On Approximate Controllability of Evolution Systems with Nonlocal

Conditions

Noyiessie Ndebeka Rostant

Institute of mathematics and physical sciences (IMSP)
Benin

Supervised by: Prof. KHALIL EZZINBI and Prof. GUY DEGLA

03 August 2018
Submitted in Partial Fulfillment of a Masters Degree at IMSP-Benin
Abstract
In this essay, we are concerned with the approximate controllability of the abstract evolution equations with
nonlocal conditions. We start by making more explicitely what Mahmudov [11] have done on the approximate
controllability of certain classes of abstract evolution equations of the form
(
x0 (t) = Ax(t) + Bu(t) + f (x(t))
x(0) = x0 + g(x),

where A is the infinitesimal generator a of strongly continuous semigroup (C0 −semigroup), B a bounded operator.
In fact, assuming the approximate controllability of the corresponding linearized equation Mahmodov [11] obtains
sufficient conditions for the approximate controllability of the abstract evolution equation. Since the results of
Mahmudov [11] are a generalization and continuation of the recent results on this issue, we are going to see how
to extend it on others classes of evolution equations and take a look at the necessaries conditions.

Declaration
I, the undersigned, hereby declare that the work contained in this essay is my original work, and that any work
done by others or by myself previously has been acknowledged and referenced accordingly.
Noyiessie Ndebeka Rostant, 03 August 2018.

i
Contents

Abstract i

1 Introduction 2
1.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Preliminaries 3
2.1 Fréchet derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
∗
2.2 Weak and Weak Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 C0 − semigroup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.4 The Arzela-Ascoli Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Approximate controllability of the semilinear parabolic evolution equations with nonlocal condi-
tions 7
3.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 The nonlocal initial condition does not depends on the solution . . . . . . . . . . . . . . . . . . . 7
3.3 The nonlocal initial condition depends on the solution . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Conclusion and Perspectives 27

A Proof of some theorems 28

References 30

ii
 Page 1

1
1. Introduction
1.1 Notations

2
2. Preliminaries
In this chapter, we shall give some mathematical tools that is, definitions, theorems, lemmas, corollaries, remarks
etc., while formulating some proofs as far as possible. The purpose of those mathematical tools is to enable us to
give the proof of controllability of semilinear evolution equations under some sufficient conditions. Observe that
some proofs are given in appendix.

2.1 Fréchet derivative

2.1.1 Definition. [15]Fréchet derivative The Fréchet derivative is a derivative defined on Banach spaces. Let X
and Y be Banach spaces, and U ⊂ X be an open subset of X. A function f : U → Y is called Fréchet differen-
kf (a + h) − f (a) − AhkY
tiable at a ∈ U if there exists a bounded linear operator A : X → Y such that lim = 0.
h→0 khkX
The limit here is meant in the usual sense of a limit of a function defined on a metric space (see functions on
metric spaces), using X and Y as the two metric spaces, and the above expression as the function of argument
h in X. As a consequence, it must exists for all sequences hhn i∞n=1 of non-zero elements of X which converge to
the zero vector hn →0.

2.2 Weak and Weak∗ Convergence

2.2.1 Definition. [8]Weak and Weak∗ Convergence Let X be a normed linear space. A sequence {fn } in X
is said to converge weakly in X to f in X provided
lim T (fn ) = T (f ) for all T ∈ X ∗ .
n→∞

We write {fn } * f in X to mean that f and each fn belong to X and {fn } converges weakly in X to f . We
continue to write {fn } → f in X to mean that lim ||fn − f ||X = 0 and, to distinguish this mode of convergence
n→∞
from weak convergence, often refer to this mode of convergence as strong convergence in X. Since
||T (fn ) − T (f )||X = ||T (fn − f )||X ≤ ||T ||X ∗ ||fn − f ||X for all T ∈ X ∗ ,
a sequence converges strongly, then it converges weakly. The converse is false.
1 1
2.2.2 Definition. [9] Let 1 ≤ p, q ≤ +∞ such that + = 1 and U be a measurable set. A sequence
p q
{fn } ⊂ Lp (U ) converges weakly to f ∈ Lp (U ), in which case we write fn * f in Lp (U ) if
Z Z
fn gdx → f gdx ∀g ∈ Lq (U ).
U U
∗
When p = +∞, we say that a sequence {fn } ⊂ Lp (U ) converges weakly∗ to f in L+∞ (U ), and we write fn * f
in L+∞ (U ), if Z Z
fn gdx → f gdx ∀g ∈ L1 (U ).
U U
We remark that when Ω is bounded the weak convergence of fn in L+∞ (Ω) to some f ∈ L+∞ (Ω) implies weak
convergence of fn to f in any Lp (Ω), 1 ≤ p < +∞.

2.3 C0 − semigroup
2.3.1 Definition. [6](C0 − semigroup) Let X be Banach space. A family {T (t)}t≥0 of bounded linear operators
from X to X is a strongly continuous semigoup (C0 − semigroup) of bounded operators if

3
Section 2.3. C0 − semigroup Page 4

• T (0) = idX ;
• T (t + s) = T (t)T (s) ∀t, s ≥ 0;
• ∀x ∈ X, R 3 t → T (t)x is continuous at 0.
2.3.2 Definition. [6]Infinitesimal generator
The linear operator A defined by
T (t)x − x
D(A) = {x ∈ X : lim exists} (2.3.1)
t→0+ t
and 
T (t)x − x d+ T (t)x 
∀x ∈ X, Ax = lim+ = (2.3.2)
t dt 

t→0
t=0

is the infinitesimal generator of the semigroup {T (t)}t≥0 ; D(A) is the domain of A. We would like to emphasize
that {T (t)}t≥0 is say to be a uniformly continuous semigroup if:

• T (0) = idX ;
• T (t + s) = T (t)T (s) ∀t, s ≥ 0;
• limt→0 ||T (t) − I||L(X) = 0. From this definition it is clear that if {T (t)}t≥0 is a uniformly continuous
semigroup of bounded linear operators then

lim ||T (t) − T (s)||L(X) = 0.

t→s

In fact
– If t > s, let h = t − s −→ 0+ and

lim ||T (t) − T (s)||L(X) = lim+ ||T (s + h) − T (s)||L(X) = ||T (s)|| lim+ ||T (h) − I||L(X) = 0;
t→s h→0 h→0

– If t < s, let h = s − t −→ 0+ and

lim ||T (t) − T (s)||L(X) = lim+ ||T (s − h) − T (s)||L(X)

t→s h→0

= lim+ ||T (s − h)||||T (h) − I||L(X) ≤ lim+ M ew(s−h) ||T (h) − I||L(X) = 0.
h→0 h→0

wt
Notice that the majoration ||S(t)||L(X) ≤ M e ∀t ≥ 0 with M > 1 and w ≥ 0 will be show in
what follows and also it is clear that uniformly continuous semigroups are particular cases of strongly
continuous semigroups.
2.3.3 Definition. [6]Compact operators
Let X, Y Banach spaces and T ∈ L(X, Y ), T is said to be compact if for each sequence {xn }n≥0 ⊂ X with
||xn || = 1 for each n ∈ N, the sequence {T (xn )}n≥0 has a subsequence which converges in Y . Equivalently,
T is compact if for each bounded sequence {xn }n≥0 ⊂ X, the sequence {T (xn )}n≥0 has a subsequence which
converges in Y. That is T is compact if and only if T (BX (0, 1)) is compact, if and only if ∀B bounded set in X,
T (B) is sequentially (relatively) compact on Y.
2.3.4 Definition. [6]compact semigroup We say that a C0 − semigroup {T (t)}t≥0 is compact if T (t) is
compact ∀t ≥ 0. Futhermore if there exists t0 ≥ 0 such that T (t0 ) is compact then we say that the C0 −
semigroup {T (t)}t≥0 is eventually compact. An important thing is that if {T (t)}t≥0 is a compact semigroup
then the function it is norm continuous, that is the function T defined as
 
T : 0, +∞ −→ L(X)
t −→ T (t) is continuous.
Section 2.3. C0 − semigroup Page 5

More precisely
lim ||T (t + h) − T (t)||L(X) = 0 ∀t > 0.
h→0

In fact let λ > 0 and M > 0 be such that ||T (t)|| ≤ M for 0 < t ≤ λ we will prove that such of λ and
M always exist. Setting Rt = {T (t)x : ||x|| ≤ 1}, for a fixed t > 0, Rt is relatively compact in X since
the operator T (t) is compact. Hence for  > 0 there exists a finite sequence {xi }N i=1 ⊂ X such that Rt =
SN 

B
i=0 X T (t)x i , . For a fixed x i the function T (.)x i ∈ C(0, T ; X) thus
2(M + 1)

lim ||T (h + t)xi − T (t)xi ||X = 0

h→0

then there exists λ0 with 0 < λ0 ≤ λ such that


|h| ≤ λ0 =⇒ ||T (h + t)xi − T (t)xi ||X < .
2
SN 

Let x ∈ X with ||x|| ≤ 1 then T (t)x ∈ Rt = i=0 BX T (t)xi , which implies that there exists
2(M + 1)
ix ∈ {xi }N
i=1 such that

||T (t)x − T (t)xix || ≤ .
2(M + 1)
For x ∈ X with ||x|| ≤ 1 and |h| ≤ λ0

||T (h + t)x − T (t)x||X ≤ ||T (h)||||T (t)x − T (t)xix ||X + ||T (h + t)xix − T (t)xix ||X + ||T (t)x − T (t)xix ||X
 
  
≤M + + = .
2(M + 1) 2 2(M + 1)

Finally

|h| ≤ λ0 =⇒ ||T (h + t)x − T (t)x||X ≤  ∀x ∈ X with ||x|| ≤ 1


=⇒ ||T (h + t) − T (t)||L(X) ≤ which concludes the proof.
2
2.3.5 Remark. If {T (t)}t≥0 is an eventually compact C0 − semigroup, let t0 ≥ 0 such that T (t0 ) is compact
then T (t) is also compact for all t ≥ t0 .

