• Aucun résultat trouvé

OF NONCONVEX DIFFERENTIAL INCLUSIONS

N/A
N/A
Protected

Academic year: 2022

Partager "OF NONCONVEX DIFFERENTIAL INCLUSIONS"

Copied!
8
0
0

Texte intégral

(1)

A VIABILITY RESULT FOR A CLASS

OF NONCONVEX DIFFERENTIAL INCLUSIONS

AURELIAN CERNEA

We prove the existence of viable solutions to the Cauchy problem x F(x) + G(t, x),x(0) =x0,x(t)K, whereKRn is a locally compact set,G(·,·) is a continuous set-valued map andF(·) is an upper semicontinuous set-valued map contained in the Fr´echet subdifferential of aφ-convex function of order two.

AMS 2000 Subject Classification: 34A60.

Key words: viable solution,φ-convex function of order two, nonconvex differential inclusion.

1. INTRODUCTION

Consider a multifunction F : Rn → P(Rn) that defines the Cauchy problem

(1.1) x∈F(x), x(0) =x0.

In the theory of differential inclusions the viability problem consists in prov- ing the existence of viable solutions, i.e., ∀t, x(t) K, to the Cauchy pro- blem (1.1).

The viability problem was first solved by Nagumo [11] for differential equations (i.e., F is a single valued map). Nagumo’s theorem was forgotten and rediscovered several times. Under the assumptions that F is an upper semicontinuous nonempty convex compact valued multifunction andK is lo- cally compact, Haddad [10] proved that a necessary and sufficient condition for the existence of viable trajectories starting fromx0 ∈K of problem (1.1) is the tangential condition∀x∈K,F(x)∩TxK =∅, whereTxKis the contingent cone toK at x∈K.

Rossi [12], proved the existence of viable solutions to problem (1.1) re- placing the convexity conditions on the images onF by

(1.2) F(x)⊂∂V(x) ∀x∈K,

REV. ROUMAINE MATH. PURES APPL.,52(2007),1, 1–8

(2)

where∂V is the subdifferential in the sense of Convex Analysis of a proper con- vex functionV. Condition (1.2) was first introduced by Bressan, Cellina and Colombo [4] for a differential inclusion problem without viability constraints.

The result in [12] was improved in [7] by assuming that

(1.3) F(x)⊂∂FV(x) ∀x∈K,

whereFV is the Fr´echet subdifferential of aφ-convex function of order twoV. In a recent paper, Duc Ha [9] showed the existence of viable solutions to the problem

(1.4) x ∈F(x) +G(t, x), x(0) =x0 ∈K, x(t)∈K, ∀t∈[0, T], whereG(·,·) is a continuous set-valued map,K Rn is locally compact and F(·) is as in [12].

In [3] the convexity assumption on the functionV in [9] is relaxed, in the sense thatV is assumed to be locally Lipschitz uniform regular and F(x)

CV(x) ∀x∈K, where∂CV is Clarke’s generalized gradient of the map V. The aim of the present paper is to prove the existence of viable solutions to the problem (1.4) withF(·) as in [7]. On one hand, our results extends the result in [7] to the more general problem and, on the other hand, our result provides an alternative improvement of the result in [9]. Unlike the approach in [3] we do not assume thatV(·) is locally Lipschitz and we use the Fr´echet subdifferential ofV(·). Note that one always has FV(x)⊂∂CV(x).

The proof of our main result follows the general ideas in [7] and [9].

The paper is organized as follows. In Section 2 we recall some preliminary facts needed in the sequel while in Section 3 we prove our main result.

2. PRELIMINARIES

We denote by P(Rn) the set of all subsets of Rn and by R+ the set of all positive real numbers. For > 0 we putB(x, ) ={y Rn;y−x< }. WithB we denote the unit ball in Rn. By cl(A) we denote the closure of the setA Rn, by co (A) the convex hull of A and putA= sup{a;a∈A}. We recall that onP(Rn) the generalized Hausdorff metric is defined by

dH(A, B) = max{d(A, B),d(B, A)}, d(A, B) = sup{d(a, B);a∈A}, where d(a, B) = infb∈Ba−b and · is the Euclidian norm onRn.

Let Ω Rn be an open set and letV : Ω R∪ {+∞}be a function with domainD(V) ={x∈Rn;V(x)<+∞}.

Definition2.1. The multifunctionFV : Ω→ P(Rn) defined byFV(x) =

α Rn,lim inf

y→x

V(y)−V(x)−α,y−x y−x 0

if V(x) < + and FV(x) = if

(3)

V(x) = +is called theFr´echet subdifferentialofV. PutD(∂FV) ={x∈Rn;

FV(x)=∅}.

According to [8] the values of FV(·) are closed and convex.

