ﻲﻤﻠﻌﻟﺍ ﺚﺤﺒﻟﺍﻭ ﱄﺎﻌﻟﺍ ﻢﻴﻠﻌﺘﻟﺍ ﺓﺭﺍﺯﻭ
BADJI MOKHTAR UNIVERSITY
OF ANNABA
UNIVERSITÉ BADJI MOKHTAR
DE ANNABA
ﺭﺎﺘﺨﻤ ﻲﺠﺎﺒ ﺔﻌﻤﺎﺠ
ﺔﺒﺎﻨﻋ
Faculté des Sciences
Département de Mathématiques
MÉMOIRE
Présenté en vue de l’obtention du diplôme de
MAGISTER EN MATHÉMATIQUES
ÉCOLE DOCTORALE EN MATHÉMATIQUES
Par
Khaled ZENNIR
Intitulé
Existence et Comportement Asymptotique des
Solutions d’une Equation de Viscoélasticité
Non Linéaire de type Hyperbolique
Dirigé par
Prof. Hocine SISSAOUI
Option
Systèmes Dynamiques et Analyse Fonctionnelle
Devant le jury
PRÉSIDENT
B. KHODJA
PROF.
U. B. M. ANNABA
RAPPORTEUR
H. SISSAOUI
PROF.
U. B. M. ANNABA
ÉXAMINATEUR
L. NISSE
M. C.
U. B. M. ANNABA
ÉXAMINATEUR
Y. LASKRI
M. C.
U. B. M. ANNABA
INVITE
B. SAID-HOUARI
Dr.
U. B. M. ANNABA
ﻲﻤﻠﻌﻟﺍ ﺚﺤﺒﻟﺍﻭ ﱄﺎﻌﻟﺍ ﻢﻴﻠﻌﺘﻟﺍ ﺓﺭﺍﺯﻭ
BADJI MOKHTAR UNIVERSITY
OF ANNABA
UNIVERSITÉ BADJI MOKHTAR
DE ANNABA
ﺭﺎﺘﺨﻤ ﻲﺠﺎﺒ ﺔﻌﻤﺎﺠ
ﺔﺒﺎﻨﻋ
Sciences Faculty
Department of Mathematics
THESIS
Submitted for the obtention of
MAGISTER DIPLOMA IN MATHEMATICS
DOCTORAL SCHOOL IN MATHEMATICS
By
Khaled ZENNIR
Existence et Comportement Asymptotique des
Solutions d’une Equation de Viscoélasticité
Non Linéaire de type Hyperbolique
Directed by
First Advisor: Prof. Hocine SISSAOUI
Second Advisor : Dr. Belkacem SAÏD-HOUARI
Option
Dynamic Systems and Functional Analysis
PRESIDENT
B. KHODJA
PROF.
U. B. M. ANNABA
SUPERVISOR
H. SISSAOUI
PROF.
U. B. M. ANNABA
EXAMINER
L. NISSE
M. C.
U. B. M. ANNABA
EXAMINER
Y. LASKRI
M. C.
U. B. M. ANNABA
2008
Existence and Asymptotic Behavior of
Solutions of a Non Linear Viscoelastic
In the first place at all, my gratitude goes to
Prof. H.
SISSAOUI
, my supervisor. He always listens, advices, encourages and
guides me, he was behind this work. I shall always be impressed by his
availability and the care with which he comes to read and correct my
manuscript.
My warm thanks go to
Dr.B.SAID-HOUARI
, who supervised this thesis
with great patience and kindness. He was able to motivate each step of my
work by relevant remarks and let me able to advance in my research,
beginning by the DES study in university of Skikda. I thank him very
sincerely for his availability even at a distance, with him I have a lot of fun
to work.
I wish to thank
Prof B. KHODJA
, who makes me the honour of chairing
the jury of my defense.
I am very grateful to
Dr. L. NISSE
and
Dr. Y. LASKRI
to have agreed to
rescind this thesis and to be part of my defense panel.
Of course, this thesis is the culmination of many years of study. I am
indebted to all those who guided me on the road of mathematics.
I thank from my heart, my teachers whose encouraged me during my
studies and have distracted me the mathematics.
I think for my parents, whose the work could not succeed without their
endless support and encouragement. They find in the realization of this
work, the culmination of their efforts and express my most affectionate
gratitude.
1 Preliminary 1
1.1 Banach Spaces - De…nition and Properties . . . 2
1.1.1 The weak and weak star topologies . . . 3
1.1.2 Hilbert spaces. . . 4
1.2 Functional Spaces . . . 6
1.2.1 The Lp( ) spaces . . . . 6
1.2.2 The Sobolev space Wm;p( ) . . . 8
1.2.3 The Lp(0; T; X) spaces . . . 10
1.2.4 Some Algebraic inequalities . . . 12
1.3 Existence Methods. . . 14
1.3.1 The Contraction Mapping Theorem . . . 14
1.3.2 Gronwell’s lemma . . . 14
1.3.3 The mean value theorem . . . 15
2 Local Existence 17 2.1 Local Existence Result . . . 18
3 Global Existence and Energy Decay 39 3.1 Global Existence Result . . . 40
3.2 Decay of Solutions . . . 46
4 Exponential Growth 59 4.1 Growth result . . . 60
Abstract
Our work, in this thesis, lies in the study, under some conditions on p, m and the functional g, the existence and asymptotic behavior of solutions of a nonlinear viscoelastic hyperbolic problem
of the form 8 > > > > > > > > > > > > > > > > < > > > > > > > > > > > > > > > > : utt u ! ut+ t R 0 g(t s) u(s; x)ds +ajutj m 2 ut = bjuj p 2 u; x2 ; t > 0 u (0; x) = u0(x) ; x2 ut(0; x) = u1(x) ; x 2 u (t; x) = 0; x2 ; t > 0 ; (P)
where is a bounded domain in RN (N 1), with smooth boundary ; a; b; w are positive
constants, m 2; p 2;and the function g satisfying some appropriate conditions.
Our results contain and generalize some existing results in literature. To prove our results many theorems were introduced.
Keywords: Nonlinear damping, strong damping, viscoelasticity, nonlinear source, local solutions,
Résumé
Notre travail, dans ce memoire consiste à étudier l’éxistence et le comportement asymptotique des solutions d’un problème de viscoelasticité non lineaire de type hyperbolique suivant:
8 > > > > > > > > > > > > > > > > < > > > > > > > > > > > > > > > > : utt u ! ut+ t R 0 g(t s) u(s; x)ds +ajutjm 2ut= bjujp 2u; x2 ; t > 0 u (0; x) = u0(x) ; x 2 ut(0; x) = u1(x) ; x2 u (t; x) = 0; x2 ; t > 0 ; (P)
où, est un domaine borné de RN (N 1), avec frontière assez regulière ; a; b; w sont des
constantes positives, m 2; p 2;et la fonction g satisfaite quelques conditions.
Nos résultats contiennent et généralisent certains résultats d’existences dans la littérature. Pour la preuve, beaucoup théorèmes ont été présentés
Mots clés: Dissipation nonlinéaire, viscoelasticité, source nonlinéaire, solutions locale, solutions
Notations
: bounded domain in RN: :topological boundary of : x = (x1; x2; :::; xN) : generic point of RN: dx = dx1dx2:::dxN :Lebesgue measuring on : ru : gradient of u: u : Laplacien of u: f+; f : max(f; 0); max( f; 0):a:e : almost everywhere.
p0 :conjugate of p; i:e 1
p +
1
p0 = 1:
D( ) : space of di¤erentiable functions with compact support in :
D0( ) : distribution space.
Ck( ) : space of functions k times continuously di¤erentiable in :
C0( ) : space of continuous functions null board in :
Lp( ) : Space of functions p th power integrated on with measure of dx:
kfkp = R jf(x)jp 1 p : W1;p( ) =nu 2 Lp( ) ; ru 2 (Lp( ))No :
kuk1;p = kukpp+krukpp
1 p :
W01;p( ) : the closure of D ( ) in W1;p( ) :
W 1;p0( ) : the dual space of W01;p( ) :
H :Hilbert space. H1 0 = W 1;2 0 : If X is a Banach space Lp(0; T ; X) = f : (0; T ) ! X is measurable ; T R 0 kf(t)kpXdt <1 : L1(0; T ; X) = (
f : (0; T )! X is measurable ;ess sup
t2(0;T ) kf(t)k p X <1 ) : Ck([0; T ] ; X) :
Space of functions k times continuously di¤erentiable for [0; T ] ! X:
D ([0; T ] ; X) :space of functions continuously di¤erentiable with compact support in [0; T ] :
Introduction
In this thesis we consider the following nonlinear viscoelastic hyperbolic problem 8 > > > > > > > > > > > > > > < > > > > > > > > > > > > > > : utt u ! ut+ t R 0 g(t s) u(s; x)ds +ajutj m 2 ut= bjuj p 2 u; x2 ; t > 0 u (0; x) = u0(x) ; ut(0; x) = u1(x) ; x2 u (t; x) = 0; x2 ; t > 0 ; (1)
where is a bounded domain in RN (N 1), with smooth boundary ; a; b; w are positive constants,
and m 2; p 2:The function g(t) is assumed to be a positive nonincreasing function de…ned on
R+ and satis…es the following conditions:
(G1): g : R+ ! R+ is a bounded C1-function such that
g(0) > 0; 1 1 Z 0 g(s)ds = l > 0: (G2): g(t) 0; g0(t) 0; g(t) g0(t); 8 t 0; > 0:
In the physical point of view, this type of problems arise usually in viscoelasticity. This type of problems have been considered …rst by Dafermos [12], in 1970, where the general decay was discussed. A related problems to (1) have attracted a great deal of attention in the last two decades, and many results have been appeared on the existence and long time behavior of solutions.
