The Basic Theory for Partial Functional Differential

Equations and Applications

Khalil Ezzinbi

To cite this version:

Khalil Ezzinbi. The Basic Theory for Partial Functional Differential Equations and Applications. 3rd
cycle. Damas (Syrie), 2004, pp.60. �cel-00376391�

HAL Id: cel-00376391

https://cel.archives-ouvertes.fr/cel-00376391
Submitted on 17 Apr 2009

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 The Basic Theory for Partial Functional
Differential Equations and Applications

CIMPA SCHOOL DAMAS 16-27 May 2004

Khalil EZZINBI∗
Université Cadi Ayyad
Faculté des Sciences Semlalia
Département de Mathématiques,
B.P. 2390, Marrakech, Morocco

May 25, 2004

1 Introduction
Let (X, |·|) be an infinite dimensional Banach space and L(X) be the space
of bounded linear operators from X into X. Suppose that r > 0 is a given
real number. C ([−r, 0] , X) denotes the space of continuous functions from
[−r, 0] to X with the uniform convergence topology and we will use simply
CX for C ([−r, 0] , X). For u ∈ C ([−r, b] , X), b > 0 and t ∈ [0, b], let ut
denote the element of CX defined by ut (θ) = u(t + θ), −r ≤ θ ≤ 0.
By an abstract semilinear functional differential equation on the space
X, we mean an evolution equation of the type
(
du
(t) = A0 u(t) + F (t, ut ), t ≥ 0,
dt (1)
u0 = ϕ,

where A0 : D(A0 ) ⊆ X → X is a linear operator, F is a function from

[0, +∞) × CX into X and ϕ ∈ CX are given. The initial value problem
associated with (1) is the following : given ϕ ∈ CX , to find a continuous
∗
e-mail: [email protected]

1
function u : [−r, h) → X, h > 0, differentiable on [0, h) such that u(t) ∈
D(A0 ), for t ∈ [0, h) and u satisfies the evolution equation of (1) for t ∈ [0, h)
and u0 = ϕ.
It is well-known (see for example [65] and [67]) that the classical semi-
group theory ensures the well posedness of Problem (1) when A0 is the infin-
itesimal generator of a C0 -semigroup of bounded linear operators (T0 (t))t≥0
in X or, equivalently when
(i) D(A0 ) = X,
(ii) there exist M0 , ω0 ∈ IR such that if λ > ω0 (λI − A0 )−1 ∈ L(X) and
° °
°(λ − ω0 )n (λI − A0 )−n ° ≤ M0 , ∀ n ∈ N.

In this case one can prove existence and uniqueness of a solution of (1), for
example by using the variation-of-constants formula ([65], [67])
½ Rt
T0 (t)ϕ(0) + 0 T0 (t − s)F (s, us )ds, if t ≥ 0,
u(t) =
ϕ(t), if t ∈ [−r, 0] ,

for every ϕ ∈ CX .
For related results, see for example Travis and Webb [65], Webb [66],
Fitzgibbon [43], Kunish and Schappacher ([49], [50]), Memory ([53], [54]),
Wu [67], and the references therein. In all of the quoted papers, A0 is an
operator verifying (i) and (ii). In the applications, it is sometimes con-
venient to take initial functions with more restrictions. There are many
examples in concrete situations where evolution equations are not densely
defined. Only hypothesis (ii) holds. One can refer for this to [40] for more
details. Non-density occurs, in many situations, from restrictions made on
the space where the equation is considered (for example, periodic continuous
functions, Hölder continuous functions) or from boundary conditions (e.g.,
the space C 1 with null value on the boundary is non-dense in the space of
continuous functions). Let us now briefly discuss the use of integrated semi-
groups. In the case where the mapping F in Equation (1) is equal to zero,
the problem can still be handled by using the classical semigroups theory
because A0 generates a strongly continuous semigroup in the space D(A0 ).
But, if F 6= 0, it is necessary to impose additional restrictions. A case which
is easily handled is when F takes their values in D(A0 ). On the other hand,
the integrated semigroups theory allows the range of the operators F to be
any subset of X.

2
Example 1 Consider the model of population dynamics with delay described
by

 ∂u ∂u
 (t, a) + (t, a) = f (t, a, ut (·, a)), (t, a) ∈ [0, T ] , × [0, l] ,
∂t ∂a
 u(t, 0) = 0, t ∈ [0, T ] ,
 (2)
u(θ, a) = ϕ(θ, a), (θ, a) ∈ [−r, 0] × [0, l] ,

where ϕ is a given function on CX := C ([−r, 0] , X), with X = C ([0, l] , IR).

By setting V (t) = u(t, ·), we can reformulate the partial differential problem
(2) as an abstract semilinear functional differential equation
½ ′
V (t) = A0 V (t) + F (t, Vt ), t ∈ [0, T ] ,
(3)
V0 = ϕ ∈ CX ,

where ½ © ª
D(A0 ) = u ∈ C 1 ([0, l] , IR) ; u(0) = 0 ,
A0 u = −u′ ,
and F : [0, T ] × CX → X is defined by F (t, ϕ)(a) = f (t, a, ϕ(·, a)) for t ∈
[0, T ], ϕ ∈ CX and a ∈ [0, l].

Example 2 Consider the following ecological model taken from [67] and it’s
described by the following reaction-diffusion equation with delay:

 ∂u
 (t, x) = ∆u(t, x) + f (t, x, ut (·, x)), t ∈ [0, T ] , x ∈ Ω,
∂t
 u(t, x) = 0, t ∈ [0, T ] , x ∈ ∂Ω,
 (4)
u(θ, x) = ϕ(θ, x), θ ∈ [−r, 0] , x ∈ Ω,

where Ω ⊂ IRn is a bounded open set with regular boundary ∂Ω, ∆ is the
Laplace operator in the sense of distributions
¡ ¢ on Ω and ϕ is a given function
on CX :=C([−r, 0] , X), with X = C Ω, IR .
The problem (4) can be reformulated as the abstract semilinear functional
differential equation (3), with
½ © ¡ ¢ ¡ ¢ ª
D(A0 ) = u ∈ C Ω, IR ; ∆u ∈ C Ω, IR and u = 0 on ∂Ω ,
A0 u = ∆u,

and F : [0, T ] × CX → X is defined by F (t, ϕ)(x) = f (t, x, ϕ(·, x)) for

t ∈ [0, T ], ϕ ∈ CX and x ∈ Ω.

3
 In the two examples given here, the operator A0 satisfies (ii) but the do-
main D(A0 ) is not dense in X and so, A0 does not generate a C0 −semigroup.
Of course, there are many other examples encountered in the applications
in which the operator A0 satisfies only (ii) (see [40]).
We will study here the abstract semilinear functional differential equa-
tion (1) in the case when the operator A0 satisfies only the Hille-Yosida
condition (ii). After providing some background materials in Section 2, we
proceed to establish the main results. A natural generalized notion of solu-
tions is provided in Section 3 by integral solutions. We derive a variation-
of-constants formula which allows us to transform the integral solutions of
the general equation to solutions of an abstract Volterra integral equation.
We prove the existence, uniqueness, regularity and continuous dependence
on the initial condition. These results give natural generalizations of re-
sults in [65] and [67]. In Section 4, we consider the autonomous case. We
prove that the solutions generate a nonlinear strongly continuous semigroup,
which satisfies a compactness property. In the linear case, the solutions are
shown to generate a locally Lipschitz continuous integrated semigroup. In
Section 5, we use a principle of linearized stability for strongly continuous
semigroups given by Desh and Schappacher [41] (see also [58], [59] and [61])
to study, in the nonlinear autonomous case, the stability of Equation (1). In
Section 6, we show in the linear autonomous case, the existence of a direct
sum decomposition of a state space into three subspaces : stable, unstable
and center, which are semigroup invariants. As a consequence of the results
established in Section 6, the existence of bounded, periodic and almost pe-
riodic solutions is established in the sections 7 and 8. In the end, we give
some examples.

2 Basic results
In this section, we give a short review of the theory of integrated semigroups
and differential operators with non-dense domain. We start with a few
definitions.

Definition 3 [34] Let X be a Banach space. A family (S(t))t≥0 ⊂ L(X) is

called an integrated semigroup if the following conditions are satisfied :
(i) S(0) = 0;
(ii) for any x ∈ X, S(t)x is a continuous function of t ≥ 0 with values in
X;
(iii) for any t, s ≥ 0 S(s)S(t) =s0 (S(t + τ ) − S(τ ))dτ.

4
Definition 4 [34] An integrated semigroup (S(t))t≥0 is called exponentially
bounded, if there exist constants M ≥ 0 and ω ∈ IR such that

kS(t)k ≤ M eωt for t ≥ 0.

Moreover (S(t))t≥0 is called non-degenerate if S(t)x = 0, for all t ≥ 0,

implies that x = 0.

If (S(t))t≥0 is an integrated semigroup, exponentially bounded, then the

Laplace transform R(λ) := λ+∞ 0 e
−λt S(t)dt exists for all λ with Re(λ) > ω.

R(λ) is injective if and only if (S(t))t≥0 is non-degenerate. R(λ) satisfies

the following expression

R(λ) − R(µ) = (µ − λ)R(λ)R(µ),

and in the case when (S(t))t≥0 is non-degenerate, there exists a unique

operator A satisfying (ω, +∞) ⊂ ρ(A) (the resolvent set of A) such that

R(λ) = (λI − A)−1 , for all Re(λ) > ω.

This operator A is called the generator of (S(t))t≥0 .

We have the following definition.

Definition 5 [34] An operator A is called a generator of an integrated semi-

group, if there exists ω ∈ IR such that (ω, +∞) ⊂ ρ(A), and there exists a
strongly continuous exponentially bounded family (S(t))t≥0 of linear bounded
operators such that S(0) = 0 and (λI − A)−1 = λ+∞ 0 e
−λt S(t)dt, for all

λ > ω.

Remark 1 If an operator A is the generator of an integrated semigroup

(S(t))t≥0 , then ∀λ ∈ IR, A − λI is the generator of the integrated semigroup
(Sλ (t))t≥0 given by
Z t
−λt
Sλ (t) = e S(t) + λ e−λs S(s) ds.
0

Proposition 6 [34] Let A be the generator of an integrated semigroup (S(t))t≥0 .

Then for all x ∈ X and t ≥ 0,
Z t µZ t ¶
S(s)xds ∈ D(A) and S(t)x = A S(s)xds + tx.
0 0

5
Moreover, for all x ∈ D(A), t ≥ 0

S(t)x ∈ D(A) and AS(t)x = S(t)Ax,

and Z t
S(t)x = tx + S(s)Ax ds.
0

Corollary 7 [34] Let A be the generator of an integrated semigroup (S(t))t≥0 .

Then for all x ∈ X and t ≥ 0 one has S(t)x ∈ D(A).
Moreover, for x ∈ X, S(·)x is right-sided differentiable in t ≥ 0 if and only
if S(t)x ∈ D(A). In that case

S ′ (t)x = AS(t)x + x.

An important special case is when the integrated semigroup is locally

Lipschitz continuous (with respect to time).

Definition 8 [52] An integrated semigroup (S(t))t≥0 is called locally Lip-

schitz continuous, if for all τ > 0 there exists a constant k(τ ) > 0 such
that
kS(t) − S(s)k ≤ k(τ ) |t − s| , for all t, s ∈ [0, τ ] .

In this case, we know from [52], that (S(t))t≥0 is exponentially bounded.

Definition 9 [52] We say that a linear operator A satisfies the Hille-Yosida

condition (HY) if there exist M ≥ 0 and ω ∈ IR such that (ω, +∞) ⊂ ρ(A)
and
© ° ° ª
sup (λ − ω)n °(λI − A)−n ° , n ∈ N, λ > ω ≤ M. (HY)

The following theorem shows that the Hille-Yosida condition character-

izes generators of locally Lipschitz continuous integrated semigroups.

Theorem 10 [52] The following assertions are equivalent.

(i) A is the generator of a locally Lipschitz continuous integrated semigroup,
(ii) A satisfies the condition (HY).

In the sequel, we give some results for the existence of solutions of the
following Cauchy problem
(
du
(t) = Au(t) + f (t), t ≥ 0,
dt (5)
u(0) = x ∈ X,

6
where A satisfies the condition (HY), without being densely defined.
By a solution of Problem (5) on [0, T ] where T > 0, we understand a
function u ∈ C 1 ([0, T ] , X) satisfying u(t) ∈ D(A) (t ∈ [0, T ]) such that the
two relations in (5) hold.
The following result is due to Da Prato and Sinestrari.

Theorem 11 [40] Let A : D(A) ⊆ X → X be a linear operator, f : [0, T ] →

X, x ∈ D(A) such that
(a) A satisfies the Rcondition (HY).
t
(b) f (t) = f (0) + 0 g(s) ds for some Bochner-integrable function g.
(c) Ax + f (0) ∈ D(A).
Then there exists a unique solution u of Problem (5) on the interval [0, T ],
and for each t ∈ [0, T ]
µ Z t ¶
ωt −ωs
|u(t)| ≤ M e |x| + e |f (s)| ds .
0

In the case where x is not sufficiently regular (that is, x is just in D(A))
there may not exist a strong solution u(t) ∈ X but, following the work of Da
Prato and Sinestrari [40], Problem (5) may still have an integral solution.

Definition 12 [40] Given f ∈ L1loc (0, +∞; X) and x ∈ X, we say that

u : [0, +∞) → X is an integral solution of (5) if the following assertions
are true
(i) u ∈ C([0, +∞) ; X),
(ii) t0 u(s)ds ∈ D(A), for t ≥ 0,
(iii) u(t) = x + At0 u(s)ds +t0 f (s)ds, for t ≥ 0.

From this definition, we deduce that for an integral solution u, we

t+h
have u(t) ∈ D(A), for all t > 0, because u(t) = h → 0lim h1 t u(s)ds and
t+h
t u(s)ds ∈ D(A). In particular, x ∈ D(A) is a necessary condition for the
existence of an integral solution of (5).

Theorem 13 [37] Suppose that A satisfies the condition (HY), x ∈ D(A)

and f : [0, +∞) → X is a continuous function. Then the problem (5) has a
unique integral solution which is given by
Z
d t
u(t) = S ′ (t)x + S(t − s)f (s)ds, pour t ≥ 0,
dt 0

7
where S(t) is the integrated semigroup generated by A.
Furthermore, the function u satisfies the inequality
µ Z t ¶
|u(t)| ≤ M eωt |x| + e−ωs |f (s)| ds , for t ≥ 0.
0
Rt
Note that Theorem 13 also says that 0 S(t − s)f (s)ds is differentiable
with respect to t.

3 Existence, uniqueness and regularity of solutions

We restate the problem
(
du
(t) = A0 u(t) + F (t, ut ), t ≥ 0,
dt (1)
u0 = ϕ ∈ CX ,
where F : [0, T ] × CX → X is a continuous function.
Throughout this work, we assume that A0 satisfies the Hille-Yosida condition
on X :
(HY) there exist M0 ≥ 0 and ω0 ∈ IR such that (ω0 , +∞) ⊂ ρ(A0 ) and
© ° ° ª
sup (λ − ω0 )n °(λI − A0 )−n ° , n ∈ N, λ > ω0 ≤ M0 .
We know from Theorem 10 that A0 is the generator of a locally Lip-
schitz continuous integrated semigroup (S0 (t))t≥0 on X, (and (S0 (t))t≥0 is
exponentially bounded).
In view of the remark following Definition 5, we will sometimes assume
without loss of generality that ω0 = 0.
Consider first the linear Cauchy problem
½ ′
u (t) = A0 u(t), t ≥ 0,
u0 = ϕ ∈ CX .
This problem can be reformulated as a special case of an abstract semilinear
functional differential equation with delay. This is
½ ′
u (t) = (Aut ) (0) , t ≥ 0,
u0 = ϕ,
where
½ © ª
D(A) = ϕ ∈ C 1 ([−r, 0] , X) ; ϕ(0) ∈ D(A0 ), ϕ′ (0) = A0 ϕ(0) ,
Aϕ = ϕ′ .

