on volterra integral equations in banach spaces proof. put

12
Funkcialaj Ekvacioj, 20 (1977) 247-258 On Volterra Integral Equations in Banach Spaces By Stanislaw SZUFLA (A. Mickiewicz University, Poland) Let $J=[0, a]$ be a compact interval in $R$ , and let $E$ be a Banach space with the norm $|| cdot||$ . Denote by $C=C(J, E)$ the Banach space of all continuous functions $u:J rightarrow E$ with the norm $||u||_{c}= sup$ $ {||u(t)||:t in J }$ . For the case where $g$ is a bounded continuous function of $(t, s, x)$ on $ {0 leq s leq t leq a } times R^{n}$ , M. Hukuhara [2] has shown that the -approximate solutions set of the Volterra integral equation (1) $x(t)=f(t)+ int_{0}^{t}g(t, s, x(s))ds$ is connected in $C(J, R^{n})$ . In this paper we prove the connectedness of the set $ {x in C:||x-F(x)||_{c}<_{ epsilon} }$ for a certain class of non-linear mappings $F:C rightarrow C$ . As an easy application of our result, we show that Hukuhara’ theorem is also true when $g$ satisfies the Caratheodory conditions. Moreover, this note contains an existence theorem for the equation (1) in Banach spaces. 1. First we shall show the following main theorem. Theorem 1. Assume that $F:C rightarrow C$ is a continuous mapping such that 1o $F(C)$ is an equiunijormly continuous set of mappings $J rightarrow E$ ; 2o there exists $b in E$ such that $F(x)(0)=b$ for every $x in C$ ; 3o $x|[0, d]=y|[0, d] Rightarrow F(x)|[0, d]=F(y)|[0, d]$ for every $x$ , $y in C$ and $0<d leq a$ . Then for any $ eta>0$ the set $S_{ eta}= {u in C:u(0)=b, ||u-F(u)||_{c}<_{ eta} }$ is non-empty and connected. Proof. Put $ w(h)= sup$ $ {||F(x)(t)-F(x)(s)||:x in C, t, s in J, |t-s| leq h }$ . From 1o it follows that $ lim_{h-0+}w(h)=0$ . Let $x in S_{ eta}$ . Choose $ epsilon>0$ such that $||x-F(x)||_{c}+w( epsilon)<_{ eta}$ . For any $p$ , $0 leq p leq a$ ,

Upload: others

Post on 15-Mar-2022

2 views

Category:

Documents


0 download

TRANSCRIPT

Funkcialaj Ekvacioj, 20 (1977) 247-258

On Volterra Integral Equations in Banach Spaces

By

Stanislaw SZUFLA

(A. Mickiewicz University, Poland)

Let $J=[0, a]$ be a compact interval in $R$ , and let $E$ be a Banach space withthe norm $||¥cdot||$ . Denote by $C=C(J, E)$ the Banach space of all continuous functions$u:J¥rightarrow E$ with the norm $||u||_{c}=¥sup$ $¥{||u(t)||:t¥in J¥}$ . For the case where $g$ is abounded continuous function of $(t, s, x)$ on $¥{0¥leq s¥leq t¥leq a¥}¥times R^{n}$, M. Hukuhara [2] hasshown that the $¥epsilon$ -approximate solutions set of the Volterra integral equation

(1) $x(t)=f(t)+¥int_{0}^{t}g(t, s, x(s))ds$

is connected in $C(J, R^{n})$ . In this paper we prove the connectedness of the set$¥{x ¥in C:||x-F(x)||_{c}<_{¥epsilon}¥}$ for a certain class of non-linear mappings $F:C¥rightarrow C$ . As aneasy application of our result, we show that Hukuhara’ $¥mathrm{s}$ theorem is also true when$g$ satisfies the Caratheodory conditions. Moreover, this note contains an existencetheorem for the equation (1) in Banach spaces.

1. First we shall show the following main theorem.

Theorem 1. Assume that $F:C¥rightarrow C$ is a continuous mapping such that1o $F(C)$ is an equiunijormly continuous set of mappings $J¥rightarrow E$ ;2o there exists $b¥in E$ such that $F(x)(0)=b$ for every $x¥in C$ ;3o $x|[0, d]=y|[0, d]¥Rightarrow F(x)|[0, d]=F(y)|[0, d]$

for every $x$ , $y¥in C$ and $0<d¥leq a$ .Then for any $¥eta>0$ the set

$S_{¥eta}=¥{u¥in C:u(0)=b, ||u-F(u)||_{c}<_{¥eta}¥}$

is non-empty and connected.

Proof. Put

$ w(h)=¥sup$ $¥{||F(x)(t)-F(x)(s)||:x¥in C, t, s¥in J, |t-s|¥leq h¥}$ .

