This article was downloaded by: [National Sun Yat-Sen University]
On: 29 July 2012, At: 19:51 Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Applicable Analysis: An International Journal
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/gapa20
On a class of fractional order
differential inclusions with infinite delays
Tran Dinh Ke a , Valeri Obukhovskii b , Ngai-Ching Wong c & Jen- Chih Yao d
a Department of Mathematics, Hanoi National University of Education, 136 Xuan Thuy, Cau Giay, Hanoi, Vietnam
b Faculty of Mathematics, Voronezh State University, Universitetskays Pl. 1, 394006 Voronezh, Russia
c Department of Applied Mathematics, National Sun Yat-Sen University, Kaohsiung 804, Taiwan
d Center for General Education, Kaohsiung Medical University, Kaohsiung 807, Taiwan
Version of record first published: 28 Jul 2011
To cite this article: Tran Dinh Ke, Valeri Obukhovskii, Ngai-Ching Wong & Jen-Chih Yao (2011):
On a class of fractional order differential inclusions with infinite delays, Applicable Analysis: An International Journal, DOI:10.1080/00036811.2011.601454
To link to this article: http://dx.doi.org/10.1080/00036811.2011.601454
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.tandfonline.com/page/terms-and- conditions
This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any
sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
2011, 1–23, iFirst
On a class of fractional order differential inclusions with infinite delays
Tran Dinh Kea, Valeri Obukhovskiib, Ngai-Ching Wongc* and Jen-Chih Yaod
aDepartment of Mathematics, Hanoi National University of Education, 136 Xuan Thuy, Cau Giay, Hanoi, Vietnam;bFaculty of Mathematics, Voronezh State University, Universitetskays Pl. 1, 394006 Voronezh, Russia;cDepartment of Applied Mathematics, National Sun Yat-Sen University, Kaohsiung 804, Taiwan;dCenter for General Education,
Kaohsiung Medical University, Kaohsiung 807, Taiwan Communicated by D.-C. Chang
(Received 6 December 2010; final version received 25 June 2011) Our aim is to study fractional order differential inclusions with infinite delays in Banach spaces. We impose the regularity condition on multi- valued nonlinearity in terms of measures of noncompactness to get the existence result. Some properties of the solution map are proved.
Keywords: fractional order differential inclusion; fractional derivative;
functional differential inclusion; infinite delay; initial value problem;
continuous dependence; multivalued map; measure of noncompactness;
fixed point; condensing map
AMS Subject Classifications: 34K37; 26A33; 34A60; 34G25; 34K09;
34K30; 47H04; 47H08; 47H10; 47H11
1. Introduction
LetEbe a Banach space. We are concerned with the following problem:
CDuðtÞ 2Fðt,ut,rNuÞ, t2J:¼ ½0,T, ð1:1Þ
rNuð0Þ ¼U0, ð1:2Þ
uðsÞ ¼’ðsÞ, s2 ð1, 0Þ, ð1:3Þ
whereN1 is an integer,2(N1,N],u: (1,T]!Eis the unknown function,
CD denotes the Caputo fractional derivative, rNu¼(u,u0,. . .,u(N1)) and F: [0,T] B EN! P(E) is a multivalued map with nonempty compact convex values. HereP(E) stands for the collection of all subsets ofE,Bis a phase space of delays and ut2 B is the history of the state function u up to the time t, that is
*Corresponding author. Email: wong@math.nsysu.edu.tw
ISSN 0003–6811 print/ISSN 1563–504X online ß2011 Taylor & Francis
DOI: 10.1080/00036811.2011.601454 http://www.informaworld.com
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
ut(s)¼u(tþs) fors2(1, 0]. The initial dataU0¼ ðeu0,eu1,. . .,euN1Þare given inEN and the initial function’2 Bis such that’ð0Þ ¼eu0.
The subject of fractional differential equations has recently received much attention due to its important applications in modelling phenomena of science and engineering. The employment of differential equations with fractional order allows to deal with many problems in numerous areas including fluid flow, rheology, electrical networks, viscoelasticity, electrochemistry, etc. For complete references, we refer to some significant works, e.g., the monographs of Kilbas et al. [1], Kiryakova [2], Miller and Ross [3] and Podlubny [4]. In the past few years, there has been a great contribution in fractional differential equations. Let us refer to some relevant works in [5–18]. With initial or boundary conditions, some particular cases of (1.1) without delay were studied. The equation
CDuðtÞ ¼fðt,uðtÞÞ or the inclusion
CDuðtÞ 2Fðt,uðtÞÞ
in the cases 2(0, 1] or 2(1, 2] were considered in [6,8–11,13]. Similar problems with the Riemann–Liouville fractional derivatives were also investigated, for instance, in [12,17,18]. The readers can find more works in the survey of Argawal et al. [19].
In addition, inclusion (1.1) can be seen as a generalized model of high-order ordinary differential equations, an example of which is the equation considered in [20]:
uðnÞðtÞ ¼fðt,uðtÞ,u0ðtÞ,. . .,uðn1ÞðtÞ,ðTuÞðtÞÞ, whereT is an integral operator.
There are several approaches that can be used to get the solvability for local, global or extremal solutions of fractional differential equations or inclusions. One can use the method of upper and lower solutions to obtain the results as in [9].
Another approach develops a comparison principle and iteration schemes to receive the existence results as in [14–16]. The method that is widely used consists of transforming problems to corresponding fixed point equations or inclusions, followed by applying some known fixed point theorems.
Our approach is employing the fixed point theory technique for multivalued condensing maps under assumptions expressed in terms of the measure of noncompactness (MNC). The method used in this note allows us to solve the problems of differential inclusions in infinite-dimensional spaces with a general form of nonlinearity. Furthermore, by using this method, we need not impose the Lipschitz condition on the nonlinearity. Instead of Lipschitz assumptions, we suppose that the nonlinearityFsatisfies a regularity condition expressed in terms of the Hausdorff MNC. In the sequel, we define a suitable MNC, prove that the solution multioperator is condensing with respect to this new MNC and find solutions by using the fixed point theory of condensing multimaps presented in [21].