8
 We can show, by using the next result and Theorem 10, that the operator
A satisfies the condition (HY).

Proposition 14 The operator A is the generator of a locally Lipschitz con-

tinuous integrated semigroup on CX given by
( R
0
θ ϕ(s) ds + S0 (t + θ)ϕ(0), t + θ ≥ 0,
(S(t)ϕ) (θ) = R t+θ
θ ϕ(s) ds, t + θ < 0.

for t ≥ 0, θ ∈ [−r, 0] and ϕ ∈CX .

Proof. It is easy to see that (S(t))t≥0 is an integrated semigroup on CX .

Consider τ > 0, t, s ∈ [0, τ ] and ϕ ∈CX .
If t + θ ≥ 0 and s + θ ≥ 0, we have

(S (t) ϕ) (θ) − (S(s)ϕ)(θ) = S0 (t + θ)ϕ(0) − S0 (s + θ)ϕ(0).

It follows immediately that there exists a constant k := k(τ ) > 0 such that

|(S (t) ϕ) (θ) − (S(s)ϕ) (θ)| ≤ k |t − s| |ϕ(0)| ,

because (S0 (t))t≥0 is Lipschitz continuous on [0, τ ].

If t + θ ≤ 0 and s + θ ≤ 0, we have

|(S (t) ϕ) (θ) − (S(s)ϕ) (θ)| = 0.

If t + θ ≥ 0 and s + θ ≤ 0, we obtain
Z 0
(S (t) ϕ) (θ) − (S(s)ϕ) (θ) = S0 (t + θ)ϕ(0) + ϕ(u)du,
s+θ

then

|(S (t) ϕ) (θ) − (S(s)ϕ) (θ)| ≤ k (t + θ) |ϕ(0)| − (s + θ) kϕk ,

This implies that

kS (t) − S(s)k ≤ (k + 1) |t − s| .
It may be concluded that (S(t))t≥0 is locally Lipschitz continuous.
In order to prove that A is the generator of (S(t))t≥0 , we calculate the
spectrum and the resolvent operator of A.

9
 Consider the equation

(λI − A) ϕ = ψ,

where ψ is given in CX , and we are looking for ϕ ∈ D(A). The above

equation reads
λϕ(θ) − ϕ′ (θ) = ψ(θ), θ ∈ [−r, 0] .
Whose solutions are such that
Z 0
ϕ(θ) = eλθ ϕ(0) + eλ(θ−s) ψ(s) ds, θ ∈ [−r, 0] .
θ

ϕ is in D(A) if ϕ(0) ∈ D(A0 ) and ϕ′ (0) = A0 ϕ(0), that is

ϕ(0) ∈ D(A0 ) and (λI − A0 ) ϕ(0) = ψ(0).

By assumption on A0 , we know that (0, +∞) ⊂ ρ(A0 ). So, for λ > 0, the
above equation has a solution ϕ(0) = (λI − A0 )−1 ψ(0).
Therefore, (0, +∞) ⊂ ρ(A) and
³ ´ Z 0
−1 λθ −1
(λI − A) ψ (θ) = e (λI − A0 ) ψ(0) + eλ(θ−s) ψ(s)ds,
θ

for θ ∈ [−r, 0] and λ > 0.

On the other hand, from the formula stated in Proposition 14, it is clear
that t → (S(t)ϕ) (θ) has at most exponential growth, not larger than ω0 = 0.
Therefore, one can defined its Laplace transform, for each λ > 0. We obtain
R +∞ −λt R −θ −λt R t+θ R +∞ −λt R 0
0 e (S (t) ϕ) (θ) dt = e ϕ(s)ds dt + −θ e θ ϕ(s)ds dt
R0+∞ θ
+ −θ e−λt S0 (t + θ)ϕ(0) dt.

Integrating by parts the first expression, it yields

R +∞ −λt R
eλθ 0
R
1 0 λ(θ−s)
0 e (S (t) ϕ) (θ) dt = − λ ϕ(s)ds + θ e ϕ(s)ds
θ
λθ R 0 Rλ +∞
+ eλ θ ϕ(s)ds + 0 eλ(θ−s) S0 (s)ϕ(0)ds,
³R R +∞ ´
eλθ 0
= λ θ e−λs ϕ(s)ds + 0 e−λs S0 (s)ϕ(0)ds ,
³R ´
eλθ 0 −λs
= λ θ e ϕ(s)ds + (λI − A0 )−1 ϕ(0) ,

1³ ´
= (λI − A)−1 ϕ (θ).
λ

10
So, A is related to (S(t))t≥0 by the formula which characterizes the infini-
tesimal generator of an integrated semigroup. The proof of Proposition 14
is complete
Our next objective is to construct an integrated version of Problem (1)
using integrated semigroups. We need to extend the integrated semigroup
(S(t))t≥0 to the space CeX = CX ⊕ hX0 i, where
hX0 i = {X0 c, c ∈ X and (X0 c) (θ) = X0 (θ)c} and X0 denotes the function
defined by X0 (θ) = 0 if θ < 0 and X0 (0) = IdX . We shall prove that this
extension determines a locally Lipschitz continuous integrated semigroup on
CeX .
e
Proposition 15 The family of operators (S(t)) e
t≥0 defined on CX by

e
S(t)ϕ = S(t)ϕ, for ϕ ∈ CX
and
³ ´ ½
e S0 (t + θ)c, if t + θ ≥ 0,
S(t)X 0 c (θ) =
0, if t + θ ≤ 0, for c ∈ X,

is a locally Lipschitz continuous integrated semigroup on CeX generated by

the operator Ae defined by
( © ª
e = ϕ ∈ C 1 ([−r, 0] , X) ; ϕ(0) ∈ D(A0 ) ,
D(A)
e = ϕ′ + X0 (A0 ϕ(0) − ϕ′ (0)) .
Aϕ

Proof. Using the same reasoning as in the proof of Proposition 14, one can
e
show that (S(t)) t≥0 is a locally Lipschitz continuous integrated semigroup
on CeX .
The proof will be completed by showing that
(
(0, +∞) ⊂ ρ(A) e and
³ ´−1 R +∞
e
λI − A ϕ e ϕ
e = λ 0 e−λt S(t) e ∈ CeX .
e dt, for λ > 0 and ϕ

For this, we need the following lemma.

Lemma 16 For λ > ­ 0,®one has ­ ® © ª

(i) D(A) e = D(A)⊕ eλ· , where eλ· = eλ· c; c ∈ D(A0 ), (eλ· c)(θ) = eλθ c ,
e and
(ii) (0, +∞) ⊂ ρ(A)
e −1 (ϕ + X0 c) = (λI − A)−1 ϕ + eλ· (λI − A0 )−1 c,
(λI − A)
for every (ϕ, c) ∈ CX × X .

11
 Proof of the lemma. For the proof of (i), we consider the following
operator
e →X
l : D(A)
ϕ → l(ϕ) = A0 ϕ(0) − ϕ′ (0).
e ∈ D(A)
Let Ψ e and λ > 0. Setting Ψ = Ψe − eλ· (λI − A0 )−1 l(Ψ),
e we deduce
that Ψ ∈ Ker(l) = D(A), and the decomposition is clearly unique.
(ii) Consider the equation
³ ´³ ´
λI − Ae ϕ + eλ· c = ψ + X0 a,

where (ψ, a) is given, ψ ∈ CX , a ∈ X, and we are looking for (ϕ, c) , ϕ ∈

D(A), c ∈ D(A0 ).
This yields

(λI − A) ϕ + λeλ· c − λeλ· c − X0 (A0 c − λc) = ψ + X0 a,

which has the solution

½
ϕ = (λI − A)−1 ψ,
c = (λI − A0 )−1 a.
Consequently,
(
e
(0, +∞) ⊂ ρ(A),
³ ´−1
e
λI − A (ϕ + X0 c) = (λI − A)−1 ϕ + eλ· (λI − A0 )−1 c.

This complete the proof of the lemma

We now turn to the proof of the proposition. All we want to show is
1³ e
´−1
e ϕ.
that λI − A ϕe can be expressed as the Laplace transform of S(t) e
λ
In view of the decomposition and what has been already done for A, we may
e = X0 c. In this case, we have
restrict our attention to the case when ϕ
³ ´
e −1 X0 c (θ) = eλθ (λI − A0 )−1 c,
(λI − A)
R +∞
= λeλθ 0 e−λt S0 (t)c dt,
R +∞
= λ −θ e−λt S³0 (t + θ)c´dt,
R +∞
e
= λ 0 e−λt S(t)X 0 c (θ) dt,

which completes the proof of the proposition

We will need also the following general lemma.

12
Lemma 17 Let (U (t))t≥0 be a locally Lipschitz continuous integrated semi-
group on a Banach space (E, |·|) generated by (A, D(A)) and G : [0, T ] →
E (0 < T ), a Bochner-integrable function. Then, the function K : [0, T ] →
E defined by Z t
K(t) = U (t − s)G(s) ds
0
is continuously differentiable on [0, T ] and satisfies, for t ∈ [0, T ] ,
Z Z t
1 t ′
(i) K ′ (t) = h ց 0lim U (t − s)U (h)G(s) ds = λ → +∞lim U ′ (t −
h 0 0
s)(Aλ G(s)) ds,
with Aλ = λ (λI − A)−1 ,
(ii) K(t) ∈ D(A), Rt
(iii) K ′ (t) = AK(t) + 0 G(s) ds.

Proof. Theorem 13 implies that K is continuously differentiable on [0, T ],

and for the prove of (ii) and (iii), see [37], [47] and [63]. On the other hand,
we know by the definition of an integrated semigroup that
Z t
U (t)U (h)x = (U (s + h)x − U (s)x) ds,
0

for t, h ≥ 0 and x ∈ E. This yields that the function t → U (t)U (h)x is

continuously differentiable on [0, T ] , for each h ≥ 0; x ∈ E, and satisfies
U ′ (t)U (h)x = U (t + h)x − U (t)x.
Furthermore, we have
µ Z t
′ 1
K (t) = h ց 0lim (U (t + h − s) − U (t − s)) G(s) ds
h 0
Z ¶
1 t+h
+ U (t + h − s)G(s) ds .
h t
If we put, in the second integral of the right-hand side, u = h1 (s − t), we
obtain
Z Z 1
1 t+h
U (t + h − s)G(s) ds = U (h(1 − u))G(t + hu) du.
h t 0

This implies that

Z t+h
1
h ց 0lim U (t + h − s)G(s) ds = 0.
h t

13
Hence
Z t
1
K ′ (t) = h ց 0lim (U (t + h − s) − U (t − s)) G(s) ds.
h 0

But
U (t + h − s) − U (t − s) = U ′ (t − s)U (h).
It follows that, for t ∈ [0, T ]
Z t
′ 1
K (t) = h ց 0lim U ′ (t − s) U (h)G(s) ds.
0 h

For the prove of the last equality of (i), see [64]

We are now able to state a first result of existence and uniqueness of
solutions.

Theorem 18 Let F : [0, T ] × CX → X be continuous and satisfy a Lipschitz

condition

|F (t, ϕ) − F (t, ψ)| ≤ L kϕ − ψk , t ∈ [0, T ] and ϕ, ψ ∈ CX ,

where L is a positive constant. Then, for given ϕ ∈ CX , such that ϕ(0) ∈

D(A0 ), there exists a unique function y : [0, T ] → CX which solves the
following abstract integral equation
Z
′ d te
y(t) = S (t)ϕ + S(t − s)X0 F (s, y(s)) ds, for t ∈ [0, T ] , (6)
dt 0
³ ´
with (S(t))t≥0 given in Proposition 14 and S(t)e in Proposition 15.
t≥0

Proof. Using the results of Theorem 13, the proof of this theorem is stan-
dard.
Since ϕ(0) ∈ D(A0 ), we have ϕ ∈ D(A), where A is the generator of
the integrated semigroup (S(t))t≥0 on CX . Then, we deduce from Corol-
lary 7 that S(·)ϕ is differentiable and (S ′ (t))t≥0 can be defined to be a
C0 -semigroup on D(A).
Let (y n )n∈N be a sequence of continuous functions defined by

y 0 (t) = S ′ (t)ϕ, t ∈ [0, T ]

Rt
y n (t) = S ′ (t)ϕ + d e n−1 (s)) ds,
dt 0 S(t − s)X0 F (s, y t ∈ [0, T ] , n ≥ 1.

14
By
¯ virtue of the continuity of F and S ′ (·)ϕ, there exists α ≥ 0 such that
¯
¯F (s, y 0 (s)¯ ≤ α, for s ∈ [0, T ] . Then, using Theorem 13, we obtain
Z t
¯ 1 ¯ ¯ ¯
¯y (t) − y 0 (t)¯ ≤ M0 ¯F (s, y 0 (s)¯ ds.
0
Hence ¯ 1 ¯
¯y (t) − y 0 (t)¯ ≤ M0 αt.
In general case we have
Z
¯ n ¯ t¯ ¯
¯y (t) − y n−1 (t)¯ ≤ M0 L ¯y n−1 (s) − y n−2 (s)¯ ds.
0
So,
¯ n ¯ n
¯y (t) − y n−1 (t)¯ ≤ M0n Ln−1 α t .
n!
n
Consequently, the limit y := n → ∞limy (t) exists uniformly on [0, T ] and
y : [0, T ] → CX is continuous.
In order to prove that y is a solution of Equation (6), we introduce the
function v defined by
¯ Z t ¯
¯ d ¯
v(t) = ¯¯y(t) − S (t)ϕ −
′ e − s)X0 F (s, y(s)) ds¯ .
S(t
dt 0 ¯
We have
¯ ¯ ¯¯ n+1 R
d t e
¯
¯
¯
v(t) ≤ y(t) − y n+1 ¯
(t) + ¯y ′
(t) − S (t)ϕ − dt 0 S(t − s)X0 F (s, y(s)) ds¯ ,
¯ ¯
¯ ¯d Rt ¯
e ¯
≤ ¯y(t) − y n+1 (t)¯ + ¯ dt n
0 S(t − s)X0 (F (s, y(s)) − F (s, y (s))) ds¯ ,
¯ ¯ Rt
≤ ¯y(t) − y n+1 (t)¯ + M0 L 0 |y(t) − y n (t)| .
Moreover, we have
∞
X ¡ ¢
y(t) − y n (t) = p = n y p+1 (t) − y p (t) .
This implies that
P
∞
tp+1 tn+1
α
v(t) ≤ (1 + M0 L) L p = n (M0 L)p+1 (p+1)! + α
L (M0 L)n+1 (n+1)! , for n ∈ N.
Consequently we obtain v = 0 on [0, T ].
To show uniqueness, suppose that z(t) is also a solution of Equation (6).
Then Z t
|y(t) − z(t)| ≤ M0 L |y(s) − z(s)| ds.
0
By Gronwall’s inequality, z = y on [0, T ]

15
Corollary 19 Under the same assumptions as in Theorem 18, the solution
y : [0, T ] → CX of the abstract integral equation (6) is the unique integral
solution of the equation
½ ′
e
y (t) = Ay(t) + X0 F (t, y(t)), t ≥ 0,
y(0) = ϕ ∈ CX ,
i.e.
(i) y ∈ C([0, T ] ; CX ),
(ii) t0 y(s)ds ∈ D(A),e for t ∈ [0, T ] ,
e
(iii) y(t) = ϕ + A0 y(s)ds + X0 t0 F (s, y(s))ds,
t for t ∈ [0, T ] ,
where the operator A e is given in Proposition 15.
Furthermore, we have, for t ∈ [0, T ],
µ Z t ¶
ky(t)k ≤ M0 kϕk + |F (s, y(s))| ds .
0

Proof. If we define the function f : [0, T ] → CeX by f (s) = X0 F (s, y(s)),

we can use Theorem 13 and Theorem 18 to prove this result
Corollary 20 Under the same assumptions as in Theorem 18, the solution
y of the integral equation (6) satisfies, for t ∈ [0, T ] and θ ∈ [−r, 0] the
translation property
½
y(t + θ)(0) if t + θ ≥ 0,
y(t)(θ) =
ϕ(t + θ) if t + θ ≤ 0.
Moreover, if we consider the function u : [−r, T ] → X defined by
½
y(t)(0) if t ≥ 0,
u(t) =
ϕ(t) if t ≤ 0.
Then, u is the unique integral solution of the problem (1), i.e.