From 1o it follows that $¥lim_{h-0+}w(h)=0$ .Let $x¥in S_{¥eta}$ . Choose $¥epsilon>0$ such that $||x-F(x)||_{c}+w(¥epsilon)<_{¥eta}$ . For any $p$ , $0¥leq p¥leq a$ ,

248 S. SZUFLA

we define a function $y(¥cdot, p)$ by the formula

(2) $y(t, p)=z_{k}(t, p)$ for $0¥leq t¥leq a$ ,

where $k$ is an integer $¥geq 1$ such that $ p+(k-1)¥epsilon¥leq a<p+k¥epsilon$, and $z_{i}(t, p)$ , $i=1$ , $¥cdots$ , $k$ ,are defined inductively by

$z_{1}(t, p)=¥left¥{¥begin{array}{l}x(t)¥mathrm{f}¥mathrm{o}¥mathrm{r}0¥leq t¥leq p¥¥x(p)¥mathrm{f}¥mathrm{o}¥mathrm{r}p¥leq t¥leq a¥end{array}¥right.$

and

$z_{i+1}(t, p)=¥left¥{¥begin{array}{l}z_{i}(t,p)¥mathrm{f}¥mathrm{o}¥mathrm{r}0¥leq t¥leq p+i¥epsilon¥¥ x(p)-F(x)(p)+F(z_{i}(¥cdot,p))(t-¥epsilon)¥mathrm{f}¥mathrm{o}¥mathrm{r}p+i¥epsilon¥leq t¥leq a.¥end{array}¥right.$

Then $y(¥cdot, p)¥in C$ and

$y(t, p)=¥{xxx(p)-F(x)(p)+F(y(¥cdot, p))(t-¥epsilon)(t)¥mathrm{f}¥mathrm{o}¥mathrm{r}0¥leq t¥leq p(p)¥mathrm{f}¥mathrm{o}¥mathrm{r}p¥leq t¥leq¥min(a,p+¥epsilon)$

for $¥min(a, p+¥epsilon)¥leq t¥leq a$ .

Moreover

$||y(t, p)-F(y(¥cdot, p))(t)||=||x(t)-F(x)(t)||<_{¥eta}$ for $0¥leq t¥leq p$ ;$||y(t, p)-F(y(¥cdot, p))(t)||=||x(p)-F(x)(p)+F(y(¥cdot, p))(p)$

$-F(y(¥cdot, p))(t)||¥leq||x-F(x)||_{c}+w(¥epsilon)<_{¥eta}$

for $p¥leq t¥leq¥min(a, p+¥epsilon)$ ;$||y(t, p)-F(y(¥cdot, p))(t)||=||x(p)-F(x)(p)+F(y(¥cdot, p))(t-¥epsilon)$

$-F(y(¥cdot, p))(t)||¥leq||x-F(x)||_{c}+w(¥epsilon)<_{¥eta}$

for $¥min(a, p+¥epsilon)¥leq t¥leq a$ .

This proves that $y(¥cdot, p)¥in S_{¥eta}$ . We shall now show that

(3) $¥lim$ $||y(¥cdot, p)-y(¥cdot, q)||_{c}=0$ .$p-q$

Let $0¥leq p¥leq q¥leq a$ and $ q-p¥leq¥epsilon$ . We see that

$||y(t, p)-y(t, q)||=0$ for $0¥leq t¥leq p$ ;

$||y(t, p)-y(t, q)||=||x(p)-x(t)||$ for $p¥leq t¥leq q$ ;

$||y(t, p)-y(t, q)||=||x(p)-x(q)||$ for $q¥leq t¥leq¥min(a, p+¥epsilon)$ .

Moreover,

On Volterra Integral Equations in Banach Spaces 249

a) if $ 0¥leq p¥leq q¥leq a-¥epsilon$ , then

$||y(t, p)-y(t, q)||=||x(p)-F(x)(p)+F(y(¥cdot, p))(t-¥epsilon)-x(q)||$

$=||x(p)-x(q)+F(y(¥cdot, p))(t-¥epsilon)-F(y(¥cdot, p))(p)||$

$¥leq||x(p)-x(q)||+w(q-p)$ for $ p+¥epsilon¥leq t¥leq q+¥epsilon$ ,

and

$||y(t, p)-y(t, q)||=||x(p)-F(x)(p)+F(y(¥cdot, p))(t-¥epsilon)$

$-x(q)+F(x)(q)-F(y(¥cdot, q))(t-¥epsilon)||¥leq||x(p)-x(q)||+w(q-p)$

$+||F(y(¥cdot, p))(t-¥epsilon)-F(y(¥cdot, q))(t-¥epsilon)||$ for $q+¥epsilon¥leq t¥leq a$ .

b) if $0¥leq p¥leq a-¥epsilon¥leq q¥leq a$ , then

$||y(t, p)-y(t, q)||=||x(p)-F(x)(p)+¥Gamma’(y(¥cdot, p))(t-¥epsilon)-x(q)||$

$¥leq||x(p)-x(q)||+w(q-p)$ for $p+¥epsilon¥leq t¥leq a$ .

c) if $a-¥epsilon¥leq p¥leq q¥leq a$ , then

$||y(t, p)-y(t, q)||=||x(p)-x(q)||$ for $q¥leq t¥leq a$ .