Besides [21], an employment of this approach can be found, e.g., in [17,22] and other works. For more applications of multivalued analysis to differential equations
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
and inclusions, the readers are referred to [23–25]. The reader may also find a complete reference to MNCs in the monographs [26,27].
The rest of this article is organized as follows. In the next section we recall some basic facts related to fractional calculus, measures of noncompactness, multivalued maps and the phase space for delay differential equations. Section 3 is devoted to the formulations and proofs of the local and global existence results. In the last section, we study the continuous dependence of the solution set of the stated problem on initial data.
2. Preliminaries 2.1. Fractional calculus
We start this section with some notation and definitions in fractional calculus. For the motivations of these definitions, see, for example [1–4].
Definition 2.1 The fractional integral of order40 of a functionf2L1(0,T;E) is defined by
I0 fðtÞ ¼ 1 ðÞ
Zt
0
ðtsÞ1fðsÞds, whereis the Gamma function.
We use Bochner integral in the foregoing definition and in the rest of this work.
Definition 2.2 For a functionf2CN([0,T];E), the Caputo fractional derivative of order2(N1,N] is defined by
CD0 fðtÞ ¼ 1 ðNÞ
Z t
0
ðtsÞN1fðNÞðsÞds:
It should be noted that there are some notions of fractional derivatives in which the Riemann–Liouville and Caputo definitions have been used widely. Many application problems, expressed by differential equations of fractional order, require initial conditions related to u(0), u0(0), etc., and the Caputo fractional derivative satisfies these demands. Foru2CN([0,T];E), we have the following formulae:
CD0I0uðtÞ ¼uðtÞ, ð2:1Þ
I0CD0uðtÞ ¼uðtÞ XN1
k¼0
uðkÞð0Þ
k! tk: ð2:2Þ
2.2. Phase space
LetBbe a linear space, with a seminormj jB, consisting of functions mapping (1, 0] intoE. The definition of the phase spaceB, introduced by Hale and Kato [28], can be given by the following axioms. Ifv: (1,T]!Eis such thatvj[0,T]2C([0,T];E) andv02 B, then
(B1)vt2 B for allt2[0,T];
(B2) the functiont°vtis continuous on [0,T];
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
(B3) jvtjB4K(t)sup{kv(s)kE: 04s4t}þM(t)jv0jB, where K,M: [0,T]![0,1), K is continuous,Mis bounded and they are independent ofv.
We may consider the following examples of phase spaces satisfying all the above properties.
(1) For 40, let B ¼C be the space of continuous functions : (1; 0]!E having a limit lim!1e ðÞwith
j jB¼ sup
150
ek ðÞk:
(2) (Spaces of ‘fading memory’). Let B ¼C be the space of functions : (1; 0]!Esuch that
(a) is continuous on [r; 0],r40;
(b) is Lebesgue measurable on (1;r) and there exists a nonnegative Lebesgue integrable function : (1;r)!Rþ such that is Lebesgue integrable on (1;r); moreover, there exists a locally bounded functionP: (1; 0]!Rþsuch that, for all0,(þ)P()() a.e.
2(1;r). Then, j jB¼ sup
r0
k ðÞk þ Z r
1
ðÞk ðÞkd:
A simple example of such a space is given by ()¼e,2R. For more examples of phase spaces, see [28].
2.3. Measures of noncompactness and multivalued maps
Let us recall some basic facts from the multivalued analysis, which will be used in this article. LetEbe a Banach space. We denote
. P(E)¼{A E:A6¼ ;},
. Pv(E)¼{A2 P(E) :Ais convex}, . K(E)¼{A2 P(E) :Ais compact}, . Kv(E)¼Pv(E)\K(E).
We will use the following definition of the MNC (see, e.g. [21]).
Definition 2.3 Let (A,5) be a partially ordered set. A function :P(E)! A is called an MNC inEif
ðcoÞ ¼ðÞ for every2 PðEÞ,
wherecois the closure of the convex hull of. An MNC is called (i) monotone, if0,12 P(E),0 1implies(0)4(1);
(ii) nonsingular, if({a}[)¼() for anya2 E,2 P(E);
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
(iii) invariant with respect to union with compact set, if (K[)¼() for every relatively compact setK Eand2 P(E).
IfAis a cone in a normed space, we say that is
(iv) algebraically semi-additive, if (0þ1)4(0)þ(1) for any 0, 12 P(E);
(v) regular, if()¼0 is equivalent to the relative compactness of.
An important example of MNC is the Hausdorff MNC, which satisfies all the above properties:
ðÞ ¼inff": has a finite"-netg:
It should be mentioned that the Hausdorff MNC has also the following properties:
. Semihomogeneity: (t)4jtj () for any2 P(E) andt2R;
. in a separable Banach spaceE, ðÞ ¼limm!1supx2dðx,EmÞ, where {Em} is a sequence of finite-dimensional subspaces of E such that Em Emþ1, m¼1, 2,. . . andS1
m¼1Em¼ E.
Let Xbe a metric space.
Definition 2.4 A multivalued map (multimap)F:X! P(E) is said to be:
(i) upper semicontinuous (u.s.c) if F1(V)¼{x2X:F(x) V} is an open subset ofXfor every open setV E;
(ii) closed if its graphF¼{(x,y) :y2 F(x)} is a closed subset ofX E;
(iii) compact if its rangeF(X) is relatively compact inE;
(iv) quasicompact if its restriction to any compact subset A Xis compact.
Definition 2.5 A multimapF:X E !K(E) is said to be condensing with respect to an MNC (-condensing) if for every bounded set X that is not relatively compact, we have
ðF ðÞÞ 64
¼ ðÞ:
Suppose that D E is a nonempty closed convex subset of E and UD is a nonempty relatively open subset ofD. We denote byUDand@UD, the closure and the boundary ofUDin the relative topology ofD, respectively.
