NONAUTONOMOUS DIFFERENTIAL INCLUSION WITHOUT CONVEXITY
AURELIAN CERNEA
We prove the existence of viable solutions to the Cauchy problemx0 ∈F(t, x), x(0) =x0, whereF is a multifunction defined on a locally compact set M and contained in the Clarke subdifferential of a uniformly regular function.
AMS 2000 Subject Classification: 34A60.
Key words: viable solution, uniform regular function, differential inclusion, Clarke’s subdifferential.
1. INTRODUCTION
Consider a multifunction F : M ⊂ Rm → P(Rm) that defines the Cauchy problem
(1.1) x0∈F(x), x(0) =x0.
In the theory of differential inclusions the viability problem consists in proving the existence of viable solutions, i.e.∀t,x(t)∈M, to the Cauchy problem (1.1).
Under the assumptions that F is an upper semicontinuous nonempty convex compact valued multifunction and M is locally compact, Haddad [5]
proved that a necessary and sufficient condition for the existence of viable trajectories starting from x0 ∈M of problem (1.1) is the tangential condition
(1.2) ∀x∈M F(x)∩TxM 6=∅,
where TxM is the contingent cone to M atx∈M.
Rossi [7] proved the existence of viable solutions to problem (1.1) replac- ing the convexity conditions on the images on F by
(1.3) F(x)⊂∂V(x) ∀x∈M,
where ∂V is the subdifferential, in the sense of Convex Analysis, of a proper convex function V. According to Theorem 4.2 in [2] the result in [7] can be improved by assuming that
(1.4) F(x)⊂∂CV(x) ∀x∈M,
MATH. REPORTS10(60),1 (2008), 11–16
where ∂CV is the Clarke subdifferential (generalized gradient) of a uniform regular function V(·).
The aim of the present note is to extend this result to the case of nonau- tonomous problem
(1.5) x0 ∈F(t, x), x(0) =x0.
We note that in [6] a similar type of result is proved for a functionV(·) whose upper Dini directional derivative coincides with the Clarke directional derivative.
The idea of the proof of our result is to use the regularizing technique in [6] and to apply the known result for autonomous problems in [2].
The paper is organized as follows. In Section 2 we recall some preliminary facts that we need in the sequel while in Section 3 we prove our main result.
2. PRELIMINARIES
Let Ω⊂Rm be a given set. A multifunction F : Ω→ P(Rm) is called (Hausdorff) upper semicontinuous at x0 ∈ Ω, if ∀ > 0 there exists δ > 0 such that kx−x0k< δ implies F(x)⊂F(x0) +B, where B is the unit ball in Rm and P(Rm) denotes the set of all subsets of Rm. For > 0 we put B(x, ) ={y∈Rm;ky−xk< }.
Consider a locally Lipschitz continuous functionV :Rm→R. For every direction v ∈ Rm its Clarke directional derivative at x in the direction v is defined by
DCV(x;v) = lim sup
y→x, t→0+
V(y+tv)−V(y)
t .
TheClarke subdifferential (generalized gradient) of V atx is defined by
∂CV(x) ={q∈Rm, hq, vi ≤DCV(x;v) ∀v∈Rm}.
We recall that theproximal subdifferential ofV(·) is defined by
∂PV(x) =
q∈Rm, ∃δ, σ >0 such that V(y)−V(x) +σky−xk2≥ hq, y−xi ∀y∈B(x, δ) .
Definition 2.1. LetV :Rm→R∪ {∞} be a lower semicontinuous func- tion and let O ⊂ dom(V) be a nonempty open subset. We say that V is uniformly regular overO if there exists a positive numberβ >0 such that for all x∈O and for all ξ∈∂PV(x) one has
hξ, y−xi ≤V(y)−V(x) +βky−xk2 ∀y∈O.
We say thatV is uniformly regular over a closed set K if there exists an open set O containing K such that V is uniformly regular over O.
In [2] there are several examples and properties of such maps. For exam- ple, any lower semicontinuous proper convex function V is uniformly regular over any nonempty compact subset of its domain with β = 0; any lower-C2 functionV is uniformly regular over any nonempty convex compact subset of its domain.
The next result will be used in the proof of our main result.