From this we deduce that if $0¥leq p$ , $q¥leq a$ and $|q-p|¥leq¥epsilon$ , then

(4) $||y(t, p)-y(t, q)||¥leq r(x, |p-q|)+w(|p-q|)$ for $ 0¥leq t¥leq¥epsilon$ ,

and

$||y(t, p)-y(t, q)||¥leq r(x, |p-q|)+w(|p-q|)$(5)

$+||F(y(¥cdot, p))(t-¥epsilon)-F(y(¥cdot, q))(t-¥epsilon)||$ for $¥epsilon¥leq t¥leq a$ ,

where $ r(x, h)=¥sup$ $¥{||x(t)-x(s)||:t, s¥in J, |t-s|¥leq h¥}$ . From (4) it is clear that

$¥lim$ $y(t, p)=y(t, q)$ uniformly on $[0, ¥epsilon]$ .$p-q$

By 3o and the continuity of $F$ from this it follows that

$¥lim$ $||F(y(¥cdot, p))(t-¥epsilon)-F(y(¥cdot, q))(t-¥epsilon)||=0$ uniformly on $[¥epsilon, 2¥epsilon]$ ,$p-q$

and hence, by (5),

$¥lim$ $y(t, p)=y(t, q)$ uniformly on $[0, 2¥epsilon]$ .$p-q$

By repeating this argument we find $¥lim_{p-q}y(t, p)=y(t, q)$ uniformly on $[0, n¥epsilon]$ , $n=$

$1,2$ , $¥cdots$ , $[a/¥epsilon]$ , which proves (3).Choose a positive number $¥beta$ such that $ w(¥beta)<¥eta$ and $¥beta<a$ . For any $¥epsilon$ , $0<_{6}$

250 S. SZUFLA

$<¥beta$ , we define a function $v(¥cdot, ¥epsilon)$ by the formula

$v(t, ¥epsilon)=w_{k}(t, ¥epsilon)$ for $0¥leq t¥leq a$ ,

where $k$ is an integer $¥geq 1$ such that $(k-1)¥epsilon¥leq a<k¥epsilon$ , and $w_{i}(t, ¥epsilon)$ , $i=1$ , $¥cdots$ , $k$ , aredefined inductively by

$w_{1}(t, ¥epsilon)=b$ for $0¥leq t¥leq a$

and

$w_{i+1}(t, ¥epsilon)=¥left¥{¥begin{array}{l}w_{i}(t,¥epsilon)¥mathrm{f}¥mathrm{o}¥mathrm{r}0¥leq t¥leq i¥epsilon¥¥ F(w_{i}(¥cdot,¥epsilon))(t-¥epsilon)¥mathrm{f}¥mathrm{o}¥mathrm{r}i¥epsilon¥leq t¥leq a.¥end{array}¥right.$

Obviously $v(¥cdot, ¥epsilon)$ is continuous on $J$ and

$v(t, ¥epsilon)=¥left¥{¥begin{array}{l}b¥¥F(v(¥cdot,¥epsilon))(t-¥epsilon)¥end{array}¥right.$ $¥mathrm{f}¥mathrm{f}¥mathrm{o}¥mathrm{r}¥epsilon¥leq t¥leq a¥mathrm{o}¥mathrm{r}0¥leq t¥leq¥epsilon$

.

Moreover $v(¥cdot, ¥epsilon)¥in S_{¥eta}$ , because

$||v(t, ¥epsilon)-F(v(¥cdot, ¥epsilon))(t)||=||b-F(v(¥cdot, ¥epsilon))(t)||$

$=||F(v(¥cdot, ¥epsilon))(0)-F(v(¥cdot, ¥epsilon))(t)||¥leq w(¥epsilon)<¥eta$ for $ 0¥leq t¥leq¥epsilon$ ,

and

$||v(t, ¥epsilon)-F(v(¥cdot, ¥epsilon))(t)||$

$=||F(v(¥cdot, ¥epsilon))(t-¥epsilon)-F(v(¥cdot, ¥epsilon))(t)||¥leq w(¥epsilon)<_{¥eta}$ for $¥epsilon¥leq t¥leq a$ .

Furthermore, if $ 0<_{¥epsilon}<¥delta<¥beta$ , then

$||v(t, ¥delta)-v(t, ¥epsilon)||=0$ for $ 0¥leq t¥leq¥epsilon$ ,$||v(t, ¥delta)-v(t, ¥epsilon)||=||b-F(v(¥cdot, ¥epsilon))(t-¥epsilon)||$

$=||F(v(¥cdot, ¥epsilon))(0)-F(v(¥cdot, ¥epsilon))(t-¥epsilon)||¥leq w(¥delta-¥epsilon)$ for $¥epsilon¥leq t¥leq¥delta$ ,

and

$||v(t, ¥delta)-v(t, ¥epsilon)||=||F(v(¥cdot, ¥delta))(t-¥delta)-F(v(¥cdot, ¥epsilon))(t-¥epsilon)||$

$¥leq||F(v(¥cdot, ¥delta))(t-¥delta)-F(v(¥cdot, ¥delta))(t-¥epsilon)||$

$+||F(v(¥cdot, ¥delta))(t-¥epsilon)-F(v(¥cdot, ¥epsilon))(t-¥epsilon)||$

$¥leq w(¥delta-¥epsilon)+||F(v(¥cdot, ¥delta))(t-¥epsilon)-F(v(¥cdot, ¥epsilon))(t-¥epsilon)||$ for $¥delta¥leq t¥leq a$ .

