See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/251047278

Kamke's Theorem and the Dependence of Solutions for Delay Differential

Equations

Article in Tokyo Journal of Mathematics · June 1983

DOI: 10.3836/tjm/1270214329

CITATION READS

1 2,863

1 author:

Jong Son Shin

Hosei University
62 PUBLICATIONS 721 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jong Son Shin on 04 July 2014.

The user has requested enhancement of the downloaded file.

 TOKYO J. MATH.
VOL. 6, No. 1, 1983

Kamke’s Theorem and the Dependence of Solutions

for Delay Differential Equations

Jong-Son SHIN

Korea University
(Communicated by T. Saito)

Introduction
We consider the delay differential equations including state dependent
lags,
(E’) $x_{i}(t)=f_{i}(t, x(t),$ $x(g_{2}(t, x(t))),$ $\cdots,$ $x(g_{m}(t, x(t))))$ ,
$i=1,$ , under the assumption that
$\cdots,$
$d$
satisfies the Caratheodory con- $f_{i}$

dition, where $x(t)$ stands for $(x_{1}(t), \cdots, x_{d}(t))$ and prime denotes differentia-
tion with respect to . We assume the existence of a finite number
$t$
$\alpha$

such that for each $j=2,$ $m,$ , whenever $g_{j}(t, x)$ is defined;
$\cdots,$ $\alpha\leqq g_{j}(t, x)\leqq t$

the delays $t-g_{j}(t, x)$ may be unbounded. This type of system arises in
studying a two-body problem of classical electrodynamics $[6, 7]$ . Driver
[4] developed the basic theory (existence, uniqueness and dependence of
solutions, etc.) for the initial value problem for delay differential
equations (E) with continuous $f_{i}[5]$ . Since then the theory of delay
differential equations (E) has been studied by many authors. Among
them Bullock [1] showed the existence theorem and uniqueness theorem
for delay differential equations (E’) of Caratheodory type. On the other
hand, Strauss and Yorke [13] constructed a fundamental theory for
ordinary differential equations by using the convergence theorem which
is a generalization of Kamke’s theorem (see [8], Theorem 3.2). Their
method proves to be very important in studying the fundamental theory
of functional (or delay) differential equations. Costello [3] extended their
results to functional differential equations of Caratheodory type with
finite delay,
$(E^{\prime})$
$x^{\prime}(t)=F(t, x_{t})$ ,
where $x_{t}(\theta)=x(t+\theta)$ , $-r\leqq\theta\leqq 0$
. Rybakowski [14] also states without
Received August 10, 1981
110 JONG-SON SHIN

proof Kamke’s theorem for the system $(E$ “ under the very mild $)$

condition of Caratheodory type. The object of this paper is to give

an extension of these results [13] to delay differential equations (E’)
together with the convergence theorem, the existence of noncontinu-
able solutions and the dependence of solutions on initial conditions and
on the functions . We emphasize that the equations (E’) cannot be
$f_{i}$

represented by the above form, since the delays $t-g_{j}(t, x)$ may be
unbounded. We introduce a modification of uniform convergence for the
sequence of continuous functions , where and $\phi_{n};[a_{n}, b_{n}]\rightarrow R^{d}$ $\{a_{n}\}$ $\{b_{n}\}$

converge. This modification plays an essential role in proving the

convergence theorem, which is very advantageous when we consider the
dependence of solutions on initial conditions and on the right-hand side
of (E’).
In section 1, we explain explicit discription of the notations and give
the initial value problem for delay differential equations (E’). In section
2, we introduce a modification of uniform convergence mentioned above,
and by using its properties, the convergence theorem is established,
which is one of our main theorem (Theorem 2.1) in the paper. In
section 3, by using a nonstandard argument as pointed out in [13], the
existence of noncontinuable solutions and the extension of solutions are
given. In section 4, from the convergence theorem, the dependence of
solutions is obtained.

\S 1. Definition and Notations.

Let $R$ (or ) denote the set of real numbers and
$R^{1}$ $R^{d}$
denote the
d-dimensional real Euclidean space. Let
$(t, X)=(t, x_{11}, \cdots, x_{1d}, x_{21}, \cdots, x_{2d}, \cdots, x_{nd})$

be a vector in $R^{dm+1}$ and $(t, x)=(t, x_{11}, \cdots, x_{1d})$ be the $d+1$ -dimensional
vector consisting of the first $d+1$ coordinates of $(t, X)$ . Let $D$ be a
domain (an open connected set) of and the set of vectors $(t, x)$
$R^{d_{1*}+1}$ $D^{*}$

in $R^{d+1}$
such that $(t, x, y)$ lies in $D$ for some in . In system (E’), $y$
$R^{d(f\hslash-1)}$

the function $f(t, X)$ is a mapping from into $D$ and the function $g(t, x)=$ $R^{d}$

$(g_{1}(t, x),$ $g_{n}(t, x))$ is a mapping from

$\cdots,$ into , where for
$D^{*}$ $R^{m}$ $g_{1}(t, x)\equiv t$

$(t, x)\in D^{*}$ . For a continuous function $x(t)$ mapping an interval into , $I$ $R^{\delta}$

we define a function by $xg(t, \xi)=(x(g_{1}(t, \xi)),$

$xg:D^{*}\rightarrow R^{dn}f$ $x(g_{n}(t, \xi)))$ $\cdots,$

for $(t, \xi)\in D^{*}$ , whenever every composition $x(g_{j}(t, \xi)),$ $j=1,$ $m$ , has a $\cdots,$

meaning. Using this notation, we can rewrite the system (E) as

(E) $x’(t)=f(t, xg(t, x(t)))$ .
 DELAY DIFFERENTIAL EQUATIONS 111

We impose the following conditions on the functions and ; $f$ $g$

(F-1) $f(t, X)$ is continuous in $X$ for each fixed . $t$

(F-2) $f(t, X)$ is Lebesgue measurable in for each fixed $X$. $t$

(F-3) For each compact set in $D,$ $f(t, X)$ is bounded on . $Q$ $Q$

(G-1) $g(t, x)$ is continuous in $(t, x)\in D^{*}$ .

(G-2) for all $(t, x)\in D^{*}$ .
$g_{1}(t, x)\equiv t$

(G-3) there exists an such that $j=2,$ $m$ , for $\alpha\in R$ $\alpha\leqq g_{j}(t, x)\leqq t,$ $\cdots,$

all $(t, x)eD^{*}$ .

Let be the class of functions $f$, defined on $D$ in $R^{dm+1}$ , which
$M_{D}$

satisfy the conditions (F-1), (F-2) and (F-3) above.