 (i) u ∈ C([−r, T ] ; X),

(ii) t0 u(s)ds ½
∈ D(A0 ), for t ∈ [0, T ] ,
t t
ϕ(0) + A0 0 u(s)ds +0 F (s, us )ds, for t ∈ [0, T ] ,


 (iii) u(t) = (7)
ϕ(t), for t ∈ [−r, 0] .
The function u ∈ C([−r, T ] ; X) is also the unique solution of
 Z
 ′ d t
S0 (t)ϕ(0) + S0 (t − s)F (s, us )ds, for t ∈ [0, T ] ,
u(t) = dt 0

ϕ(t), for t ∈ [−r, 0] . (8)

16
Furthermore, we have for t ∈ [0, T ]
µ Z t ¶
kut k ≤ M0 kϕk + |F (s, us )| ds .
0

Conversely, if u is an integral solution of Equation (1), then the function

t 7→ ut is a solution of Equation (6).

Proof. From Proposition 14 and Proposition 15, we have for t + θ ≥ 0

Z
′ d t+θ
y(t)(θ) = S0 (t + θ)ϕ(0) + S0 (t + θ − s)F (s, y(s)) ds,
dt 0
= y(t + θ)ϕ(0),
Rt³ ´
and for t + θ ≤ 0, we obtain 0 S(t e − s)X0 F (s, y(s)) (θ) ds = 0 and
(S ′ (t)ϕ) (θ) = ϕ(t + θ). Hence

y(t)(θ) = ϕ(t + θ).

The second part of the corollary follows from Corollary 19

Remark 2 A continuous function u from [−r, T ] into X is called an integral

solution of Equation (1) if the function t 7→ ut satisfies Equation (6) or
equivalently u satisfies (7) or equivalently u satisfies (8).

Corollary 21 Assume that the hypotheses of Theorem 18 are satisfied and

let u, u ϕ ∈ CX , respectively.
b be the functions given by Corollary 20 for ϕ, b
Then, for t ∈ [0, T ]

bt k ≤ M0 eLt kϕ − ϕk
kut − u b .

Proof. This is just a consequence of the last inequality stated in Corollary

20. After the equation (6) has been centered near u, we obtain the equation
Z
d te
v(t) = S ′ (t) (ϕ
b − ϕ) + S(t − s)X0 (F (s, us + v(s)) − F (s, us ) ) ds,
dt 0
with
bt − ut .
v(t) = u
This yields
µ Z t ¶
kv(t)k ≤ M0 kϕ − ϕk
b + |F (s, us ) − F (s, u
bs )| ds ,
0

17
this is µ Z t ¶
kut − u
bt k ≤ M0 kϕ − ϕk
b +L kus − u
bs k ds .
0
By Gronwall’s inequality, we obtain

bt k ≤ M0 eLt kϕ − ϕk
kut − u b

We give now two results of regularity of the integral solutions of (1).

Theorem 22 Assume that F : [0, T ] × CX → X is continuously differen-

tiable and there exist constants L, β, γ ≥ 0 such that

|F (t, ϕ) − F (t, ψ)| ≤ L kϕ − ψk ,

|Dt F (t, ϕ) − Dt F (t, ψ)| ≤ β kϕ − ψk ,
|Dϕ F (t, ϕ) − Dϕ F (t, ψ)| ≤ γ kϕ − ψk ,

for all t ∈ [0, T ] and ϕ, ψ ∈ CX , where Dt F and Dϕ F denote the derivatives.

Then, for given ϕ ∈ CX such that

ϕ(0) ∈ D(A0 ), ϕ′ ∈ CX , ϕ′ (0) ∈ D(A0 ) and ϕ′ (0) = A0 ϕ(0) + F (0, ϕ),

let y : [0, T ] → CX be the solution of the abstract integral equation (6) such
that y(0) = ϕ. Then, y is continuously differentiable on [0, T ] and satisfies
the Cauchy problem
½ ′
e
y (t) = Ay(t) + X0 F (t, y(t)), t ∈ [0, T ] ,
y(0) = ϕ.

Moreover, the function u defined on [−r, T ] by

½
y(t)(0) if t ≥ 0,
u(t) =
ϕ(t) if t < 0,

is continuously differentiable on [−r, T ] and satisfies the Cauchy problem

(1).

Proof. Let y be the solution of Equation (6) on [0, T ] such that y(0) = ϕ.
We deduce from Theorem 18 that there exists a unique function v : [0, T ] →
CX which solves the following integral equation
Z
d te
v(t) = S ′ (t)ϕ′ + S(t − s)X0 (Dt F (s, y(s)) + Dϕ F (s, y(s))v(s)) ds,
dt 0

18
such that v(0) = ϕ′ .
Let w : [0, T ] → CX be the function defined by
Z t
w(t) = ϕ + v(s) ds, for t ∈ [0, T ] .
0

We will show that w = y on [0, T ].

Using the expression satisfied by v, we obtain
Rt
w(t) = ϕ + 0 S ′ (s)ϕ′ ds
Rt
e − s)X0 (Dt F (s, y(s)) + Dϕ F (s, y(s))v(s)) ds,
+ 0 S(t
Rt
= ϕ + S(t)ϕ′ + e − s)X0 (Dt F (s, y(s)) + Dϕ F (s, y(s))v(s)) ds.
S(t
0

On the other hand, we have ϕ ∈ D(A) e and ϕ′ (0) = A0 ϕ(0) + F (0, ϕ), then
′ e
ϕ = Aϕ + X0 F (0, ϕ). This implies that
e
S(t)ϕ′ = S(t)ϕ′ e Aϕ
= S(t) e + S(t)X
e 0 F (0, ϕ).

Using Corollary 7, we deduce that

e
S(t)ϕ′ = S ′ (t)ϕ − ϕ + S(t)X 0 F (0, ϕ).

Furthermore, we have
R
d t e
Rt
d e
dt 0 S(t − s)X0 F (s, w(s)) ds = dt 0 S(s)X 0 F (s, w(t − s)) ds

Rt
= S(t e
e − s)X0 [Dt F (s, w(s)) + Dϕ F (s, w(s))] v(s)ds + S(t)X 0 F (0, ϕ).
0

Then
R
d t e
w(t) = S ′ (t)ϕ + dt 0 S(t − s)X0 F (s, w(s))ds
Rt
− 0 S(te − s)X0 (Dt F (s, w(s)) + Dϕ F (s, w(s))v(s)) ds
Rt
+ 0 S(te − s)X0 (Dt F (s, y(s)) + Dϕ F (s, y(s))v(s)) ds,

R
d t e
= S ′ (t)ϕ + dt 0 S(t − s)X0 F (s, w(s))ds
Rt
e
+ 0 S(t − s)X0 (Dt F (s, y(s)) − Dt F (s, w(s))) ds
Rt
+ 0 S(te − s)X0 (Dϕ F (s, y(s)) − Dϕ F (s, w(s))) v(s) ds.

We obtain
R
d t e
w(t) − y(t) = dt R 0 S(t − s)X0 (F (s, w(s)) − F (s, y(s))) ds
t e
+ 0 S(t − s)X0 (Dt F (s, y(s)) − Dt F (s, w(s))) ds
Rt
e
+ 0 S(t − s)X0 (Dϕ F (s, y(s)) − Dϕ F (s, w(s))) v(s) ds.

19
So, we deduce
Z t
|w(t) − y(t)| ≤ M0 (L + β + γ |v(s)|) |w(s) − y(s)| ds.
0

By Gronwall’s inequality, we conclude that w = y on [0, T ]. This implies

that y is continuously differentiable on [0, T ].
Consider now the function g : [0, T ] → CeX defined by g(t) = X0 F (t, y(t))
and consider the Cauchy problem
½ ′
e
z (t) = Az(t) + g(t), t ∈ [0, T ] ,
(9)
z(0) = ϕ.

The assumptions of Theorem 22 imply that ϕ ∈ D(A), e Aϕ

e + g(0) ∈ D(A)e
and g is continuously differentiable on [0, T ]. Using Theorem 11, we deduce
that there exists a unique solution on [0, T ] of Equation (9). By Theorem
13, we know that this solution is given by
Z
′ d te
z(t) = S (t)ϕ + S(t − s)g(s)ds.
dt 0

Theorem 18 implies that z = y on [0, T ].

If we consider the function u defined on [−r, T ] by
½
y(t)(0) if t ≥ 0,
u(t) =
ϕ(t) if t < 0.
Rt
By virtue of Corollary 20, we have 0 u(s)ds ∈ D(A0 ) and
Z t Z t
u(t) = ϕ(0) + A0 u(s)ds + F (s, us )ds, for t ∈ [0, T ] .
0 0

We have also the existence of

µ Z t+h ¶
1
h → 0limA0 u(s) ds = u′ (t) − F (t, ut ).
h t

Furthermore, the operator A0 is closed. Then, we obtain u(t) ∈ D(A0 ) and

u′ (t) = A0 u(t) + F (t, ut ), for t ∈ [0, T ] .

The second part of the theorem is a consequence of Corollary 19

20
 Assume that T > r and A0 : D(A0 ) ⊆ X → X satisfies (with not
necessarily dense domain) the condition
 ¤π £

 there exist β ∈ 2 , π and M0 > 0 such that if
λ
° ∈ C − {0} ° |arg λ| < β, then
and (10)

 ° −1 °
°(λI − A0 ) ° ≤ M |λ|
0
.

The condition (10) is stronger than (HY).

We have the following result.

Theorem 23 Suppose that A0 satisfies (10) (non-densely defined) on X

and there exist a constant L > 0 and α ∈ ]0, 1[ such that

|F (t, ψ) − F (s, ϕ)| ≤ L (|t − s|α + kψ − ϕk)

for t, s ∈ [0, T ] and ψ, ϕ ∈ CX .

Then, for given ϕ ∈ CX , such that ϕ(0) ∈ D(A0 ), the integral solution u of
Equation (1) on [0, T ] is continuously differentiable on (r, T ] and satisfies

u(t) ∈ D(A0 ), u′ (t) ∈ D(A0 ) and

u′ (t) = A0 u(t) + F (t, ut ), for t ∈ (r, T ] .

Proof. We know, from ([48], p.487) that A0 is the generator of an analytic

semigroup (not necessarily C0 −semigroup) defined by
Z
1
eA0 t
= eλt (λI − A0 )−1 dλ, t > 0
2πi +C

where +C is a suitable oriented path in the complex plan.

Let u be the integral solution on [0, T ] of Equation (1), which exists by
virtu of Theorem 18, and consider the function g : [0, T ] → X defined by
g(t) = F (t, ut ). We deduce from ([60], p.106) that
Z t
u(t) = eA0 t ϕ(0) + eA0 (t−s) g(s) ds, for t ∈ [0, T ] .
0

Using (Theorem 3.4, [62]), we obtain that the function u is γ−Hölder con-
tinuous on [ε, T ] for each ε > 0 and γ ∈ ]0, 1[ . Hence, there exists L1 ≥ 0
such that
kut − us k ≤ L1 |t − s|γ , for t, s ∈ (r, T ] .

21
On the other hand, we have

|g(t) − g(s)| ≤ L(|t − s|α + kut − us k), for t, s ∈ [0, T ] .

Consequently, the function t ∈ (r, T ] → g(t) is locally Hölder continuous.

By virtue of (Theorem 4.4 and 4.5, [62]), we deduce that u is continuously
differentiable on (r, T ] and satisfies

u(t) ∈ D(A0 ), u′ (t) ∈ D(A0 ) and

u′ (t) = A0 u(t) + F (t, ut ), for t ∈ (r, T ]

We prove now the local existence of integral solutions of Problem (1)

under a locally Lipschitz condition on F .

Theorem 24 Suppose that F : [0, +∞) × CX → X is continuous and sat-

isfies the following locally Lipschitz condition : for each α > 0 there exists
a constant C0 (α) > 0 such that if t ≥ 0, ϕ1 , ϕ2 ∈ CX and kϕ1 k , kϕ2 k ≤ α
then
|F (t, ϕ1 ) − F (t, ϕ2 )| ≤ C0 (α) kϕ1 − ϕ2 k .
Let ϕ ∈ CX such that ϕ(0) ∈ D(A0 ). Then, there exists a maximal interval of
existence [−r, Tϕ [ , Tϕ > 0, and a unique integral solution u(., ϕ) of Equation
(1), defined on [−r, Tϕ [ and either

Tϕ = +∞ or t → Tϕ− lim sup |u(t, ϕ)| = +∞.

Moreover, u(t, ϕ) is a continuous function of ϕ, in the sense that if ϕ ∈ CX ,

ϕ(0) ∈ D(A0 ) and t ∈ [0, Tϕ [ , then there exist positive constants L and ε
such that, for ψ ∈ CX , ψ(0) ∈ D(A0 ) and kϕ − ψk < ε, we have

t ∈ [0, Tψ [ and |u(s, ϕ) − u(s, ψ)| ≤ L kϕ − ψk , for all s ∈ [−r, t] .

Proof. Let T1 > 0. Note that the locally Lipschitz condition on F implies
that, for each α > 0 there exists C0 (α) > 0 such that for ϕ ∈ CX and
kϕk ≤ α, we have

|F (t, ϕ)| ≤ C0 (α) kϕk + |F (t, 0)| ≤ αC0 (α) + s ∈ [0, T1 ]sup |F (s, 0)| .

Let ϕ ∈ CX , ϕ(0) ∈ D(A0 ), α = kϕk+1 and c1 = αC0 (α)+s ∈ [0, T1 ]sup |F (s, 0)| .
Consider the following set

Zϕ = {u ∈ C ([−r, T1 ] , X) : u (s) = ϕ (s) if s ∈ [−r, 0]

and 0 ≤ s ≤ T1 sup |u(s) − ϕ(0)| ≤ 1} ,

22
where C ([−r, T1 ] , X) is endowed with the uniform convergence topology.
It’s clear that Zϕ is a closed set of C ([−r, T1 ] , X) . Consider the mapping

H : Zϕ → C ([−r, T1 ] , X)

defined by

 dt
H (u) (t) = S0′ (t)ϕ(0) + S0 (t − s)F (s, us )ds, for t ∈ [0, T1 ] ,
 ϕ(t), dt 0
for t ∈ [−r, 0] .