Lemma 2.2 ([3]). Let V :Rm → R be a locally Lipschitz function and let ∅ 6=K ⊂dom(V) a closed set. If V is uniformly regular over K then
i)∂CV(x) =∂PV(x) ∀x∈K;
ii) if z(·) : [0, τ]→ Rm is absolutely continuous and z0(t) ∈ ∂CV(z(t)) a.e. ([0, τ]), then
(V ◦z)0(t) =kz0(t)k2, a.e.([0, τ]).
We recall that the contingent cone to a set K ⊂ Rm at x ∈ K is de- fined by
TxK = n
v∈Rm; lim inf
h→0+
1
hd(x+hv, K) = 0 o
.
The following viability result for differential inclusions is due to Bounk- hel [2].
Theorem 2.3 ([2]). Let K ⊂ Rm be a nonempty closed set, G : K → P(Rm) an upper semicontinuous multifunction with nonempty closed values and V(·) :Rm →Ra locally Lipschitz function that is uniformly regular over K such that
G(x)⊂∂CV(x) ∀x∈K, G(x)∩TxK6=∅ ∀x∈K.
Then for any x0 ∈K there exists τ >0 such that the problem x0 ∈G(x), x(t0) =x0∈K, x(t)∈K ∀t≥t0
admits a solution on [t0, τ].
In what follows we assume
Hipothesis 2.4. i) F : [0,∞)×M ⊂ Rm → P(Rm) is a bounded set valued map, measurable with respect to t, upper semicontinuous with respect tox, with nonempty closed values.
ii) There exists a locally Lipschitz function V :Rm → R∪ {+∞} uni- formly regular over the locally compact set M ⊂Rm with a constant β > 0 such that
(2.1) F(t, x)⊂∂CV(x) ∀x∈M, a.e. t∈[0,∞), (2.2) F(t, x)∩TxM 6=∅ ∀x∈M, a.e. t∈[0,∞).
3. THE MAIN RESULTS
Our main result is the following
Theorem3.1. Assume that Hypothesis 2.4 is satisfied. Then for every x0 ∈M there exists a >0 such that problem (1.5)admits a solution on [0, a]
satisfying x(t)∈M, ∀t∈[0, a].
Proof.Letx0 ∈M. SinceM ⊂Rm is locally compact, there existsR >0 such that M0:=M∩B(x0, R) is compact. Consider
L:= sup
(t,x)∈[0,∞)×Rm
kF(t, x)k,
define T := L+1R and taken∈Nsuch that n1 < T.
By regularizing the set valued map F of the Cauchy problem (1.5) we reduce the nonautonomous problem to the autonomous case (see [6]). We can find a countable collection of disjoint subintervals (aj, bj)⊂[0, T],j= 1,2, . . . such that their total length is less then n1 and a set valued map Fn defined on D := ([0, T]\S∞
j=1(aj, bj))×M that is jointly upper semicontinuous and Fn(t, x)⊂F(t, x) for each (t, x)∈D. Moreover, ifu(·) andv(·) are measurable functions on [0, T] such that u(t) ∈ F(t, v(t)) a.e. t ∈ [0, T] then for a.e.
t ∈ ([0, T]\S∞
j=1(aj, bj)) we have u(t) ∈ Fn(t, v(t)) (we refer to [8] for this Scorza Dragoni type theorem). It is obvious that all trajectories of F are also trajectories of Fn. We extendFn to the whole [0, T]×M and define
F˜n(t, x) =
Fn(t, x) ift∈[0, T]\S∞
j=1(aj, bj) Fn(aj, x) ifaj < t < aj+b2 j
Fn(bj, x) if aj+b2 j < t < bj Fn(aj, x)∪Fn(bj, x) ift= aj+b2 j.
It is easy to see that ˜Fn(·,·) still satisfies the tangential condition (2.2). On the other hand, according to Lemma 4 in [6], ˜Fn(·,·) is upper semicontinuous on [0, T]×M.
By extending the state space from Rm to R×Rm we can reduce our problem to the autonomous case. For every (t, x)∈[0, T]×M we define
V˜(t, x) =t+V(x).
