonlocal Integrodifferential Equations 45

t2 t1
Z Z


R(t2 − s)Cux (s)ds − R(t1 − s)Cux (s)ds

0 0

Z t2
≤ M1 Cux (s) ds

t1

Z t1
+ R(t2 − s) − R(t1 − s) Cux (s) ds

0
h   i
≤ M2 M3 kx1 k + M1 kx0 k + gr + M1 φ(r)kLf kL1
 Z t1 
M1 (t2 − t1 ) + R(t2 − s) − R(t1 − s) ds

0

and the right hand side tends to 0 as t2 → t1 by Lebesgue dominated con-

vergence theorem, showing that the family
Z t 
R(t − s)Cux (s) ds ; x ∈ D
0

is equicontinuous and therefore the set K(D) is equicontinuous on I.

We now show that n K(D) is equibounded.
o To do this, we show that for all
t ∈ [0, b], the set K(x)(t); x ∈ D is relatively compact. We achieve this
by using the measure of noncompactness. We have that for t = 0, the set
n o n o
(Kx)(0) ; x ∈ D = x0 + g(x) ; x ∈ D = x0 + g(D)

is relatively compact in X. Since g is compact, then g(D) is compact also.

For t ∈ (0, b], we have that,
   
β {(K1 xn )(t)}n≥1 ≤ β {R(t)(x0 + g(xn ))}n≥1 = 0,

by the compactness of g. Also we have by (H3 ) − (ii) that

Controllability for Nonlocal Integrodifferential Equations 46

 
β {uxn (t) }n≥1 (t)

 n   Z b o 
−1
=β W x1 − R(b) x0 + g(xn ) − R(t − s)f (s, xn (s)) ds (t)
0 n≥1

n   Z b o 
≤ LW (t)β x1 − R(b) x0 + g(xn ) − R(t − s)f (s, xn (s)) ds (t)
0 n≥1

n  o 
≤ LW (t)β x1 − R(b) x0 + g(xn ) (t)
n≥1

n Z b o 
+LW (t)β R(t − s)f (s, xn (s)) ds (t) .
0 n≥1

By Lemma 2.3.3 and (H4 ) − (iii), we deduce that

  Z b   
β {uxn (t) }n≥1 (t) ≤ 2M1 LW (t) h(s) ds β {xn (s)}n≥1 (t)
0
Z b   
≤ 2M1 LW (t) h(s) ds β D(t) .
0

It follows that
 
β {(K2 xn )(t)}n≥1
!
Z t Z t 
=β R(t − s)f (s, xn (s)) ds + R(t − s)Cuxn (s) ds (t)
0 0 n≥1

Z b   
≤ 2M1 h(s) ds β D(t)
0

Z b  Z b   
+2M1 M2 LW (s) ds 2M a1 h(s) ds β D(t)
0 0
   
≤ 2M1 khkL1 β D(t) + 2M1 M2 kLW kL1 2M1 khkL1 β D(t)
    
≤ 1 + 2M1 M2 kLW kL1 2M1 khkL1 β D(t) .
Controllability for Nonlocal Integrodifferential Equations 47

It follows that
     
β K(D)(t) ≤ β K1 (D)(t) + β K2 (D)(t)
    
≤ 1 + 2M1 M2 kLW kL1 2M1 khkL1 β D(t) .
   
That is β K(D(t)) ≤ γβ D(t) . But from Mönch’s condition, we have
        
β D(t) ≤ β co {0} ∪ K(D(t)) = β K(D(t)) ≤ γβ D(t) .
   
This implies that β D(t) = 0, since γ < 1, which implies that β K(D)(t) =
0. This shows that K(D)(t) is compact, that is {K(x)(t); x ∈ D} is compact
as desired. So K(D) is equicontinuous and equibounded and therefore by
Ascoli-Arzela’s Theorem, we have that K(D) is relatively compact.
But
      
β D ≤ β co {0} ∪ K(D) = β K(D) = 0.
 
This implies that β D = 0, showing that D is compact in X as desired.
Thus D is relatively compact and the Mönch condition is satisfied. Therefore
by Theorem 2.4.2, K has a fixed point x in Br , which is a mild solution of
equation (3.1.1) and satisfies x(b) = x1 . And the proof is complete.
2
we now illustrate our main result by the following example.

3.3 Example of Application

Let Ω be bounded domain in Rn with smooth boundary and consider the
following nonlinear integrodifferential equation.
 Z t
∂v(t,ξ) e−t
 ∂t = ∆v(t, ξ) + ζ(t − s)∆v(s, ξ) ds + k+e t sin(v(t, ξ)) + aω(t, ξ)




 0



for t ∈ [0, b] = I and ξ ∈ Ω





v = 0 on ∂Ω







 Z Z b
  
1

ρ(t, ξ) log 1 + |v(t, η)| 2 dt dη for ξ ∈ Ω,

 v(0, ξ) = v0 (ξ) +

Ω 0
(3.3.1)
Controllability for Nonlocal Integrodifferential Equations 48

where a > 0, k ≥ 1, ω : I ×Ω → Ω continuous in t and ω(t, ξ) = 0 for all ξ ∈

∂Ω , ρ ∈ C(I × Ω) and ρ(t, ξ) = 0 for all ξ ∈ ∂Ω, and ζ ∈ W 1,1 (R+ , R)

Let X = U = C0 (Ω), the space of all continuous functions from Ω to R

vanishing on the boundary. We define A : D(A) ⊂ X → X by:
 n o
1

 D(A) = v ∈ C0 (Ω) ∩ H0 (Ω); ∆v ∈ C0 (Ω)
(3.3.2)

Av = ∆v,


for each v ∈ D(A).

Theorem 3.3.1 (Theorem 4.1.4, p. 82 of [76]) If Ω has a C 1 -boundary,

then the operator A defined as above, is the infinitesimal generator of a C0 -
semigroup of contractions on C0 (Ω).

By Theorem 8.4.1 above, A generates a C0 -semigroup T (t) t≥0 of contrac-

tions on C0 (Ω). Moreover, T (t) t≥0 generated by A above, is compact for

t > 0 and operator-norm continuous for t > 0. Then by Theorem 2.2.6, the
corresponding resolvent operator is operator-norm continuous. Define

∂v(t, ξ)
x(t)(ξ) = v(t, ξ), x0 (t)(ξ) = .
∂t
e−t
f (t, x)(ξ) = sin(x(t)(ξ)) for t ∈ I and ξ ∈ Ω
k + et
C : X → X be defined by Cu = aω.

(B(t)x)(ξ) = ζ(t)∆x(t)(ξ) for t ∈ I, x ∈ D(A) and ξ ∈ Ω.

g : C(I, X) → X be defined by
Z Z b  
1
g(x)(ξ) = ρ(t, ξ) log 1+|x(t)(η)| 2 dt dη for ξ ∈ Ω and x ∈ C(I, X).
Ω 0

Equation (3.3.1) is then transformed into the following form

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, x(t)) + Cu(t) for t ∈ I = [0, b],


0


x(0) = x0 + g(x).

(3.3.3)
Controllability for Nonlocal Integrodifferential Equations 49

f is Lipschitz continuous with respect to the second variable and moreover

we have that e−t
f (t, x) ≤ for (t, x) ∈ I × Ω.

k + et
Consequently, f satisfies (H4 ) − (i), (H4 ) − (ii) and (H4 ) − (iii), with
φ : R+ → R+ defined by φ(x) = 1.

Moreover,
    21
g(x) ≤ b mes(Ω) Mρ kxk ,


C0 (Ω)

where Mρ = max |ρ(t, ξ)|.

(t,ξ)∈I×Ω
n o gr
It is clear that with gr = sup kg(x)k : kxk∞ ≤ r , we have lim = 0.
r→+∞ r

Lemma 3.3.2 The map g : C(I, C0 (Ω)) → C0 (Ω) defined by

Z Z b  
1
g(x)(ξ) = ρ(t, ξ) log 1+|x(t)(η)| 2 dt dη for ξ ∈ Ω and x ∈ C(I, X),
Ω 0

is compact.

Proof. Let E ⊂ C(I, C0 (Ω)) be bounded. Then, by computing as above, we

have that
    12
g(x) ≤ b mes(Ω) M kxk , for all x ∈ E.

ρ
C0 (Ω)

So g(E) is bounded.
Now since ρ is uniformly continuous on I × Ω, it follows that g(E) is equicon-
tinuous on Ω. Therefore, by Ascoli-Arzela’s theorem, g(E) is relatively com-
pact in C0 (Ω). Hence, g is compact.

We have by Lemma 3.3.2 that g is compact and therefore satisfies (H5 ). Now
for ξ ∈ Ω, the operator W is given by
Z 1
(W u)(ξ) = R(1 − s)ωu(s, ξ) ds.
0

Assuming that W satisfies (H3 ), then all the conditions of Theorem 3.2.1
hold and equation (3.3.3) is controllable.
 CHAPTER 4

Controllability for some Partial Functional

Integrodifferential Equations with Finite Delay in Banach
Spaces

4.1 Introduction
In this chapter, we study the controllability for some systems that arise in
the analysis of heat flow in a rigid body of a material with memory [53, 85].
Materials with memory are interesting because they act adaptively to their
environment. They can be shaped easily into different forms at low temper-
ature, but return to their original forms on heating. Steering such systems
from an initial state (initial condition) to a desired terminal one (boundary
condition) by an appropriate choice of a control, is of interest to many sci-
entists and engineers.
These systems take the form of the following abstract model of partial func-
tional integrodifferential equation with finite delay in a Banach space (X, k·k)
1
The results of this chapter are contents of the following paper
- Khalil Ezzinbi and Patrice Ndambomve [44]

50
Controllability for integrodifferential Equations with Finite Delay 51

:
 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) + Cu(t) for t ∈ I = [0, b]


0


x0 = ϕ ∈ C = C([−r, 0]; X),

(4.1.1)
where f : I ×C → X is a function satisfying some conditions; A : D(A) →X
is the infinitesimal generator of a C0 -semigroup T (t) t≥0 on X; for t ≥ 0,

B(t) is a closed linear operator with domain D(B(t)) ⊃ D(A). The control
u belongs to L2 (I, U ) which is a Banach space of admissible controls, where
U is a Banach space. The operator C ∈ L(U, X), where L(U, X) denotes the
Banach space of bounded linear operators from U into X and C([−r, 0], X)
denotes the Banach space of continuous functions ϕ : [−r, 0] → X with
supremum norm kϕk = sup kϕ(θ)kX , xt denotes the history function that
θ∈[−r,0]
is defined by
xt (θ) = x(t + θ), for − r ≤ θ ≤ 0.
The particular cases in which B(t) = 0 and A = A(t) were considered by K.
Balachandran and R. Sakthivel; S. Baghli, M. Benchohra and K. Ezzinbi; S.
Selvi and M. M. Arjunan and many others. In this work, we extend the work
of K. Balachandran and R. Sakthivel by considering some integrodifferential
equation when B(t) 6= 0, and without any compactness assumption.
The controllability problem of nonlinear systems described by integrodiffer-
ential equations in infinite dimensional Banach spaces has been studied by
several authors: see for instance [8]-[16], [40, 55, 56, 69, 70, 83, 71] and the
references therein. Many authors have also studied the controllability prob-
lem of nonlinear systems with delay in inifinite dimensional Banach spaces:
see for instance [8], [55], [70], [83], [71], etc and the references therein. For
example in [83], the authors proved the controllability of semilinear func-
tional evolution equations with infinite delay using the nonlinear alternative
of Leray-Schauder type. In [56], M. Li, M. Wang and F. Zhang proved the
controllability of an impulsive functional differential system with finite delay
using Schaefer’s fixed-point theorem. In [71], S. Selvi and M. M. Arjunan
proved the controllability for impulsive differential systems with finite delay
using Mönch’s fixed-point Theorem, and in [8], K. Balachandran and R. Sak-
thivel studied the controllability of functional semilinear integrodifferential
systems in Banach spaces using Schaefer’s fixed point Theorem and the au-
thors assumed that the semigroup generated by the linear part is compact.
In many areas of applications such as Engineering, Electronics, Fluid Dy-
namics, Physical Sciences, etc..., integrodifferential equations appear and
Controllability for integrodifferential Equations with Finite Delay 52

have received considerable attention during the last decades. In recent years,
much work on the existence and regularity of solutions of nonlinear inte-
grodifferential equations with finite delay has been done by many authors
by applying the resolvent operator theory, see e.g., [43] and the references
therein.
In [85], R. Grimmer proved the existence and uniqueness of resolvent opera-
tors for these integrodifferential equations that give the variation of parame-
ters formula for the solution. In [24], W. Desch, R. Grimmer and W. Schap-
pacher proved the equivalence of the compactness of the resolvent operator
and that of the operator semigroup. In this work, we use the equivalence be-
tween the operator-norm continuity of the associated resolvent operator and
that of the operator semigroup. This property allows us to drop the com-
pactness assumption on the operator semigroup, considered by the authors
in [8, 83], and prove that the operator solution satisfies the Mönch condition.
The variation of parameters formula for the mild solutions of equation (4.1.1)
is given by the resolvent operator, and we prove the controllability result us-
ing the Mönch’s fixed-point Theorem and the measure of noncompactness.
This method enables us overcome the resolvent operator case considered in
this work. In contrary to the evolution semigroup case considered in [55, 71],
here the semigroup property can not be used because resolvent operators in
general do not form semigroups.
To the best of our knowledge, up to now no work has reported on control-
lability of partial functional integrodifferential equation (4.1.1) with finite
delay. It has been an untreated topic in the literature, and this fact is the
main aim and motivation of the present work.

4.2 Preliminary Results

Definition 4.2.1 A mild solution of equation (4.1.1) is a function x ∈
C([−r, b], X) satisfying the relation
 Z t
 
 R(t)ϕ(0) + R(t − s) f (s, xs ) + Cu(s) ds for t ∈ I,


x(t) = 0


ϕ(t) for − r ≤ t ≤ 0.

(4.2.1)

Definition 4.2.2 Equation (4.1.1) is said to be controllable on the interval

I if for every ϕ ∈ C and x1 ∈ X, there exists a control u ∈ L2 (I, U ) such
that a mild solution x of equation (4.1.1) satisfies the condition x(b) = x1 .
Controllability for integrodifferential Equations with Finite Delay 53

We now state the following useful result for equicontinuous subsets of C([a, b]; X),
where X is a Banach space.

Lemma 4.2.3 [12] Let M ⊂ C([a, b]; X) be bounded and equicontinuous.

Then β(M (t)) is continuous and

β(M ) = sup{β(M (t)); t ∈ [a, b]}, where M (t) = {x(t); x ∈ M }.

Lemma 4.2.4 [12] Let M ⊂ C([a, b]; X) be bounded and equicontinuous.

Then the set co(M ) is also bounded and equicontinuous.

To prove the controllability for equation (4.1.1), we need the following results.
Lemma 4.2.5 [40] If (un )n≥1 is a sequence of Bochner integrable functions
from I into a Banach space Y with the estimation kun (t)k ≤ µ(t) for almost
all t ∈ I and every n ≥ 1, where µ ∈ L1 (I, R), then the function

ψ(t) = β({un (t) : n ≥ 1})

belongs to L1 (I, R+ ) and satisfies the following estimation

Z t  Z t
β un (s)ds : n ≥ 1 ≤ 2 ψ(s)ds.
0 0

4.3 Controllability result

In this section, we give sufficient conditions ensuring the controllability of
equation (4.1.1). For that goal, we need to assume that:

(H3 )

(i) The following linear operator W : L2 (I, U ) → X defined by

Z b
Wu = R(b − s)Cu(s) ds,
0

is surjective so that it induces an isomorphism between L2 (I, U ) /KerW

and X again denoted by W with inverse W −1 taking values in L2 (I, U ) /KerW ,
(see e.g.,[68]).
Controllability for integrodifferential Equations with Finite Delay 54

(ii) There exists a function LW ∈ L1 (I, R+ ) such that for any bounded
set Q ⊂ X we have

β((W −1 Q)(t)) ≤ LW (t)β(Q),

where β is the Hausdorff MNC.

(H4 ) The function f : I × C −→ X satisfies the following two conditions:

(i) f (·, ϕ) is measurable for ϕ ∈ C and f (t, ·) is continuous for a.e t ∈ I,

(ii) for every positive integer q, there exists a function lq ∈ L1 (I, R+ )

such that
Z b
lq (t)
sup kf (t, ϕ)k ≤ lq (t) for a.e. t ∈ I and lim inf dt = l < +∞,
kϕkC ≤q q→+∞ 0 q

(iii) there exists a function h ∈ L1 (I, R+ ) such that for any bounded set
D ⊂ C([−r, 0], X),

β(f (t, D)) ≤ h(t)β(D) for a.e t ∈ I.

(H5 )   
γ = 1 + 2Mb M2 kLW kL1 2Mb khkL1 < 1,
where Mb = sup kR(t)k and M2 is such that M2 = kCk.
0≤t≤b

Theorem 4.3.1 Suppose that hypotheses
(H3 ) − (H5 ) hold and equation
(2.2.1) has a resolvent operator R(t) t≥0 that is continuous in the operator-
norm topology for t > 0. Then equation (4.1.1) is controllable on I provided
that
Mb (1 + Mb M2 M3 b)l < 1, (4.3.1)
where M3 is such that M3 = kW −1 k.

Proof. Using (H3 ) and given an arbitrary function x ∈ C([−r, b], X), we
define the control as usual by the following formula:
 Z b 
ux (t) = W −1
x1 − R(b)ϕ(0) − R(b − s)f (s, xs ) ds (t) for t ∈ I.
0
Controllability for integrodifferential Equations with Finite Delay 55

We define the following space

n o
Eb = x : [0, b] → X continuous such that x(0) = ϕ(0) .

For each x ∈ Eb we define its continuous extension xe from [−r, b] to X as

follows

 x(t) if t ∈ [0, b]


x
e(t) =
ϕ(t) if t ∈ [−r, 0]


We show using this control that the operator P : Eb → Eb defined by

Z t
es ) + Cux (s) ds for t ∈ I = [0, b]
 
(P x)(t) = R(t)ϕ(0) + R(t − s) f (s, x
0

has a fixed-point. This fixed point is then a mild solution of equation (4.1.1).
Observe that (P x)(b) = x1 . This means that the control ux steers the in-
tegrodifferential equation from ϕ to x1 in time b which implies the equation
(4.1.1) is controllable on I.
For each ϕ ∈ C, we define the function y ∈ C([0, b], X) by y(t) = R(t)ϕ(0)
and its extension ye in C([−r, b], X) by

 y(t) if t ∈ [0, b]


ye(t) =
ϕ(t) if t ∈ [−r, 0]


For each z ∈ C([0, b], X), set x

e(t) = ze(t) + ye(t), where ze is the extension by
zero of the function z on [−r, 0]. Observe that x satifies (4.2.1) if and only if
z(0) = 0 and
Z t
 
z(t) = R(t − s) f (s, zes + yes ) + Cuz (s) ds for t ∈ [0, b],
0

where
 Z b 
−1
uz (t) = W x1 − R(b)ϕ(0) − R(b − s)f (s, zes + yes ) ds (t).
0

Now let
n o
Eb0 = z : [0, b] → X continuous such that z(0) = 0 .

Thus Eb0 is a Banach space provided with the supremum norm. Define the
operator K : Eb0 → Eb0 by
Controllability for integrodifferential Equations with Finite Delay 56

Z t
R(t − s) f (s, zes + yes ) + Cuz (s) ds for t ∈ [0, b]
 
(Kz)(t) =
0

Note that the operator P has a fixed point if and only if K has one. So to
prove that P has a fixed point, we only need to prove that K has one.
For each positive number q, let Bq = {z ∈ Eb0 : kzk ≤ q}. We shall prove the
theorem in the following steps.

Step1. We claim that there exists q > 0 such that K(Bq ) ⊂ Bq . We proceed
by contradiction. Assume that it is not true. Then for each positive number
q, there exists a function z q ∈ Bq , such that K(z q ) ∈
/ Bq , i.e., k(Kz q )(τ )k > q
for some τ ∈ [0, b]. Now we have that

q
q < (Kz )(τ )

Z b Z b
q
≤ Mb f (s, z̃s + yes ) ds + Mb kCuzq (s)k ds

0 0
Z b Z b h Z b i
q −1 q
≤ Mb f (s, z̃s + yes ) ds + Mb BW x1 − R(b)ϕ(0) − R(b − s)f (s, z̃s ) ds ds

0 0 0
 Z b  Z b
≤ bMb M2 M3 kx1 k + Mb kϕk + Mb kf (s, z̃sq )k ds + Mb f (s, z̃sq + yes ) ds

0 0
 Z b  Z b
≤ bMb M2 M3 kx1 k + Mb kϕk + Mb lq0 (s) ds + Mb lq0 (s) ds,
0 0

where q 0 = q + q0 , with q0 = sup ke

y (s)k.
s∈[−r,b]
Hence
  Z b  
q ≤ 1 + Mb M2 M3 b Mb lq0 (s) ds + Mb M2 M3 b kx1 k + Mb kϕkC .
0

Dividing both sides by q and noting that q 0 = q + q0 → +∞ as q → +∞, we

obtain that

Z b   
   lq0 (s) ds  Mb M2 M3 b kx1 k + Mb kϕkC
0
1 ≤ 1 + Mb M2 M3 b Mb  +
 q  q
Controllability for integrodifferential Equations with Finite Delay 57

and Z b  Z b 
l (s) ds 
q0 l (s) ds
q0
 0
 0 q0 
lim inf   = lim inf   = l.
q→+∞  q  q→+∞  q0 q

 
Thus we have, 1 ≤ 1 + Mb M2 M3 b Mb l, and this contradicts (4.3.1). Hence
for some positive number q, we must have K(Bq ) ⊂ Bq .