We will show that

H (Zϕ ) ⊆ Zϕ .
Let u ∈ Zϕ and t ∈ [0, T1 ], we have, for suitable constants M0 and ω0
¯ t ¯
¯d ¯
|H (u) (t) − ϕ(0)| ≤ |S0 (t)ϕ(0) − ϕ(0)| + ¯¯
′ S0 (t − s)F (s, us )ds¯¯ ,
dt 0
≤ |S0′ (t)ϕ(0) − ϕ(0)| + M0 e0ω0 t t e−ω0 s |F (s, us )| ds.

We can assume, without loss of generality, that ω0 > 0. Then,

¯ ¯
|H (u) (t) − ϕ(0)| ≤ ¯S0′ (t)ϕ(0) − ϕ(0)¯ + M0 e0ω0 t t |F (s, us )| ds.

Since |u(s) − ϕ(0)| ≤ 1, for s ∈ [0, T1 ] , and α = kϕk + 1, we obtain |u(s)| ≤

α, for s ∈ [−r, T1 ] . Then, kus k ≤ α, for s ∈ [0, T1 ] and

|F (s, us )| ≤ C0 (α) kus k + |F (s, 0)| ,

≤ c1 .

Consider T1 > 0 sufficiently small such that

©¯ ¯ ª
0 ≤ s ≤ T1 sup ¯S0′ (s)ϕ(0) − ϕ(0)¯ + M0 eω0 s c1 s < 1.

So, we deduce that

|H (u) (t) − ϕ(0)| ≤ |S0′ (t)ϕ(0) − ϕ(0)| + M0 eω0 t c1 t

< 1,

for t ∈ [0, T1 ]. Hence,

H (Zϕ ) ⊆ Zϕ .

23
On the other hand, let u, v ∈ Zϕ and t ∈ [0, T1 ], we have
¯ t ¯
¯d ¯
¯
|H (u) (t) − H (v) (t)| = ¯ S0 (t − s)(F (s, us ) − F (s, vs ))ds¯¯ ,
dt 0
≤ M0 eω0 0 t t |F (s, us ) − F (s, vs )| ds,
≤ M0 eω0 t C0 (α)t0 kus − vs k ds,
≤ M0 eω0 T1 C0 (α) T1 ku − vkC([−r,T1 ],X) .

Note that α ≥ 1, then

M0 eω0 T1 C0 (α)T1 ≤ M0 eω0 T1 c1 T1 ,

≤ 0 ≤ s ≤ T1 sup {|S0′ (s)ϕ(0) − ϕ(0)| + M0 eω0 s c1 s} ,
< 1.

Then, H is a strict contraction in Zϕ . So, H has one and only one fixed
point u in Zϕ . We conclude that Equation (1) has one and only one integral
solution which is defined on the interval [−r, T1 ].
Let u(., ϕ) be the unique integral solution of Equation (1), defined on
it’s maximal interval of existence [0, Tϕ [ , Tϕ > 0.
Assume that Tϕ < +∞ and t → Tϕ− lim sup |u(t, ϕ)| < +∞. Then,
there exists a constant α > 0 such that ku(t, ϕ)k ≤ α, for t ∈ [−r, Tϕ [. Let
t, t + h ∈ [0, Tϕ [ , h > 0, and θ ∈ [−r, 0].
If t + θ ≥ 0, we obtain

|u(t ′ ′
¯ + θ + h, ϕ) − u(t + θ, ϕ)| ≤ |(S0 (t + θ + h) − S0 (t + θ))ϕ(0)| ¯
¯ d t+θ+h d t+θ ¯
¯ ¯
+¯ S0 (t + θ + h − s)F (s, us (., ϕ))ds − S0 (t + θ − s)F (s, us (., ϕ))ds¯ ,
¯ dt 0 dt 0 ¯
¯ ¯
¯ d t+θ+h ¯
¯ ¯
≤ kS0′ (t + θ)k |S0′ (h)ϕ(0) − ϕ(0)| + ¯ S0 (s)F (t + θ + h − s, ut+θ+h−s (., ϕ))ds¯
¯ dt t+θ ¯
¯ ¯
¯ d t+θ ¯
¯ ¯
+¯ S0 (s) (F (ut+θ+h−s , ϕ) − F (t + θ − s, ut+θ−s (., ϕ))) ds¯ .
¯ dt 0 ¯
This implies that,

|ut+h (θ, ϕ) − ut (θ, ϕ)| ≤ M0 eω0 Tϕ |S0′ (h)ϕ(0) − ϕ(0)| + M0 eω0 Tϕ c1 h

+M0 eω0 Tϕ C0 (α)t0 kus+h (., ϕ) − us (., ϕ)k ds.

If t + θ < 0. Let h0 > 0 sufficiently small such that, for h ∈ ]0, h0 [

|ut+h (θ, ϕ) − ut (θ, ϕ)| ≤ −r ≤ σ ≤ 0sup |u(σ + h, ϕ) − u(σ, ϕ)| .

24
Consequently, for t, t + h ∈ [0, Tϕ [ , h ∈ ]0, h0 [ ;

kut+h (., ϕ) − ut (, ϕ)k ≤ δ(h) + M0 eω0 Tϕ (|S0′ (h)ϕ(0) − ϕ(0)| + c1 h)

+M0 eω0 Tϕ C0 (α)t0 kus+h (., ϕ) − us (., ϕ)k ds,

where
δ(h) = −r ≤ σ ≤ 0sup |u(σ + h, ϕ) − u(σ, ϕ)| .
By Gronwall’s Lemma, it follows
£ ¤
kut+h (., ϕ) − ut (, ϕ)k ≤ β(h) exp C0 (α)M0 eω0 Tϕ Tϕ ,

with £¯ ¯ ¤
β(h) = δ(h) + M0 eω0 Tϕ ¯S0′ (h)ϕ(0) − ϕ(0)¯ + c1 h .
Using the same reasoning, one can show the same result for h < 0.
It follows immediately, that

t → Tϕ− limu(t, ϕ) exists.

Consequently, u(., ϕ) can be extended to Tϕ , which contradicts the maxi-

mality of [0, Tϕ [ .
We will prove now that the solution depends continuously on the initial
data. Let ϕ ∈ CX , ϕ(0) ∈ D(A0 ) and t ∈ [0, Tϕ [. We put

α = 1 + −r ≤ s ≤ tsup |u(s, ϕ)|

and ¡ ¢
c(t) = M0 eω0 t exp M0 eω0 t C0 (α)t .
Let ε ∈ ]0, 1[ such that c(t)ε < 1 and ψ ∈ CX , ψ(0) ∈ D(A0 ) such that

kϕ − ψk < ε.

We have
kψk ≤ kϕk + ε < α.
Let
T0 = sup {s > 0 : kuσ (., ψ)k ≤ α for σ ∈ [0, s]} .
If we suppose that T0 < t, we obtain for s ∈ [0, T0 ] ,
Z s
kus (., ϕ) − us (., ψ)k ≤ M0 eω0 t kϕ − ψk+M0 eω0 t C0 (α) kuσ (., ϕ) − uσ (., ψ)k dσ.
0

25
By Gronwall’s Lemma, we deduce that

kus (, ϕ) − us (, ψ)k ≤ c(t) kϕ − ψk . (11)

This implies that

kus (., ψ)k ≤ c(t)ε + α − 1 < α, for all s ∈ [0, T0 ] .

It follows that T0 cannot be the largest number s > 0 such that kuσ (., ψ)k ≤
α, for σ ∈ [0, s] . Thus, T0 ≥ t and t < Tψ . Furthermore, kus (., ψ)k ≤ α,
for s ∈ [0, t], then using the inequality (11) we deduce the dependence
continuous with the initial data. This completes the proof of Theorem

Theorem 25 Assume that F is continuously differentiable and satisfies the

following locally Lipschitz condition: for each α > 0 there exists a constant
C1 (α) > 0 such that if ϕ1 , ϕ2 ∈ CX and kϕ1 k , kϕ2 k ≤ α then

|F (t, ϕ1 ) − F (t, ϕ2 )| ≤ C1 (α) kϕ1 − ϕ2 k ,

|Dt F (t, ϕ1 ) − Dt F (t, ϕ2 )| ≤ C1 (α) kϕ1 − ϕ2 k ,
|Dϕ F (t, ϕ1 ) − Dϕ F (t, ϕ2 )| ≤ C1 (α) kϕ1 − ϕ2 k .
1 := C 1 ([−r, 0] , X) such that
Then, for given ϕ ∈ CX

ϕ(0) ∈ D(A0 ), ϕ′ (0) ∈ D(A0 ) and ϕ′ (0) = Aϕ(0) + F (ϕ),

let u(., ϕ) : [−r, Tϕ [ → X be the unique integral solution of Equation (1).

Then, u(., ϕ) is continuously differentiable on [−r, Tϕ [ and satisfies Equation
(1).

Proof. The proof is similar to the proof of Theorem 22

4 The semigroup and the integrated semigroup in

the autonomous case
Let E be defined by
CX
n o
E := D(A) = ϕ ∈ CX : ϕ(0) ∈ D(A0 ) .

Throughout this section we will suppose the hypothesis of Theorem 18 ex-

cept that we require F to be autonomous, that is, F : CX → X. By virtue

26
of Theorem 18, there exists for each ϕ ∈ E a unique continuous function
y(., ϕ) : [0, +∞) → X satisfying the following equation
Z
d te
y(t) = S ′ (t)ϕ + S(t − s)X0 F (y(s)) ds, for t ≥ 0. (12)
dt 0
Let us consider the operator T (t) : E → E defined, for t ≥ 0 and ϕ ∈ E, by
T (t)(ϕ) = y(t, ϕ).
Then, we have the following result.
Proposition 26 Under the same assumptions as in Theorem 18, the family
(T (t))t≥0 is a nonlinear strongly continuous semigroup of continuous oper-
ators on E, that is
i) T (0) = Id,
ii) T (t + s) = T (t)T (s), for all t, s ≥ 0,
iii) for all ϕ ∈ E, T (t)(ϕ) is a continuous function of t ≥ 0 with values in
E,
iv) for all t ≥ 0, T (t) is continuous from E into E.
Moreover,
v) (T (t))t≥0 satisfies, for t ≥ 0, θ ∈ [−r, 0] and ϕ ∈ E, the translation
property
½
(T (t + θ)(ϕ))(0), if t + θ ≥ 0,
(T (t)(ϕ))(θ) = (13)
ϕ(t + θ), if t + θ ≤ 0,
and
vi) there exists γ > 0 and M ≥ 0 such that,
kT (t)(ϕ1 ) − T (t)(ϕ2 )k ≤ M eγt kϕ1 − ϕ2 k , for all ϕ1 , ϕ2 ∈ E.
Proof. The proof of this proposition is standard.
We recall the following definition.
Definition 27 A C0 −semigroup (T (t))t≥0 on a Banach space (X, |·|) is
called compact if all operators T (t) are compact on X for t > 0.
Lemma 28 Let (S(t))t≥0 a locally Lipschitz continuous integrated semi-
group on a Banach space (X, |·|), (A, D(A)) its generator, B a bounded sub-
set of X, {Gλ , λ ∈ Λ} a set of continuous functions from the finite interval
[0, T ] into B, c > 0 a constant and
Z
d t−c
Pλ (t) = S(t − s)Gλ (s) ds, for c < t ≤ T + c and λ ∈ Λ.
dt 0

27
Assume that S ′ (t) : (D(A), |·|) → (D(A), |·|) is compact for each t > 0.
Then, for each t ∈ (c, T + c], {Pλ (t), λ ∈ Λ} is a precompact subset of (D(A), |·|).

Proof. The proof is similarnto the proof of a fundamental result of Travis

o
and Webb [65]. Let H = S ′ (t)x; t ∈ [c, T + c] , x ∈ D(A), |x| ≤ kN ,
where k is the Lipschitz constant of S(·) on [0, T ] and N is a bound of
B. Using the same reasoning as in the proof of Lemma 2.5 [65], one can
prove that H is precompact in (D(A), |·|). Hence, the convex hull of H is
precompact. On the other hand, we have, from Lemma 17
Z
1 t−c ′
Pλ (t) = h ց 0lim S (t − s)S(h)Gλ (s) ds,
h 0

and, for each h > 0 small enough and t ∈ (c, T + c] fixed, the set
½ ¾
′ 1
S (s) S(h)Gλ (t − s), s ∈ [c, t] , λ ∈ Λ
h
is contained in H. Then, the set
½ Z t ¾
1 ′
S (s)S(h)Gλ (t − s) ds, λ ∈ Λ
h c

is contained in the closed convex hull of (t − c)H. Letting h tend to zero,

the set
{Pλ (t), λ ∈ Λ}
is still a precompact set of D(A). Hence the proof is complete

Theorem 29 Assume that S0′ (t) : (D(A0 ), |·|) → (D(A0 ), |·|) is compact
for each t > 0 and the assumptions of Theorem 18 are satisfied. Then, the
nonlinear semigroup (T (t))t≥0 is compact on (E, k.kCX ) for every t > r.

Proof. Let {ϕλ , λ ∈ Λ} be a bounded subset of E and let t > r. For

θ ∈ [−r, 0], we have t + θ > 0. For each λ ∈ Λ, define yλ by

yλ (t)(θ) = y(t, ϕλ )(θ).

Then, we obtain
R
d t+θ
yλ (t)(θ) = S0′ (t + θ)ϕλ (0) + dt 0 S0 (t + θ − s)F (yλ (s)) ds,
R t+θ
= S0 (t + θ)ϕλ (0) + h ց 0lim h1 0 S0′ (t + θ − s)S0 (h)F (yλ (s)) ds.
′

28
First, we show that the family {yλ (t), λ ∈ Λ} is equicontinuous, for each
t > r. Using the Lipschitz condition on F , one shows that
{F (yλ (s)), s ∈ [0, t] and λ ∈ Λ} is bounded by a constant, say K.
Let λ ∈ Λ, 0 < c < t − r and −r ≤ θb < θ ≤ 0. Observe that
¯ ¯ ¯ ¯
¯ b ¯¯ ≤ ¯¯S ′ (t + θ)ϕλ (0) − S ′ (t + θ)ϕ
b λ (0)¯¯
¯yλ (t)(θ) − yλ (t)(θ)
¯R 0 0
¯
1 ¯ t+θ ′ ¯
+h ց 0lim h ¯ t+θb S0 (t + θ − s)S0 (h)F (yλ (s)) ds¯
¯R b ³ ´ ¯
¯ t+θ b − s) S0 (h)F (yλ (s)) ds¯¯
+h ց 0lim h1 ¯ t+θ−c S ′ (t + θ − s) − S ′ (t + θ
¯R bb ³ 0 0
´ ¯
1 ¯ t+θ−c ¯
+h ց 0lim h ¯ 0 S0 (t + θ − s) − S0′ (t + θb − s) S0 (h)F (yλ (s)) ds¯ ,
′

° ° ¯ ¯
° b° ¯ ¯
≤ °S0′ (t + θ) − S0′ (t + θ)° |ϕλ (0)| + M0 eω0 t kK ¯θ − θb¯ + c2M0 eω0 t kK
h i ° °
° °
+ ts ∈ 0, t + θb − c sup °S0′ (t + θ − s) − S0′ (t + θb − s)° kK.
¯ ¯
¯ ¯
If we take ¯θ − θb¯ small enough, it follows from the uniform continuity of
S0′ (.) : [a, t] → L(D(A0 )) for 0 < a < t, the claimed equicontinuity of
{yλ (t), λ ∈ Λ}.
On the other hand, the family {S0′ (t + θ)ϕλ (0); λ ∈ Λ} is precompact,
for each t > r and θ ∈ [−r, 0]. We will show that, for each t > r and
θ ∈ [−r, 0].
½ Z t+θ ¾
d
S0 (t + θ − s)F (yλ (s)) ds; λ ∈ Λ
dt 0
is precompact.
Observe that for 0 < c < t + θ and λ ∈ Λ
¯Z ¯
1 ¯¯ t+θ ′ ¯
¯ S0 (t + θ − s)S0 (h)F (yλ (s)) ds¯¯ ≤ cM0 eω0 t kK.
h t+θ−c

Then, if h tends to zero we obtain

¯ Z t+θ ¯
¯d ¯
¯ S (t + θ − s)F (y (s)) ds¯ ≤ cM0 eω0 t kK. (14)
¯ dt 0 λ ¯
t+θ−c

For 0 < c < t + θ and λ ∈ Λ,

Z t+θ−c Z t+θ
S0 (t + θ − s)F (yλ (s)) ds = S0 (s)F (yλ (t + θ − s)) ds.
0 c

29
By lemma 28, if 0 < c < t + θ and λ ∈ Λ, then
½ Z t+θ−c ¾
d
S0 (t + θ − s)F (yλ (s)) ds, λ ∈ Λ
dt 0

is precompact in X, for each t > r and θ ∈ [−r, 0]. This fact together with
(14) yields the precompactness of the set
½ Z t+θ ¾
d
S0 (t + θ − s)F (yλ (s)) ds, λ ∈ Λ .
dt 0

Using the Arzela-Ascoli theorem, we obtain the result of Theorem 29

Consider now the linear autonomous functional differential equation
½ ′
u (t) = A0 u(t) + L(ut ), t ≥ 0,
(15)
u0 = ϕ,

where L is a continuous linear functional from CX into X.