A real d-dimensional vector function $\phi(t)=(\phi_{1}(t), \cdots, \phi_{d}(t))$ defined on
, is said to be an initial function if it is continuous and the
$[\alpha, \sigma_{0}],$ $\alpha\leqq\sigma_{0}$

point lies in $D$ . For such a function , an initial value

$(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))$
$\phi$

problem for the delay differential equations (E) is to find an -valued $R^{d}$

function $x(t)=(x_{1}(t), \cdots, x_{d}(t))$ defined for , where , which $\alpha\leqq t<\beta$ $\sigma_{0}<\beta\leqq\infty$

satisfies the following conditions;

(I-1) $x(t)=\phi(t)$ for . $\alpha\leqq t\leqq\sigma_{0}$

(I-2) $x(t)$ is locally absolutely continuous for . $\sigma_{0}\leqq t<\beta$

(I-3) $(t, xg(t, x(t)))\in D$ for $\sigma_{0}\leqq t<\beta$

(I-4) $x^{\prime}(t)=f(t, xg(t, x(t)))$ almost everywhere on ). $[\sigma_{0}, \beta$

Such a function $x(t)$ (or $x(t,$ ) is called a solution of the system (E) $\sigma_{0},$
$\phi)$

for the initial function on or simply a solution of (E) through $\phi$

$[\alpha, \sigma_{0}]$

. Clearly, a function $x(t)$ defined on

$(\sigma_{0}, \phi)$ ) is a solution of (E) $[\alpha, \beta$

through if and only if it satisfies

$(\sigma_{0}, \phi)$

$x(t)=\left\{\begin{array}{ll}\phi(t) & for t\in[\alpha, \sigma_{0}]\\\phi(\sigma_{0})+\int_{\sigma_{0}}^{t}f(s, xg(s, x(s)))ds & for t\in[\sigma_{0}, \beta).\end{array}\right.$

a convenience’s sake, we introduce some notations. For

Finally, for
a continuous function mapping $[a, b]$ into , we set $\phi$
$R^{d}$ $|\phi|_{[a,b]}=$

$sup\{|\phi(s)|:s\in[a, b]\}$ , where . is an arbitrary norm in . For a $|$

$R^{d}$
$\underline{s}etQ$

in , int $Q$ denote the interior of $Q,$

$R^{k+1}$ $\partial Q$
the boundary of $Q,$ $Q$ the
closure of $Q$ and $prQ=$ { $t:(t,$ $x)\in Q$ for some $x\in R^{k}$
}.

\S 2. Convergence theorem.
The result in this section is the main tool to develop the fundamental
theory on the initial value problem of (E) in the following sections.
We say that a sequence of compact intervals converges to $\{[a_{n}, b_{n}]\}$

a compact interval $[a, b]$ (or ) if and as . $[a_{n},$ $b_{n}]\rightarrow[a,$ $b]$ $a_{n}\rightarrow a$
$b_{n}\rightarrow b$ $ n\rightarrow\infty$

The function mapping $[a, b]$ onto is defined as

$\tau_{n}$
$[a_{n}, b_{n}]$
112 JONG-SON SHIN

$\tau_{n}t=\tau_{n}(b_{n}, a_{n}, b, a)t=\frac{b_{n}-a_{n}}{b-a}t+\frac{a_{n}b-ab_{n}}{b-a}$ for all $t\in[a, b]$ ,

if .
To emphasize the domain of a function, we denote by $\phi|[a, b]$
$a<b$
a function defined on $[a, b]$ . Also, this symbol means the restriction
$\phi$

of on $[a, b]$ , where is defined on a domain containing $[a, b]$ . First,

$\phi$ $\phi$

we introduce a modification of uniform convergence.

DEFINITION 1. A sequence of (continuous) functions, , is $\{\phi_{n}|[a_{n}, b_{n}]\}$

said to converge uniformly to a function $\phi|[a, b]$ if the following condi-

tions are satisfied;
(1) as ,
$[a_{n}, b_{n}]\rightarrow[a, b]$ $ n\rightarrow\infty$

(2) in case $a<b,$ as , where , $|\tilde{\phi}_{n}-\phi|_{[a,b]}\rightarrow 0$ $ n\rightarrow\infty$ $\tilde{\phi}_{n}(t)=\phi_{n}(\tau_{n}(b_{n},$

$a_{n}$

$b,$) ) for $te[a, b]$ ; in case $a=b$ ,

$a$ $t$

$\sup_{a\leq t\leq b}|\phi_{n}(t)-\phi(a)|\rightarrow 0$ as $ n\rightarrow\infty$

.
It is obvious from the definition that, if a sequence of continuous
functions converges uniformly to a function $\phi|[a, b]$ , then the
$\{\phi_{n}|[a_{n}, b_{n}]\}$

function is continuous on $[a, b]$ . In the ca8e $a_{n}=a,$ $b_{n}=b$ for all ,
$\phi(t)$
$n$

the uniform convergence of the sequence coincides with the $\{\phi_{n}|[a_{n}, b_{n}]\}$

usual uniform convergence.

EXAMPLE 1. If the function $\phi_{n}(t),$
$n=1,2,$ $\cdots$ , is given by

$\phi_{n}(t)=\left\{\begin{array}{ll}-nt & for te[-1/n, 0]\\0 & for te[0,1]\\nt- n & for t\in[1,1+1/n],\end{array}\right.$

then converges uniformly to

$\{\phi_{n}|[0,1]\}$
$0|[0,1]$ , but $\{\phi_{n}|[-1/n, 1+1/n]\}$
does not converge uniformly.
EXAMPLJ 2. If the function $\psi_{n}(t),$
$n=1,2,$ $\cdots$ , is given by

$\psi_{n}(t)=\left\{\begin{array}{ll}-nt+1 & for t\in[0,1/n]\\0 & for t\in[1/n, 1-1/n]\\nt-n+1 & for te[1-1/n, 1],\end{array}\right.$

then converges uniformly to

$\{\psi_{n}|[1/n, 1-1/n]\}$ $0|[0,1]$ , but $\{\psi_{n}|[0,1]\}$ does
not converge uniformly.
DEFINITION 2. A sequence of continuous functions, , $\{\phi_{n}|[a_{n}, b_{n}]\}$

$n=1,2,$ , is said to be uniformly bounded if there is a constant $M\geqq 0$

$\cdots$

such that for all . The sequence is said to be equicon-

$|\phi_{n}|_{[a_{\hslash},b_{\hslash}]}\leqq M$ $n$

tinuous if for any , there exists a such that $|t-s|<\delta,$ $\epsilon>0$ $\delta>0$ $t,$ $ s\in$

, implies
$[a_{n}, b_{n}]$
for all . $|\phi_{n}(t)-\phi_{n}(s)|<\epsilon$ $n$
 DELAY DIFFERENTIAL EQUATIONS 113
The following lemma is immediately obtained from Definitions 1, 2.
LEMAA 2.1. Suppose as , where $a<b$ . Then $[a_{n}, b_{n}]\rightarrow[a, b]$ $ n\rightarrow\infty$ $a_{n}<b_{n},$

the sequence is equicontinuous if and only if the sequence

$\{\phi_{n}|[a_{n}, b_{n}]\}$ $\{\tilde{\phi}_{n}\}$

is equicontinuous on $[a, b]$ , where . $\tilde{\phi_{n}}(t)=\phi_{n}(\tau_{n}t),$

$\tau_{n}t=\tau_{n}(b_{n}, a_{n}, b, a)t$

If we set
PROOF. , then . Since as$\sigma_{n}=\tau_{n}^{-1}$ $\phi_{n}=\tilde{\phi}_{n}\circ\sigma_{n}$
$[a_{n}, b_{n}]\rightarrow[a, b]$

, all
$ n\rightarrow\infty$
and are Lipschitz continuous with a common Lipschitz
$\tau_{n}$ $\sigma_{n}$

constant, that is, $|\tau_{n}t-\tau_{n}s|\leqq M|t-s|$ for $s\in[a, b]$ and $|\sigma_{n}t-\sigma_{n}s|\leqq M|t-s|$ $t,$

for $s\in[a_{n}, b_{n}]$ , where $M$ is a positive constant. Therefore the relations
$t,$

and imply the equivalence between the equicontinuous

$\tilde{\phi}_{n}=\phi_{n}\circ\tau_{n}$ $\phi_{n}=\tilde{\phi}_{n}\circ\sigma_{n}$

of and {
$\{\phi_{n}|[a_{n}, b_{n}]\}$
This completes the proof. $\tilde{\phi}_{n}|[a,$ $ bIf\cdot$