Step2. K : Eb0 → Eb0 is continuous. In fact let K := K1 + K2 , where

Z t Z t
(K1 z)(t) = ys ) ds and
R(t−s)f (s, zes +e (K2 z)(t) = R(t−s)Cuz (s) ds.
0 0

Let {z n }n≥1 ⊂ Eb0 with z n → z in Eb0 . Then there exists a number q >
1 such that kz n (t)k ≤ q for all n and a.e. t ∈ I. So z n , z ∈ Bq . By
(H4 ) − (i), f (t, z̃tn + yet ) → f (t, zet + yet ) for each t ∈ [0, b]. And by (H4 ) − (ii),

kf (t, z̃tn + yet ) − f (t, zet + yet )k ≤ 2lq0 (t).

Then we have
Z b
n
kK1 z − K1 zkC ≤ Mb kf (s, z̃sn + yes ) − f (s, zes + yes )k ds −→ 0, as n → +∞
0

by dominated convergence Theorem. Also we have that

Z b
n 2
kK2 z − K2 zkC ≤ Mb M2 M3 b kf (s, z̃sn ) − f (s, zes )k ds −→ 0, as n → +∞
0

by dominated convergence Theorem. Thus

kKz n − Kzk ≤ kK1 z n − K1 zk + kK2 z n − K2 zk −→ 0, as n → +∞.

Hence K is continuous on Eb0 .

Controllability for integrodifferential Equations with Finite Delay 58

Step3. K(Bq ) is equicontinuous on [0, b]. In fact let t1 , t2 ∈ I, t1 < t2 and

z ∈ Bq , we have
Z t1
k(Kz)(t2 ) − (Kz)(t1 )k ≤ kR(t2 − s) − R(t1 − s)kkf (s, zes + yes ) + Cuz (s)k ds
0
Z t2
+ kR(t2 − s)kkf (s, zes + yes ) + Cuz (s)k ds
t
Z 1t1
≤ kR(t2 − s) − R(t1 − s)klq0 (s) ds
0
Z t1
+ kR(t2 − s) − R(t1 − s)kM2 M3 ×
0
 Z b 
kx1 k + Mb kϕk + Mb lq0 (τ ) dτ ds
0
Z t2
+ kR(t2 − s)klq0 (s) ds
t1
Z t2  Z b 
+ kR(t2 − s)kM2 M3 kx1 k + Mb kϕk + Mb lq0 (τ ) dτ ds
t1 0

By the continuity of R(t) t≥0 in the operator-norm toplogy, the dominated

convergence Theorem, we conclude that the right hand side of the above
inequality tends to zero and independent of z as t2 → t1 . Hence K(Bq ) is
equicontinuous on I.

Step4. We show that the Mönch’s condition holds.

Suppose that D ⊆ Bq is countable and D ⊆ co({0} ∪ K(D)). We shall show

that β(D) = 0, where β is the Hausdorff MNC. Without loss of generality,
we may assume that D = {z n }n≥1 . Since K maps Bq into an equicontinuous
family, K(D) is also equicontinuous on I.
Controllability for integrodifferential Equations with Finite Delay 59

By (H3 ) − (ii), (H4 ) − (iii) and Lemma 4.2.5, we have that

 Z b !
    
−1 n
β {uzn (t)}n≥1 = β W x1 − R(b)ϕ(0) − R(t − b)f s, {z̃s + yes }n≥1 ds (t)
0 n≥1

≤ LW (t)β ({x1 − R(b)ϕ(0)})

Z b !
  
n
+LW (t)β R(t − b)f s, {z̃s + yes }n≥1 ds
0 n≥1
Z b 
n


≤ 2Mb LW (t) h(s)β {z̃s }n≥1 + {e
ys } ds
0
Z h b  i 
n


≤ 2Mb LW (t) h(s) β {z̃s }n≥1 + β {e ys } ds
0
Z b  n o
h(s)β {z̃s }n≥1 ds , since yes : s ∈ [0, b] is compact
n


≤ 2Mb LW (t)
0
Z b 
n


≤ 2Mb LW (t) h(s) sup β {z̃s (θ)}n≥1 ds
0 −r≤θ≤0
 
by Lemma 4.2.3, since D = {z n }n≥1 is equicontinuous
Z b 
h(s) ds sup β {z n (t)}n≥1


≤ 2Mb LW (t)
0 0≤t≤b

This implies that

!
  Z t 
β {(Kz n )(t)}n≥1 ≤ β n
R(t − s)f (s, {z̃s + yes }n≥1 ) ds
0 n≥1
!
Z t 
+β R(t − s)uzn (s) ds
0 n≥1
Z b 
h(s) ds sup β {z n (t)}n≥1


≤ 2Mb
0 0≤t≤b
Z  b
Z b 
h(s) ds sup β {z n (t)}n≥1


+ 2Mb M2 LW (s) ds 2Mb
0 0 0≤t≤b
n


≤ 2Mb khkL1 sup β {z (t)}n≥1
0≤t≤b

+ 2Mb M2 kLW kL1 2Mb khkL1 sup β {z n (t)}n≥1 .

0≤t≤b
Controllability for integrodifferential Equations with Finite Delay 60

It follows that
     
β K(D)(t) ≤ 2Mb khkL1 sup β D(t) + 2Mb M2 kLW kL1 2Mb khkL1 sup β D(t)
0≤t≤b 0≤t≤b
   
≤ 1 + 2Mb M2 kLW kL1 2Mb khkL1 sup β D(t)
0≤t≤b
 
= γ sup β D(t)
0≤t≤b

Since D and K(D) are equicontinuous

  on 
[0, b] and D is bounded, it follows
by Lemma 4.2.3 that β K(D) ≤ γβ D , where γ is as defined in (H5 ).
Thus from the Mönch condition, we get that
       
β D ≤ β co({0} ∪ K(D) = β K(D) ≤ γβ D ,
 
and since γ < 1, this implies β D = 0, which implies that D is relatively
compact as desired in Bq and the Mönch condition is satisfied. We conclude
by Theorem 2.4.2, that K has a fixed point z in Bq . Then x = z + y is a
fixed point of P in Eb and thus equation (4.1.1) is controllable on [0, b].

we now illustrate our main result by the following example.

4.4 Example
Consider the following nonlinear integrodifferential equation.
 Z t
∂v ∂v ∂v
(t, ξ) = (t, ξ) + ζ(t − s) (s, ξ) ds + α(t) sin(v(t − r, ξ)) + ωu(t, ξ)





 ∂t ∂ξ 0 ∂ξ



for t ∈ [0, 1] = I and ξ ∈ [0, π]



v(t, 0) = v(t, π) = 0 for t ∈ [0, 1]








v(θ, ξ) = ψ(θ, ξ) for θ ∈ [−r, 0] and ξ ∈ [0, π].


(4.4.1)
where ω > 0, α ∈ C([0, 1]; R), u : I × [0, π] → [0, π] continuous in t and
u(t, 0) = u(t, π) = 0 and ζ ∈ W 1,1 (R+ , R+ )

Let X = U = C0 ([0, π]) be the space of all continuous functions from [0, π]
to R vanishing at 0 and at π.
Controllability for integrodifferential Equations with Finite Delay 61

We define A : D(A) ⊂ X → X by:

 n o
 D(A) = y ∈ X : y 0 exists and y 0 ∈ X



Ay = y 0 .


Then, A is the infinitesimal generator of a C0 -semigroup. Moreover, the

C0 -semigroup T (t) t≥0 generated by A above and defined by

T (t)y(s) = y(t + s) for y ∈ X,

is not compact but is operator-norm continuous for t > 0. Then by Theo-

rem 2.2.6, the corresponding resolvent operator is operator-norm continuous.
Define
x(t)(ξ) = v(t, ξ)
ϕ(θ)(ξ) = ψ(θ, ξ), θ ∈ [−r, 0], ξ ∈ [0, π].
f (t, ϕ)(ξ) = α(t) sin(ϕ(−r)(ξ)) for t ∈ I and ξ ∈ [0, π]
C : X → X be defined by Cu = ωu.

(B(t)x)(ξ) = ζ(t)∆v(t, ξ) for t ∈ I, x ∈ D(A) and ξ ∈ [0, π].

Equation (4.4.1) is then transformed into the following form

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) + Cu(t) for t ∈ I,


0


x0 = ϕ.

(4.4.2)
f is Lipschitz continuous with respect to the second variable, moreover we
have
sup f (t, ϕ) ≤ q |α(t)|.

kϕk≤q

So f satisfies (H4 ) − (i) and (H4 ) − (ii) with lq (t) = q |α(t)|. Also f satisfies
(H4 ) − (iii), since f is Lipschitz. Now for ξ ∈ , the operator W is given by
Z 1
(W u)(ξ) = ω R(1 − s)u(s, ξ) ds.
0

Assuming that W satisfies (H3 ), then all the conditions of Theorem 4.3.1
hold and equation (4.4.2) is controllable.
 CHAPTER 5

Controllability Results for some Partial Functional

Integrodifferential Equations with Infinite Delay in
Banach Spaces

5.1 Introduction
In this chapter, we study the controllability of the following partial functional
integrodifferential equation with infinite delay in a Banach space (X, k · k) :

 Z t
0
 x (t) = Ax(t) + γ(t − s)x(s)ds + f (t, xt ) + Cu(t) for t ∈ I = [0, b]


0


x0 = ϕ ∈ B,

(5.1.1)
where
A : D(A) → X is the infinitesimal generator of a C0 -semigroup
T (t) t≥0 on a Banach space X; for t ≥ 0, γ(t) is a closed linear opera-
tor with domain D(γ(t)) ⊃ D(A). The control u belongs to L2 (I, U ) which
is a Banach space of admissible controls, where U is a Banach space. The
operator C ∈ L(U, X), where L(U, X) denotes the Banach space of bounded
1
The results of this chapter are contents of the following paper
- Khalil Ezzinbi and Patrice Ndambomve [45]

62
Controllability for integrodifferential Equations with Finite Delay 63

linear operators from U into X, and the phase space B is a linear space of
functions mapping ] − ∞, 0] into X satisfying axioms which will be described
later, for every t ≥ 0, xt denotes the history function of B defined by

xt (θ) = x(t + θ) for − ∞ ≤ θ ≤ 0,

f : I × B → X is a continuous function satisfying some conditions. In the

literature devoted to equations with finite delay, the phase space is the space
of continuous functions on [−r, 0], for some r > 0, endowed with the uniform
norm topology. But when the delay is unbounded, the selection of the phase
space B plays an important role in both qualitative and quantitative theories.
A usual choice is a normed space satisfying some suitable axioms, which was
introduced by Hale and Kato [82].
The particular cases in which γ(t) = 0 and A = A(t) were considered by K.
Balachandran and R. Sakthivel; S. Baghli, M. Benchohra and K. Ezzinbi; S.
Selvi and M. M. Arjunan and many others. In this work, we extend the work
of K. Balachandran and R. Sakthivel by considering some integrodifferential
equation when γ(t) 6= 0, and without any compactness assumption.
The controllability problem of nonlinear systems described by integrodiffer-
ential equations in infinite dimensional Banach spaces has been studied by
several authors: see for instance [8]-[16], [40, 55, 56, 69, 70, 83, 71] and the
references therein. Many authors have also studied the controllability prob-
lem of nonlinear systems with delay in inifinite dimensional Banach spaces:
see for instance [8], [55], [70], [83], [71], etc and the references therein. For
example in [83], the authors proved the controllability of semilinear func-
tional evolution equations with infinite delay using the nonlinear alternative
of Leray-Schauder type. In [56], M. Li, M. Wang and F. Zhang proved the
controllability of an impulsive functional differential system with finite delay
using Schaefer’s fixed-point theorem. In [71], S. Selvi and M. M. Arjunan
proved the controllability for impulsive differential systems with finite delay
using Mönch’s fixed-point Theorem, and in [8], K. Balachandran and R. Sak-
thivel studied the controllability of functional semilinear integrodifferential
systems in Banach spaces using Schaefer’s fixed point Theorem and the au-
thors assumed that the semigroup generated by the linear part is compact.
In many areas of applications such as Engineering, Electronics, Fluid Dy-
namics, Physical Sciences, etc..., integrodifferential equations appear and
have received considerable attention during the last decades. In recent years,
much work on the existence and regularity of solutions of nonlinear inte-
grodifferential equations with finite delay has been done by many authors
by applying the resolvent operator theory, see e.g., [43] and the references
therein.
Controllability for integrodifferential Equations with Finite Delay 64

In [85], R. Grimmer proved the existence and uniqueness of resolvent opera-

tors for these integrodifferential equations that give the variation of parame-
ters formula for the solution. In [24], W. Desch, R. Grimmer and W. Schap-
pacher proved the equivalence of the compactness of the resolvent operator
and that of the operator semigroup. In this work, we use the equivalence be-
tween the operator-norm continuity of the associated resolvent operator and
that of the operator semigroup. This property allows us to drop the com-
pactness assumption on the operator semigroup, considered by the authors
in [8, 83], and prove that the operator solution satisfies the Mönch condition.
The variation of parameters formula for the mild solutions of equation (5.1.1)
is given by the resolvent operator, and we prove the controllability result us-
ing the Mönch’s fixed-point Theorem and the measure of noncompactness.
This method enables us overcome the resolvent operator case considered in
this work. In contrary to the evolution semigroup case considered in [55, 71],
here the semigroup property can not be used because resolvent operators in
general do not form semigroups.
To the best of our knowledge, up to now no work has reported on control-
lability of partial functional integrodifferential equation (5.1.1) with infinite
delay. It has been an untreated topic in the literature, and this fact is the
main aim and motivation of the present work.

5.2 Preliminary Results

In this section we introduce some definitions and Lemmas that will be used
throughout the paper.
In this work, we will employ an axiomatic definition of the phase space B
introduced by Hale and Kato in [82]. Thus, (B, k · kB ) will be a normed linear
space of functions mapping ] − ∞, 0] into X and satisfying the following
axioms:

(A1 ) There exist positive constant H and functions K : R+ → R+ con-

tinuous and M : R+ → R+ locally bounded, such that for a > 0, if
x : ] − ∞, a] → X is continuous on [0, a] and x0 ∈ B, then for every
t ∈ [0, a], the following conditions hold:

(i) xt ∈ B,

(ii) kx(t)k ≤ Hkxt kB , which is equivalent to kϕ(0)k ≤ HkϕkB for every

ϕ ∈ B,
Controllability for integrodifferential Equations with Finite Delay 65

(iii) kxt kB ≤ K(t) sup kx(s)k + M (t)kx0 kB .

0≤s≤t

(A2 ) For the function x in (A1 ), t → xt is a B-valued continuous function

for t ∈ [0, a].
(A3 ) The space B is complete.
Example [43] Let the spaces
BC the space of bounded continuous functions defined from (−∞, 0] to X;
BU C the space of bounded uniformly continuous functions defined from
(−∞, 0]nto X; o
C ∞ := φ ∈ BC : limθ→−∞ φ(θ) exists ;
n o
C := φ ∈ BC : limθ→−∞ φ(θ) = 0 , be endowed with the uniform norm
0

kφk = sup kφ(θ)k.

θ≤0

We have that the spaces BU C, C ∞ and C 0 satisfy conditions (A1 ) − (A3 ).

Definition 5.2.1 Let u ∈ L2 (I, U ) and ϕ ∈ B. A function x : ] − ∞, b] → X

is called a mild solution of equation (5.1.1) if x ∈ C([0, b]; X) and satisfies
the following integral equation
 Z t
 R(t)ϕ(0) + R(t − s) [f (s, xs ) + Cu(s)] ds for t ∈ I


x(t) = 0 (5.2.1)


ϕ(t) for − ∞ ≤ t ≤ 0.


Definition 5.2.2 Equation (5.1.1) is said to be controllable on the interval

I if for every ϕ ∈ B and x1 ∈ X, there exists a control u ∈ L2 (I, U ) such
that a mild solution x of equation (5.1.1) satisfies the condition x(b) = x1 .

Let
Rb = sup kR(t)k, Kb = sup kK(t)k, Mb = sup kM (t)k.
t∈[0,b] t∈[0,b] t∈[0,b]

We now state the following useful result for equicontinuous subsets of C([a, b]; X),
where X is a Banach space.

Lemma 5.2.3 [12] Let M ⊂ C([a, b]; X) be bounded and equicontinuous.

Then β(M (t)) is continuous and

β(M ) = sup{β(M (t)); t ∈ [a, b]}, where M (t) = {x(t); x ∈ M }.

Controllability for integrodifferential Equations with Finite Delay 66

Lemma 5.2.4 [12] Let M ⊂ C([a, b]; X) be bounded and equicontinuous.

Then the set co(M ) is also bounded and equicontinuous.

To prove the controllability for equation (5.1.1), we need the following results.

Lemma 5.2.5 [40] If (un )n≥1 is a sequence of Bochner integrable functions

from I into a Banach space Y with the estimation kun (t)k ≤ µ(t) for almost
all t ∈ I and every n ≥ 1, where µ ∈ L1 (I, R), then the function

ψ(t) = β({un (t) : n ≥ 1})

belongs to L1 (I, R+ ) and satisfies the following estimation

Z t  Z t
β un (s)ds : n ≥ 1 ≤ 2 ψ(s)ds.
0 0

5.3 Controllability result

In this section, we give sufficient conditions ensuring the controllability of
equation (5.1.1). For that goal, we need to assume that:

(H3 )

(i) The following linear operator W : L2 (I, U ) → X defined by

Z b
Wu = R(b − s)Cu(s) ds,
0

is surjective so that it induces an isomorphism between L2 (I, U ) /KerW

and X again denoted by W with inverse W −1 taking values in L2 (I, U ) /KerW ,
(see e.g.,[68]).

(ii) There exists a function LW ∈ L1 (I, R+ ) such that for any bounded
set Q ⊂ X we have

β((W −1 Q)(t)) ≤ LW (t)β(Q),

where β is the Hausdorff MNC.

(H4 ) The function f : I × B −→ X satisfies the following two conditions:

Controllability for integrodifferential Equations with Finite Delay 67

(i) f (·, ϕ) is measurable for ϕ ∈ B and f (t, ·) is continuous for a.e t ∈ I,

(ii) for every positive integer q, there exists a function lq ∈ L1 (I, R+ )

such that
Z b
lq (t)
sup kf (t, ϕ)k ≤ lq (t) for a.e. t ∈ I and lim inf dt = l < +∞,
kϕkB ≤q q→+∞ 0 q

(iii) there exists a function h ∈ L1 (I, R+ ) such that for any bounded set
D ⊂ B,

β(f (t, D)) ≤ h(t) sup β(D(θ)) for a.e t ∈ I,

−∞<θ≤0

where
D(θ) = {φ(θ) : φ ∈ D}.

(H5 )   
τ = 1 + 2Rb M2 kLW k L1 2Rb khk L1 < 1,
where Rb = sup kR(t)k and M2 is such that M2 = kCk.
0≤t≤b

Theorem 5.3.1 Suppose that hypotheses
(H3 ) − (H5 ) hold and equation
(2.2.1) has a resolvent operator R(t) t≥0 that is continuous in the operator-
norm topology for t > 0. Then equation (5.1.1) is controllable on I provided
that
Rb (1 + Rb M2 M3 b)Kb l < 1, (5.3.1)
where M3 is such that M3 = kW −1 k.

Proof. Using (H3 ) and given an arbitrary function x, we define the control
as usual by the following formula:
 Z b 
ux (t) = W −1
x1 − R(b)ϕ(0) − R(b − s)f (s, xs ) ds (t) for t ∈ I.
0

For each x ∈ C([0, b], X) such that x(0) = ϕ(0), we define its extension x
e
from ] − ∞, b] to X as follows

 x(t) if t ∈ [0, b]


x
e(t) =
ϕ(t) if t ∈] − ∞, 0]

Controllability for integrodifferential Equations with Finite Delay 68

We define the following space

n o
Eb = x :] − ∞, b] → X such that x|I ∈ C([0, b], X) and x0 ∈ B .

where x|I is the restriction of x to I.

