Funkcialaj Ekvacioj, 25 (1982) 153-162

A Baire Category Approach to the Existence of Solutions of

Multivalued Differential Equations in Banach Spaces
By

F. S. De BLASI and G. PIANIGIANI

(Universita degli Studi, Firenze, Italy)

1.

Introduction.

be the set of all closed convex bounded

Let $X$ be a real Banach space. Let
$X$
In this note we study the solution sets
interior.
nonempty
subsets of which have
of the multivalued differential equations
$ovalbox{ttsmall REJECT}$

(1.1)

$dot{x}in F(t, x)$

(1.2)

$dot{x}inpartial F(t, x)$

$x(t_{0})=x_{0}$

$x(t_{0})=chi_{0}$

$(cdot=frac{d}{dt})$

denotes the
and
Here, $F$ is a mapping from an open subset of $Rtimes X$ into
$F(t,
x)$
boundary of
.
Our main result states that, if $X$ is reflexive and $F$ continuous in the Hausdorff
metric, then (1.2) has at least one solution. We obtain this as an immediate consequence of a more general theorem which establishes that almost all (in the sense of
the Baire category) solutions of (1.1) are actually solutions of (1.2).
If $X$ has finite dimension, our existence result is a special case of Filippov
theorem [5]; but it is new when $X$ is infinite dimensional. In this case most existence
theorems refer to equation (1.1) and are obtained under compactness assumptions
on $F([1], [3])$ . Further properties of multivalued differential equations (1.1) with
nonconvex $F(t, x)subset R^{n}$ can be found in [7].
We adapt here a method used by Cellina [2] in the study of a differential inclusion in $R$ .
Denote by $M_{F}$ (resp.
) the set of all solutions of (1.1) (resp. (1.2)). We
is nonempty and that, under the metric of uniform
shall prove, first of all, that
can be exconvergence, is a complete metric space. We show, next, that
is a
. Thus
pressed as a countable intersection of open dense subsets of
, hence it is nonempty and (1.2) has solutions.
dense -subset of
$F$
When is single valued, (1.1) and (1.2) reduce to the same ordinary differential
equation which, as is well known, has not necessarily solutions if $F$ is only continuous and $X$ is an infinite dimensional space ([6], [9]). This shows that for con$iovalbox{ttsmall REJECT}$

$partial F(t, x)$

$mathrm{s}$

$ovalbox{ttsmall REJECT}_{partial F}$

$veeovalbox{ttsmall REJECT}_{F}$

$ovalbox{ttsmall REJECT}_{partial F}$

$veeovalbox{ttsmall REJECT}_{F}$

$mathrm{G}_{delta}$

$ovalbox{ttsmall REJECT}_{F}$

$ovalbox{ttsmall REJECT}_{partial F}$

154

F. S. DE BLASI and G. PIANIGIANI

tinuous $F$, in infinite dimensional Banach spaces, the existence of solutions of (1.2)
can fail, without the assumption that $F(t, x)$ have nonempty interior.
Notations and main results.

2.

. In any Banach space we denote

be a real Banach space with norm
by $S(u, r)$ the open ball with centre at and radius $r>0$ . We put $S=S(0,1)subset X$.
stands for the boundary of .
,
For any set
Denote by $ff$ (resp. ) the space of all nonempty subsets of $X$ which are
bounded (resp. closed convex bounded with nonempty interior). $ff$ is endowed
with the Hausdorff pseudometric
Let

$X$

$|cdot|$

$u$

$A subset X$

$A$

$partial A$

$ovalbox{ttsmall REJECT}$

$h(A, B)=inf{t>0|Asubset B+tS, Bsubset A+tS}$,

$A$

$Binovalbox{ttsmall REJECT}$

and
As well known, becomes a metric when is restricted to . For any
$R

times
X$
we use the norm
, we set $d(x, A)=inf{|x-a||ain A}$ . In the space
,
$|(t, x)|=max$ ${|t|, |x|}$ , $(t, x)in Rtimes X$.
Let $F$ be a continuous mapping from a nonempty open subset of $Rtimes X$ into .
be in the domain of $F$. We wish to prove the existence of (local) soluLet
tions of (1.1) and (1.2). To this end, if we consider the restriction of $F$ to a nonempty
, where $J_{2a}=(t_{0}-2a, t_{0}+2a)$ and $D_{2R}=$
open subset of its domain, say
$S(chi_{0},2R)$ , we can assume without loss of generality, that:
and satisfies $h(F(t, x), mathrm{O})<M$ for
into
$(*)F$ is a continuous mapping from
.
each
By a solution of (1.1) (resp. (1.2)) we mean any function $x:[t_{0}, T]rightarrow X(t_{0}<T)$
. and satisfies (1.1) (resp. (1.2)) for almost all
which is Lipschitzean, has derivative
$tin[t_{0}, T]$ .
$h$