Given ϕ ∈ CX . It is clear that there exists a unique function
z := z(., ϕ) : [0, +∞) → CX which solves the following abstract integral
equation
µZ t ¶
d e
z(t) = S(t)ϕ + S(t − s)X0 L(z(s)) ds , for t ≥ 0.
dt 0 (16)

Theorem 30 The family of operators (U (t))t≥0 defined on CX by

U (t)ϕ = z(t, ϕ),

is a locally Lipschitz continuous integrated semigroup on CX generated by

the operator P defined by
 ©
 D(P) = ϕ ∈ C 1 ([−r, 0] , X) ; ϕ(0) ∈ D(A0 ),
ϕ′ (0) = A0 ϕ(0) + L (ϕ)} ,
 ′
Pϕ = ϕ .

Proof. Consider the operator

e : CX → CeX
L

defined by
e (ϕ) = X0 L(ϕ).
L

30
 e defined
Using a result of Kellermann [51], one can prove that the operator G
in CeX by (
D(G)e = D(A), e
e
G = A + L,e e

where Ae is defined in Proposition 15, is the generator of a locally Lipschitz

continuous integrated semigroup on C eX , because Ae satisfies (HY) and Le∈
e
L(CX , CX ).
Let us introduce the part G of G e in CX , which is defined by :
( n o
D(G) = ϕ ∈ D(G);e Gϕ e ∈ CX ,
G (ϕ) = G e (ϕ) .

It is easy to see that

G = P.
Then, P is the generator of a locally Lipschitz continuous integrated semi-
group (V (t))t≥0 on CX .
On the other hand, if we consider, for each ϕ ∈ CX , the nonhomogeneous
Cauchy problem
(
dz e
(t) = Az(t) + h(t), for t ≥ 0,
dt (17)
z(0) = 0,

where h : [0, +∞[ → CeX is given by

e (t)ϕ).
h(t) = ϕ + L(V

By Theorem 13, the nonhomogeneous Cauchy problem (17) has a unique

integral solution z given by
µZ t ¶
d e − s)h(s) ds ,
z(t) = S(t
dt µZ0 ¶ µZ t ¶
t
d e d e
= S(t − s)ϕ ds + S(t − s)X0 L (V (s)ϕ) ds .
dt 0 dt 0

Then, µZ ¶
t
d e − s)X0 L (V (s)ϕ) ds .
z(t) = S(t)ϕ + S(t
dt 0

31
On the other hand, Proposition 6 gives
µZ t ¶
V (t)ϕ = P V (s)ϕ ds + tϕ.
0

Moreover, for ψ ∈ D(P ), we have

e + X0 L (ψ) .
P ψ = Aψ

Then, we obtain
µZ t ¶ µZ t ¶
e
V (t)ϕ = A V (s)ϕ ds + X0 L V (s)ϕ ds + tϕ.
0 0

So, µZ ¶ Z
t t
e
V (t)ϕ = A V (s)ϕ ds + h(s) ds.
0 0
Hence, the function t → V (t)ϕ is an integral solution of Equation (17). By
uniqueness, we conclude that V (t)ϕ = z(t), for all t ≥ 0. Then we have
U (t) = V (t) on CX . Thus the proof of Theorem 30
Let B be the part of the operator P on E. Then,
 n

 D(B) = ϕ ∈ C 1 ([−r, 0] , X) ; ϕ(0) ∈ D(A0 ), ϕ′ (0) ∈ D(A0 )
 ϕ′ (0) = A0 ϕ(0) + L (ϕ)} ,

Bϕ = ϕ′ .

Corollary 31 B is the infinitesimal generator of the C0 −semigroup (U ′ (t))t≥0

on E.

Proof. See [64]

Corollary 32 Under the same assumptions as in Theorem 29, the linear

C0 −semigroup (U ′ (t))t≥0 is compact on E, for every t > r.

To define a fundamental integral solution Z(t) associated to Equation

e ∈ CeX , the integral equation
(15), we consider, for ϕ
µZ t ¶
e d e
z(t) = S(t)ϕe+ S(t − s)X0 L(z(s)) ds , for t ≥ 0.
dt 0 (18)

One can show the following result.

32
Proposition 33 Given ϕ e ∈ CeX , the abstract integral equation (18) has a
unique solution z := z(., ϕ)
e which is a continuous
³ ´ mapping from [0, +∞) →
e
CX . Moreover, the family of operators U (t) defined on CeX by
t≥0

e (t)ϕ
U e = z(t, ϕ)
e

is a locally Lipschitz continuous integrated semigroup on CeX generated by

the operator Ge defined by
(
e = D(A),
D(G) e
e e
Gϕ = A + X0 L(ϕ).

For each complex number λ, we define the linear operator

∆(λ) : D(A0 ) → X

by ³ ´
∆(λ)x := λx − A0 x − L eλ· x , x ∈ D(A0 ).

where eλ· x : [−r, 0] → CX , is defined for x ∈ X by (note that we consider

here the comlexification of CX )
³ ´
eλ· x (θ) = eλθ x, θ ∈ [−r, 0] .

We will call λ a characteristic value of Equation (15) if there exists x ∈

D(A0 )\{0} solving the characteristic equation ∆(λ)x = 0. The multiplicity
of a characteristic value λ of Equation (15) is defined as dim Ker∆(λ).
We have the following result.

Corollary 34 There exits ω ∈ IR, such that for λ > ω and c ∈ X, one has
e −1 (X0 c) = eλ· ∆(λ)−1 c.
(λI − G)

Proof. We have, for λ > ω0

³ ´
∆(λ) = (λI − A0 ) I − (λI − A0 )−1 L(eλ· I) .

Let ω > max(0, ω0 + M0 kLk) and

Kλ = (λI − A0 )−1 L(eλ· I).

33
Then,
M0 kLk
kKλ k ≤ < 1, for λ > ω.
λ − ω0
Hence the operator ∆(λ) is invertible for λ > ω.
Consider the equation
e λ· a) = X0 c,
(λI − G)(e

where c ∈ X is given and we are looking for a ∈ D(A0 ). This yields

³ ³ ´´
λeλ· c − λeλ· c + X0 λc − A0 c − L eλ· c = X0 a.

Then, for λ > ω

c = ∆(λ)−1 a.
Consequently,
³ ´−1
e
λI − G (X0 c) = eλ· ∆(λ)−1 c

Corollary 35 For each c ∈ X, the function U e (.)(X0 c) satisfies, for t ≥ 0

and θ ∈ [−r, 0], the following translation property
( ³ ´
³ ´ Ue (t + θ)(X0 c) (0), t + θ ≥ 0,
e (t)(X0 c) (θ) =
U
0, t + θ ≤ 0.

We can consider the following linear operator

Z(t) : X → X

defined, for t ≥ 0 and c ∈ X, by

e (t)(X0 c)(0)
Z(t)c = U

Corollary 36 Z(t) is the fundamental integral solution of Equation (15);

that is Z +∞
∆(λ)−1 = λ e−λt Z(t) dt, for λ > ω.
0

Proof. We have, for c ∈ X

³ ´−1
e
λI − G (X0 c) = eλ· ∆(λ)−1 c.

34
Then
Z +∞ ³ ´ Z +∞
λθ
e ∆(λ) −1
c=λ −λs
e e
U (s)(X0 c) (θ) ds = λ e−λs Z(s + θ)c ds.
0 −θ

So, Z +∞
−1
∆(λ) c=λ e−λt Z(t)c dt
0
It is easy to prove the following result.

Corollary 37 If c ∈ D(A0 ) then the function

t → Z(t)c

is differentiable for all t > 0 and we have

Z +∞
−1
∆(λ) c = e−λt Z ′ (t)c dt.
0

Hence the name of fundamental integral solution.

The fundamental solution Z ′ (t) is defined only for c ∈ D(A0 ) and is

discontinuous at zero. For these reasons we use the fundamental integral
e (.)(X0 ).
solution Z(t) or equivalently U
We construct now a variation-of-constants formula for the linear nonho-
mogeneous system

u′ (t) = A0 u(t) + L(ut ) + f (t), t ≥ 0, (19)

or its integrated form

µZ t ¶
′ d e
y(t) = S (t)ϕ + S(t − s)X0 L [(y(s)) + f (s)] ds , for t ≥ 0,
dt 0 (20)

where f : [0, +∞) → X is a continuous function.

Theorem 38 For any ϕ ∈ CX , such that ϕ(0) ∈ D(A0 ), the function

y : [0, +∞) → X defined by
Z
′ d te
y(t) = U (t)ϕ + U (t − s)X0 f (s) ds, (21)
dt 0
satisfies Equation (20).

35
Proof. It follows immediately from Theorem 18 and Corollary 19 that, for
ϕ ∈ CX and ϕ(0) ∈ D(A0 ), Equation (20) has a unique solution y which is
the integral solution of the equation
½ ′
e
y (t) = Ay(t) + X0 [L(y(t)) + f (t)] , t ≥ 0,
y(0) = ϕ.

On the other hand, D(A)e = D(G) e and Ge = A+X

e 0 L. Then, y is the integral
solution of the equation
½ ′
e
y (t) = Gy(t) + X0 f (t), t ≥ 0,
y(0) = ϕ.

Moreover, we have
n o
e = D(P ) = ϕ ∈ CX ,
D(G) ϕ(0) ∈ D(A0 ) = E,

e in CX is the operator P . Then t → U (.)ϕ is differentiable

and the part of G
in t ≥ 0 and Theorem 13 implies
Z
d te
y(t) = U ′ (t)ϕ + U (t − s)X0 f (s) ds
dt 0

5 Principle of linearized stability

In this section, we give a result of linearized stability near an equilibrium
point of Equation (1) in the autonomous case, that is
(
du
(t) = A0 u(t) + F (ut ), t ≥ 0,
dt (22)
u0 = ϕ.

We make the following hypothesis :

F is continuously differentiable, F (0) = 0 and F is satisfies the Lipschitz
condition

|F (ϕ1 ) − F (ϕ2 )| ≤ L kϕ1 − ϕ2 k , for all ϕ1 , ϕ2 ∈ CX ,

where L is a positive constant.

Let T (t) : E → E, for t ≥ 0, be defined by

T (t)(ϕ) = ut (., ϕ),

36
where u(., ϕ) is the unique integral solution of Equation (22) and
n o
E = ϕ ∈ CX , ϕ(0) ∈ D(A0 ) .

We know that the family (T (t))t≥0 is a nonlinear strongly continuous semi-

groups of continuous operators on E.
Consider the linearized equation of (22) corresponding to the Fréchet-
derivative Dϕ F (0) = F ′ (0) at 0 :
(
du
(t) = A0 u(t) + F ′ (0)ut , t ≥ 0,
dt (23)
u0 = ϕ ∈ CX ,

and let (U ′ (t))t≥0 be the corresponding linear C0 −semigroup on E.

Proposition 39 The Fréchet-derivative at zero of the nonlinear semigroup

T (t), t ≥ 0, associated to Equation (22), is the linear semigroup U ′ (t), t ≥ 0,
associated to Equation (23).

Proof. It suffices to show that, for each ε > 0, there exists δ > 0 such that
° °
°T (t)(ϕ) − U ′ (t)ϕ° ≤ ε kϕk , for kϕk ≤ δ.

We have
kT (t)(ϕ) − U ′ (t)ϕk = θ ∈ [−r, 0]sup |(T (t)(ϕ)) (θ) − (U ′ (t)ϕ) (θ)| ,
= t + θ ≥ 0θ ∈ [−r, 0]sup |(T (t)(ϕ)) (θ) − (U ′ (t)ϕ) (θ)| ,

and, for t + θ ≥ 0
Z t+θ
¡ ¢
′ d ¡ ¢
(T (t)(ϕ)) (θ)− U (t)ϕ (θ) = S0 (t+θ−s) F (T (s)(ϕ)) − F ′ (0)(U ′ (s)ϕ) ds.
dt 0

It follows that
Z
° ° t ¯ ¯
°T (t)(ϕ) − U ′ (t)ϕ° ≤ M0 eω0 t e−ω0 s ¯F (T (s)(ϕ)) − F ′ (0)(U ′ (s)ϕ)¯ ds
0

and
¡
kT (t)(ϕ) − U ′ (t)ϕk ≤ M0 eω0 t t0 e−ω0 s |F (T (s)(ϕ)) − F (U ′ (s)ϕ))|
¢ ds
t
+0 e−ω 0 s ′ ′ ′
|F (U (s)ϕ) − F (0)(U (s)ϕ)| ds .

37
By virtue of the continuous differentiability of F at 0, we deduce that for
ε > 0, there exists δ > 0 such that
Z t
¯ ¯
e−ω0 s ¯F (U ′ (s)ϕ) − F ′ (0)(U ′ (s)ϕ)¯ ds ≤ ε kϕk , for kϕk ≤ δ.
0

On the other hand, we obtain

Z t Z t
¯
−ω0 s ¯
¯ ° °
e ′ ¯
F (T (s)(ϕ)) − F (U (s)ϕ) ds ≤ L e−ω0 s °T (s)(ϕ) − U ′ (s)ϕ° ds.
0 0

Consequently,
µ Z t ¶
° ° ° °
°T (t)(ϕ) − U ′ (t)ϕ° ≤ M0 eω0 t ε kϕk + L e−ω0 s ° ′ °
T (s)(ϕ) − U (s)ϕ ds .
0

By Gronwall’s Lemma, we obtain

° °
°T (t)(ϕ) − U ′ (t)ϕ° ≤ M0 ε kϕk e(LM0 +ω0 )t .

We conclude that T (t) is differentiable at 0 and Dϕ T (t)(0) = U ′ (t), for each

t≥0

Definition 40 Let Y be a Banach space, (V (t))t≥0 a strongly continuous

semigroup of operators V (t) : W ⊆ Y → W, t ≥ 0, and x0 ∈ W an equilib-
rium of (V (t))t≥0 (i.e., V (t)x0 = x0 , for all t ≥ 0). The equilibrium x0 is
called exponentially asymptotically stable if there exist δ > 0, µ > 0, k ≥ 1
such that

kV (t)x − x0 k ≤ ke−µt kx − x0 k for all x ∈ W with kx − x0 k ≤ δ and all t ≥ 0.