We show using this control that the operator P : Eb → Eb defined by

Z t
es ) + Cux (s) ds for t ∈ I = [0, b]
 
(P x)(t) = R(t)ϕ(0) + R(t − s) f (s, x
0

has a fixed-point. This fixed point is then a mild solution of equation (5.1.1).
Observe that (P x)(b) = x1 . This means that the control ux steers the in-
tegrodifferential equation from ϕ to x1 in time b which implies the equation
(5.1.1) is controllable on I.
For each ϕ ∈ B, we define the function y ∈ C([0, b], X) by y(t) = R(t)ϕ(0)
and its extension ye on ] − ∞, 0] by

 y(t) if t ∈ [0, b]


ye(t) =
ϕ(t) if t ∈] − ∞, 0]


For each z ∈ C([0, b], X), set x

e(t) = ze(t) + ye(t), where ze is the extension by
zero of the function z on ] − ∞, 0]. Observe that x satifies (5.2.1) if and only
if z(0) = 0 and
Z t
 
z(t) = R(t − s) f (s, zes + yes ) + Cuz (s) ds for t ∈ [0, b],
0

where
 Z b 
−1
uz (t) = W x1 − R(b)ϕ(0) − R(b − s)f (s, zes + yes ) ds (t).
0

Now let n o
Eb0 = z ∈ Eb such that z0 = 0 .
Thus Eb0 is a Banach space provided with the supremum norm. Define the
operator Γ : Eb0 → Eb0 by

Z t
R(t − s) f (s, zes + yes ) + Cuz (s) ds for t ∈ [0, b]
 
(Γz)(t) =
0

Note that the operator P has a fixed point if and only if Γ has one. So to
prove that P has a fixed point, we only need to prove that Γ has one.
Controllability for integrodifferential Equations with Finite Delay 69

For each positive number q, let Bq = {z ∈ Eb0 : kzk ≤ q}. Then, for any
z ∈ Bq , we have by axiom (A1 ) that
kzs + ys k ≤ kzs kB + kys kB
≤ K(s)kz(s)k + M (s)kz0 kB + K(s)ky(s)k + M (s)ky0 kB
≤ Kb kz(s)k + Kb kR(t)kkϕ(0)k + Mb kϕkB
≤ Kb kz(s)k + Kb Rb HkϕkB + Mb kϕkB
 
≤ Kb kz(s)k + Kb Rb H + Mb kϕkB
 
≤ Kb q + Kb Rb H + Mb kϕkB

Thus,  
kzs + ys k ≤ Kb q + Kb Rb H + Mb kϕkB =: q 0 .
We shall prove the theorem in the following steps.

Step1. We claim that there exists q > 0 such that Γ(Bq ) ⊂ Bq . We proceed
by contradiction. Assume that it is not true. Then for each positive number
q, there exists a function z q ∈ Bq , such that Γ(z q ) ∈
/ Bq , i.e., k(Γz q )(t)k > q
for some t ∈ [0, b]. Now we have that

q
q < (Γz )(t)

Z b Z b
q
≤ Rb f (s, z̃s + yes ) ds + Rb kCuzq (s)k ds

0 0
Z b Z b h Z b i
q −1 q
≤ Rb f (s, z̃s + yes ) ds + Rb CW x1 − R(b)ϕ(0) − R(b − s)f (s, z̃s ) ds ds

0 0 0
 Z b  Z b
≤ bRb M2 M3 kx1 k + Rb kϕ(0)k + Rb kf (s, z̃sq )k ds + Rb f (s, z̃sq + yes ) ds

0 0
 Z b  Z b
≤ bRb M2 M3 kx1 k + Rb HkϕkB + Rb lq0 (s) ds + Rb lq0 (s) ds,
0 0
 
where q 0 := Kb q + q0 , with q0 := Kb Rb H + Mb kϕkB .
Hence
  Z b  
q ≤ 1 + Rb M2 M3 b Rb lq0 (s) ds + Rb M2 M3 b kx1 k + Rb HkϕkB .
0

Dividing both sides by q and noting that q 0 = Kb q + q0 → +∞ as q → +∞,

we obtain that
Controllability for integrodifferential Equations with Finite Delay 70

Z b   
   l (s) ds  Rb M2 M3 b kx1 k + Rb HkϕkB
q0
0
1 ≤ 1 + Rb M2 M3 b Rb  +
 q  q

and Z b  Z b 
lq0 (s) ds  lq0 (s) ds
 0
 0 q0 
lim inf   = lim inf   = lKb .
q→+∞  q  q→+∞  q0 q

 
Thus we have, 1 ≤ 1+Rb M2 M3 b Rb Kb l, and this contradicts (5.3.1). Hence
for some positive number q, we must have Γ(Bq ) ⊂ Bq .

Step2. Γ : Eb0 → Eb0 is continuous. In fact let Γ := Γ1 + Γ2 , where

Z t Z t
(Γ1 z)(t) = ys ) ds and
R(t−s)f (s, zes +e (Γ2 z)(t) = R(t−s)Cuz (s) ds.
0 0

Let {z n }n≥1 ⊂ Eb0 with z n → z in Eb0 . Then there exists a number q >
1 such that kz n (t)k ≤ q for all n and a.e. t ∈ I. So z n , z ∈ Bq . By
(H4 ) − (i), f (t, z̃tn + yet ) → f (t, zet + yet ) for each t ∈ [0, b]. And by (H4 ) − (ii),

kf (t, z̃tn + yet ) − f (t, zet + yet )k ≤ 2lq0 (t).

Then we have
Z b
n
kΓ1 z − Γ1 zkC ≤ Rb kf (s, z̃sn + yes ) − f (s, zes + yes )k ds −→ 0, as n → +∞
0

by dominated convergence Theorem. Also we have that

Z b
n 2
kΓ2 z − Γ2 zkC ≤ Rb M2 M3 b kf (s, z̃sn ) − f (s, zes )k ds −→ 0, as n → +∞
0

by dominated convergence Theorem. Thus

kΓz n − Γzk ≤ kΓ1 z n − Γ1 zk + kΓ2 z n − Γ2 zk −→ 0, as n → +∞.

Hence Γ is continuous on Eb0 .

Controllability for integrodifferential Equations with Finite Delay 71

Step3. Γ(Bq ) is equicontinuous on [0, b]. In fact let t1 , t2 ∈ I, t1 < t2 and

z ∈ Bq , we have
Z t1
k(Γz)(t2 ) − (Γz)(t1 )k ≤ kR(t2 − s) − R(t1 − s)kkf (s, zes + yes ) + Cuz (s)k ds
0
Z t2
+ kR(t2 − s)kkf (s, zes + yes ) + Cuz (s)k ds
t
Z 1t1
≤ kR(t2 − s) − R(t1 − s)klq0 (s) ds
0
Z t1
+ kR(t2 − s) − R(t1 − s)kM2 M3 ×
0
 Z b 
kx1 k + Rb HkϕkB + Rb lq0 (τ ) dτ ds
0
Z t2
+ kR(t2 − s)klq0 (s) ds
t1
Z t2  Z b 
+ kR(t2 − s)kM2 M3 kx1 k + Rb HkϕkB + Rb lq0 (τ ) dτ ds
t1 0

By the continuity of R(t) t≥0 in the operator-norm toplogy, the dominated

convergence Theorem, we conclude that the right hand side of the above
inequality tends to zero and independent of z as t2 → t1 . Hence Γ(Bq ) is
equicontinuous on I.

Step4. We show that the Mönch’s condition holds.

Suppose that D ⊆ Bq is countable and D ⊆ co({0} ∪ Γ(D)). We shall show

that β(D) = 0, where β is the Hausdorff MNC. Without loss of generality,
we may assume that D = {z n }n≥1 . Since Γ maps Bq into an equicontinuous
family, Γ(D) is also equicontinuous on I.
Controllability for integrodifferential Equations with Finite Delay 72

By (H3 ) − (ii), (H4 ) − (iii) and Lemma 5.2.5, we have that

 Z b !
    
−1 n
β {uzn (t)}n≥1 = β W x1 − R(b)ϕ(0) − R(t − b)f s, {z̃s + yes }n≥1 ds (t)
0 n≥1

≤ LW (t)β ({x1 − R(b)ϕ(0)})

!
Z b   
+LW (t)β R(t − b)f s, {z̃sn + yes }n≥1 ds
0 n≥1
Z b 
h(s)β {z̃sn }n≥1 + {e


≤ 2Rb LW (t) ys } ds
0
Z h b  i 
n


≤ 2Rb LW (t) h(s) β {z̃s }n≥1 + β {e ys } ds
0
Z b  n o
h(s)β {z̃s }n≥1 ds , since yes : s ∈ [0, b] is compact
n


≤ 2Rb LW (t)
0
Z b 
n


≤ 2Rb LW (t) h(s) sup β {z̃s (θ)}n≥1 ds
0 −∞<θ≤0
 
by Lemma 5.2.3, since D = {z n }n≥1 is equicontinuous
Z b 
h(s) ds sup β {z n (t)}n≥1


≤ 2Rb LW (t)
0 0≤t≤b

This implies that

!
  Z t 
β {(Γz n )(t)}n≥1 ≤ β R(t − s)f (s, {z̃sn + yes }n≥1 ) ds
0 n≥1
!
Z t 
+β R(t − s)uzn (s) ds
0 n≥1
Z b 
sup β {z n (t)}n≥1


≤ 2Rb h(s) ds
0 0≤t≤b
Z b  Z b 
h(s) ds sup β {z n (t)}n≥1


+ 2Rb M2 LW (s) ds 2Rb
0 0 0≤t≤b
n


≤ 2Rb khkL1 sup β {z (t)}n≥1
0≤t≤b

+ 2Rb M2 kLW kL1 2Rb khkL1 sup β {z n (t)}n≥1 .

0≤t≤b
Controllability for integrodifferential Equations with Finite Delay 73

It follows that
     
β Γ(D)(t) ≤ 2Rb khkL1 sup β D(t) + 2Rb M2 kLW kL1 2Rb khkL1 sup β D(t)
0≤t≤b 0≤t≤b
   
≤ 1 + 2Rb M2 kLW kL1 2Rb khkL1 sup β D(t)
0≤t≤b
 
= τ sup β D(t)
0≤t≤b

Since D and Γ(D) are equicontinuous

  on 
[0, b] and D is bounded, it follows
by Lemma 5.2.3 that β Γ(D) ≤ τ β D , where τ is as defined in (H5 ).
Thus from the Mönch condition, we get that
       
β D ≤ β co({0} ∪ Γ(D) = β Γ(D) ≤ τ β D ,
 
and since τ < 1, this implies β D = 0, which implies that D is relatively
compact as desired in Bq and the Mönch condition is satisfied. We conclude
by Theorem 2.4.2, that Γ has a fixed point z in Bq . Then x = z + y is a fixed
point of P in Eb and thus equation (5.1.1) is controllable on [0, b].
we now illustrate our main result by the following example.

5.4 Example
Consider the following nonlinear integrodifferential equation.

 Z t Z 0
∂v ∂v ∂v
(t, ξ) = (t, ξ) + ζ(t − s) (s, ξ) ds + α(θ)g(t, v(t + θ, ξ)) dθ + ηω(t, ξ)


∂t ∂ξ ∂ξ



 0 −∞



for t ∈ I = [0, 1] and ξ ∈ [0, π]



v(t, 0) = v(t, π) = 0 for t ∈ [0, 1]









v(θ, ξ) = φ(θ, ξ) for θ ∈] − ∞, 0] and ξ ∈ [0, π],

(5.4.1)
where η > 0, g : [0, 1] × R → R is continuous and Lipschitzian with respect
to the second variable, the initial data function φ : R− × [0, π] → R is a given
function, ω : [0, 1] × [0, π] → R continuous in t and ω(t, 0) = ω(t, π) = 0,
α : R− → R is continuous, α ∈ L1 (R− , R) and ζ ∈ W 1,1 (R+ , R+ ).
Let X = U = C0 ([0, π], R), the space of all continuous functions from [0, π]
to R vanishing at 0 and π equipped with the uniform norm topology, and
Controllability for integrodifferential Equations with Finite Delay 74

the phase space B = BU C(R− , X), the the space of uniformly bounded
continuous functions endowed with the following norm

kϕkB = sup kϕ(θ)k.

θ≤0

Then, the space BU C(R− , X) satisfies axioms (A1 ), (A2 ) and (A3 ).

We define A : D(A) ⊂ X → X by:

 n o
 D(A) = y ∈ X : y 0 exists and y 0 ∈ X



Ay = y 0 .


Then, A is the infinitesimal generator of a C0 -semigroup. Moreover, the

C0 -semigroup T (t) t≥0 generated by A above and defined by

T (t)y(s) = y(t + s) for y ∈ X,

is operator-norm continuous for t > 0. Thus by Theorem 2.2.6, the corre-

sponding resolvent operator is operator-norm continuous for t > 0.
Now define
∂v(t, ξ)
x(t)(ξ) = v(t, ξ), x0 (t)(ξ) = , ω(t, ξ) = u(t)(ξ).
∂t
ϕ(θ)(ξ) = φ(θ, ξ) for θ ∈] − ∞, 0] and ξ ∈ [0, π].
Z 0
f (t, ψ)(ξ) = α(θ)g(t, ψ(θ)(ξ)) dθ for θ ∈] − ∞, 0] and ξ ∈ [0, π].
−∞
 
C : X → X be defined by Cu(t) (ξ) = Cu(t)(ξ) = ηω(t, ξ).

(γ(t)x)(ξ) = ζ(t)∆v(t, ξ) for t ∈ [0, 1], x ∈ D(A) and ξ ∈ [0, π].

We suppose that ϕ ∈ BU C(R− , X). Then, equation (5.4.1) is then trans-
formed into the following form:
 Z t
0
 x (t) = Ax(t) + γ(t − s)x(s)ds + f (t, xt ) + Cu(t) for t ∈ I = [0, 1],


0


x0 = ϕ ∈ B.

(5.4.2)
Suppose there exists a continuous function p ∈ L (I; R ) such that
1 +

|g(t, y1 ) − g(t, y2 )| ≤ p(t)|y1 − y2 | for t ∈ I and y1 , y2 ∈ R.

Controllability for integrodifferential Equations with Finite Delay 75

and
g(t, 0) = 0 for t ∈ I.
One can see that f is Lipschitz continuous with respect to the second variable
and moreover for ϕ ∈ B, we have we have

sup f (t, ϕ) ≤ q kαk p(t).

kϕkB ≤q

So f satisfies (H4 ) − (i) and (H4 ) − (ii) with lq (t) = q kαk p(t). Also f
satisfies (H4 ) − (iii) by condition (viii) of Theorem 2.3.2, since f is Lipschitz.
Now for ξ ∈ [0, π], the operator W is given by
Z 1
(W u)(ξ) = η R(1 − s)ω(s, ξ) ds.
0

Assuming that W satisfies (H3 ), then all the conditions of Theorem 5.3.1
hold and equation (5.4.2) is controllable.
 Part II

Optimal Controls of some Partial

Functional Integrodifferential
Equations in Banach Spaces

76
 CHAPTER 6

Solvability and Optimal Control for some Partial

Functional Integrodifferential Equations with Finite Delay

6.1 Introduction
The aim of this chapter is to study the existence of mild solutions and the
optimal controls of some systems that take the form of the following partial
functional integrodifferential equation with finite delay in a Banach space
(X, k · k).

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) + C(t)u(t) for t ∈ I = [0, b]


0


x0 = ϕ ∈ C = C([−r, 0]; X),

(6.1.1)
where f : I ×C → X is a function satisfying some conditions; A : D(A) → X
is the infinitesimal generator of a C0 -semigroup T (t) t≥0 on a separable re-

flexive Banach space X; for t ≥ 0, B(t) is a closed linear operator with

1
The results of this chapter are contents of the following paper
- Khalil Ezzinbi and Patrice Ndambomve [91]

77
Optimal Controls for Integrodifferential Equations with Finite Delay 78

domain D(B(t)) ⊃ D(A). The control u(t) takes values from another sepa-
rable reflexive Banach space U . The operator C(t) belongs to L(U, X), the
Banach space of bounded linear operators from U into X, and C([−r, 0], X)
denotes the Banach space of continuous functions ϕ : [−r, 0] → X with
supremum norm kϕk = sup kϕ(θ)k, xt denotes the history function of C
θ∈[−r,0]
defined by
xt (θ) = x(t + θ) for θ ∈ [−r, 0].
In many areas of applications such as engineering, electronics, fluid dynam-
ics, physical sciences, etc..., integrodifferential equations appear and have
received considerable attention during the last decades. In [85], R. Grimmer
has proved the existence and uniqueness of resolvent operators for these in-
tegrodifferential equations that give the variation of parameter formula for
the solution. In recent years, much work has been done on the existence and
regularity of solutions of nonlinear integrodifferential equations with finite
delay by many authors by applying the resolvent operator theory, for inte-
gral equations see e.g., [86] and the references therein. Problems of control-
lability and existence of optimal controls for nonlinear differential equations
have been studied extensively by many authors under various hypotheses (see
e.g., [87], [88],[89],[90],[94] [95]), but little is known and done about the ex-
istence of optimal controls for integrodifferential equations with delay using
the resolvent operator theory. In [96], the authors studied the existence and
continuous dependence of mild solutions and the optimal controls of a La-
grange problem for some fractional integrodifferential equation with infinite
delay in Banach spaces using the using the techniques of a priori estimation
and extension of step by steps. Wand and Zhou [97] discussed the optimal
controls of a Lagrange problem for fractional evolution equations. In [98],
Wei et al. studied the optimal controls for nonlinear impulsive integrodiffer-
ential equations of mixed type on Banach spaces. In [99], the authors studied
the existence of mild solutions and the optimal controls of a Lagrange prob-
lem for some impulsive fractional semilinear differential equations, using the
techniques of a priori estimation. Motivated by these works, we investigate
the solvability and the existence of optimal controls of a Lagrange problem
for equation (6.1.1), using the techniques of a priori estimation of mild solu-
tions. The existence and uniqueness of mild solutions is obtained using the
theory of resolvent operator for integral equations. Furthermore, to the best
of our knowledge, the optimal controls for partial functional integrodifferen-
tial equation (6.1.1) with finite delay are untreated in the literature, and this
fact motivates us to extend the existing ones and make new development of
the present work on this issue.
Optimal Controls for Integrodifferential Equations with Finite Delay 79

6.2 Existence of mild solutions for equation (6.1.1)

We make the following assumptions.

(H3 ) The function f : I × C → X satisfies the following conditions:

(i) f (·, ψ) is measurable for ψ ∈ C,

(ii) for any ρ > 0, there exists Lf (ρ) > 0 such that

kf (t, ψ1 )−f (t, ψ2 )k ≤ Lf (ρ)kψ1 −ψ2 k for kψ1 k ≤ ρ, kψ2 k ≤ ρ and t ∈ [0, b],

(iii) there exists af > 0 such that

kf (t, ψ)k ≤ af (1 + kψk) for all ψ ∈ C and t ∈ [0, b].

(H4 ) Let U be the separable reflexive Banach space from which the control
u takes values and assume C ∈ L∞ (I; L(U, X)).

(H5 ) The multivalued map Γ : I → 2U \{∅} has closed, convex, and bounded
values, Γ is graph measurable, and Γ (·) ⊆ Ω where Ω is a bounded set
in U .
We denote by Uad the set of admissible controls defined by:
n o
Uad = u : I → U such that u is measurable and u(t) ∈ Γ (t), a.e. .

Then, we have the following:

Theorem 6.2.1 [96] Uad 6= ∅ and Uad ⊂ L2 (I, U ) is bounded, closed and
convex. Also, Cu ∈ L2 (I, U ) for all u ∈ Uad .

Definition 6.2.2 Let u ∈ Uad and ϕ ∈ C. A function x ∈ C([−r, b], X) is

called a mild solution of equation (6.1.1) if

 Z t
 R(t)ϕ(0) + R(t − s) [f (s, xs ) + C(s)u(s)] ds for t ∈ I


x(t) = 0


ϕ(t) for − r ≤ t ≤ 0.

(6.2.1)

We have the following Theorem on existence of mild solutions to equation

(6.1.1) with respect to a given control u ∈ Uad .
Optimal Controls for Integrodifferential Equations with Finite Delay 80

Theorem 6.2.3 Assume that (H1 ) − (H5 ) hold. Then for each u ∈ Uad ,
equation (6.1.1) has a unique mild solution on [−r, b].