Definition 2.2. Let V : Ω R∪ {+∞} be a lower semicontinuous function. We say that V is φ-convex of order 2 if there exists a continuous mapφV : (D(V))2×R2 R+ such that for everyx, y ∈D(∂FV) and every α∈∂FV(x) we have

(2.1) V(y)≥V(x) +α, x−y −φV(x, y, V(x), V(y))(1 +α2)x−y2. In [6], [8] there are several examples and properties of such maps. For example, according to [6], ifK R2 is a closed and bounded domain whose boundary is aC2 regular Jordan curve, then the indicator function ofK,

V(x) =IK(x) =

0, ifx∈K +∞, otherwise, isφ-convex of order 2.

In what follows we assume

Hypothesis2.3. i)K Rn is locally compact.

ii) G(·,·) : [0, a]×Rn → P(Rn), a > 0,is a continuous set-valued map with compact values.

iii)F :Rn→ P(Rn) is upper semicontinuous with compact values.

iv)There exists a proper lower semicontinuousφ-convex function of order twoV :RnR∪ {+∞} such that

F(x)⊂∂FV(x) ∀x∈K.

v)For any (t, x)[0, a]×K, the tangential condition lim inf

h→0

1

hd(x+h[F(x) +G(t, x)], K) = 0, holds.

Finally, we recall the following result proved in [9].

Lemma2.4 ([9]). Assume that

i) K Rn is nonempty, x0 K, and there exists r > 0 such that K0 :=K∩B(x0, r) is compact;

ii) P,·) : [0, a]×Rn → P(Rn) is an upper semicontinuous set-valued map with nonempty compact values;

iii)for any (t, x) [0, a]×K, the tangential condition lim inf

h→0

1

hd(x+hP(t, x), K) = 0 holds.

(4)

Let T∈(0,min{a,Mr+1}), where M := sup{P(t, x); (t, x)[0, a]×K0}. Then for any ε (0,1), any set N ={ti; 0 = t0 <· · · < tν = a} and any u0 P(0, x0), there exist a set N = {ti; 0 = t0 < · · · < tν = T}, step functionsf(·),z(·) andx(·) defined on[0, T]such that for every i∈ {1, . . . , ν} the conditions below hold.

1){t0, . . . , tk(i)} ⊂ {t0, . . . , ti}, where k(i) is the unique integer such that k(i)∈ {0,1, . . . , ν1} andtk(i)≤ti ≤tk(i)+1.

2) 0< tj+1−tj ≤α for allj∈ {0,1, . . . , i1}, where α:=εmin{1, t1 t0, . . . , tν−tν−1}.

3) p(0) = u0, p(t) = p(θ(t))∈P(θ(t), x(θ(t))) on [0, ti], where θ(t) = tj ift∈[tj, tj+1), for allj∈ {0,1, . . . , i1} and θ(ti) =ti.

4) z(0) = 0, z(t) = z(tl+1) if t (tl, tl+1], l i−1 and z(t) ≤ 2ε(M + 1)T.

5) x(t) = x0+t

0 p(s)ds+z(t) ∀t [0, ti], x(tj) = xj K0 and xj xj(M+ 1)|tj −tj|for j, j ∈ {0,1, . . . , i}.

3. MAIN RESULT Our main result is as follows.

Theorem 3.1. Assume that Hypothesis 2.3 holds. Then for all x0 ∈K there existsT (0, a) such that problem(1.1) admits a solution on [0, T].

Proof. Let r > 0 be such that K0 := K∩B(x0, r) is a compact set in Rn. Let M, T >0 be such that F(x)+G(t, x) ≤M ∀(t, x)∈[0, a]×K0 andT (0,min{a,Mr+1}). LetN0 :={0, T} and εm = 21m,m= 1,2, . . . .

First, the uniform continuity ofG(·) on the compact set K0 ensures the existence ofδm>0 such that∀t, t[0, T] and∀x, x ∈K0if(t, x)(t, x) (M+ 2)δm, then dH(G(t, x), G(t, x))≤εm.

Apply Lemma 2.4 withP(t, x)≡F(x) +G(t, x) and εm,m= 1,2, . . . . Therefore, for every m = 1,2, . . . there exist a set Nm ={tmi ; 0 =tm0 <

· · · < tmνm = T}, and step functions fm(·), gm(·), zm(·) and xm(·) defined on [0, T] with the following properties:

i)Nm ⊂Nm+1 ∀m≥0;

ii) 0 < tmi+1 −tmi αm for all i ∈ {0,1, . . . , νm 1}, where αm :=

εmmin{1, δm, tm−11 −tm−10 , . . . , tm−1νm−1 −tm−1νm−1−1};

iii) gm(t) =gmm(t)) G(θm(t), xmm(t))) and fm(t) = fmm(t)) F(xmm(t))) on [0, T], whereθm(t) =tmi ift∈[tmi , tmi+1], for all i∈ {0,1, . . . , νm1}and θm(T) =T;

(5)

iv) zm(0) = 0, zm(t) = z(ti+1) if t (ti, ti+1], 0 i νm 1 and zm(t) ≤2εm(M+ 1)T;

v)xm(t) =x0+t

0[fm(s) +gm(s)]ds+zm(t),xmm(t))∈K0 ∀t∈[0, T], and

xm(tmi )−xm(tmj ) ≤(M+ 1)|tmi −tmj | fori, j∈ {0,1, . . . , νm}.