$subset X$

$xsubset X$

$ovalbox{ttsmall REJECT}$

$A$

$ Aneqphi$

$ovalbox{ttsmall REJECT}$

$(t_{0}, x_{0})$

$Omega_{2}=J_{2a}times D_{2R}$

$Omega_{2}$

$ovalbox{ttsmall REJECT}$

$(t, x)inOmega_{2}$

$mathrm{a}.mathrm{e}$

. Then (1.1) has at least one solution $x:[t_{0}, T]$

Proposition 2.1. Let $F$ satisfy
, where $0<T-t_{0}<min{a, R/M}$ . Moreover, if $X$ is reflexive, the uniform limit
of solutions is a solution of (1.1).
$(,* )$

$rightarrow X$

(resp. , ) the set of all solutions of (1.1) (resp. (1.2)) which

is a nonempty closed
. If $X$ is reflexive, by Proposition 2. 1,
are defined on
$C([t_{0},
T],
X)$
nonempty complete
a
is
Consequently,
.
subset of the Banach space
metric space under the metric induced by the norm of uniform convergence of
Denote by

$ovalbox{ttsmall REJECT}_{partial F}$

$ovalbox{ttsmall REJECT}/_{F}$

$[t_{0}, T]$

$ovalbox{ttsmall REJECT}_{F}$

$ovalbox{ttsmall REJECT}_{F}$

$C([t_{0}, T], X)$

Our purpose is to show that

any
, we set

$ovalbox{ttsmall REJECT}_{partial F}$

is a dense

$mathrm{G}_{delta^{neg}}mathrm{s}mathrm{u}mathrm{b}mathrm{s}mathrm{e}mathrm{t}$

of

$ovalbox{ttsmall REJECT}_{F}$

$theta>0$

$theta={xinovalbox{ttsmall REJECT}_{F}|int_{tmathrm{o}}^{T}d(dot{x}(s), partial F(s, x(s)))ds<theta}$

To this end, for

Multivalued

Differential

155

Equations

is a continuous mapping from

Since
the function under the integral is measurable.
$(t, x)rightarrowpartial F(t, x)$

$Omega_{2}$

into

Proposition 2.2. Let $F$ satisfy $(*)$ . Then, for any

is open in
. If, in addition, $X$ is reflexive, the set

$ff$

$theta>0$

$ovalbox{ttsmall REJECT}_{F}$

$chi_{theta}$

and

, the

$ovalbox{ttsmall REJECT}_{F}$

$dot{X}$

is measurable,

$setswarrow mathrm{r}_{theta}$

is dense in

By virtue of Propositions 2.1 and 2.2 we obtain immediately the following

Theorem 2.3. Let $X$ be a reflexive real Banach space. Let

is a dense -subset
and hence, in particular,
the
$G_{delta}$

$set_{vee}ovalbox{ttsmall REJECT}_{partial F}$

$of_{vee}ovalbox{ttsmall REJECT}_{F}$

$F$

satisfy $(*)$ . Then

is nonempty.

$ovalbox{ttsmall REJECT}_{partial F}$

as
be such that
Let
. By Proposition 2.1
are open and
is a complete metric space and, by Proposition 2.2, the sets
. Therefore
dense in

Proof.

$theta_{1}>theta_{2}>--$

$ nrightarrow+infty$

$theta_{n}rightarrow 0$

$Lambda_{theta_{n}}^{prime}$

$ovalbox{ttsmall REJECT}_{F}$

$ovalbox{ttsmall REJECT}_{F}$

$mathrm{A}^{prime}=bigcap_{n=1}^{infty}Lambda_{theta_{n}}^{nearrow}$

is a dense
proved.
3.

$mathrm{G}_{delta}$

-subset of

$ovalbox{ttsmall REJECT}_{F}$

and so

$Lambda^{prime}$

is nonempty.

Since

$/V=ovalbox{ttsmall REJECT}_{partial F}$

, the theorem is

Proof of Proposition 2.1.

Suppose that
defined by

$F$

satisfies

$sigma(t, x)=frac{1}{2}sup$

$(*)$

Let us introduce the following function

{ $r>0|$ there is $yin F(t,$

$x)$

such that

$S(y,$

$r)subset F(t$

$x)$

$sigma:Omega_{2}rightarrow R$

}.