2.3.6 Lemma. [7],[6]Uniform Boundedness Principle

Let K ⊂ L(X, Y ): ∀x ∈ X,
sup{||T (x)|| : T ∈ K} < +∞.
Then K is uniformly bounded, that is
sup{||T || : T ∈ K} < +∞.

Proof.
 
If {T (t)}t≥0 is a strongly continuous
  semigroup, then the finite orbits {T (t)x : t ∈ 0, t0 } are continuous
images of a compact interval ( 0, t0 ), hence also compact, hence bounded ∀x ∈ X. It follows from the Uniform
Boundedness Principle that every C0 −semigroup is uniformly bounded in every compact interval. We are going to


prove that this has as a consequence that every C0 − semigroup is exponentially bounded on the whole 0, +∞ .
2.3.7 Proposition. [6],[7]
For every C0 − semigroup {T (t)}t≥0 , there exist constants ω ∈ R+ and M ≥ 1 such that

||T (t)|| ≤ M etω ∀t ≥ 0.

Proof.
Section 2.4. The Arzela-Ascoli Theorem Page 6

2.3.8 Theorem. [4]Gronwall inequality    

Letf, gand h be real continuous functions defined in a, b , such that g(t) ≥ 0 ∀t ∈ a, b . We suppose that
on a, b we have the inequality
Z t
f (t) ≤ h(t) + f (s)g(s)ds.
a
Then Z t Rt
g(r)dr
f (t) ≤ h(t) + h(s)g(s)e s ds.
a

Proof.

2.4 The Arzela-Ascoli Theorem

The Arzela-Ascoli Theorem gives sufficient conditions for compactness in certain function spaces. Among other
things, it helps provide some additional perspective on what compactness means.
2.4.1 Definition.
 [12]equicontinuous set of functions
A set F ⊂ C( 0, 1 ) is (uniformly) equicontinuous if and only if for any  > 0 there exists a δ > 0 such that for
all f ∈ F, if |x − y| < δ then |f (x) − f (y)| < . That is, if F is (uniformly) equicontinuous then every f ∈ F is
uniformly continuous and for every  > 0 and every f ∈ F, we can use the same δ > 0.
 
2.4.2 Theorem. [12]Basic Arzela-Ascoli If F ⊂ C( 0, 1 ) is closed, bounded, and equicontinuous then it is
compact.

Proof.
2.4.3 Theorem. [12]Generalized Arzela-Ascoli Theorem
Let (X, dX ) and (Y, dY ) be compact metric spaces and let F be a subset of C(X, Y ). If F is closed and
equicontinuous then it is compact. Notice that in case where X is not compact if G ⊂ X is a compact metric
subspace of X and F ⊂ C(G, Y ) such that for each x ∈ G, the set {f (x) : f ∈ F } is relatively compact in Y.
Moreover, F is uniformly bounded and equicontinuous then F is relatively compact.

Proof.
2.4.4 Theorem. [1]Banach Theorem
Every contraction mapping of a complete metric space into itself has a unique fixed point.

Proof.
2.4.5 Theorem. [1]Schauder Theorem
Every continuous operator that maps a closed convex subset of a Banach space into a compact subset of itself
has at least one fixed point.

Proof.
2.4.6 Theorem. [1]Schaefer Theorem
Let E be a normed linear space, let F : E → E be a completely continuous operator, and let

χ(F ) = {x ∈ E : x = λF (x), for some 0 < λ < 1}.

Then, either χ(F ) is unbounded or F has a fixed point.

Proof.
3. Approximate controllability of the semilinear
parabolic evolution equations with nonlocal
conditions
In this chapter we are going to deal we the approximate controllability of the semilinear parabolic evolution
equation with nonlocal condition. Two cases will be solved, first of us the case that the nonlocal initial condition
does not depends on the solution and conversely. We recall that the initial condition of a Cauchy problem is say
to be local when the corresponding solution is not defined on the entire time interval but on and open part of
that and nonlocal otherwise.

3.1 Hypotheses
In all what follow,

1. X and U are Hilbert spaces representing the states space and the controls
  space respectively, A is the
infinitesimal generator
 of a
 strongly continuous
  semigroup S(t) t ∈ 0, T , T > 0, g : C(0, T ; X) → X,
1
f : X → X, x ∈ C( 0, T , D(A)) ∩ C ( 0, T , X) is the state function, B : U → X is a linear bounded
operator, u ∈ L2 (0, T ; U ) is the control function;
 
2. S(t) t ∈ 0, T , T > 0 is a compact semigroup and B ∈ L(U, X);

3. f ∈ C 1 (X, X) and his Frechet derivative fulfilled

0
||f (x)||L(X) ≤ L ∀x ∈ X where L > 0;

3.2 The nonlocal initial condition does not depends on the solution
We consider the following semilinear parabolic equation
(
x0 (t) = Ax(t) + Bu(t) + f (x(t))
 
t ∈ 0, T , T > 0,
A= (3.2.1)
x(0) = x0 ,

3.2.1 Proposition.
If x0 ∈ D(A) then under the contions 1, 2 and 3 every solution of the equation(3.2.1) reads as,
Z t  
x(t) = x(t, x0 , u) = S(t)x0 + S(t − s) Bu(s) + f (x(s)) ds. (3.2.2)
0

Morever the solution is unique.

Proof.

1. Suppose that x(.) = x(.,x0 , u) ∈ C 1 (0, T ; X) ∩ C(0, T ; D(A)) is a solution of equation(3.2.1).

 
Let t ∈ 0, T , r ∈ 0, t , h ∈ 0, t − r , we set h(r) = S(t − r)x(r) ∀r ∈ 0, t , since {S(t)}t≥0 is a

7
Section 3.2. The nonlocal initial condition does not depends on the solution Page 8

 
strongly continuous semigroup then h ∈ C( 0, t , X) futhermore

h(r + h) − h(r) S(t − r − h)x(r + h) − S(t − r)x(r)

=
h h
S(t − r − h)x(r + h) − S(t − r)x(r) − S(t − r − h)x(r) + S(t − r − h)x(r)
=
h
S(t − r − h)x(r + h) − S(t − r − h)S(h)x(r) − S(t − r − h)x(r) + S(t − r − h)x(r)
=
 h 
x(r + h) − x(r) T (h) − I
=S(t − r − h) − x(r) .
h h

Hence
 
h(r + h) − h(r) x(r + h) − x(r) T (h) − I
lim = lim+ S(t − r − h) − x(r)
h→0+ h h→0 h h
 
0
=S(t − r) x (r) − Ax(r)
 
=S(t − r) Bu(r) + f (x(r)) .