Letbe a monotone nonsingular MNC inE. The application of the topological degree theory for condensing multimaps (see, e.g. [21]) yields the following fixed point theorems.
THEOREM 2.1 [21, Corollary 3.3.1] Let M be a bounded convex closed subset ofE and F:M !Kv(M) an u.s.c. -condensing multimap. Then the fixed point set FixF:¼{x:x2 F(x)}is a nonempty compact set.
The following theorem presents a version for multimaps of the classical Leray–Schauder alternative.
THEOREM 2.2 [21, Corollary 3.3.3] LetUDbe a bounded open neighbourhood of a2D andF :UD!KvðDÞan u.s.c-condensing multimap satisfying the boundary condition
xa 62ðF ðxÞ aÞ
for all x2@UDand0541.Then FixF is a nonempty compact set.
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
Definition 2.6 Let G: [0,T]!K(E) be a multifunction and p51.Then G is said to be
. Lp-integrable, if it admits a Bochner Lp-integrable selection. That is there exists g: [0,T]!E, g(t)2G(t) for a.e. t2[0,T] such that RT
0 kgðsÞkpEds51;
. Lp-integrably bounded, if there exists a function2Lp([0,T]) such that kGðtÞk:¼supfkgkE:g2GðtÞg4ðtÞ for a.e.t2 ½0,T:
The set of allLp-integrable selections ofGwill be denoted bySpG.
The multifunctionGis calledmeasurableifG1(V) measurable (with respect to the Lebesgue measure onJ:¼[0,T]) for any open subsetVof E. We say thatGis strongly measurableif there exists a sequenceGn: [0,T]!K(E),n¼1, 2,. . .of step multifunctions such that
n!1lim HðGnðtÞ,GðtÞÞ ¼0 for a.e.t2 ½0,T, whereHis the Hausdorff metric inK(E).
It is known that, whenEis a separable Banach space, the notion of measurable multifunctions has some equivalences. More precisely, for a multifunction G: [0,T]!K(E), the following conditions are equivalent to each other (see, e.g. [21]):
(1) Gis measurable;
(2) for every countable dense set {xn} of E, the functions ’n: [0,T]!R, defined by
’nðtÞ ¼dðxn,GðtÞÞ are measurable;
(3) Ghas a Castaing presentation: there is a countable family {gn} of measurable selections ofGsuch that
[1
n¼1
gnðtÞ ¼GðtÞ for a.e.t2[0,T];
(4) Gis strongly measurable.
Furthermore, if G is measurable and Lp-integrably bounded, then it is Lp-integrable. If G is Lp-integrable on [0,d] for some p51, then G is also L1-integrable. In this case, we have a multifunctiont°Rt
0GðsÞdsdefined by Zt
0
GðsÞds:¼ Zt
0
gðsÞdx:g2S1G
8t2 ½0,d:
The following -estimate ( is the Hausdorff MNC), which is similar to [21, Theorem 4.2.3] will be used in the sequel.
LEMMA 2.3 Assume that E is a separable Banach space. Let G: [0,d]! P(E) be Lp-integrable,Lp-integrably bounded multifunction such that
ðGðtÞÞ4qðtÞ
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
for a.e. t2[0,d].Here,q2Lpþð½0,dÞ.Then Zt
0
GðsÞds
4 Zt
0
qðsÞds
for all t2[0,d].In particular,if the multifunction G: [0,d]!K(E)is measurable and Lp-integrably bounded then the function (G())is integrable and,moreover,
Zt
0
GðsÞds
4 Zt
0
ðGðsÞÞds for all t2[0,d].
Now we consider the multimap F: [0,T] B EN!Kv(E) in our problem (1.1)–(1.3).
Definition 2.7 We say thatFsatisfies theupper Carathe´odory conditions if
(1) the multifunction F(,,U) : [0,T]!Kv(E) admits a strongly measurable selection for each (,U)2 B EN, and
(2) the multimapF(t,,) :B EN!Kv(E) is u.s.c for a.e.t2[0,T].
The multimapF is said to beLp-locally integrably bounded if for eachr40, there exists a function!r2Lp([0,T]) such that
kFðt,,UÞk ¼supfkzkE:z2Fðt,,UÞg4!rðtÞ for all (,U)2 B ENsatisfyingjjBþ kUkEN4r.
Let CE(1,T) denote the linear topological space of functionsu: (1,T]!E satisfying that
u02 B and uj½0,T2CN1ð½0,T;EÞ, endowed with the seminorm
kukCEð1,TÞ¼ ju0jBþ kukCN1ð½0,T;EÞ: Foru2 CE(1,T), consider the superposition multifunction
F:½0,T !KvðEÞ, FðtÞ ¼Fðt,ut,rNuðtÞÞ:
By the axioms of the phase space, we see thatt°ut2 Bis a continuous function.
Further, the functionrNu: [0,T]!ENis continuous, too. ThenFisLp-integrable providedFsatisfies upper Carathe´odory conditions and isLp-locally bounded. The proof can be made in the same way as in [21, Theorem 1.3.5].
As the consequence, we can define on CE(1,T) the superposition multi- operatorPFby
PFðuÞ ¼ f2Lpð0,T;EÞ: ðtÞ 2Fðt,ut,rNuðtÞÞfor a.e.t2 ½0,Tg:
We have the following property of weak closedness of PF, whose proof can be proceeded as in [21, Lemma 5.1.1].
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
LEMMA 2.4 Let{un}be a sequence inCE(1,T)converging to u2 CE(1,T)and suppose that the sequence {n} Lp(0,T;E), n2 PF(un) weakly converges to a function.Then2 PF(u).
3. Existence results
In the sequel, let Ebe a separable Banach space. We give the definition of a mild solution for (1.1)–(1.3), in accordance with formula (2.2), as follows.