See in this directions [2; 3; 5 8; 18; 29; 33; 34; 38] and references therein.
In the absence of the strong damping ut; that is for w = 0; and when the function g vanishes
identically ( i.e: g = 0); then problem (1) reduced to the following initial boundary damped wave equation with nonlinear damping and nonlinear sources terms.
utt u + ajutj
m 2
ut = bjuj
p 2
u: (2)
Some special cases of equation (2) arise in quantum …eld theory which describe the motion of charged mesons in an electromagnetic …eld.
Equation (2) together with initial and boundary conditions of Dirichlet type, has been extensively studied and results concerning existence, blow up and asymptotic behavior of smooth, as well
as weak solutions have been established by several authors over the past three decades. Some interesting results have been summarized by Said-Houari in his master thesis [42]:
For b = 0, that is in the absence of the source term, it is well known that the damping term ajutj
m 2
ut assures global existence and decay of the solution energy for arbitrary initial data ( see
for instance [17] and [21]):
For a = 0; the source term causes …nite-time blow-up of solutions with a large initial data ( negative initial energy). That is to say, the norm of our solution u(t; x) in the energy space reaches +1 when the time t approaches certain value T called " the blow up time", ( see [1] and [20] for more details).
The interaction between the damping term a jutjm 2ut and the source term b jujp 2u makes the
problem more interesting. This situation was …rst considered by Levine [23; 24] in the linear damping case (m = 2), where he showed that solutions with negative initial energy blow up in …nite time T . The main ingredient used in [23] and [24] is the " concavity method" where the basic idea of this method is to construct a positive function L(t) of the solution and show that for some
> 0; the function L (t) is a positive concave function of t: In order to …nd such ; it su¢ ces to
verify that:
d2L (t)
dt2 = L
2(t) LL00 (1 + )L02(t) 0; 8t 0:
This is equivalent to prove that L(t) satis…es the di¤erential inequality
LL00 (1 + )L02(t) 0; 8t 0:
Unfortunately, this method fails in the case of nonlinear damping term (m > 2):
Georgiev and Todorova in their famous paper [14], extended Levine’s result to the nonlinear damp-ing case (m > 2): More precisely, in [14] and by combindamp-ing the Galerkin approximation with the
contraction mapping theorem, the authors showed that problem (2) in a bounded domain with
initial and boundary conditions of Dirichlet type has a unique solution in the interval [0; T ) provided that T is small enough. Also, they proved that the obtained solutions continue to exist globally in
time if m p and the initial data are small enough. Whereas for p > m the unique solution of
problem (2) blows up in …nite time provided that the initial data are large enough. ( i.e: the initial energy is su¢ ciently negative).
This later result has been pushed by Messaoudi in [35] to the situation where the initial energy
E(0) < 0: For more general result in this direction, we refer the interested reader to the works of
In the presence of the viscoelastic term (g 6= 0) and for w = 0; our problem (1) becomes 8 > > > > > > > > > > > > > > < > > > > > > > > > > > > > > : utt u + t R 0 g(t s) u(s; x)ds +ajutjm 2ut= bjujp 2u; x2 ; t > 0 u (0; x) = u0(x) ; ut(0; x) = u1(x) ; x2 u (t; x) = 0; x2 ; t > 0 : (3)
For a = 0; problem (3) has been investigated by Berrimi and Messaoudi [3]: They established the local existence result by using the Galerkin method together with the contraction mapping theorem. Also, they showed that for a suitable initial data, then the local solution is global in time and in addition, they showed that the dissipation given by the viscoelastic integral term is strong enough to stabilize the oscillations of the solution with the same rate of decaying ( exponential or polynomial) of the kernel g: Also their result has been obtained under weaker conditions than those used by Cavalcanti et al [7]; in which a similar problem has been addressed.
Messaoudi in [29]; showed that under appropriate conditions between m; p and g a blow up and global existence result, of course his work generalizes the results by Georgiev and Todorova [14] and Messaoudi [29]:
One of the main direction of the research in this …eld seems to …nd the minimal dissipation such that the solutions of problems similar to (3) decay uniformly to zero, as time goes to in…nity. Consequently, several authors introduced di¤erent types of dissipative mechanisms to stabilize these
problems. For example, a localized frictional linear damping of the form a(x)utacting in sub-domain
w has been considered by Cavalcanti et al [7]: More precisely the authors in [6] looked into
the following problem
utt u +
t Z
0
g(t s) u(s; x)ds + a(x)ut+juj u = 0: (4)
for > 0; g a positive function and a : ! R+ a function, which may be null on a part of the
domain :
By assuming a(x) a0 > 0 on the sub-domain w ; the authors showed a decay result of an
exponential rate, provided that the kernel g satis…es
and kgkL1(0;1) is small enough.
This later result has been improved by Berrimi and Messaoudi [2], in which they showed that the viscoelastic dissipation alone is strong enough to stabilize the problem even with an exponential rate.
In many existing works on this …eld, the following conditions on the kernel
g0(t) gp(t); t 0; p 1; (6)
is crucial in the proof of the stability.
For a viscoelastic systems with oscillating kernels, we mention the work by Rivera et al [36]. In that work the authors proved that if the kernel satis…es g(0) > 0 and decays exponentially to zero, that is for p = 1 in (6), then the solution also decays exponentially to zero. On the other hand, if the kernel decays polynomially, i.e. (p > 1) in the inequality (6), then the solution also decays polynomially with the same rate of decay.
In the presence of the strong damping (w > 0) and in the absence of the viscoelastic term (g = 0), the problem (1) takes the following form
8 > > > > > > > < > > > > > > > : utt u ! ut+ ajutjm 2ut= bjujp 2u; x2 ; t > 0 u (0; x) = u0(x) ; ut(0; x) = u1(x) ; x 2 u (t; x) = 0; x2 ; t > 0 : (7)
Problem (7) represents the wave equation with a strong damping ut. When m = 2, this problem
has been studied by Gazzola and Squassina [13]: In their work, the authors proved some results on well posedness and asymptotic behavior of solutions. They showed the global existence and polynomial decay property of solutions provided that the initial data is in the potential well. The proof in [13] is based on a method used in [19]. Unfortunately their decay rate is not opti-mal, and their result has been improved by Gerbi and Said-Houari [16], by using an appropriate modi…cation of the energy method and some di¤erential and integral inequalities.
Introducing a strong damping term ut makes the problem from that considered in [42] and [14],
for this reason less results where known for the wave equation with strong damping and many problem remain unsolved. ( See [13] and the recent work by Gerbi and Said-Houari [15]):
In this thesis, we investigated problem (1), in which all the damping mechanism have been considered
form those studied in the literature, specially the blow up result / exponential growth of solutions (chapter4).
This thesis is organized as follows: Chapter1:
In this chapter we introduce some notation and prepare some material needed for our work. The
main results of this chapter such as: the Lp spaces, the Sobolev spaces, di¤erential and integral
inequalities and other theorems of functional analysis, can found in the books [4] and [43]: Chapter2:
This chapter is devoted to the study of the local existence result, the main ingredient used in this chapter is the Galerkin approximations ( the compactness method) introduced in the book of Lions
[26]; together with the …x point method.
Indeed, we consider …rst for u 2 C ([0; T ] ; H1
0) given, the following problem
vtt v ! vt+
t Z 0
g(t s) v(s; x)ds + ajvtjm 2vt= bjujp 2u; x2 ; t > 0 (8)
with the initial data
v (0; x) = u0(x) ; vt(0; x) = u1(x) ; x 2 (9)
and boundary conditions of the form
v (t; x) = 0; x2 ; t > 0; (10)
and we will show that problem (8) (10) has a unique local solutions v by the Faedo-Galerkin
method, which consists in constructing approximations of the solution, then we obtain a priori estimates necessary to guarantee the convergence of these approximations. We recall here that the
presence of nonlinearity on the damping term a jvtj
m 2
vt forces us to go until the second a priori
estimate. We point out that the contraction semigroup method fails here, because of the presence of the nonlinear terms.