 

0 


Thus h ∈ C 1 ( 0, t , X) and h (r) = S(t − r) Bu(r) + f (x(r)) ∀r ∈ 0, t . Let ∈ 0, t we have

Z s Z s  
0
h (r)dr = S(t − r) Bu(r) + f (x(r)) dr
0 0
Z s  
h(s) − h(0) = S(t − r) Bu(r) + f (x(r)) dr
0
Z s  
S(t − s)x(s) − S(t − 0)x(0) = S(t − r) Bu(r) + f (x(r)) dr
0
Z s  
S(t − s)x(s) =S(t)x0 + S(t − r) Bu(r) + f (x(r)) dr
0
Z s  
lim S(t − s)x(s) =S(t)x0 + lim S(t − r) Bu(r) + f (x(r)) dr
s→t s→t 0
Z t  
x(t) =S(t)x0 + S(t − r) Bu(r) + f (x(r)) dr.
0

2. Uniqueness

Since the function F : R+ × X −→ X

(t, x) 7−→ F (t, x(t)) = Ax(t) + Bu(t) + f (x(t)) is Lipschit with respect to x(t)

then the equation(3.2.1) has a unique solution given by

Z t  
x(t) = x(t, x0 , u) = S(t)x0 + S(t − s) Bu(s) + f (x(s)) ds.
0
Section 3.2. The nonlocal initial condition does not depends on the solution Page 9

In fact using the hypothesis 3 and the fact that A is bounded we have
   
   
F (t, x(t)) − F (t, y(t)) = Ax(t) + Bu(t) + f (x(t)) − Ay(t) − Bu(t) − f (y(t))
   
 
 
= A(x(t) − y(t)) + f (x(t)) − f (y(t))
    
≤ Ax(t) − y(t) + f (x(t)) − f (y(t))
    0   
≤ Ax(t) − y(t) + sup f (z) x(t) − y(t)
z∈X
    0   
≤ Ax(t) − y(t) + sup f (z) x(t) − y(t)
z∈X
    
≤ Ax(t) − y(t) + Lx(t) − y(t)
 
 
≤ A + L x(t) − y(t) then F is Lipschit with respect to x(t).
Now let define the operator,
P : C(0, T ; X) −→ C(0, T ; X)
Z t  
x 7−→ P [x] such that P [x](t) = S(t)x0 + S(t − s) Bu(s) + f (x(s)) ds
0

Then is it clear that x ∈ C(0, T ; X) is a fix point of P if and only if x is a solution of the equation(3.2.1),
thus to conclude the uniqueness of the solution of the equation(3.2.1) it is sufficients to show that the
operator P has a unique fix point. In order to do this we state the following theorem.
3.2.2 Theorem. [6]
Let X be a complete metric space g : X −→ X be a map such that there exists k ∈ N∗ , c ∈ 0, 1 and

d(g k (x), g k (y)) ≤ cd(x, y) ∀x, y ∈ X. Then g has a unique fix point in X.

We will prove this theorem in the appendix, let us apply it here for that, let x, y ∈ C(0, T ; X) we assume
that C(0, T ; X) is equipped with the supremun norm. We have
   Z t Z t 
       
P [x](t) − P [y](t) = S(t)x0 + S(t − s) Bu(s) + f (x(s)) ds − S(t)x0 − S(t − s) Bu(s) + f (y(s)) ds
  
0 0
 Z t Z t 
     
= 
  S(t − s) Bu(s) + f (x(s)) ds − S(t − s) Bu(s) + f (y(s)) ds
0 0
 Z t 
   
=  S(t − s) f (x(s)) − f (y(s)) ds
0
Z t
 0       
≤M sup f (z) x(s) − y(s)ds where S(s) ≤ M ∀s ∈ 0, T
0 z∈X
Z t
x(s) − y(s)ds where sup f 0 (z) ≤ L by hypothesis
   
≤ ML
0 z∈X
Z t  
≤ ML   x(s) − y(s) ds
sup  
0 s∈ 0,T
   
≤ M Lx − y t ∀t ∈ 0, T .

Similarly we have
  Z t
 2
P [x](t) − P 2 [y](t) ≤ M L
  
P [x](s) − P [y](s)ds
 
0
 t (M L)2 t2 
Z
2 
 
≤ (M L) x − y   sds = x − y .
0 2
Section 3.2. The nonlocal initial condition does not depends on the solution Page 10

By induction, we deduce that

n n 
 
P [x](t) − P n [y](t) ≤ (M L) t x − y  ∀t ∈ 0, T .
 n    
  n!

Then
n n 
 
P [x](t) − P n [y](t) ≤ (M L) T x − y  ∀n ∈ N∗ ∀t ∈ 0, T .
 n    
  n!
Hence
n 
 
P [x] − P n [y] ≤ (T M L) x − y  ∀n ∈ N∗ .
 n  
  n!
Since
(T M L)n (T M L)n
lim = lim  n =0
n→+∞ n! n→+∞ n √
2πn
e
1 (T M L)n 1
thus for  = there exists k ∈ N∗ such that n > k ⇒ < this allows us to say that
8 n! 8
k 
 
P [x] − P k [y] ≤ (T M L) x − y  ≤ 1 x − y .
 k    
  k! 8

This therefore permits us to say that P admits a unique fix point in C(0, T ; X).

R1 0
3.2.3 Proposition. Setting F as F (z) = 0
f (rz)dr ∀z ∈ X we have the followings

1. F : X −→ L(X);
2. f (z) = F (z)z + f (0) ∀z ∈ X;
 
3. ||F (z(t))|| ≤ L ∀z ∈ C(0, T ; X) ∀t ∈ 0, T ;
4. F ∈ C(X, L(X)).

Proof. 1. First point

Let z, x, y ∈ X and t ∈ R we have
Z 1
0
F (z)(x + ty) = f (rz)(x + ty)dr
0
Z 1 Z 1
0 0 0
= f (rz)(x)dr + t f (rz)(y)dr since f (rz) is linear
0 0
= F (z)(x) + tF (z)(y).

Moreover
Z 1
0
||F (z)|| = || f (rz)dr||
0
Z 1 Z 1 Z 1
0 0
≤ ||f (rz)||dr ≤ ||f (rz)||dr ≤ L dr = L.
0 0 0

Then F (z) ∈ L(X) ∀z ∈ X.

Section 3.2. The nonlocal initial condition does not depends on the solution Page 11

2. Second point
Z 1
0
F (z)z = f (rz)zdr
0
Z 1 Z 1
0 0 0 0
= f (rz) ◦ g (t)dr = f (g(r)) ◦ g (r)dr where g(r) = rz
0 0
Z 1
0
= (f ◦ g) (r)dr = (f ◦ g)(1) − (f ◦ g)(0) = f (g(1)) − f (g(0)) = f (z) − f (0)
0

then F (z)z = f (z) − f (0) ∀z ∈ X.

3. Third point  
By the first point F (z) ∈ L(X) ∀z ∈ X. Then ||F (z(t))|| ≤ L ∀z ∈ C(0, T ; X) ∀t ∈ 0, T .
4. Fourd point
Let {yn }n≥0 ⊂ X and y ∈ X be such that yn −→ y in X as n −→ +∞. Let us show that F (yn ) −→
0
F (y) in L(X) as n −→ +∞. Remark that f ∈ C 1 (X, X), from this f : X −→ L(X) is continuous hence
   
0 0
yn −→ y in X =⇒ f (yn ) −→ f (y) in L(X) .

Futhermore
   Z 1 Z 1 
   0 0 
F (yn ) − F (y) =  f (ryn )dr − f (ry)dr
 
L(X) 0 0 L(X)
Z 1
0 0
≤ ||f (ryn ) − f (ry)||L(X) dr
0

thus by the convergent dominated theorem, we have

  Z 1 Z 1
  0 0 0 0
lim F (yn ) − F (y) ≤ lim ||f (ryn ) − f (ry)||L(X) dr = lim ||f (ryn ) − f (ry)||L(X) dr = 0
n−→+∞ n−→+∞ 0 0 n−→+∞
L(X)

that is F ∈ C(X, L(X)).

3.2.4 Proposition.    
The unique solution x ∈ C( 0, T , D(A)) ∩ C 1 ( 0, T , X) of equation (3.2.1) can rewritten as
Z t  
x(t) = x(t, x0 , u) = S(t)x0 + S(t − s) Bu(s) + F (x(s))x(s) + f (0) ds. (3.2.3)
0

Proof.
We have
Z t  
x(t) = S(t)x0 + S(t − s) Bu(s) + f (x(s) ds
0
Z t  
= S(t)x0 + S(t − s) Bu(s) + F (x(s))x(s) + f (0) ds since f (z) = F (z)z + f (0) ∀z ∈ X.
0
Section 3.2. The nonlocal initial condition does not depends on the solution Page 12

3.2.5 Proposition.
Defining the operator
   
P : C( 0, T , X) −→ C( 0, T , X)
Z t  
z 7−→ P [z] such that P [z](t) = S(t)x0 + S(t − s) Bu(s) + F (z(s))P [z](s) + f (0) ds
0

P can be rewritten as
Z t  
P [z](t) = Q(t, 0; F (z))x0 + Q(t, s; F (z)) Bu(s) + f (0) ds (3.2.4)
0

where
Z t
Q(t, s; F (z))x = S(t − s)x + S(t − r)F (z(r))Q(r, s; F (z))xdr with 0 ≤ s ≤ t ≤ T ∀x ∈ X.
s

Proof.
Z t
 
R[z](t) = Q(t, 0; F (z))x0 + Q(t, s; F (z)) Bu(s) + f (0) ds
0
Z t
= S(t)x0 + S(t − r)F (z(r))Q(r, 0; F (z))x0 dr
0
Z t Z tZ t
   
+ S(t − s) Bu(s) + f (0) ds + S(t − r)F (z(r))Q(r, s; F (z)) Bu(s) + f (0) dsdr
0 0 s
Z t
= S(t)x0 + S(t − r)F (z(r))Q(r, 0; F (z))x0 dr
0
Z t Z tZ r
   
+ S(t − s) Bu(s) + f (0) ds + S(t − r)F (z(r))Q(r, s; F (z)) Bu(s) + f (0) drds
0 0 0
Z t
 
= S(t)x0 + S(t − s) Bu(s) + f (0) ds
0
Z t  Z r 
 
+ S(t − r)F (z(r)) Q(r, 0; F (z))x0 + Q(r, s; F (z)) Bu(s) + f (0) ds dr
0 0
Z t Z t 
 
= S(t)x0 + S(t − s) Bu(s) + f (0) ds + S(t − r)F (z(r))P [z](r) dr
0 0
Z t  
= S(t)x0 + S(t − s) Bu(s) + F (z(s))P [z](s) + f (0) ds
0
= P [z](t).