Combining Definition 1, Lemma 2.1 and Ascoli-Arzela’s theorem, we

obtain immediately the following lemma.
LEMMA 2.2. Suppose that the function $n=1,2,$ , is $\phi_{n}|[a_{n}, b_{n}],$ $\cdots$

continuous, where as . Then every subsequence of $[a_{n}, b_{n}]\rightarrow[a, b]$ $ n\rightarrow\infty$

contains a subsequence which converges uniformly if and

$\{\phi_{n}|[a_{n}, b_{n}]\}$

only if $\{\phi_{n}|[a_{n}, b_{n}]\}$

is uniformly bounded and equicontinuous.
LEMMA 2.3. Suppose that the function $n=1,2,$ , defined on $\phi_{n}(t),$ $\cdots$

is continuous and
$[a_{n}, b_{n}]$
and as , where $a_{n}\rightarrow a,$ $b_{n}\rightarrow b$
$c_{n}\rightarrow c$ $ n\rightarrow\infty$ $ a_{n}\leqq$

. Then
$c_{n}\leqq b_{n}$
converges uniformly to $\phi|[a, b]$ if and only if
$\{\phi_{n}|[a_{n}, b_{n}]\}$

and
$\{\phi_{n}|[a_{n}, c_{n}]\}$
converge uniformly to $\phi|[a, c]$ and $\phi|[c, b]$ ,
$\{\phi_{n}|[c_{n}, b_{n}]\}$

respectively. In particular, if converges uniformly to $\phi|[a, b]$ , $\{\phi_{n}|[a_{n}, b_{n}]\}$

then converges to .
$\phi_{n}(c_{n})$ $\phi(c)$

PROOF. We prove only in the case $a<c<b$ . Set $\tau_{n}t=\tau_{n}(b_{n}, a_{n}, b, a)t$ ,
$\tau_{n}^{1}t=\tau_{n}(c_{n}, a_{n}, c, a)t$ and . Then we have $\tau_{n}^{2}t=\tau_{n}(b_{n}, c_{n}, b, c)t$

(2.1) $\tilde{\phi}_{n}(t)-\phi(t)=\left\{\begin{array}{ll}\phi_{n}(\tau_{n}t)-\phi_{n}(\tau_{n}^{1}t)+\phi_{n}(\tau_{n}^{1}t)-\phi(t) & for t\in[a, c]\\\phi_{n}(\tau_{n}t)-\phi_{n}(\tau_{n}^{2}t)+\phi_{n}(\tau_{n}^{2}t)-\phi(t) & for t\in[c, b] ,\end{array}\right.$

where Suppose that converges uniformly to

$\tilde{\phi}_{n}(t)=\phi_{n}(\tau_{n}t)$
. $\{\phi_{n}|[a_{n}, b_{n}]\}$

$\phi|[a, b]$ . Then Lemma 2.2 says that the sequence is equi- $\{\phi_{n}|[a_{n}, b_{n}]\}$

continuous. Therefore, since as , we $|\tau_{n}-\tau_{n}^{1}|_{[a,c]},$ $|\tau_{n}-\tau_{n}^{2}|_{[c,b]}\rightarrow 0$ $ n\rightarrow\infty$

have as . Relation (2.1)

$|\phi_{n}\circ\tau_{n}-\phi_{n}\circ\tau_{n}^{1}|_{[a,c]},$ $|\phi_{n}\circ\tau_{n}-\phi_{n}\circ\tau_{n}^{2}|_{[c,b]}\rightarrow 0$ $ n\rightarrow\infty$

implies as . This shows the “only

$|\phi_{n}\circ\tau_{n}^{1}-\phi|_{[a,c]},$ $|\phi_{n}\circ\tau_{n}^{2}-\phi|_{[c,b]}\rightarrow 0$ $ n\rightarrow\infty$

if“ part of the lemma, and vice versa.

The following lemma is important to prove the main theorem.
LEMMA 2.4. Let $f\in M_{Q}$ , where is a compact subset of D. Suppose $Q$

that a continuous function $n=1,2,$ , defined on satisfies $x_{n}(t),$ $\cdots$ $[\alpha, \beta_{n}]$
114 JONG-SON SHIN

$x_{n}(t)=x_{n}(\sigma_{n})+\int_{\sigma_{\hslash}}^{t}f(s, x_{n}g(s, x_{n}(s)))ds+G_{n}(t)$

for $t\in[\sigma_{n}, \beta_{n}]$ ,

where $\alpha\leqq\sigma_{n}<\beta_{n}$ , $\{[\sigma_{n}, \beta_{\hslash}]\}$

converges to $[\sigma_{0}, \beta]$
, $\sigma_{0}<\beta$ , and that
$(t, x_{n}g(t, x_{n}(t)))\in Q$
for all $t\in[\sigma_{n}, \beta_{n}]$
. If and $\{x.|[a, \sigma.]\}$ $\{|G_{n}||[\sigma_{n}, \beta_{n}]\}$

converge uniformly to $\phi|[\alpha, \sigma_{0}]$

and , respectively, then there
$0|[\sigma_{0}, \beta]$

exists a subsequence of $\{x_{n}|[\alpha, \beta_{n}]\}$

which converges uniformly to a solution
$x|[\alpha, \beta]$
of (E) through $(\sigma_{0}, \phi)$
.
PROOF. We set
$\psi_{n}(t)=\int_{\sigma_{t}}^{t}f(s, x_{n}g(s, x_{n}(t)))ds$ for $t\in[\sigma_{n}, \beta_{n}]$
.
Since the function is bounded on the compact set $Q$ , the sequence
$f$

is uniformly bounded and equicontinuous. Hence, from

$\{\psi_{n}|[\sigma_{n}, \beta_{n}]\}$

Lemma 2.2, by taking a subsequence if necessary, we may assume that

converges uniformly. Lemma 2.3 implies
$\{\psi_{n}|[\sigma_{n}, \beta_{n}]\}$ as $x_{n}(\sigma_{n})\rightarrow\phi(\sigma_{0})$

, while
$ n\rightarrow\infty$ converges uniformly to
$\{|G_{n}||[\alpha_{n}, \beta_{n}]\}$ . Since $0|[\sigma_{0}, \beta]$

$x_{n}(t)=x_{n}(\sigma_{n})+\psi_{n}(t)+G_{n}(t)$ for the sequence converges$\sigma_{n}\leqq t\leqq\beta_{n}$ $\{x_{n}|[\sigma_{n}, \beta_{n}]\}$

uniformly to a function . From Lemma 2.3, it follows that $x|[\sigma_{0}, \beta]$

converges to
$\{x_{n}|[\alpha, \beta_{n}]\}$ , where . $x|[\alpha, \beta]$ $x|[\alpha, \sigma_{0}]=\phi|[\alpha, \sigma_{0}]$

for any $(1/3)(t-\sigma_{0})>\epsilon>0$ ,

Let be a fixed number $t$
in ]. Then $(\sigma_{0}, \beta$ $\epsilon,$

there exists an such that $n>N_{1}$ implies

$N_{1}$ and , $|\sigma_{n}-\sigma_{0}|<\epsilon$ $|\tau_{n}t-t|<\epsilon$

where . Thus for all $n>N_{1}$ , the function

$\tau_{n}t=\tau_{n}(\beta_{n}, \sigma_{n}, \beta, \sigma_{0})t$ is $x_{n}(t)$

defined on . For a fixed

$[\alpha, t-\epsilon]$ , it follows from the $s,$ $\sigma_{0}+\epsilon\leqq s\leqq t-\epsilon$

hypotheses (G-1) and (G-3) that and $\alpha\leqq g_{j}(s, x_{n}(s))\leqq s$ $g_{\dot{f}}(s, x_{n}(s))\rightarrow g_{j}(s, x(s))$

as . Hence from Lemma

$ n\rightarrow\infty$ 2.3 we have as $x_{n}g_{j}(s, x_{n}(s))\rightarrow xg_{j}(s, x(s))$