Obviously, ˜V(·,·) is a locally Lipschitz uniformly regular function over [0, T]×
M and (1, v)∈∂CV˜(t, x) if and only ifv∈∂CV(x) for all (t, x)∈[0, T]×M. At the same time, standard arguments show that (1, v)∈T(t,x)([0, T]×M) if and only if v∈TxM. Therefore, conditions (2.1) and (2.2) imply that
(3.1) (1,F˜n(t, x))⊂∂CV˜(t, x) ∀(t, x)∈[0, T]×M,
(3.2) (1,F˜n(t, x))∩T(t,x)([0, T]×M)6=∅ ∀(t, x)∈[0, T]×M.
Thus, applying Theorem 2.3, we obtain the existence of an absolutely continu- ous function xn(·) : [0, a]→Rm,a∈(0, T), that satisfies
(1, x0n(t))∈(1,F˜n(t, xn(t)))∩∂CV˜(t, xn(t)) a.e.[0, a], xn(0) =x0
and
(t, xn(t))∈[0, T]×M ∀t∈[0, a].
It follows that xn(·) verifies
(3.3) x0n(t)∈Fn(t, xn(t))∩∂CV(xn(t)) a.e.[0, a], xn(0) =x0 and
(3.4) xn(t)∈M ∀t∈[0, a].
Therefore, from (3.3) we have
(3.5) kx0n(t)k ≤L.
On the other hand, by (3.4), graph(xn(·)) is contained in [0, T]×M andxn(·) also is a solution to the inclusion (1.5) except for a set (say)Enof measure not exceeding 1n for eachn∈N. Hence, by (3.5) and Theorem III.27 in [4] there exists a subsequence (again denoted by xn(·)) and an absolutely continuous function x(·) : [0, a]→Rm such that
xn(·) converges uniformly tox(·),
x0n(·) converges weakly in L2([0, T],Rm) to x0(·).
Since graph(∂CV) is closed, by (3.3) one has
(3.6) x0(t)∈∂CV(x(t)) a.e.[0, a].
We apply Lemma 2.2 and deduce that
(V(x(t)))0=hx0(t), x0(t)i=kx0(t)k2 a.e.[0, a].
Therefore, V(x(a))−V(x0) = Ra
0 kx0(t)k2dt. On the other hand, from (3.3) we deduce that
Z a 0
kx0n(t)k2dt= Z a
0
(V ◦xn)0(t)dt=V(xn(a))−V(x0).
Hence, by the lower semicontinuity of V, we get
n→∞lim Z a
0
kx0n(t)k2dt=V(x(a))−V(x0) = Z a
0
kx0(t)k2dt,
so {x0n(·)} converges strongly in L2([0, a],Rm). Hence, there exists a subse- quence (still denoted) x0n(·) which converges pointwise almost everywhere to x0(·). From (3.3) and the fact that graph(F) is closed we have
x0(t)∈F(t, x(t)) a.e.[0, a]
and from (3.4) we obtain that ∀t∈[0, a], x(t)∈M.
Remark3.2. The result in Theorem 3.1 is still valid if instead ofRm we work on a real Hilbert space X. The main tool in the proof this time is an infinite dimensional version of Theorem 2.3, namely, Theorem 2.3 in [3].
REFERENCES
[1] J.P. Aubin and A. Cellina,Differential Inclusions. Springer, Berlin, 1984.
[2] M. Bounkhel,Existence results of nonconvex differential inclusions. Port. Math. (N.S.) 59(2002), 283–309.
[3] M. Bounkhel and T. Haddad, Existence of viable solutions for nonconvex differential inclusions. Electron. J. Diff. Equations2005, No. 50, 10 pp.
[4] H. Brezis,Analyse fonctionelle, th´eorie et applications. Masson, Paris, 1983.
[5] G. Haddad, Monotone trajectories of differential inclusions and functional inclusions with memory. Israel J. Math.39(1981), 83–100.
[6] Z. Kannai and P. Tallos,Viable solutions to nonautonomous inclusions without conve- xity. CEJOR Cent. Eur. J. Oper. Res.11(2003), 47–55.
[7] P. Rossi, Viability for upper semicontinuous inclusions without convexity. Differential Integral Equations5(1992), 455–459.
[8] T. Rzezuchowski, Scorza-Dragoni type theorems for upper semicontinuous multivalued functions. Bull. Acad. Polon. Sci. S´er. Sci. Math.28(1980), 61–66.
Received 8 January 2007 University of Bucharest
Faculty of Mathematics and Computer Sciences Str. Academiei 14
010014 Bucharest, Romania