Once the local solution v exists, we will use the contraction mapping theorem to show the local existence of our problem (1). This will be done under the assumption that T is required to be small enough (see formula (2:77)):
Chapter 3:
Our main purpose in this chapter is tow-fold:
First, we introduce a set W de…ned in (3:5) called " the potential well " or " stable set" and we show that if we restrict our initial data in this set, then our solution obtained in chapter 2 is global in time, that is to say, the norm
in the energy space L2( ) H1
0( ) of our solution is bounded by a constant independent of the
time t:
Second, We show that, if our solution is global in time, ( i.e: by assuming that the initial data
u0 2 W ) and if our function g satis…es the condition (6) ( for p = 1); then our solution decays time
asymptotically to zero. More precisely we prove that the decay rate is of the form (1 + t)2=(2 m)
if m > 2; whereas for m = 2; we obtain an exponential decay rate. (See Theorem 3.2.1). The main tool used in our proof is an inequality due to Nakao [37]; in which this inequality has been introduced in order to study the stability of the wave equation, but it is still works in our problem. Chapter 4:
In this chapter we will prove that if the initial energy E(0) of our solution is negative ( this means that our initial data are large enough), then our local solutions in bounded and
kutk2+kruk2 ! 1
as t tends to +1: In fact it will be proved that the Lp norm of the solution grows as an exponential
function. An essential tool of the proof is an idea used by Gerbi and Said-Houari [15], which based on an auxiliary function ( which is a small perturbation of the total energy), in order to obtain a di¤erential inequality leads to the exponential growth result provided that the following conditions
1 Z 0 g(s)ds < p 2 p 1; holds.
Preliminary
Abstract
In this chapter we shall introduce and state some necessary materials needed in the proof of our results, and shortly the basic results which concerning the Banach spaces, the weak and weak star
topologies, the Lp space, Sobolev spaces and other theorems. The knowledge of all this notations
1.1
Banach Spaces - De…nition and Properties
We …rst review some basic facts from calculus in the most important class of linear spaces "Banach spaces".
De…nition 1.1.1 A Banach space is a complete normed linear space X: Its dual space X0 is the
linear space of all continuous linear functional f : X ! R:
Proposition 1.1.1 ([43])
X0 equipped with the norm k:kX
0 de…ned by
kfkX0 = supfjf(u)j : kuk 1g ; (1.1)
is also a Banach space.
We shall denote the value of f 2 X0 at u 2 X by either f(u) or hf; ui
X0; X:
Remark 1.1.1 ([43]) From X0 we construct the bidual or second dual X00 = (X0)0: Furthermore,
with each u 2 X we can de…ne '(u) 2 X00 by '(u)(f ) = f (u); f 2 X0, this satis…es clearly
k'(x)k kuk : Moreover, for each u 2 X there is an f 2 X0 with f (u) = kuk and kfk = 1, so it
follows that k'(x)k = kuk :
De…nition 1.1.2 Since ' is linear we see that
' : X ! X00;
is a linear isometry of X onto a closed subspace of X00, we denote this by
X ,! X00:
De…nition 1.1.3 If ' ( in the above de…nition) is onto X00 we say X is re‡exive, X u X00:
Theorem 1.1.1 ([4]; Theorem III.16)
Let X be Banach space. Then, X is re‡exive, if and only if,
BX =fx 2 X : kxk 1g ;
is compact with the weak topology (X; X0) : (See the next subsection for the de…nition of (X; X0))
De…nition 1.1.4 Let X be a Banach space, and let (un)n2N be a sequence in X. Then un converges
strongly to u in X if and only if
lim
n!1kun ukX = 0;
and this is denoted by un! u; or lim
1.1.1
The weak and weak star topologies
Let X be a Banach space and f 2 X0: Denote by
'f : X ! R
x7! 'f(x); (1.2)
when f cover X0, we obtain a family '
f f2X0 of applications to X in R:
De…nition 1.1.5 The weak topology on X; denoted by (X; X0) ; is the weakest topology on X for
which every 'f f2X0 is continuous.
We will de…ne the third topology on X0, the weak star topology, denoted by (X0; X) : For all
x2 X: Denote by
'x : X0 ! R
f 7! 'x(f ) =hf; xiX0; X;
(1.3)
when x cover X, we obtain a family ('x)x2X0 of applications to X0 in R:
De…nition 1.1.6 The weak star topology on X0 is the weakest topology on X0 for which every
('x)x2X0 is continuous.
Remark 1.1.2 ([4]) Since X X00; it is clear that, the weak star topology (X0; X) is weakest
then the topology (X0; X00), and this later is weakest then the strong topology.
De…nition 1.1.7 A sequence (un) in X is weakly convergent to x if and only if
lim
n!1f (un) = f (u);
for every f 2 X0; and this is denoted by un * u.
Remark 1.1.3 ([42]; Remark 1.1.1)
1. If the weak limit exist, it is unique.
2. If un ! u 2 X (strongly), then un * u (weakly):
3. If dim X < +1; then the weak convergent implies the strong convergent.
Proposition 1.1.2 ([43])
On the compactness in the three topologies in the Banach space X : 1- First, the unit ball
in X is compact if and only if dim(X) < 1:
2- Second, the unit ball B0 in X0 (The closed subspace of a product of compact spaces) is weakly
compact in X0 if and only if X is re‡exive.
3- Third, B0 is always weakly star compact in the weak star topology of X0.
Proposition 1.1.3 ([4]; proposition III.12)
Let (fn) be a sequence in X0. We have:
1. hfn * f in (X0; X)
i
, [fn(x)! f(x); 8x 2 X] :
2. If fn ! f (strongly), then fn * f, in (X0; X00),
If fn * f in (X0; X00), then fn* f, in (X0; X) :
3. If fn * f, in (X0; X), then kfnk is bounded and kfk lim infkfnk :
4. If fn * f, in (X0; X) and x
n ! x (strongly) in X, then fn(xn)! f(x):
1.1.2
Hilbert spaces
The proper setting for the rigorous theory of partial di¤erential equation turns out to be the most important function space in modern physics and modern analysis, known as Hilbert spaces. Then, we must give some important results on these spaces here.
De…nition 1.1.8 A Hilbert space H is a vectorial space supplied with inner product hu; vi such that
kuk =phu; ui is the norm which let H complete.
Theorem 1.1.2 ([42], Theorem 1.1.1)
Let (un)n2N is a bounded sequence in the Hilbert space H, then it possess a subsequence which
converges in the weak topology of H:
Theorem 1.1.3 ([42], Theorem 1.1.2)
In the Hilbert space, all sequence which converges in the weak topology is bounded.
Theorem 1.1.4 ([42], Corollary 1.1.1)
Let (un)n2N be a sequence which converges to u; in the weak topology and (vn)n2N is an other sequence
which converge weakly to v; then
lim
n!1hvn; uni = hv; ui: (1.5)
Theorem 1.1.5 ([42], Theorem 1.1.3)
Let X be a normed space, then the unit ball
B0 =nl 2 X0 :klk 1o; (1.6)
Proposition 1.1.4 ([42]; Proposition 1.1.1)
Let X and Y be two Hilbert spaces, let (un)n2N 2 X be a sequence which converges weakly to u 2 X;
let A 2 L(X; Y ): Then, the sequence (A (un))n2N converges to A(u) in the weak topology of Y:
Proof. For all u 2 X, the function
u7! hA(u); vi
is linear and continuous, because
jhA(u); vij kAkL(X; Y )kukXkvkY ; 8u 2 X; v 2 Y:
So, according to Riesz theorem, there exists w 2 X such that hA(u); vi = hu; wi, 8u 2 X: Then,
lim
n!1hA (un) ; vi = limn!1hun; wi
= hu; wi = hA(u); vi: (1.7)
1.2
Functional Spaces
1.2.1
The
L
p( ) spaces
De…nition 1.2.1 Let 1 p 1; and let be an open domain in Rn; n2 N: De…ne the standard
Lebesgue space Lp( ); by Lp( ) = 8 < :f : ! R : f is measurable and Z jf(x)jpdx <1 9 = ;: (1.8)
Notation 1.2.1 For p 2 R and 1 p < 1, denote by
kfkp = 0 @Z jf(x)jp dx 1 A 1 p : (1.9) If p = 1; we have
L1( ) =ff : ! R : f is measurable and there exists a constant C
such that, jf(x)j C a.e in g:
(1.10)
Also, we denote by
kfk1=Inf fC; jf(x)j C a.e in g : (1.11)
Notation 1.2.2 Let 1 p 1; we denote by q the conjugate of p i.e. 1
p +
1
q = 1:
Theorem 1.2.1 ([48])
It is well known that Lp( ) supplied with the norm
k:kp is a Banach space, for all 1 p 1:
Remark 1.2.1 In particularly, when p = 2, L2( ) equipped with the inner product
hf; giL2( ) =
Z
f (x)g(x)dx; (1.12)
is a Hilbert space.
Theorem 1.2.2 ([43], Corollary 3.2)
Some integral inequalities
We will give here some important integral inequalities. These inequalities play an important role in applied mathematics and also, it is very useful in our next chapters.
Theorem 1.2.3 ([48], Hölder’s inequality )
Let 1 p 1. Assume that f 2 Lp( ) and g
2 Lq( ); then, f g
2 L1( ) and
Z
jfgj dx kfkpkgkq: (1.13)
Corollary 1.2.1 (Hölder’s inequality - general form)
Lemma 1.2.1 Let f1; f2; :::fk be k functions such that, fi 2 Lpi( ); 1 i k, and
1 p = 1 p1 + 1 p2 + ::: + 1 pk 1:
Then, the product f1f2:::fk 2 Lp( ) and kf1f2:::fkkp kf1kp1:::kfkkpk:
Lemma 1.2.2 ( [48], Young’s inequality)
Let f 2 Lp (R) and g 2 Lq (R) with 1 < p < 1; 1 < q < 1 and 1 r = 1 p + 1 q 1 0. Then f g 2 Lr (R) and kf gkLr(R) kfkLp(R)kgkLq(R): (1.14)
Lemma 1.2.3 ([43]; Minkowski inequality)
For 1 p 1; we have ku + vkLp kukLp +kvkLp: (1.15) Lemma 1.2.4 ([43]) Let 1 p r q; 1 r = p + 1 q ; and 1 1: Then
kukLr kukLpkuk
1 Lq : (1.16) Lemma 1.2.5 ([43]) If ( ) <1; 1 p q 1; then Lq , ! Lp; and kukLp ( ) 1 p 1 q kukL q:
1.2.2
The Sobolev space W
m;p( )
Proposition 1.2.1 ([26])
Let be an open domain in RN;
Then the distribution T 2 D0( ) is in Lp( ) if there exists a
function f 2 Lp( ) such that
hT; 'i = Z
f (x)'(x)dx; for all '2 D( );
where 1 p 1; and it’s well-known that f is unique.