, that is, $x_{n}g(s, x_{n}(s))\rightarrow xg(s, x(s))$ as

$ n\rightarrow\infty$ . Using the hypothesis $ n\rightarrow\infty$

(F-1), we obtain $f(s, x_{n}g(s, x_{n}(s)))\rightarrow f(s, xg(s, x(s)))$ as . Moreover, $ n\rightarrow\infty$

by using the Lebesgue dominated convergence theorem, for any , $\epsilon_{1}>0$

we can find such that $n>N_{2}$ implies

$N_{2}$

$|\int_{\sigma_{0}+\epsilon}^{t-\epsilon}f(s, x_{n}g(s, x_{n}(s)))ds-\int_{\sigma_{0}+\epsilon}^{t-\epsilon}f(s, xg(s, x(s)))ds|<\epsilon_{1}$ .

If we set
$\psi(t)=\int_{\sigma_{0}}^{t}f(s, xg(s, x(s)))ds$ for $te[\sigma_{0}, \beta]$ ,

then, by dividing the intervals of integrals, for all $n>N$, $N=$

max $\{N_{1}, N_{2}\}$ , we have
$|\psi_{n}(\tau_{n}t)-\psi(t)|$

$\leqq|\int_{\sigma_{n}}^{\tau_{\#}t}f(s, x_{n}g(s, x_{n}(s)))ds-\int_{\sigma_{0}}^{t}f(s, xg(s, x(s)))ds|$

 DELAY DIFFERENTIAL EQUATIONS 115
$<6M\epsilon+\epsilon_{1}$ ,
where $M=\sup_{Q}|f|$ . Since is arbitrary, we have $t\in[\sigma_{0}, \beta]$
$x(t)=\phi(\sigma_{0})+\psi(t)$

for . Namely $x(t)$ is a solution of (E) through

$t\in[\sigma_{0}, \beta]$
$(\sigma_{0}, \phi)$
. This proves
the lemma.
PROPOSITION. Let , where $Q$ is a compact subset of domain
$f\in M_{Q}$
$D\subset R^{d+1}$
and int Q. Let
$(\sigma_{0}, x_{0})\in$
be defined on and let $x_{n}(t)$ $[a_{n}, b_{n}]$ $(\sigma_{n}, x_{n})\rightarrow$

$(\sigma_{0}, x_{0}),$
$a<b$ , as
$[a_{n}, b_{n}]\rightarrow[a, b],$
, where . Suppose a $ n\rightarrow\infty$ $\sigma_{n}\in[a_{n}, b_{n}]$

function $G_{n}(t)$
is defined by

$G_{n}(t)=x_{n}(t)-x_{n}-\int_{\sigma_{n}}^{t}f(s, x_{n}(s))ds$

for all and that $(t, x_{n}(t))\in Q$ for $t\in[a_{n}, b_{n}]$ . If

$t\in[a_{n}, b_{n}]$
is measur- $x_{n}(t)$

able on and if as
$[a_{n}, b_{n}]$
, then $\{x_{n}|[a_{n}, b_{n}]\}contain\epsilon$
$|G_{n}|_{[a_{n},b_{n}]}\rightarrow 0$ $ n\rightarrow\infty$

a subsequence which converges uniformly on $[a, b]$ to a solution of $x_{0}(t)$

(E) $x^{\prime}(t)=f(t, x(t))$ , $x(\sigma_{0})=x_{0}$

.
PROOF. Since is measurable on $x_{n}(t)$
$x_{n}-G_{n}|[a_{n}, b_{n}]$ is con- $[a_{n}, b_{n}],$

tinuous. Therefore $\{x_{n}-G_{n}|[a_{n}, b_{n}]\}$ is uniformly bounded and equi-

continuous. From Lemma 2.2, by taking a subsequence if necessary,
we may assume that $\{x_{n}-G_{n}|[a_{n}, b_{n}]\}$ converges uniformly to a function
defined on $[a, b]$ . From the same argument used in the proof of
$x_{0}(t)$

Lemma 2.4, it follows that is a solution of . $x_{0}(t)$ $(E_{0})$

This proposition generalizes Lemma 1 in [13]. In fact, if $[a_{n}, b_{n}]=$

$[a, b]$ for all , proposition coincides with Lemma 1 in [13].

$n$

DEFINITION 3. Let $D$ be a domain of $R^{dm+1}$ and let be an $g$ $R^{m}-$

valued function defined on , which satisfies (G-l, 2, 3). Let $x(t)$ be a $D^{*}$

continuous function mapping some interval into . Then $x(t)$ is said $I$ $R^{d}$

to be a noncontinuable function (with respect to $D$ and g) if

$(t, xg(t, x(t)))\in D$ for all $t\in I$ and if there exists a sequence in such $\{t_{n}\}$ $I$

that as and that for each compact subset of $D$ , there

$t_{n}\rightarrow\sup I$ $ n\rightarrow\infty$ $Q$

exists a number $N=N(Q)$ such that

$(t_{n}, xg(t_{n}, x(t_{n})))\in D-Q$ for all $n\geqq N$ .
Furthermore, if $x(t)$ is a solution of (E) and $D$ is the domain of $f$,
then it is called a noncontinuable solution of (E). If a noncontinuable
solution $x(t)$ of (E) through is defined on ), , then we $(\sigma_{0}, \phi)$ $[\sigma_{0}, \beta$ $\sigma_{0}<\beta$

sometimes write or ). $D.(\sigma_{0})=[\sigma_{0}, \beta)$ $ D_{x}=[\sigma_{0}, \beta$

Although Definition 3 above may look slightly different from the

116 JONG-SON SHIN

usual definition for noncontinuable solution (cf. [7, 9]), but as we shall
see later they are equivalent to each other (Theorem 3.2).
Let $x(t)$ be a continuous function mapping some
DEFINITION 4 [13].
(possibly unbounded) interval into . A sequence of continuous $I$ $R^{d}$

functions is said to converge compactly to $x(t)$ if for every compact

$\{x_{n}(t)\}$

subset of is defined on
$J$
for sufficiently large
$I,$ and
$x_{n}(t)$ $J$ $n$ $\{x_{n}|J\}$

converges uniformly to $x|J$.

The following is the main result of this section.
THEOREM 2.1. Let $f\in M_{D}$ and $x_{n}(t)$
be a noncontinuable function,
defined on ), which satisfies $[\alpha, \beta_{n}$

(E) $x_{n}(t)=x_{n}(\sigma_{n})+\int_{\sigma_{\iota}}^{t},f(s, x_{n}g(s, x_{n}(s)))ds+G_{n}(t)$

for $t\in[\sigma_{n}, \beta_{n}$ ),

where , and
$\alpha\leqq\sigma_{n}<\beta_{n},$ as
$\alpha\leqq\sigma_{0}$
. Suppose that the sequence
$\sigma_{n}\rightarrow\sigma_{0}$ $ n\rightarrow\infty$

converges uniformly to
$\{x_{n}|[\alpha, \sigma_{n}]\}$ , and that $\phi|[\alpha, \sigma_{0}],$ $(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))\in D$

for every compact subset of with $Q$ $D$ int $Q$ , there exists a $(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))\in$

sequence such that

$\{\omega_{n}(Q)\}$ on and as ,
$|G_{n}(t)|\leqq\omega_{n}(Q)$ $[\sigma_{n}, t_{n}^{*}]$ $\omega_{n}(Q)\rightarrow 0$ $ n\rightarrow\infty$

$t_{n}^{*}=t_{n}^{*}(Q)=\inf\{t>\sigma_{n};(t, x_{n}g(t, x_{n}(t)))\in\partial Q\}$ . Then there exists a non-

where
continuable solution $x(t)$ of (E) through and a subsequence of $(\sigma_{0}, \phi)$ $\{x_{n}(t)\}$

which converges compactly to $x(t)$ .