We have the following result.

Theorem 41 Suppose that the zero equilibrium of (U ′ (t))t≥0 is exponen-

tially asymptotically stable, then zero is exponentially asymptotically stable
equilibrium of (T (t))t≥0 .

The proof of this theorem is based on the following result.

Theorem 42 (Desh and Schappacher [41]) Let (V (t))t≥0 be a nonlinear

strongly continuous semigroup of type γ on a subset W of a Banach space
Y , i.e.

kV (t)x1 − V (t)x2 k ≤ M ′ eγt kx1 − x2 k , for all x1 , x2 ∈ W

38
and assume that x0 ∈ W is an equilibrium of (V (t))t≥0 such that V (t) is
Fréchet-differentiable at x0 for each t ≥ 0, with Y (t) the Fréchet-derivative
at x0 of V (t), t ≥ 0. Then, (Y (t))t≥0 is a strongly continuous semigroup of
bounded linear operators on Y. If the zero equilibrium of (Y (t))t≥0 is expo-
nentially asymptotically stable, then x0 is an exponentially asymptotically
stable equilibrium of (V (t))t≥0 .

6 Spectral Decomposition
In this part, we show in the linear autonomous case (Equation (15)), the
existence of a direct sum decomposition of the state space
n o
E = ϕ ∈ CX , ϕ(0) ∈ D(A0 )

into three subspaces : stable, unstable and center, which are semigroup
invariants. We assume that the semigroup (T0 (t))t≥0 on D(A0 ) is compact.
It follows from the compactness property of the semigroup U ′ (t), for t > r,
on E, the following results.

Corollary 43 [60] For each t > r, the spectrum σ(U ′ (t)) is a countable set
and is compact with the only possible accumulation point 0 and if µ 6= 0 ∈
σ(U ′ (t)) then µ ∈ P σ(U ′ (t)), (where P σ(U ′ (t)) denotes the point spectrum).

Corollary 44 [60] There exists a real number δ such that Re λ ≤ δ for all
λ ∈ σ(B). Moreover, if β is a given real number then there exists only a
finite number of λ ∈ P σ(B) such that Re λ > β.

We can give now an exponential estimate for the semigroup solution.

Proposition 45 Assume that δ is a real number such that Re λ ≤ δ for

all characteristic values of Equation (15). Then, for γ > 0 there exists a
constant k(γ) ≥ 1 such that
° ′ °
°U (t)ϕ° ≤ k(γ)e(δ+γ)t kϕk , for all t ≥ 0, ϕ ∈ E.

Proof. Let ω1 be defined by

© ¡ ° °¢ ª
ω1 := inf ω ∈ lR : t ≥ 0sup e−ωt °U ′ (t)° < +∞ .

The compactness property of the semigroup (see [55]) implies that

ω1 = s1 (B) := sup {Re λ : λ ∈ P σ(B)} .

39
On the other hand, if λ ∈ P σ(B) then there exists ϕ 6= 0 ∈ D(B) such that
Bϕ = λϕ. This implies that

ϕ(θ) = eλt ϕ(0) and ϕ′ (0) = A0 ϕ(0) − L(ϕ) with ϕ(0) 6= 0.

Then, ∆(λ)ϕ(0) = 0. We deduce that λ is a characteristic value of Equation

(15).
We will prove now the existence of λ ∈ P σ(B) such that Re λ = s1 (B).
Let (λn )n be a sequence in P σ(B) such that Re λn → s1 (B) as n → +∞.
Then, there exists β such that Re λn > β for n ≥ n0 with n0 large enough.
From Corollary 44, we deduce that {λn : Re λn > β} is finite. So, the
sequence (Re λn )n is stationary. Consequently, there exists n such that
Re λn = s1 (B). This completes the proof of Proposition 45
The asymptotic behavior of solutions can be now completely obtained
by the characteristic equation.

Theorem 46 Let δ be the smallest real number such that if λ is a char-

acteristic value of Equation (15), then Re λ ≤ δ. If δ < 0, then for all
ϕ ∈ E, kU ′ (t)ϕk → 0 as t → +∞. If δ = 0 then there exists ϕ ∈ E\{0}
such that kU ′ (t)ϕk = kϕk for all t ≥ 0. If δ > 0, then there exists ϕ ∈ E
such that kU ′ (t)ϕk → +∞ as t → +∞.

Proof. Assume that δ < 0, then we have ω1 = s1 (B) < 0 and the stability
holds. If δ = 0, then there exists x 6= 0 and a complex λ such that Re λ = 0
and ∆(λ)x = 0. Then, λ ∈ P σ(B) and eλt ∈ P σ(U ′ (t)). Consequently, there
exists ϕ 6= 0 such that
U ′ (t)ϕ = eλt ϕ.
° °
This implies that kU ′ (t)ϕk = °eλt ϕ° = kϕk . Assume now that δ > 0. Then,
there exists x 6= 0 and a complex λ such that Re λ = δ and ∆(λ)x = 0. Then,
there exists ϕ 6= 0 such that kU ′ (t)ϕk = eδt kϕk → +∞, as t → +∞. This
completes the proof of Theorem 46
Using the same argument as in [67], [section 3.3, Theorem 3.1], we obtain
the following result.

Theorem 47 Suppose that X is complex. Then, there exist three linear

subspaces of E denoted by S, U S and CN , respectively, such that

E = S ⊕ U S ⊕ CN

and
(i) B(S) ⊂ S, B(U S) ⊂ U S and B(CN ) ⊂ CN ;

40
(ii) U S and CN are finite dimensional;
(iii) P σ(B |U S ) = {λ ∈ P σ(B) : Re λ > 0} , P σ(B |CN ) = {λ ∈ P σ(B) : Re λ = 0} ;
(iv) U ′ (t) (U S) ⊂ U S, U ′ (t) (CN ) ⊂ CN, for t ∈ lR, U ′ (t) (S) ⊂ S, for
t ≥ 0;
(v) for any 0 < γ < α := inf {|Re λ| : λ ∈ P σ(B) and Re λ 6= 0} , there ex-
ists M = M (γ) > 0 such that

kU ′ (t)PU S ϕk ≤ M eγt kPU S ϕk , t ≤ 0,

γ
kU ′ (t)PCN ϕk ≤ M e 3 t kPCN ϕk , t ∈ lR,
kU ′ (t)PS ϕk ≤ M e−γt kPS ϕk , t ≥ 0,

where PS , PU S and PCN are projections of E into S, U S and CN respec-

tively. S, U S and CN are called stale, unstable and center subspaces of the
semigroup (U ′ (t))t≥0 .

7 Existence of bounded solutions

We reconsider now the equation (19) mentioned in the introduction. For the
convenience of the reader, we restate this equation
(
du
(t) = A0 u(t) + L(ut ) + f (t), t ≥ 0,
dt
u0 = ϕ ∈ CX ,

and its integrated form

Z t
d e (t − s)X0 f (s)ds,
ut = U ′ (t)ϕ + U for t ≥ 0. (24)
dt 0

where f is a continuous function form lR into X.

Thanks to Lemma 17, we will use the following integrated form of Equa-
tion (19), which is equivalent to (24) :
Z t
′
ut = U (t)ϕ + λ → +∞lim fλ X0 f (s)ds, for t ≥ 0,
U ′ (t − s)B
0 (25)

fλ : hX0 i → CX is defined by
where the operator B
³ ´−1
fλ X0 c = λ λI − G
B e (X0 c) = λeλ· ∆(λ)−1 c, c ∈ X.

We need the following definition.

41
Definition 48 We say that the semigroup (U ′ (t))t≥0 is hyperbolic if

σ(B) ∩ ilR = ∅.

Theorem 47 implies that, in the hyperbolic case the center subspace CN

is reduced to zero. Thus, we have the following result.

Corollary 49 If the semigroup (U ′ (t))t≥0 is hyperbolic, then the space E is

decomposed as
E = S ⊕ US
and there exist positive constants M and γ such that

kU ′ (t)ϕk ≤ M e−γt kϕk , t ≥ 0, ϕ ∈ S,

kU ′ (t)ϕk ≤ M eγt kϕk , t ≤ 0, ϕ ∈ U S.

We give now the first main result of this section.

Theorem 50 Assume that the semigroup (U ′ (t))t≥0 is hyperbolic. Let B

represent B(lR− ), B(lR+ ) or B(lR), the set of bounded continuous functions
from lR− , lR+ or lR respectively to X. Let π : B → B be a projection onto
the integral solutions of Equation (15) (for any ϕ ∈ E) which are in B.
Then, for any f ∈ B, there is a unique solution Kf ∈ B of Equation (25)
(for some ϕ ∈ E) such that πKf = 0 and K : B → B is a continuous linear
operator. Moreover,
(i) for B = B(lR− ), we have
©
π(B) = x : lR− → X, there exists ϕ ∈ U S
such that x(t) = (U ′ (t)ϕ) (0), t ≤ 0}

and
Rt ³ ´S
(Kf )t = s → −∞limλ → +∞lim s U ′ (t − τ ) B fλ X0 f (τ ) dτ +
Rt ³ ´U S
λ → +∞lim 0 U ′ (t − τ ) Bfλ X0 f (τ ) dτ.

(ii) For B = B(lR+ ), we have

©
π(B) = x : lR+ → X, there exists ϕ ∈ S
such that x(t) = (U ′ (t)ϕ) (0), t ≥ 0}

42
and
Rt ³ ´S
(Kf )t = λ → +∞lim U ′ (t − τ ) Bfλ X0 f (τ ) dτ +
0
Rt ³ ´U S
s → +∞limλ → +∞lim s U ′ (t − τ ) B fλ X0 f (τ ) dτ.

(iii) For B = B(lR), we have

π(B) = {0}

and
Rt ³ ´S
(Kf )t = s → −∞limλ → +∞lim U ′ (t − τ ) Bfλ X0 f (τ ) dτ +
s
Rt ′ ³ ´U S
s → +∞limλ → +∞lim s U (t − τ ) B fλ X0 f (τ ) dτ.

Proof. Theorem 47 implies that the space U S is finite dimensional and

′ S) ⊆ U S. Then,
© (t) (U
U ª
x : lR− → X, there exists ϕ ∈ U S such that x(t) = (U ′ (t)ϕ) (0), t ≤ 0
⊆ π(B(lR− )). Conversely, let ϕ ∈ S and u(., ϕ) be the integral solution
of Equation (15) in S, which is bounded on lR− . Assume that there is a
t ∈ (−∞, 0] such that ut (., ϕ) 6= 0. Then, for any s ∈ (−∞, t), we have

ut (., ϕ) = U ′ (t − s)us (., ϕ).

Thanks to Corollary 49 we have

kut (., ϕ)k ≤ M e−γ(t−s) kus (., ϕ)k , s ≤ t.

Since us (., ϕ) is bounded, we deduce that us (., ϕ) = 0. Therefore,

©
π(B(lR− )) ⊆ x : lR− → X, there exists ϕ ∈ U S
such that x(t) = (U ′ (t)ϕ) (0), t ≤ 0} .

In the same manner, one can prove the same relations for B(lR+ ) and B(lR).
Let f ∈ B(lR− ) and u = u(., ϕ, f ) be a solution of Equation (25) in
B(lR− ), with initial value ϕ ∈ E. Then, the function u can be decomposed
as
ut = uU
t
S
+ uSt ,
where uUt
S ∈ U S and uS ∈ S are given by
t
Z t ³ ´U S
uU S
= U ′
(t − s)uUS
+ λ → +∞lim U ′
(t − τ ) Bfλ X0 f (τ ) dτ, for t, s ∈ lR,
t s
s (26)

43
 Z t ³ ´S
uSt ′
= U (t − s)uSs + λ → +∞lim U ′ (t − τ ) Bfλ X0 f (τ ) dτ, for s ≤ t ≤ 0,
s (27)

since U ′ (t) is defined on U S for all t ∈ lR. By Corollary 49, we deduce that
Z t ³ ´S
S
ut = s → −∞limλ → +∞lim U ′ (t − τ ) Bfλ X0 f (τ ) dτ, for t ≤ 0.
s (28)

By Lemma 16, we get B fλ X0 = λeλ· ∆−1 (λ). Thus,

° ³ ´S °
° ′ °
°U (t − τ ) Bfλ X0 f (τ ) ° ≤ M e−γ(t−τ ) M λ τ ∈ (−∞, 0]sup kf (τ )k .
° ° λ − ω1

Consequently, we get
° S° MM
°ut ° ≤ τ ∈ (−∞, 0]sup kf (τ )k , t ≤ 0. (29)
γ
We have proved that
Rt ³ ´U S
ut = U ′ (t)ϕU S + λ → +∞lim U ′ (t − τ ) Bfλ X0 f (τ ) dτ +
0
Rt ³ ´S
s → −∞limλ → +∞lim s U ′ (t − τ ) Bfλ X0 f (τ ) dτ, for t ≤(30)
0.

We obtain also, for t ≤ 0, the following estimate

° ³ ´U S °
° ′ R °
°U (t)ϕU S + λ → +∞lim t U ′ (t − τ ) B fλ X0 f (τ ) dτ °≤
° 0 °
° US° MM
M e °ϕ ° + γ τ ∈ (−∞, 0]sup kf (τ )k .
γt
(31)

Conversely, we can verify that the expression (30) is a solution of Equation

(25) in B(lR− ) satisfying the estimates (29) and (31) for every ϕ ∈ E.
Let u = u(., ϕU S , f ) be defined by (30) and let K : B(lR− ) → B(lR− ) be
defined by Kf = (I − π) u(., 0, f ). We can easily verify that

 ut (., ϕU S , 0) = U ′ (t)ϕU S ,
(I − π) u(., ϕU S , 0) = 0,

u(., ϕU S , f ) = u(., ϕU S , 0) + u(., 0, f ).

Therefore, K is a continuous linear operator on B(lR− ), Kf satisfies Equation

(25) for every f ∈ B(lR− ), πK = 0 and K is explicitly given by Theorem 50.

44
 For B(lR+ ) and B(lR) the proofs are similar. This completes the proof
of Theorem 50
We consider now the nonlinear equation
du
(t) = A0 u(t) + L(ut ) + g(t, ut ), (32)
dt
and its integrated form
Z t
′
ut = U (t)ϕ + λ → +∞lim fλ X0 g(s, us )ds,
U ′ (t − s)B (33)
0

where g is continuous from lR × CX into X. We will assume that

(H1) g(t, 0) = 0 for t ∈ lR and there exists a nondecreasing function α :
[0, +∞) → [0, +∞) with h → 0limα(h) = 0 and
kg(t, ϕ1 ) − g(t, ϕ2 )k ≤ α(h) kϕ1 − ϕ2 k for ϕ1 , ϕ2 ∈ E, kϕ1 k , kϕ2 k ≤ h and
t ∈ lR.

Proposition 51 Assume that the semigroup (U ′ (t))t≥0 is hyperbolic and the

assumption (H1) holds. Then, there exists h > 0 and ε ∈ ]0, h[ such that
for any ϕ ∈ S with kϕk ≤ ε, Equation (33) has a unique bounded solution
u : [−r, +∞) → X with kut k ≤ h for t ≥ 0 and uS0 = ϕ.

Proof. Let ϕ ∈ S. By Theorem 50, it suffices to establish the existence of

a bounded solution u : [−r, +∞) → X of the following equation
Rt ³ ´S
ut = U ′ (t)ϕ + λ → +∞lim U ′ (t − τ ) Bfλ X0 g(τ, uτ ) dτ +
0
Rt ³ ´U S
s → +∞limλ → +∞lim s U ′ (t − τ ) B fλ X0 g(τ, uτ ) dτ.