3.2.6 Lemma. [3],[2],[11]

For each α >, h ∈ X and z ∈ C(0, T ; X) the linear quadratic problem
 t 
R 
S(t − s) Bu(s) + F (z(s))P [z](s) + f (0) ds



 P [z](t) = S(t)x0 +
 0 R
T 
(S) = 1  2 α
u(s)2 ds → min
 (3.2.5)
 Jα (u) = P [z](T ) − h +


 2 2 0
u ∈ L2 (0, T ; U ),


has a unique solution of the form

1
uα (t) = − B ∗ Q(T, t; F (z))∗ Pα [z](T ) − h .
 
(3.2.6)
α
Section 3.2. The nonlocal initial condition does not depends on the solution Page 13

Where
Z t
Q(t, s; F (z))∗ x = S ∗ (t − s)x + S ∗ (t − r)F ∗ (z(r))Q(r, s; F (z))∗ xdr with 0 ≤ s ≤ t ≤ T ∀x ∈ X.
s
Also we have

−1
Z T
uα (t) = B ∗ Q(T, t; F (z))∗ αI + ΓT0 (F (z)) h − Q(T, 0; F (z))x0 − Q(t, s; F (z))f (0)ds. (3.2.7)
0
Where Z T
ΓT0 (F (z))x = Q(T, r; F (z))BB ∗ Q(T, r; F (z))∗ xdr ∀x ∈ X.
0
And the following is fulfilled

−1
Z T
Pα [z](T ) − h = −α αI + ΓT0 (F (z)) h − Q(T, 0; F (z))x0 − Q(t, s; F (z))f (0)ds. (3.2.8)
0
Moreover, there exists a constant C > 0 such that for all α > 0 we have
Z T
Pα [z](T ) − h2 + α uα (s)2 ds ≤ C.
   
(3.2.9)
0

Before starting the proof of lemma 3.2.6, let us make some comments on it.On the system (3.2.5) we are looking for R
t 
the unique control uα which will give us the solution Pα [z] of the equation P [z](t) = S(t)x0 + S(t−s) Bu(s)+
0 R
T 
 1  2 α
u(s)2 ds

F (z(s))P [z](s)+f (0) ds such that the functional Jα defined by Jα (u) = P [z](T )−h +
2 2 0
reachs its minimal value for a fix z ∈ C(0, T ; X) and α > 0. We will give a another view of the proof of the
uniqueness and the existence of solution of system (3.2.5) nevertheless we would like to emphasize that for a
beautiful and more convincing proof see [3] and [2].

Proof. Let us compute the variational derivative of the functional Jα , for that let v ∈ L2 (0, T ; U ) and r ∈ R.
We set
Z t
 
Pr [z](t) = S(t)x0 + S(t − s) Bu(s) + rBv(s) + F (z(s))P [z](s) + f (0) ds
0
Z t
 
= Q(t, 0; F (z))x0 + Q(t, s; F (z)) Bu(s) + Bv(s) + f (0) ds,
0
 
then, it is clear that P0 [z](t) = P [z](t) ∀t ∈ 0, T .
 
2 α T 
 Z
d  d 1  2
Jα (u + rv) = Pr [z](T ) − h + u(s) + rv(s) ds 

dr  dr 2 2 0 
r=0 r=0
"  Z T  #
d  d 
= Pr [z](T ) − h, Pr [z](T ) − h + α u(s) + rv(s), u(s) + rv(s) ds 

dr 0 dr 
r=0
" Z T  Z T  #
= Pr [z](T ) − h, Q(T, s; F (z))Bv(s)ds + α u(s) + rv(s), v(s) ds 

0 0 
r=0
 Z T  Z T 
= P [z](T ) − h, Q(T, s; F (z))Bv(s)ds + α u(s), v(s) ds
0 0
Z T   Z T  
= B ∗ Q(T, s; F (z))∗ (P [z](T ) − h), v(s) ds + +α u(s), v(s) ds
0 0
Z T  
∗ ∗
= B Q(T, s; F (z)) (P [z](T ) − h) + αu(s), v(s) ds.
0
Section 3.2. The nonlocal initial condition does not depends on the solution Page 14

Hence
"  # "Z #
T  
d 
Jα (u + rv) = 0 =⇒ B ∗ Q(T, s; F (z))∗ (P [z](T ) − h) + αu(s), v(s) ds ∀v ∈ L2 (0, T ; U )

dr  0
r=0
" #
∗ ∗
 
=⇒ B Q(T, s; F (z)) (P [z](T ) − h) + αu(s) = 0 almost everywhere in 0, T
" #
1 ∗
=⇒ uα (s) = − B Q(T, s; F (z))∗ (Pα [z](T ) − h) for almost all s ∈ 0, T .
 
α
We obtain a unique control uα , we will show in appendix that it is truly the minimun of the function Jα . Now
let us replace the value of uα in the expression of Pα [z](T ) that is let us compute Pα [z](T ) − h.
Z T
 
Pα [z](T ) − h = Q(T, 0; F (z))x0 + Q(T, s; F (z)) Buα (s) + f (0) ds − h
0
Z T  1
Q(T, s; F (z)) − BB ∗ Q(T, s; F (z))∗ (Pα [z](T ) − h) + f (0) ds − h

= Q(T, 0; F (z))x0 +
0 α
1 T
Z Z T
= Q(T, 0; F (z))x0 − Q(T, s; F (z))BB ∗ Q(T, s; F (z))∗ (Pα [z](T ) − h)ds + Q(T, s; F (z))f (0)ds − h
α 0 0
Z T
1
= Q(T, 0; F (z))x0 − ΓT0 (z)(Pα [z](T ) − h) + Q(T, s; F (z))f (0)ds − h.
α 0

Thus
Z T
1 T

I+ Γ0 (z) (Pα [z](T ) − h) = Q(T, 0; F (z))x0 + Q(T, s; F (z))f (0)ds − h
α 0
 Z T 
1
−1
Pα [z](T ) − h = I + ΓT0 (z) Q(T, 0; F (z))x0 + Q(T, s; F (z))f (0)ds − h
α 0
 Z T 
T

−1
= −α αI + Γ0 (z) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

this prove equation (3.2.8) and we obtain also equation (3.2.7) as follow
1
uα (s) = − B ∗ Q(T, s; F (z))∗ (Pα [z](T ) − h)
α
 Z T 
1
−1
= B ∗ Q(T, s; F (z))∗ αI + ΓT0 (z) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds .
α 0

After all, since Jα (uα ) is the minimal value of Jα there exists a constant C > 0 such that
Jα (uα ) ≤ C
that is Z T

1  2 α
uα (s)2 ds ≤ C
 
Pα [z](T ) − h +
2 2
0

3.2.7 Lemma. [10]

Let Fn , F ∈ L+∞ (0, T ; L(X)), and xn , x ∈ X, be such that
∗

+∞
Fn * F in L (0, T ; L(X)),

xn * x in X,

as n −→ +∞.

Section 3.2. The nonlocal initial condition does not depends on the solution Page 15

Then there exists a subsequence of {Q(T, .; Fn )∗ xn }n≥0 and {Q(T, .; Fn )xn }n≥0 denoted again by {Q(T, .; Fn )∗ xn }n≥0
and {Q(T, .; Fn )xn }n≥0 respectively such that,

∗ ∗
Q(T, .; Fn ) xn −→ Q(T, .; F ) x in C(0, T ; X)

Q(T, .; Fn )xn −→ Q(T, .; F )x in C(0, T ; X)

as n −→ +∞.


