\documentclass[twoside]{article}
\usepackage{amssymb,amsmath}
\pagestyle{myheadings}
\markboth{\hfil Existence principles for inclusions \hfil EJDE--2002/04}
{EJDE--2002/04\hfil Jean-Fran\c{c}ois Couchouron \& Radu Precup \hfil}
\begin{document}
\title{\vspace{-1in}\parbox{\linewidth}{\footnotesize\noindent
{\sc Electronic Journal of Differential Equations},
Vol. {\bf 2002}(2002), No. 04, pp. 1--21. \newline
ISSN: 1072-6691. URL: http://ejde.math.swt.edu or http://ejde.math.unt.edu
\newline ftp ejde.math.swt.edu (login: ftp)}
\vspace{\bigskipamount} \\
%
Existence principles for inclusions of Hammerstein type involving
noncompact acyclic multivalued maps
%
\thanks{ {\em Mathematics Subject Classifications:} 47H10, 45N05, 47J35.
\hfil\break\indent
{\em Key words:} Fixed point, multivalued map, acyclic set, integral
inclusion, \hfil\break\indent
Hammerstein equation, evolution equation, boundary value problem.
\hfil\break\indent
\copyright 2002 Southwest Texas State University. \hfil\break\indent
Submitted June 18, 2001. Published January 3, 2002.} }
\date{}
%
\author{Jean-Fran\c{c}ois Couchouron \& Radu Precup}
\maketitle
\begin{abstract}
We apply M\"{o}nch type fixed point theorems for acyclic
multivalued maps to the solvability of inclusions of
Hammerstein type in Banach spaces. Our approach makes
possible to unify and improve the existence theories for
nonlinear evolution problems and abstract integral inclusions
of Volterra and Fredholm type.
\end{abstract}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{remark}{Remark}[section]
\renewcommand{\theequation}{\thesection.\arabic{equation}}
\catcode`@=11
\@addtoreset{equation}{section}
\catcode`@=12
\section{Introduction}
In \cite{orp}, the following two fixed point theorems of M\"{o}nch
type for multivalued maps with convex and compact values were proved:
\begin{theorem}
Let $D$ be a closed, convex subset of a Banach space
$X$ and $N:D\to 2^{D}\setminus\{\emptyset\}$ be a map with convex values.
Assume $\mathop{\rm graph}(N)$ is closed,
$N$ maps compact sets into relatively compact sets and
that for some $x_{0}\in D$ one has
\[
\left.
\begin{array}[c]{c}%
M\subset D,\quad M=\mathop{\rm conv}(\{x_{0}\}\cup N(M)) \\
\text{and }\overline{M}=\overline{C} \text{ with $C$ a countable
subset of $M$}\end{array} \right\} \Longrightarrow
\overline{M}\text{ is compact.}
\]
Then there exists $x\in D$ with $x\in N(x)$.
\end{theorem}
\begin{theorem}
Let $K$ be a closed, convex subset of a Banach space
$X$, $U$ be a relatively open subset of $K$, and $N:\overline{U}\to
2^{K}\setminus\{\emptyset\}$ a map with convex values.
Assume $\mathop{\rm graph}(N)$ is closed, $N$ maps compact sets
into relatively compact sets and that for some $x_{0}\in U$, the following two conditions
are satisfied:
\[
\left.
\begin{array}[c]{c}
M\subset\overline{U},\quad M\subset\mathop{\rm conv}(\{x_{0}\}\cup N(M))\\
\text{and } \overline{M}=\overline{C}\text{ with $C$ a countable subset of
$M$}\end{array}
\right\} \Longrightarrow \overline{M}\text{ is compact;}
\]
\[
x\not \in(1-\lambda) x_{0}+\lambda N(x) \text{ for all
$x\in\overline{U}\setminus U$, $\lambda\in ]0,1[$.}
\]
Then there exists $ x\in\overline{U}$ with $x\in N(x)$.
\end{theorem}
In \cite{orp} and \cite{orp2}, some applications of Theorems 1.1 and 1.2
are presented for Hammerstein integral inclusions of the form
\begin{equation}
u(t) \in\int_{0}^{T}k(t,s) g(s,u(s)) ds\;\,\text{ a.e. on } [0,T]
. \label{eq1}
\end{equation}
Here $k$ is a real single-valued function, while $g$ is a
set-valued map with convex, compact values in a Banach space $E$.
Equation (\ref{eq1}) can be written in
the operator form%
\begin{equation}
u\in S G(u) \label{eq2}%
\end{equation}
where $G$ is the Nemitsky multivalued operator associated
to $g$, and $S$ is the linear integral operator of kernel $k$.
The aim of this paper is to present a unified existence theory for
inclusions of type (\ref{eq2}) with linear and nonlinear operators
$S$. Such inclusions arise naturally in the theory of evolution
inclusions of the form
\[
u'(t) \in f(t,u(t))+g(t,u(t) )
\]
subject to initial conditions. They also arise in the theory of
boundary-value problems for second order differential inclusions of
the form
\[
u''(t) \in f(t,u(t))+g(t,u(t)) .
\]
In both cases $S$ is the solution operator assigning to each
function $w$ the solution (assuming its existence and uniqueness)
of the corresponding problem for
\[
u'(t) \in f(t,u(t) )+w(t) ,
\]
respectively
\[
\,u''(t) \in f(t,u(t) )
+w(t) .
\]
If $S$ is nonlinear, we can not assume that the map
$N:=SG$ has convex values and so Theorems 1.1 and 1.2 do
not apply. This was the motivation in \cite{precup} to give extensions of
Theorems 1.1-1.2 for multivalued maps with non-convex values. These extensions
are based on the Eilenberg-Montgomery fixed point theorem
\cite{eilenberg} and generalize previous results obtained by Fitzpatrick
and Petryshyn \cite{fitzpatrick} for condensing set-valued maps.
Our approach to (\ref{eq2}) and several hypotheses are inspired
from \cite{couchouron,couchouron2}. Notice in \cite{couchouron}
and \cite{couchouron2} it is assumed that the nonlinear operator
$S$ can be compared (in some sense explained latter) with a
Volterra linear integral operator. This assumption together with a
suitable compactness property of $g$ guarantees that $N$ is
condensing with respect to a specific measure of non-compactness
in the space of continuous functions on $[0,T]$. In the present
paper, the hypotheses on $S$ and $g$ are more general, so that $N$
have not to be condensing, but just to satisfy a M\"{o}nch type
compactness condition. Moreover, in this paper we discuss not only
continuous solutions but also $L^p$-solutions, and this is done by
a common existence theory. Our results improve and extend those in
\cite{couchouron,couchouron2,orp,orp2}. They also extend a lot of
classical results on perturbed evolution problems and abstract
integral inclusions of both Volterra and Fredholm type.
\section{Preliminaries}
First we recall some definitions. Let $H_{\ast}=\{H_{n}\}_{n\geq0}$
denote the \v{C}ech homology functor with
compact carriers and coefficients in the field of rational numbers
$\mathbb{Q}$ (see Gorniewicz \cite{gorniewicz}). A
nonempty metric space $X$ is said to be \textit{acyclic} if
\[
H_{n}(X) =\left\{ \begin{array}[c]{ll}
\mathbb{Q} &\text{if }n=0\\
0 &\text{if }n\geq 1
\end{array}\right.