PROOF. Since $(\sigma_{n}, x_{n}g(\sigma_{n}, x_{n}(\sigma_{n})))\rightarrow(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))\in D$
as $ n\rightarrow\infty$ , we can
choose a sequence of compact sets such that int $\{Q_{k}\}$ $ Q_{k}\in$ $Q_{k+1},$ $D=\bigcup_{k=0}^{\infty}Q_{k}$

and for all . Put

$(\sigma_{n}, x_{n}g(\sigma_{n}, x_{n}(\sigma_{n})))\in intQ_{0}$ . Then there $n$ $t_{n}(k)=t_{n}^{*}(Q_{k})$

exists a subsequence such that converges as for

$\{x_{n_{i}}\}\subset\{x_{n}\}$ $\{t_{n_{i}}(k)\}$
$ i\rightarrow\infty$

all . In fact, since

$k$

(2.2) $(t_{n}(k), x_{n}g(t_{n}(k), x_{n}(t_{n}(k))))\in\partial Q_{k}$ for all , $n$

is contained in the compact set $prQ_{k}$ . Thus first we can choose

$\{t_{n}(k)\}_{n=1}^{\infty}$

a subsequence of which converges. Next we can also

$\{t_{n(i,0)}(0)\}_{i=1}^{\infty}$ $\{t_{n}(0)\}$

choose a subsequence of such that con-

$\{n(i, 1)\}_{i=1}^{\infty}$ $\{n(i, 0)\}_{i=1}^{\infty}$ $\{t_{n(i,1)}(1)\}_{l=1}^{\infty}$

verges. Continuing in this fashion, one obtains the required subsequence

by the well known diagonal method. We denote such a subsequence by
again. Furthermore, since $Q_{k-1}\subset intQ_{k}$ , we have $t_{n}(k-1)<t_{n}(k)$ for
$\{x_{n}\}$

all $n$ . If we set $W_{k}=\lim_{n\rightarrow\infty}t_{n}(k)$ , then we obtain

(2.3) $\sigma_{0}\leqq W_{0}\leqq W_{1}\leqq\cdots\leqq W_{k}\leqq\cdots$

Put $W=\lim_{k\rightarrow\infty}W_{k}$
.
Next, by taking a subsequence if necessary, we shall show that for
all , $\{x_{n}|[\alpha, t_{n}(k)]\}$ converges uniformly. Since
$k$ as , $t_{n}(k)\rightarrow W_{k}$ $ n\rightarrow\infty$
 DELAY DIFFERENTIAL EQUATIONS 117
Lemma 2.4 assures that there exists a subsequence of which $\{x_{n}|[\alpha, t_{n}(k)]\}$

converges uniformly. Thus, by the familiar diagonal method, we can

extract a convergent sequence. Now, let the limit of the sequence
be denoted by
$\{x_{n}|[\alpha, t_{n}(k)]\}$
. Since for all
$x^{k}|[\alpha, W_{k}]$ $\sigma_{0}\leqq t_{n}(k-1)<t_{n}(k)$

$n$, we have $x^{k-1}|[a, W_{k-1}]=x^{k}|[\alpha, W_{k-1}]$ by Lemma 2.3. From this fact,
there exists a unique function $x|[\alpha, W$ ) such that $x|[a, W_{k}]=x^{k}|[\alpha, W_{k}]$
for all . Consequently, for all
$k$
the sequence converges $k$ $\{x_{n}|[\alpha, t_{n}(h)]\}$

uniformly to .
$x|[\alpha, W_{k}]$

By using this fact and Lemma 2.3, we have $(t, xg(t, x(t)))\in Q_{k}$ for
all , and hence $(t, xg(t, x(t)))\in D$ for all
$t\in[\sigma_{0}, W_{k}]$
). Moreover, $t\in[\sigma_{0}, W$

since as
$x_{n}(t_{n}(k))\rightarrow x(W_{k})$ , it follows from the continuity of
$ n\rightarrow\infty$
that $g$

$g(t_{n}(k), x_{n}(t_{n}(k)))\rightarrow g(W_{k}, x(W_{k}))$ as

. However, from the hypothesis $ n\rightarrow\infty$

(G-3), we have . Thus by Lemma 2.3 again, we

$\alpha\leqq g(t_{n}(k), x_{n}(t_{n}(k)))\leqq t_{n}(k)$

have $x_{n}g(t_{n}(k), x_{n}(t_{n}(k)))\rightarrow xg(W_{k}, x(W_{k}))$ as . Therefore it follows $ n\rightarrow\infty$

from (2.2) that

(2.4) $(W_{k}, xg(W_{k}, x(W_{k})))\in\partial Q_{k}$ .
From (2.3) and $Q_{k-1}$
cint $Q_{k}$
, we have
(2.5) $\sigma_{0}\leqq W_{0}<W_{1}<\cdots<W_{k}<W_{k+1}<\cdots\rightarrow W$ .
Thus from (2.5) and as , it follows that
$t_{n}(k+1)\rightarrow W_{k+1}$ $ n\rightarrow\infty$ $[\alpha, W_{k}]\subset$

for safficiently large . By Lemma 2.3 again,

$[a, t_{n}(k+1)]$ $n$ $\{x_{n}|[\alpha, W_{k}]\}$

converges uniformly to $x|[\alpha, W_{k}]$ , and hence converges com- $\{x_{n}|[\alpha, \beta_{n})\}$

pactly to $x|[\alpha, W$ ).
Finally, we show that $x(t)$ is a noncontinuable solution of (E).
For
any compact set $Q$ in $D$ there exists an $N=N(Q)$ such that $Q$ is con-
tained in int for all $k\geqq N$. Then the relation (2.4) means that
$Q_{k}$

$(W_{k}, xg(W_{k}, x(W_{k})))\in D-Q$ for all $k\geqq N$. In view of Lemma 2.4 the
function $x(t)$ is a noncontinuable solution of (E). This completes the
proof.

REMARK. Under the same assumptions as in Theorem 2.1, if a sub-

sequence of converges compactly to noncontinuable solution
$\{x_{n_{i}}(t)\}$ $\{x_{n}(t)\}$

$x(t)$ of (E) through , then we have .