De…nition 1.2.2 Let m 2 N and p 2 [0; 1] : The Wm;p( ) is the space of all f
2 Lp( ), de…ned as Wm;p( ) = ff 2 Lp( ); such that @ f 2 Lp( ) for all 2 Nm such that j j =Pnj=1 j m; where, @ = @ 1 1 @22::@nng: (1.17) Theorem 1.2.4 ([9])
Wm;p( ) is a Banach space with their usual norm
kfkWm;p( ) =
X j j m
k@ fkLp; 1 p < 1, for all f 2 Wm;p( ): (1.18)
De…nition 1.2.3 Denote by W0m;p( ) the closure of D( ) in Wm;p( ):
De…nition 1.2.4 When p = 2, we prefer to denote by Wm;2( ) = Hm( ) and W0m;2( ) = H0m( )
supplied with the norm
kfkHm( ) = 0 @X j j m (k@ fkL2) 2 1 A 1 2 ; (1.19)
which do at Hm( ) a real Hilbert space with their usual scalar product
hu; viHm( ) = X j j m Z @ u@ vdx (1.20) Theorem 1.2.5 ([42]; Proposition 1.2.1)
1) Hm( ) supplied with inner product
h:; :iHm( ) is a Hilbert space.
2) If m m0, Hm( ) ,
! Hm0
( ), with continuous imbedding :
Lemma 1.2.6 ([26])
Since D( ) is dense in H0m( ) ; we identify a dual H m( ) of H0m( ) in a weak subspace on ;
and we have
D( ) ,! H0m( ) ,! L
2( ) ,
Lemma 1.2.7 (Sobolev-Poincaré’s inequality) If 2 q 2n n 2; n 3 q 2; n = 1; 2; then kukq C(q; )kruk2; (1.21) for all u 2 H1 0( ) :
The next results are fundamental in the study of partial di¤erential equations
Theorem 1.2.6 ([9] Theorem 1.3.1)
Assume that is an open domain in RN (N 1), with smooth boundary . Then,
(i) if 1 p n; we have W1;p Lq( ); for every q
2 [p; p ] ; where p = np
n p:
(ii) if p = n we have W1;p Lq( ); for every q
2 [p; 1) :
(iii) if p > n we have W1;p L1( )\ C0; ( ); where = p n
p :
Theorem 1.2.7 ([9] Theorem 1.3.2)
If is a bounded, the embedding (ii) and (iii) of theorem 1.1.4 are compacts. The embedding (i) is
compact for all q 2 [p; p ) :
Remark 1.2.2 ([26])
For all ' 2 H2( ); '
2 L2( ) and for su¢ ciently smooth, we have
k'(t)kH2( ) Ck '(t)kL2( ): (1.22)
Proposition 1.2.2 ([43], Green’s formula)
For all u 2 H2( ); v 2 H1( ) we have Z uvdx = Z rurvdx Z @ @u @ vd ; (1.23) where @u @ is a normal derivation of u at .
1.2.3
The
L
p(0; T; X) spaces
De…nition 1.2.5 Let X be a Banach space, denote by Lp(0; T; X) the space of measurable functions
f : ]0; T [! X t 7 ! f(t) such that 0 @ T Z 0 kf(t)kpXdt 1 A 1 p =kfkLp(0;T;X)<1; for 1 p < 1: (1.24) If p = 1, kfkL1(0;T;X) = sup t2]0;T [ esskf(t)kX: (1.25) Theorem 1.2.8 ([42])
The space Lp(0; T; X) is complete.
We denote by D0(0; T; X) the space of distributions in ]0; T [ which take its values in X, and let us
de…ne
D0(0; T; X) =L (D ]0; T [ ; X) ;
where L ( ; ') is the space of the linear continuous applications of to ': Since u 2 D0(0; T; X) ;
we de…ne the distribution derivation as @u
@t(') = u
d'
dt ; 8' 2 D (]0; T [) ; (1.26)
and since u 2 Lp(0; T; X) ; we have
u(') = T Z 0
u(t)'(t)dt; 8' 2 D (]0; T [) : (1.27)
We will introduce some basic results on the Lp(0; T; X)space. These results, will be very useful in
the other chapters of this thesis.
Lemma 1.2.8 ([26] Lemma 1.2 )
Let f 2 Lp(0; T; X) and @f
@t 2 L
p(0; T; X); (1 p
1) ; then, the function f is continuous from
Lemma 1.2.9 ([26])
Let ' = ]0; T [ an open bounded domain in R Rn;and let g ; g are two functions in Lq(]0; T [ ; Lq( )) ;
1 < q < 1 such that kg kLq(0;T;Lq( )) C; 8 2 N (1.28) and g ! g in '; then g * g in Lq(') : Theorem 1.2.9 ([9]; Proposition 1.4.17)
Lp(0; T; X) equipped with the norm
k:kLp(0;T;X), 1 p 1 is a Banach space.
Proposition 1.2.3 ([14])
Let X be a re‡exive Banach space, X0 it’s dual, and 1 p <1; 1 q <1; 1
p +
1
q = 1: Then the
dual of Lp(0; T; X) is identify algebraically and topologically with Lq(0; T; X0):
Proposition 1.2.4 ([9])
Let X; Y be to Banach space, X Y with continuous embedding, then we have Lp(0; T; X) Lp(0;
T; Y ) with continuous embedding.
The following compactness criterion will be useful for nonlinear evolution problems, especially in the limit of the non linear terms.
Proposition 1.2.5 ([26]).
Let B0; B; B1 be Banach spaces with B0 B B1; assume that the embedding B0 ,! B is compact
and B ,! B1 are continuous. Let 1 < p < 1; 1 < q < 1; assume further that B0 and B1 are
re‡exive. De…ne
W fu 2 Lp(0; T; B0) : u0 2 Lq(0; T; B1)g : (1.29)
1.2.4
Some Algebraic inequalities
Since our study based on some known algebraic inequalities, we want to recall few of them here.
Lemma 1.2.10 ([48];The Cauchy-Schwarz inequality)
Every inner product satis…es the Cauchy-Schwarz inequality
hx1; x2i kx1k kx2k : (1.30)
The equality sign holds if and only if x1 and x2 are dependent.
Young’s inequalities :
Lemma 1.2.11 For all a; b 2 R+, we have
ab a2+ b
2
4 ; (1.31)
where is any positive constant.
Proof. Taking the well-known result
(2 a b)2 0 8a; b 2 R
for all > 0, we have
4 2a2+ b2 4 ab 0: This implies 4 ab 4 2a2+ b2 consequently, ab a2+ 1 4 b 2:
This completes the proof.
Lemma 1.2.12 ([43])
For all a; b 0; the following inequality holds
ab a p p + bq q; where, 1 p + 1 q = 1:
Proof. Let G = (0; 1), and f : G ! R is integrable, such that f (x) = 8 > < > : p log a, 0 x 1 p q log b, 1 p x 1 ;
for all a; b 0and 1
p +
1
q = 1:
Since '(t) = et is convex, and using Jensen’s inequality
' 0 @ 1 (G) Z G f (x)dx 1 A 1 (G) Z G ' (f (x)) dx: (1.32) Consequently, we have 1 (G) Z G ' (f (x)) dx = 1 Z 0 ef (x)dx = 1=p Z 0 ep log adx + 1 Z 1=p eq log bdx = 1=p Z 0 apdx + 1 Z 1=p bqdx = 1 p(a p) + 1 1 p b q: (1.33) = a p p + bq q ; where, (G) = 1 and ' 0 @ 1 (G) Z G f (x)dx 1 A = e(R01f (x)dx) = e( R1=p 0 p log adx+ R1 1=pq log bdx)
= e(log a+log b) = elog ab
= ab: (1.34)
1.3
Existence Methods
1.3.1
The Contraction Mapping Theorem
Here we prove a very useful …xed point theorem called the contraction mapping theorem. We will apply this theorem to prove the existence and uniqueness of solutions of our nonlinear problem.
De…nition 1.3.1 Let f : X ! X be a map of a metric space to itself. A point x 2 X is called a
…xed point of f if f (x) = x.
De…nition 1.3.2 Let (X; dX) and (Y; dY) be metric spaces. A map ' : X ! Y is called a
contrac-tion if there exists a positive number C < 1 such that
dY('(x); '(y)) CdX(x; y); (1.35)
for all x; y 2 X.
Theorem 1.3.1 (Contraction mapping theorem [45] )
Let (X; d) be a complete metric space. If ' : X ! X is a contraction, then ' has a unique …xed point.