Thus for each $¥epsilon$ , $¥delta$ , $0<_{¥epsilon}$ , $¥delta<¥beta$ , we have

$||v(t, ¥delta)-v(t, ¥epsilon)||¥leq w(|¥delta-¥epsilon|)$ for $ 0¥leq t¥leq¥epsilon$

On Volterra Integral Equations in Banach Spaces 251

and

$||v(t, ¥delta)-v(t, ¥epsilon)||¥leq w(|¥delta-¥epsilon|)+||F(v(¥cdot, ¥delta))(t-¥epsilon)-F(v(¥cdot, ¥epsilon))(t-¥epsilon)||$ for $¥epsilon¥leq t¥leq a$ .

Hence, using the same argument as in the proof of (3), we can prove that

$¥lim_{¥delta-¥mathrm{e}}||v(¥cdot, ¥delta)-v(¥cdot, ¥epsilon)||_{c}=0$ .

From this we conclude that the set $V=¥{v(¥cdot, ¥epsilon):0<_{¥epsilon}<¥beta¥}$ is connected in C.Further, for any $x¥in S_{¥eta}$ we choose $¥epsilon=¥epsilon_{x}$ such that $ 0<_{¥epsilon}<¥beta$ and $||x-F(x)||_{c}+$

$w(¥epsilon)<_{¥eta}$ . Put $T_{x}=¥{y(¥cdot, p):0¥leq p¥leq a¥}$ , where $y(¥cdot, p)$ denotes the function definedby (2). From (3) it follows that $T_{x}$ is a connected subset of $C$ . As $y(¥cdot, 0)=$

$v(¥cdot, ¥epsilon)¥in V¥cap T_{x}$ , the set $V¥cup T_{x}$ is connected, and therefore the set $W=¥bigcup_{x¥in S¥eta}T_{x}$

$¥cup V$ is connected in C. Moreover $S_{¥eta}¥subset W$ , because $x=y(¥cdot, a)¥in T_{x}$ for every $x¥in S_{¥eta}$ .

On the other hand $W¥subset S_{¥eta}$ , since $T_{x}¥subset S_{¥eta}$ and $V¥subset S_{¥eta}$ . Consequently $S_{¥eta}=W$, whichends the proof.

2. Now we consider the Volterra integral equation.

(1) $x(t)=i(t)+¥int_{0}^{t}g(t, s, x(s))ds$ ,

where $t¥in J$ and $x$ , $f$ and $g$ are functions with values in E.Assume thatI. $f:J¥rightarrow E$ is a continuous function,

$¥mathrm{I}¥mathrm{I}$ . $(t, s, x)¥rightarrow g(t, s, x)$ is a function of the set $¥{0¥leq s¥leq t¥leq a, x¥in E¥}$ into $E$ , whichsatisfies the following conditions:

1o for each fixed $x¥in E$ and $t¥in J$ the function $s¥rightarrow g(t, s, x)$ is strongly L-measurable on $[0, t]$ ;

2o for each fixed $t$ , $s$ , $0¥leq s¥leq t¥leq a$ , the function $x¥rightarrow g(t, s, x)$ is continuous on$E$ ;

3o there exist real-valued functions $(¥tau, t, s)¥rightarrow r(¥tau, t, ¥mathrm{s})$ and $(t, s)¥rightarrow m(t, s)$

$(0¥leq s¥leq t¥leq¥tau¥leq a)$ such that(i) for each fixed $t$ , $¥tau$ the functions $s¥rightarrow r(¥tau, t, ¥mathrm{s})$ and $s¥rightarrow m(t, s)$ are $¥mathrm{L}$ -integrable

on $[0, t]$ ;(ii) $¥sup$ $¥{||g(¥tau, s, x)-g(t, s, x)||:x¥in E¥}¥leq r(¥tau, t, ¥mathrm{s})$ and $¥sup$ $¥{||g(t, s, x)||:x¥in E¥}$

$¥leq m(t, s)$ ;

(iii) $¥lim_{¥tau-t-0+}[¥int_{t}^{¥tau}m(¥tau, s)ds+¥int_{0}^{t}r(¥tau, t, s)ds]=0$ for fixed $t$ or $¥tau$ .

A function $x:J¥rightarrow E$ is called an $¥epsilon$-approximate solution of (1) if $x(0)=f(0)$ and

$||x(t)-f(t)-¥int_{0}^{t}g(t, s, x(s))ds||<_{¥epsilon}$ for every $t¥in J$ .

252 S. SZUFLA

Theorem 2. Under the above assumptions the set of all $¥epsilon-$approximate solu-tions of (1) is non-empty and connected in $C$ .