Note that from the last inequality for i, j∈ {0,1, . . . , νm} we get (tmi , xm(tmi ))(tmj , xm(tmj ))(M+ 2)|tmi −tmj |.

Remark that the sequence gm(·) can be constructed using the relative compactness property in the space of bounded functions. The proof of this fact can be found in [9]. Therefore, without loss of generality, one may assume that there exists a bounded function g(·) such that

(3.1) lim

m→∞ sup

t∈[0,T]gm(t)−g(t)= 0.

Define qm(t) = x0 + t

0[fm(s) +gm(s)]ds. By property iv), one has zm (t)= 0 a.e. on [0, T] and thus qm (t)=xm(t) ≤M a.e. on [0, T] and the sequence qm(·) is equicontinuous while the sequence of derivatives qm (·) is equibounded. Therefore, a subsequence (still denoted) qm(·) converges in the sup-norm topology to an absolutely continuous functionx(·) : [0, T]Rn while the sequence of derivatives qm(·) converges weakly in L2([0, T],Rn) to x(·). At the same time,qm(t)−xm(t)=zm(t),zm(t)= 0 a.e. on [0, T].

Hence by iv) it follows that

(3.2) xm(·) converges uniformly to x(·),

xm(·) converges weakly in L2([0, T],Rn) tox(·).

On the other hand,gm(·) converges pointwise a.e. on [0, T] tog(·). Then the continuity ofG(·,·) and the fact that the values ofG(·,·) are closed yield g(t)∈G(t, x(t)) a.e. on [0, T]. Further, by the properties of the sequencexm(·) and the closedness ofK0, we getx(t)∈K0 ⊂K.

Definef(t) :=x(t)−g(t). By construction,fm(t) :=xm(t)−gm(t) and (3.3) fm(t)∈F(xmm(t)))⊂∂FV(xmm(t))) a.e.([0, T]).

Since fm(t) M ∀t [0, T], m 1, it follows (e.g. Theorem III.27 in [5]) that there exists a subsequence (again denoted) fm(·) such that fm(·) converges weakly inL2([0, T],Rn) to f(·).

Therefore from (3.3) and Theorem 1.4.1 (“Aubin-Cellina convergence theorem”) in [2] we obtain

f(t) =x(t)−g(t)∈coF(x(t))⊂∂FV(x(t)) a.e.([0, T]).

(6)

Next, apply Theorem 2.2 in [6] to deduce that

(V(x(t))) =f(t), x(t)=x(t)2− g(t), x(t) a.e.([0, T]).

Thus

(3.4) V(x(T))−V(x0) = T

0 x(t)2dt T

0 g(t), x(t)dt.

On the other hand, by construction we have xm(t) = fim +gim with fim ∈F(xmi )⊂∂F(xmi ). On account of Definition 2.2, one has

V(xmi+1)−V(xmi )≥ fim, xmi+1−xmi

−φV(xmi+1, xmi , V(xmi+1), V(xmi ))(1 +fim2)xmi+1−xmi 2 =

=

xm(t)−gm(t), tmi+1

tmi xm(t)dt

−φV(xmi+1, xmi , V(xmi+1), V(xmi ))(1 +fim2)xmi+1−xmi 2

tm

i+1

tmi xm(t)2dt tm

i+1

tmi xm(t), gm(t)dt

−φV(xmi+1, xmi , V(xmi+1), V(xmi ))(1 +fim2)xmi+1−xmi 2. Summing over ithe last inequalities we get

V(xm(T))−V(x0) T

0 xm(t)2dt T

0 xm(t), gm(t)dt−a(m), where

a(m) =

νm

i=0

φV(xmi+1, xmi , V(xmi+1), V(xmi ))(1 +fim2)xmi+1−xmi 2. Denote

S:= sup{φV(x1, x2, y1, y2);xi∈K0, yi [V(x0)−1, V(x0)+1], i= 1,2}<+∞.

According to Remark 1.14 and Theorem 1.18 in [9] (or Theorem 2.1 in [6]),V(·) is continuous onD(V). So, we have the estimate

|a(m)| ≤

νm

i=0

S(1 +F(K0)2)xmi+1−xmi 2 =

=S(1 +F(K0)2)

νm

i=0

tmi+1

tmi xm(t)dt 2

≤S(1 +F(K0)2)

νm

i=0

1 2m

tm i+1

tmi xm(t)2dt

(7)

≤S(1 +F(K0)2) 1 2m

T

0 xm(t)2dt 1

2mS(1 +M2)T(M+ 1)2, so that lim

m→∞a(m) = 0.