\]
i.e., $X$ has the same homology as a single point space.
A metric space $X$ is said to be \textit{contractible} if
there is a homotopy $h:X\times[0,1] \to X$
such that $h(x,0) =x$ and
$h(x,1) =x_{0}$ for every $x\in X$ and with $x_{0}\in X$ given.
The space $X$ is an \textit{absolute retract} (AR for
short) if for every metric space $Z$ and closed set
$A\subset Z$, every continuous map $f:A\to X$ has a continuous
extension $\widehat{f}:Z\to X$.
We say that $X$ is an \textit{absolute neighborhood retract}
(ANR for short) if the above $f$ has a continuous extension to some
neighborhood of $A$.
It is well known that AR's spaces as well as contractible spaces are acyclic.
So are $R_{\delta}$-sets, i.e. compact metric spaces
$X$ for which there exists a decreasing sequence
$(A_{n})_{n\geq1}$ of compact absolute
retracts such that $X=\bigcap_{n\geq1}A_{n}$. Also, every convex subset
of a normed space is contractible and
every compact and convex subset of a normed space is an ANR and is acyclic.
If $X$ is a Hausdorff topological space, we let
\begin{gather*}
P_{f}(X) =\{A\subset X: A\text{ is nonempty, closed}\}\,,\\
P_{k}(X) =\{ A\subset X: A\text{ is nonempty, compact}\} .
\end{gather*}
If $X$ is a metric space we define
\begin{gather*}
P_{a}(X) =\{ A\subset X: A\text{ is nonempty, acyclic}\}\,,\\
P_{ka}(X) =\{ A\subset X: A\text{ is nonempty, compact, acyclic}\}.
\end{gather*}
If $X$ is a closed, convex subset of a normed space
$(E,|\cdot |)$, then we define
\begin{gather*}
P_{c}(X) =\{ A\subset X: A\text{ is nonempty, convex}\} , \\
P_{kc}(X) =\{ A\subset X: A\text{ is nonempty, compact,convex}\} ,
\end{gather*}
and for any nonempty subset $A\subset E$ we let
$| A| =\sup\{ | x| : x\in A\} $, By $\mathop{\rm conv}(A) $
we mean the convex hull of $A$.
Now we state the Eilenberg-Montgomery fixed point theorem \cite{eilenberg}.
\begin{theorem}
Let $\Xi$ be acyclic and ANR, $\Theta$ a
compact metric space, $\Phi:\Xi\to P_{a}(
\Theta) $ an upper semi-continuous map and $\Gamma
:\Theta\to \Xi$ a continuous single-valued map. Then the map
$\Gamma\Phi:\Xi\to 2^{\Xi}$ has a fixed point.
\end{theorem}
An extension of this theorem for condensing (noncompact) acyclic maps is due
to Fitzpatrick-Petryshyn \cite{fitzpatrick}.
Next we recall a well-known result of set-valued analysis (see \cite{hu},
Proposition 1.2.17, Proposition 1.2.23 and Corollary 1.2.20).
\begin{theorem}
Let $X$, $Y$ be Hausdorff topological spaces.
\begin{enumerate}
\item[(a)] Let $N:X\to P_{f}(Y) $, If
$N$ is upper semicontinuous, then
$\mathop{\rm graph}(N) $ is closed in
$X\times Y$, Conversely, if $\mathop{\rm graph}(N)$ is closed and
$\overline{N(X) }$ is compact, then $N$ is upper semicontinuous.
\item[(b)] Let $N:X\to P_{k}(Y) $ be upper semicontinuous.
Then $N(A)$ is compact for each compact $A\subset X$.
\end{enumerate}
\end{theorem}
Throughout this paper $E$ will be a real Banach space
with norm $|\cdot |$. A function $u:[a,b] \to E$ is said to be
\textit{strongly measurable }on $[a,b]$ if there exists a sequence of
finitely-valued functions $u_{n}$ with
\[
u_{n}(t) \to u(t) \quad\text{as }n\to \infty,\quad\text{a.e. on }[a,b] .
\]
By $\int_{a}^{b}u(t)dt$ we mean the
Bochner integral of $u$, assuming its existence. Recall
that a strongly measurable function $u$ is Bochner integrable if
and only if $| u|$ is Lebesgue integrable.
For any real $p\in\lbrack 1,\infty\lbrack$, we consider the space
$L^p([a,b] ;E) $ of all strongly measurable functions $u:[a,b] \to
E$ such that $|u|^p$ is Lebesgue integrable on $[a,b]$. Then
$L^p([a,b];E)$ is a Banach space under the norm
\[
| u|_p=(\int_{a}^{b}| u(s) |^pds) ^{1/p}.
\]
Also for $p=\infty$, we let $L^{\infty}([a,b] ;E)$
be the space of all strongly measurable function $u:[a,b] \to E$
which are essentially bounded, i.e.
\[
\mathop{\rm ess}\sup_{t\in[a,b] }|u(t)| :=\inf\{c\geq0:\;|u(t)|
\leq c\quad \text{a.e. on }[a,b] \} <\infty.
\]
$L^{\infty}([a,b] ;E)$ is a Banach space under the norm
$|u|_{\infty}=\mathop{\rm ess}\sup_{t\in[a,b] }| u(t) |$.
When $E=\mathbb{R}$ the space
$L^p([a,b] ;\mathbb{R}) $ is simply denoted by $L^p[a,b]$. By
$|u|_{\infty}$ we also denote the max-norm on the space $C([a,b]
;E) $ of all continuous functions $u:[a,b] \to E$.
For a function $u:[a,b] \to E$, we
define the \textit{translation} by $h$ ($00$
there exists $c_{\rho}\in L^{q}([a,b];\mathbb{R}_{+}) $ such that
$| \psi(t,x) |_{Y}\leq c_{\rho}(t)$
a.e. on $[a,b]$, for all $x\in D$ with $|x|_{X}\leq\rho$.
\end{enumerate}\end{enumerate}
%
A map $\psi$ which satisfies (C1)-(C2) is said to be a
Carath\'{e}odory function.
\section{Fixed point theorems}
In this section, we present the extensions of Theorems 1.1 and 1.2 to
set-valued maps with acyclic values, which were established in
\cite{precup}.
For the reader convenience, we also reproduce here their proofs.
\begin{theorem}
Let $D$ be a closed, convex subset of a Banach space
$X$, $Y$ a metric space, $N:D\to P_{a}(Y) $ and
$J:Y\to D$ continuous. Assume $\mathop{\rm graph}\,(N)$
is closed, $N$ maps compact sets into relatively compact sets and
that for some $x_{0}\in D$ one has
\begin{equation}
\left.
\begin{array}[c]{c}%
M\subset D,\quad M=\mathop{\rm conv}(\{ x_{0}\} \cup JN(M) ) \\
\text{and }\overline{M}=\overline{C}\text{ with $C$ a countable
subset of $M$} \end{array}
\right\} \Longrightarrow \overline{M}\text{ is compact.}\label{eq8}
\end{equation}
Then there exists $x\in D$ with $x\in JN(x)$.