Definition 3.1 For a given 2(0,T], a function u2 CE(1,) is called a mild solution to problem (1.1)–(1.3) on the interval (1, ] if it satisfies the integral equation
uðtÞ ¼
’ðtÞ, fort40, X
N1 k¼0
tk
k!eukþ 1 ðÞ
Zt
0
ðtsÞ1ðsÞds fort2 ½0,, 8>
<
>: where2 PF(u).
We assume that the multimap F in problem (1.1)–(1.3) satisfies the following hypotheses:
(F1)F: [0,T] B EN!Kv(E) satisfies the upper Carathe´odory conditions;
(F2)FisLp-locally integrably bounded forp4Nþ11 ;
(F3) there exists a functionk2Lp(0,T;E) such that for any bounded subsetsQ B andj E,j¼0,. . .,N1, we have
F t,Q,N1Y
j¼0
j
!!
4kðtÞ ðQÞ þXN1
j¼0
ðjÞ
! ,
where
ðQÞ ¼sup
40
ðQðÞÞ, ð3:1Þ
QðÞ ¼ fqðÞ:q2 Qg:
Remark 3.1 In the case E¼Rm, condition (F3) follows from (F2). In fact, the Lp-locally integrably boundedness of F implies that the set Fðt,Q,QN1
j¼0 jÞ is bounded inRmfor a.e.t2[0,T], and hence it is a precompact set.
If dim(E)¼ þ1, then a particular case of fulfilling (F3) is the following condition:
Fðt, ,Þ:B EN!KvðEÞ
is completely continuous for a.e. t2[0,T], i.e. F(t,,) maps each bounded set in B ENinto a precompact set inE.
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
For a given2(0,T], we set
S:Lpð0,;EÞ !Cð½0,;EÞ, SðÞðtÞ ¼ 1
ðÞ Zt
0
ðtsÞ1ðsÞds, ð3:2Þ
uðtÞ ¼
’ðtÞ, ift40, X
N1 k¼0
tk
k!euk, if 04t4, 8>
><
>>
: and
GðuÞ ¼uþS PFðuÞ: ð3:3Þ One sees that u2 CE(1,) is a mild solution of problem (1.1)–(1.3) on interval (1,] if and only if it is a fixed point ofG. From now on, we can restrictGon the subsetD CN1([0,];E) defined as
D ¼ fv2CN1ð½0,;EÞ,vð0Þ ¼eu0¼’ð0Þg, ð3:4Þ by settingG(v)¼ G(v[’]), where
v½’ðtÞ ¼ ’ðtÞ, ift40, vðtÞ, if 04t4:
(
We first verify some features of S which will be used to obtain some important properties of the multioperator G. Define the following collection of operators Sj:Lp(0,;E)!C([0,];E),j¼0, 1,. . .,N1:
S0¼S: ð3:5Þ
IfN41 then SjðÞðtÞ ¼ dj
dtjSðÞðtÞ ¼ 1 ðjÞ
Z t
0
ðtsÞj1ðsÞds, j¼1,. . .,N1: ð3:6Þ PROPOSITION 3.1 The operators Sj,j¼0, 1,. . .,N1have the following properties:
(S1)There exist constants Cj40,j¼0, 1,. . .,N1,such that kSjðÞðtÞ SjðÞðtÞkpE4Cpj
Zt
0
kðsÞ ðsÞkpEds, ,2Lpð0,;EÞ:
(S2)For each compact set K E and sequence{n} Lp(0,;E)such that{n(t)} K for a.e. t2[0,],the weak convergencen* 0implies Sj(n)!Sj(0)in C([0,];E).
Proof
(i) By using the Ho¨lder inequality we have kSjðÞðtÞ SjðÞðtÞkE4 1
ðjÞ Zt
0
ðtsÞj1kðsÞ ðsÞÞkEds 4 1
ðjÞ h Z t
0
ðtsÞðj1Þp=ðp1Þdsip1ph Z t
0
kðsÞ ðsÞkpEdsi1p :
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
Then
kSjðÞðtÞ SjðÞðtÞkpE4Cpj Z t
0
kðsÞ ðsÞkpEds, where
Cj¼
h p1 ðjÞp1
ip1p Tj1p ðjÞ: Notice that (j)p140,j¼0,. . .,N1 sincep4Nþ11 .
(ii) To prove (S2), notice that, without loss of generality, {n(t)} E0 for all t2[0,], where E0¼spK is the separable Banach space spanned by the compact setK. Moreover, it is clear also that {Sj(n)(t)} E0for allt2[0,] andj¼0, 1,. . .,N1. Then, applying Lemma 2.3 we obtain
fSjð Þn ð Þgt
1
ðjÞ Zt
0
ðtsÞj1 ðfnðsÞgÞds¼0:
Hence, the sequencefSjðnÞðtÞg1n¼1 Eis relatively compact for everyt2[0,].
On the other hand, we have kSjðnÞðt2Þ SjðnÞðt1ÞkE
¼ 1
ðjÞ Z t2
0
ðt2sÞj1nðsÞds Zt1
0
ðt1sÞj1nðsÞds
E
4 1 ðjÞ
Zt2
t1
ðt2sÞj1nðsÞds
E
þ 1
ðjÞ Zt1
0
½ðt2sÞj1 ðt1sÞj1nðsÞds
E
:
Since {n(s)} Kfor a.e. s2[0,], the right side of this inequality tends to zero as t2!t1uniformly with respect to n. So {Sj(n)} is an equicontinuous set. Thus from the Arzela–Ascoli theorem, we obtain that the sequence {Sj(n)} C([0,];E) is relatively compact.
Property (S1) ensures that eachSj:Lp(0,;E)!C([0,];E),j¼0,. . .,N1, is a bounded linear operator. Then it is continuous with respect to the topology of weak sequential convergence, that is the weak convergencen* 0ensuringSj(n)*Sj(0).