Let (u(n) )n∈N be a sequence of continuous functions from [−r, +∞) to X,

defined by
(0)
ut = U ′ (t)ϕ
Rt ³ ´S
(n+1) fλ X0 g(τ, u(n)
ut = U ′ (t)ϕ + λ → +∞lim
0 U ′ (t − τ ) B
τ ) dτ +
Rt ′ ³ ´U S
s → +∞limλ → +∞lim s U (t − τ ) B fλ X0 g(τ, u(n)
τ ) dτ.
³ ´
(n) S
It is clear that u0 = ϕ. Moreover, we can choose h > 0 and ε ∈ ]0, h[
° °
° (n) °
small enough such that, if kϕk ≤ ε then °ut ° < h for t ≥ 0.

45
 On the other hand, we have
° ° Rt ° °
° (n+1) (n) ° ° (n) (n−1) °
°ut − ut ° ≤ 0 M M e−γ(t−τ ) α(h) °uτ − uτ ° dτ +
R +∞ ° °
° (n) (n−1) °
t M M eγ(t−τ ) α(h) °uτ − uτ ° dτ.

So, by induction we get

° ° µ ¶n
° (n+1) (n) ° 2M M α(h)
°ut − ut ° ≤ 2h , t ≥ 0.
γ
We choose h > 0 such that
2M M α(h) 1
< .
γ 2

Consequently, the limit u := n → +∞limu(n) exists uniformly on [−r, +∞)

and u is continuous on [−r, +∞). Moreover, kut k < h for t ≥ 0 and uS0 = ϕ.
In order to prove that u is a solution of Equation (33), we introduce the
following notation
° ³ ´S
° R
′ (t)ϕ − λ → +∞lim t U ′ (t − τ ) B
v(t) = ° u − U f X g(τ, u ) dτ
° t 0 λ 0 τ
Rt ′ ³ ´U S ° °
f
−s → +∞limλ → +∞lim s U (t − τ ) Bλ X0 g(τ, uτ ) dτ °
°.

We obtain
° °
° (n+1) °
v(t) ≤ °ut − ut °
° µ³ ´S ³ ´ ¶ °
° Rt ′ (n) S °
° f f
+ °λ → +∞lim 0 U (t − τ ) Bλ X0 g(τ, uτ ) − Bλ X0 g(τ, uτ ) dτ °
°
° µ³ ´U S ³ ´ ¶ °
° Rt ′ (n) U S °
° f
+ °s → +∞limλ → +∞lim s U (t − τ ) Bλ X0 g(τ, uτ ) f
− Bλ X0 g(τ, uτ ) dτ °
°.

Moreover, we have
+∞ X ³ ´
(n+1) (k+1) (k)
ut − ut =k = n + 1 ut − ut .

It follows that
· ¸ +∞ X µ ¶k
2M M α(h) 2M M α(h)
v(t) ≤ 2h 1 + k =n+1 .
γ γ

46
Consequently, v = 0 on [0, +∞).
To show the uniqueness suppose that w is also a solution of Equation
(33) with kwt k < h for t ≥ 0. Then,
° °
° (n+1) °
°wt − ut °≤
° µ³ ´S ³ ´ ¶ °
° R (n) S °
°λ → +∞lim t U ′ (t − τ ) f f
Bλ X0 g(τ, wτ ) − Bλ X0 g(τ, uτ ) dτ °
° 0 °
° µ³ ´ ³ ´ ¶ °
° Rt ′ US (n) U S °
° f
+ °s → +∞limλ → +∞lim s U (t − τ ) Bλ X0 g(τ, wτ ) f
− Bλ X0 g(τ, uτ ) dτ °
°.

This implies
° ° µ ¶n
° (n+1) ° 2M M α(h)
°wt − ut ° ≤ 2h → 0, as n → +∞.
γ
This proves the uniqueness and completes the proof

8 Existence of periodic or almost periodic solu-

tions
In this section, we are concerned with the existence of periodic (or almost
periodic) solutions of Equations (25) and (33).
As a consequence of the existence of a bounded solution we obtain the
following result.
Corollary 52 Assume that the semigroup (U ′ (t))t≥0 is hyperbolic. If the
function f is ω−periodic, then the bounded solution of Equation (25) given
by Theorem 50 is also ω−periodic.
Proof. Let u be the unique bounded solution of Equation (25). The func-
tion u(. + ω) is also a bounded solution of Equation (25). The uniqueness
property implies that u = u(.+ω). This completes the proof of the corollary

We are concerned now with the existence of almost periodic solution of

Equation (25). We first recall a definition.
Definition 53 A function g ∈ B(lR) is said to be almost periodic if and
only if the set
{gσ : σ ∈ lR} ,
where gσ is defined by gσ (t) = g(t + σ), for t ∈ lR, is relatively compact in
B(lR).

47
Theorem 54 Assume that the semigroup (U ′ (t))t≥0 is hyperbolic. If the
function f is almost periodic, then the bounded solution of Equation (25) is
also almost periodic.

Proof. Let AP (lR) be the Banach space of almost periodic functions from
lR to X endowed with the uniform norm topology. Define the operator Q
by
Rt ³ ´S
(Qf ) (t) = s → −∞limλ → +∞lim s U ′ (t − τ ) Bfλ X0 f (τ ) (0) dτ
Rt ³ ´U S
+s → +∞limλ → +∞lim s U ′ (t − τ ) B fλ X0 f (τ ) (0) dτ , for t ∈ lR.

Then, Q is a bounded linear operator from AP (lR) into B(lR). By a sample

computation, we obtain

(Qf )σ = Q (fσ ) , for σ ∈ lR.

By the continuity of the operator Q, we deduce that Q({fσ : σ ∈ lR}) is

relatively compact in B(lR). This implies that if the function f is almost
periodic, then the bounded solution of Equation (25) is also almost periodic.
This completes the proof of the theorem.
We are concerned now with the existence of almost periodic solutions of
Equation (33). We will assume the followings.
(H2) g is almost periodic in t uniformly in any compact set of CX . This
means that for each ε > 0 and any compact set K of CX there exists lε > 0
such that every interval of length lε contains a number τ with the property
that
t ∈ lRlR, ϕ ∈ Ksup kg(t + τ, ϕ) − g(t, ϕ)k < ε.
We know from [39] that if the function g is almost periodic in t uniformly
in any compact set of CX and if v is an almost periodic function, then the
function t 7→ g(t, vt ) is also almost periodic.
(H3) kg(t, ϕ1 ) − g(t, ϕ2 )k ≤ K1 kϕ1 − ϕ2 k , t ∈ lR, ϕ1 , ϕ2 ∈ CX .
We have the following result.

Proposition 55 Assume that the semigroup (U ′ (t))t≥0 is hyperbolic and

(H2), (H3) hold. Then, if in the assumption (H3) K1 is chosen small
enough, Equation (33) has a unique almost periodic solution.

Proof. Consider the operator H defined on AP (lR) by

Hv = u,

48
where u is the unique almost periodic solution of Equation (25) with f =
g(., v. ). We can see that there exists a positive constant K2 such that

kHv1 − Hv2 k ≤ K1 K2 kv1 − v2 k , v1 , v2 ∈ AP (lR).

If K1 is chosen such that K1 K2 < 1, then the map H is a strict contrac-

tion in AP (lR). So, H has a unique fixed point in AP (lR). This gives an
almost periodic solution of Equation (33). This completes the proof of the
proposition

9 Applications
We now consider the two examples mentioned in the introduction. For the
convenience of the reader, we restate the equations.
Example 1

 ∂u ∂u
 (t, a) + (t, a) = f (t, a, ut (·, a)), t ∈ [0, T ] , a ∈ [0, l] ,
∂t ∂a
 u(t, 0) = 0, t ∈ [0, T ] ,

u(θ, a) = ϕ(θ, a), θ ∈ [−r, 0] , a ∈ [0, l] . (34)

where T, l > 0, ϕ ∈ CX := C ([−r, 0] , X) and X = C ([0, l] , IR) .

By setting U (t) = u(t, ·), Equation (34) reads
½ ′
V (t) = A0 V (t) + F (t, Vt ), t ∈ [0, T ] ,
V (0) = ϕ,

where A0 : D(A0 ) ⊆ X → X is the linear operator defined by

½ © ª
D(A0 ) = u ∈ C 1 ([0, l] , IR) ; u(0) = 0 ,
A0 u = −u′ ,

and F : [0, T ] × CX → X is the function defined by

F (t, ϕ)(a) = f (t, a, ϕ(·, a)), for t ∈ [0, T ] , ϕ ∈ CX and a ∈ [0, l] .

We have D(A0 ) = {u ∈ C([0, l] , IR); u(0) = 0} 6= X. Moreover,

(
° 0 ) = C, °
ρ(A
° −1 °
°(λI − A0 ) ° ≤ λ1 , for λ > 0,

this implies that A0 satisfies (HY) on X (with M0 = 1 and ω0 = 0).

We have the following result.

49
Theorem 56 Assume that F is continuous on [0, T ] × CX and satisfies a
Lipschitz condition

|F (t, ϕ) − F (t, ψ)| ≤ L kϕ − ψk , t ∈ [0, T ] and ϕ, ψ ∈ CX ,

with L ≥ 0 constant. Then, for a given ϕ ∈ CX , such that

ϕ(0, 0) = 0,

there exists a unique function u : [0, T ] → X solution of the following initial

value problem
 Ra
 0 f (τ − a + Rt , τ , uτ −a+t (·, τ )) dτ, if a ≤ t,
a
u(t, a) = ϕ(0, a − t) + a−t f (τ − a + t , τ , uτ −a+t (·, τ )) dτ, if a ≥ t ≥ 0,

ϕ(t, a), if t ≤ 0. (35)

Furthermore, u is the unique integral solution of the partial differential equa-

tion (34), i.e.
Rt Rt
(i) u ∈ CX , 0 u(s, ·)ds ∈ C 1 ([0, l] , IR) and 0 u(s, 0)ds = 0, for t ∈ [0, T ] ,
(ii)
 Z t Z t
 ∂
ϕ(0, a) + u(s, a) ds + f (s, a, ut (·, a) ds, if t ∈ [0, T ] ,
u(t, a) = ∂a 0 0

ϕ(t, a), if t ∈ [−r, 0] .

Proof. The assumptions of Theorem 56 imply that ϕ ∈ CX and ϕ(0, ·) ∈

D(A0 ). Consequently, from Theorem 18, we deduce that there exists a
unique function v : [0, T ] → CX which solves the integral equation (6).
It suffices to calculate each term of the integral equation (6).
Let (S0 (t))t≥0 be the integrated semigroup on X generated by A0 . In
view of the definition of (S0 (t))t≥0 , we have
³ ´ Z +∞
−1
(λI − A0 ) x (a) = λ e−λt (S0 (t)x)(a) dt.
0

On the other hand, solving the equation

(λI − A0 ) y = x, where λ > 0, y ∈ D(A0 ) and x ∈ X,

we obtain
³ ´ Z a
(λI − A0 )−1 x (a) = y(a) = e−λt x(a − t) dt.
0

50
Integrating by parts one obtains
³ ´ Z a Z a µZ a ¶
−1 −λa −λt
(λI − A0 ) x (a) = e x(t)dt + e x(s)ds dt.
0 0 a−t

By uniqueness of Laplace transform, we obtain

½ Ra
x(s) ds, if a ≤ t,
(S0 (t)x) (a) = R0a
a−t x(s) ds, if a ≥ t.

Using Proposition 14, we obtain

 R0 Ra

 Rθ ϕ(s, a) ds + R0 ϕ(0, τ ) dτ, if a ≤ t + θ,
0 a
(S(t)ϕ) (θ, a) = θ ϕ(s, a) ds + a−t−θ ϕ(0, τ ) dτ, if a ≥ t + θ ≥ 0,
 R
 t+θ
θ ϕ(s, a) ds, if t + θ ≤ 0.

The assumption ϕ(0, 0) = 0 implies that S(·)ϕ ∈ C 1 ([0, T ] , CX ) and we have


 0, if a ≤ t + θ,
d
(S(t)ϕ) (θ, a) = ϕ(0, a − t − θ), if a ≥ t + θ ≥ 0,
dt 
ϕ(t + θ, a), if t + θ ≤ 0.

Remark that the condition ϕ(0, 0) = 0 is necessary to have S(·)ϕ ∈ C 1 ([0, T ] , CX ).

Let G : [0, T ] → X (T > 0) be a Bochner-integrable function and con-
sider the function K : [0, T ] → CeX defined by
Z t
K(t) = Se0 (t − s)X0 G(s) ds.
0

For a ≤ t + θ, we have
R t+θ
K(t)(θ, a) = (S0 (t + θ − s) G (s)) (a) ds,
R0t+θ−a ¡R a ¢ R t+θ ³R a ´
= 0 0 G (s) (τ ) dτ ds + t+θ−a a−t−θ+s G (s) (τ ) dτ ds.

For a ≥ t + θ ≥ 0, we obtain
Z t+θ µZ a ¶
K(t)(θ, a) = G (s) (τ ) dτ ds,
0 a−t−θ+s

and for t + θ ≤ 0, we have K(t)(θ, a) = 0.

51
 The derivation of K is easily obtained
 Ra
dK  R0 G(τ − a + t + θ)(τ ) dτ, a ≤ t + θ,
a
(t)(θ, a) = G(τ − a + t + θ)(τ ) dτ, a ≥ t + θ ≥ 0,
dt  a−t−θ
0, t + θ ≤ 0.

If we consider the function u : [−r, T ] → X defined by

½
v(t)(0, ·), if t ≥ 0,
u(t, ·) =
ϕ(t, ·), if t ≤ 0.

Corollary 20 implies that ut = v(t). Hence, u is the unique solution of (35).

The second part of Theorem 56 follows from Corollary 20

Theorem 57 Assume that F is continuously differentiable and there exist

constants L, β, γ ≥ 0 such that
|F (t, ϕ) − F (t, ψ)| ≤ L kϕ − ψk ,
|Dt F (t, ϕ) − Dt F (t, ψ)| ≤ β kϕ − ψk ,
|Dϕ F (t, ϕ) − Dϕ F (t, ψ)| ≤ γ kϕ − ψk .
Then, for given ϕ ∈ CX such that
∂ϕ
∈ CX , ϕ(0, ·) ∈ C 1 ([0, l] , IR) ,
∂t
∂ϕ
ϕ(0, 0) = (0, 0) = 0 and
∂t
∂ϕ ∂ϕ
(0, a) + (0, a) = f (0, a, ϕ(·, a)), for a ∈ [0, l] ,
∂t ∂a
the solution u of Equation (35) is continuously differentiable on [0, T ] × [0, l]
and is equal to the unique solution of Problem (34).

Proof. One can use Theorem 22 (all the assumptions of this theorem are
satisfied).
Example 2

 ∂u
 (t, x) = ∆u(t, x) + f (t, x, ut (·, x)), t ∈ [0, T ] , x ∈ Ω,
∂t
 u(t, x) = 0, t ∈ [0, T ] , x ∈ ∂Ω,
 (36)
u(θ, x) = ϕ(θ, x), θ ∈ [−r, 0] , x ∈ Ω,
where Ω ⊂ IRn is a bounded open set with regular boundary ∂Ω, ∆ is the
Laplace operator in the sense of distributions
¡ ¢ on Ω and ϕ is a given function
on CX := C([−r, 0] , X), with X = C Ω, IR .