\end{theorem}
\paragraph{Proof}
Since $J$ is continuous, the map
$JN$ also has a closed graph and maps compact sets into
relatively compact sets.
Following the steps (a) and (b) of the proof of Theorem 3.1 in \cite{orp}, we
find a convex set $M\subset D$ with $x_{0}\in M$,
$M=\mathop{\rm conv}(\{ x_{0}\} \cup JN(M))$ and
$K:=\overline{M}$ compact. Next, instead of steps (c)-(d) of the above
mentioned proof, we follow:
\noindent(c$^{\ast}$)\quad Proof of inclusion $JN(K) \subset K$.
Let $\varepsilon>0$ be fixed. According to
Theorem 2.2, $JN$ is upper semicontinuous.
Consequently, for each $x\in M$ there exists an open
neighborhood $V_{x}$ of $x$ such
that $JN(y) \subset JN(x)+B_{\varepsilon}(0) $ for all $y\in V_{x}$.
Since for $x\in M$, one has
$JN(x) \subset K$, it follows that $JN(y) \subset
K_{\varepsilon}:=K+B_{\varepsilon}( 0) $ for every $y\in V_{x}$.
Now $M$
being dense in $K$, it results that
$\{ V_{x}: x\in M\} $ is a cover of $K$. Consequently, $JN(K)
\subset K_{\varepsilon}$. Hence $JN(K)
\subset\bigcap_{\varepsilon>0}K_{\varepsilon}=K$.
\noindent(d$^{\ast}$)\quad Application of the Eilenberg-Montgomery theorem. Since every
compact and convex subset of a Banach space is an ANR and is acyclic, we may
apply Theorem 2.1 to: $\Xi:=K$, $\Theta:=N(K) $,
$\Phi=N$ and $\Gamma=J$.
\hfill $\Box$
\begin{remark} \rm
(a) Under the assumptions of Theorem 3.1,
$N:D\to P_{ka}(Y) $.
\\
(b) According to Theorem 2.2, Theorem 3.1 is true under
the following assumptions:
$N:D\to P_{ka}(Y) $ and $N$ is upper semicontinuous.
\end{remark}
The next result is a version of Theorem 1.2 for set-valued maps
with acyclic values.
\begin{theorem}
Let $K$ be a closed, convex subset of a Banach space
$X$, $U$ a convex, relatively open subset of
$K$, $Y$ a metric space, $N:\overline{U}\to P_{a}(Y) $ and
$\,J:Y\to K$ continuous.
Assume $\mathop{\rm graph}(N)$ is closed, $N$
maps compact sets into relatively compact sets and that for some
$x_{0}\in U$, the following two conditions are
satisfied:
\begin{equation}
\left.
\begin{array}[c]{c}
M\subset\overline{U},\quad M\subset\mathop{\rm conv}(\{x_{0}\}
\cup JN(M) ) \\
\text{and }\overline{M}=\overline{C}\text{ with $C$ a countable
subset of $M$}
\end{array}
\right\} \Longrightarrow \overline{M}\text{ is compact;}\label{eq9}
\end{equation}
\begin{equation}
x\not \in(1-\lambda) x_{0}+\lambda JN(x) \text{ for all }
x\in\overline{U}\setminus U,\;\lambda\in]0,1[. \label{eq6}
\end{equation}
Then there exists $x\in\overline{U}$ with $x\in
JN(x) $.
\end{theorem}
\paragraph{Proof.}
Let $D=\overline{\mathop{\rm conv}}(\{ x_{0}\}\cup
JN(\overline{U}))$. Clearly, $x_{0}\in D\subset K$. Let $P:K\to
\overline{U}$ be
\[
P(x) =\left\{
\begin{array} [c]{ll}
x & \text{for }x\in\overline{U}\\
\overline{x} &\text{for }x\in K\setminus\overline{U}%
\end{array} \right.
\]
Here $\overline{x}=(1-\lambda) x_{0}+\lambda
x\in\overline{U}\setminus U$, $\lambda\in]0,1[$. It is
easy to see that $P$ is single valued and continuous.
Let $\widetilde{N}:D\to P_{a}(Y)$, $\widetilde{N}(x)=N(P(x))$.
It is easily seen that $\mathop{\rm graph}(\widetilde{N})$
is closed and $\widetilde{N}$ maps compact
sets into relatively compact sets. Next we check (\ref{eq8}) for
$J\widetilde{N}$. Let $M\subset D$ be such that $M=\mathop{\rm
conv}(\{ x_{0}\}\cup J\widetilde{N}(M) )$ and
$\overline{M}=\overline{C}$ for some countable subset $C$ of $M$.
Since
\begin{align*}
P(M) & \subset\text{conv}(\{x_{0}\} \cup M) \subset
\mathop{\rm conv}(\{x_{0}\} \cup J\widetilde{N}(M) ) \\
& =\mathop{\rm conv}(\{ x_{0}\} \cup JNP(M)),
\end{align*}
$\overline{P(M)}=\overline{P(C)}$, $P(C) \subset P(M)$,
and $P(C)$ is countable,
from (\ref{eq9}) we deduce that $P(M) $ is
relatively compact. Then $J\widetilde{N}(M)=JNP(M)$ is
relatively compact and Mazur's lemma implies that $\overline{M}$
is compact. Thus (\ref{eq8}) holds for $J\widetilde{N}$.
Now we apply Theorem 3.1 to deduce that there exists an $x\in D$
with $x\in J\widetilde{N}(x)$. We
claim that $x\in D\cap\overline{U}$. Assume the contrary, that is
$x\in D\setminus\overline{U}$. Then $x\in JN(\overline{x}) $,
where $\overline{x}=(1-\lambda) x_{0}+\lambda x\in
\overline{U}\setminus U$, $\lambda\in]0,1[$. Then $x=(1/\lambda)
\overline{x}+(1-1/\lambda) x_{0}\in JN( \overline{x})$. Hence
$\overline{x}\in(1-\lambda) x_{0}+\lambda JN(\overline{x}) $,
which contradicts (\ref{eq6}). Thus $x\in D\cap\overline{U}$
and so $x\in JN(x)$. \hfill $\Box$
\section{Inclusions of Hammerstein type}
Let $00$ for which $A$
can be covered by finitely many balls of $E_{0}$ with
radius not greater than $\varepsilon$.
\item[(SG)] For every $u\in K$ the set $SG(u)$ is acyclic in
$K$.
\end{enumerate}
Now we can state the main result of this section.
\begin{theorem}
Assume (S1)-(S2), (g1)-(g5) and (SG) hold. In addition suppose
(H3). Then (\ref{eq10}) has at least one solution $u$ in
$K\subset L^p(I;E)$ with $|u|_p\leq R$.
\end{theorem}
The proof is based on Theorem 4.1 and the following two lemmas
that extend some results in \cite{couchouron}.
\begin{lemma}
Let $S:L^{q}(I;E) \to L^p(I;E) $
satisfy (S1)-(S2), $q\in\lbrack1,\infty\lbrack$
and $p\in[1,\infty] $.