Taking into account that {Sj(n)} is relatively compact, we arrive at the conclusion thatSj(n)!Sj(0) strongly in C([0,];E). g We have the following technical result, whose proof constitutes of the arguments from [21, Theorem 4.2.1, Corollary 4.2.1, Remarks 4.2.1 and 4.2.2].
PROPOSITION 3.2 Let the sequence of functions {n} Lp(0,;E) be Lp-integrably bounded:
knðtÞkE4ðtÞ, for all n¼1, 2,. . ., and a.e. t2 ½0,, where2Lp(0,).Assume that
ðfnðtÞgÞ4qðtÞ
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
for a.e. t2[0,], where q2Lp(0,). Then for every 40 there exists a compact set K E,a set m [0,], meas(m)5and a sequence of functions G Lp(0,;E)with values in K,such that for every n51 there exists bn2Gfor which
knðtÞ bnðtÞkE42qðtÞ þ, t2 ½0,nm:
In addition,one can choose the sequence{bn}so that bn¼0on mand this sequence is weakly compact.
Now using this result, we will prove the following proposition.
PROPOSITION 3.3 Let the sequence {n} Lp(0,;E) satisfy the conditions of Proposition3.2.Then we have
ðfSjðnÞðtÞgÞ42Cj Zt
0
jqðsÞjpds
1p
, j¼1,. . .,N1, for any t2[0,].
Proof We follow the arguments in [21, Theorem 4.2.2] with some slight modifications. For any 40, choose 2(0,) such that for all m [0,], with meas(m)5, we have
Z
m
jðsÞjpds5:
Taking m and {bn} corresponding to {n} from Proposition 3.2, we have by Proposition 3.1 that the sequences {Sj(bn)},j¼0,. . .,N1 are relatively compact in C([0,];X). Furthermore
kSjðnÞðtÞ SjðbnÞðtÞkpE 4Cpj
Zt
0
knðsÞ bnðsÞkpEds 4Cpj
Z
½0,tnm
knðsÞ bnðsÞkpEdsþCpj Z
½0,t\m
knðsÞkpEds 4Cpj
Z
½0,tnm
½2qðsÞ þpdsþCpj Z
m
jðsÞjpds 4Cpj
Zt
0
j2qðsÞ þjpdsþ
:
Therefore, the relatively compact setSjG(t) forms aCjðRt
0j2qðsÞ þjpdsþÞ1p-net for the set {Sj(n)(t)}. This proves the proposition since40 is arbitrary. g Definition 3.2 A sequence {n} in Lp(0,;E) is called semicompact if it is Lp-integrably bounded and the set {n(t)} is relatively compact inEfor a.e.t2[0,].
PROPOSITION 3.4 Let {n} be a semicompact sequence in Lp(0,;E). Then {n} is weakly compact in Lp(0,;E).For each04j4N1, {Sj(n)}is relatively compact in C([0,];E).Moreover,ifn* 0then Sj(n)!Sj(0),j¼0,. . .,N1.
Proof The weak compactness of {n} inLp(0,;E) is the consequence of the results in [29, Corollary 3.4]. Since {n(t)} is relatively compact in E for a.e. t2[0,],
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
from Proposition 3.3, we have that for eachj¼0,. . .,N1, the sequence {Sj(n)(t)}
is relatively compact inEfor a.e.t2[0,].
On the other hand, from the assumptions, there exists a function 2Lp(0,) such that
knðtÞk4ðtÞ, for alln¼1, 2,. . ., and t2 ½0,: Thanks to the Ho¨lder inequality, one thus has
kSjðnÞðt2Þ SjðnÞðt1ÞkE
¼ 1
ðjÞ Zt2
0
ðt2sÞj1nðsÞds Zt1
0
ðt1sÞj1nðsÞds
E
4 1 ðjÞ
Zt2
t1
ðt2sÞj1nðsÞds
E
þ 1
ðjÞ Zt1
0
½ðt2sÞj1 ðt1sÞj1nðsÞds
E
4 1 ðjÞ
Zt2
t1
ðt2sÞðj1Þp0ds p10 Zt2
t1
jðsÞjpds
1p
þ 1
ðjÞ Zt1
0
jðt2sÞj1 ðt1sÞj1jp0ds
p10 Zt1
0
jðsÞjpds
1p
, where p0 is the conjugate of p. The last inequality implies that {Sj(n)} is equicontinuous in C([0,];E), and then it is relatively compact in C([0,];E).
The last conclusion follows the same lines as in the proof of Proposition 3.1. g LEMMA 3.5 Suppose that F satisfies(F1)–(F3). Then the composition
G ¼uþS PF is a closed multioperator with compact values.
Proof It suffices to prove the assertions forS PF. Let {vn} be such thatvn!vin D. Assume that n2 PF(vn[’]) and zn¼S(n)2S(PF(vn[’])), zn!z in CN1([0,];E). We prove thatz2S(PF(v[’])). Since
fnðtÞg Fðt,ðvn½’Þt,rNvnðtÞÞ,
we see that {n} is integrably bounded by (F2) and the following inequality holds by (F3)
ðfnðtÞgÞ4kðtÞ ðfðvn½’ÞtgÞ þXN1
j¼0
ðfvðnjÞðtÞgÞ
! :
For each j¼0,. . .,N1, the sequence fvðnjÞg converges in C([0,];E). Then ðfvðnjÞðtÞgÞ ¼0 for a.e.t2[0,]. On the other hand,
ðfðvn½’ÞtgÞ ¼sup
40
ðfvn½’ðtþÞgÞ 4sup
s2½0,t
ðfvnðsÞgÞ ¼0: ð3:7Þ
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
Thus
ðfnðtÞgÞ ¼0, for a.e.t2 ½0,,
and then {n} is a semicompact sequence. By Proposition 3.4, we may assume, without loss of generality, that there exists2Lp(0,;E) such that
n* and zn¼SðnÞ !SðÞ ¼z:
Using Lemma 2.4, we obtain 2 PF(v[’]) and we deduce that z¼S()2S(PF(v[’])).