52
 Problem (36) can be reformulated as an abstract semilinear functional
differential equation
½ ′
V (t) = A0 V (t) + F (t, Vt ), t ∈ [0, T ] ,
V (0) = ϕ,

with
© ¡ ¢ ¡ ¢ ª
D(A0 ) = u ∈ C Ω, IR ; ∆u ∈ C Ω, IR and u = 0 on ∂Ω ,
A0 u = ∆u,

and F : [0, T ] × CX → X is defined by

F (t, ϕ)(x) = f (t, x, ϕ(·, x)) for t ∈ [0, T ] , ϕ ∈ CX and x ∈ Ω.

© ¡ ¢ ª
We have D(A0 ) = u ∈ C Ω, IR ; u = 0 on ∂Ω , 6= X.
Moreover (
ρ(A0 )⊂ (0, +∞)
° °
° −1 °
°(λI − A0 ) ° ≤ λ1 , for λ > 0,

this implies that A0 satisfies (HY) on X (with M0 = 1 and ω0 = 0).

Using the results of Section ??, we obtain the following theorems (the
all assumptions are satisfied).

Theorem 58 Assume that F is continuous on [0, T ] × CX and satisfies a

Lipschitz condition

|F (t, ϕ) − F (t, ψ)| ≤ L kϕ − ψk , t ∈ [0, T ] and ϕ, ψ ∈ CX ,

with L ≥ 0 constant. Then, for a given ϕ ∈ CX , such that

∆ϕ(0, ·) = 0, on ∂Ω,

there exists a unique integral solution u : [0, T ] → X of the partial differential

equation (36), ³i.e. ´
Rt ¡ ¢ Rt
(i) u ∈ CX , ∆ 0 u(s, ·)ds ∈ C Ω, IR and 0 u(s, ·)ds = 0, on ∂Ω,
(ii)
 ³R ´ R

 t t if t ∈ [0, T ] ,
 ϕ(0, x) + ∆ 0 u(s, x) ds + 0 f (s, x, ut (·, x) ds,
and x ∈ Ω
u(t, x) =

 if t ∈ [−r, 0] ,
 ϕ(t, x),
and x ∈ Ω.

53
Theorem 59 Assume that F is continuously differentiable and there exist
constants L, β, γ ≥ 0 such that

|F (t, ϕ) − F (t, ψ)| ≤ L kϕ − ψk ,

|Dt F (t, ϕ) − Dt F (t, ψ)| ≤ β kϕ − ψk ,
|Dϕ F (t, ϕ) − Dϕ F (t, ψ)| ≤ γ kϕ − ψk .

Then, for given ϕ ∈ CX such that

∂ϕ ¡ ¢
∈ CX , ∆ϕ(0, ·) ∈ C Ω, IR ,
∂t
ϕ(0, ·) = ∂ϕ
∂t (0, ·) = 0 on ∂Ω and
∂ϕ
(0, x) = ∆ϕ(0, x) + f (0, x, ϕ(·, x)), for x ∈ Ω.
∂t
There is a unique function x defined on [−r, T ] × Ω, such that x = ϕ on
[−r, 0] × Ω and satisfies Equation (36) on [0, T ] × Ω.

References
[1] M.Adimy, Integrated semigroups and delay differential equations, Jour-
nal of .Mathematical .Analysis and .Applications., 177, (1993), 125-
134.

[2] M. Adimy et K. Ezzinbi, Semi-groupes intégrés et équations

différentielles à retard en dimension infinie, C. R. Acad. Sci. Paris,
t.323, Série I, (1996), 481-486.

[3] M. Adimy et K. Ezzinbi, Local existence and linearized stability for

partial functional differential equations, Dynamic Systems and applica-
tions, 7, (1998), 389-404.

[4] M. Adimy and K. Ezzinbi, Some results on existence, asymptotic be-

havior and stability of solutions for a class of partial neutral functional
differential equations, Hiroshima Mathematical Journal, ( to appear
2004).

[5] M Adimy and K. Ezzinbi, Strict solutions of nonlinear hyperbolic neu-

tral differential equations, Applied Mathematics Letters, 12, 107-112,
(1999).

54
 [6] M Adimy and K. Ezzinbi, A class of linear partial neutral functional
differential equations with non-dense domain, Journal of Differential
Equations, 147, 285-332, (1998).
[7] M Adimy and K. Ezzinbi, Existence and linearized stability for partial
neutral functional differential equations, Journal of Differential Equa-
tions and Dynamical Systems Vol. 7, Number 4, October, 371-417,
(1999).
[8] M Adimy, K. Ezzinbi and M. Laklach, Local existence and global con-
tinuation for a class of partial neutral functional differential equations,
C.R.Acad. Sci. Paris, t.330, Série I, 957-962, (2000).
[9] M Adimy, H.Bouzahir and K. Ezzinbi, Existence for a class of partial
functional differential equations with infinite delay, Journal of Nonlinear
Analysis, Theory, Methods and Applications, Vol 46, No 1, 91-112,
(2001).
[10] M Adimy, H. Bouzahir and K. Ezzinbi, Local existence and stability
for a class of partial functional differential equations with infinite delay,
Journal of Nonlinear Analysis, Theory, Methods and Applications, 48,
323-348, (2002).
[11] M. Adimy, H. Bouzahir and K. Ezzinbi, Existence and stability for some
partial functional differential equations with infinite delay, Journal of
Mathematical Analysis and Applications, ( to appear 2004).
[12] R.Benkhalti and H. Bouzahir and K. Ezzinbi, Existence of Periodic
solutions for some partial functional differential equations with infinite
delay, Journal of Mathematical Analysis and Applications, Vol. 256,
257-280, (2001) .
[13] R.Benkhalti and K.Ezzinbi, A Massera type criterion for some partial
functional differential equations, Dynamic Systems and Applications,
Vol. 9, 221-228, (2000).
[14] R.Benkhalti and K.Ezzinbi, A Hartman-Grobman Theorem for some
partial functional differential equations, Journal of Bifurcation and
Chaos, Vol. 10, No. 5, 1165-1169, (2000).
[15] O. Arino, R.Benkhalti and K.Ezzinbi, Existence results for initial value
problems of neutral functional differential equations, Journal of Differ-
ential Equations, Vol. 138, No. 1, (1998).

55
[16] K. Ezzinbi and H. Tamou, Abstract semilinear functional differential
equations of retarded type, Dynamics of Continuous, Discrete and Im-
pulsive Systems, Serie A, Mathematical Analysis, Vol. 8, No. 2, (2001),
291.

[17] J. H. Liu and K. Ezzinbi, Periodic solutions of Non-Densely defined de-

lay evolution equations, Journal of Applied Mathematics and Stochastic
Analysis, , Vol. 15, No. 2, 113-123, (2002).

[18] K. Ezzinbi and S.L.Rhali, Positivity and stability for some partial func-
tional differential equations, Nonlinear Differential Equations and Ap-
plications, Vol. 10, No. 1, 1-13, (2003).

[19] H. Bouzahir and K. Ezzinbi, Global attractor for a class of partial func-
tional differential equations with infinite delayn, Topics in Functional
Differential Equations, AMS, Fields institute communications, T. Faria
and P. Freitas (Editors), 63-71, (2001).

[20] K.Ezzinbi anf J.H. Liu, Non-densely defined evolution equations with
non local conditions, Mathematical and Computer Modeling, 36, 1027-
1038, (2002).

[21] E. Ait Dads and K. Ezzinbi, Boundedness and almost period-

icity for some state-dependent delay differential equations, Elec-
tronic Journal of Differential Equations, Vol. 2002, No 67, 1-13,
http://ejde.math.swt.edu , (2002).

[22] M. Adimy, K.Ezzinbi and M. Laklach, Spectral decompostion for

some neutral partial functional differential equations, Canadian Ap-
plied Mathematics Quarterly, Vol. 9, No4, 1-34, Winter (2001).

[23] R. Benkhalti and K.Ezzinbi, Existence of periodic solutions for some

neutral functional differential equations, Dynamic Systems and Appli-
cations, Vol. 11, No. 2, (2002), 231-240.

[24] M. Adimy, K. Ezzinbi and M. Laklach, Nonlinear Semigroup of a class

of abstract semilinear functional differential equations with non dense
domain, Acta Mathematic Senica, English series, (to appear 2003).

[25] K.Ezzinbi and Xianlong Fu, Existence of solutions for neutral functional
evolution equations with nonlocal conditions, Nonlinear Analysis, The-
ory, Methods and Applications, Vol. 54, 215-227, (2003).

56
[26] J. H. Liu and K. Ezzinbi, Non-autonomous intergrodifferential equa-
tions with nonlocal conditions, Journal of Integral Equations and Ap-
plications, Vol. 15, Number 1, Spring, 79-93, (2003).
[27] K. Ezzinbi, T.Naito, V. Minh and J. H. Liu, Periodic solutions of evo-
lution equations, Dynamics of Continuous, Discrete and Impulsive Sys-
tems, ( to appear 2004).
[28] K. Ezzinbi and R. Benkhalti, Periodic solutions for some partial func-
tional differential equations, Journal of Applied Mathematics and Sto-
chastic Analysis, (to appear 2004).
[29] M. Adimy, H. Bouzahir and K. Ezzinbi, Local existence for a class
of partial neutral functional differential equations with infinite delay,
Journal of Differential Equations and Dynamical Systems, ( to appear
2004).
[30] K. Ezzinbi, Existence and stability for some partial functrional differ-
ential equations with infinite delay, Electronic Journal of Differential
Equations, Vol. 2003, No. 116, 1-13, http://ejde.math.swt.edu, (2003).
[31] K. Ezzinbi and F. Xianlong, Existence and regularity of solutions for
some neutral partial differential equations with nonlocal conditions,
Nonlinear Analysis, Theory, Methods and Applications, ( to appear
2004).
[32] K. Ezzinbi and M. Jazar, New criteria for the existence of periodic
and almost periodic solutions for some evolution equations in Banach
spaces, Electronic Journal of Qualitative Theory of Differential Equa-
tions, No.6, 1-12, (2004), http://www.math.u-szeged.hu/ejqtde/.
[33] R. Benkhalti and K. Ezzinbi, Existence and stability for some partial
neutral functional differential equations, Differential and Integral Equa-
tions, ( to appear 2004).
[34] W.Arendt, Resolvant positive operators and integrated semigroup,
Proc. London Math. Soc., 3, vol.54, (1987), 321-349.
[35] W.Arendt, Vector valued Laplace transforms and Cauchy problems,
Israel J.Math., 59, (1987), 327-352.
[36] O.Arino et E.Sanchez, Linear theory of abstract functional differential
equations of retarded type, J.M.A.A., 191, (1995), 547-571.

57
[37] S.Busenberg and B.Wu, Convergence theorems for integrated semi-
group, Diff. Int. Eq. 5, vol. 3, May, (1992), 509-520.

[38] P.Clément, O.Diekmann, M.Gyllenberg, H.J.A.M. Heijmans and

H.R.Thieme, Perturbation theory for dual semigroups. I. the sun-
reflexive case, Math. Ann., 277, (1987), 709-725.

[39] C. Corduneanu, Almost Periodic Functions, 2nd edition, Chelsea , New

York (1989).

[40] G.Da Prato and E.Sinestrari, Differential operators with non-dense do-
mains, Ann., Scuola, Norm.,Sup. Pisa Cl. Sci., 14, (1987), 285-344.

[41] W. Desch and W. Schappacher, Linearized stability for nonlinear

semigroups, in “Differential Equations in Banach Spaces” ( A.Favini
and E.Obrecht, Eds.) pp. 61-73, Lecture Notes in Math., Vol. 1223,
Springer-Verlag, New York / Berlin (1986).

[42] O. Diekmann, S. A. Van Gils, S. M. Verduyn Lunel and H. O.

Walther, Delay Equations, Functional, Complex and Nonlinear Analy-
sis, Springer-Verlag, (1995).

[43] W.E.Fitzgibbon, Semilinear functional differential equations in Banach

space, J. Differential Equ. 29, (1978), 1-14.

[44] A.Grabosch and U.Moustakas, A semigroup approach to retarded differ-

ential equations, in ” One-Parameter Semigroups of Positive Operators
(R.Nagel, ed.), p 219-232, Lecture Notes in Mathematics, Vol. 1184,
Springer-Verlag, Berlin/New York, (1986).

[45] J.K.Hale, Theory of functional differential equations Springer-Verlag,

New York, (1977).

[46] J.K.Hale and S.Lunel, Introduction to functional differential equations,

Springer-Verlag, New York (1993).

[47] M.Hieber, Integrated semigroups and diff. operators on Lp , Disserta-

tion, (1989).

[48] T.Kato, Perturbation theory for linear operators, Springer, Berlin

(1966).

58
[49] K.Kunish and W.Schappacher, Variation-of-constants formula for par-
tial differential equations with delay, Nonlinear Anal. TMA 5, (1981),
123-142.

[50] K.Kunish and W.Schappacher, Necessary conditions for partial differ-

ential equations with delay to generate a C0 −semigroups, J.D.E 50,
(1983), 49-79.

[51] H. Kellermann, Integrated Semigroups, Dissertation, Tübingen, (1986).

[52] H.Kellermann and M.Hieber, Integrated semigroup, J.Fun.Anal., 15,

(1989), 160-180.

[53] M.Memory, Stable and unstable manifolds for partial functional differ-
ential equations, Nonlinear Analysis, 16, (1991), 131-142.

[54] M.Memory, Invariant manifolds for partial functional differential

equations, In Mathematical Population Dynamics (Eds. O.Arino,
D.E.Axelrod and M.Kimmel) (1991), 223-232, Marcel-Dekker.

[55] R. Nagel, One-parameter Semigroups of Positive Operators, Lect. Notes

in Mathematics, Springer-Verlag, 1184, (1986).

[56] R. Nagel and E. Sinestrari, Nonlinear hyperbolic Volterra integrodiffer-

ential equations, Nonlinear Anal. TMA, Vol.27, No.2, (1996), 167-186.

[57] F.Neubrander, Integrated semigroups and their applications to the ab-

stract Cauchy problems, Pacific, J. Math., 135, (1988), 111-155.

[58] M.E. Parrrot, Positivity and a principle of linearized stability for delay
differential equations, J. Diff. and Integ. Equa., 2, (1989), 170-182.

[59] M.E. Parrrot, Linearized stability and irreducibility for a functional

differential equation, SIAM J. Math. Anal., 23, (1992), 649-661.

[60] A.Pazy, Semigroups of linear operators and applications to partial dif-

ferential equations, Springer, Berlin (1983).

[61] W.M. Ruess and W. H. Summers, Linearized stability for abstract dif-
ferential equations with delay, J. of Mathematical Analysis and Appli-
cations, 198, (1996), 310-336.

[62] E.Sinestrari, On the abstract Cauchy problem of parabolic type in

spaces of continuous functions; J.M.A.A., 107 (1985), 16-66.

59
[63] H.Thieme, Integrated semigroups and integrated solutions to abstract
Cauchy problems, J.M.A.A., 152, (1990), 416-447.

[64] H.Thieme, Semiflows generated by Lipschitz perturbations of non-

densely defined operators, Diff. Integr. Equations 3, No.6, (1990),
1035-1066.

[65] C.C.Travis and G.F.Webb, Existence and stability for partial functional
differential equations, Trans. Amer. Math. Soc. 200, (1974), 395-418.

[66] G.F.Webb, Asymptotic stability for abstract functional differential

equations, Proc. Amer. Math.Soc. 54 (1976), 225-230.

[67] J. Wu, Theory and Applications of Partial Functional Differential Equa-

tions, Springer-Verlag, (1996).

60