Let $M\subset L^{q}(I;\,E)$ be countable with
\begin{equation}
|u(t)|\leq\nu(t) \label{eq12}%
\end{equation}
a.e. on $I$, for all $u\in M$, where $\nu\in L^{q}(I) $.
Let $E_{0}$ be a separable closed subspace of $E$
with $u(t) \in E_{0}$ a.e. on $I$, for every $u\in M\cup S(M)$.
Then the function $\varphi(t)=\beta_{E_{0}}(M(t) ) $ belongs to
$L^{q}(I) $ and satisfies
\begin{equation}
\beta_{E_{0}}(S(M) (t) ) \leq
\int_{I}k(t,s) \varphi(s) ds\quad \text{a.e. on }I.\label{eq13}
\end{equation}
\end{lemma}
\paragraph{Proof}
Let $M=\{ u_{n}: n\in\mathbb{N}\}$. The space $E_{0}$ being separable,
we may represent it as $\overline{\bigcup_{k\geq1}E_{k}}$ where for each
$k$, $E_{k}$ is a $k$-dimensional subspace of $E_{0}$ with
$E_{k}\subset E_{k+1}$.
The fact that $\varphi$ is measurable follows from the formula
of representation of $\beta$ for separable spaces which yields
\begin{equation}
\varphi(t) =\lim_{k\to \infty}\sup_{n\geq 1}\,d(u_{n}(t) ,E_{k}) .\label{eq14}%
\end{equation}
Now $\varphi\in L^{q}(I) $ since $\varphi(t) \leq\nu(t)$ a.e.
on $I$.
Since $M$ is countable, we may suppose that (\ref{eq12})
hold for all $t\in I$ and $u\in M$.
We will prove (\ref{eq13}) for any fixed $t_{0}\in I$.
Let $\varepsilon>0$ and choose $\delta>0$
such that for every measurable subset $\Theta$
of $I$ we have
\[
| \Theta| \leq\delta\,\Longrightarrow\,\int_{\Theta}k(
t_{0},s) \nu(s) ds<\varepsilon.
\]
Here $| \Theta| $ is the Lebesgue measure of $\Theta$. Also choose
a constant $\rho>0$
such that $| \Theta_{1}| <\delta/3$ for
\[
\Theta_{1}=\{ t\in I:\, \nu(t) >\rho\} .
\]
So we have $d(u_{n}(t) ,E_{k})\leq| u_{n}(t) | \leq\rho$ for $t\in
I\setminus\Theta_{1}$ and $n,k\in \mathbb{N}$. Consequently,
$d(u_{n}(t) ,E_{k}) =d(u_{n}(t) ,\overline{B}_{k}) $
with $\overline{B}_{k}=\{ x\in E_{k}: | x| \leq\rho\} $.
From (\ref{eq14}) and Egorov's Theorem there is a set
$\Theta_{2}\subset I\setminus\Theta_{1}$ with $|\Theta_{2}|
<\delta/3$ and an integer $k_{0}$
such that
\begin{equation}
\sup_{n\geq1} d(u_{n}(t) ,\overline{B}_{k})
\leq\varphi(t) +\varepsilon\label{eq15}
\end{equation}
for $t\in I\setminus(\Theta_{1}\cup\Theta_{2})$, $n\geq1$ and
$k\geq k_{0}$. Since $M$ is a countable set of strongly
measurable functions, we may find a set $\Theta_{3}\subset I$
with $|\Theta_{3}|<\delta/3$ and a
countable set $\widetilde{M}=\{ \widetilde{u}_{n}:n\geq1\}$
of finitely-valued functions from $I$ to $E$ with
\begin{equation}
|u_{n}(t)-\widetilde{u}_{n}(t)| \leq\varepsilon\label{eq16}
\end{equation}
for $t\in I\setminus\Theta_{3}$ and $n\geq1$.
From (\ref{eq15}) and (\ref{eq16}) we obtain
\[
d(\widetilde{u}_{n}(t) ,\overline{B}_{k})
\leq\varphi(t) +2\varepsilon
\]
for $n\in\mathbb{N}$, $k\geq k_{0}$ and $t\in I\setminus \Theta$
with $\Theta=\Theta_{1}\cup\Theta_{2}\cup\Theta_{3}$. Then there
exists a finitely-valued function $\widehat{u}_{n,k}$ from $I$ to
$\overline{B}_{k}$ with
\[
|u_{n}(t) -\widehat{u}_{n,k}(t) | \leq\varphi(t) +3\varepsilon
\]
for $n\geq1$, $k\geq k_{0}$ and $t\in I\setminus\Theta$.
We put $\widehat{u}_{n,k}(t) =0$
for $t\in\Theta$. Note that $|\Theta| \leq\delta$.
For each fixed $k\geq k_{0}$, Theorem 2.4 guarantees that the
sequence $(\widehat{u}_{n,k})_{n\geq1\text{ }}$ is relatively
compact in $L_{w}^{q}(I;\overline{B}_{k})$. Then, from (S2) the
sequence $(S(\widehat{u}_{n,k}))_{n\geq1}$
is relatively compact in $L^p(I;E) $.
Therefore, for every $t\in I$ the set $(S(\widehat{u}_{n,k}(t)))_{n\geq1}$
is relatively compact in $E$. Now using (S1),
we obtain
\begin{eqnarray*}
\lefteqn{| S(u_{n}) (t_{0}) -S(\widehat
{u}_{n,k}) (t_{0}) | }\\
& \leq&\int_{I}k(
t_{0},s) | u_{n}(s) -\widehat{u}_{n,k}(
s) | ds \\
&\leq& \int_{I\setminus\Theta}k(t_{0},s) (\varphi(
s) +3\varepsilon) ds+\int_{\Theta}k(t_{0},s)
| u_{n}(s) | ds \\
&\leq&\int_{I}k(t_{0},s) \varphi(s) ds+3\varepsilon
| k(t_{0},.) |_{1}+\int_{\Theta}k(
t_{0},s) \nu(s) ds \\
&\leq&\int_{I}k(t_{0},s) \varphi(s) ds+3\varepsilon
| k(t_{0},.) |_{1}+\varepsilon.
\end{eqnarray*}
Hence $\{S(\widehat{u}_{n,k})(t_{0}) : n\geq1\}$ is a relatively compact
$\gamma$-net of the set
$\{ S(u_{n})(t_{0}): n\geq1\}$ with
\[
\gamma=\int_{I}k(t_{0},s) \varphi(s)
ds+3\varepsilon| k(t_{0},.) |_{1}+\varepsilon
\to \int_{I}k(t_{0},s) \varphi(s) ds
\]
as $\varepsilon\to 0$.
\hfill $\Box$
\begin{lemma}
Assume (S1) and (S2). Let $M$ be a countable subset of $L^{q}(I;E) $
such that $M(t) $ is relatively compact for a.a.
$t\in I$ and there is a function $\nu\in L^{q}(I)$ with
$| u(t)| \leq\nu(t)$ a.e. on $I$, for every $u\in M$.