According to (3.1) and (3.2), we obtain

m→∞lim T

0 xm(t), gm(t)dt= T

0 x(t), g(t)dt.

Lettingm→ ∞ in (3.5), the continuity of V(·) yields lim sup

m→∞

T

0 xm(t)2dt T

0 x(t), g(t)dt≤V(x(T))−V(x(0)).

Using (3.4) we obtain lim sup

m→∞

T

0 xm(t)2dt T

0 x(t)2dt

and by the lower semicontinuity of the norm inL2([0, T],Rn) (e.g. Proposi- tion III.30 in [5]) we find that

m→∞lim T

0 xm(t)2dt= T

0 x(t)2dt, so thatxm(·) converges strongly in L2([0, T],Rn).

Hence, there exists a subsequence (still denoted) xm(·) which converges pointwise almost everywhere to x(·). It remains to note that (xm(t), xm(t) gm(t))graphF(·) a.e. ([0, T]) and the closedness of graphF(·) ensures that (x(t), x(t)−g(t))∈graphF(·) a.e. ([0, T]) and the proof is complete.

Remark3.2. IfG≡0 in Theorem 3.1, then we obtain the result in [7].

If K = Rn, G(·,·) is single valued and V(·) is convex, in Theorem 3.1 then we obtain the result in [1].

IfV(·) is convex in Theorem 3.1, then we obtain the result in [9].

REFERENCES

[1] F. Ancona and G. Colombo,Existence of solutions for a class of nonconvex differential inclusions. Rend. Sem. Mat. Univ. Padova83(1990), 71–76.

[2] J.P. Aubin and A. Cellina,Differential Inclusions. Springer, Berlin, 1984.

[3] M. Bounkhel,Existence results of nonconvex differential inclusions. Portugaliae Math.

59(2002), 283–310.

[4] A. Bressan, A. Cellina and G. Colombo, Upper semicontinuous differential inclusions without convexity. Proc. Amer. Math. Soc.106(1989), 771–775.

[5] H. Br´ezis,Analyse fonctionelle, theorie et applications. Masson, Paris, 1983.

[6] T. Cardinali, G. Colombo, F. Papalini and M. Tosques,On a class of evolution equations without convexity. Nonlinear Anal.28(1996), 217–234.

(8)

[7] A. Cernea,Existence of viable solutions for a class of nonconvex differential inclusions.

Math. Reports6(56)(2004), 217–224.

[8] M. Degiovanni, A. Marino and M. Tosques, Evolution equations with lack of convexity.

Nonlinear Anal.9(1995), 1401–1443.

[9] T. X. Duc Ha, Existence of viable solutions of nonconvex differential inclusions. Atti Sem. Mat. Fis. Univ. Modena47(1999), 457–471.

[10] G. Haddad, Monotone trajectories of differential inclusions and functional inclusions with memory. Israel J. Math.39(1981), 83–100.

[11] M. Nagumo, Uber die Lage der Integralkurven gew¨¨ ohnlicher Differentialgleichungen.

Proc. Phys. Math. Soc. Japan24(1942), 551–559.

[12] P. Rossi,Viability for upper semicontinuous inclusions without convexity. Differential Integral Equations5(1992), 455–459.

Received 1 March 2006 University of Bucharest

Faculty of Mathematics and Computer Science Str. Academiei 14

010014 Bucharest, Romania

Références

Documents relatifs

Recently a large number of case-control and prospec- ti ve observat ional studies from around the world have supported the hypothesis that light-to-moderate levels

for proofs and refinements of these assertions. It is in general a very difficult problem to determine the irrationality exponent of a given transcendental real number ξ, unless ξ

The set of assumptions (1.6), (1.7), (1.9) &amp; (1.8) are supposed to be a good model of the real dynamics if the particles are small enough so that the Reynolds number of the fluid

- In Section 2, we prove that, if the support of γ(x) satisfies the geometric control con- dition introduced by C.Bardos, G.Lebeau and J.Rauch in [3], Equation (1.1) generates

Institut de la de la Main Main – – Clinique Jouvenet Clinique Jouvenet

Toute utilisation commerciale ou impression systématique est constitutive d’une infrac- tion pénale.. Toute copie ou impression de ce fichier doit contenir la présente mention

Toute utilisation commerciale ou impression systématique est constitutive d’une infrac- tion pénale.. Toute copie ou impression de ce fichier doit contenir la présente mention

The proof of the existence results for nonconvex variational problems considered in this paper hinges on the local Lipschitz continuity of minimizers of the relaxed problem