1.3.2
Gronwell’s lemma
Theorem 1.3.2 ( In integral form)
Let T > 0; and let ' be a function such that, ' 2 L1(0; T ); ' 0; almost everywhere and be
a function such that, 2 L1(0; T ); 0; almost everywhere and '
2 L1[0; T ] ; C 1; C2 0. Suppose that (t) C1+ C2 t Z 0 '(s) (s)ds; for a.e t2 ]0; T [ ; (1.36) then, (t) C1exp 0 @C2 t Z 0 '(s)ds 1 A ; for a.e t 2 ]0; T[ : (1.37) Proof. Let F (t) = C1+ C2 t Z 0 '(s) (s)ds; for t 2 [0; T ] ; (1.38) we have, (t) F (t);
From (1:38) we have F0(t) = C2'(t) (t) C2'(t) (t); for a.e t 2 ]0; T [ : (1.39) Consequently, d dt 8 < :F (t) exp 0 @ t Z 0 C2'(s)ds 1 A 9 = ; 0; (1.40) then, F (t) C1exp 0 @C2 t Z 0 '(s)ds 1 A ; for a.e t 2 ]0; T[ : (1.41)
Since F; then our result holds.
In particle, if C1 = 0; we have = 0 for almost everywhere t 2 ]0; T [ :
1.3.3
The mean value theorem
Theorem 1.3.3 Let G : [a; b] ! R be a continues function and ' : [a; b] ! R is an integral positive
function, then there exists a number x in (a; b) such that b Z a G(t)'(t)dt = G(x) b Z a '(t)dt: (1.42)
In particular for '(t) = 1; there exists x 2 (a; b) such that b Z a G(t)dt = G(x) (b a) : (1.43) Proof. Let m = inffG(x); x 2 [a; b]g (1.44) and M = supfG(x); x 2 [a; b]g (1.45)
of course m and M exist since [a; b] is compact. Then, it follows that
m b Z a '(t)dt b Z a G(t)'(t)dt M b Z a '(t)dt: (1.46)
By monotonicity of the integral. dividing through byRab'(t)dt; we have that m Rb a G(t)'(t)dt Rb a '(t)dt M: (1.47)
Since G(t) is continues, the intermediate value theorem implies that there exists x 2 [a; b] such that G(x) = Rb a G(t)'(t)dt Rb a '(t)dt : (1.48)
Local Existence
Abstract
Our goal in this chapter is to study the local existence ( local well-possedness) of the problem (P );
for u in C ([0; T ] ; H1
0( )). For this purpose we consider, …rst the related problem for u …xed in
C ([0; T ] ; H01( )) 8 > > > > > > > > > > > > < > > > > > > > > > > > > : vtt vt v + Rt 0 g(t s) v(s; x)ds +jvtj m 2 vt=juj p 2 u; x2 ; t > 0 v (0; x) = u0(x) ; vt(0; x) = u1(x) ; x2 v (t; x) = 0; x2 ; t > 0 ; (2.1)
and we will prove the local existence of this problem by using the Faedo-Galerkin method. Then, by using the well-known contraction mapping theorem, we can show the local existence of (P ): Our techniques of proof follows carefully the techniques due to Georgiev and Todorova [14]; with necessary modi…cations imposed by the nature of our problem. The …rst step of our proof is the choice of the space where the local solution exists. The minimal requirement for this space is that
u(t; x)be time continuous. The weak space satisfying the above requirement is C ([0; T ] ; H), where
H = H1
2.1
Local Existence Result
In order to prove our local existence results, let us introduce the following space
YT = 8 > > < > > : u : u2 C ([0; T ] ; H1 0( )) ; ut 2 C([0; T ] ; H01( ))\ Lm([0; T ] ) 9 > > = > > ; : (2.2)
Our main result in this chapter reads as follows:
Theorem 2.1.1 Let (u0; u1)2 (H01( ))
2
be given. Suppose that m 2; p 2 be such that
maxfm; pg 2 (n 1)
n 2 ; n 3: (2.3)
Then, under the conditions (G1) and (G2), the problem (P ) has a unique local solution u(t; x) 2 YT,
for T small enough.
The proof of theorem 2.1.1 will be established through several lemmas. The presence of the term
jujP 2u in the right hand side of our problem (P ) ; gives us negative values of the energy. For this
purpose we …xed u 2 C ([0; T ] ; H1
0( )) in the right hand side of (P ) and we will prove that our
problem (2:1) ; admits a solution.
Lemma 2.1.1 ([14]; Theorem 2.1) Let (u0; u1)2 (H01( ))
2
, assume that m 2; p 2 and (2:3)
holds. Then, under the conditions (G1) and (G2), there exists a unique weak solution v 2 YT to the
problem (2:1); for any u 2 C ([0; T ] ; H01( )) given.
The proof of the above Lemma follows the techniques due to Lions [26], in order to deal with the
convergence of the non linear terms in our problem, we must take …rst our initial data (u0; u1) in a
high regularity (that is, u0 2 H2( )
\ H1
0( ); and u1 2 H01( )\ L2(m 1)( )).
Lemma 2.1.2 ([26]; Theorem 3.1) Let u 2 C ([0; T ] ; H01( )). Suppose that
u0 2 H2( )\ H01( ); (2.4)
u1 2 H01( )\ L
2(m 1)( ); (2.5)
assume further that m 2; p 2. Then, under the conditions (G1) and (G2) there exists a unique
solution v of the problem (2:1) such that
vt2 L1 [0; T ] ; H01( ) ; (2.7)
vtt 2 L1 [0; T ] ; L2( ) ; (2.8)
vt2 Lm([0; T ] ( )) : (2.9)
The following technical Lemma will play an important role in the sequel.
Lemma 2.1.3 For any v 2 C1(0; T; H2( )) we have
Z Zt 0 g(t s) v(s):v0(t)dsdx = 1 2 d dt (g rv) (t) 1 2 d dt 2 4 t Z 0 g(s) Z jrv(t)j2dxds 3 5 1 2(g 0 rv) (t) +1 2g(t) Z jrv(t)j2dxds: where (g u)(t) = t Z 0 g(t s) Z ju(s) u(t)j2dxds:
The proof of this result is given in [31], for the reader’s convenience we repeat the steps here.
Proof. It’s not hard to see
Z Zt 0 g(t s) v(s):v0(t)dsdx = t Z 0 g(t s) Z rv0(t):rv(s)dxds = t Z 0 g(t s) Z rv0(t): [rv(s) rv(t)] dxds t Z 0 g(t s) Z rv0(t):rv(t)dxds: Consequently, Z Zt 0 g(t s) v(s):v0(t)dsdx = 1 2 t Z 0 g(t s)d dt Z jrv(s) rv(t)j2dxds t Z 0 g(s) 0 @d dt 1 2 Z jrv(t)j2dx 1 A ds
which implies, Z Zt 0 g(t s) v(s):v0(t)dsdx = 1 2 d dt 2 4 t Z 0 g(t s) Z jrv(s) rv(t)j2dxds 3 5 1 2 d dt 2 4 t Z 0 g(s) Z jrv(t)j2dxds 3 5 1 2 t Z 0 g0(t s) Z jrv(s) rv(t)j2dxds +1 2g(t) Z jrv(t)j2dxds:
This completes the proof. Proof of Lemma 2.1.2. Existence:
Our main tool is the Faedo-Galerkin’s method, which consist to construct approximations of the so-lutions, then we obtain a prior estimates necessary to guarantee the convergence of approximations. Our proof is organized as follows. In the …rst step, we de…ne an approach problem in bounded
dimension space Vn which having unique solution vn and in the second step we derive the various
a priori estimates. In the third step we will pass to the limit of the approximations by using the compactness of some embedding in the Sobolev spaces.
1. Approach solution:
Let V = H1
0( )\ H2( ) the separable Hilbert space. Then there exists a family of subspaces fVng
such that
i) Vn V (dimVn<1); 8n 2 N:
ii) Vn ! V; such that, there exist a dense subspace # in V and for all v 2 #; we can get sequence
fvngn2N 2 Vn;and vn! v in V:
iii) Vn Vn+1 and [n2N Vn= H01( )\ H2( ):
For every n 1; let Vn = Spanfw1; :::; wng ; where fwig, 1 i n; is the orthogonal complete
system of eigenfunctions of such that kwjk2 = 1, wj 2 H2( )\ L2(m 1)( ) for all j = 1; :::; n.
Denote by f jg the related eigenvalues, where wj are solutions of the following initial boundary
value problem 8 > > < > > : wj = jwj wj = 0 on j = 1; :::; in : (2.10)
According to (iii) ; we can choose vn0; vn12 [w1; :::; wn] such that vn0 n X j=1 jnwj ! u0 in H01( )\ H 2( ), (2.11) vn1 n X j=1 jnwj ! u1 in H01( )\ L 2(m 1)( ); (2.12) where jn= R u0wjdx jn= R u1wjdx: We seek n functions 'n 1; :::; 'nn2 C2[0; T ] ; such that vn(t) = n X j=1 'nj(t)wj(x); (2.13)
solves the problem 8 > > > > > > > > > > > > > > > > < > > > > > > > > > > > > > > > > : R ((v00 n(t) v0n(t) vn(t)) dx +R t R 0 g(t s) vn(s)ds +jvn0(t)j m 2 v0 n(t) dx =R ju(t)jp 2u(t) dx vn(0) = vn0; v0n(0) = vn1 ; (2.14)
where the prime "0" denotes the derivative with respect to t. For every 2 V
n and t 0: Taking
= wj; in (2:14) yields the following Cauchy problem for a ordinary di¤erential equation with
unknown 'n j: 8 > > > > > > > > > < > > > > > > > > > : '00n j (t) + j'0nj (t) + j'nj(t) + j t R 0 g(t s)'n j(s)ds + '0h j (t) m 2 '0n j (t) = j(t) 'nj(0) =R u0wj; '0nj (0) =R u1wj; j = 1; :::; n ; (2.15)
for all j; where
j(t) =
Z
By using the Caratheodory theorem for an ordinary di¤erential equation, we deduce that, the
above Cauchy problem yields a unique global solution 'n
j 2 H3[0; T ] ; and by using the embedding
Hm[0; T ] ,
! Cm 1[0; T ], we deduce that the solution 'n
j 2 C2[0; T ]. In turn, this gives a unique
vn de…ned by (2:13) and satisfying (2:14):
2. The a priori estimates:
The next estimate prove that the energy of the problem (2:1) is bounded and by using a result in
[9];we conclude that; the maximal time tn of existence of (2:15) can be extended to T .