Then the set $S(M)$ is relatively compact in $L^p(I;E)$.
In addition $S$ is continuous from $M$ equipped with the
relative weak topology of $L^{q}(I;E) $ to $L^p(I;E)$ equipped with
its strong topology.
\end{lemma}
\paragraph{Proof.}
Let $M=\{u_{n}: n\geq1\} $. Let $\varepsilon>0$. As in the proof
of Lemma 4.3, we can find functions $\widehat{u}_{n,k}$ with
values in the compact $\overline{B}_{k}\subset E$ such that
\[
| u_{n} -\widehat{u}_{n,k} |_{q} \leq\varepsilon
\]
for every $n\geq1$. Then (S1) implies via H\"{o}lder's inequality
that
\begin{equation}
| S(u_{n}) -S(\widehat{u}_{n,k})|_p\leq\gamma:=\varepsilon | |
k(t,.)|_{r}|_p.\label{eq17}
\end{equation}
On the other hand, from Theorem 2.4 the set
$\{ \widehat{u}_{n,k}: n\geq1\} \subset L^{q}(I;E) $ is
weakly relatively compact in $L^{q}(I;E) $.
Next (S2) guarantees that $\{S(\widehat{u}_{n,k}): n\geq1\}$
is relatively compact in $L^p(I;E)$.
Hence from (\ref{eq17}) we see that $\{S(\widehat{u}_{n,k}): n\geq1\}$
is a relatively compact $\gamma$-net of $S(M)$.
Since $\varepsilon$ is arbitrary, we conclude that $S(M)$ is
relatively compact.
Now suppose that $(w_{m})_{m}$ converges weakly in
$L^{q}(I;E) $ to $w$ and $w_{m}\in M$. In view of the relative
compactness of $S(M) $, we may assume that $(S(w_{m}))_{m}$ converges in
$L^p(I;E)$ towards some function $h_{\infty}$. We have to prove
\begin{equation}
h_{\infty}=S(w) .\label{hifSgif}
\end{equation}
For each fixed $\varepsilon>0$, we have already seen that the
proof of Lemma 4.3 again provides a compact set $K_{\varepsilon}$
and a sequence $(w_{m}^{\varepsilon})_{m}$ of
$K_{\varepsilon}$-valued functions satisfying
\begin{equation}
|w_{m}-w_{m}^{\varepsilon} |_{q}\leq\varepsilon\label{gngene}
\end{equation}
for every $m\geq1$. The sequence $(w_{m}^{\varepsilon})_{m}$ being
weakly precompact in $L^{q}(I,E)$, a suitable subsequence
$(w_{m_{j}}^{\varepsilon})_{j}$
must be weakly convergent in $L^{q}(I,E) $ towards some
$w_{\infty}^{\varepsilon}$. Then the Masur's
Theorem and (\ref{gngene}) provide
\begin{equation}
| w -w_{\infty}^{\varepsilon} |_{q} \leq\varepsilon.\label{gifge}%
\end{equation}
The triangle inequality yields
\begin{equation}
\begin{aligned}
| h_{\infty}-S(w) |_p \leq&| h_{\infty
}-S(w_{m_{j}}) |_p+| S(w_{m_{j}})
-S(w_{m_{j}}^{\varepsilon}) |_p\\
& + | S(w_{m_{j}}^{\varepsilon}) -S(w_{\infty
}^{\varepsilon}) |_p+| S(w_{\infty}^{\varepsilon}) -S(w) |_p.
\end{aligned}
\label{ShifSge}
\end{equation}
Passing to the limit when $j$ approaches infinity in (\ref{ShifSge})
and using Assumption (S2) we obtain
\begin{equation}
| h_{\infty}-S(w) |_p\leq\lim\sup_{j}|
S(w_{m_{j}}) -S(w_{m_{j}}^{\varepsilon}) |_p
+| S(w_{\infty}^{\varepsilon}) -S(w)
|_p.\label{hSginf}
\end{equation}
According to (\ref{gngene}) and (\ref{gifge}) we deduce from (\ref{hSginf})
and Assumption (S1)
\[
| h_{\infty}-S(w) |_p\leq2\varepsilon | | k(t,.) |_{r}|_p.
\]
Since $\varepsilon$ is arbitrary the proof of Lemma 4.4 is ended.\
\hfill $\Box$
\paragraph{Proof of Theorem 4.2} (a) First we show that
$G(u)\neq\emptyset$ and so $SG(u) \neq\emptyset$
for every $u\in\overline{B}_{R}$. Indeed,
since $g$ takes nonempty, compact values and satisfies
(g2)-(g3), for each strongly measurable function $u$
there exists a strongly measurable selection $w$
of $g(.,u(.) ) $
(see \cite{deimling}, Proof of Proposition 3.5 (a)). Next, if
$u\in L^p([0,T] ;E) $, (g4)
guarantees $w\in L^{q}([0,T] ;E) $. Hence $w\in G(u)$.
\\
(b) The values of $SG$ are acyclic according to condition (SG).
\\
(c) The graph of $SG$ is closed. To show this, let
$(u_{n},v_{n})\in \mathop{\rm graph}(SG)$, $n\geq1$,
with $| u_{n}-u|_p$, $| v_{n}-v|_p \to 0$ as $n\to \infty$.
Let $\ v_{n}=S(w_{n}) $, $w_{n}\in L^{q}([0,T];E)$,
$w_{n}\in G( u_{n})$.
Since $| u_{n}-u|_p\to 0$, by Theorem 2.3 we may suppose that for
every $t\in I$, there exists a compact
set $C\subset E$ with $\{ u_{n}( t) ;\,n\geq1\}\subset C$.
Furthermore, since $g$ satisfies (g3) and has compact values,
Theorem 2.2 (b) guarantees that $g(t,C) $ is
compact. Consequently, $\{ w_{n}(t): n\geq1\}$ is relatively
compact in $E$. If we also take into account (g4) we
may apply Theorem 2.4 to conclude that (at least for a
subsequence) $(w_{n})$ converges weakly in $L^{q}(I;E) $
to some $w$. As in \cite{frigon}, since
$g$ has convex values and satisfies (g3), we
can show that $w\in G(u) $. Furthermore,
by using Lemma 4.4 and a suitable subsequence we deduce
$S(w_{n}) \to S(w) $. Thus $v=S(w)$ and so
$(u,v)\in\mathop{\rm graph}(SG)$.
\\
(d) We show that $SG(M) $ is relatively compact for each compact
$M\subset\overline{B}_{R}$. Let $M\subset\overline{B}_{R}$ be a
compact set and $(v_{n}) $ be any sequence of elements of $SG(M)
$. We prove that $(v_{n})$ has a convergent subsequence. Let
$u_{n}\in M$ and $w_{n}\in L^{q}([0,T] ;E)$ with
\[
v_{n}=S(w_{n}) \;\,\text{and\ }w_{n}\in
G(u_{n}) .