Proof. Since xn * x in X then the sequence {xn }n≥0 is uniformly bounded in X. That is

sup{||xn ||} < +∞.

n≥0

In fact,
   
xn * x in X ⇒ f (xn ) → f (x) in R ∀f ∈ L(X, R) = X ∗
 
⇒ sup{|f (xn )|} < +∞ ∀f ∈ L(X, R) = X ∗
n≥0
 
⇒ sup{|xn (f )|} < +∞ ∀f ∈ L(X, R) = X ∗
n≥0
 
⇒ sup{||xn ||} < +∞ Uniform boundedness principle, see lemma 2.3.6
n≥0

In the same sense the weak start convergence of the sequence {Fn }n≥0 to F in L+∞ (0, T ; L(X)) implies that
the sequence {Fn }n≥0 is uniformly bounded in L+∞ (0, T ; L(X)). That is

sup{||Fn ||L+∞ (0,T ;L(X)) } < +∞.

n≥0

The proof of this is too longs, thus we will keep it here and do it in appendix, in the other hand one can see
[5],[13] and [9] for the proof. We notice that, since
 
∀n ≥ 0 ||Fn ||L+∞ (0,T ;L(X)) = {α : ||Fn (t)||L(X) ≤ α a.e. t ∈ 0, T }

then

n≥0
 {||Fn (t)||L(X) } ≤ ||Fn ||L+∞ (0,T ;L(X)) .
sup sup (3.2.10)
t∈ 0,T

This inequation will be helpful in what follows. Now we set φn = Q(T, .; Fn )xn and φ = Q(T, .; F )x then since
Z t
Q(t, s; F )x = S(t − s)x + S(t − r)F (r)Q(r, s; F )xdr
s

then Z t
Q(t, s; Fn )xn = S(t − s)xn + S(t − r)Fn (r)Q(r, s; Fn )xn dr
s
thus replacing t by T we have
Z T
Q(T, s; Fn )xn = S(T − s)xn + S(T − r)Fn (r)Q(T, r; Fn )xn dr This is where the problem is
s

that is Z T
φn (s) = S(T − s)xn + S(T − r)Fn (r)φn (r)dr. (3.2.11)
s
In the same sence we have Z T
φ(s) = S(T − s)x + S(T − r)F (r)φ(r)dr. (3.2.12)
s
Section 3.2. The nonlocal initial condition does not depends on the solution Page 16

Let us show that sequence {ψn }n≥0 is uniformly bounded in C(0, T ; X) equipped with the norm of the uniform
convergence. To do this end, we have
     Z T 
     
φn (s) ≤ S(T − s)xn  +  S(T − r)Fn (r)φn (s)dr
    
X X s X
     
Z T      
φn (s)
X
≤ S(T − s)L(X) xn X + S(T − r)
L(X)
Fn (r)
L(X)
φn (s) dr
X
s
 
Z T  
φn (s)
X
≤ AB + AC φn (s)X dr
s

where      
||S(t)||L(X) ≤ A, xn X ≤ B and Fn (t)L(X) ≤ C ∀n ≥ 0 ∀t ∈ 0, T .
We obtain an integral inequality of Gronwall type then by the Gronwall theorem 2.3.8 we have
Z T RT
A2 BCe ACda
dr ≤ AB + A2 BCeT AC T = D.
 
φn (s)
X
≤ AB + r

We then deduce that  

 { φn (s) X } ≤ D ∀n ≥ 0
sup  
s∈ 0,T

hence  
  φn  
C(0,T ;X)
≤ D ∀n ≥ 0.
we conclude that  
sup φn C(0,T ;X) ≤ +∞.
n≥0
 
Secondly, let us show that ∀t ∈ 0, T the sequence {φn (t)}n≥0 is relatively compact.

• For t = T then {φn (t)}n≥0 = {xn }n≥0 and since xn * x in X then as we have shown beforehand the
sequence {xn }n≥0 is bounded and therefore it is relatively compact.


• For t ∈ 0, T we have
Z T
lim S(T − r)Fn (r)φn (r)dr = 0
x−→0+ T −x

then ∀ > 0 there exists δ such that 0 < δ ≤ T and

 Z T 
 ≤  .
 

 S(T − r)F n (r)φ n (r)dr  2 (3.2.13)
T −δ

Moreover setting
Z T −δ
yn = S(T − t − δ)xn + S(T − r − δ)Fn (r)φn (r)dr
t

since the sequences {φn }n≥0 and {xn }n≥0 are bounded and the semigroup {S(t)}   is uniformly
t∈ 0,T
bounded then the sequence {yn }n≥0 is bounded. And since S(δ) is a compact linear operator then
{S(δ)yn }n≥0 is relatively compact, and since X is a Banach space then there exists a finite sequence
{zn }kn=1 such that
 Z T −δ  k
[ 
{S(δ)yn }n≥0 = S(T − t)xn + S(T − r)Fn (r)φn (r)dr ⊂ BX (zn , ). (3.2.14)
t n≥0 n=1
2
Section 3.2. The nonlocal initial condition does not depends on the solution Page 17

Finally, we have
 Z T 
   
φn (t) − zn  = S(T − t)xn + S(T − r)Fn (r)φ n (r)dr − zn 


t
 Z T −δ Z T 
 
= S(T − t)xn +
  S(T − r)Fn (r)φn (r)dr − zn + S(T − r)Fn (r)φn (r)dr
t T −δ


Z T −δ   Z T
 


≤ S(T − t)xn + S(T − r)Fn (r)φn (r)dr − zn  +  S(T − r)Fn (r)φn (r)dr
t T −δ
 Z T 
     
≤ S(δ)yn − zn  +  S(T − r)Fn (r)φn (r)dr ≤ + = .
T −δ 2 2

Hence
k
[
{φn (t)}n≥0 BX (zn , )
n=1
 
that is the sequence {φn (t)}n≥0 is relatively compact in X ∀t ∈ 0, T .

Let us provethat the sequence {φn }n≥0 is equicontinuous compact in C(0, T ; X) to do this end, let t1 and t2 be
elements of 0, T such that 0 ≤ t1 ≤ t2 ≤ T. We have
 Z T Z T 
   
φn (t1 ) − φn (t2 ) = S(T − t1 )xn + S(T − r)Fn (r)φn (r)dr − S(T − t2 )xn − S(T − r)Fn (r)φn (r)dr

t1 t
 Z t2  2
 
= S(T − t1 )xn − S(T − t2 )xn + S(T − r)Fn (r)φn (r)dr
t1
≤ AB + ACD|t1 − t2 |
 
this shows the equicontinuity of the sequence {φn }n≥0 . Since {φn }n≥0 ⊂C(0, T ; X) with 0, T compact subset
of R, the sequence {φn (t)}n≥0 is relatively conpact in X for any t ∈ 0, T , {φn }n≥0 is uniformly bounded
in C(0, T ; X) and equicontinuous, then by the Arzela-Ascoli theorem 2.4.3, {φn }n≥0 . is relatively compact in
C(0, T ; X). It therefore exists a subsequence of {φn }n≥0 denoted again by {φn }n≥0 which is convergent in
C(0, T ; X). h a limit point of the {φn }n≥0 in C(0, T ; X) then since
Z T
φn (s) = S(T − s)xn + S(T − r)Fn (r)φn (r)dr
s

we have through the convergent dominated theorem

Z T
h(s) = S(T − s)x + S(T − r)Fn (r)h(r)dr
s

and since we were having

Z T
φ(s) = S(T − s)x + S(T − r)Fn (r)φ(r)dr
s

hence h and φ are two solutions of the equation(3.2.12) then by the uniqueness of the solution of equation(3.2.12)
we have h = φ. By replacing in the above proof φn = Q(T, .; Fn )xn by φn = Q(T, .; Fn )∗ xn and φ = Q(T, .; F )x
by φ = Q(T, .; F )∗ x we complete the proof.
3.2.8 Lemma. [11]
Let Fn , F ∈ L+∞ (0, T ; L(X)), and xn , x ∈ X, be such that
∗

+∞
Fn * F in L (0, T ; L(X)),

xn * x in X,

as n −→ +∞.

Section 3.2. The nonlocal initial condition does not depends on the solution Page 18

Then there exists a subsequence of {xn }n≥0 denoted egain by {xn }n≥0 such that

ΓT0 (Fn )xn −→ ΓT0 (F )x in X as −→ +∞

where Z T
ΓT0 (Fn )xn = Q(T, r; Fn )BB ∗ Q(T, r; Fn )∗ xn dr
0
and Z T
ΓT0 (F )x = Q(T, r; F )BB ∗ Q(T, r; F )∗ xdr.
0

Proof.
   Z T Z T 
∗ ∗ ∗ ∗
 T
Γ0 (Fn )xn − ΓT0 (F )x = 
  
   Q(T, r; Fn )BB Q(T, r; Fn ) xn dr − Q(T, r; F )BB Q(T, r; F ) xdr
0 0
 Z T Z T 
Q(T, r; Fn )BB ∗ Q(T, r; Fn )∗ xn dr − Q(T, r; Fn )BB ∗ Q(T, r; F )∗ xdr
 
= 
0 0
 Z T Z T 
∗ ∗ ∗ ∗
 
+ 
  Q(T, r; Fn )BB Q(T, r; F ) xdr − Q(T, r; F )BB Q(T, r; F ) xdr
0 0
 Z T   
∗ ∗ ∗
 
= 
  Q(T, r; Fn )BB Q(T, r; Fn ) xn − Q(T, r; F ) x dr
0
 Z T   
Q(T, r; Fn ) − Q(T, r; F ) BB ∗ Q(T, r; F )∗ xdr
 
+ 
0
Z T   
∗  ∗ ∗
  
≤ Q(T, r; Fn )BB  Q(T, r; Fn ) xn − Q(T, r; F ) x dr
  
0
Z T   
∗ ∗ 
 
+  Q(T, r; Fn ) − Q(T, r; F ) BB Q(T, r; F ) xdr
 
0
Z T  
≤ D||B||2  Q(T, r; Fn )∗ xn − Q(T, r; F )∗ x dr
 
 
0
Z T   
∗ ∗ 
 
+  Q(T, r; Fn ) − Q(T, r; F ) BB Q(T, r; F ) xdr
 
0

Since
∗

+∞
Fn * F in L (0, T ; L(X)),

xn * x in X,

as n −→ +∞.


the by lemma 3.2.7 we have


∗ ∗
Q(T, .; Fn ) xn −→ Q(T, .; F ) x in C(0, T ; X)

Q(T, .; Fn )xn −→ Q(T, .; F )x in C(0, T ; X)

as n −→ +∞.


in terms of subsequence. Thus  

 T T

lim   Γ0 (Fn )xn − Γ0 (F )x = 0
n−→+∞ 

which complete the proof.