Proof. Let b1 ≤ b, ρ > 0, and ψ ∈ C such that kψk ≤ ρ. For t ∈ [0, b1 ], we

have by the local Lipschitz condition on f that

kf (t, ψ)k ≤ Lf (ρ)kψk + kf (t, 0)k ≤ Lf (ρ)ρ + sup kf (s, 0)k.

s∈[0,b1 ]

b1 will be chosen sufficiently small enough to get the local existence of mild
solutions.
Let ϕ ∈ C, ρ = kϕk + 1 and ρ∗ = Lf (ρ)ρ + sups∈[0,b1 ] kf (s, 0)k.
We define the following space
n o
Eϕ = x ∈ C([−r, b1 ]; X) such that x(θ) = ϕ(θ) for θ ∈ [−r, 0] and sup kx(s)−ϕ(0)k ≤ 1 .
s∈[0,b1 ]

For x ∈ Eϕ , one can see that kxt k ≤ 1 + kϕk = ρ.

Then, Eϕ is a closed subset of C([−r, b1 ]; X) which is endowed with the
uniform norm topology. Let

Mb = sup kR(t)k.
t∈[0,b]

Define the operator K : Eϕ → C([−r, b1 ]; X) by

 Z t
R(t − s) f (s, xs ) + C(s)u(s) ds for t ∈ [0, b1 ]
 
 R(t)ϕ(0) +


(Kx)(t) = 0

ϕ(t) for t ∈ [−r, 0]



 
We claim that K Eϕ ⊂ Eϕ . In fact let x ∈ Eϕ and t ∈ [0, b1 ]. Then,

k(Kx)(t) − ϕ(0)k ≤ kR(t)ϕ(0) − ϕ(0)k

Z t
 
+ R(t − s) f (s, xs ) + C(s)u(s) ds

0
√
≤ kR(t)ϕ(0) − ϕ(0)k + Mb ρ∗ t + Mb kCk kukL2 t.

Now, choose b1 sufficiently small such that

n √ o
sup kR(s)ϕ(0) − ϕ(0)k + Mb ρ∗ s + Mb kCk kukL2 s < 1. (6.2.2)
s∈[0,b1 ]
Optimal Controls for Integrodifferential Equations with Finite Delay 81

Consequently,
√
k(Kx)(t)−ϕ(0)k ≤ kR(t)ϕ(0)−ϕ(0)k+Mb ρ∗ t+Mb kCk kukL2 t < 1 for t ∈ [0, b1 ].
 
Hence, K Eϕ ⊂ Eϕ .
Let x, y ∈ Eϕ and t ∈ [0, b1 ]. Then, kxs k, kys k ≤ ρ for s ∈ [0, b1 ] and we
have
Z t
k(Kx)(t) − (Ky)(t)k ≤ Mb kf (s, xs ) − f (s, ys )k ds
0
Z t
≤ Mb Lf (ρ) kxs − ys k ds
0
Z t
≤ Mb Lf (ρ) sup kx(τ ) − y(τ )k dτ
0 τ ∈[0,s]

≤ Mb Lf (ρ) b1 kx − yk

Now, since

∗
n
∗
√ o
Mb Lf (ρ) b1 ≤ Mb ρ b1 < sup kR(s)ϕ(0)−ϕ(0)k+Mb ρ s+Mb kCk kukL2 s .
s∈[0,b1 ]

Condition (6.2.2) implies that

Mb Lf (ρ) b1 < 1.

Thus, K is a strict contraction on Eϕ . It follows from the contraction map-

ping principle that K has a unique fixed point x ∈ Eϕ , which is the unique
mild solution of equation (6.1.1) with respect to u on [−r, b1 ].
Using the same arguments, we can show that x can be extended to a maximal
interval of existence [0, tmax [.

Lemma 6.2.4 [86] If tmax < b, then, lim sup kx(t)k = ∞.

t→tmax

We show that tmax = b. Assume on the contrary that tmax < b. Then for
t ∈ [0, tmax ] we have that
Z t
 
x(t) = R(t)ϕ(0) + R(t − s) f (s, xs ) + C(s)u(s) ds.
0
Optimal Controls for Integrodifferential Equations with Finite Delay 82

It follows that
Z t Z t
kx(t)k ≤ Mb kϕ(0)k + Mb kf (s, xs )k ds + Mb kC(s)u(s)k ds
0 0
Z t Z t
≤ Mb kϕk + Mb tmax af + Mb af kxs k ds + Mb kCk ku(s)k ds
0 0

√
Z t
≤ Mb kϕk + Mb tmax af + Mb tmax kCkkukL2 + Mb af kxs k ds
0

√
Z t
≤ Mb kϕk + Mb tmax af + Mb tmax kCkkukL2 + Mb af sup kx(τ )k dτ
0 τ ∈[−r,s]

This implies that

√
Z t
sup kx(s)k ≤ Mb kϕk+Mb tmax af +Mb tmax kCkkukL2 +Mb af sup kx(τ )k dτ
s∈[−r,t] 0 τ ∈[−r,s]

It follows by Gronwall’s inequality that

kx(t)k ≤ β ∗ eMb af t for t ∈ [0, tmax ],

√
where β ∗ = Mb kϕk + Mb tmax af + Mb tmax kCkkukL2 .
Thus
lim kx(t)k ≤ β ∗ eMb af tmax < ∞.
t→tmax

This contradicts Lemma 6.2.4. Therefore, tmax = b and hence, equation

(6.1.1) has a unique mild solution on [−r, b].

6.3 Continuous Dependence

In this section, we discuss the continuous dependence of the mild solutions
of equation (6.1.1) on the controls and initial states.
We have the following a priori estimation.

Lemma 6.3.1 Suppose (H1 ) − (H3 ) hold and assume that equation (6.1.1)
has a mild solution xu on [−r, b] with respect to u ∈ Uad . Then, there exists
a constant ρ > 0 independent of u such that kxu (t)k ≤ ρ for t ∈ [0, b], (ρ
depends only on Uad and ϕ).
Optimal Controls for Integrodifferential Equations with Finite Delay 83

Proof. Let ϕ ∈ C. We define the following space

n o
Eb = x : [0, b] → X continuous such that x(0) = ϕ(0) .

For x ∈ Eb , we define its continuous extension x

e over [−r, b] by

 x(t) for t ∈ [0, b]



x
e(t) =
ϕ(t) for t ∈ [−r, 0]


For ϕ ∈ C, we define the function y ∈ C([0, b], X) by y(t) = R(t)ϕ(0) and its
extension ye in C([−r, b], X) by

 y(t) for t ∈ [0, b]



ye(t) =
ϕ(t) for t ∈ [−r, 0]


For each z ∈ C([0, b], X), set x

e(t) = ze(t) + ye(t), where ze is the extension by
zero of the function z on [−r, 0]. Observe that x satifies (6.2.1) if and only if
z(0) = 0 and
Z t
R(t − s) f (s, zes + yes ) + C(s)u(s) ds for t ∈ [0, b],
 
z(t) =
0

Let
Mb = sup kR(t)k and kCk = sup kC(t)kL(U,X) .
t∈[0,b] t∈I

Since Uad is bounded, let K

e > 0 be such that kukL2 ≤ K e for all u ∈ Uad .
Z t Z t
kz(t)k ≤ Mb kf (s, zes + yes )k ds + Mb kC(s)u(s)k ds
0 0
Z t Z b
≤ Mb b af + Mb af ke
zs + yes k ds + Mb kCk ku(s)k ds
0 0

√ Z t
≤ Mb b af + Mb b kCkkukL2 + Mb af ke
zs + yes k ds
0

√ Z t
≤ Mb b af + Mb b kCkK
e + Mb af ke
zs + yes k ds,
0

Thus
√ Z t
kz(t)k ≤ Mb b af + Mb b kCkK
e + Mb af ke
zs + yes k ds. (6.3.1)
0
Optimal Controls for Integrodifferential Equations with Finite Delay 84

ke
zs + yes k ≤ ke
zs k + ke
ys k

≤ ke
zs k + Mb kϕk

≤ sup kz(τ )k + Mb kϕk

τ ∈[0,s]

This implies that (6.3.1) can be rewritten as follows

√ Z t
sup kz(s)k ≤ Mb b af + Mb e + Mb2 af b kϕk + Mb af
b kCkK sup kz(τ )k dτ
s∈[0,t] 0 τ ∈[0,s]
Z t
= M + Mb af sup kz(τ )k dτ.
0 τ ∈[0,s]

√
with M = Mb b af + Mb e + M 2 af b kϕk.
b kCkK b

It follows by Gronwall’s inequality that

kz(t)k ≤ M eb af Mb =: M
f.

As a result, for t ∈ I, we have

kxu (t)k ≤ kz(t)k + kR(t)ϕ(0)k

≤ M
f + Mb kϕk := ρ

That is kxu (t)k ≤ ρ for all t ∈ I. This completes the proof of the Lemma.

We have the following theorem on continuous dependence of the mild solu-

tions of equation (6.1.1) on the controls and initial states.

Theorem 6.3.2 For all λ > 0, there exists γ ∗ (λ) > 0 such that for all
ϕ1 , ϕ2 ∈ B(0, λ),
 
kx1t − x2t k ≤ γ ∗ (λ) kϕ1 − ϕ2 k + ku1 − u2 kL2 for t ∈ [0, b],
Optimal Controls for Integrodifferential Equations with Finite Delay 85

where
 Z t
R(t − s) f (s, xis ) + C(s)ui (s) ds for t ∈ I
i
 
 R(t)ϕ (0) +


xi (t) = 0


ϕi (t) for − r ≤ t ≤ 0,

(6.3.2)
and ui ∈ Uad , for i = 1, 2.

Proof. Let xi , for i = 1, 2, be two mild solutions of equation (6.1.1),

corresponding to the controls ui ∈ Uad and λ > 0 such that ϕ1 , ϕ2 ∈ B(0, λ).

 Z t
R(t − s) f (s, xis ) + C(s)ui (s) ds for t ∈ I
i
 
 R(t)ϕ (0) +


xi (t) = 0

ϕi (t) for − r ≤ t ≤ 0,



From the proof of Lemma 6.3.1, one can see that for ρλ = M f + Mb λ > 0, we
have kxs k ≤ ρλ , i = 1, 2.
i

Now, for t ∈ [0, b], we have

Z t
1 2 1 2
kx (t) − x (t)k ≤ Mb kϕ (0) − ϕ (0)k + Mb kf (s, x1s ) − f (s, x2s )k ds
0
Z t
+ Mb kC(s)u1 (s) − C(s)u2 (s)k ds
0
Z t 
1 2
≤ Mb kϕ − ϕ k + Mb Lf (ρλ ) kx1s − x2s k ds
0
Z t
+ Mb kC(s)u1 (s) − C(s)u2 (s)k ds
0
Z t
1 2
≤ Mb kϕ − ϕ k + Mb Lf (ρλ ) kx1s − x2s k ds
0
√
+ Mb Lf (ρλ ) b kCk ku1 − u2 kL2

That is

1 2 1 2
Z t √
kx (t)−x (t)k ≤ Mb kϕ −ϕ k+Mb Lf (ρλ ) kx1s −x2s k ds+Mb Lf (ρλ ) b kCk ku1 −u2 kL2
0
(6.3.3)
Optimal Controls for Integrodifferential Equations with Finite Delay 86

But we have that

kx1s − x2s k ≤ sup kx1 (τ ) − x2 (τ )k.
τ ∈[−r,s]

It follows that
√
sup kx1 (s) − x2 (s)k ≤ Mb kϕ1 − ϕ2 k + Mb Lf (ρ) b kCk ku1 − u2 kL2
s∈[−r,t]
Z t
+ Mb Lf (ρλ ) sup kx1 (τ ) − x2 (τ )k dτ.
0 τ ∈[−r,s]

By Gronwall’s inequality, we have that

h √ i
sup kx (s) − x (s)k ≤ Mb kϕ − ϕ k + Mb Lf (ρλ ) b kCk ku − u kL2 eMb Lf (ρλ )b .
1 2 1 2 1 2
s∈[−r,t]

This implies that

h √ i
kx1t − x2t k ≤ Mb kϕ1 − ϕ2 k + Mb Lf (ρλ ) b kCk ku1 − u2 kL2 eMb Lf (ρλ )b ,

Let n √ o
γ ∗ (λ) := max Mb eMb Lf (ρλ )b , Mb Lf (ρλ ) b kCk eMb Lf (ρλ )b .
Then, we have that
 
kx1t − x2t k ≤ γ ∗ (λ) kϕ1 − ϕ2 k + ku1 − u2 kL2 for t ∈ [0, b].

And the proof is complete.

2

6.4 Existence of the Optimal Controls

Now, we study the existence of solutions to the following Lagrange problem

 Find a control u0 ∈ Uad such that



(LP)
J (u0 ) ≤ J (u) for all u ∈ Uad ,


where Z b  
J (u) := L t, xut , xu (t), u(t) dt,
0
and xu denotes the mild solution of (6.1.1) corresponding to the control
u ∈ Uad and the initial data ϕ.
For the existence of solutions to problem (LP), we make the following as-
sumptions.
Optimal Controls for Integrodifferential Equations with Finite Delay 87

(HL )

(i) The functional L : I × C × X × U → R ∪ {∞} is Borel measurable.

(ii) L(t, ·, ·, ·) is sequencially lower semicontinuous on C ×X ×U for almost

all t ∈ I.

(iii) L(t, ψ, y, ·) is convex on U for each ψ ∈ C, y ∈ X and almost all

t ∈ I.

(iv) There exist constants ν, β ≥ 0, γ > 0, and µ ∈ L1 (I) nonnegative

such that

L(t, ψ, y, u) ≥ µ(t) + νkψk + βkyk + γkuk.

We have the following result on the existence of optimal controls for problem
(LP).

Theorem 6.4.1 Assume that hypotheses (H1 ) − (H5 ) and (HL ) hold. Then
the Lagrange problem (LP) admits at least one optimal pair, that is there
exists an admissible control pair (x0 , u0 ) ∈ C([−r, b], X) × Uad such that
Z b   Z b  
0 0 0 0 u u
J (u ) = L t, xt , x (t), u (t) dt ≤ L t, xt , x (t), u(t) dt = J (u) for u ∈ Uad .
0 0

n o
Proof. If inf J (u) : u ∈ Uad = ∞, we are done.
n o
Without loss of generality, assume that inf J (u) : u ∈ Uad = δ < ∞.
Suppose that δ = −∞, then for each n ∈ N, there exists (un )n≥1 ⊂ Uad such
that
J (un ) < −n (∗)
Boundedness of Uad implies that (un )n≥1 is bounded and so there exists a
subsequence (unk )k≥1 of (un )n≥1 that converges weakly to some u0 in L2 (I, U ),
since L2 (I, U ) is reflexive. But Uad is closed and convex, so by Mazur’s
Lemma, it is weakly closed and therefore, u0 ∈ Uad . By hypothesis (HL ),
L(t, ψ, y, ·) is weakly lower semicontinuous, so we have that

L(t, ψ, y, u0 ) ≤ lim inf L(t, ψ, y, unk ) < −∞,

k→∞

which implies that J (u0 ) < −∞ using (∗). And this is a contradiction since
J (u0 ) ∈ R ∪ {∞}. Hence δ ∈ R.
Optimal Controls for Integrodifferential Equations with Finite Delay 88

Now by the definition of δ, there exists a minimizing sequence, a feasible pair

((xn , un ))n≥1 ⊂ Sad such that
Z b  
L t, xnt , xn (t), un (t) dt −→ δ as n → ∞,
0

where
n o
Sad := (x, u) : x is a mild solution of equation (6.1.1) corresponding to the control u ∈ Uad .

Boundedness of Uad and the fact that L2 (I, U ) is reflexive imply that (un )n≥1
has a subsequence denoted for simplicity by (uk )k≥1 , that converges weakly
to some u0 in L2 (I, U ). But Uad is closed and convex, so by Mazur’s Lemma,
it is weakly closed and therefore, u0 ∈ Uad .
Let
 Z t
R(t − s) f (s, xks ) + C(s)uk (s) ds for t ∈ I
 
 R(t)ϕ(0) +


xk (t) = 0

ϕ(t) for − r ≤ t ≤ 0.



denote the subsequence of (xn )n≥1 corresponding to the control sequence

(uk )k≥1 and x0 be the mild solution corresponding to the control u0 ∈ Uad .
We show that xk → x0 .
For t ∈ [0, b], we have
Z t h i
k 0
x (t) − x (t) ≤ R(t − s) f (s, xks ) − f (s, x0s ) ds

0
Z t h i
+ R(t − s) C(s)uk (s) − C(s)u0 (s) ds

0
Z t Z t
k 0 k 0
≤ Mb Lf (ρ) xs − xs ds + Mb C(s)u (s) − C(s)u (s) ds

0 0

Z t √ Z t
 2  12
k 0 k 0
≤ Mb Lf (ρ) xs − xs ds + Mb b C(s)u (s) − C(s)u (s) ds

0 0
Z t
k

0
√ k

0
≤ Mb Lf (ρ) x
s − x s ds + Mb b Cu − Cu
0 L2 (I,U )
Optimal Controls for Integrodifferential Equations with Finite Delay 89

That is

k 0
Z t
k
√
x (t) − x (t) ≤ Mb Lf (ρ) xs − x0s ds + Mb b Cuk − Cu0

0 L2 (I,U )
(6.4.1)
But we have that

kxks − x0s k ≤ sup kxk (τ ) − x0 (τ )k.

τ ∈[0,s]

This implies that

k 0
√ k

0
Z t
k 0

sup x (s)−x (s) ≤ Mb b Cu −Cu +Mb Lf (ρ) sup x (τ )−x (τ ) dτ.

s∈[0,t] L2 (I,U ) 0 τ ∈[0,s]

It follows from Gronwall’s inequality that

√
x (t) − x0 (t) ≤ M ∗∗ Cuk − Cu0 , where M ∗∗ = Mb b eMb bLf (ρ) .
k
L2 (I,U )
(6.4.2)
We have the following Lemma.
Lemma 6.4.2 [96] Let (un )n≥1 ⊂ Uad and u0 ∈ Uad such that (un )n≥1 con-
verges weakly to u0 . Then,

−→ 0 as k → ∞, if C ∈ L∞ (I; L(U, X)).
k
Cu − Cu0 2

L (I,U )

We have by (6.4.2) that

k 0 ∗∗ k 0
x − x ≤ M Cu − Cu ,
L2 (I,U )

and therefore, it follows by Lemma 6.4.2 that

xk −→ x0 as k → ∞.

We note that (HL ) implies the assumptions of Balder’s Theorem. Hence by

Z b  
using Balder’s Theorem, we can conclude that (xt , x, u) 7→ L t, xt , x(t), u(t) dt
0
is sequencially lower semicontinuous in the strong topology of C([−r, 0], X) ×
L1 (I, X) × L1 (I, U ).
Now, since C([−r, 0], X) × L2 (I, X) × L2 (I, U ) ⊂ C([−r, 0], X) × L1 (I, X) ×
L1 (I, U ), J is also sequencially lower semicontinuous on C([−r, 0], X) ×
Optimal Controls for Integrodifferential Equations with Finite Delay 90

L2 (I, X) × L2 (I, U ), and in the strong topology of L1 (I, Eϕ × X × U ).

Hence, J is weakly lower semicontinuous on L2 (I, U ), and since by (HL ) − (iv),
J > −∞, J attains its infimum at u0 ∈ Uad , that is
Z b   Z b  
k k k
δ = lim L t, xt , x (t), u (t) dt ≥ L t, x0t , x0 (t), u0 (t) dt = J (u0 ) ≥ δ.
k→∞ 0 0

Thus, δ = J (u0 ), and hence there exists an admissible control u0 ∈ Uad such
that
J (u0 ) ≤ J (u) for all u ∈ Uad .
This completes the proof.

We now illustrate our main result by the following example. We observe that
in [96], the Langrangian function L defined by the authors in the example
does not satisfy condition (HL ) − (iv), as they claimed. We correct that
here.

6.5 Example
Let Ω be bounded domain in Rn with smooth boundary and consider the
following nonlinear integrodifferential equation.

 Z t
∂v(t, ξ)
= ∆v(t, ξ) + ζ(t − s)∆v(s, ξ) ds + α(t) sin(v 2 (t − r, ξ)) + β(t)ω(t, ξ)





 ∂t 0



for t ∈ I = [0, 1] and ξ ∈ Ω



v(t, ξ) = 0 for t ∈ [0, 1] and ξ ∈ ∂Ω








v(θ, ξ) = φ(θ, ξ) for θ ∈ [−r, 0] and ξ ∈ Ω,


(6.5.1)
where α, β ∈ C([0, 1]; R), ω : I×Ω → R continuous in t and ζ ∈ W 1,1 (R+ , R+ )

Let X = U = L2 (Ω). For η > 0, we define the set of admissible controls

Uad by
n o
Uad := u : I → U such that u is measurable and kukL2 (I,U ) ≤ η ,
Optimal Controls for Integrodifferential Equations with Finite Delay 91

where Z 1 Z 
kuk2L2 (I,U ) = 2
u (s)(ξ) dξ ds.
0 Ω
We define A : D(A) ⊂ X → X by:

 D(A) = H 2 (Ω) ∩ H01 (Ω)

Av = ∆v for v ∈ D(A).