\]
The set $M$ being compact, we may assume that $|u_{n}-u|_p\to 0$
for some $u\in\overline{B}_{R}$. As above, there exists a $w\in
G(u)$ with $w_{n} \rightharpoonup w$ weakly in $L^{q}([0,T] ;E)$
(at least for a subsequence) and $S(w_{n}) \to S(w) $. Hence
$v_{n}\to $ $S(w) $. Thus (H1) is completely verified.
\\
(e) Finally, we check (H2). Suppose $M\subset\overline{B}_{R}$,
$M\subset\mathop{\rm conv}(\{ 0\} \cup SG(M)) $ and
$\overline{M}=\overline{C}$ for some countable set $C\subset M$.
Since
\[
C\subset M\subset\mathop{\rm conv}(\{ 0\} \cup SG(
M) ) \quad \text{and}\quad C\text{ is countable,}%
\]
we can find a countable set $V=\{ v_{n}: n\geq1\} \subset SG(M)$
with $C\subset\mathrm{conv\,}(\{ 0\} \cup V)$.
Then, there exists $u_{n}\in M$ and $w_{n}\in L^{q}([0,T];E)$
with
\[
v_{n}=S(w_{n}) \quad \text{and} \quad w_{n}\in G(u_{n}).
\]
From (S2) and (g4) with $v_{n}\in V$ and
$v_{0}=S(0) $, we have
\begin{eqnarray*}
| v_{n}(t) -v_{0}(t) |
&=&| S(w_{n}) (t) -S(0) (t)| \\
&\leq& \int_{I}k(t,s) | w_{n}(s)| ds\\
&\leq& \int_{I}k(t,s) (a(s) +b| u_{n}(s) | ^{p/q}) ds\\
&\leq& (| a|_{q}+bR^{p/q}) | k(t,.) |_{r}.
\end{eqnarray*}
Hence
\begin{equation}
| v_{n}(t) | \leq| S(0) (t) | +(| a|_{q}+bR^{p/q}) |
k(t,.) |_{r}\;\;\text{a.e. on }I\label{eq90}
\end{equation}
for every $n\geq1$. From
$M\subset\overline{C}\subset\overline{\mathrm{conv}}\,(\{ 0\} \cup V)$
it follows that (\ref{eq90}) is also true with any
$u\in M$ instead of $v_{n}$.
Since $V$ and $\{ w_{n}: n\geq1\} $ are countable sets of strongly
measurable functions, we may suppose that their
values belong to a separable closed subspace $E_{0}$ of
$E$. Clearly, the same is true for
$\overline{M}=\overline{C}$. Now Lemma 4.3 guarantees
\begin{eqnarray*}
\beta_{E_{0}}(M(t) ) &=&\beta_{E_{0}}(
C(t) ) \leq\beta_{E_{0}}(V(t)) \\
& =&\beta_{E_{0}}(\{ S(w_{n}) (t) : \,n\geq1\} ) \\
&\leq& \int_{I}k(t,s) \beta_{E_{0}}(\{ w_{n}(s) :\;n\geq1\} ) ds
\end{eqnarray*}
while (g5) gives
\[
\beta_{E_{0}}(\{ w_{n}(s) :\;n\geq1\}
) \leq\beta_{E_{0}}(g(s,M(s) ) \cap
E_{0}) \leq\omega(s,\beta_{E_{0}}(M(s)) ) .
\]
It follows
\[
\beta_{E_{0}}(C(t) ) \leq\int_{I}k(
t,s) \omega(s,\beta_{E_{0}}(C(s) )) ds.
\]
Moreover, the function $\varphi$ given by $\varphi(t)
=\beta_{E_{0}}(C(t) ) $ belongs to $L^p(I;\mathbb{R}_{+}) $.
Consequently, $\varphi=0$, and so
$\varphi(t) =\beta_{E_{0}}(M(t) ) =0$ a.e.
$t\in[0,T] $. Moreover, according to (\ref{eq90}) and Assumption
(g4) we have
\[
| w_{n}(t) | \leq a(t) +b(| S(0) (t) |
+(|a|_{q}+bR^{p/q}) | k(t,.) |_{r}) ^{p/q}:=\nu(t)
\]
a.e. on $I$, and $\nu\in L^{q}(I) $. Let $( v_{n_{k}})_{k\geq1}$
be any subsequence of $V$.
Then, as at step (c), $(w_{n_{k}})_{k\geq1}$ has a weakly
convergent subsequence in $L^q(I;E), $ say to $w$.
Owing to Lemma 4.4 the corresponding subsequence of
$(S(w_{n_{k}}) )_{k\geq1}=(v_{n_{k}})_{k\geq1}$ converges to
$S(w)$ in $L^p(I;E) $. Hence $V$
is relatively compact. Now Mazur's Lemma guarantees
$\overline{\mathop{\rm conv}}(\{ 0\} \cup V) $
is compact and so $\overline{C}=\overline{M}$
is compact too. Thus (H2) also holds and Theorem 4.1 applies.
\hfill $\Box$
\begin{remark} \rm
The following condition is sufficient for \emph{(SG)} to hold:
\begin{enumerate}
\item[(S3)] $S$ is affine, i.e.
\[
S(\lambda w_{1}+(1-\lambda) w_{2}) =\lambda
S(w_{1}) +(1-\lambda) S(w_{2})
\]
for all $w_{1},w_{2}\in L^{q}(I;E) $, or for all
$w_{0},\,w_{1},\,w_{2}\in L^{q}(I;E) $, the
relation $S(w_{1}) =S(w_{2}) $ implies
\[
S(1_{[0,\lambda] }w_{1}+1_{[\lambda,T]
}w_{0}) =S(1_{[0,\lambda] }w_{2}+1_{[
\lambda,T] }w_{0})
\]
for every $\lambda\in I$.
Here $1_{[a,b]}$ is the characteristic function of the
interval $[a,b]$.
\end{enumerate}
\end{remark}
Indeed, let $u\in K$ and $v_{0}\in SG(u) $. Then $v_{0}=S(w_{0}) $
for some $w_{0}\in G(u) $. Define $H:[0,1] \times SG(u)\to $ $SG(u)$,
\[
H(\lambda,v) =S(1_{[0,(1-\lambda) T] }w+1_{[(1-\lambda) T,T] }w_{0})
\]
where $w\in G(u)$ and $v=S(w)$.
According to (S3), the definition of $H(\lambda,v)$ does not depend on the
choice of $w$. Clearly,
\[
H(0,v) =v\;\quad \text{and}\quad H(1,v) =v_{0}.
\]
It remains to prove the continuity of $H$. Let $\lambda_{n}\to
\lambda$ and $v_{n}\to v$ with $v_{n}=S(w_{n}) $
and $w_{n}\in G(u) $. As at step
(c) in the Proof of Theorem 4.2, we show that a subsequence of
$(w_{n}) $ converges weakly in
$L^{q}(I;E) $ to some $w$, and
$w\in G(u) $. Finally, by Lemma 4.4 we obtain
\[
S(1_{[0,(1-\lambda_{n}) T] } w_{n}+1_{[(1-\lambda_{n}) T,T] }w_{0})
\to S(1_{[0,(1-\lambda) T]}w+1_{[(1-\lambda) T,T] }w_{0})
\]
in $L^p(I;E)$. Hence $H(\lambda_{n},v_{n}) \to H(\lambda,v) $.