The …rst a priori estimate:
Substituting = v0 n(t) into (2:14), we obtain Z v00n(t)v0n(t)dx Z vn0(t)vn0(t)dx Z vn(t)vn0(t)dx + Z Zt 0 g(t s) vn(s)dsvn0(t)dx + Z jvn0(t)j m 2 vn0(t)vn0(t)dx (2.16) = Z f (u)v0n(t)dx:
for every n 1; where f (u) = ju(t)jp 2u(t);Since the following mapping
L2( ) ! L2( )
u 7! jujp 2u
is continues, we deduce that,
jujp 2u2 L1 [0; T ] ; L2( ) :
So, f 2 H1([0; T ] ; H1( )). Consequently, by using the Lemma 2.1.3 and (G1) we get easily
1 2 d dtkv 0 n(t)k 2 2+krvn0(t)k 2 2+ 1 2 d dtkrvn(t)k 2 2 +1 2 d dt 2 4 t Z 0 g(t s) Z jrvn(s) rvn(t)j 2 dxds 3 5 1 2 d dt 0 @krvn(t)k 2 2 t Z 0 g(s):ds 1 A 1 2 t Z 0 g0(t s) Z jrvn(s) rvn(t)j2dxds + 1 2g(t)krvn(t)k 2 2 +kv0n(t)k m m = Z f (u):vn0(t)dx:
Therefore, we obtain d dtEn(t) Z f (u):v0n(t)dx +krvn0(t)k22+kv0n(t)kmm = 1 2(g 0 rvn)(t) 1 2g(t)krvn(t)k 2 2; where En(t) = 1 2 8 < :kv0n(t)k 2 2+ 0 @1 t Z 0 g(s)ds 1 A krvn(t)k 2 2+ (g rvn)(t) 9 = ;;
is the functional energy associated to the problem (2:1): It’s clear by using (G2), that
d
dtEn(t)
Z
f (u):vn0(t)dx +krv0n(t)k22+kvn0(t)kmm 0; 8t 0:
Which implies, by using Young’s inequality, for all > 0
d dtEn(t) +krv 0 n(t)k 2 2+kv0n(t)k m m 1 4 kf(u)k 2 2+ kvn0(t)k 2 2: (2.17)
Integrating (2:17) over [0; t] ; (t < T ), we obtain
En(t) + t Z 0 krvn0(s)k 2 2ds + t Z 0 kvn0(s)k m mds 1 4 T Z 0 kf(u)k22ds + T Z 0 kv0n(s)k 2 2ds + 1 2 ku1nk 2 2+kru0nk 2 2 : Since f 2 H1(0; T; H1 0( )); we deduce En(t) + t Z 0 krvn0(s)k 2 2ds + t Z 0 kvn0(s)k m mds CT ku1nk 2 2+kru0nk 2 2 ; (2.18)
for, small enough and every n 1, where CT > 0 is positive constant independent of n:
Then, by the de…nition of En(t) and by using (2:11) ; (2:12), we get
kvn0(t)k 2 2+ 0 @1 t Z 0 g(s)ds 1 A krvn(t)k 2 2+ (g rvn)(t) KT; (2.19) and t Z 0 kvn0(s)k m mds KT; (2.20)
and t Z 0 krv0n(s)k 2 2ds KT; (2.21)
where KT = CT ku1nk22+kru0nk22 ; by (2:19), we get tn= T; 8n:
However, the insu¢ cient regularity of the nonlinear operator, jvtjm 2vt; with the presence of the
viscoelastic term and strong damping, we must prove in the next a prior estimates that, the family of
approximations vnde…ned in (2:13) is compact in the strong topology and by using compactness of
the embedding H1([0; T ] ; H1( )) ,
! L2([0; T ] ; L2( )); we can extract a subsequence of vn denoted
also by v such that v0 converges strongly in L2([0; T ] ; L2( )): To do this, it’s su¢ ces to prove
that v0n is bounded in L1([0; T ] ; H1
0( )) and v
00
n is bounded in L1([0; T ] ; L2( )), then by using
Aubin-Lions Lemma, our conclusion holds. The second a priori estimate:
Substituting = wj in (2:14) and taking wj = jwj, multiplying by '
0n
j (t)and summing up the
product result with respect to j; we get by Green’s formula. Z rvn00:rv0ndx + Z vn0: vn0dx + Z vn: vn0dx Z Zt 0 g(t s) vn: vn0dsdx + n X i=1 Z @ @xi jv 0 njm 2vn0 @v 0 n @xi dx (2.22) = Z rf(u):rvn0dx: As in [26], we have n X i=1 Z @ @xi jv 0 njm 2vn0 @v 0 n @xi dx = (m 1) n X i=1 Z @ @xi jv 0 nj m 2 2 @v 0 n @xi 2 dx (2.23) = 4(m 1) m2 n X i=1 Z @ @xi jv 0 nj m 2 2 v0 n 2 dx:
Also, the fourth term in the left hand side of (2:22) can be written as follows Z Zt 0 g(t s) vn: v0ndsdx = 1 2g(t)k vnk 2 2+ 1 2(g 0 v n) 1 2 d dt 8 < :(g vn) k vnk 2 2 t Z 0 g(s)ds 9 = ;: (2.24) Therefore, (2:22) becomes 1 2 d dt 2 4krv0nk2 2+ (g vn) +k vnk 2 2 0 @1 t Z 0 g(s)ds 1 A 3 5 +4(m 1) m2 n X i=1 Z @ @xi jv 0 nj m 2 2 v0 n 2 dx (2.25) +k v0nk22+ 1 2g(t)k vnk 2 2 1 2(g 0 v n) = Z rf(u):rvn0dx:
Let us de…ne the energy term
Kn(t) = 1 2 2 4krv0nk2 2+ (g vn) +k vnk 2 2 0 @1 t Z 0 g(s)ds 1 A 3 5 (2.26)
Then, it’s clear that (2:25) takes the form d dtKn(t) Z rf(u):rvn0dx +k vnk0 2 2 + n X i=1 Z @ @xi jv 0 nj m 2 2 v0 n 2 dx (2.27) = 1 2(g(t)k vnk 2 2) + 1 2(g 0 v n): Using (G1) ; (G2) and integrating (2:27) over [0; t] ; we obtain
Kn(t) + n X i=1 t Z 0 Z @ @xi jv 0 nj m 2 2 v0 n 2 dxds + t Z 0 k v0nk22ds T Z 0 Z rf(u):rv0ndxds + 1 2(krv1nk 2 2+k v0nk 2 2): (2.28)
Obviously, by using Young’s inequality, we get T Z 0 Z rf(u):rvn0dxds 1 4 T Z 0 krf(u)k22ds + T Z 0 krv0nk22ds:
Inserting the above estimate into (2:28); to get
Kn(t) + n X i=1 t Z 0 Z @ @xi jv 0 nj m 2 2 v0 n 2 dxds + t Z 0 k vnk0 22ds CT(krv1nk22+k v0nk22): Thus, Kn(t) + n X i=1 t Z 0 Z @ @xi jv 0 nj m 2 2 v0 n 2 dxds + t Z 0 k v0nk22ds CT;
for, small enough and every n 1, where CT > 0is positive constant independent of n: Therefore,
this equivalent by the de…nition of Kn(t)
krv0nk22+ (g vn) +k vnk 2 2 0 @1 t Z 0 g(s)ds 1 A CT; (2.29) and n X i=1 t Z 0 Z @ @xi jv 0 nj m 2 2 v0 n 2 dxds CT; (2.30) and t Z 0 k v0nk22ds CT: (2.31)
Then, from (2:29), (2:30) and (2:31) ; we conclude
vn0 is bounded in L1([0; T ] ; H01( )); (2.32) vn is bounded in L1([0; T ] ; H2( )); (2.33) @ @xi jv 0 nj m 2 2 v0 n is bounded in L 2([0; T ] ; L2( )); i = 1; :::; n: (2.34)
The third a priori estimate: It’s clear that
d dt 2 4 t Z 0 g(t s) vn(s)ds 3 5 = g(0) vn+ t Z 0 g0(t s) vn(s)ds: (2.35)
Performing an integration by parts in (2:35) we …nd that d dt 2 4 t Z 0 g(t s) vn(s)ds 3 5 = g(t) un0+ t Z 0 g(t s) v0n(s)ds: (2.36)
Now, returning to (2:14), di¤erentiating throughout with respect to t, and using (2:36), we obtain Z (vn000(t) vn00(t) v0n(t)) dx + Z 0 @ t Z 0 g(t s) v0n(s)ds + g(t) un0+ (m 1)(jvn0(t)j m 2 vn00(t)) 1 A dx = Z (f (u))0) dx; (2.37) where, (f (u))0 = @f @t. By substitution of = v 00 h(t) in (2:37); yields Z vn000(t)vn00(t)dx Z vn00(t)vn00(t)dx Z vn0(t)vn00(t)dx + Z 0 @ t Z 0 g(t s) vn0(s)ds + g(t) un0 1 A v00 n(t)dx +(m 1) Z jv0n(t)j m 2 v00n(t)v00n(t)dx (2.38) = 0 Z (f (u))0vn00(t)dx:
The fourth term in the left hand side of (2:38) can be analyzed as follows. It’s clear that Lemma 2.1.3 implies Z Zt 0 g(t s) vn0(s):v00n(t)dsdx = Z Zt 0 g(t s)rv0n(s):rvn00(t)dsdx = 1 2 d dt (g rv 0 n) (t) 1 2 d dt t Z 0 g(s)krv0n(t)k22ds (2.39) 1 2(g 0 rv0 n) (t) + 1 2g(t)krv 0 n(t)k 2 2:
Also, (m 1)hjvn0(t)jm 2vn00(t); v00n(t)i = 4(m 1) m2 Z @ @t jv 0 nj m 2 2 v0 n(t) 2 dx: (2.40)
Inserting (2:39) and (2:40) into (2:38), we obtain 1 2 d dt kv 00 nk22 +krv00nk 2 2+ 1 2 d dtkrv 0 nk22+ 1 2 d dt 8 < :(g rvn0) (t) krv0nk 2 2 t Z 0 g(s)ds 9 = ; +4(m 1) m2 Z @ @t jv 0 nj m 2 2 v0 n(t) 2 dx + (g(t) un0) Z v00n(t)dx (2.41) = Z (f (u))0v00n(t)dx + 1 2(g 0 rv0n) (t) 1 2g(t)krv 0 nk22: Let us denote by n(t) = 1 2 8 < :kv00nk 2 2+ + (g rv0n) (t) + (1 t Z 0 g(s)ds)krv0nk22 9 = ;: (2.42)
We obtain, from (2:41) ; (G2) that d dt n(t) Z (f (u))0:vn00(t)dx +4(m 1) m2 Z @ @t jv 0 nj m 2 2 v0 n(t) 2 dx +krvnk00 22+ (g(0) un0) Z vn00(t)dx = 1 2(g 0 rv0 n) (t) 1 2g(t)krv 0 n(t)k 2 2 0:
Integration the above estimate over [0; t], we conclude
n(t) + 4(m 1) m2 t Z 0 Z @ @t jv 0 nj m 2 2 v0 n(t) 2 dxds + t Z 0 krvnk00 22ds + (g(0) un0) t Z 0 Z vn00(s)dxds 1 2kv 00 n0k22+ 1 2krvn1k 2 2+ 1 4 T Z 0 (f (u))0 22ds + T Z 0 kvnk00 22ds:
Then, for small enough, we deduce n(t) + 4(m 1) m2 t Z 0 Z @ @t jv 0 nj m 2 2 v0 n 2 dxds + t Z 0 krvnk00 22ds + (g(0) un0) t Z 0 Z v00ndxds (2.43) CT kv00n0k 2 2+krvn1k 2 2 :
In order to estimate the term kvn0k00 2
2, taking t = 0 in (2:14), we …nd kv00n0k22 = Z vn1:vn000 dx + Z vn0:vn000 dx Z jvn1jm 2vn1:vn000 dx + Z f (u(0)):vn000 dx:
Thanks to Cauchy-Schwartz inequality (Lemma 1.2.10), we write
kv00n0k22 kv00n0k2 0 B B @k vn1k2+k vn0k2+kf(u(0))k2+ 0 @Z jv1nj2(m 1)dx 1 A 1 2 1 C C A ; which implies, by using (2:11) and (2:12) ; that
kv00n0k2 k vn1k2+k vn0k2+kf(u(0))k2+ 0 @Z jv1nj 2(m 1) dx 1 A 1 2 C: (2.44) Then, n(t) LT; (2.45) 4(m 1) m2 t Z 0 Z @ @t jv 0 nj m 2 2 v0 n 2 dxds LT; (2.46) and t Z 0 krv00nk22ds LT; (2.47) and t Z 0 Z vn00dxds LT: (2.48)
where LT > 0:
From (2:11); (2:12) and (2:44) (2:48) ; we deduce
vn00 is bounded in L1([0; T ] ; L2( )); (2.49) vn is bounded in L1([0; T ] ; H01( )); (2.50) @ @t jv 0 nj m 2 2 v0 n is bounded in L 2([0; T ] ; L2( )); i = 1; :::; n: (2.51)
3.Pass to the limit:
By the …rst, the second and the third estimates, we obtain
vn is bounded in L1([0; T ] ; H2( )\ H01( )); (2.52)
vn0 is bounded in L1([0; T ] ; H01( )); (2.53)
vn00 is bounded in L1([0; T ] ; L2( )); (2.54)
v0n is bounded in Lm((0; T ) ): (2.55)
Therefore, up to a subsequence, and by using the (Theorem 1.1.5), we observe that there exists a
subsequence v of vn and a function v that we my pass to the limit in (2:14); we obtain a weak
solution v of (2:1) with the above regularity
v * v in L1([0; T ] ; H01( )\ H2( ));
v0 * v0in L1([0; T ] ; H01( )\ Lm( ));
v00* v00in L1([0; T ] ; L2( )):
By using the fact that
L1([0; T ] ; L2( )) ,! L2([0; T ] ; L2( )); L1([0; T ] ; H01( )) ,! L2([0; T ] ; H01( )): We get vn0 is bounded in L2([0; T ] ; H01( )); vn00 is bounded in L2([0; T ] ; L2( )); therefore, vn0 is bounded in H1([0; T ] ; H1( )): (2.56)
Consequently, since the embedding
is compact, then we can extract a subsequence v0 such that v0 ! v0 in L2([0; T ] ; L2( )). (2.57) which implies v0 ! v0 a.e on (0; T ) . By (2:20), we have vn0 is bounded in Lm([0; T ] ):
and by using Theorem 1.2.2
jv0jm 2v0 * # in Lm0([0; T ] ; Lm0( ))
where m
m 1 = m
0:
The estimates (2:34) and (2:46) imply that jv0j
m 2
2 v0 * in H1([0; T ] ; H1( )
by using the fact that, the mapping u 7 ! jujm 2uis continuous, (Lemma 1.2.9) and since the weak
topology is separate, we deduce
# =jv0jm 2v0 (2.58)
=jv0j
m 2
2 v0 (2.59)
Then, by using the uniqueness of limit, we deduce
jv0jm 2v0 *jv0jm 2v0 in Lm0([0; T ] ; Lm0( )) (2.60)
jv0jm22 v0 *jv0j
m 2
2 v0 in H1([0; T ] ; H1( ) (2.61)
Now, we will pass to the limit in (2:14), by the same techniques as in [26]:
Taking = wj, n = and …xed j < ;
Z v00(t):wjdx + Z rv (t):rwjdx + Z rv0(t):rwjdx Z Zt 0 g(t s)rv (s):rwjdsdx + Z jv0(t)jm 2v0(t):wjdx (2.62) = Z f (u):wjdx:
We obtain, by using the property of continuous of the operator in the distributions space Z v00(t):wjdx * Z v00(t):wjdx; in D0(0; T ) Z rv (t):rwjdx * Z rv(t):rwjdx; in L1(0; T ) Z rv0(t):rwjdx * Z rv0(t):rwjdx; in L1(0; T ) Z Zt 0 g(t s)rv (s):rwjdsdx * Z Zt 0 g(t s)rv(s):rwjdsdx; in L1(0; T ) Z jv0(t)jm 2v0(t):wjdx * Z jv0(t)jm 2v0(t):wjdx; in L1(0; T )
We deduce from (2:62), that Z v00(t):wjdx + Z rv(t):rwjdx + Z rv0(t):rwjdx Z Zt 0 g(t s)rv(s):rwjdsdx + Z jv0(t)jm 2v0(t):wjdx (2.63) = Z f (u):wjdx:
Since, the basis wj (j = 1; :::)is dense in H1
0( )\ H2( ) ; we can generalize (2:63) ; as follows
Z v00(t):'dx + Z rv(t):r'dx + Z rv0(t):r'dx Z Zt 0 g(t s)rv(s):r'dsdx + Z jv0(t)jm 2v0(t):'dx = Z f (u):'dx; 8' 2 H01( )\ H2( ) : Then, v 2 L1 [0; T ] ; H2( )\ H01( ) ; vt 2 L1 [0; T ] ; H01( ) ; vtt 2 L1 [0; T ] ; L2( ) ; vt 2 Lm([0; T ] ( )) :
Uniqueness:
Let v1; v2 two solutions of (2:1), and let w = v1 v2 satisfying :
w00 w w0+
t Z
0
g(t s) wds + jv1j0 m 2v01 jv2j0 m 2v20 = 0: (2.64)
Multiplying (2:64), by w0 and integrating over , we get
1 2 d dt 0 @kw0 n(t)k 2 2 + 0 @1 t Z 0 g(s)ds 1 A krwn(t)k22+ (g rwn)(t) 1 A + Z jv1j0 m 2v10 jv2j0 m 2v20 w0dx = krw0n(t)k22+ 1 2(g 0 rw n)(t) 1 2g(t)krwn(t)k 2 2: Denote by J (t) =kw0n(t)k22+ 0 @1 t Z 0 g(s)ds 1 A krwn(t)k22+ (g rwn)(t): (2.65)
Since the function y 7! jyjm 2y is increasing, we have
Z jv01jm 2v01 jv2j0 m 2v02 w0dx 0 and since 1 2(g 0 rw n)(t) 0; we deduce d dtJ (t) 0 : (2.66)
This implies that J (t) is uniformly bounded by J (0) and is decreasing in t; since w(0) = 0; we
obtain w = 0 and v1 = v2:
Proof of Lemma 2.1.1.