Theorem 6.5.1 (Theorem 4.1.2, p. 79 of [76]) The linear operator A

defined above, is the infinitesimal generator of a C0 -semigroup on L2 (Ω).
A generates a C0 -semigroup T (t) t≥0 on L2 (Ω).

Define
∂v(t, ξ)
x(t)(ξ) = v(t, ξ), x0 (t)(ξ) = , ω(t, ξ) = u(t)(ξ).
∂t
ϕ(θ)(ξ) = φ(θ, ξ) for θ ∈ [−r, 0] and ξ ∈ Ω.
f (t, ϕ)(ξ) = α(t) sin (ϕ2 (−r)(ξ)) for t ∈ [0, 1] and ξ ∈ Ω.


 
C(t) : X → X be defined by C(t)u(t) (ξ) = C(t)u(t)(ξ) = β(t)ω(t, ξ).

(B(t)x)(ξ) = ζ(t)∆v(t, ξ) for t ∈ [0, 1], x ∈ D(A) and ξ ∈ Ω.

Equation (6.5.1) is then transformed into the following form
 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, xt ) + C(t)u(t) for t ∈ I = [0, 1],


0


x0 = ϕ.

(6.5.2)
One can see that, f satisfies (H3 ). Now we consider the following cost func-
tion: Z 1  
J (u) := L t, xut , xu (t), u(t) dt,
0
where
L : [0, 1] × C([−r, 0], L2 (Ω)) × L2 (Ω) × L2 (Ω) −→ R is defined by:
 
L t, ψ, x, u = kψk + kxk + kuk.
L satisfies all the conditions of hypothesis (HL ). Then,
Z 1 
u u
J (u) = kxt k + kx (t)k + ku(t)k dt.
0
Hence, all the conditions of Theorem 6.4.1 are satisfied, and therefore, equa-
tion (6.5.2) has at least one optimal pair.
 CHAPTER 7

On the Solvability and Optimal Control of some Partial

Functional Integrodifferential Equations with Infinite
Delay in Banach Spaces

7.1 Introduction
In this chapter, we study the existence of mild solutions and the optimal con-
trols of some systems that take the form of the following partial functional
integrodifferential equation with infinite delay in a Banach space (X, k · k):

 Z t
0
 x (t) = Ax(t) + γ(t − s)x(s)ds + f (t, xt ) + C(t)u(t) for t ∈ I = [0, b]


0


x0 = ϕ ∈ B,

(7.1.1)
where
A : D(A) → X is the infinitesimal generator of a C0 -semigroup
T (t) t≥0 on a separable reflexive Banach space X; for t ≥ 0, γ(t) is a
closed linear operator with domain D(γ(t)) ⊃ D(A). The control u takes
1
The results of this chapter are contents of the following paper
- Khalil Ezzinbi and Patrice Ndambomve [93]

92
Optimal Controls for Integrodifferential Equations with Finite Delay 93

values from another separable reflexive Banach space U . The operator C(t)
belongs to L(U, X) which is the Banach space of bounded linear operators
from U into X, and the phase space B is a linear space of functions mapping
] − ∞, 0] into X satisfying axioms which will be described later, for every
t ≥ 0, xt denotes the history function of B defined by

xt (θ) = x(t + θ) for − ∞ ≤ θ ≤ 0,

f : I × B → X is a continuous function satisfying some conditions. In the

literature devoted to equations with finite delay, the phase space is the space
of continuous functions on [−r, 0], for some r > 0, endowed with the uniform
norm topology. But when the delay is unbounded, the selection of the phase
space B plays an important role in both qualitative and quantitative theories.
A usual choice is a normed space satisfying some suitable axioms, which was
introduced by Hale and Kato [82].
Problems of controllability and existence of optimal controls for nonlinear
differential equations have been studied extensively by many authors under
various hypotheses (see e.g., [87], [88],[89],[90],[94] [95]), but little is known
and done about the existence of optimal controls for integrodifferential equa-
tions with delay using the resolvent operator theory.
In [96], the authors considered the following fractional integrodifferential
equation with infinite delay in Banach spaces

  Z t 
C
Dtq x(t) = Ax(t) + f t, xt , g(t, s, xs )ds + C(t)u(t) for t ∈ I = [0, b]



0


x0 = ϕ ∈ B,


where C Dtq denotes the Caputo fractional derivative of order q ∈ (0, 1). Using
the techniques of a priori estimation and extension of step by steps, they
studied the existence and continuous dependence of mild solutions and the
optimal controls of the associated Lagrange problem. Wand and Zhou [97]
discussed the optimal controls of a Lagrange problem for fractional evolution
equations. In [98], the authors studied the optimal controls for nonlinear im-
pulsive integrodifferential equations of mixed type on Banach spaces. In [99],
the authors studied the existence of mild solutions and the optimal controls
of a Lagrange problem for some impulsive fractional semilinear differential
equations, using the techniques of a priori estimation. Motivated by these
works, we investigate the solvability and the existence of optimal controls
of a Lagrange problem for equation (7.1.1), using the techniques of a priori
estimation of mild solutions. The existence and uniqueness of mild solutions
Optimal Controls for Integrodifferential Equations with Finite Delay 94

is obtained using the theory of resolvent operator for integral equations. Fur-
thermore, to the best of our knowledge, existence results for optimal controls
of partial functional integrodifferential equation (7.1.1) with infinite delay are
untreated in the literature, and this fact motivates us to extend the existing
ones and make new development of the present work on this issue.

7.2 Existence of mild solutions for equation (7.1.1)

In this work, we will employ an axiomatic definition of the phase space B
introduced by Hale and Kato in [82]. Thus, (B, k · kB ) will be a normed linear
space of functions mapping ] − ∞, 0] into X and satisfying the following
axioms:

(A1 ) There exist positive constant H and functions K : R+ → R+ con-

tinuous and M : R+ → R+ locally bounded, such that for a > 0, if
x : ] − ∞, a] → X is continuous on [0, a] and x0 ∈ B, then for every
t ∈ [0, a], the following conditions hold:

(i) xt ∈ B,

(ii) kx(t)k ≤ Hkxt kB , which is equivalent to kϕ(0)k ≤ HkϕkB for every

ϕ ∈ B,

(iii) kxt kB ≤ K(t) sup kx(s)k + M (t)kx0 kB .

0≤s≤t

(A2 ) For the function x in (A1 ), t → xt is a B-valued continuous function

for t ∈ [0, a].

(A3 ) The space B is complete.

We assume the following hypotheses.

(H3 ) The function f : I × B → X satisfies the following conditions:

(i) f (·, ψ) is measurable for ψ ∈ B,

(ii) for any ρ > 0, there exists Lf (ρ) > 0 such that

kf (t, ψ1 )−f (t, ψ2 )k ≤ Lf (ρ)kψ1 −ψ2 kB for kψ1 kB ≤ ρ, kψ2 kB ≤ ρ and t ∈ [0, b],
Optimal Controls for Integrodifferential Equations with Finite Delay 95

(iii) there exists af > 0 such that

kf (t, ψ)k ≤ af (1 + kψkB ) for all ψ ∈ B and t ∈ [0, b].

(H4 ) Let U be the separable reflexive Banach space from which the control
u takes values and assume C ∈ L∞ (I; L(U, X)).

(H5 ) The multivalued map Γ : I → 2U \{∅} has closed, convex, and bounded
values, Γ is graph measurable, and Γ (·) ⊆ Ω where Ω is a bounded set
in U .
We denote by Uad the set of admissible controls defined by:
n o
Uad = u : I → U ; u is measurable and u(t) ∈ Γ (t), a.e. .

Then, we have the following:

Theorem 7.2.1 [96] Uad 6= ∅ and Uad ⊂ L2 (I, U ) is bounded, closed and
convex. Also, Cu ∈ L2 (I, U ) for all u ∈ Uad .

Definition 7.2.2 Let u ∈ Uad and ϕ ∈ B. A function x : ] − ∞, b] → X is

called a mild solution of equation (7.1.1) if x ∈ C([0, b]; X) and
 Z t
 R(t)ϕ(0) + R(t − s) [f (s, xs ) + C(s)u(s)] ds for t ∈ I


x(t) = 0


ϕ(t) for − ∞ ≤ t ≤ 0.

(7.2.1)

Let

Rb = sup kR(t)k, Kb = sup kK(t)k, Mb = sup kM (t)k, kCk = sup kC(t)kL(U,X) .

t∈[0,b] t∈[0,b] t∈[0,b] t∈[0,b]

We have the following theorem on existence of mild solutions of equation

(7.1.1) with respect to a given control u ∈ Uad .

Theorem 7.2.3 Assume that (H1 ) − (H5 ) hold. Then for each u ∈ Uad ,
equation (7.1.1) has a unique mild solution on ] − ∞, b].

Proof. Let b1 ≤ b, ρ > 0, and ψ ∈ B such that kψkB ≤ ρ. For t ∈ [0, b1 ], we

have by the local Lipschitz condition on f that

kf (t, ψ)k ≤ Lf (ρ)kψkB + kf (t, 0)k ≤ Lf (ρ)ρ + sup kf (s, 0)k.

s∈[0,b1 ]
Optimal Controls for Integrodifferential Equations with Finite Delay 96

b1 will be chosen sufficiently small enough to get the local existence of mild
solutions.
Let ϕ ∈ B, ρ = kϕkB + 1 and ρ∗ = Lf (ρ)ρ + sups∈[0,b1 ] kf (s, 0)k.
Define the function

 R(t)ϕ(0) for t ∈ [0, b]



y(t) =
ϕ(t) for t ∈] − ∞, 0].


By axioms (A1 ) − (i) and (A2 ), we deduce that yt ∈ B and t 7→ yt is contin-

uous for t ∈ [0, b]. Then, for τ1 ∈]0, 1[, there exists τ2 ∈]0, 1[ such that

kyt − ϕkB ≤ τ1 for t ∈ [0, τ2 ].

Let b1 ∈ [0, τ2 ] be such that

 p 
Kb Rb ρ∗ b1 + kCk kukL2 b1 < 1 − τ1 . (7.2.2)

For x ∈ C([0, b], X) such that x(0) = ϕ(0), we define its extension x
e on
] − ∞, b] by

 x(t) for t ∈ [0, b]



x
e(t) =
ϕ(t) for t ∈] − ∞, 0]


We introduce the following space

n o
Eϕ = x ∈ C([0, b1 ]; X) such that x(0) = ϕ(0) and sup ke
xs − ϕkB ≤ 1 ,
s∈[0,b1 ]

provided with the uniform norm topology.

Define the operator P : Eϕ → C([0, b1 ]; X) by

Z t
es ) + C(s)u(s) ds for t ∈ [0, b1 ]
 
(P x)(t) = R(t)ϕ(0) + R(t − s) f (s, x
0
 
We claim that P Eϕ ⊂ Eϕ .
In fact let x ∈ Eϕ . By (H3 ) and axiom (A1 ), the function s 7→ f (s, x
es ) is
continuous on [0, b1 ]. Then, Definition 2.2.2 implies that
Z t
 
s 7→ R(t − s) f (s, x
es ) + C(s)u(s) ds
0
Optimal Controls for Integrodifferential Equations with Finite Delay 97

is continuous on [0, b1 ], consequently, v = P x is continuous on [0, b1 ]. We

claim that v ∈ Eϕ . In fact, for any t ∈ [0, b1 ], we have

ke
vt − ϕkB ≤ ke
vt − yt kB + kyt − ϕkB ≤ ke
vt − yt kB + τ1 .

Since ke
xs − ϕkB ≤ 1 for s ∈ [0, b1 ], then ke
xs kB ≤ ρ for s ∈ [0, b1 ]. Then,

kf (s, x xs kB + kf (s, 0)k ≤ ρ∗ .

es )k ≤ Lf (ρ)ke

By axiom (A3 ) − (iii), for any t ∈ [0, b1 ], we have

ke
vt − yt kB ≤ Kb sup kv(s) − y(s)k
s∈[0,t]

and
Z t
 
kv(t) − y(t)k ≤ R(t − s) f (s, x
es ) + C(s)u(s) ds

0
p
≤ Rb ρ∗ b1 + Rb kCk kukL2 b1
1
< (1 − τ1 ).
Kb
Consequently, ke vt − yt kB < 1 − τ1 . Thus, we deduce that
  vt − ϕkB < 1 for
ke
any t ∈ [0, b1 ], which implies that v ∈ Eϕ . Therefore P Eϕ ⊂ Eϕ .
We now show that P is a strict contraction on [0, b1 ]. In fact, let x1 , x2 ∈ Eϕ
and t ∈ [0, b1 ].
xis k ≤ ρ for i = 1, 2. We have that
Then, ke
Z t
1 2
k(P x )(t) − (P x )(t)k ≤ Rb e1s ) − f (s, x
kf (s, x e2s )k ds
0
Z t
≤ Rb Lf (ρ) x1s − x
ke e2s k ds
0
Z t
≤ Rb Lf (ρ)Kb sup kx1 (σ) − x2 (σ)k dσ
0 σ∈[0,s]

≤ Kb Rb Lf (ρ) b1 kx1 − x2 k

Then,
k(P x1 )(t) − (P x2 )(t)k ≤ Kb Rb Lf (ρ) b1 kx1 − x2 k.
Now, since
p
Kb Rb Lf (ρ) b1 ≤ Kb Rb ρ∗ b1 < Kb Rb ρ∗ b1 + Kb Rb kCk kukL2 b1 ,
Optimal Controls for Integrodifferential Equations with Finite Delay 98

condition (7.2.2) implies that

Kb Rb Lf (ρ) b1 < 1.

Thus, P is a strict contraction on Eϕ .

It follows from the contraction mapping principle that P has a unique fixed
point x ∈ Eϕ , which is the unique mild solution of equation (7.1.1) with re-
spect to u on ] − ∞, b1 ].
Using the same arguments, we can show that x can be extended to a maximal
interval of existence [0, tmax [.

Lemma 7.2.4 [109] If tmax < b, then, lim sup kx(t)k = ∞.

t→tmax

We show that tmax = b.

Assume on the contrary that tmax < b. Then for t ∈ [0, tmax ] we have that
Z t
 
x(t) = R(t)ϕ(0) + R(t − s) f (s, xs ) + C(s)u(s) ds.
0

It follows that
Z t Z t
kx(t)k ≤ Rb kϕ(0)k + Rb kf (s, xs )k ds + Rb kC(s)u(s)k ds
0 0
Z t Z t
≤ Rb kϕ(0)k + Rb tmax af + Rb af kxs k ds + Rb kCk ku(s)k ds
0 0

√
Z t
≤ Rb kϕ(0)k + Rb tmax af + Rb tmax kCkkukL2 + Rb af kxs k ds
0
√
≤ Rb kϕ(0)k + Rb tmax af + Rb tmax kCkkukL2 + Rb Mtmax af tmax kϕ(0)k
Z t
+Rb Ktmax af sup kx(τ )k dτ.
0 τ ∈[0,s]

This implies that

√
sup kx(s)k ≤ Rb kϕ(0)k + Rb tmax af + Rb tmax kCkkukL2 + Rb Mtmax af tmax kϕ(0)k
s∈[0,t]
Z t
+Rb Ktmax af sup kx(τ )k dτ,
0 τ ∈[0,s]
Optimal Controls for Integrodifferential Equations with Finite Delay 99

where
Ktmax = sup K(s) and Mtmax = sup M (s).
s∈[0,tmax ] s∈[0,tmax ]

It follows by Gronwall’s inequality that

kx(t)k ≤ β ∗ e(Rb Ktmax af t) for t ∈ [0, tmax ],

√
where β ∗ = Rb kϕ(0)k+Rb tmax af +Rb tmax kCkkukL2 +Rb Mtmax af tmax kϕ(0)k.
Thus
lim kx(t)k ≤ β ∗ e(Rb Ktmax af tmax ) < ∞.
t→tmax

This contradicts Lemma 7.2.4. Therefore, tmax = b and hence, equation

(7.1.1) has a unique mild solution on (−∞, b].

7.3 Continuous Dependence and Existence of

the Optimal Control
In this section, we discuss the continuous dependence of the mild solutions
of equation (7.1.1) on the controls and initial states, and the existence of
solutions of the Lagrange problem associated to equation (7.1.1).
We have the following a priori estimation.

Lemma 7.3.1 Suppose (H3 ) holds and assume that equation (7.1.1) has a
mild solution xu on ] − ∞, b] with respect to u ∈ Uad . Then, there exists a
constant ρ > 0 independent of u such that kxu (t)k ≤ ρ for t ∈ [0, b], (ρ
depends only on Uad and ϕ).

Proof. Let ϕ ∈ B. For x ∈ C([0, b], X) such that x(0) = ϕ(0), we define its
extension x
e on ] − ∞, b] by

 x(t) for t ∈ [0, b]



x
e(t) =
ϕ(t) for t ∈] − ∞, 0]


Also, we define the function y ∈ C([0, b], X) by y(t) = R(t)ϕ(0) and its
extension ye on ] − ∞, b] by

 y(t) for t ∈ [0, b]



ye(t) =
ϕ(t) for t ∈] − ∞, 0]

Optimal Controls for Integrodifferential Equations with Finite Delay 100

For each z ∈ C([0, b], X), set x

e(t) = ze(t) + ye(t), where ze is the extension by
zero of the function z on ] − ∞, 0]. Observe that x satifies (7.2.1) if and only
if z(0) = 0 and
Z t
R(t − s) f (s, zes + yes ) + C(s)u(s) ds for t ∈ [0, b].
 
z(t) =
0

Since Uad is bounded, let N

e > 0 be such that kukL2 ≤ K e for all u ∈ Uad .
Z t Z t
kz(t)k ≤ Rb kf (s, zes + yes )k ds + Rb kC(s)u(s)k ds
0 0
Z t Z b
≤ Rb b af + Rb af ke
zs + yes k ds + Rb kCk ku(s)k ds
0 0

√ Z t
≤ Rb b af + Rb b kCkkukL2 + Rb af ke
zs + yes kB ds
0

√ Z t
≤ Rb b af + Rb b kCkN
e + Rb af ke
zs + yes kB ds,
0

Thus
√ Z t
kz(t)k ≤ Rb b af + Rb b kCkN + Rb af
e ke
zs + yes kB ds. (7.3.1)
0

ke
zs + yes kB ≤ ke
zs kB + ke
ys kB

≤ K(s) sup kz(τ )k + M (s)kz0 kB + K(s) sup ke y 0 kB

y (τ )k + M (s)ke
0≤τ ≤s 0≤τ ≤s

≤ Kb sup kz(τ )k + Kb Rb kϕkB + Mb kϕ(0)k.

τ ∈[0,s]

This implies that (7.3.1) can be rewritten as follows

√  
sup kz(s)k ≤ Rb b af + Rb b kCkN + Rb af b Kb Rb kϕkB + Mb kϕ(0)k
e
s∈[0,t]
Z t
+Kb Rb af sup kz(τ )k dτ
0 τ ∈[0,s]
Z t
∗
= N + Kb Rb af sup kz(τ )k dτ.
0 τ ∈[0,s]
Optimal Controls for Integrodifferential Equations with Finite Delay 101

√  
with N ∗ = Rb b af + Rb e + Rb af b Kb Rb kϕkB + Mb kϕ(0)k .
b kCkN

It follows by Gronwall’s inequality that

kz(t)k ≤ N ∗ e(b af Rb Kb ) =: N
f∗ .

As a result, for t ∈ I, we have

kxu (t)k ≤ kz(t)k + kR(t)ϕ(0)k

f∗ + Rb kϕ(0)k := ρ
≤ N

That is kxu (t)k ≤ ρ for all t ∈ I. This completes the proof of the Lemma.

We have the following continuous dependence Theorem.

Theorem 7.3.2 For all r > 0, there exists λ∗ (r) > 0 such that for all
ϕ1 , ϕ2 ∈ B(0, r),
 
kx1 (t)−x2 (t)k ≤ λ∗ (r) kϕ1 (0)−ϕ2 (0)k+kϕ1 −ϕ2 kB +ku1 −u2 kL2 for t ∈ [0, b],

where
 Z t
R(t − s) f (s, xis ) + C(s)ui (s) ds for t ∈ I
i
 
 R(t)ϕ (0) +


xi (t) = 0


ϕi (t) for − ∞ ≤ t ≤ 0,

(7.3.2)
and ui ∈ Uad , for i = 1, 2.