Since $F$ is continuous and takes values in , it follows that is continuous and
positive [4, Lemma 3.1].
We denote by $L^{p}([t_{0}, T], X)$ , $ 1leq p<+infty$ , the Banach space of all (strongly
such that
measurable functions
$ovalbox{ttsmall REJECT}$

$sigma$

$u:[t_{0}, T]rightarrow X$

$int_{tmathrm{o}}^{T}|u(t)|^{p}dt<+infty$

equipped with norm

$(int_{tmathrm{o}}^{T}|u(t)|^{p}dt)^{1/p}$

We set
Let
$x_{1}$

$Omega_{1}=J_{a}times D_{R}$

where,

Proof of Proposition 2. 1.
( ,
{

$ t_{1}=sup$

$[t_{0}, t_{1}]rightarrow X$

$t_{0}leqtauleq T|d$

by

$v_{0}$

$J_{a}=(t_{0}-a, t_{0}+a)$

Let
(,

$partial F$

$t$

$v_{0}in F(t_{0}, x_{0})$

and

$D_{R}=S(x_{0}, R)$

be such that
, for each

$x_{0}+(t-t_{0})u_{0}))>0$

$d(u_{0}, partial F(t_{0}, x_{0}))>sigma(t_{0}, x_{0})$

$tin[t_{0},$

$tau]$

}.

Define

F. S. DE BLASI and G. PIANIGIANI

156

$x_{mathrm{J}}(t)=x_{0}+(t-t_{0})v_{0}$

for
Note that
been defined and satisfies
be such that
$x_{n}(t_{n})+(t-t_{n})v_{n}))>0$ for each

$tin[t_{0}, t_{1}]$

$(t, x_{1}(t))inOmega_{1}$

$tin[t_{0}, t_{1}]$

, $ngeq 1$ , has
Now, suppose that :

. Let
for each
. Let $t_{n+1}=sup{t_{n}leqtauleq T|d(u_{n},$
,
:
by
. Then, define
$[t_{n-1}, t_{n}]rightarrow X$

$x_{n}$

$(t, x_{n}(t))inOmega_{1}$

$t in[t_{n1} _ t_{n}]$

$v_{n} in F(t_{n}, x_{n}(t_{n}))$

$d(v_{n}, partial F(t_{n}, x_{n}(t_{n})))>sigma(t_{n}, x_{n}(t_{n}))$

$partial F(t$

$tin[t_{n}, tau]}$

$x_{n+1}$

$x_{n+1}(t)=x_{n}(t_{n})+(t-t_{n})U_{n}$

$[t_{n}, t_{n+1}]rightarrow X$

$tin[t_{n}, t_{n+1}]$

. Thus the sequence of functions

if
on
Denote by the piecewise linear function which is equal to
;
moreover,
$n=1,2$ ,?. Observe that, by construction,
Clearly
defined.

$(t, x_{n+1}(t))inOmega_{1}$

$t in[t_{n}, t_{n+1}]$

${x_{n}}$

$x$

$x_{n}$

$t_{1}leq t_{2}leq--$

ever

is well
,
when-

$[t_{n-1}, t_{n}]$

$t_{n}<t_{n+1}$

$t_{n}<T$

is strictly
We claim that, for some , $t_{n}=T$. Suppose the contrary. Then
. Set
and fix
increasing and, since it is bounded, it has a limit
$F$
$|t-t|<

delta/[2(M+1)]$
such
that
there
is
,
and
continuity
of
. By the
$|x-x|<

delta$
imply
and
${t_{n}}$

$mathrm{w}$

$hat{t},hat{t}leq T$

$<sigma(hat{t},hat{x})$

$ 0<epsilon$

$delta>0$

$sigma$

$h(partial F(t, x), partial F(hat{t},hat{x}))<frac{epsilon}{4}$

$hat{x}=x(hat{t})$

Fix such that $|t_{n}-hat{t}|<delta/[2(M+1)]$ and

have $|x_{n}(t_{n})+(t-t_{n})U_{n}-hat{x}|<delta$ , thus
$n$

$|sigma(t, x)-sigma(hat{t},hat{x})|<frac{epsilon}{4}$

$|x_{n}(t_{n})-hat{x}|<delta/2$

. For each

$tin[t_{n},hat{t}]$

we

$d(u_{n}, partial F(t, x_{mathit{7}/}(t_{n})+(t-t_{n})u_{n}))geq d(u_{n}, partial F(t_{n}, x_{n}(t_{n})))$

$-h(partial F(t_{n}, x_{n}(t_{n})), partial F(hat{t},hat{x}))-h(partial F(hat{t},hat{x}), partial F(t, x_{n}(t_{n})+(t-t_{n})u_{n}))$

$>sigma(t_{n}, x_{n}(t_{n}))-frac{epsilon}{4}frac{epsilon}{4}>sigma(hat{t},hat{x})-frac{epsilon}{4}frac{epsilon}{4}frac{epsilon}{4}>0---$

, which is a contradiction. Thus there is such that $t_{n}=T$ and, clearly,

is a solution of (1.1).
be a sequence of solutions of (1.1) conAssume, now, $X$ reflexive and let
,
verging uniformly to . We want to prove that is a solution of (1.1). In fact
$L^{2}([t_{0},
T],
X)$
as a bounded set contained in the reflexive Banach space
(see [8], p. 89),
say
subsequence,
, cona
is weakly precompact. By Eberlein-Smulians theorem
verges weakly to a measurable function $omegain L^{2}([t_{0}, T], X)$ ; hence, by Mazurs theorem
converges
([8], p. 36, Corollary) a sequence of convex combinations
$L^{1}([t_{0},
T],
X)$
. As a consequence of this,
strongly to in $L^{2}([t_{0}, T], X)$ and so also in
from
Hence