Section 3.2. The nonlocal initial condition does not depends on the solution Page 19

3.2.9 Lemma. [11]l

Let Fn , F ∈ L+∞ (0, T ; L(X)) be such that
∗
Fn * F in L+∞ (0, T ; L(X)).

−1
−1
Then there exists a subsequence of { αI + ΓT0 (Fn ) }n≥0 denoted again by { αI + ΓT0 (Fn ) }n≥0 such that

−1
−1
αI + ΓT0 (Fn ) −→ αI + ΓT0 (F ) weakly in L(X) as n −→ +∞.

Proof. We start by recalling that saying that

−1
−1
αI + ΓT0 (Fn ) −→ αI + ΓT0 (F ) weakly in L(X) as n −→ +∞.
means

−1
−1
αI + ΓT0 (Fn ) x * αI + ΓT0 (F ) x in X as n −→ +∞ ∀x ∈ X (see definition 2.2.1).
That is
   

−1
−1
f αI + ΓT0 (Fn ) x −→ f αI + ΓT0 (F ) x in R as n −→ +∞ ∀f ∈ L(X, R) = X ∗ ∀x ∈ X.

That is    

−1
−1
αI + ΓT0 (Fn ) x, y −→ αI + ΓT0 (F ) x, y in R as n −→ +∞ ∀x, y ∈ X

because of the Riesz representation theorem on the Hilbert space X. Let

Rn = αI + ΓT0 (Fn ), R = αI + ΓT0 (F )
∗
and x, y ∈ X. Since Fn * F in L+∞ (0, T ; L(X)), then by lemma 3.2.8
Rn R−1 x * RR−1 x = x hence hRn R−1 x, yi −→ hx, yi
in such a way that
hRn (Rn−1 x − R−1 x), yi = hx − Rn R−1 x, yi = hx, yi − hRn R−1 x, yi −→ hx, yi − hx, yi = 0.
Moreover, since Rn is uniformly bounded, so is Rn−1 . In fact,
+∞  n +∞  n
Rn  =  αI + ΓT0 (Fn ) −1  = 1  I + 1 ΓT0 (Fn ) −1  = 1  X 1 T 1 X  1 T
 −1  
(−1)n

 
  
Γ0 (Fn )  ≤ Γ0 (Fn ) 
α α α n=0 α α n=0 α
+∞   n +∞
1 X  1 T  1 X  1 T n 1 1 1 1
≤ Γ0 (Fn )  ≤ Γ0 (Fn ) =  1  ≤ α 1 − G
α n=0 α α n=0 α α
1 −  ΓT0 (Fn )
α
 1
where  ΓT0 (Fn ) ≤ G < 1 ∀n ≥ 0.

α

By using the that Rn−1 is uniformly bounded, we have

hRn (Rn−1 x − R−1 x), yi = hRn (Rn−1 x − R−1 x), yi + hR(Rn−1 x − R−1 x), yi − hR(Rn−1 x − R−1 x), yi
= hRn (Rn−1 x − R−1 x), yi − hR(Rn−1 x − R−1 x), yi + hR(Rn−1 x − R−1 x), yi
= hRn (Rn−1 x − R−1 x) − R(Rn−1 x − R−1 x), yi + hR(Rn−1 x − R−1 x), yi
= h(Rn − R)(Rn−1 x − R−1 x), yi + hR(Rn−1 x − R−1 x), yi
= hRn−1 x − R−1 x, (Rn − R)∗ yi + hR(Rn−1 x − R−1 x), yi
= hRn−1 x − R−1 x, (Rn∗ − R∗ )yi + hR(Rn−1 x − R−1 x), yi
= hRn−1 x − R−1 x, (Rn − R)yi + hR(Rn−1 x − R−1 x), yi since Rn∗ = Rn and R∗ = R.
Section 3.2. The nonlocal initial condition does not depends on the solution Page 20

Hence, since Rn y −→ Ry, hR(Rn−1 x − R−1 x), yi −→ 0 as n −→ +∞ and {Rn−1 x}n≥0 is uniformly bounded
then
hRn (Rn−1 x − R−1 x), yi −→ 0 as n −→ +∞.
Futhermore, since {Rn−1 x}n≥0 is uniformly bounded and X is an Hilbert space therefore reflexive, it admits a
subsequence denoted {Rn−1 x}n≥0 which weakly convergent in X. We notice that we will emphasize on the proof
of this in appendix, nevertheless one can see [14] for the proof. Let λ ∈ X such that Rn−1 x * λ as n −→ +∞,
since A is a bounded operator RRn−1 x * Rλ as n −→ +∞ in such a way that

R(Rn−1 x − R−1 x) * R(λ − R−1 x) = Rλ − x as n −→ +∞.

That is hR(Rn−1 x − R−1 x), yi −→ hRλ − x, yi. But we have shown up that hR(Rn−1 x − R−1 x), yi −→ 0. Thus
by the uniqueness of the limit we have we have

hRλ − x, yi = 0 ∀y ∈ X
Rλ − x = 0
λ = R−1 x

Finally, we conclude that

Rn−1 x * λ = R−1 x as n −→ +∞.

3.2.10 Lemma. [11]

Let z ∈ C(0, T ; X) and let Q(t, s; F (z)) be the evolution operator generated by A + F (z), where F is defined at
proposition 3.2.3 and let
 Z T 

−1
uα (t, z) = B ∗ Q(T, t; F (z))∗ αI + ΓT0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds .
0

Then, there exists L1 > 0 such that for all z ∈ C(0, T ; X),
 
Q(t, s; F (z)) ≤ L1 , 0 ≤ s ≤ t ≤ T,

and there exists L2 > 0 such that for any α > 0 and for all z ∈ C(0, T ; X),

uα (t, z) ≤ 1 L2 , 0 ≤ t ≤ T.
 
α
Moreover
uα (t, z) : C(0, T ; X) −→ C(0, T ; X) is continuous.
 
Proof. Let s ∈ 0, T , x ∈ X such ||x|| = 1 we consider
Z t  
Q(t, s; F (z))x = S(t − s)x + S(t − r)F (z(r))Q(r, s; F (z))xdr ∀t ∈ s, T .
s

We have
  t
 Z 
   
Q(t, s; F (z))x ≤ S(t − s)x +  S(t − r)F (z(r))Q(r, s; F (z))xdr 
 
s
Z t
        
≤ S(t − s) x + S(t − r)F (z(r))Q(r, s; F (z))xdr
s
Z t
       
≤ A + AL Q(r, s; F (z))dr where S(t) ≤ A, F (z(t)) ≤ L ∀t ∈ 0, T ,
s
Section 3.2. The nonlocal initial condition does not depends on the solution Page 21

we obtain an integral inequality of Gronwall type, hence by the Gronwall theorem 2.3.8 we have
Z t Rt
Q(t, s; F (z))x ≤ A + A2 L eAL r da dr ≤ A + A2 L eALT T = L1 ∀x ∈ X, x = 1.
   
s

Hence    
Q(t, s; F (z))x ≤ L1 ∀x ∈ X, x = 1.

That is  
Q(t, s; F (z)) ≤ L1 0 ≤ s ≤ t ≤ T.