Proof. Let xi , for i = 1, 2, be two mild solutions of equation (7.1.1),

corresponding to the controls ui ∈ Uad and initial conditions ϕi ∈ B(0, r).
 Z t
R(t − s) f (s, xis ) + C(s)ui (s) ds for t ∈ I
i
 
 R(t)ϕ (0) +


xi (t) = 0

ϕ (t) for − ∞ ≤ t ≤ 0,

 i

By Lemma 7.3.1, we have that, there exists a constant ρr = N

f∗ + Rb r > 0
such that kxs k ≤ ρr , i = 1, 2.
i
Optimal Controls for Integrodifferential Equations with Finite Delay 102

Now, for t ∈ [0, b], we have

Z t
1 2 1 2
kx (t) − x (t)k ≤ Rb kϕ (0) − ϕ (0)k + Rb kf (s, x1s ) − f (s, x2s )k ds
0
Z t
+ Rb kC(s)u1 (s) − C(s)u2 (s)k ds
0
Z t 
1 2
≤ Rb kϕ (0) − ϕ (0)k + Rb Lf (ρr ) kx1s − x2s kB ds
0
Z t
+ Rb kC(s)u1 (s) − C(s)u2 (s)k ds
0
Z t
1 2
≤ Rb kϕ (0) − ϕ (0)k + Rb Lf (ρr ) kx1s − x2s kB ds
0
√
+ Rb Lf (ρr ) b kCk ku1 − u2 kL2
Z t
1 2
≤ Rb kϕ (0) − ϕ (0)k + Rb Lf (ρr )Kb sup kx1 (τ ) − x2 (τ )k dτ
0 τ ∈[0,s]
√
+ Rb Lf (ρr )Mb b kx10 − x20 kB + Rb Lf (ρ) b kCk ku1 − u2 kL2
Z t
1 2
≤ Rb kϕ (0) − ϕ (0)k + Rb Lf (ρr )Kb sup kx1 (τ ) − x2 (τ )k dτ
0 τ ∈[0,s]
√
+ Rb Lf (ρr )Mb b kϕ1 − ϕ2 kB + Rb Lf (ρr ) b kCk ku1 − u2 kL2

It follows that
√
sup kx1 (s) − x2 (s)k ≤ Rb kϕ1 (0) − ϕ2 (0)k + Rb Lf (ρr ) b kCk ku1 − u2 kL2
s∈[0,t]
Z t
1 2
+ Rb Lf (ρ)Mb b kϕ − ϕ kB + Rb Lf (ρr )Kb sup kx1 (τ ) − x2 (τ )k dτ.
0 τ ∈[0,s]

By Gronwall’s inequality, we have that

sup kx1 (s) − x2 (s)k ≤ N ∗∗ eRb Lf (ρr )Kb b ,

s∈[0,t]

where
∗∗
h
1 2 1 2
√ 1 2
i
N = Rb kϕ (0)−ϕ (0)k+Rb Lf (ρr )Mb b kϕ −ϕ kB +Rb Lf (ρr ) b kCk ku −u kL2 .
Optimal Controls for Integrodifferential Equations with Finite Delay 103

This implies that

h
kx1 (t) − x2 (t)k ≤ Rb kϕ1 (0) − ϕ2 (0)k + Rb Lf (ρr )Mb b kϕ1 − ϕ2 kB

√ i
+Rb Lf (ρr ) b kCk ku1 − u2 kL2 eRb Lf (ρr )Kb b

Let
n √ o
λ∗ (r) := max Rb eRb Lf (ρr )Kb b , Rb Lf (ρr )Mb b eRb Lf (ρ)Kb b , Rb Lf (ρr ) b kCk eRb Lf (ρr )Kb b .

Then, we have that

 
kx1 (t)−x2 (t)k ≤ λ∗ (r) kϕ1 (0)−ϕ2 (0)k+kϕ1 −ϕ2 kB +ku1 −u2 kL2 for t ∈ [0, b].

And the proof is complete.

2
We now study the existence of solutions to the following Lagrange problem

 Find a control u0 ∈ Uad such that



(LP)
J (u0 ) ≤ J (u) for all u ∈ Uad ,


where Z b  
u u
J (u) := L t, xt , x (t), u(t) dt,
0
and x denotes the mild solution of (7.1.1) corresponding to the control
u

u ∈ Uad .
For the existence of solutions to problem (LP), we make the following as-
sumptions.
(HL )
(i) The functional L : I × B × X × U → R ∪ {∞} is Borel measurable.
(ii) L(t, ·, ·, ·) is sequencially lower semicontinuous on B×X ×U for almost
all t ∈ I.
(iii) L(t, ψ, y, ·) is convex on U for each ψ ∈ B, y ∈ X and almost all
t ∈ I.
(iv) There exist constants ν, β ≥ 0, γ > 0, and µ ∈ L1 (I) nonnegative
such that
L(t, ψ, y, u) ≥ µ(t) + νkψkB + βkyk + γkukU .
Optimal Controls for Integrodifferential Equations with Finite Delay 104

We have the following result on the existence of optimal controls for problem
(LP).

Theorem 7.3.3 Assume that hypotheses (H1 ) − (H5 ) and (HL ) hold. Then
the Lagrange problem (LP) admits at least one optimal pair, that is there
exists an admissible control pair (x0 , u0 ) ∈ C([0, b], X) × Uad such that
Z b   Z b  
0 0 0 0
J (u ) = L t, xt , x (t), u (t) dt ≤ L t, xut , xu (t), u(t) dt = J (u) ∀u ∈ Uad .
0 0

n o
Proof. If inf J (u) : u ∈ Uad = ∞, we are done.
n o
Without loss of generality, assume that inf J (u) : u ∈ Uad = δ < ∞.
Suppose that δ = −∞, then for each n ∈ N, there exists (un )n≥1 ⊂ Uad such
that
J (un ) < −n (∗)
Boundedness of Uad implies that (un )n≥1 is bounded and so there exists a
subsequence (unk )k≥1 of (un )n≥1 that converges weakly to some u0 in L2 (I, U ),
since L2 (I, U ) is reflexive. But Uad is closed and convex, so by Mazur’s
Lemma, it is weakly closed and therefore, u0 ∈ Uad . By hypothesis (HL ),
L(t, x, y, ·) is weakly lower semicontinuous, so we have that

L(t, ψ, y, u0 ) ≤ lim inf L(t, ψ, y, unk ) < −∞,

k→∞

which implies that J (u0 ) < −∞ using (∗). And this is a contradiction since
J (u0 ) ∈ R ∪ {∞}. Hence δ ∈ R.
Now by the definition of δ, there exists a minimizing sequence, a feasible pair
((xn , un ))n≥1 ⊂ Sad such that
Z b  
L t, xnt , xn (t), un (t) dt −→ δ as n → ∞,
0

where
n o
Sad := (x, u) : x is a mild solution of equation (7.1.1) corresponding to the control u ∈ Uad .

Boundedness of Uad and the fact that L2 (I, U ) is reflexive imply that (un )n≥1
has a subsequence denoted for simplicity by (uk )k≥1 , that converges weakly
to some u0 in L2 (I, U ). But Uad is closed and convex, so by Mazur’s lemma,
Optimal Controls for Integrodifferential Equations with Finite Delay 105

it is weakly closed and therefore, u0 ∈ Uad .

Let
 Z t
R(t − s) f (s, xks ) + C(s)uk (s) ds for t ∈ I
 
 R(t)ϕ(0) +


xk (t) = 0

ϕ(t) for − r ≤ t ≤ 0.



denote the subsequence of (xn )n≥1 corresponding to the control sequence

(uk )k≥1 and x0 be the mild solution corresponding to the control u0 ∈ Uad .
We show that xk → x0 .
For t ∈ [0, b], we have
Z t h i
k 0 k 0
x (t) − x (t) ≤ R(t − s) f (s, xs ) − f (s, xs ) ds

0
Z t h i
k 0
+ R(t − s) C(s)u (s) − C(s)u (s) ds


0
Z t Z t
k 0
≤ Rb Lf (ρ) xs − xs ds + Rb C(s)uk (s) − C(s)u0 (s) ds

0 0

Z t √ Z t
 2  12
k 0 k 0
≤ Rb Lf (ρ) xs − xs ds + Rb b C(s)u (s) − C(s)u (s) ds

0 0
Z t √
≤ Rb Lf (ρ) Kb sup xk (τ ) − x0 (τ ) dτ + Rb b Cuk − Cu0

0 τ ∈[0,s] L2 (I,U )

This implies that

k 0
√ k

0
Z t
k 0

sup x (s)−x (s) ≤ Rb b Cu −Cu +Rb Lf (ρ) Kb sup x (τ )−x (τ ) dτ.

s∈[0,t] L2 (I,U ) 0 τ ∈[0,s]

It follows from Gronwall’s inequality that

√
∗∗
, where λ∗∗ = Rb b eRb b Lf (ρ) Kb .
k 0 k 0
x (t) − x (t) ≤ λ Cu − Cu

L2 (I,U )
(7.3.3)
We have the following Lemma.
Lemma 7.3.4 [96] Let (un )n≥1 ⊂ Uad and u0 ∈ Uad such that (un )n≥1 con-
verges weakly to u0 . Then,

−→ 0 as k → ∞, if C ∈ L∞ (I; L(U, X)).
k
Cu − Cu0 2

L (I,U )
Optimal Controls for Integrodifferential Equations with Finite Delay 106

We have by (7.3.3) that

x − x0 ≤ λ∗∗ Cuk − Cu0
k
,

L2 (I,U )

and therefore, it follows by Lemma 7.3.4 that

xk −→ x0 as k → ∞.

We note that (HL ) implies the assumptions of Balder’s Theorem (see Theo-
rem 2.4.3). Hence by Balder’s Theorem, we can conclude that (xt , x, u) 7→
Z b  
L t, xt , x(t), u(t) dt is sequencially lower semicontinuous in the strong
0
topology of B × L1 (I, X) × L1 (I, U ).
Now, since B × L2 (I, X) × L2 (I, U ) ⊂ B × L1 (I, X) × L1 (I, U ), J is also
sequencially lower semicontinuous on B × L2 (I, X) × L2 (I, U ), and in the
strong topology of L1 (I, B × X × U ).
Hence, J is weakly lower semicontinuous on L2 (I, U ), and since by (HL ) − (iv),
J > −∞, J attains its infimum at u0 ∈ Uad , that is
Z b   Z b  
k k k
δ = lim L t, xt , x (t), u (t) dt ≥ L t, x0t , x0 (t), u0 (t) dt = J (u0 ) ≥ δ.
k→∞ 0 0

Thus, δ = J (u0 ), and hence there exists an admissible control u0 ∈ Uad such
that
J (u0 ) ≤ J (u) for all u ∈ Uad .
This completes the proof.

We now illustrate our main result by the following example.

7.4 Example
Let Ω be bounded domain in Rn with smooth boundary and consider the
following nonlinear integrodifferential equation.
Optimal Controls for Integrodifferential Equations with Finite Delay 107

t 0
 Z Z
∂v(t,ξ)
= ∆v(t, ξ) + ζ(t − s)∆v(s, ξ) ds + α(θ)g(t, v(t + θ, ξ)) dθ + β(t)ω(t, ξ)



 ∂t

 0 −∞



for t ∈ I = [0, 1] and ξ ∈ Ω



v(t, ξ) = 0 for t ∈ [0, 1] and ξ ∈ ∂Ω









v(θ, ξ) = φ(θ, ξ) for θ ∈] − ∞, 0] and ξ ∈ Ω,

(7.4.1)
where β ∈ C([0, 1]; R), g : [0, 1]×R → R is continuous and Lipschitzian with
respect to the second variable, the initial data function φ : R− × Ω → R is a
given function, ω : [0, 1] × Ω → R continuous in t, α : R− → R is continuous,
α ∈ L1 (R− , R) and ζ ∈ W 1,1 (R+ , R+ ).
Let X = U = L2 (Ω) and the phase space B = BU C(R− , X), the the space of
uniformly bounded continuous functions endowed with the following norm

kϕkB = sup kϕ(θ)k.

θ≤0

Then, the space BU C(R− , X) satisfies axioms (A1 ), (A2 ) and (A3 ).
For η > 0, we define the set of admissible controls Uad by
n o
Uad := u : I → U : u is measurable and kukL2 (I,U ) ≤ η .

where Z 1 Z 
kuk2L2 (I,U ) = 2
u (s)(ξ) dξ ds.
0 Ω

We define A : D(A) ⊂ X → X by:


 D(A) = H 2 (Ω) ∩ H01 (Ω)

Av = ∆v for v ∈ D(A).


Theorem 7.4.1 (Theorem 4.1.2, p. 79 of [76]) A is the infinitesimal

generator of a C0 -semigroup on L2 (Ω).

Define
∂v(t, ξ)
x(t)(ξ) = v(t, ξ), x0 (t)(ξ) = , ω(t, ξ) = u(t)(ξ).
∂t
ϕ(θ)(ξ) = φ(θ, ξ) for θ ∈] − ∞, 0] and ξ ∈ Ω.
Optimal Controls for Integrodifferential Equations with Finite Delay 108

Z 0
f (t, ψ)(ξ) = α(θ)g(t, ψ(θ)(ξ)) dθ for θ ∈] − ∞, 0] and ξ ∈ Ω.
−∞
 
C(t) : X → X be defined by C(t)u(t) (ξ) = C(t)u(t)(ξ) = β(t)ω(t, ξ).

(γ(t)x)(ξ) = ζ(t)∆v(t, ξ) for t ∈ [0, 1], x ∈ D(A) and ξ ∈ Ω.

We suppose that ϕ ∈ BU C(R− , X). Then, equation (7.4.1) is then trans-
formed into the following form
 Z t
0
 x (t) = Ax(t) + γ(t − s)x(s)ds + f (t, xt ) + C(t)u(t) for t ∈ I = [0, 1],


0


x0 = ϕ.

(7.4.2)
Suppose there exist a continuous function p ∈ L (I; R ) such that
1 +

|g(t, y)| ≤ p(t)|y| for t ∈ I and y ∈ R.

One can see that, f satisfies (H3 ). Now we consider the following cost func-
tion: Z 1  
u u
J (u) := L t, xt , x (t), u(t) dt,
0
where
L : [0, 1] × B × L2 (Ω) × L2 (Ω) −→ R is defined by:
 
L t, ψ, x, u = kψkB + kxk + kuk.

L satisfies all the conditions of hypothesis (HL ). Then,

Z 1 
J (u) = kxut k + kxu (t)k + ku(t)k dt.
0

Hence, all the conditions of Theorem 7.3.3 are satisfied, and therefore, equa-
tion (7.4.2) has at least one optimal pair.
 CHAPTER 8

Solvability and Optimal Controls for some Partial

Functional Integrodifferential Equations with Classical
Initial Conditions in Banach Spaces

8.1 Introduction
The aim of this chapter is to study the existence of mild solutions and the
optimal controls of some systems that take the form of the following partial
functional integrodifferential equation in a Banach space (X, k · k):

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, x(t)) + C(t)u(t) for t ∈ I = [0, b]


0


x(0) = x0 ∈ X,

(8.1.1)
where f : I ×X → X is a function satisfying some conditions; A : D(A) → X
is the infinitesimal generator of a C0 -semigroup T (t) t≥0 on a separable re-

flexive Banach space X; for t ≥ 0, B(t) is a closed linear operator with

1
The results of this chapter are contents of the following paper
- Khalil Ezzinbi and Patrice Ndambomve [92]

109
Optimal Controls for Integrodifferential Equations with Finite Delay 110

domain D(B(t)) ⊃ D(A). The control u(t) takes values from another sepa-
rable reflexive Banach space U . The operator C(t) belongs to L(U, X), the
Banach space of bounded linear operators from U into X.
In many areas of applications such as engineering, electronics, fluid dynam-
ics, physical sciences, etc..., integrodifferential equations appear and have
received considerable attention during the last decades. In [85], R. Grimmer
proved the existence and uniqueness of resolvent operators for these inte-
grodifferential equations that give the variation of parameter formula for the
solution. In recent years, much work has been done on the existence and
regularity of solutions of nonlinear integrodifferential equations with various
initial conditions by many authors by applying the resolvent operator the-
ory, for integral equations see e.g., [86] and the references therein. Problems
of controllability and existence of optimal controls for nonlinear differential
equations have been studied extensively by many authors under various hy-
potheses (see e.g., [87], [88],[89],[90],[94] [95]), but little is known and done
about the existence of optimal controls for integrodifferential equations us-
ing the resolvent operator theory. Wang and Zhou [97] discussed the optimal
controls of a Lagrange problem for the following fractional evolution equa-
tions:

 D x(t) = −Ax(t) + f (t, x(t)) + C(t)u(t) for t ∈ [0, b]

 q

x(0) = x0 ∈ X,


where Dq denotes the Caputo fractional derivative of order q ∈ (0, 1) and

−A : D(A) → X is the infinitesimal generator of a compact analytic semi-
group of uniformly bounded linear operators. In [99], the authors studied the
existence of mild solutions and the optimal controls of a Lagrange problem
for the following impulsive fractional semilinear differential equations,
  Z t 
C
Dtα x(t) = Ax(t) + f t, x(t), g(t, s, x(s))ds + C(t)u(t) for t ∈ [0, b], t 6= tk





 0


 ∆x(tk ) = Ik (x(t−
k )), k = 1, 2, · · · m




x(0) = x0 ∈ X,


where C Dtα denotes the Caputo fractional derivative of order α ∈ (0, 1] with
lower limit zero and A : D(A) → X is the infinitesimal generator of a C0 -
semigroup. They used the techniques of a priori estimation. In [98], the
authors studied the optimal controls for nonlinear impulsive integrodifferen-
tial equations of mixed type on Banach spaces.
Optimal Controls for Integrodifferential Equations with Finite Delay 111

Motivated by these works, we investigate the solvability and the existence of

optimal controls of a Lagrange problem for equation (8.1.1), using the tech-
niques of a priori estimation of mild solutions. The existence and uniqueness
of mild solutions is obtained using the theory of resolvent operator for in-
tegral equations. Furthermore, to the best of our knowledge, the optimal
controls for partial functional integrodifferential equation (8.1.1) with clas-
sical Cauchy initial conditions are untreated in the literature, and this fact
motivates us to extend the existing ones and make new development of the
present work on this issue.

8.2 Existence of mild solutions for equation (8.1.1)

We make the following assumptions.

(H3 ) The function f : I × X → X satisfies the following conditions:

(i) f (·, x) is measurable for x ∈ X,

(ii) for any ρ > 0, there exists Lf (ρ) > 0 such that

kf (t, x) − f (t, y)k ≤ Lf (ρ)kx − yk for kxk ≤ ρ, kyk ≤ ρ and t ∈ [0, b],

(iii) there exists af > 0 such that

kf (t, xk ≤ af (1 + kxk) for all x ∈ X and t ∈ [0, b].

(H4 ) Let U be the separable reflexive Banach space from which the control
u takes values and assume C ∈ L∞ (I; L(U, X)).

(H5 ) The multivalued map Γ : I → 2U \{∅} has closed, convex, and bounded
values, Γ is graph measurable, and Γ (·) ⊆ Ω where Ω is a bounded set
in U .
We denote by Uad the set of admissible controls defined by:
n o
Uad = u : I → U such that u is measurable and u(t) ∈ Γ (t), a.e. .

Then, we have the following:

Theorem 8.2.1 [96] Uad 6= ∅ and Uad ⊂ L2 (I, U ) is bounded, closed and
convex. Also, Cu ∈ L2 (I, U ) for all u ∈ Uad .
Optimal Controls for Integrodifferential Equations with Finite Delay 112

Definition 8.2.2 Let u ∈ Uad . A function x ∈ C(I; X) is called a mild

solution of equation (8.1.1) if

Z t
x(t) = R(t)x0 + R(t − s) [f (s, x(s)) + C(s)u(s)] ds for t ∈ I (8.2.1)
0

We have the following theorem on existence of mild solutions to equation

(8.1.1) with respect to a given control u ∈ Uad .

Theorem 8.2.3 Assume that (H1 ) − (H5 ) hold. Then for each u ∈ Uad ,
equation (8.1.1) has a unique mild solution on [0, b].