It remains to show that, for v2 D and {n} chosen in PF(v[’]), the sequence {S(n)} is relatively compact inCN1([0,];E). Hypotheses (F2)–(F3) imply that {n} is semicompact. Using Proposition 3.4, we obtain that Sj(n), j¼0,. . .,N1, is relatively compact in C([0,];E). Therefore, {S(n)} is relatively compact in
CN1([0,];E). The proof is completed. g
In order to prove the u.s.c. property ofS PF, we need the following result.
THEOREM 3.6 [21] Let X and Y be metric spaces and F:X!K(Y) a closed quasicompact multimap.Then F is u.s.c.
LEMMA 3.7 Let the conditions of Lemma3.5hold.Then the multioperatorGis u.s.c.
Proof In view of Theorem 3.6 and the result in Lemma 3.5, it suffices to check that Gis a quasicompact multimap. LetA Dbe a compact set. We prove thatG(A) is a relatively compact subset of CN1([0,];E). Assume that {zn} G(A). Then zn¼uþS(n), where n2 PF(vn[’]), for a certain sequence {vn} A. Hypotheses (F2)–(F3) yield the fact that {n} is semicompact and then it is a weakly compact sequence inLp(0,;E). Similar arguments as in the proof of Lemma 3.5 imply that {S(n)} is relatively compact in CN1([0,];E). Thus we have the desired
conclusion. g
We are in a position to prove thatGis a condensing multioperator. We first need an MNC constructed suitably for our problem. Denote
:PðCN1ð½0,;EÞÞ !Rþ, ðÞ ¼ sup
t2½0,
eLtXN1
j¼0
dj dtjðtÞ
: ð3:8Þ
Here,
dj
dtjðtÞ:¼ dj
dtjvðtÞ: v2
, and the constantLis chosen so that
‘:¼4XN1
j¼0
Cj sup
t2 ½0,T
Z t
0
eLpðtsÞkpðsÞds
1p
51, ð3:9Þ
wherek() is the function from condition (F3).
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
Further, consider
modC: PðCN1ð½0,;EÞÞ !Rþ, modCðÞ ¼lim
!0sup
v2
04j4N1max max
jt1t2j5kvðjÞðt1Þ vðjÞðt2Þk, ð3:10Þ which is called the modulus of equicontinuity of in CN1([0,];E). Consider the function
:PðCN1ð½0,;EÞÞ !R2
þ, ðÞ ¼ max
D2DðÞððDÞ, modCðDÞÞ, ð3:11Þ whereD() is the collection of all countable subsets ofand the maximum is taken in the sense of the ordering in the coneR2
þ. By the same arguments as in [21], one can see thatis well-defined. That is, the maximum archives inD(), andis an MNC in the space CN1([0,];E), which fulfils all properties in Definition 2.3 (see [21, Example 2.1.3] for details).
LEMMA 3.8 Under conditions of Lemma 3.5, the multioperator G:D!K(D) is -condensing.
Proof Let Dbe such that
ðGðÞÞ5ðÞ: ð3:12Þ We show that is relatively compact. Let (G()) be achieved on a sequence {zn} G(), i.e.
ðGðÞÞ ¼ ððfzngÞ, modCðfzngÞÞ:
Then
zn¼uþSðnÞ, n2 PFðvn½’Þ wherefvng : Now inequality (3.12) implies
ðfzngÞ5ðfvngÞ: ð3:13Þ It follows from (F3) that
ðfnðsÞgÞ4kðsÞ ðfðvn½’ÞsgÞ þXN1
j¼0
ðfvðnjÞðsÞgÞ
!
for everys2[0,]. In view of (3.1), we have ðfðvn½’ÞsgÞ ¼sup
40
ðfvn½’ðsþÞgÞ ¼ sup
2 ½0,s
ðfvnðÞgÞ:
Then
ðfnðsÞgÞ4kðsÞeLs sup
2 ½0,s
eL ðfvnðÞgÞ þeLsXN1
j¼0
ðfvðnjÞðsÞgÞ
!
42eLskðsÞðfvngÞ:
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
Now the application of Proposition 3.3 forSj,j¼0,. . .,N1, yields ðfSjðnÞðtÞgÞ44Cj
Z t
0
epLskpðsÞds
1p
ðfvngÞ for anyt2 ½0,: Recalling that
SjðnÞðtÞ ¼ dj
dtjSðnÞðtÞ ¼ dj
dtjznðtÞ dj
dtjuðtÞ ¼zðnjÞðtÞ dj
dtjuðtÞ, t50, we arrive at
eLtN1X
j¼0
ðfzðnjÞðtÞgÞ ¼eLtN1X
j¼0
zðnjÞðtÞ dj dtjuðtÞ
¼eLtN1X
j¼0
ðfSjðnÞðtÞgÞ44XN1
j¼0
Cj
Zt
0
eLpðtsÞkpðsÞds
1p
ðfvngÞ:
Putting this relation together with (3.13), we obtain
ðfvngÞ4ðfzngÞ ¼ sup
t2 ½0,
eLtXN1
j¼0
ðfzðnjÞðtÞgÞ4‘ðfvngÞ:
Therefore({vn})¼0. This implies
ðfvðnjÞðtÞgÞ ¼0, j¼0,. . .,N1 for all t2 ½0,:
Using (F2)–(F3) again, one gets that {n} is a semicompact sequence. Then, Proposition 3.4 ensures that {Sj(n)} is relatively compact in C([0,];E) for j¼0,. . .,N1. This yields that {zn} is relatively compact in CN1([0,];E). Hence modC({zn})¼0. Finally,
ðÞ ¼ ð0, 0Þ:
g Now we go to the main results of this section. The following assertion is the local existence result for (1.1)–(1.3).