$t_{n+1}geqhat{t}$

$n$

$x:[t_{0}, T]rightarrow X$

${z_{n}}$

$z$

${dot{z}_{n}}$

$z$

${dot{z}_{n}}$

${sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{z}_{n+iota}}$

$omega$

$sum_{i=1}^{k_{n}}alpha_{i}^{n}z_{n+i}(t)=X_{0}+int_{tmathrm{o}}^{t}(sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{z}_{n+i}(s))ds$

letting

$ nrightarrow+infty$

, it follows

$tin[t_{0}, T]$

157

Multivalued Differential Equations

$z(t)=x_{0}+int_{tmathrm{o}}^{t}omega(s)ds$

$t$

$in[t_{0}, T]$

Since $F$ is continuous and takes closed convex values, by a standard argument one
shows that the Lipschitzean function is a solution of (1.1). This completes the
proof.
$z$

Proof of Proposition 2.2 (

4.

set

is dense).

$mathscr{N}_{theta}$

In this section we prove the first statement of Proposition 2.2 namely, that the
.
is dense in
Let $F$ satisfy $(*)$ . For $mu>0$ and
, set

$Lambda_{theta}^{nearrow}$

$ovalbox{ttsmall REJECT}_{F}$

$(t, x)inOmega_{2}$

,
$Phi_{mu}(t, x)={uin F(t, x)|d(u, partial F(t, x))>mu}$ ,
$F_{mu}(t, x)={uin F(t, x)|d(u, partial F(t, x))leqmu}$

$G_{mu}(t, x)={uin F(t, x)|d(u, partial F(t, x))=mu}$

and let
. Then
and
are in $ff$ and,
is a nonempty convex open bounded subset of $X$ [4, Remark 3.3]. Furthermore, it follows from [4, Remark 3.8] that there is a neighborhood $V$ of
such
$V$
(respectively, from
to the
and
that the mappings
$V$
$X$
are well defined
nonempty convex open bounded subsets of and, f.rom to
$V$
and continuous in .
and $mu>0$ , put
For any
Let

$theta<mu<sigma(hat{t},hat{x})$

$(hat{t},hat{x})inOmega_{2}$

$F_{mu}(hat{t},hat{x})$

$G_{mu}(hat{t},hat{x})$

$Phi_{mu}(hat{t},hat{x})$

$(hat{t},hat{x})$

$(t, x)rightarrow G_{mu}(t, x)$

$(t, x)rightarrowPhi_{mu}(t, x)$

$ffmathrm{I}$

$xinovalbox{ttsmall REJECT}_{F}$

$Delta_{x}^{mu}={tin[t_{0}, T]|d(dot{x}(t), partial F(t, x(t))>mu}$

Lemma 4.1.

Let

$F$

satisfy

$(*)$

$ 0<mu<min$

Let

$x$

$inovalbox{ttsmall REJECT}_{F}$

and fix

$epsilon>0$

Let

${sigma(t, x(t))|tin[t_{0}, T]}$

. Let $t_{0}<tau<T$ be a point of

and suppose that
has Lebesgue measure
there is
density of . Then, there exists
such that for each
:
, which is Lipschitzean, differentiable
,
function
$a.e$ . and such that
$m(Delta_{x}^{mu})>0$

$Delta_{x}^{mu}$

$z_{tau,lambda}$

$0<lambda<lambda_{0}(tau)$

$lambda_{0}(tau)>0$

$Delta_{x}^{mu}$

$J_{tau,lambda}rightarrow X$

$a$

$J_{tau,lambda}=[tau-lambda, tau+lambda]subset[t_{0}, T]$

(4. 1)

$z_{tau,lambda}(taupmlambda)=x(taupmlambda)$

(4.2)

$|z_{tau,lambda}(t)-x(t)|<epsilon$

(4.3)

$dot{z}_{tau,lambda}(t)in F_{mu}(t, z_{tau,lambda}(t))$

Proof.

tinuity of
that:

Let
and

$G_{mu/2}$

$epsilon/backslash 0$

$Phi_{mu}$

Let

at

$tau$

for each
$mathrm{a}.mathrm{e}$

$tin J_{tau,lambda}$

. in

$J_{tau,lambda}$

be a point of density of . From this and the consuch

it follows that there is $ 0<delta(tau)<min$
$Delta_{x}^{mu}$

$(tau, x(tau))$

$frac{m(J_{tau,lambda}backslash Delta_{x}^{mu})}{m(J_{tau,lambda})}<frac{mu}{8M}$