Proof. Let b1 , ρ > 0, and x ∈ X such that kxk ≤ ρ. For t ∈ [0, b1 ], we have
by the local Lipschitz condition on f that

kf (t, x)k ≤ Lf (ρ)kxk + kf (t, 0)k ≤ Lf (ρ)ρ + sup kf (s, 0)k.

s∈[0,b1 ]

b1 will be chosen sufficiently small enough to get the local existence of mild
solutions.
Let x ∈ X, ρ = kxk + 1 and ρ∗ = Lf (ρ)ρ + sups∈[0,b1 ] kf (s, 0)k.
We define the following space
n o
E0 = x ∈ C([0, b1 ]; X) such that sup kx(s) − x(0)k ≤ 1 .
s∈[0,b1 ]

Then, E0 is a closed subset of C([0, b1 ]; X) which is endowed with the uniform

norm topology. Let
Mb = sup kR(t)k.
t∈[0,b]

Define the operator K : E0 → C([0, b1 ]; X) by

Z t
R(t − s) f (s, x(s)) + C(s)u(s) ds for t ∈ [0, b1 ]
 
(Kx)(t) = R(t)x0 +
0
 
We claim that K E0 ⊂ E0 . In fact let x ∈ E0 and t ∈ [0, b1 ]. Then,

k(Kx)(t) − x(0)k ≤ kR(t)x(0) − x(0)k

Z t
 
+ R(t − s) f (s, x(s)) + C(s)u(s) ds

0
√
≤ kR(t)x(0) − x(0)k + Mb ρ∗ t + Mb kCk kukL2 t.
Optimal Controls for Integrodifferential Equations with Finite Delay 113

Now, choose b1 sufficiently small such that

n √ o
sup kR(s)x(0) − x(0)k + Mb ρ∗ s + Mb kCk kukL2 s < 1. (8.2.2)
s∈[0,b1 ]

Consequently,
√
k(Kx)(t)−x(0)k ≤ kR(t)x(0)−x(0)k+Mb ρ∗ t+Mb kCk kukL2 t < 1 for t ∈ [0, b1 ].
 
Hence, K E0 ⊂ E0 .
Let x, y ∈ E0 and t ∈ [0, b1 ]. Then, there exists ρ > 0 such that kxk, kyk ≤ ρ.
We have that
Z t
k(Kx)(t) − (Ky)(t)k ≤ Mb kf (s, x(s)) − f (s, y(s))k ds
0
Z t
≤ Mb Lf (ρ) kx(s) − y(s)k ds
0
Z t
≤ Mb Lf (ρ) sup kx(τ ) − y(τ )k dτ
0 τ ∈[0,s]

≤ Mb Lf (ρ) b1 kx − yk

Now, since
∗
n
∗
√ o
Mb Lf (ρ) b1 ≤ Mb ρ b1 < sup kR(s)x(0)−x(0)k+Mb ρ s+Mb kCk kukL2 s .
s∈[0,b1 ]

Condition (8.2.2) implies that

Mb Lf (ρ) b1 < 1.

Thus, K is a strict contraction on E0 . It follows from the contraction mapping

principle that K has a unique fixed point x ∈ E0 , which is the unique mild
solution of equation (8.1.1) with respect to u on [0, b1 ].
Using the same arguments, we can show that x can be extended to a maximal
interval of existence [0, tmax [.
Lemma 8.2.4 [86] If tmax < b, then, lim sup kx(t)k = ∞.
t→tmax

We show that tmax = b.

Assume on the contrary that tmax < b. Then for t ∈ [0, tmax ] we have that
Z t
 
x(t) = R(t)x0 + R(t − s) f (s, x(s)) + C(s)u(s) ds.
0
Optimal Controls for Integrodifferential Equations with Finite Delay 114

It follows that
Z t Z t
kx(t)k ≤ Mb kx0 k + Mb kf (s, x(s))k ds + Mb kC(s)u(s)k ds
0 0
Z t Z t
≤ Mb kx0 k + Mb tmax af + Mb af kx(s)k ds + Mb kCk ku(s)k ds
0 0

√
Z t
≤ Mb kx0 k + Mb tmax af + Mb tmax kCkkukL2 + Mb af kx(s)k ds
0

This implies that

√
Z t
kx(t)k ≤ Mb kx0 k + Mb tmax af + Mb tmax kCkkukL2 + Mb af kx(s)k ds
0

It follows by Gronwall’s inequality that

kx(t)k ≤ β ∗ eMb af t for t ∈ [0, tmax ],

√
where β ∗ = Mb kx0 k + Mb tmax af + Mb tmax kCkkukL2 .
Thus
lim kx(t)k ≤ β ∗ eMb af tmax < ∞.
t→tmax

This contradicts Lemma 8.2.4. Therefore, tmax = b and hence, equation

(8.1.1) has a unique mild solution on [0, b].

8.3 Continuous Dependence and Existence of

the Optimal Control
In this section, we discuss the continuous dependence of the mild solutions
of equation (8.1.1) on the controls and initial states, and the existence of
solutions of the Lagrange problem associated to equation (8.1.1).
We have the following a priori estimation.
Lemma 8.3.1 Suppose (H1 ) − (H3 ) holds and assume that equation (8.1.1)
has a mild solution xu on [0, b] with respect to u ∈ Uad . Then, there exists
a constant ρ > 0 independent of u such that kxu (t)k ≤ ρ for t ∈ [0, b], (ρ
depends only on Uad and x0 ).
Optimal Controls for Integrodifferential Equations with Finite Delay 115

Proof. Let

t ∈ [0, b], Mb = sup kR(t)k and kCk = sup kC(t)kL(U,X) .

t∈[0,b] t∈I

Since Uad is bounded, let K

e > 0 be such that kukL2 ≤ K e for all u ∈ Uad .
Then, we have that
Z t Z t
kx(t)k ≤ Mb kx0 k + Mb kf (s, x(s))k ds + Mb kC(s)u(s)k ds
0 0
Z t Z b
≤ Mb kx0 k + Mb b af + Mb af kx(s)k ds + Mb kCk ku(s)k ds
0 0

√ Z t
≤ Mb kx0 k + Mb b af + Mb b kCkkukL2 + Mb af kx(s)k ds
0

√ Z t
≤ Mb kx0 k + Mb b af + Mb b kCkK
e + Mb af kx(s)k ds.
0

Thus
√ Z t
kx(t)k ≤ Mb kx0 k + Mb b af + Mb b kCkK
e + Mb af kx(s)k ds. (8.3.1)
0

It follows by Gronwall’s inequality that

kx(t)k ≤ M eb af Mb =: M
f,
√
with M = Mb kx0 k + Mb b af + Mb b kCkK. e
As a result, for t ∈ I, we have

kxu (t)k ≤ M
f := ρ

That is kxu (t)k ≤ ρ for all t ∈ I. This completes the proof of the Lemma.

We have the following theorem on continuous dependence of the mild solu-

tions of equation (8.1.1) on the controls and initial states.

Theorem 8.3.2 For all λ > 0, there exists γ ∗ (λ) > 0 such that for all
x10 , x20 ∈ B(0, λ),
 
1 2 ∗ 1 2 1 2
kx (t) − x (t)k ≤ γ (λ) kx0 − x0 k + ku − u kL2 for t ∈ [0, b],
Optimal Controls for Integrodifferential Equations with Finite Delay 116

where
Z t
i
R(t)xi0 R(t − s) f (s, xi (s)) + C(s)ui (s) ds for t ∈ I (8.3.2)
 
x (t) = +
0

and ui ∈ Uad , for i = 1, 2.

Proof. Let xi , for i = 1, 2, be two mild solutions of equation (8.1.1),

corresponding to the controls ui ∈ Uad and λ > 0 such that x10 , x20 ∈ B(0, λ).

Z t
i
R(t)xi0 R(t − s) f (s, xi (s)) + C(s)ui (s) ds for t ∈ I
 
x (t) = +
0

By Lemma 8.3.1, we have that, there exists a constant ρλ = M f > 0 such

that kx (s)k ≤ ρλ , i = 1, 2.
i

Now, for t ∈ [0, b], we have

Z t
1 2 1 2
kx (t) − x (t)k ≤ Mb kx0 − x0 k + Mb kf (s, x1 (s)) − f (s, x2 (s))k ds
0
Z t
+ Mb kC(s)u1 (s) − C(s)u2 (s)k ds
0
Z t 
≤ Mb kx10 − x20 k + Mb Lf (ρλ ) kx1 (s) − x2 (s)k ds
0
Z t
+ Mb kC(s)u1 (s) − C(s)u2 (s)k ds
0
Z t
≤ Mb kx10 − x20 k + Mb Lf (ρλ ) kx1 (s) − x2 (s)k ds
0
√
+ Mb Lf (ρλ ) b kCk ku1 − u2 kL2

That is

1 2
√ Z t
kx (t)−x (t)k ≤ Mb kx10 −x20 k+Mb Lf (ρλ ) 1 2
b kCk ku −u kL2 +Mb Lf (ρλ ) kx1 (s)−x2 (s)k ds.
0
(8.3.3)
By Gronwall’s inequality, we have that
h √ i
kx1 (t) − x2 (t)k ≤ Mb kx10 − x20 k + Mb Lf (ρλ ) b kCk ku1 − u2 kL2 eMb Lf (ρλ ) b .
Optimal Controls for Integrodifferential Equations with Finite Delay 117

This implies that

h √ i
kx1 (t) − x2 (t)k ≤ Mb kx10 − x20 k + Mb Lf (ρλ ) b kCk ku1 − u2 kL2 eMb Lf (ρλ ) b ,

Let n √ o
∗ Mb Lf (ρλ )b Mb Lf (ρλ ) b
γ (λ) := max Mb e , Mb Lf (ρλ ) b kCk e .
Then, we have that
 
kx1 (t) − x2 (t)k ≤ γ ∗ (λ) kx10 − x20 k + ku1 − u2 kL2 for t ∈ [0, b].

And the proof is complete.

Now, we study the existence of solutions to the following Lagrange problem

 Find a control u0 ∈ Uad such that



(LP)
J (u0 ) ≤ J (u) for all u ∈ Uad ,


where Z b  
J (u) := L t, xu (t), u(t) dt,
0

and x denotes the mild solution of (8.1.1) corresponding to the control

u

u ∈ Uad and the initial data x0 .

For the existence of solutions to problem (LP), we make the following as-
sumptions.

(HL )

(i) The functional L : I × X × U → R ∪ {∞} is Borel measurable.

(ii) L(t, ·, ·) is sequencially lower semicontinuous on X × U for almost all

t ∈ I.

(iii) L(t, y, ·) is convex on U for each y ∈ X and almost all t ∈ I.

(iv) There exist constants β ≥ 0, γ > 0, and µ ∈ L1 (I) nonnegative such

that
L(t, y, u) ≥ µ(t) + βkyk + γkuk.

We have the following result on the existence of optimal controls for problem
(LP).
Optimal Controls for Integrodifferential Equations with Finite Delay 118

Theorem 8.3.3 Assume that hypotheses (H1 ) − (H5 ) and (HL ) hold. Then
the Lagrange problem (LP) admits at least one optimal pair, that is there
exists an admissible control pair (x0 , u0 ) ∈ C([0, b], X) × Uad such that
Z b   Z b  
0 0 0
J (u ) = L t, x (t), u (t) dt ≤ L t, xu (t), u(t) dt = J (u) for u ∈ Uad .
0 0

n o
Proof. If inf J (u) : u ∈ Uad = ∞, we are done.
n o
Without loss of generality, assume that inf J (u) : u ∈ Uad = δ < ∞.
Suppose that δ = −∞, then for each n ∈ N, there exists (un )n≥1 ⊂ Uad such
that
J (un ) < −n (∗)
Boundedness of Uad implies that (un )n≥1 is bounded and so there exists a
subsequence (unk )k≥1 of (un )n≥1 that converges weakly to some u0 in L2 (I, U ),
since L2 (I, U ) is reflexive. But Uad is closed and convex, so by Mazur’s
Lemma, it is weakly closed and therefore, u0 ∈ Uad . By hypothesis (HL ),
L(t, y, ·) is weakly lower semicontinuous, so we have that

L(t, y, u0 ) ≤ lim inf L(t, y, unk ) < −∞,

k→∞

which implies that J (u0 ) < −∞ using (∗). And this is a contradiction since
J (u0 ) ∈ R ∪ {∞}. Hence δ ∈ R.
Now by the definition of δ, there exists a minimizing sequence, a feasible pair
((xn , un ))n≥1 ⊂ Sad such that
Z b  
L t, xn (t), un (t) dt −→ δ as n → ∞,
0

where
n o
Sad := (x, u) : x is a mild solution of equation (8.1.1) corresponding to the control u ∈ Uad .

Boundedness of Uad and the fact that L2 (I, U ) is reflexive imply that (un )n≥1
has a subsequence denoted for simplicity by (uk )k≥1 , that converges weakly
to some u0 in L2 (I, U ). But Uad is closed and convex, so by Mazur’s Lemma,
it is weakly closed and therefore, u0 ∈ Uad .
Let
Z t
k
R(t − s) f (s, xk (s)) + C(s)uk (s) ds for t ∈ I
 
x (t) = R(t)x0 +
0
Optimal Controls for Integrodifferential Equations with Finite Delay 119

denote the subsequence of (xn )n≥1 corresponding to the control sequence

(uk )k≥1 and x0 be the mild solution corresponding to the control u0 ∈ Uad .
We show that xk → x0 .
For t ∈ [0, b], we have
Z t h i
k 0
x (t) − x (t) ≤ R(t − s) f (s, xk (s)) − f (s, x0 (s)) ds

0
Z t h i
+ R(t − s) C(s)uk (s) − C(s)u0 (s) ds

0
Z t Z t
k 0 k 0
≤ Mb Lf (ρ) x (s) − x (s) ds + Mb C(s)u (s) − C(s)u (s) ds

0 0
Z t
k 0
≤ Mb Lf (ρ) x (s) − x (s) ds

0

√
Z t 2  21
k 0
+Mb b C(s)u (s) − C(s)u (s) ds

0
Z t
k
√
≤ Mb Lf (ρ) x (s) − x0 (s) ds + Mb b Cuk − Cu0

0 L2 (I,U )

That is

k 0
Z t
k 0
√ k

0
x (t) − x (t) ≤ M L (ρ) x (s) − x (s) ds + M b Cu − Cu

b f b
0 L2 (I,U )
(8.3.4)
By Gronwall’s inequality, we have that
√ M bL (ρ)
∗∗
, where M ∗∗ = Mb
k 0 k 0
x (t) − x (t) ≤ M Cu − Cu be b f .

2
L (I,U )
(8.3.5)
We have the following Lemma.

Lemma 8.3.4 [96] Let (un )n≥1 ⊂ Uad and u0 ∈ Uad such that (un )n≥1 con-
verges weakly to u0 . Then,

−→ 0 as k → ∞, if C ∈ L∞ (I; L(U, X)).
k
Cu − Cu0 2

L (I,U )
Optimal Controls for Integrodifferential Equations with Finite Delay 120

We have by (8.3.5) that

x − x0 ≤ M ∗∗ Cuk − Cu0
k
,

L2 (I,U )

and therefore, it follows by Lemma 8.3.4 that

xk −→ x0 as k → ∞.

We note that (HL ) implies the assumptions of Balder’s Theorem. Hence by

Z b  
using Balder’s Theorem, we can conclude that (x, u) 7→ L t, x(t), u(t) dt
0
is sequencially lower semicontinuous in the strong topology of L1 (I, X) ×
L1 (I, U ).
Now, since L2 (I, X) × L2 (I, U ) ⊂ L1 (I, X) × L1 (I, U ), J is also sequencially
lower semicontinuous on L2 (I, X) × L2 (I, U ), and in the strong topology of
L1 (I, X × U ).
Hence, J is weakly lower semicontinuous on L2 (I, U ), and since by (HL ) − (iv),
J > −∞, J attains its infimum at u0 ∈ Uad , that is
Z b   Z b  
k k
δ = lim L t, x (t), u (t) dt ≥ L t, x (t), u (t) dt = J (u0 ) ≥ δ.
0 0
k→∞ 0 0

Thus, δ = J (u0 ), and hence there exists an admissible control u0 ∈ Uad such
that
J (u0 ) ≤ J (u) for all u ∈ Uad .
This completes the proof.

we now illustrate our main result by the following example.

Optimal Controls for Integrodifferential Equations with Finite Delay 121

8.4 Example
Let Ω be bounded domain in Rn with smooth boundary and consider the
following nonlinear integrodifferential equation.
 Z t
∂v(t, ξ)
= ∆v(t, ξ) + ζ(t − s)∆v(s, ξ) ds + α(t) sin(v 2 (t, ξ)) + β(t)ω(t, ξ)





 ∂t 0



for t ∈ [0, 1] = I and ξ ∈ Ω



v(t, ξ) = 0 for t ∈ [0, 1] and ξ ∈ ∂Ω









v(0, ξ) = v0 (ξ),

(8.4.1)
where α, β ∈ C(I, R), ω : I × Ω → R continuous in t, and ζ ∈ W 1,1 (R+ , R).

Let X = U = L2 (Ω). For η > 0, we define the set of admissible controls

Uad by
n o
Uad := u : I → U such that u is measurable and kukL2 (I,U ) ≤ η ,

where Z 1 Z 
kuk2L2 (I,U ) = 2
u (s)(ξ) dξ ds.
0 Ω

We define A : D(A) ⊂ X → X by:


 D(A) = H 2 (Ω) ∩ H01 (Ω)
(∗∗)
Av = ∆v for v ∈ D(A).


Theorem 8.4.1 (Theorem 4.1.2,
p. 79 2of [76]) A is the infinitesimal

generator of a C0 -semigroup T (t) t≥0 on L (Ω).

Define
∂v(t, ξ)
x(t)(ξ) = v(t, ξ), x0 (t)(ξ) = , ω(t, ξ) = u(t)(ξ).
∂t
f (t, x(t))(ξ) = α(t) sin x2 (t)(ξ) for t ∈ [0, 1] and ξ ∈ Ω.


 
C(t) : X → X be defined by C(t)u(t) (ξ) = C(t)u(t)(ξ) = β(t)ω(t, ξ).

(B(t)x)(ξ) = ζ(t)∆v(t, ξ) for t ∈ [0, 1], x ∈ D(A) and ξ ∈ Ω.

Optimal Controls for Integrodifferential Equations with Finite Delay 122

Equation (8.4.1) is then transformed into the following form

 Z t
0
 x (t) = Ax(t) + B(t − s)x(s)ds + f (t, x(t)) + C(t)u(t) for t ∈ I = [0, 1],


0


x(0) = x0 .

(8.4.2)
The operator A defined in (**) is closed, linear and densely
defined and so
satisfies hypothesis (H1 ). Also, the family of operators B(t) t≥0 defined by
B(t)y = ζ(t)∆y is linear and belongs to W 1,1 (R+ , R) since ζ ∈ W 1,1 (R+ , R),
and it follows that hypothesis (H2 ) is satisfied. Therefore, equation (8.4.2)
satisfies hypotheses (H1 ) and (H2 ), which guarantees the existence of a
unique resolvent operator.
One can see that, f satisfies (H3 ). Now we consider the following cost func-
tion: Z 1  
u
J (u) := L t, x (t), u(t) dt,
0
where
L : [0, 1] × L2 (Ω) × L2 (Ω) −→ R is defined by:
 
L t, x, u = kxk + kuk.

L satisfies all the conditions of hypothesis (HL ). Then,

Z 1 
J (u) = kxu (t)k + ku(t)k dt.
0

Hence, all the conditions of Theorem 8.3.3 are satisfied, and therefore, equa-
tion (8.4.2) has at least one optimal pair.
 Conclusion and Perspectives

In this work, we have been able to make the following contributions. In the
first part, we investigated the controllability results for some partial func-
tional integrodifferential equations with nonlocal initial conditions ( equation
(1.3.4)), with finite delay (equation (1.3.5)) and with infinite delay (equation
(1.3.6)) in Banach spaces, using fixed point techniques and measure of non-
compactness. In the second part, we established the existence of optimal
pairs for the Lagrange problems associated with some partial functional in-
tegrodifferential equations with bounded delay, with unbounded delay and
with classical Cauchy initial condition in Banach spaces, using techniques of
convex optimization. The main results are obtained by supposing that the
linear homogeneous and undelayed parts of these equations admit a resolvent
operator in the sense of Grimmer. We note that this class of integrodifferen-
tial equations have not been investigated (to the best of our knowledge) for
controllability and optimal controls.
The importance of the problems considered in this work is that they have
opened doors for new deep research problems such as approximate controlla-
bility which is a weaker notion than the axact controllability treated in this
thesis, for these partial functional integrodifferential equations. For future
work, we shall consider equations (1.3.4), (1.3.5) and (1.3.6) for approximate
controllability without compactness assumption on the resolvent operator,
we shall also extend them by a Wiener process in the stochastic frame work
and use iterative methods for existence of solutions and possibly also for
existence of optimal controls.

123
 Bibliography

[1] A. Aghajani, J. Banas, Y. Jalilian, Existence of solutions for a class of

nonlinear Volterra singular integral equations, Comput. Math. Appl. 62
(2011) 1215-1227.

[2] Agrachev, A.A. (ed.): Mathematical Control Theory. Summer School on

Mathematical Control Theory, First Edition, The Abdus Salam ICTP
Publications and Printing Section (2002)

[3] R.R. Akhmerov, M.I. Kamenskii, A.S. Potapov, A.E. Rodkina, B.N.
Sadovskii, Measures of noncompactness and condensing operators,
Translated from the Russian by A. Iacob, Birkhäuser Verlag, (1992).