Proof. It is sufficient to show that the mapping $F$ , defined by the formula

$F(x)(t)=f(t)+¥int_{0}^{t}g(t, s, x(s))ds$ for $x¥in C$ and $t¥in J$ ,

satisfies the assumptions of Theorem 1.Let us fix $t_{0}¥in J$ . For any $x¥in C$ and $t¥in J$ we have

$||F(x)(t)-F(x)(t_{0})||¥leq||f(t)-f(t_{0})||$

$+¥int_{t¥mathrm{o}}^{t}||g(t, s, x(s))||ds+¥int_{0}^{t¥mathrm{o}}||g(t, s, x(s))-g(t_{0}, s, x(s))||ds$

$¥leq||f(t)-f(t_{0})||+¥int_{t¥mathrm{o}}^{t}m(t, s)ds+¥int_{0}^{t¥mathrm{o}}r(t, t_{0}, s)ds$ when $t_{0}¥geq t$ ,

and

$||F(x)(t)-F(x)(t_{0})||¥leq||f(t)-f(t_{0})||+¥int_{t}^{t¥mathrm{o}}m(t_{0}, s)ds+¥int_{0}^{t}r(t_{0}, ¥mathrm{t}, s)ds$ when $t¥leq t_{0}$ .

By 3o and the continuity of $f$ from this it follows that

$¥lim_{t}$ $F(x)(t)=F(x)(t_{0})$ uniformly in $x¥in C$ ,

which proves that $F(C)$ is an equicontinuous subset of $C$ . This implies that the set$F(C)$ is equiuniformly continuous, since $J$ is compact.

Assume that $x_{n}$ , $x¥in C$ and $¥lim_{n-¥infty}||x_{n}-x||_{c}=0$ . Since

$¥lim_{n¥rightarrow¥infty}g(t, s, x_{n}(s))=g(t, s, x(s))$ and $||g(t, s, x_{n}(s))-g(t, s, x(s))||¥leq 2m(t, s)$

for each $0¥leq s¥leq t¥leq a$ , the Lebesgue theorem proves that $¥lim_{n¥rightarrow¥infty}¥int_{0}^{t}||g(t, s, x_{n}(s))-$

$g(t, s, x(s))||ds=0$, i.e. $¥lim_{n-¥infty}F(x_{n})(t)=F(x)(t)$ for every $t¥in J$ , and hence, by theequicontinuity of $F(C)$ , $¥lim_{n-¥infty}||F(x_{n})-F(x)||_{c}=0$ . This shows that $F$ is a contin-uous mapping $C¥rightarrow C$ .

Remark. The use of the time delay $ t-¥epsilon$ for constructing the $¥eta$-approximatesolutions is due to Caratheodory in the case of ordinary differential equations (seealso [5] $)$ .

Now we shall show an existence theorem for the equation (1). Assume thata non-negative real-valued function $(¥mathrm{t}, s, z)¥rightarrow h(t, s, z)$ defined on $¥{0¥leq s¥leq t¥leq a¥}¥times R^{+}$

is a Kamke function, i.e. $h$ satisfies the Caratheodory conditions 1o-3o and the

On Volterra Integral Equations in Banach Spaces 253

function identically equal to zero is the unique continuous solution of the inequality$u(t)¥leq¥int_{0}^{t}h(t, s, u(s))ds$ for $t¥in J$ satisfying the condition $u(0)=0$ .

Furthermore, for any bounded subset $A$ of $E$ we denote by $¥alpha(A)$ the infimumof all $¥epsilon>0$ such that there exists a finite covering of $A$ by sets of diameter $¥leq¥epsilon$ (cf.[4] $)$ . The number $¥alpha(A)$ is called the measure of non-compactness of the set $A$ .For properties of $¥alpha$ see [1].

Theorem 3. Assume that for any $t¥in J$ the inequality

(7) $¥varliminf_{¥delta 0}¥alpha(g(t, I_{S,¥delta}, X))¥leq h(t, s, ¥alpha(X))$ ,

where $I_{S,¥delta}=(s-¥delta, s+¥delta)¥cap[0, t]$ , is satisfied for almost every $s¥in[0, t]$ and for everybounded subset $X$ of E. Then there exists at least one solution of (1) defined on$J$ .

The proof of Theorem 3 is suggested by a paper of Pianigiani [7] concerningdifferential equations.

Proof. Let $F$ denote the mapping defined in the proof of Theorem 2. Thereexists a sequence $(u_{n})$ such that $u_{n}¥in C$ and

(8) $n¥varliminf¥infty||u_{n}-F(u_{n})||_{c}=0$.

Put $V=¥{u_{n} : n=1,2, --¥}$ . Denote by I the identity mapping on $C$ . From (8) itfollows that $(I-F)(V)$ is an equiuniformly continuous subset of $C$ . Since

(9) $V¥subset(I-F)(V)+F(V)$

and the set $F(V)$ is equiuniformly continuous, we see that the set $V$ is also equiuni-formly continuous.