THEOREM 3.9 Suppose that conditions (F1)–(F3) are satisfied. Then there exists 2(0,T]such that problem(1.1)–(1.3)has at least one mild solution inCE(1,).
Proof We take an arbitrary number40 and denote 0¼ ðKþ1ÞþMj’jB, K¼ max
t2 ½0,TKðtÞ, M¼ sup
t2 ½0,T
MðtÞ,
p0¼ p
p1, and Q¼ ZT
0
j!0ðsÞjpds
1p
,
where!0is given in Definition 2.7. Taking into account thatp4Nþ11 , it is obvious thatj1p40 for allj¼0,. . .,N1. Thus we can take2(0,T], which is small
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
so that
1
ðjÞQ 1 ðj1Þp0þ1
p10
j1p4
N, ð3:14Þ
for allj¼0,. . .,N1.
Denote byBthe closed convex bounded subset of Ddefined by B¼ fu2 D :kuukCN1g:
Then foru2Bandvchosen from
GðuÞ ¼uþS PFðu½’Þ, we have
kvðtÞ uðtÞkE4 1 ðÞ
Z t
0
ðtsÞ1kFðs,us,rNuðsÞÞkds 4 1
ðÞ Z t
0
ðtsÞ1!0ðsÞds, ð3:15Þ for allt2(0,]. Here we use assumption (F2) with a note that
jusjBþ krNuðsÞkEN4KðsÞkukCð½0,s;EÞþMðsÞj’jBþ kukCN1ð½0,;EÞ
4ðKþ1ÞkukCN1ð½0,;EÞþMj’jB 40:
An application of the Ho¨lder inequality to (3.15) gives kvðtÞ uðtÞkE4 1
ðÞ Zt
0
ðtsÞð1Þp0ds p10 Zt
0
j!0ðsÞjpds
1p
4 1 ðÞQ
h 1 ð1Þp0þ1
ip10
t1p4 N, thanks to (3.14). Similarly, forj¼1,. . .,N1, we have
dj
dtjvðtÞ dj dtjuðtÞ
E
4 1 ðjÞ
Z t
0
ðtsÞj1!0ðsÞds 4 1
ðjÞ Z t
0
ðtsÞðj1Þp0ds p10 Zt
0
j!0ðsÞjpds
1p
4 1
ðjÞQ 1 ðj1Þp0þ1
p10
tj1p 4
N: So, the following estimate holds
X
N1 j¼0
sup
t2 ½0,
dj
dtjðvðtÞ uðtÞÞ
4,
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
or equivalently,
kvukCN1ð½0,;EÞ4:
Therefore, G maps B into B. Hence, the proof is completed by invoking the
conclusion of Theorem 2.1. g
In order to get the global existence result, we need to replace assumption (F2) with a stronger one. Precisely, we impose the assumption that
(F20) There exists a function!2Lp([0,T]) such that
kFðt,,UÞk:¼supfkfkE: f2Fðt,,UÞg4!ðtÞð1þ jjBþ kUkENÞ, for all2 BandU 2EN.
In addition, we need the following version of generalized Bellman–Gronwall inequality (see, e.g. [30]).
LEMMA 3.10 Assume that f(t),g(t)and y(t)are non-negative,integrable functions on [0,T]satisfying the integral inequality
yðtÞ4gðtÞ þ Zt
0
fðsÞyðsÞds, t2 ½0,T: Then we have
yðtÞ4gðtÞ þ Zt
0
expn Z t
s
fðÞdo
fðsÞgðsÞds, t2 ½0,T:
THEOREM 3.11 Under assumptions (F1), (F20) and (F3), the set of solutions for problem(1.1)–(1.3) on the interval[0,T]is nonempty and compact.
Proof We apply Theorem 2.2 to prove that FixG is a nonempty compact set.
Combining the results of Lemmas 3.5, 3.7 and 3.8, it remains to check that, ifu2 DT is such that
uu2ðGðuÞ uÞ
for2(0, 1], thenumust belong to a bounded set inCN1([0,T];E). Indeed, we have the inequality by using (F20) forj¼0,. . .,N1
dj
dtjuðtÞ dj dtjuðtÞ
E
4 ðjÞ
Zt
0
ðtsÞj1kFðs,us,rNuðsÞÞkds 4 1
ðjÞ Zt
0
ðtsÞj1!ðsÞð1þ jusjBþ krNuðsÞkENÞds:
Using the fact that
jusjBþ krNuðsÞkEN4Mj’jBþ ðKþ1ÞkukCN1ð½0,s;EÞ, we have
dj dtjuðtÞ
E
4kU0kEN
X
N1 k¼0
Tk
k! þMj’jBþ1 ðjÞ
Zt
0
ðtsÞj1!ðsÞds þ Kþ1
ðjÞ Zt
0
ðtsÞj1!ðsÞkukCN1ð½0,s;EÞÞds
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
forj¼0,. . .,N1. By the Ho¨lder inequality, we get dj
dtjuðtÞ
E
4kU0kENXN1
k¼0
Tk k!