${epsilon, a, R}$

for each

$0<lambdaleqdelta(tau)$

158

F. S. DE BLASI and G, PIANIGIANI

moreover, for each

,
, the sets
nonempty bounded, nonempty convex open bounded and satisfy
$(t, y)in S((tau, x(tau)), delta(tau))$

(4.4)

$G_{mu/2}(t, y)$

$G_{mu/2}(tau, x(tau))subset G_{mu/2}(t,y)+frac{mu}{2}Ssubset F_{mu}(t, y)$

$Phi_{mu}(t, y)$

are respectively

$Phi_{mu}(t, y)subsetPhi_{mu}(tau, x(tau))+frac{mu}{8}S$

Fix

Note that for each

any
closed interval
be

$0<lambda_{0}(tau)<delta(tau)/[8(M+1)]$

$S((tau, x(tau)), delta(tau)/2)$

Let

$J_{tau,lambda}$

$sin J_{tau,lambda_{0}(tau)}$

$[ tau-1, tau+lambda]$

we have
,

$(s, x(s))in$

$0<lambda<lambda_{0}(tau)$

We

have
$q=int_{J_{tau.lambda}}dot{x}(s)ds=int_{J_{tau.lambdacap}Delta_{x}^{mu}}dot{x}(s)ds+int_{J_{tau.lambda}backslash Delta_{x}^{mu}}dot{x}(s)ds$

For each

, we have

$sin J_{tau,lambda}capDelta_{x}^{mu}$

.
, thus

$dot{x}(s)inPhi_{mu}(s, x(s))subsetPhi_{mu}(tau, x(tau))+(mu/8)S$

$int_{J_{tau.lambda}capDelta_{x}^{mu}}dot{x}(s)dsin m(J_{tau,lambda}capDelta_{x}^{mu})[Phi_{mu}(tau, x(tau))+frac{mu}{8}S]$

$subset m(J_{tau,lambda})[Phi_{mu}(tau, x(tau))+frac{mu}{8}S]+m(J_{tau,lambda}backslash Delta_{x}^{mu})MS$

$subset m(J_{tau,lambda})[Phi_{mu}(tau,x(tau))+frac{mu}{4}S]$

On the other hand,

$|int_{J_{tau,lambda}backslash Delta_{x}^{mu}}dot{x}(s)ds|leq Mm(J_{tau,lambda}backslash Delta_{x}^{mu})<frac{mu}{8}m(J_{tau,lambda})$

Therefore
$qin m(J_{tau,lambda})[Phi_{mu}(tau, x(tau))+frac{3}{8}mu S]$

and so,
$frac{q}{m(J_{tau,lambda})}inPhi_{mu}(tau, x(tau))+frac{mu}{2}SsubsetPhi_{mu/2}(tau, x(tau))$

such
is open, there are points ,
Since
, for some $0<alpha<1$ . Hence, by a suitable partition of
that
and , we have $q=q_{1}m(J_{1})+q_{2}m(J_{2})$ . Set, now,
in two intervals
denotes the characteristic function of , $i=1,2$ .
, where
,
and satisfies
Observe that
is a measurable function with values
$Phi_{mu/2}(tau, x(tau))$

$q_{1}$

$q_{2}inpartialPhi_{mu/2}(tau, x(tau))=G_{mu/2}(tau, x(tau))$

$q/m(J_{tau,lambda})=alpha q_{1}+(1-alpha)q_{2}$
$J_{1}$

$J_{tau,lambda}$

$q_{1}chi_{J_{1}}(s)+q_{2}chi_{J_{2}}(s)$

$omega_{tau,lambda}$

$sin J_{tau,lambda}$

$omega_{tau,lambda}(s)=$

$J_{2}$

$J_{i}$

$chi_{J_{i}}$

$omega_{tau,lambda}(s)in G_{mu/2}(tau, x(tau))$

Multivalued

Differential

159

Equations

$int_{J_{tau.lambda}}omega_{tau,lambda}(s)ds=int_{J_{tau.lambda}}dot{x}(s)ds$

Define
$z_{tau,lambda}(t)=x(tau-lambda)+int_{tau-lambda}^{t}omega_{tau,lambda}(s)ds$

Clearly,

$z_{tau,lambda}(taupmlambda)=x(taupmlambda)$

$tin J_{tau,lambda}$

Furthermore, for each

$tin J_{tau,lambda}$

, we have

$|z_{tau,lambda}(t)-x(t)|leqint_{J_{tau.lambda}}|omega_{tau,lambda}(s)-dot{x}(s)|ds<2Mm(J_{tau,lambda})<frac{delta(tau)}{2}<epsilon$

Since

, it follows that

$|z_{tau,lambda}(t)-x(tau)|leq|z_{tau,lambda}(t)-x(t)|+|x(t)-x(tau)|<delta(tau)$

$S((tau, x(tau)), delta(tau))$

.
$(t, z_{tau,lambda}(t))in$

and so, by (4.4),

$dot{z}_{tau,lambda}(t)=omega_{tau,lambda}(t)in G_{mu/2}(tau, x(tau))subset F_{mu}(t, z_{tau,lambda}(t))$

$mathrm{a}.mathrm{e}$

. in

$J_{tau,lambda}$

This completes the proof.