Moreover we would like to emphasize on this fact, since

Z T
ΓT0 (F (z))x = Q(T, r; F (z))BB ∗ Q(t, s; F (z))∗ xdr,
0

then
Z T
Q(T, r; F (z))BB ∗ Q(t, s; F (z))∗ x ≤ L2 B 2 T.
 T     
Γ0 (F (z))x ≤
0

 2  1 
Thus we choose α > L2 B  T in such a way that  ΓT0 (F (z)) < 1. In fact in all this work as we have stated
α
+
 1 T 
in the beginning α is picking in R \ {0} such that  Γ (F (z)) < 1; such a α always axists in this work since
α 0
F is in L+∞ (0, T ; L(X)) thus bounded. The boudedness of F always permits us to have a such α. We notice
also that in the case that Fn turn to F weakly star in L+∞ (0, T ; L(X)) the sequence {ΓT0 (Fn (z))}n≥0 is clearly
uniformly bounded. Let us take the case of the control uα we have
 Z T 
∗ ∗ T

−1
uα (t, z) = B Q(T, t; F (z)) αI + Γ0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

hence,
  Z T 
uα (t, z) = B ∗ Q(T, t; F (z))∗ αI + ΓT0 (F (z)) −1 h − Q(T, 0; F (z))x0 −
  

 Q(T, s; F (z))f (0)ds 
0
 Z T 
 ∗   ∗  
    T

−1     
≤ B
     Q(T, t; F (z)) αI + Γ0 (F (z))  
h − Q(T, 0; F (z))x0 −
  Q(T, s; F (z))f (0)ds
0
 Z T 
1 1 −1
≤ B  Q(T, t; F (z))  I + ΓT0 (F (z))
         
    
 h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds 
α α  
0

 Z T 
1    1 T
−1           
≤ B L1   I + Γ0 (F (z))     h + Q(T, 0; F (z))
        x0 +
    Q(T, s; F (z))    f (0) ds
 
α α 0
 
1   1
−1 
≤ B  L1 h + L1 x0  + T L1 f (0)  I + ΓT0 (F (z))
      
,
α α

 1 
since F (z) ∈ L+∞ (0, T ; L(X)) then ΓT0 (F (z)) is bounded and α is choosing such that  ΓT0 (F (z)) ≤ G < 1.
α
 1  ≤ 1 . We therefore conclure that

−1 
thus  I + ΓT0 (F (z))
α 1−G
  
uα (t, z) ≤ 1 B  L1 h + L1 x0  + T L1 f (0) 1 1
         
= L2 .
α 1−G α
Section 3.2. The nonlocal initial condition does not depends on the solution Page 22

 
Finally let us prove that the control uα (t) is continuous ∀t ∈ 0, T . To do this end, let {zn }n≥0 ⊂ C(0, T ; X)
and z ∈ C(0, T ; X) be such that zn −→ z as n −→ +∞ in C(0, T ; X). As we have proved in proposition 3.2.3
that F ∈ C(X, L(X)) hence we have
   
 
zn −→ z in C(0, T ; X) =⇒ zn (t) −→ z(t) in X ∀t ∈ 0, T
 
 
=⇒ F (zn (t)) −→ F (z(t)) in L(X) ∀t ∈ 0, T

hence for g ∈ L1 (0, T ; L(X)) and by the convergent dominated theorem we have
 Z T    Z T
     
lim   F (zn (t)) − F (z(t)) g(t)dt
  ≤ lim F (zn (t)) − F (z(t))
L(X)
g(t)
L(X)
dt
n−→+∞ 0 n−→+∞ 0
L(X)
Z T    
= lim F (zn (t)) − F (z(t)) g(t) dt = 0.
n−→+∞ L(X) L(X)
0

We therefore have the weak star

 comvergent
 of the sequence {F (zn )}n≥0 to F (z). This will allow us to use the
above lemma. Moreover ∀t ∈ 0, T the control uα (t) is defined as

uα (t) : C(0, T ; X) −→ C(0, T ; X)

z 7−→ uα (t)(z) = uα (t, z) such that

−1
 Z T 
∗ ∗
uα (t, z) = B Q(T, t; F (z)) αI + ΓT0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

and from lemma 3.2.9 and lemma 3.2.7 we have the following

−1
 Z T 
∗ ∗
uα (t, zn ) − uα (t, z) = B Q(T, t; F (zn )) αI + ΓT0 (F (zn )) h − Q(T, 0; F (zn ))x0 − Q(T, s; F (zn ))f (0)ds
0

−1
 Z T 
∗ ∗
− B Q(T, t; F (z)) αI + ΓT0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

−1
 Z T 
= B ∗ Q(T, t; F (zn ))∗ αI + ΓT0 (F (zn )) h − Q(T, 0; F (zn ))x0 − Q(T, s; F (zn ))f (0)ds
0

−1
 Z T 
∗ ∗
− B Q(T, t; F (zn )) αI + ΓT0 (F (zn )) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

−1
 Z T 
∗ ∗
+ B Q(T, t; F (zn )) αI + ΓT0 (F (zn )) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

−1
 Z T 
− B ∗ Q(T, t; F (zn ))∗ αI + ΓT0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

−1
 Z T 
∗ ∗ T
+ B Q(T, t; F (zn )) αI + Γ0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0

−1
 Z T 
∗ ∗ T
− B Q(T, t; F (z)) αI + Γ0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0
Section 3.2. The nonlocal initial condition does not depends on the solution Page 23

that is

−1
uα (t, zn ) − uα (t, z) = B ∗ Q(T, t; F (zn ))∗ αI + ΓT0 (F (zn ))
 Z T Z T 
× Q(T, 0; F (z))x0 − Q(T, 0; F (zn ))x0 + Q(T, s; F (z))f (0)ds − Q(T, s; F (zn ))f (0)ds
0 0
 

−1
−1
+ B ∗ Q(T, t; F (zn ))∗ αI + ΓT0 (F (zn )) − αI + ΓT0 (F (z))
 Z T 
× h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds
0
 
+ B Q(T, t; F (zn )) − B Q(T, t; F (z))∗
∗ ∗ ∗



−1
 Z T 
× αI + ΓT0 (F (z)) h − Q(T, 0; F (z))x0 − Q(T, s; F (z))f (0)ds .
0

Thus it is clear that by passing the norm symbol throughout the above expressions and by using the above
hypotheses on weak star convergent we deduce that
uα (., zn ) −→ uα (., z) in C(0, T, X) as n −→ +∞.
3.2.11 Lemma. [11] For any α > 0 the operator Pα defined by replacing the control u in the definition of the
operator P of the proposition 3.2.5 by the control uα defined in lemma 3.2.10, that is
Z t
 
Pα [z](t) = S(t)x0 + S(t − s) Buα [z](s) + F (z(s))Pα [z](s) + f (0) ds
0
Z t
 
= Q(t, 0; F (z))x0 + Q(t, s; F (z)) Buα [z](s) + f (0) ds
0

where
Z t
Q(t, s; F (z))x = S(t − s)x + S(t − r)F (z(r))Q(r, s; F (z))xdr with 0 ≤ s ≤ t ≤ T ∀x ∈ X
s

has a fixed point in C(0, T ; X).

Proof. To prove this we will use the Schauder fixed point theorem, that is we will prove that
Pα : C(0, T ; X) −→ C(0, T ; X)
is a continuous operator that maps a closed convex subset of the banach sapce C(0, T ; X) into a compact subset
of itself (see theorem 2.4.5). To do this end, we pick the closed convex subset of C(0, T ; X) itself.

• Let us prove that Pα is a continuous operator. First remark that there exists some contant Mα > 0 such
that  
 
Pα [z] ≤ Mα ∀z ∈ C(0, T ; X).
 
In fact
   Z t 
     
Pα [z](t) = Q(t, 0; F (z))x0 + Q(t, s; F (z)) Bu α [z](s) + f (0) ds
   
0
Z t  
         
≤ Q(t, 0; F (z)) x0 +
         Q(t, s; F (z))     B uα [z](s) + f (0) ds
         
0
 
  1      
≤ L1 x0  + L1 B L2 + f (0) T = Mα ∀t ∈ 0, T using lemma 3.2.10
α
Section 3.2. The nonlocal initial condition does not depends on the solution Page 24

that is  
   
  Pα [z](t) ≤ Mα ∀z ∈ C(0, T ; X).
Pα [z] = sup  
 
C(0,T ;X) t∈ 0,T

This mean that the operator Pα send its domain C(0, T ; X) into a bouded subset Pα (C(0, T ; X)) of itself,
in others words Pα uniformly bounded. Let {zn }n≥0 and z be a sequence of elements of C(0, T ; X) and
element of C(0, T ; X) respectively, such that zn −→ z as n −→ +∞.
   Z t
    
Pα [zn ](t) − Pα [z](t) = S(t)x0 + S(t − s) Buα [zn ](s) + F (zn (s))Pα [zn ](s) + f (0) ds
  
X 0
Z t 
  
− S(t)x0 − S(t − s) Buα [z](s) + F (z(s))Pα [z](s) + f (0) ds
0
Z t
  
≤ S(t − r)B uα [zn ](r) − uα [z](r)dr
0
Z t
   
+ S(t − r)F (zn (r))Pα [zn ](r) − Pα [z](r)dr
0
Z t
   
+ S(t − r)F (zn (r)) − F (z(r))Pα [z](r)dr
0
Z t
      
≤ T A B uα [zn ] − uα [z] C(0,T ;X) + A F (zn ) C(0,T ;L(X)
           Pα [zn ](r) − Pα [z](r)dr
0
     
+ Mα AF (zn ) − F (z)C(0,T ;L(X)) where S(t)L(X) ≤ A ∀t ∈ 0, T

we obtain an integral inequality of Gronwall type, hence by the Gronwall theorem 2.3.8 we have
 
      
Pα [zn ](t) − Pα [z](t) ≤ Mα AF (zn ) − F (z) + T AB uα [zn ] − uα [z]C(0,T ;X)
  C(0,T ;L(X))
X
 
+ AF (zn )C(0,T ;L(X)
 
    
× Mα AF (zn ) − F (z)C(0,T ;L(X)) + T AB uα [zn ] − uα [z]C(0,T ;X)
 
T AF (zn )  
× Te C(0,T ;L(X) ∀t ∈ 0, T ∀n ≥ 0.