Thus $SG(u) $ is contractible and so acyclic for every $u\in K$.
Note that (S3) holds whenever $S$ is one-to-one. An open
problem is to find weaker conditions to guarantee (SG) in order to extend the
applicability of Theorems 4.1-4.2. For example, we may think to find
conditions such that the values of $SG$ are
$R_{\delta}$-sets. Such conditions are known for particular classes
of problems (see \cite{bader}).
\begin{remark} \rm
A sufficient condition for (H3) is
\begin{equation}
| S(0) |_p+(| a|_{q}+bR^{p/q}) | | k(t,.) |_{r}|_p\leq R \label{eq21}
\end{equation}
if $p<\infty$ and respectively,
\[
| S(0) |_{\infty}+| a|_{q}| | k(t,.) |_{r}|_{\infty}\leq R
\]
if $p=\infty$.
\end{remark}
Indeed, if $u\in\overline{B}_{R}$ is any solution of
$u\in\lambda SG(u)$ for some $\lambda\in]0,1[$ and $u=\lambda S(w) $
with $w\in G(u)$, then for almost every $t\in[0,T]$, we have
\begin{eqnarray*}
| u(t) | &=&\lambda| S(w) (t) | \leq\lambda| S(0) (t)
| +\lambda\int_{I}k(t,s) (a(s)+b| u(s) | ^{p/q}) ds \\
&\leq&\lambda| S(0) (t) |+\lambda| k(t,.) |_{r}| a+b| u|
^{p/q}|_{q} \\
&\leq&\lambda[| S(0) (t) |+| k(t,.) |_{r}(| a|_{q}+b| u|_p^{p/q}) ] .
\end{eqnarray*}
This and (\ref{eq21}) yield
\begin{eqnarray*}
| u|_p&\leq&\lambda[| S(0) |_p+| | k(t,.) |_{r}|_p(
| a|_{q}+b| u|_p^{p/q}) ]\\
&\leq&\lambda[| S(0) |_p+| |k(t,.) |_{r}|_p(| a|_{q}+bR^{p/q}) ] r$ such that
$k(t,.) \in L^{r'}[0,T]$ for a.a. $t\in I$ and the map
$t\mapsto k(t,.)$ belongs to $L^p(I;L^{r'}(I))$.
In addition suppose that for every separable closed subspace
$E_{0}$ of $E$, there exists a
function $\delta\in L^{\infty}(I) $ such that for almost every
$t\in I,$
\begin{equation}
\beta_{E_{0}}(g(t,M) \cap E_{0}) \leq\delta(
t) \beta_{E_{0}}(M)
\end{equation}
for every subset $M\subset E_{0}$ satisfying
\[
\,| M| \leq| S(0) (t) |+(| a|_p+bR) | k(t,.)|_{r}\,.
\]
Then (\ref{eq10}) has at least one solution $u$ in $K$.
\end{corollary}
\paragraph{Proof}
We apply Theorem 4.1 to $U=\{u\in K: \|u\| |S(0)|_p$
and a suitable equivalent norm $\|\cdot \| $ on $L^p(I;E)$.
According to the proof of Theorem 4.2 and of Corollary 4.6, the assumptions
(H1)-(H2) are fulfilled. It remains to guarantee (H3).
Let $u\in K$ be any solution of $u\in\lambda SG(u)$
for some $\lambda\in]0,1[$. Then, for any $\theta>0$, we have
\[
| u(t) | \leq\lambda| S(0)(t) | +\lambda\int_{0}^{t}k(t,s) e^{\theta
s}(| a(s) | +b| u(s)| e^{-\theta s}) ds.
\]
Define an equivalent norm on $L^p(I;E)$, by
\[
\| u\| =| u(t) e^{-\theta t}|_p.
\]
Then, since $1/r'+(r'-r) /(
rr') +1/p=1$, H\"{o}lder's inequality guarantees
\begin{eqnarray*}
| u(t) | & \leq& \lambda| S(0)(t) | +\lambda| k(t,.) |
_{r'}(| a|_p+b\left\| u\right\| )(\int_{0}^{t}e^{\theta rr'/(r'-r)
s}ds) ^{(r'-r) /(rr') } \\
&\leq&\lambda| S(0) (t) |+\lambda| k(t,.) |_{r'}(|
a|_p+b\left\| u\right\| ) (\frac{r'-r}{\theta rr'}) ^{(r'-r) /(
rr') }e^{\theta t}.
\end{eqnarray*}
Consequently
\begin{equation}
\left\| u\right\| \leq\lambda[| S(0) |_p+(| a|_p+b\left\| u\right\| )(
\frac{r'-r}{\theta rr'}) ^{(r'-r)
/(rr') }| | k(t,.) |
_{r'}|_p] . \label{eq222}%
\end{equation}
Now we choose $\theta>0$ so large that%
\[
| S(0) |_p+(| a|_p+bR) (\frac{r'-r}{\theta rr'})^{(r'-r) /(rr') }| |
k(t,.) |_{r'}|_p\leq R.
\]
Then, since $\lambda<1$, from (\ref{eq222}) we have
$\left\| u\right\| 0$, there exists a
positive integer $n_{0}=n_{0}(M,\varepsilon) $ such that
\[
| S_{n_{0}}(w) -S(w) |_{\infty}\leq\varepsilon\quad \text{for all }w\in M,
\]
that is $S_{n_{0}}(M) $ is an $\varepsilon$-net for $S(M) $. We
omit here the details.
\\
(3) From (\ref{eq78}) we see that for each $n$ and any
bounded $M\subset L^{2}(I;E) $, the set $S_{n}(M) $ is bounded in
$C(I;E) $. In addition, using
\[
u_{n}(t) =-\int_{0}^{T}G(t,s) [A_{n}u_{n}(s) +w(s) ] ds
\]
and the Lipschitz property of $A_{n}$, we obtain
\[
| u_{n}(t) -u_{n}(t') |\leq\int_{0}^{T}| G(t,s) -G(t',s)
| [2n| u_{n}(s) | +| w(s) | ] ds.
\]
This implies the equicontinuity of $S_{n}(M) $.
Now we consider a compact, convex subset $C$ of
$E$ and a countable set $M\subset L^{2}(I;C) $, We claim that
$S_{n}(M)(t) $ is relatively compact in $E$ for every $t\in I$.
Indeed, for any $w\in M$, the unique solution $u_{n}=S_{n}(w) $
of (\ref{eq79}) satisfies
\[
-u_{n}''+nu_{n}=nJ_{n}u_{n}-w\quad \text{a.e. on }I.