$(\sigma_{0}, \phi)$ $supD_{x}\leqq\lim\inf_{i\rightarrow\infty}$
$supD_{x_{n_{i}}}$

COROLLARY 2.1. Let $f^{n},$ $f\in M_{D},$ $n=1,2,$ $\cdots$

. Let a noncontinuable
function defined on $x_{n}(t)$ $[\alpha, \beta_{n}$
) satisfies
(E) $x_{n}(t)=x_{n}(\sigma_{n})+\int_{\sigma_{n}}^{t}f^{n}(s, x_{n}g(s, x_{n}(s)))ds$
for $t\in[\sigma_{n}, \beta_{n}$ )

where, $\alpha\leqq\sigma_{n}<\beta_{n},$ $\alpha\leqq\sigma_{0}$

and $\sigma_{n}\rightarrow\sigma_{0}$
as $ n\rightarrow\infty$
. If the sequence $\{x_{n}|[\alpha, \sigma_{n}]\}$
118 JONG-SON SHIN

converges uniformly to $\phi|[\alpha, \sigma_{0}],$ $(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))eD$ , and if

(2.6) $\lim_{n\rightarrow\infty}\int_{prQ}\sup_{X\in Q_{t}}|f^{n}(t, X)-f(t, X)|dt=0,$ $Q_{t}=\{X:(t, X)\in Q\}$ ,

for every compact set such that is in int $Q$ , then $Q\in D$ $(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))$

there exists a subsequence of which converges compactly to a non- $\{x_{n}(t)\}$

continuable solution of (E) through . $(\sigma_{0}, \phi)$

PROOF. If we set
$ G_{n}(t)=\int_{\sigma_{n}}^{t}|f^{n}(s, x_{n}g(s, x_{n}(s)))-f(s, x_{n}g(s, x_{n}(s)))|d\epsilon$ ,

then we can rewrite $(E_{2})$

as $(E_{1})$ . Given any such compact set $Q\subset D$ ,
put
$\omega_{n}(Q)=\int_{prQ}\sup_{XeQ_{t}}|f^{n}(t, X)-f(t, X)|dt$ .
Since $\omega_{n}(Q)\rightarrow 0$
as $ n\rightarrow\infty$ , from (2.6) and $G_{n}(t)\leqq\omega_{n}(Q)$ , we have the
conclusion.
COROLLARY 2.2. Let be continuous functions on a domain $D$ $f^{n},$ $f$

in . If we replace condition (2.6) in Corollary 2.1 by the condition

$R^{dn+1}$

that converges compactly to on $D$ , then the conclusion of Corollary

$f^{n}$ $f$

2.1 is true.
We note that recently, Kato [12] shows Kamke’s theorem in func-
tional differential equations with infinite delay on an abstract phase
space (cf. [10]).

\S 3. Existence of noncontinuable solutions.

In this section, by applying Theorem 2.1 we show that a noncon-
tinuable solution of (E) through always exists. $(\sigma_{0}, \phi)$

THEOREM 3.1. Suppose $f\in M_{D}$ . Then for every initial function
, there exists a noncontinuable solution of (E) through
$\phi|[\alpha, \sigma_{0}]$ $(\sigma_{0}, \phi)$
.
$PR\infty F$
. For each $n$ , define $x_{n}(t)$ as follows;

$x_{n}(t)=\left\{\begin{array}{ll}\phi(t) & for \alpha\leqq t\leqq\sigma_{0}\\\phi(\sigma_{0}) & for \sigma_{0}\leqq t\leqq\sigma_{0}+\frac{1}{n}\\\phi(\sigma_{0})+\int_{\sigma_{0}}^{t-1/n}f(s, x_{n}g(s, x_{n}(s)))ds & for \sigma_{0}+\frac{1}{n}\leqq t.\end{array}\right.$
 DELAY DIFFERENTIAL EQUATIONS 119
Note first that this formula is meaningful for large enough since $n$

functions are continuous and

$\phi,$ $g$
. In this recursive $(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))\in D$

way is defined on
$x_{n}$ $[\sigma_{0}+k/n, \sigma_{0}+(k+1)/n]$ for $k=0,1,$ , provided $\cdots$

$(s, x_{n}g(s, x_{n}(s)))$ belongs to $D$ on the previous

interval. Thus this process
can be continued either for all or until the first point at which $t$
$\beta_{n}$

. Hence each
$(\beta_{n}, x_{n}g(\beta_{n}, x_{n}(\beta_{n})))\in\partial D$
is continuous on ) and $x_{n}(t)$ $[\alpha, \beta_{n}$

noncontinuable with respect to $D$ and . Furthermore, satisfies $g$ $x_{n}(t)$

$x_{n}(t)=\left\{\begin{array}{ll}\phi(t) & for \alpha\leqq t\leqq\sigma_{0}\\\phi(\sigma_{0})+\int_{\sigma_{0}}^{t}f(s, x_{n}g(s, x_{n}(s)))ds+G_{n}(t) & for \sigma_{0}\leqq t<\beta_{n} ,\end{array}\right.$

where

$G_{n}(t)=\left\{\begin{array}{ll}-\int_{\sigma_{0}}^{t}f(s, x_{n}g(s, x_{n}(s)))ds & for \sigma_{0}\leqq t\leqq\sigma_{0}+\frac{1}{n}\\-\int_{t-1/n}^{t}f(s, x_{n}g(s, x_{n}(s)))ds & for \sigma_{0}+\frac{1}{n}\leqq t<\beta_{n}.\end{array}\right.$

For every compact set $Q$ in $D$ with $(\sigma_{0}, \phi g(\sigma_{0}, \phi(\sigma_{0})))\in intQ$ , we have
$|G_{n}(t)|\leqq(1/n)M_{Q}$ for , where $t\in[\sigma_{0}, t_{n}^{*}(Q)]$ $M_{Q}=\sup_{Q}|f|$ . Therefore we
obtain the conclusion from Theorem 2.1.
Bullock [1] stated without proof the local existence of solution for
(E) of Caratheodory type, but we obtain immediately the existence of
noncontinuable solutions of (E) without appealing to the local existence.
REMARK. Let $f\in M_{D},$ $D=(a, b)\times R^{dm}$ , and let $x(t)$ is a noncontinu-
able solution of (E). We note from Definition 3 that if $c=\sup D_{x}<b$ ,
then $\lim\sup_{t\rightarrow c-0}|x(t)|=+\infty$ . This conclusion does not necessarily imply
. Papers $[2, 11]$ pointed out that, in general, for
$\lim_{t\rightarrow c-0}|x(t)|=+\infty$

delay differential equations a noncontinuable solution $x(t)$ may not have

the property
$\lim_{t\rightarrow c-0}x(t)=\lim_{t\rightarrow c}\sup_{-0}x(t)=\infty$

In particular, Herdman [11] constructed a counter example.

THEOREM 3.2. be a domain of $R^{dm+1}$ and let $x(t)$ be a contin-
Let $D$

uous function mapping some interval I into . Then the following $R^{d}$

statements are equivalent:

(1) $x(t)$ is not a noncontinuable function with respect to $D$ and $g$ .
(2) Put $S=\{(t, xg(t, x(t))):t\in I\}\subset D$ . Then the closure $\overline{S}$

of $S$
is
compact and $\overline{S}\subset D$
.
120 JONG-SON SHIN

Moreover, and $x(t)$ is a solution of (E), then they are equivalent

if $f\in M_{D}$

to the following statement:

(3) $\lim_{t\rightarrow b-0}xg(t, x(t))=X_{0}$ and $(b, X_{0})\in D$ , where $b=\sup I$.

$PR\infty F$ . If (1) is satisfied, then is bounded, and hence is compact.