Lemma 4.2.
solution

Let the hypotheses

of (1.1) such that

of Lemma 4.1 be satisfied.

Then, there is

$a$

$z:[t_{0}, T]rightarrow X$

(4.5)

$|z(t)-x(t)|<epsilon$

(4.6)

$dot{z}(t)in F_{mu}(t, z(t))$

Proof.

for each
$a.e$

. in

$tin[t_{0}, T]$

$[t_{0}, T]$

Let
be the set of the points of density of . It is well known that
$m(
. If Delta^{*})=0$ there is nothing to prove. So let $m(Delta^{*})>0$ and let
, $t_{0}<tau<T$. By Lemma 4.1 there is
such that for each
there is a Lipschitzean function
:
which is differentiable
. and satisfies
where
(4. 1), (4.2), (4.3). Likewise in [2], consider the family of all closed intervals
$t_{0}<

tau<T$
,
, and
. Since the intervals
are a Vitalis covering of
, by Vitalis theorem there is a countable subcovering of
by pairwise disjoint
intervals
such that
.
Set
$Delta^{*}$

$Delta_{x}^{mu}$

$m(Delta^{*})=m(Delta_{x}^{mu})$
$tauinDelta^{*}$

$0<lambda<lambda_{0}(tau)$

$lambda_{0}(tau)>0$

$z_{tau,lambda}$

$J_{tau,lambda}rightarrow X$

$mathrm{a}.mathrm{e}$

$J_{tau,lambda}$

$tauinDelta^{*}$

$0<lambda<lambda_{0}(tau)$

$J_{tau,lambda}$

$Delta^{*}$

$Delta^{*}$

$J_{i}=J_{tau}i^{lambda},i$

$m(Delta^{*}backslash bigcup_{i}J_{i})=0$

$omega(t)=sum_{i}dot{z}_{tau}i^{lambda_{i}},(t)chi_{J_{i}}(t)+dot{x}(t)chi_{[t_{0},T]backslash cup i^{J_{i}}}(t)$

$t$

$in[t_{0}, T]$

$mathrm{a}.mathrm{e}.$

and define
$z(t)=x_{0}+int_{tmathrm{o}}^{t}omega(s)ds$

$tin[t_{0}, T]$

Evidently is Lipschitzean and

. Moreover and are equal at the
end points of every interval and at each
. We prove only the first
statement (the proof of the second is similar). To this end, set
and denote
by
the union of all intervals
(of the Vitalis subcovering of ) which are
contained in
. We have
$dot{z}(t)=omega(t)mathrm{a}.mathrm{e}.$

$z$

$z$

$x$

$tin[t_{0}, T]backslash bigcup_{i}J_{i}$

$J_{i}$

$J_{i}=[a_{i}, b_{i}]$

$bigcup_{k}J_{k}$

$J_{k}$

$[t_{0}, a_{i}]$

$Delta^{*}$

160

F. S. DE BLASI and G. PIANIGIANI

$z(a_{i})=x_{0}+sum_{k}int_{J_{k}}omega(s)ds+int_{[t_{0},a_{i}]backslash bigcup_{k}J_{k}}omega(s)ds$

$=x_{0}+sum_{k}int_{J_{h}}dot{z}_{tau}k^{lambda},k(s)ds+int_{[tmathrm{o},i]backslash bigcup_{k}J_{k}}adot{x}(s)ds$

Thus
$z(a_{i})=x_{0}+sum_{k}int_{J_{k}}dot{x}(s)ds+int_{[ta}mathrm{o},i]backslash bigcup_{k}J_{k}dot{x}(s)ds=x(a_{i})$

and since, clearly,

the statement is proved.

$z(b_{i})=x(b_{i})$

Therefore

$z(t)=left{begin{array}{l}z_{tau i^{lambda}i},(t),tin J_{i},i=1,2,cdots x(t),tin[t_{0},T]backslash bigcup_{i}J_{i}end{array}right.$

and so, by (4.2), we obtain (4.5).