þMj’jBþ1 ðjÞ
h 1
ðj1Þp0þ1 ip10
Tj1p ZT
0
j!ðsÞjpds
1p
þ Kþ1 ðjÞ
h 1
ðj1Þp0þ1 ip10
Tj1p Zt
0
j!ðsÞjpkukpCN1ð½0,s;EÞds
1p
:
Let
Cj¼ 1 ðjÞ
h 1
ðj1Þp0þ1 ip10
Tj1p, C¼maxfCj: j¼0,. . .,N1g,
g0 ¼NkU0kENXN1
k¼0
Tk
k!þNCðMj’jBþ1Þ ZT
0
j!ðsÞjpds
1p
, fðsÞ ¼ ½NCðKþ1Þ1p!ðsÞ, s2 ½0,T:
Then
kukCN1ð½0,t;EÞ4g0þ Z t
0
jfðsÞjpkukpCN1ð½0,s;EÞds
1p
:
This implies
vðtÞ42pgp0þ2p Zt
0
jfðsÞjpvðsÞds,
wherevðtÞ ¼ kukpCN1ð½0,t;EÞ. In accordance with Lemma 3.10, we obtain the estimate
vðtÞ ¼ kukpCN1ð½0,t;EÞ42pgp0 1þ ZT
0
exp 2p Z T
s
jfðÞjpd
jfðsÞjpds
for allt2[0,T]. The last inequality leads to the fact that kukCN1ð½0,T;EÞ4R0, where
R0¼2g0
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1þ
ZT
0
exp 2p Z T
s
jfðÞjpd
jfðsÞjpds
p
s
:
Finally, takinga¼u2 DTand applying Theorem 2.2 for UDT¼ fu2 DT:kukCN1ð½0,T;EÞRg
withR4R0, we conclude that FixGis nonempty compact set. g
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
4. Properties of solution map
In this section we will study the continuous dependence of the solution set F of problem (1.1)–(1.3) on the initial data (’,U)2 B EN. To attain these ends we will need two additional hypotheses on the phase spaceB.
It will be supposed that
(B4) there existsl40 such thatk (0)k4lk kB for all 2 B;
(B5) there exists m, 04m4þ1, such that for every sequence { n} B with k n 0kB!0 the sequence { n()} is relatively compact inEfor every2[m, 0].
It is easy to see that both spaces presented as examples in Section 2.2 satisfy these properties. In fact,l¼1 in both cases, andm¼ þ1in case (1) andm¼rin case (2).
Further, we assume that the multimap F: [0,T] B EN!Kv(E) satisfies conditions (F1), (F20) and the following slightly modified condition of -regularity:
(F30) there exists a function k2Lp(0,T;E) such that for any bounded subsets Q Bandj E,j¼0,. . .,N1, we have
Fðt,Q,N1Y
j¼0
jÞ
!
4kðtÞ sup
m440
ðQðÞÞ þXN1
j¼0
ðjÞ
! :
Now, we consider in the spaceB ENCN1([0,T];E) the subset
D¼ fð ;eu0,. . .,euN1;vÞ: ð0Þ ¼eu0¼vð0Þ,eu1¼v0ð0Þ,. . .,euN1¼vðN1Þð0Þg, which is closed by (B4). We define a family of multioperators
:D! PðCN1ð½0,T;EÞÞ by
ð ;eu0,. . .,euN1;vÞ ¼euþS PFðv½ Þ, where
euðtÞ ¼XN1
k¼0
tk
k!euk, 04t4T:
It is clear that v2ð ;eu0,. . .,euN1;vÞ implies that the function v[ ]2 CE(1,T) belongs to the solution setF( ,U), whereU¼ ðeu0,. . .,euN1Þ:
LEMMA 4.1 The multioperator is closed, i.e.,for any sequences {( n,Un,vn)} D and wn2( n,Un,vn), the conditions k n 0kB!0, kUnU0kEN!0, kvnv0kCN1!0, andkwnw0kCN1!0imply w02( 0,U0,v0).
Proof Consider a sequence {fn} Lp(0,T;E) such that fn2 PF(vn[ n]) and wn¼eunþSðfnÞ: ð4:1Þ Here,
eunðtÞ ¼XN1
k¼0
tk
k!ðeukÞn, 04t4T, andðeukÞn, k¼0,. . .,N1 are the components ofUn.
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012
From hypothesis (B3) it follows that the sequence {vn[ n]t} is uniformly bounded with respect tot2[0,T]. Then (F20) implies that the sequence {fn} isLp-integrably bounded.
Further, from condition (F30) we get for a.e.t2[0,T]:
ffnðtÞg
ð Þ4 F t, fvn½ ntg,fUng 4kðtÞ sup
m440
fvn½ ntðÞg
þN1X
j¼0
fðeujÞng
!
¼kðtÞ sup
m440
fvn½ ntðÞg
¼
kðtÞmax n
sup044t ðfvnðÞgÞ, suptm4040 ðf nð0ÞgÞ o
, if 04t5m, kðtÞ sup
tm44t
fvnðÞg
ð Þ, ifm4t:
8<
:
Applying hypothesis (B5), we conclude that both values are vanishing and so the sequence {fn} is semicompact. By Proposition 3.4 and Lemma 2.4, we may assume, with loss of generality, that fn*f0, where f02 PF(v0[ 0]). Applying again Proposition 3.4, and passing to the limit in (4.1), we obtain
w0¼eu0þSðf0Þ,
concluding the proof. g
Now we are in position to prove the main result of this section. Denote bythe closed subset of the spaceB ENdefined by
:¼ fð ,UÞ ¼ ð ;eu0,. . .,euN1Þ: ð0Þ ¼eu0g:
For (’,U)2, let F(’,U) CE(1,T) be the set of solutions of problem (1.1)–(1.3).
THEOREM 4.2 Under conditions (B1)–(B5), (F1), (F20)and (F30),the multimap F: ! PðCEð1,TÞÞ
is upper semicontinuous.
Proof At first, we prove the upper semicontinuity of the multimap :!K(CN1([0,T];E)) defined as
ð’,UÞ ¼ fv2ð’,U,vÞg:
Suppose to the contrary that there exist "0, and sequences {(’n,Un)} , with k’n’0kB!0,kUnU0kEN!0, {vn} CN1([0,T];E),vn2(’n,Un) such that
vn62W"0ðð’0,U0ÞÞ, n1, ð4:2Þ whereW"0 denotes the "0-neighbourhood of a set. We have
vn2ð’n,Un,vnÞ, n1, ð4:3Þ i.e.,
vn¼eunþSðfnÞ,
Downloaded by [National Sun Yat-Sen University] at 19:51 29 July 2012