$S$ $\overline{S}$

Let now as , where $Y_{n}=xg(t_{n}, x(t_{n}))$ . Then

$(t_{n}, Y_{n})\in S\rightarrow(b, Y_{0})\in\partial\overline{S}$ $ n\rightarrow\infty$

it follows from (1) that the point $(b, Y_{0})$ belongs to $D$ , that is, . $\overline{S}\subset D$

One easily sees that (2) implies (1) and that (3) implies (2). Thus we
show that (2) implies (3). Since is bounded on the compact set from $\overline{S}$

$f$

(F-3), we have $|x(t)-x(s)|\leqq M(\overline{S})|t-s|$ for all $s\in I$, where . $t,$ $M(\overline{S})=\sup_{\overline{s}}|f|$

Hence, by the Cauchy convergence criterion exists, so that $\lim_{t\rightarrow b-0}x(t)$

$X_{0}=\lim_{t\rightarrow b-0}xg(t, x(t))$ exists and $(b, X_{0})$ is in $D$ . This completes the proof.
REMARK. It follows from Theorem 3.2 that Definition 3 is equivalent
to the standard definition for noncontinuable solutions (cf. [7, 9]).
THEOREM 3.3. (Extension). Let $f\in M_{D}$ . Then every solution of (E)
can be extended to a noncontinuable solution.
PROOF. Let $x(t)$ be not a noncontinuable solution
, defined on $[\sigma_{0}, b$
),
of (E) through . Then
it follows from Theorem 3.2 that
$(\sigma_{0}, \phi)$

$\lim_{t\rightarrow b-0}xg(t, x(t))=X_{0}$ and $(b, X_{0})\in D$ . Thus by Theorem 3.1 we obtain
that there is a noncontinuable solution of (E) through , where $(b,\tilde{x})$

$\tilde{x}(t)=x(t)$ for ) and

$t\in[\alpha, b$ . $\tilde{x}(b)=x_{0}$

THEOREM 3.4. Suppose $f\in M_{D}$ . Then of all intervals on which some
noncontinuable solution (E) through is defined, there is a smallest $(\sigma_{0}, \phi)$

interval on which such a noncontinuable solution is defined.

$PR\infty F$ . Let $NS$ denote the family of all the noncontinuable solutions
of (E) through $(\sigma_{0}, \phi)$
and set $J=\bigcap_{xeNS}D_{x}$ . If we set $\beta=\sup J$, then there
exists a sequence such that as
$\{\beta_{n}\}$ , where $\beta_{1}\geqq\beta_{2}\geqq\cdots\geqq\beta_{n}\geqq\cdots\rightarrow\beta$ $ n\rightarrow\infty$

is the right end point of a noncontinuable solution

$\beta_{n}$ , that is, $x_{n}(t)$ $D_{x_{\hslash}}=$

. By applying Theorem
$[\sigma_{0}, \beta_{n})$ 2.1, we find a subsequence of which $\{x_{n}\}$

converges compactly to a noncontinuable solution $x(t)$ of (E) through

. Then in view of Remark of Theorem 2.1, we have
$(\sigma_{0}, \phi)$ . From $D_{x}\subseteqq J$

the definition of we have , and hence $Dae=J$

$J$ . This implies that $J\subseteqq D_{x}$

a noncontinuable solution of (E) through defined on exists. This $(\sigma_{0}, \phi)$ $J$

completes the proof.

\S 4. Dependence of solutions.
In this section, we shall discuss the dependence of solutions on the
 DELAY DIFFERENTIAL EQUATIONS 121
initial condition and on the right hand sides of the delay differential
equations (E). In particular, the result on the dependence of solutions
due to Yoshizawa [15. Theorem 5.1] for ordinary differential equations
shall be extended to our system (E).
THEOREM 4.1. Let $f\in M_{D}$ and let , be the subinterval $I=[\sigma_{0}, b],$ $\sigma_{0}<b$

of , where
$J$
is the smallest interval given in Theorem 3.4. Then for
$J$

any , there exists a

$\epsilon>0$
for which the following condition holds: if $\delta>0$

in case
$|\sigma-\sigma_{0}|<\delta,$
( in case ),
$|\tilde{\phi}-\phi_{0}|_{\iota\alpha,\sigma_{0}}j<\delta$ $\alpha<\sigma_{0}$ $|\phi(\alpha)-\phi_{0}(\alpha)|<\delta$ $\alpha=\sigma_{0}$

where , and if $q(t),$ , is a bounded

$\tilde{\phi}=\phi(\tau(\sigma, \alpha, \sigma_{0}, \alpha)t)$ $t\in J_{1}=[\alpha, \sigma_{0}]\cup J$

and integrable function with , then every noncontinuable $\int_{J_{1}}|q(t)|dt<\delta$

solution $y(t, \sigma, \phi)$

of
(E) $y^{\prime}(t)=f(t, yg(t, y(t)))+q(t)$ ,
is always continued beyond $t=b$ and satisfies
(4.1) $|y(t, \sigma, \phi)-x(t, \sigma_{0}, \phi_{0})|<\epsilon$
for $t\in[\max\{\sigma, \sigma_{0}\}, b]$

for some noncontinuable solution $x(t, \sigma_{0}, \phi_{0})$

of (E) which may depend on
$y(t, \sigma, \phi)$ .
It is sufficient to prove only in the case
PROOF. . First, we $\alpha<\sigma_{0}$

show that there exists a such that, if $\delta_{0}>0$ $|\sigma-\sigma_{0}|<\delta_{0},$ $|\tilde{\phi}-\phi_{0}|_{[\alpha,\sigma_{0}]}<\delta_{0}$

and , then $supD_{y}>b$ for every noncontinuable solution

$\int_{J_{1}}|q(t)|dt<\delta_{0}$

of . Suppose not. Then there exist sequences

$y(t, \sigma, \phi)$ $(E_{3})$ $\{\sigma_{k}\},$ $\{\phi_{k}|[\alpha, \sigma_{k}]\}$

and such that

$\{q_{k}(t)\}$

(4.2) $\sigma_{k}\rightarrow\sigma_{0},$ $\phi_{k}|[\alpha, \sigma_{k}]\rightarrow\phi|[\alpha, \sigma_{0}]$

and $\int_{J_{1}}|q_{k}(t)|dt\rightarrow 0$ as $ k\rightarrow\infty$

and moreover $supD_{\nu_{k}}\leqq b$ , where $y_{k}(t)$ is a noncontinuable solution of

(4.3) $y(t)=\left\{\begin{array}{ll}\phi_{k}(t) & for \alpha\leqq t\leqq\sigma_{k}\\\phi_{k}(\sigma_{k})+\int_{\sigma_{k}}^{t}\hat{f}(s, yg(s, y(s)))ds+\int_{\sigma_{k}}^{t}q_{k}(s)ds & for \sigma_{k}<t,\end{array}\right.$

where $\hat{f}$

denotes the restriction of $f$ on $D\cap(J_{1}\times R^{dm})$ . Since

$|\int_{\sigma_{k}}^{t}q_{k}(s)ds|\leqq\int_{J_{1}}|q_{k}(s)|ds$

$\equiv\omega_{k}\rightarrow 0$
as $ k\rightarrow\infty$
,
it follows from Theorem 2.1 that there exists a subsequence of , $\{y_{k}(t)\}$

denoted by again, which converges compactly to a noncontinuable

$\{y_{k}(t)\}$

solution of (E). By Remark of Theorem 2.1, we have

$y_{0}(t, \sigma_{0}, \phi_{0})$
122 JONG-SON SHIN

$supD_{l_{0}}\leqq\lim_{k\rightarrow\infty}\inf[\sup D_{u_{k}}]\leqq b$ ,
a contradiction.
Next we show the inequality holds. Now suppose the conclusion (4.1)
is false. Then there exist and such that the condition (4.2) $\{\sigma_{k}\}\{\phi_{k}\}$ $\{q_{k}\}$

holds and that the equation (4.3) has a noncontinuable solution $y_{k}(t)$

having the property that for any noncontinuable solution of $x(t, \sigma_{0}, \phi_{0})$