Furthermore,

$dot{z}(t)=left{begin{array}{l}dot{z}_{tau}i^{lambda},i(t),tin J_{i}mathrm{a}.mathrm{e}.,i=1,2,cdotsdot{x}(t),tin[t_{0},T]backslash bigcup_{i}J_{i}mathrm{a}.mathrm{e}..end{array}right.$

and the set

of measure zero such that

, hence
, we have
we have
the lemma is proved.
Now we are ready to prove that

Since

$m(Delta^{*}backslash bigcup_{i}J_{i})=0$

$Delta^{*}subsetDelta_{x}^{mu}$

, there is a set
. Thus, for almost all
On the other hand, for almost all
. Therefore satisfies (4.6) and

satisfies

$hat{J}subset$

$m(Delta_{x}^{mu}backslash Delta^{*})=0$

$Delta_{x}^{mu}subsethat{J}cup(bigcup_{i}J_{i})$

$[t_{0}, T]$

$ dot{x}(t) in F_{ mu}(t, x(t))$

$tnotinDelta_{x}^{mu}$

$bigcup_{i}J_{i}$

$tin[t_{0}, T]backslash $

$ tin$

$dot{z}(t)=dot{z}_{tau i^{lambda_{i}}},(t)in F_{mu}(t, z_{tau}i^{lambda},i(t))=F_{mu}(t, z(t))$

$J_{i}$

$swarrow mathrm{r}_{theta}$

Proof of Proposition 2.2 (

$z$

is dense in

is dense).

$ovalbox{ttsmall REJECT}_{F}$

and fix
Let
satisfying (4.5) and (4.6). By (4.6),
4.2 there is
provided $mu<theta/(T-t_{0})$ . Since $|z(t)-x(t)|<epsilon$ for each
so
.
is dense in
$A_{theta}^{nearrow}$

$x$

$inovalbox{ttsmall REJECT}_{F}$

$d(dot{z}(t),$

$Zinveeovalbox{ttsmall REJECT}_{F}$

$partial F(t, z(t))leqmu mathrm{a}.mathrm{e}$

$ZinLambda_{theta}^{nearrow}$

$swarrow mathrm{r}_{theta}$

5.

By Lemma
. and
$tin[t_{0}, T]$ , the set

$epsilon>0$

$ovalbox{ttsmall REJECT}_{F}$

Proof of Proposition 2.2 (

$mathscr{N}_{theta}$

is open).

In this section we prove the second statement of Proposition 2.2, namely that
.
is open in
(if $X$ is reflexive) the set
By a simple application of Lebesgues covering lemma it is easy to prove the
following
$swarrowpsi_{theta}$

$veeovalbox{ttsmall REJECT}_{F}$

.
Lemma 5.1. Let $F$ satisfy $(*)$ . Let $K$ be a compact subset of . Let
$(s,
u)

in$
$(t,
u)

in
K$
$(delta<min{a, R})$ such that for each
and all
Then there is
$S((t, u), delta)$ we have
$Omega_{1}$

$epsilon>0$

$delta>0$

$F(S, U)subset F(t, u)+epsilon S$

Lemma 5.2.

Let

$Ainovalbox{ttsmall REJECT}$

If

$u_{1}$

$u_{2}$

$cdots$

$u_{n}in A$

.
and

$sum_{i=1}^{n}alpha_{i}=1$

$alpha_{i}geq 0$

, then

Multivalued

Differential

161

Equations

$d(sum_{i=1}^{n}alpha_{i}u_{i},partial mathrm{A})geqsum_{i=1}^{n}alpha_{i}d(u_{i}, partial A)$

We observe that, for any $uin A$ ,

. For each $i=1,2$ ,?, , there is
$A$
. Since is convex

Proof.

$epsilon>0$

$ d(u, partial A)=sup$

$beta_{i}>d(u_{i}, partial A)-epsilon$

$n$

${betageq 0|Udaggerbeta Ssubset A}$

$beta_{i}geq 0$

, such that

. Let

$ u_{i}+beta_{i}Ssubset$

$A$

$Asupsetsum_{i=1}^{n}alpha_{i}(u_{i}+beta_{i}S)=sum_{i=1}^{n}alpha_{i}u_{i}+(sum_{i=1}^{n}alpha_{i}beta_{i})S$

which implies
.

$ d(sum_{i=1}^{n}alpha_{i}u_{i},partial mathrm{A})geqsum_{i=1}^{n}alpha_{i}beta_{i}>sum_{i=1}^{n}alpha_{i}d(u_{i}, partial A)-epsilon$

Since

$epsilon$

is arbitrary, the lemma is proved.

Proof of Proposition 2.2 (

$chi_{theta}$

is open).

It is enough to prove that the set

$A_{theta}^{tilde{nearrow}}={xinovalbox{ttsmall REJECT}_{F}|int_{tmathrm{o}}^{T}d(dot{x}(t), partial F(t, x(t)))dtgeqtheta}$

is closed in
.
Indeed, suppose that
converges uniformly to
. Let
.
$K$
Set $K={(t, x(t))|tin[t_{0}, T]}$ and let correspond (to and ) according to Lemma
5.1. There is
such that for each
and all $tin[t_{0}, T]$ we have $(t, x_{m}(t))in$
. Hence, by Lemma 5. 1,
$veeovalbox{ttsmall REJECT}_{F}$

${x_{n}}subset A_{theta}^{tilde{nearrow}}$

$epsilon>0$

$xinovalbox{ttsmall REJECT}_{F}$

$delta$

$epsilon$

$mgeq n_{0}$

$n_{0}$

$S((t, x(t)), delta)subsetOmega_{2}$

(5.1)