Put $ w(V, ¥delta)=¥sup$ $¥{||u(t)-u(s)||:u¥in V, t, s¥in J, |t-s|¥leq¥delta¥}$ and $V(t)=¥{u(t)$ :$u¥in V¥}$ . Since

$||u(t)-v(t)||¥leq||u(s)-v(s)||+2w(V, |t-s|)$ for $u$ , $v¥in V$ and $t$ , $s¥in J$ ,

we see that

$|¥alpha(V(t))-¥alpha(V(s))|¥leq 2w(V, |t-s|)$ for $t$ , $s¥in J$ ,

and therefore the function $t¥rightarrow¥alpha(V(t))$ is continuous on /. Moreover, let $b=$

$¥sup$ $¥{||u(t)||:u¥in V, t¥in J¥}$ .Fix $t¥in J$ and $¥epsilon>0$ . Scorza-Dragoni’ $¥mathrm{s}$ theorem [10] proves that there is an

open set $J_{¥epsilon}¥subset[0, t]$ such that

254 S. SZUFLA

$¥mu(J_{¥mathrm{e}})<_{¥epsilon}$ ($¥mu-$the Lebesgue measure); $¥int_{J_{¥mathrm{g}}}m(t, s)ds<_{6}$ ;

the function $(s, u)¥rightarrow h(t, s, u)$ is uniformly continuous on $([0, t]¥backslash J_{¥epsilon})¥times[0,2b]$ ;the inequality $¥lim_{¥delta-0}¥alpha(g(t, I_{S,¥delta}, X))¥leq h(t, s, ¥alpha(X))$ is satisfied for every $s¥in[0, t]¥backslash J_{¥epsilon}$

and each bounded subset $X$ of $E$ . Let $T=[0, t]¥backslash J_{¥mathrm{e}}$ . Putting $¥int_{Z}g(t, s, V(s))ds=$

$¥{¥int_{Z}g(t, s, u(s))ds:u¥in V¥}$ , we see that

$F(V)(t)¥subset f(t)+¥int_{T}g(t, s, V(s))ds+¥int_{J_{¥mathrm{e}}}g(t, s, V(s))ds$,

and, consequently,

(10) $¥alpha(F(V)(t))¥leq¥alpha(¥int_{T}g(t, s, V(s))ds)+¥alpha(¥int_{J_{¥xi}}g(t, s, V(s))ds)$ .

Choose $¥delta>0$ in such a way that $|s_{1}-s_{2}|<¥delta$. $|u_{1}-u_{2}|<¥delta$, $s_{1}$ , $s_{2}¥in T$ , $u_{1}$ , $u_{2}¥in[0,2b]$ ,implies $|h(t, s_{1}, u_{1})-h(t, s_{2}, u_{2})|<_{¥epsilon}$ , and choose $¥eta$ , $ 0<¥eta<¥delta$ , such that $|s_{1}-s_{2}|<¥eta$ ,$s_{1}$ , $s_{2}¥in T$ , implies $|¥alpha(V(s_{1}))-¥alpha(V(s_{2}))|<¥delta$ . We divide the interval $[0, t]$ into $n$ parts$0=t_{0}<t_{1}<¥cdots<t_{n}=t$ in such a way that $ t_{i}-t_{i-1}<¥eta$ for $i=1$ , $¥cdots$ , $n$ . Let $T_{i}=$

$[t_{i-1}, t_{i}]¥backslash J_{¥mathrm{e}}$ . Because

$¥int_{T}g(t, s, V(s))ds¥subset¥sum_{i=1}^{n}¥int_{¥tau_{i}}g(t, s, V(s))ds$,

we get

(11) $¥alpha(¥int_{T}g(t, s, V(s))ds)¥leq¥sum_{i=1}^{n}¥alpha(¥int_{¥tau_{i}}g(t, s, V(s))ds)$ .

Set $V_{i}=¥{v(s):v¥in V, s¥in T_{i}¥}$ . For any $s¥in T_{i}$ there exists $d_{s}>0$ such that

(12) $¥alpha(g(t, I_{s,a_{s}}, V_{i}))¥leq h(t, s, ¥alpha(V_{i}))+¥epsilon$.

Since the family $¥{I_{s,l_{S}}( : s¥in T_{i}¥}$ covers $T_{i}$ , there is a finite cover $I_{s_{1},d_{s_{1}}}$ , $¥cdots$ , $I_{s_{m},a_{s_{m}}}$ of$T_{i}$ . Choose sets $P_{1}$ , $¥cdots$ , $P_{m}$ such that $P_{k}¥subset I_{s_{k},a_{Sk}}$ , $¥bigcup_{k=1}^{m}P_{h}=T_{i}$ , $ P_{j}¥cap P_{k}=¥emptyset$ for$j¥neq k$ . Then

$¥int_{¥tau_{i}}g(t, s, V(s))ds¥subset¥sum_{k=1}^{m}¥int_{P_{h}}g(t, s, V(s))ds¥subset¥sum_{h=1}^{m}¥mu(P_{k})¥overline{¥mathrm{c}¥mathrm{o}¥mathrm{n}¥mathrm{v}}g(t, P_{k}, V_{i})$ .

By (12) and the corresponding properties of $¥alpha$ from this it follows that

On Volterra Integral Equations in Banach Spaces 255

$¥alpha(¥int_{¥tau_{i}}g(t, s, V(s))ds)¥leq¥sum_{h=1}^{m}¥mu(P_{k})¥alpha(g(t, P_{k}, V_{i}))$

(13)$¥leq¥sum_{k=1}^{m}¥mu(P_{k})h(t, s_{k}, ¥alpha¥iota^{¥prime}V_{i}))+¥epsilon¥mu(T_{i})$ .