[4] R. P. Argawal, M. Benchohra, D. Seba, On the application of measure

of noncompactness to the existence of solutions for fractional differential
equations, Results Math. 55 (2009) 221-230.

[5] O. Arino, M. L. Hbid, E. Ait Dads (eds), Delay Differential Equations

and Applications, Springer, (2006).

[6] J. M. Ayerbee Toledano, T. Dominguez Benavides, G. Lopez Acedo,

Measures of Noncompactness in Metric Fixed Point Theory, volume
99 of Operator Theory, Advances and Applications, Birkhäuser, Basel,
1997.

[7] K Balachandran and J. P. Dauer, Controllability of Nonlinear Systems

in Banach Spaces: A Survey, Journal of Optimization Theory and Ap-
plications, Vol. 115, No. 1, 7-28, (2002).

124
Bibliography 125

[8] K. Balachandran, R. Sakthivel, Controllability of Functional Semilinear

Integrodifferential Systems in Banch Spaces, Journal of Mathematical
Analysis and Applications, 255, 447-457, (2001).
[9] K. Balachandran and J. Y. Park, Existence of Solutions and Controlla-
bility of Nonlinear Integrodifferential Systems in Banach Spaces, Math-
ematical Problems in Engineering 2003:2, 65-79, (2003) .
[10] E. Balder, Necessary and sufficient conditions for L1 -strong-weak lower
semicontinuity of integral functional, Nonlinear Anal., Real World Appl.
11, (1987), 1399-1404.
[11] J. Banas, Measures of noncompactness in the study of solutions of non-
linear differential and integral equations, Cent. Eur. J. Math. 10(6)
(2012) 2003-2011.
[12] J. Banas, K. Goebel, Measures of Noncompactness in Banach Spaces,
vol. 60 of Lecture Notes in Pure and Applied Mathematics, Marcel
Dekker, New York, (1980).
[13] A. Boyarsky, On the existence of optimal controls for nonlinear systems,
Journal of Optimization Theory and Applications, Vol. 20, No. 2, (1976),
205-213.
[14] L. Byszewski, Theorems about the existence and uniqueness of solutions
of a semilinear evolution nonlocal Cauchy problem, Journal of Mathe-
matical Analysis and Applications, 162, 494-505 ( 1991)
[15] B. Cahlon, D. M. Kulkarni, P. Shi, Stepwise stability for the heat equa-
tion with a nonlocal constraint, SIAM J. NUMER. ANAL. Vol. 32, No.
2, 571-593 (1995)
[16] Y.K. Chang, J.J. Nieto, W.S. Li, Controllability of Semilinear Differ-
ential Systems with Nonlocal Initial Conditions in Banach Spaces, J.
Optim. Theory Appl. 142, 267-273, (2009).
[17] G. Chen and R. Grimmer, Semigroups and integral equations, J. Integral
Equations, 2, 133-154, (1980).
[18] W. L. Chow, Uber systeme von linearen partiellen differentialgleichungen
erster ordnung, Math. Ann. 117, 98-105, (1939).
[19] Cristóbal González, A. Jiménez-Melado and E. Llorens-Fuster, A Mönch
Type Fixed Point Theorem Under The Interior Condition, J. Math.
Anal. Appl. 352, 816-821, (2009).
Bibliography 126

[20] C. M. Dafermos & A. Nohelj, Energy methods for non linear hyperbolic
Volterra integrodifferential equations, Communs partial diff. Eqns. 4,
219-278 (1979).
[21] G. Darbo, Punti uniti in transformazioni a condominio non compatto,
Rend. Sem. Math. Univ. Padova 24 (1955) 84-92.
[22] M.C. Delfour, J. P. Zolésio, Shapes and Geometries: Metrics, Analysis,
Differential Calculus, and Optimization, SIAM series on Advances in
Design and Control, Society for Industrial and Applied Mathematics,
Philadelphia, Second Edition (2011)
[23] W. Desch, R. Grimmer, W. Schappacher, Well-posedness and wave prop-
agation for a class of integrodifferential equations in Banach space, Jour-
nal of Differential Equations. 74, No 2, 391-411 (1988)
[24] W. Desch, R. Grimmer, W. Schappacher, Some Considerations for Lin-
ear Integrodifferential Equations, Journal of Mathematical analysis and
Applications 104, 219-234, (1984)
[25] I. Z. Djolović, E. Malkowsky, The Hausdorff measure of noncompact-
ness of operators in the matrix domains of triangles in the spaces of
C1 summable and bounded sequences, Appl. Math. Comp. 216 (2010)
1122-1130.
[26] Dugundij, J., Granas, A Fixed Point Theory, Mongrafie Mat. PWN,
Warsaw (1982)
[27] E. Flytzanis, N. S. Papageorgiou, On the existence of optimal controls
for a class of nonlinear infinite dimensional systems, Mathematische
Nachrichten, 150, (1991), 203-217.
[28] R. Grimmer, Resolvent operators for integral equations in a Banach
space, Transaction of American Mathematical society. 273, 333-349
(1983)
[29] R. Grimmer & A. J. Pritchard, Analytic resolvent operators for integral
equations in Banach space, J. diff. Eqns. 50, 234-259 (1983).
[30] R. Grimmer & Kappelf, Series expansions for resolvents of Volterra in-
tegrodifferential equations in Banach space, SIAM J. math. Analysis 15,
595-604 (1984).
[31] M. E. Gurtin & A. C. Pipkin, A genera1 theory of heat conduction with
finite wave speeds, Archs ration. Mech. Analysis 31, 113-126 (1968).
Bibliography 127

[32] K. B. Hannsgen, The resolvent kernel of an integrodifferential equation

in Hilbert spaces, SIAM J. Math. Anal., 7 (1976), 481-490.
[33] K. B. Hannsgen, Uniform L1 behavior for an integrodifferential equation
with parameter, SIAM J. Math. Anal., 8 (1977), 626-639.
[34] R. Hermann, On the accessibility problem in control theory, Int. Symp.
Diff. Eqs. and Mechanics, Academic Press, NY, 1973.
[35] Hongwei Lou, Existence of optimal controls for semilinear parabolic
equations without Cesari-type conditions, Applied Mathematics and Op-
timization, 47, (2003), 121-142 .
[36] M. Jakszto, A. Skowron, Existence of optimal controls via continuous
dependence on parameters, Computers and Mathematics with Applica-
tions, 43, (2003), 1657-1669.
[37] Jerzy Zabczyk, Mathematical Control Theory: An Introduction, Modern
Birkhäuser Classics, Boston, (2008).
[38] Jin Liang, James H. Liu, Ti-Jun Xiao, Nonlocal Problems for Integrod-
ifferential Equations, Dynamics of Continuous, Discrete and Impulsive
Systems Series A: Mathematical Analysis 15, 815-824, (2008).
[39] JinRong Wang, Yong Zhou, Milan Medved, On the solvability and op-
timal controls of fractional integrodifferential evolution systems with in-
finite delay Journal of Optimization Theory and Applications, 152,
(2012), 31-50.
[40] JinRong Wang, Zhenbin Fan, Yong Zhou, Nonlocal Controllability of
Semilinear Dynamic Systems with Fractional Derivative in Banach
Spaces, J. Optim. Theory Appl. 154: 292-302,(2012).
[41] R. E. Kalman, Y.C. Ho, and K.S. Narendra, Controllability of linear
dynamical systems, Contr. Diff. Eqs. 1, 189-213, (1963).
[42] Khalil Ezzinbi, Guy Degla, Patrice Ndambomve, Controllability for
some partial functional integrodifferential equations with nonlocal con-
dition in Banach spaces, Discussiones Mathematicae, Differential In-
clusions, Control and Optimization, Vol. 35., No. 1, 1-22, (2015), (to
appear ).
[43] Khalil Ezzinbi, Hamidou Toure, Issa Zabsonre, Existence and regularity
of solutions for some partial functional integrodifferential equations in
Banach spaces, Nonlinear Analysis, 70, 2761-2771, (2009).
Bibliography 128

[44] Khalil Ezzinbi and Patrice Ndambomve, Controllability for some

Partial Functional Integrodifferential Equations in Banach Spaces, (Ac-
cepted to Appear: American Journal of Mathematical Analy-
sis).

[45] Khalil Ezzinbi and Patrice Ndambomve, On the Controllability of

some Partial Functional Integrodifferential Equations with infinite delay
in Banach Spaces, (Accepted to Appear: Advances in Fixed Point
Theory ).

[46] J. Klamka, Controllability of dynamical systems. A survey, Bulletin of

the Polish Academy of Sciences, Technical Sciences, Vol. 61, No. 2,
(2013).

[47] A. Krener, A generalization of Chow’s theorem and the bang-bang theo-

rem to nonlinear control systems, SIAM J. Control 12, 43-52, (1974).

[48] K. Kuratowski, Sur les espaces complets, Fund. Math. 15 (1930) 301-309.

[49] D. La Torre, H. Kunze, M. Ruiz-Galan, T. Malik, and S. Marsiglio,

Optimal control: theory and application to science, engineering, and
social sciences, Abstract and Applied Analysis, (2015), 90527-1-90527-
2.

[50] S. Lenhart and J. T. Workman, Optimal Control Applied to Biological

Models, Mathematical and Computational Biology, Chapman & Hall,
Boca Raton, Fla, USA ;CRC Press, London, UK, (2007).

[51] J. Liang, J. H. Liu, Ti-Jun Xiao, Nonlocal problems for integrodifferen-

tial equations, Dynamics of Continuous, Discrete and Impulsive Systems
Series A: Mathematical Analysis 15, (2008), 815-824.

[52] C. Lobry, Controllabilité des systèmes non linèaires, SIAM J. Contr. 8,

573-605, (1970).

[53] A. Lorenzi, and E. Sinestrari, An inverse problem in the theory of materi-

als with memory, Nonlinear Analysis: Theory, Methods & Applications,
Vol. 12, No. 12, 1317-1335, (1988).

[54] A. Lunardi & E. Sinestrari, C α -regularity for non autonomous linear

integrodifferential equations of parabolic type, J. diff. Eqns. 68. 88-116
(1986).
Bibliography 129

[55] Machado et al, Controllability results for impulsive mixed-type functional

integro-differential evolution equations with nonlocal conditions, Fixed
Point Theory and Applications 2013, 2013:66.

[56] Meili Li, Miansen Wang, Fengqin Zhang, Controllability of Impulsive

Functional Differential Systems in Banach Spaces, Chaos, Solitons and
Fractals 29, 175-181, (2006).

[57] C. Meyer P. Philip and F. Tröltzsch, Optimal control of a semilinear

PDE with nonlocal radiation interface conditions, SIAM Journal of Con-
trol and Optimization, 45, (2006), 699-721.

[58] R. K. Miller, Volterra integral equations in a Banach space, Funkcial.

Ekvac. 18 (1975), 163-193.

[59] R. K. Miller, An integrodifferential equation for rigid heat conductors

with memory, J. Math. Anal. Appl. 66 (1978), 313-332.

[60] R. K. Miller and R. L. Wheeler, Asymptotic behavior for a linear Volterra

integral equation in Hilbert spaces, J. Differential Equations 23 (1978),
270-284.

[61] R. K. Miller and R. L. Wheeler, Well-posedness and stability of linear

Volterra integrodifferential equations in abstract spaces, Funkcial. Ekvac.
21 (1978), 279-305.

[62] H. Mönch, Boundary value problems for nonlinear ordinary differential

equations of second order in Banach spaces, Nonlinear Analysis, Theory,
Methods and Applications. Vol. 4, No. 5, 985-999 (1980)

[63] P. Neittaanmäki, D. Tiba, Optimal Control of Nonlinear Parabolic Sys-

tems: Theory, Algorithms and Applications, Pure and Applied Mathe-
matics, A series of Monographs and Textbooks, 179 (1994)

[64] E. S. Noussair, S. Nababan, and K. L. Teo, On the existence of optimal

controls for quasilinear parabolic partial differential equations, Journal
of Optimization Theory and Applications, Vol. 34, No. 1, (1981), 99-115.

[65] N. S. Papageorgiou, Existence of optimal controls for nonlinear systems

in Banach spaces, Journal of Optimization Theory and Applications,
Vol. 53, No. 3, (1987), 451-459.

[66] A. Pazy, Semigroups of Linear Operators and Applications to Partial

Differential Equations, Springer, 1983.
Bibliography 130

[67] Qi Xiang DONG and Gang LI, Measure of noncompactness and semi-
linear nonlocal functional differential equations in Banach spaces, Acta
Mathematica Sinica, English Series, Vol. 31, No. 1, pp. 140-150, 2015.
[68] M. D. Quinn, N. Carmichael, An approach to nonlinear control problem
using fixed point methods, degree theory and pseudo-inverses, Numer.
Funct. Anal. Optimi. 7, 197-219 (1984)
[69] Rahima Atmania, Said Mazouzi, Controllability of Semilinear Integrod-
ifferential Equations with Nonlocal conditions, Electronic Journal of Dif-
ferential Equations. Vol. 2005, No. 75, 1-9, (2005).
[70] R. Sakthivel, N.I. Mahmudov, Juan. J. Nieto, Controllability for a class
of fractional-order neutral evolution control systems, Applied Mathemat-
ics and Computation 218, 10334-10340, (2012).
[71] S. Selvi and M. Malika Arjunan, Controllability Results for Impulsive
Differential Systems With Finite Delay, J. Nonlinear Sci. Appl. 5, 206-
219, (2012).
[72] E. Sinestrari, An integrodifferential equation arising from the theory
of nonlinear heat flow with memory. in Nonlinear Parabolic Equations:
Qualitarioe Properties of Solutions, Research Notes in Mathematics
(Edited by L. Boccardo and A. Tesei), 149. 207-218. Pitman (1987).
[73] M. Schulz, Control Theory in Physics and Other Fields of Science: Con-
cepts, Tools and Applications, Springer Tracts in Modern Physics, Vol.
215 (2006)
[74] H. J. Sussmann, and V. Jurdjevic, Controllability of nonlinear systems,
J. Diff. Eqs. 12, 95-116, (1972).
[75] C. C. Travis and G. F. Webb, Existence and stability for partial func-
tional differential equations, Transactions of the American Mathematical
Society, Vol. 200, 395- 418, (1974).
[76] I. I. Vrabie, C0 -Semigroups and Applications, Mathematics Studies. 191,
(2003)
[77] J. Wang, W. Wei, Controllability of integrodifferential systems with non-
local initial conditions in Banach spaces, Journal of Mathematical Sci-
ences, Vol. 177, No. 3, 459-465 (2011)
[78] J. R. Wang and Y. Zhou, A class of fractional evolution equations and
optimal controls, Nonlinear Anal: RWA., 12 (2011), 262-272.
Bibliography 131

[79] W. Wei, X. Xiang and Y. Peng, Nonlinear impulsive integrodifferential

equation of mixed type and optimal controls, Optim., 55 (2006), 141-156.

[80] Xiuwen Li and Zhenhai Liu, The solvability and optimal controls of im-
pulsive fractional semilinear differential equations, Taiwanese Journal of
Mathematics, Vol. 19, No. 2, (2015), 433-453.

[81] E. Zeidler, Nonlinear Functional Analysis and Its Applications II/A,

Springer, New York (1990).

[82] Hale J and Kato J. Phase space for retarded equations with infinite delay,
Funkcial. Ekvac. 21 (1978), 11-41.

[83] S. Baghli, M. Benchohra and K. Ezzinbi, Controllability Results for

Semilinear Functional and Neutral Functional Evolution Equations with
Infinite Delay, Surveys in Mathematics and its Applications, Vol.4, 15-
39, (2009).

[84] E. Sinestrari, An integrodifferential equation arising from the theory

of nonlinear heat flow with memory. in Nonlinear Parabolic Equations:
Qualitative Properties of Solutions, Research Notes in Mathematics
(Edited by L. Boccardo and A. Tesei), 149. 207-218. Pitman (1987).
PART II

[85] R. Grimmer, Resolvent operators for integral equations in a Banach

space, Transaction of American Mathematical society, 273, 333-349,
(1983).

[86] K. Ezzinbi, H. Toure and I. Zabsonre, Existence and regularity of solu-

tions for some partial functional integrodifferential equations in Banach
spaces, Nonlinear Analysis 70, 2761-2771 (2009).

[87] A. Boyarsky, On the existence of optimal controls for nonlinear systems,

Journal of Optimization Theory and Applications, Vol. 20, No. 2, 205-
213 (1976).

[88] E. Flytzanis and N.S. Papageorgiou, On the existence of optimal controls

for a class of nonlinear infinite dimensional systems, Mathematische
Nachrichten, 150, 203-217 (1991).

[89] L. Hongwei, Existence of optimal controls for semilinear parabolic equa-

tions without Cesari-type conditions, Applied Mathematics and Opti-
mization, 47, 121-142 (2003).
Bibliography 132

[90] M. Jakszto and A. Skowron, Existence of optimal controls via continuous

dependence on parameters, Computers and Mathematics with Applica-
tions, 43, 1657-1669 (2003).
[91] Khalil Ezzinbi and Patrice Ndambomve, Solvability and optimal con-
trols for some Partial Functional Integrodifferential Equations with finite
delay, (Accepted to Appear: American Journal of Modeling and
Optimization).
[92] Khalil Ezzinbi and Patrice Ndambomve, Partial Functional Inte-
grodifferential Equations and Optimal Controls in Banach Spaces, (Ac-
cepted to Appear: Communications in Optimization Theory).
[93] Khalil Ezzinbi and Patrice Ndambomve, Solvability and optimal con-
trols for some Partial Functional Integrodifferential Equations with in-
finite delay in Banach Spaces, (Under Review: Optimal Control:
Applications and Methods).
[94] E. S. Noussair, S. Nababan and K. L. Teo, On the existence of optimal
controls for quasilinear parabolic partial differential equations, Journal
of Optimization Theory and Applications, Vol. 34, No. 1, 99-115 (1981).
[95] N. S. Papageorgiou, Existence of optimal controls for nonlinear systems
in Banach spaces, Journal of Optimization Theory and Applications,
Vol. 53, No. 3, 451-459 (1987).
[96] J. Wang, Y. Zhou and M. Medved, On the solvability and optimal con-
trols of fractional integrodifferential evolution systems with infinite delay,
Journal of Optimization Theory and Applications, 152, 31-50, (2012).
[97] J. R. Wang and Y. Zhou, A class of fractional evolution equations and
optimal controls, Nonlinear Analysis: Real World Applications„ 12, 262-
272 (2011).
[98] W. Wei, X. Xiang and Y. Peng, Nonlinear impulsive integrodifferential
equation of mixed type and optimal controls, Optimization, 55, 141-156
(2006).
[99] X. Li and Z. Liu, The solvability and optimal controls of impulsive frac-
tional semilinear differential equations, Taiwanese Journal of Mathemat-
ics, Vol. 19, No. 2, 433-453 (2015).
[100] C. M. Dafermos and A. Nohelj, Energy methods for non linear hyper-
bolic Volterra integrodifferential equations, Communications in Partial
Differential Equations, 4, 219-278 (1979).
Bibliography 133

[101] M. E. Gurtin and A. C. Pipkin, A genera1 theory of heat conduction

with finite wave speeds, Archive for Rational Mechanics and Analysis,
31, 113-126, (1968).

[102] E. Sinestrari, An integrodifferential equation arising from the theory

of nonlinear heat flow with memory in Nonlinear Parabolic Equations:
Qualitative Properties of Solutions, Research Notes in Mathematics
(Edited by L. Boccardo and A. Tesei), 149. 207-218. Pitman (1987).

[103] R. Grimmer and A. J. Pritchard, Analytic resolvent operators for in-

tegral equations in Banach space, Journal of Differential Equations, 50,
234-259 (1983).

[104] R. Grimmer and C. Kappelf, Series expansions for resolvents of

Volterra integrodifferential equations in Banach space, SIAM Journal
of Mathematical Analysis, 15, 595-604 (1984).

[105] A. Lunardi and E. Sinestrari, C α -regularity for non autonomous lin-

ear integrodifferential equations of parabolic type, Journal of Differential
Equations 68, 88-116 (1986).

[106] J. Liang, J. H. Liu and Ti-Jun Xiao Nonlocal problems for integrod-
ifferential equations, Dynamics of Continuous, Discrete and Impulsive
Systems Series A: Mathematical Analysis 15, 815-824 (2008).

[107] E. Balder, Necessary and sufficient conditions for L1 -strong-weak lower

semicontinuity of integral functional, Nonlinear Analysis: Real World
Applications 11, 1399-1404 (1987).

[108] X. Li and J. Yong, Optimal Control Theory Infinite Dymensional Sys-

tems, BirkhÃďuser, p. 39 (1995).

[109] K. Ezzinbi, H. Toure and I. Zabsonre, Local existence and regularity

of solutions for some partial functional integrodifferential equations with
infinite delay in Banach spaces, Nonlinear Analysis 70 (2009), 3378-3389.