As in [14]; since D( ) = H2( ), we approximate, u0; u1 by sequences (u 0) ; (u 1) in D( ), and u
by a sequence (u ) in C ([0; T ] ; D( )), for the problem (2:1). Lemma 2.1.2 guarantees the existence
of a sequence of unique solutions (v ) satisfying (2:6) (2:9) :Now, to complete the proof of Lemma
2.1.1, we proceed to show that the sequence (v ) is Cauchy in YT equipped with the norm
kuk2YT =kuk
2
H +kutk
2
where kuk2H = max0 t T 8 < : Z u2t + ljruj2 (x; t)dx 9 = ;
Denote w = v v for ; given. Then w is a solution of the Cauchy problem:
8 > > > > > > > > > > > > < > > > > > > > > > > > > : wtt w wt+ L(w) + k(vt) k(vt) = f (u ) f (u ); x2 ; t > 0 w(0; x) = u 0 u 0; wt(0; x) = u 1 u 1; x2 w(t; x) = 0; x2 ; t > 0 ; (2.67) where, k(vt) = jvtjm 2vt f (u ) =ju jp 2u L(w) = t Z 0 g(t s) w(s; x)ds:
The energy equality reads as 1 2 d dt 8 < :kwtk 2 2+ 0 @1 t Z 0 g(s)ds 1 A krwk2 2+ (g rw)(t) 9 = ; + Z k(vt) k(vt) wtdx +krwtk22 (2.68) = Z f (u ) f (u ) wtdx + 1 2(g 0 rw) (s)ds 1 2krw(s)k 2 2 t Z 0 g(s)ds: The term, Z k(vt) k(vt) wtdx = Z jvtjm 2vt vt m 2 vt vt vt dx is nonnegative. We need to estimate Z
ful…lled for u; u; v; v 2 H1
0( ); where C is a constant depending on ; l; p only. Then, Hölder’s
inequality yields, for 1
q + 1 n + 1 2 = 1 (q = 2n n 2); Z f (u ) f (u ) wtdx = Z ju jp 2u u p 2u vt vt dx C u u Lq vt vt L2 ku k p 2 n(p 2)+ u p 2 n(p 2) : (2.69)
The Sobolev embedding Lq ,
! H1 0( ) gives u u Lq( ) C ru ru L2( ): Then, ku kp 2n(p 2)+ u p 2 n(p 2) C(ku k p 2 L2( )+ u p 2 L2( )):
The necessity to estimate ku kn(p 2) by the energy norm kukH requires a restriction on p. Namely,
we need n(p 2) 2n
n 2; then the Sobolev embedding L
q , ! H1 0( ) gives ku kp 2n(p 2) kuk p 2 H :
Therefore, (2:69) takes the form Z f (u ) f (u ) wtdx C vt vt L2( ) ru ru L2( ) kru k p 2 L2( )+ ru p 2 L2( ) (2.70)
under the fact that
t Z 0 (g0 rw) (s)ds 0; we conclude t Z 0 (g0 rw) (s)ds + (g rw) (t) + krw(t)k22 t Z 0 g(s)ds 0: Thus, 1 2kw(t; :)k 2 H 1 2kw(0; :)k 2 H + C t Z 0 ru ru L2( )kwt(s; :)kHds:
The Gronwell Lemma and Young’s inequality guarantee that
Since
v (t; :) v (t; :) H C v (0; :) v (0; :) H + CT u u C([0;T ];H); (2.71)
then fv g is a Cauchy sequence in C([0; t] ; H); since fu g and fv (0; :)g are Cauchy sequences in
C([0; T ] ; H) and H, respectively.
Now, we shall prove that fvtg is a Cauchy sequence, in Lm([0; T ] ), to control the norm
kvtk2Lm([0;T ] ): By the following algebraic inequality
j jm 2 j jm 2 ( ) Cj jm; (2.72)
which holds for any real ; and c, we get
Z k(vt) k(vt) wtdx = Z vt jvtjm 2 vt vt m 2 vt vt dx C vt vt 2 Lm([0;T ] ):
This estimate combined with (2:68) gives
vt vt 2 Lm([0;t] ) C v (0; :) v (0; :) Lm([0;t] ) +CR t Z 0 u u Lm([0;t] ) v (t; :) v (t; :) Lm([0;t] )ds:
So by using Gromwell Lemma, we obtain fvtg is a Cauchy sequence, in Lm([0; T ] )and hence
fv g is a Cauchy sequence in YT:Let v its limit in YT and by Lemma 2.1.2, v is a weak solution of
(2:1).
Now, we are ready to show the local existence of the problem (P ) Proof of Theorem 2.1.1.
Let (u0; u1)2 (H01( ))
2
; and
R2 = kru0k22+ku1k22 :
For any T > 0; consider
MT = u2 YT : u(0) = u0; ut(0) = u1 and kukYT R :
Let
: MT ! MT
We will prove as in [13] that,
(i) (MT) MT:
(ii) is contraction in MT:
Beginning by the …rst assertion. By Lemma 2.1.1, for any u 2 MT we may de…ne v = (u);
the unique solution of problem (2:1). We claim that, for a suitable T > 0, is contractive map
satisfying
(MT) MT:
Let u 2 MT; the corresponding solution v = (u)satis…es for all t 2 [0; T ] the energy identity :
1 2 8 < :kv0(t)k 2 2+ 0 @1 t Z 0 g(s)ds 1 A krv(t)k2 2+ (g rv)(t) 9 = ; + t Z 0 krv0(s)k22ds + t Z 0 kv0(s)kmm)ds (2.73) = 1 2 kv1k 2 2+krv0k 2 2 + t Z 0 Z ju(s)jp 2u(s)v0(t)dxds: We get 1 2kv(t)k 2 YT 1 2kv(0)k 2 YT + t Z 0 Z ju(s)jp 2u(s)v0(t)dxds: (2.74)
We estimate the last term in the right-hand side in (2:74) as follows: thanks to Hölder’s, Young’s
inequalities, we have Z
ju(s)jp 2u(s)v0(t)dx CkukpY
TkvkYT; then, kv(t)k2YT kv(0)k 2 YT + CR p t Z 0 kvkYT ds;
where C depending only on T; R: Recalling that u0; u1 converge, then
kv(t)kYT kv(0)kYT + CR
pT:
Choosing T su¢ ciently small, we getkvkYT R; which shows that
Now, we prove that is contraction in MT:Taking w1 and w2 in MT, subtracting the two equations
in (2:1); for v1 = (w1) and v2 = (w2);and setting v = v1 v2;we obtain for all 2 H01( ) and
a.e. t 2 [0; T ] Z vtt: dx + Z rvr dx + Z rvtr dx + Z Zt 0 g(t s)rvr dsdx + Z jvtjm 2vt dx = Z jw1jp 2w1 jw2jp 2w2 dx: (2.75)
Therefore, by taking = vt in (2:75) and using the same techniques as above, we obtain
kv(t; :)k2YT C t Z 0 kw1k p 2 YT +kw2k p 2 YT kw1 w2kYTkv(s; :)k 2 YT ds: (2.76)
It’s easy to see that
kv(t; :)k2Yt =k (w1) (w2)k
2
Yt kw1 w2k
2
Yt; (2.77)
for some 0 < < 1 where = 2CT RP 2:
Finally by the contraction mapping theorem together with (2:77); we obtain that there exists a
unique weak solution u of u = (u) and as (u) 2 YT we have u 2 YT: So there exists a unique
weak solution u to our problem (P ) de…ned on [0; T ] ; The main statement of Theorem 2.1.1 is proved.
Remark 2.1.1 Let us mention that in our problem (P ) the existence of the term source ( f (u) =
jujp 2u ) in the right hand forces us to use the contraction mapping theorem. Since we assume a
little restriction on the initial data. To this end, let us mention again that our result holds by the well depth method, by choosing the initial data satisfying a more restrictions.
Global Existence and Energy Decay
Abstract
In this chapter, we prove that the solution obtained in the second chapter (Local solution ) is global in time: In addition, we show that the energy of solutions decays exponentially if m = 2 and polynomial if m > 2, provided that the initial data are small enough. The existence of the source
term jujp 2u forces us to use the potential well depth method in which the concept of so-called
stable set appears. We will make use of arguments in [44] with the necessary modi…cations imposed by the nature of our problem.