That is
 
      
Pα [zn ] − Pα [z] ≤ Mα AF (zn ) − F (z)C(0,T ;L(X)) + T AB uα [zn ] − uα [z]C(0,T ;X)
 
C(0,T ;X)
 
+ AF (zn )C(0,T ;L(X)
 
    
× Mα AF (zn ) − F (z)C(0,T ;L(X)) + T AB uα [zn ] − uα [z]C(0,T ;X)
 
T AF (zn )
× Te C(0,T ;L(X) ∀n ≥ 0.

Finally by proposition 3.2.3, lemma 3.2.10 and the fact that zn −→ z in C(0, T ; X) as n −→ +∞. F (zn )
turns to F (z) in C(0, T ; L(X)), uα [zn ] turns to uα [z] in C(0, T ; X) as n −→ +∞ respectively, which implies

Pα [zn ] −→ Pα [z] as n −→ +∞ in C(0, T ; X)

Pα is therefore continuous.
 
• Let us prove that the set Pα (C(0, T ; X)) = Pα [z] : z ∈ C(0, T ; X) is relatively compact.

– We have shown up there that is uniformly bounded Pα (C(0, T ; X)) in C(0, T ; X).
Section 3.2. The nonlocal initial condition does not depends on the solution Page 25

 
– Let us show that Pα (C(0, T ; X)) is equicontinuous. To do this end, let t, δ ∈ 0, T be such that
δ > t and λ > 0 small enough. We have the following
 
Pα [z](δ) − Pα [z](t) ≤ ||S(δ) − S(t)||||x0 ||
Z t−λ
+ ||S(δ − r) − S(t − r)||||Buα [z](r) + F (z(r))Pα [z](r) + f (0)||dr
0
Z t
+ ||S(δ − r) − S(t − r)||||Buα [z](r) + F (z(r))Pα [z](r) + f (0)||dr
t−λ
Z δ
+ ||S(δ − r)||||Buα [z](r) + F (z(r))Pα [z](r) + f (0)||dr.
t

Remark that

||Buα [z](r) + F (z(r))Pα [z](r) + f (0)|| ≤ ||B||||uα [z](r)|| + ||F (z(r))||||Pα [z](r)|| + ||f (0)||
1
≤ L2 ||B|| + LMα + ||f (0)|| = Dα .
α
From this we obtain
 
Z t−λ
Pα [z](δ) − Pα [z](t) ≤ ||S(δ) − S(t)||||x0 || + Dα ||S(δ − r) − S(t − r)||dr
0
Z t
+ Dα ||S(δ − r) − S(t − r)||dr + Dα A|δ − t| where ||S(t)|| ≤ A.
t−λ

Since S(t) is a compact strongly continuous semigroug then it is uniform continuous (see 2.3.4), hence
when |δ − t| −→ 0, ||S(δ − r) − S(t − r)|| −→ 0 and for  > there exists r > 0 such that
 − ADα r
>0
||x0 || + Dα T + λ
and
 − ADα r
|δ − t| ≤ r =⇒ ||S(δ − r) − S(t − r)|| <
||x0 || + Dα T + λ
this permits to get

|δ − t| ≤ r =⇒ ||Pα [z](δ) − Pα [z](t)|| <  ∀z ∈ C(0, T ; X)

which concludes the equicontinuity of the subset Pα (C(0, T ; X)).

 
– Let us show that {Pα [z](t) : z ∈ C(0, T ; X)} is relatively compact in X for any t ∈ 0, T .
∗ For t = 0 then {Pα [z](t) : z ∈ C(0, T ; X)} = {x0 } which is closed and bounded in the Hilbert
space X therefore it is relatively compact.

∗ For t ∈ 0, T we have
Z t
 
lim + S(t − r) Buα [z](r) + F (z(r))Pα [z](r) + f (0) dr = 0
λ−→0 t−λ

then ∀ > 0 there exists δ such that 0 < δ ≤ T and

 Z t 
 ≤  .
   

 S(t − r) Bu α [z](r) + F (z(r))P α [z](r) + f (0) dr  2
t−δ

Setting
Z t−δ  
y(z)(t) = S(t − δ)x0 + S(t − r − δ) Buα [z](r) + F (z(r))Pα [z](r) + f (0) dr
0
Section 3.3. The nonlocal initial condition depends on the solution Page 26

we have ||y(z)(t)|| ≤ Mα ∀z ∈ C(0, T ; X) hence the subset {y(z)(t)

 : z ∈ C(0, T ; X)} is bounded


in X. Moreover S(δ) is a compact linear operator then S(δ) y(z)(t) : z ∈ C(0, T ; X) is
relatively compact in the Hilbert space X, then there exists a finite sequence {zn }kn=1 such that
    k
 [ 
S(δ) y(z)(t) : z ∈ C(0, T ; X) = S(δ)(y(z)(t)) : z ∈ C(0, T ; X) ⊂ BX (zn , ).
n=1
2

Finally, we have
 Z t 
     
Pα [z](t) − zn  = S(t)x0 + S(t − r) Bu α [z](r) + F (z(r))P α [z](r) + f (0) dr − zn 


0
 Z t−δ
  
= S(t)x0 +
  S(t − r) Buα [z](r) + F (z(r))Pα [z](r) + f (0) dr − zn
0
Z t 
  
+ S(t − r) Buα [z](r) + F (z(r))Pα [z](r) + f (0) dr
t−δ
 Z t−δ 
   
≤ S(t)x0 + S(t − r) Buα [z](r) + F (z(r))Pα [z](r) + f (0) dr − zn 
0
 Z t 
   
+  S(t − r) Buα [z](r) + F (z(r))Pα [z](r) + f (0) dr
t−δ
  t
 Z 
   
≤ S(δ)(y(z)(t)) − zn + 
      S(t − r) Buα [z](r) + F (z(r))Pα [z](r) + f (0) dr
t−δ
 
≤ + = .
2 2
Hence
  k
[
Pα [z](t) : z ∈ C(0, T ; X) ⊂ BX (zn , )
n=1
 
 
that is the sequence Pα [z](t) : z ∈ C(0, T ; X) is relatively compact in X ∀t ∈ 0, T .

3.3 The nonlocal initial condition depends on the solution

4. Conclusion and Perspectives

27
Appendix A. Proof of some theorems

28
Acknowledgements

29
References
[1] K Balachandran and JP Dauer. Controllability of nonlinear systems in banach spaces: a survey. Journal of
Optimization Theory and Applications, 115(1):7–28, 2002.

[2] Agamirza E Bashirov and Nazim I Mahmudov. On concepts of controllability for deterministic and stochastic
systems. SIAM Journal on Control and Optimization, 37(6):1808–1821, 1999.
[3] Dr. Anthony J. Pritchard (eds.) Dr. Ruth F. Curtain. Infinite Dimensional Linear Systems Theory. Lecture
Notes in Control and Information Sciences 8. Springer-Verlag Berlin Heidelberg, 1 edition, 1978.

[4] Sever Silvestru Dragomir. Some Gronwall type inequalities and applications. Nova Science Publishers New
York, 2003.
[5] Lawrence C Evans. Partial differential equations (graduate studies in mathematics, vol. 19). In American
Math. Soc, 1998.
[6] Khalil Ezzinbi. Lecture notes in functional analysis and evolution equations. African University of Science
and Technology, 2009.
[7] Spanos Georgios. Semilinear parabolic evolution equations and an application to the lotka-volterra system
with diffusion.
[8] Patrick Fitzpatrick Halsey Royden. Real Analysis (4th Edition). Prentice Hall, 2010.

[9] HK Karlsen. Notes on weak convergence. Online: http://www. uio.

no/studier/emner/matnat/math/MAT4380/v06/Weakconvergence. pdf, 31, 2006.
[10] Xungjing Li and Jiongmin Yong. Optimal control theory for infinite dimensional systems. Springer Science
& Business Media, 2012.

[11] Nazim I Mahmudov. Approximate controllability of evolution systems with nonlocal conditions. Nonlinear
Analysis: Theory, Methods & Applications, 68(3):536–546, 2008.
[12] John Nachbar. he arzela-ascoli theorem. African University of Science and Technology, 2016.
[13] HL Royden and PM Fitzpatrick. Real analysis (4th edtion). New Jersey: Printice-Hall Inc, 2010.

[14] Bengt Ove Turesson. Functional analysis weak convergence properties of weak and weak star convergence.
Handout, 2015.
[15] wikipedia. Frechet derivative, 2018.

30