$dot{x}_{m}(t)in F(t, x_{m}(t))subset F(t, x(t))+epsilon S$

$tin[t_{0}, T]$

$mgeq n_{0}$

On the other hand, since $X$ is reflexive and

converges uniformly to , by the
argument of Proposition 2. 1 it follows that a subsequence, say
, converges weakly
to in $L^{2}([t_{0}, T], X)$ . Hence a sequence of convex combinations
converges strongly to in $L^{2}([t_{0}, T], X)$ and so, in particular, in $L^{1}([t_{0}, T], X)$ .
We have
${x_{n}}$

$x$

${dot{x}_{n}}$

$dot{x}$

${sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{x}_{n+i}}$

$dot{x}$

$int_{tmathrm{o}}^{T}d(dot{x}(t), partial F(t, x(t)))dtgeqint_{tmathrm{o}}^{T}d(dot{x}(t), partial[F(t, x(t))+epsilon S])dt-epsilon(T-t_{0})$

$geqint_{tmathrm{o}}^{T}d$

$sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{x}_{n+i}(t)$

$partial[F(t, x(t))+epsilon S]$

$dt$

$-int_{tmathrm{o}}^{T}|sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{x}_{n+i}(t)-dot{x}(t)|dt-epsilon(T-t_{0})$

Let

$ngeq n_{0}$

. Then, by virtue of (5. 1) and Lemma 5.2, we have

162

F. S. DE BLASI and G. PIANIGIANI

$int_{tmathrm{o}}^{T}d$

$sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{x}_{n+i}(t)$

$partial[F(t, x(t))+epsilon S]$

$dt$

$geqsum_{i=0}^{k_{n}}alpha_{i}^{n}int_{tmathrm{o}}^{T}d(dot{x}_{n+i}(t), partial[F(t, x(t))+epsilon S])dt$

$geqsum_{i=0}^{k_{n}}alpha_{i}^{n}int_{tmathrm{o}}^{T}d(dot{x}_{n+i}(t), partial F(t, x_{n+i}(t)))dt$

Therefore, since

$X_{n+i}in A_{theta}^{tilde{nearrow}}$

, we obtain

$int_{tmathrm{o}}^{T}d(dot{x}(t), partial F(t,x(t)))dtgeqtheta-int_{tmathrm{o}}^{T}|sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{x}_{n+i}(t)-dot{x}(t)|dt-epsilon(T-t_{0})$

. Since
. Thus
it follows that
Proposition 2.2 is complete.

Let

$ nrightarrow+infty$

${sum_{i=0}^{k_{n}}alpha_{i}^{n}dot{x}_{n+i}}$

$x$

$inswarrowtilde{V}_{theta}$

Acknowledgement.
helpful discussions.

$swarrowell_{theta}^{tilde{nearrow}}$

converges to in $L^{1}([t_{0}, T], X)$ and is arbitrary,

is open and the proof of
is closed, hence
$dot{x}$

$epsilon$

$Lambda_{theta}^{prime}$

The authors are indebted to G. S. Goodman for some

References
[1]
[2]

[3]

[4]
[5]
[6]
[7]

[8]

[9]

Castaing, C., Valadirr, M., Equations differentielles multivoques dans les espaces vec-

toriels localement convexes, Rev. Franqaise Informat. Recherche Operationnelle, 3

(1969), 3-16.
Cellina, A., On the differential inclusion $dot{x}in[-1,1]$ , Atti Accad. Naz. Lincei Rend. Cl.
Sci. Fis. Mat. Natur., Sez. VIII, 69 (1980), 1-6.
letude des equations differentielles multivoques dans les
Daures, J. P., Contribution
espaces de Banach, C.R. Acad. Se. Paris Ser. A-B, 270 (1970), 769-772.
De Blasi, F. S., Dianigiani, G., Remacks on Hausdorff continuous multifunctions and
selection, (preprint).
Filippov, A. F., The existence of solutions of generalized differential equations, Math.
Notes, 10 (1971), 608-611.
Godunov, A. N., The Peanos theorem in Banach spaces (Russian), Funkcional. Anal, i
Prilozen, 9 (1974), 59-60.
Hermes, T., The generalized differential equation $dot{x}in R(t, x)$ , Advances in Math. 4
(1970), 149-169.
Hille, E. and Phillips, R. S., Functional analysis and semigroups, American Mathematical
Society, Colloquium Publications, XXXI, Providence, 1957.
Yorke, J. A., A continuous differential equation in Hilbert space without existence,
Funkcial. Ekvac., 13 (1970), 19-21.

nuna

adreso;

Istituto Matematico U. Dini

Universita degli Studi
Viale Morgagni 67/A
1-50134 Firenze
Italy

(Ricevita la 7-an de julio, 1981)

(Reviziita la 6-an de octobro, 1981)