By Ambrosetti’ $¥mathrm{s}$ lemma [1; Th. 2.2] and the continuity of $¥alpha(V(¥cdot))$ there is $q_{i}¥in T_{i}$

such that $¥alpha(V_{i})=¥sup$ $¥{¥alpha(V(s)):s¥in T_{i}¥}=¥alpha(V(q_{i}))$ . Therefore $|¥alpha(V_{i})-¥alpha(V(s_{k}))|=$

$|¥alpha(V(q_{i}))-¥alpha(V(s_{k}))|<¥delta$ , because $|q_{i}-s_{k}|<¥eta$ . Consequently,

$|h(t, s_{h}, ¥alpha(V_{i}))-h(t, s_{k}, ¥alpha(V(s_{k})))|<¥epsilon$ ,

so that

(14) $¥sum_{k=1}^{m}¥mu(P_{h})h(t, s_{k}, ¥alpha(V_{i}))¥leq¥sum_{k=1}^{m}¥mu(P_{k})h(t, s_{k}, ¥alpha(V(s_{h})))+¥epsilon¥mu(T_{i})$ .

Since $s^{¥prime}$ , $s^{¥prime¥prime}¥in T_{i}$ implies $|h(t, s^{¥prime}, ¥alpha(V(s^{¥prime})))-h(t, s^{¥prime¥prime}, ¥alpha(V(s^{¥prime¥prime})))|<_{¥epsilon}$ , we have

(15) $¥mu(P_{k})h(t, s_{lc}, ¥alpha(V(s_{k})))¥leq¥int_{P_{k}}h(t, s, ¥alpha(V(s)))ds+¥epsilon¥mu(P_{k})$ .

Now from (13)?(15) it follows that

$¥alpha(¥int_{¥tau_{i}}g(t, s, V(s))ds)¥leq¥int_{¥tau_{i}}h(t, s, ¥alpha(V(s)))ds+3¥epsilon¥mu(T_{i})$ for $i=1$ , $¥cdots$ , $n$ ,

and therefore, by (11),

$¥alpha(¥int_{T}g(t, s, V(s))ds)¥leq¥sum_{i=1}^{n}[¥int_{¥tau_{i}}h(t, s, ¥alpha(V(s)))ds+3¥epsilon¥mu(T_{i})]$

(16)$=¥int_{T}h(t, s, ¥alpha(V(s)))ds+3¥epsilon¥mu(T)$ .

On the other hand, for any $v¥in V$ we have

$||¥int_{J_{8}}g(t, s, v(s))ds||¥leq¥int_{J_{¥mathrm{e}}}m(t, s)ds<¥epsilon$ ,

which implies

$¥alpha(¥int_{J_{¥epsilon}}g(t, s, V(s))ds)¥leq 2¥epsilon$ .

Finally, by (10) and (16), we obtain

$¥alpha(F(V)(t))¥leq¥int_{T}h(t, s, ¥alpha(V(s)))ds+2¥epsilon+3¥epsilon¥mu(T)$

$¥leq¥int_{0}^{t}h(t, s, ¥alpha(V(s)))ds+2¥epsilon+3¥epsilon¥mu(T)$.

256 S. SZUFLA

Since the last inequality is satisfied for every $¥epsilon>0$ , we get

$¥alpha(F(V)(t))¥leq¥int_{0}^{t}h(t, s, ¥alpha(V(s)))ds$ .

Further, from (8) it follows that $¥alpha((I-F)(V))=0$ . Consequently, by (9) andAmbrosetti’ $¥mathrm{s}$ lemma [1; Th. 2. 3], we get

$¥alpha(V(t))¥leq¥alpha((I-F)(V)(t))+¥alpha(F(V)(t))=¥alpha(F(V)(t))$ ,

so that

$¥alpha(V(t))¥leq¥int_{0}^{t}h(t, s, ¥alpha(V(s)))ds$ for $t¥in J$ .

As $h$ is a Kamke function, this implies

$¥alpha(V(t))=0$ for $t¥in J$ .

Therefore by Ambrosetti’ $¥mathrm{s}$ lemma

$¥alpha(V)=¥sup$ $¥{¥alpha(V(t)):t¥in J¥}=0$ ,

$¥mathrm{i}.¥mathrm{e}.¥overline{V}$ is compact in $C$ . Hence we can find a subsequence $(u_{n_{k}})$ of $(u_{n})$ whichconverges in $C$ to a limit $u¥in C$ . From (8) it follows that

$||u-F(u)||_{c}=¥lim_{k-¥infty}||u_{n_{k}}-F(u_{n_{k}})||_{c}=0$ .

This shows that $u=F(u)$ , i.e. $u$ is a solution of (1).

3. In this section we present a simple theorem on the connectedness of theset $T^{-1}(y)$ , where $T$ is a continuous function from a metric space $X$ into a metricspace $¥mathrm{Y}$ , and $y$ is a fixed element of Y. Put $S_{¥mathrm{e}}=¥{x¥in X:d(y, T(x))<_{¥epsilon}¥}$ for any$¥epsilon>0$ , where $d$ denotes the metric of Y.