\]
If we denote by $\widetilde{G}$ the Green function of the
operator $-u''+nu$ corresponding to the
boundary conditions $u(0) =u(T) =0$, then
\begin{equation}
u_{n}(t) =\int_{0}^{T}\widetilde{G}(t,s) [
nJ_{n}u_{n}(s) -w(s) ] ds.\label{eq80}
\end{equation}
Using a result by Heinz, really a particular case of Lemma 4.3, the
nonexpansivity of $J_{n}$ and the inclusion $M(s) \subset C$ a.e. on $I$,
from (\ref{eq80}), we obtain
\begin{equation}
\beta_{0}(S_{n}(M) (t) ) \leq
n\int_{0}^{T}\widetilde{G}(t,s) \beta_{0}(S_{n}(M) (s) ) ds.\label{eq81}
\end{equation}
Here $\beta_{0}$ is the ball measure of noncompactness
corresponding to a suitable separable closed subspace of $E$.
Let
\[
\varphi(t) =\beta_{0}(S_{n}(M) (t) ), \quad v(t)
=\int_{0}^{T}\widetilde {G}(t,s) \varphi(s) ds.
\]
We have
\[
-v''+nv=\varphi,\quad v(0) =v(T) =0.
\]
According to (\ref{eq81}), $\varphi\leq nv$. Hence $-v''\leq0$.
This, since $v(0) =v(T) =0$, implies $v\leq0$ on $I$. The function $v$
being nonnegative, it follows $v\equiv0$. Thus
$\beta_{0}(S_{n}(M) (t) )=0$ for all $t\in I$, that is
$S_{n}(M) (t) $ is relatively compact in $E$. As a result, $S_{n}(M) $
is relatively compact in $C(I;E) $.
Therefore, we have shown that for each $\varepsilon>0$, there
exists a relatively compact $\varepsilon$-net of $S(M) $.
By Hausdorff's Theorem, $S(M) $ is relatively compact in $C(I;E) $.
\hfill $\Box$
\begin{remark} \rm
Proposition 5.1 together with Theorem 4.2 gives new existence
results for the problem (\ref{eq33}) if the multivalued
perturbation $g$ satisfies (g1)--(g5) and (SG).
\end{remark}
Similar results can be obtained for problems of type (\ref{eq33}) with some
other boundary conditions like those in \cite{pavel} and \cite{haraux}.
\begin{thebibliography}{99} \frenchspacing
\bibitem{pavel} A.R. Aftabizadeh, S. Aizicovici and N.H. Pavel, On a class of
second-order anti-periodic boundary value problems, \textit{J. Math. Anal.
Appl. }\textbf{171} (1992), 301-320.
\bibitem {bader}R. Bader, On the semilinear multi-valued flow under
constraints and the periodic problem, \textit{Comment. Math. Univ.
Carolin. }\textbf{41} (2000), 719-734.
\bibitem {barbu}V. Barbu, ``Nonlinear Semigroups and Differential Equations in
Banach Spaces'', Ed. Academiei \& Noordhoff International Publishing,
Bucure\c{s}ti-Leyden, 1976.
\bibitem {couchouron3}J.-F. Couchouron, Probl\`{e}me de Cauchy non autonome
pour des \'{e}quations d'\'{e}volution, \textit{Potential Anal.} \textbf{13}
(2000), 213-248.
\bibitem {couchouron}J.-F. Couchouron and\ M. Kamenski, A unified topological
point of view for integro-differential inclusions,\textit{\ }in ``Differential
Inclusions and Optimal Control'', Lecture Notes in Nonlinear
Analysis, Vol. 2, 1998, 123-137.
\bibitem {couchouron2}J.-F. Couchouron and\ M. Kamenski, An abstract
topological point of view and a general averaging principle in the theory of
differential inclusions,\textit{\ Nonlinear Anal. }\textbf{42} (2000), 1101-1129.
\bibitem {deimling}K. Deimling, ``Multivalued Differential Equations'', Walter
de Gruyter, Berlin-New York, 1992.
\bibitem {diestel}J. Diestel, W.M. Ruess and\ W. Schachermayer, Weak
compactness in $L^{1}(\mu,\,X) $, \textit{Proc. Amer. Math.
Soc.} \textbf{118} (1993), 447-453.
\bibitem {eilenberg}S. Eilenberg and\ D. Montgomery, Fixed point theorems for
multivalued transformations,\textit{ Amer. J. Math.} \textbf{68} (1946), 214-222.
\bibitem {fitzpatrick}P.M. Fitzpatrick and\ W.V. Petryshyn, Fixed point
theorems for multivalued noncompact acyclic mappings,\textit{ Pacific J.
Math.} \textbf{54} (1974), 17-23.
\bibitem {frigon}M. Frigon, Th\'{e}or\`{e}mes d'existence de solutions
d'inclusions diff\'{e}rentielles, in ``Topological Methods in Differential
Equations and Inclusions'' (A. Granas and M. Frigon eds.), NATO\ ASI\ Series
C, Vol. 472, Kluwer Academic Publishers, Dordrecht-Boston-London, 1995, 51-87.
\bibitem {gorniewicz}L. Gorniewicz, ``Homological Methods in Fixed Point
Theory of Multivalued Maps'', Dissertationes Math. 129, Polish Scientific
Publishers, Warsaw, 1976.
\bibitem {guo}D. Guo, V. Lakshmikantham and\ X. Liu, ``Nonlinear Integral
Equations in Abstract Spaces'', Kluwer Academic Publishers,
Dordrecht-Boston-London, 1996.
\bibitem {gutman}S. Gutman, Existence theorems for nonlinear evolution
equations, \textit{Nonlinear Anal. }\textbf{11} (1987), 1193-1206.
\bibitem {haraux}A. Haraux, Anti-periodic solutions of some nonlinear
evolution equations, \textit{Manuscripta Math.} \textbf{63} (1989), 479-505.
\bibitem {hu}S. Hu and\ N.S. Papageorgiou, ``Handbook of Multivalued Analysis,
Vol. I: Theory'', Kluwer Academic Publishers, Dordrecht-Boston-London, 1997.
\bibitem {orp}D. O'Regan and\ R. Precup, Fixed point theorems for set-valued
maps and existence principles for integral inclusions, \textit{J. Math. Anal.
Appl. }\textbf{245} (2000), 594-612.
\bibitem {orp2}D. O'Regan and\ R. Precup, Integrable solutions of Hammerstein
integral inclusions in Banach spaces, \textit{Dynam. Contin. Discrete Impuls.
Systems}, to appear.
\bibitem {precup}R. Precup, A M\"{o}nch type generalization of the
Eilenberg-Montgomery fixed point theorem, \textit{Seminar on Fixed
Point Theory Cluj-Napoca} \textbf{1} (2000), 69-71.
\bibitem {vrabie}I.I. Vrabie, ``Compactness Methods for Nonlinear
Evolutions'', Pitman Monographs and Surveys in Pure and Applied Mathematics,
Vol. 32, Longman Scientific \& Technical, 1987.
\end{thebibliography}
\noindent\textsc{Jean-Fran\c{c}ois Couchouron}\\
Universit\'{e} de Metz, Math\'{e}matiques INRIA Lorraine, \\
Ile du Saulcy, 57045 Metz, France\\
e-mail: couchour@loria.fr \smallskip
\noindent\textsc{Radu Precup}\\
University Babe\c s-Bolyai, \\
Faculty of Mathematics and Computer Science, \\
3400 Cluj, Romania\\
e-mail: r.precup@math.ubbcluj.ro
\end{document}