(E),
$|y_{k}(t_{k})-x(t_{k}, \sigma_{0}, \phi_{0})|\geqq\epsilon$
for some $t_{k}\in[\max\{\sigma_{k}, \sigma_{0}\}, b]$ ,
where may depend on
$t_{k}$
. By repeating the $x(t, \sigma_{0}, \phi_{0})$
same argument,
it follows that for a sufficiently large , $k$

$|y_{k}(t)-y_{0}(t, \sigma_{0}, \phi_{0})|<\epsilon$

for all $t\in[\alpha, b]$ ,
a contradiction. This completes the proof.
THEOREM 4.2. Let $F,$ $f\in M_{D}$ and let , be the sub- $I=[\sigma_{0}, b],$ $\sigma_{0}<b$

interval of , where is the smallest interval in Theorem 4.1. Then

$J$ $J$

for any , there exists a

$\epsilon>0$
such that, if $\delta>0$ $|\sigma-\sigma_{0}|<\delta,$ $|\tilde{\phi}-\phi_{0}|_{\mathfrak{c}\alpha,\sigma_{0}}l<\delta$

in case ( $\alpha<\sigma_{0}$
in case ), where ,
$|\phi(\alpha)-\phi_{0}(\alpha)|<\delta$ $\alpha=\sigma_{0}$ $\tilde{\phi}(t)=\phi(\tau(\sigma, \alpha, \sigma_{0}, \alpha)t)$

and if $F$
satisfies
$\int_{prQ}\sup_{X\in Q_{t}}|f(t, X)-F(t, X)|dt<\delta$

for every compact set $Q\subset D$ , then every noncontinuable solution $y(t, \sigma, \phi)$

of
$y^{\prime}(t)=F(t, yg(t, y(t)))$

satisfies
$|y(t, \sigma, \phi)-x(t, \sigma_{0}, \phi_{0})|<\epsilon$
for all $t\in[\max\{\sigma, \sigma_{0}\}, b]$ ,
where $x(t, \sigma_{0}, \phi_{0})$
is a noncontinuable solution of (E) which may depend
on $y(t, \sigma, \phi)$ .
We can easily prove this theorem by Corollary 2.1 and by the argu-
ment used in the proof of Theorem 4.1.
COROLLARY 4.1. Suppose that is in the class and that a non- $f$ $M_{D}$

continuable solution of (E) through is defined on and

$x(t, \sigma_{0}, \phi_{0})$ $(\sigma_{0}, \phi_{0})$ $D_{x}$

is unique. Let , be the subinterval of . Then for

$I=[\sigma_{0}, b],$
, $\sigma_{0}<b$ $D_{x}$ $\epsilon>0$

there exists a such that, if $\delta>0$

in case $|\sigma-\sigma_{0}|<\delta,$ $|\tilde{\phi}-\phi_{0}|_{\mathfrak{c}\alpha,\sigma_{0}}J<\delta$
$\alpha<\sigma_{0}$
 DELAY DIFFERENTIAL EQUATIONS 123
( in case ), where
$|\phi(\alpha)-\phi_{0}(\alpha)|<\delta$ , and if $\alpha=\sigma_{0}$
$\tilde{\phi}(t)=\phi(\tau(\sigma, \alpha, \sigma_{0}, a)t)$ $q(t)$ ,
, is a bounded and integrable function with
$t\in J_{1}=[\alpha, \sigma_{0}]\cup D_{x}$ $\int_{J_{1}}|q(t)|dt<\delta$ ,
then every noncontinuable solution of satisfies (4.1). $y(t, \sigma, \phi)$ $(E_{3})$

Note that the result corresponding to Corollary 4.1 for Theorem 4.2
can be obtained similary.
The following is the familiar theorem on the continuous dependence
of solutions on initial conditions.
COROLLARY 4.2. Let $f\in M_{D}$ . Suppose that a solution $x(t, \sigma_{0}, \phi_{0})$
of (E)
through is defined on $A>0$ , and is unique.
$(\sigma_{0}, \phi_{0})$ $[\sigma_{0}, \sigma_{0}+A],$ Then for
any , there exists a
$\epsilon>0$
such that, if and $\delta>0$ $|\sigma-\sigma_{0}|<\delta$ $|\tilde{\phi}-\phi_{0}|_{\mathfrak{c}\alpha.\sigma_{0}}l<\delta$

in case ( in
$a<\sigma_{0}$case ), where $|\phi(\alpha)-\phi_{0}(\alpha)|<\delta$ $\alpha=\sigma_{0}$
$\tilde{\phi}(t)=\phi(\tau(\sigma, a, \sigma_{0}, \alpha)t)$ ,
then every solution $x(t, a, \phi)$
of (E) satisfies $|x(t, \sigma, \phi)-x(t, \sigma_{0}, \phi_{0})|<\epsilon$
for
all $t\in[\max\{\sigma, \sigma_{0}\}, \sigma_{0}+A]$ .
I would like to express my sincere thanks to
ACKNOWLEDGMENT.
Professor T. Naito for his many helpful comments and suggestions.

References
[1] B. M. BULLOCK, A uniqueness theorem for delay-differential system, SIAM. Rev., 9
(1967), 737-740.
[2] T. BURTON and R. GRIMMER, Oscillation, continuation and uniqueness of solutions of
retarded differential equations, Trans. Amer. Math. Soc., 187 (1974), 193-209.
[3] T. COSTELLO, On the fundamental theory of functional differential equations, Funkcial.
Ekvac., 14 (1971), 177-190.
[4] R. DRIVER, Existence theory for a delay-differential system, Contrib. Differential Equa-
tions, 1 $(1\Re 3),$ 317-336.
[5] R. DRIVER, Existence and stability of solutions of a delay-differential system, Arch.
Rat. Mech. Anal., 1O (1962), 401-426.
[6] R. DRIVER, A two-body problem of classical electrodynamics: the one dimensional case,
Ann. Physics., 21 (1963), 122-142.
[7] R. DRIVER, Ordinary and Delay Differential Equations, Springer-Verlag, New York-
Heidelberg-Berlin, 1977.
[8] P. HARTMAN, Ordinary Differential Equations, John Wiley, New York, 1964.
[9] J. K. HALE, Theory of Functional Differential Equations, Springer-Verlag, Berlin-
Heidelberg-New York, 1977.
[10] J. K. HALE and J. KATO, Phase space for retarded equations with infinite delay,
Funkcial Ekvac., 21 (1978), 11-41.
[11] T. L. HERDMAN, A note on noncontinuable solutions of a delay diferential equa-
tion, Differential Equations, ed Ahmad, Keener and Lazer, Academic Press, 1980,
187-192.
[12] J. KATO, Kamke theorem in functional differential equations, preprint.
[13] A. STRAUSS and J. A. YORKE, On the fundamental theory of differential equations,
SIAM. Rev., 11 (1969), 236-246.
 124 JONG-SON SHIN

[14] K. P. RYBAKOWSKI, A topological principle for retarded functional differential equations

of Caratheodory type, J. Differential Equation8,39 (1981), .
$131-1\mathfrak{X}$

[15] T. YOSmzAWA, Stability Theory by Liapunov’s Second Method, Math. Soc. Japan, 9
(1966), 21-24.

Present Address:
DEPARTMENT OF MATHEMATICS
KOREA UNIVERSITY
$1-7W$ , OGAWA, KODAIRA-Sm,
TOKYO 187.

View publication stats