Theorem 4. Assume that $T:X¥rightarrow ¥mathrm{Y}$ is a continuous function such that1o for any $¥epsilon>0$ the set $S_{¥mathrm{e}}$ is connected;2o $y¥in¥overline{T(V)}¥Rightarrow y¥in T(V)$ for every closed subset $V$ of X. Then the set $T^{-1}(y)$ is

connected.

Proof. Suppose that the set $S=T^{-1}(y)$ is disconnected. Thus there are non-empty closed sets $W_{1}$ , $W_{2}$ such that $S=W_{1}¥cup W_{2}$ and $ W_{1}¥cap W_{2}=¥emptyset$ , and, consequent-ly, there are two disjoint open sets $U_{1}$ , $U_{2}$ of $X$ such that $W_{1}¥subset U_{1}$ , $W_{2}¥subset U_{2}$ . Sup-pose that for every $n¥in N=¥{1,2, --¥}$ there exists an $x_{n}¥in V_{n}¥backslash U$ , where $V_{n}=¥overline{S}_{1/n}$ and$U=U_{1}¥cup U_{2}$ . From 1o it follows that $V_{n}$ is connected for every $n¥in N$ . Put $V=$

$¥{¥overline{x_{n}..n¥in N}¥}$ . Since $d(y, T(x_{n}))¥leq 1/n$ , we see that $T(x_{n})¥rightarrow y$ , i.e. $y¥in¥overline{T(V)}$ . By 2o

On Volterra Integral Equations in Banach Spaces 257

this implies that there exists $x_{0}¥in V$ such that $T(x_{0})=y$ . Furthermore $V¥subset X¥backslash U$ ,since $U$ is open in $X$ , so that $x_{0}¥in S¥backslash U$ , a contradiction. Therefore there is $m¥in N$

such that $V_{m}¥subset U$ . Because $S¥subset V_{m}$ , we see that $U_{1}¥cap V_{m}¥neq¥emptyset¥neq U_{2}¥cap V_{m}$, and hence$V_{m}$ is disconnected, in contradiction with the connectedness of $V_{n}$ for every $n¥in N$ .This proves that $S$ is connected.

Remark. Let $B$ be a Banach space, and let $F$ be a continuous mapping $B¥rightarrow B$ .Denote by I the identity mapping on $B$ . Then the fixed points set of $F$ is connect-ed whenever the mapping $T=I-F$ satisfies the assumptions 1, $2^{¥mathrm{o}}$ of Theorem 4.In particular, the solutions set of (1) is connected whenever the mapping $T=I-F$satisfies the assumption 2o of Theorem 4, where $F$ denotes the mapping defined inthe proof of Theorem 2. In this way we obtain a simple proof of Kneser’ $¥mathrm{s}$ theoremfor a certain class of Volterra integral equations in Banach spaces.

Corollary. Under the assumptions of Theorem 3 the solutions set of (1) is $a$

continuum $¥dot{¥tau}nG$ .

Acknowledgement. The author expresses his appreciation to the referee forhis suggestions to amend the earlier version of the paper.

References

[1] Ambrosetti, A., Un teorema di esistenza per le equazioni differenziali negli spazi diBanach, Rend. Sem. Mat. Univ. Padova, 39 (1967), 349-360.

[2] Hukuhara, M., Sur l’application qui fait correspondre a un point un continu bicompact,Proc. Japan Acad., 31-1 (1955), 5-7.

[3] Kelley, W. G., A Kneser theorem for Volterra integral equations, Proc. Amer. Math.Soc., 40-1 (1973).

[4] Kuratowski, K., Topologie, Warszawa, 1958.[5] Murakami, H. and Nakagiri, S., Approximate solutions for some non-linear Volterra

integral equations, Proc. Japan Acad., 50-3 (1974), 212-217.[6] Murakami, H. and Nakagiri, S., Kneser’s property of solution families of non-linear

Volterra integral equations, ibid., 50-4 (1974), 296-300.[7] Pianigiani, G., Existence of solutions for ordinary differential equations in Banach

spaces, Bull. Acad. Polon. Sci., Ser. Sci. Math. Astronom. Phys., 23-8 (1975), 853-857.[8] Sato, T., Sur l’equation integrale non lineaire de Volterra, Compositio Math., 11-3

(1953), 271-290.[9] Sato, T., Sur l’application qui fait correspondre a une corbe une famille des courbes,

Proc. Japan Acad., 31-1 (1955), 1-4.[10] Scorza-Dragoni, G., Un teorema sulle funzioni continue rispetto a una e misurabili

rispetto all’altra variable, Rend. Sem. Mat. Univ. Padova, 17 (1948), 102-108.[11] Vidossich, G., On the structure of solutions set of nonlinear equations, J. Math. Anal.

Appl., 34 (1971), 602-617.[12] Vidossich, G., A fixed point theorem for function spaces, ibid., 36 (1971), 581-587.

S. SZUFLA

nuna adreso:Os. Poustan Narodowych 5961216 PoznanPoland

(Ricevita la 8-an de decembro, 1976)(Reviziita la 15-an de februaro, 1977)