Tόhoku Math. J.

42(1990), 109-117

REPRESENTATION OF THE SOLUTION OPERATOR GENERATED

BY FUNCTIONAL DIFFERENTIAL EQUATIONS

Li Z H I X I A N G

(Received January 25, 1989, revised June 26, 1989)

1. Introduction. Hale gave a representation theorem for the solution operator

generated by a neutral functional differential equation (NFDE) with finite delay, which
represents the solution operator as the sum of a bounded linear operator with zero
spectrum and a conditionally completely continuous operator (cf. [1]). This result proved
to be very useful in studying the existence of periodic solutions of NFDE, and has been
generalized to functional differential equations (FDE) with infinite delay (cf. [2]). In
this paper, we generalize the latter result to NFDE with infinite delay, and give some
applications to the existence of periodic solutions.
In the present paper, we denote the segment of a function x(s) for — oo <s^ί by
xv and let X be a Banach space of some real functions φ: (— oo, 0]->i?" with the norm
|| 0|| having the following properties:
(H t ) Ifx: (— oo, σ + A)->Rtt, A > 0 , σ^O, is continuous for te[σ, σ + A) and xσeX,
then xteX and xt is continuous for te[σ, σ + A).
(H 2 ) There is a positive constant k0 such that |</>(0)|^&0||</>||, for φeX, where
I I stands for a norm in Rn.
(H 3 ) There are positive constants K and M such that if x satisfies (H t ) then

\\xt\\^Ksup\x(u)\+M\\xJ, t^σ.
ue[σ,t]

n
A continuous functional D: [0, oo)x X-±R is said to be atomic (cf. [3]), if it can
be represented as

= A(ήφ(O)-L(t)φ, ί>0, φeX,

with a continuous nonsingular nxn matrix A(t) and a bounded linear operator
L(t): X^Rn which satisfy sup,^ 0 | L(t)| ^ L , sup,^ 0 (| A(t)\ + 1 A ' \ t ) \ ) < A and | L(t)φ \ ^
y(β)\\φ\\ for t^Q, β^O and φeX with compact support contained in (— oo, 0], where
L and A are positive constants and y is a nonnegative continuous function on [0, oo)
with y(0) = 0. Here and hereafter, | L(t) \ and | A(t) | stand for the operator norms of L(t)
and A(t), respectively.
For any atomic D and any He C([σ, oo), Rn), the equation

(1.1) D(ήzt = H(ή9 ί^σ^O, zσ = φeX, H(σ) = D(σ)φ,

110 Z.-X. LI

has a unique solution. Henceforth, we will denote the solution by z(σ, φ, H). If the
zero solution of (1.1) with H = 0 is uniformly asymptotically stable in the usual sense,
then D is said to be stable. The following estimate for \\zt(σ, φ, H)\\ plays an important
role in this paper:

LEMMA 1.1. (cf. [4]). If D is table, then there exist positive constants a andb such that

at
\\zσ+t(σ,φ,H)\\<be- \\φ\\+b sup \H(u)\.
ue[σ,σ + t]

2. The representation theorem. In this paper, we consider the NFDE with infinite
delay:

(2.1) ^-(D(t)xt)=f(t,xt), t^σ^O, xσ = φeX,

at
where / : [0, oo) x X->Rn is completely continuous and D is stable. Throughout this
paper, we assume that the solution x(σ, φ) of the initial value problem for (2.1) is
unique. Now, define the operators Γ, TD and To as

Γ(σ, t)φ : = * σ + ί ( σ , φ),

(2.2) TD(σ, t)φ :=zσ+t(σ, φ, D(σ)φ) ,
Γ 0 (σ, t)φ:=zσ+t(σ9 0, hφ) ,
where hφ(u): = $uσf(s9 Γ(σ, s — σ)φ)ds, u^σ. From the uniqueness of the solution of the
initial value problem for (1.1), we have that, for all σ^O, t^O and φeX,

(2.3) Γ(σ, 0 0 = ^ ( σ , t)φ + Γ 0 (σ, /)0

and that ^ ( σ , /)</) is linear in φeX. Moreover, by Lemma 1.1, we get
\\TD(σ,ήφ\\^be-at\\φ\\9 t>09 σ^0, 0eX, D(σ)φ = 09
\\T0(σ,ήφ\\^b sup |A^(iι)|, ί>0, σ^0, 06JΓ.
«e[σ,σ + ί]

σ w
LEMMA 2.1. IfBaXis bounded with the property that Uo^M^ί^( > ) ^ ^ bounded,
then T0(σ, t)B is a precompact subset of X.
PROOF. From the complete continuity of /, we can find c > 0 such that
\f(s, T(σ, s — σ)φ) K c for {s, φ) e [σ, σ +1] x B. Then, we see easily that {hφ\ φeB) is a
precompact subset in C([σ, σ + t], Rn). Now, (H 2 ) and Lemma 1.1 imply that

(2.5) \z(σ909hφ)(u)\^ko\\zu(σ909hφ)\\^kob[ \f(s9 T(σ, s-σ)φ)\ds^k0bct

Jσ

for φ e B and u e [σ, σ + ί], and that

 REPRESENTATION OF THE SOLUTION OPERATOR 111

(2.6) \z{σ,O,hί){uί)-z{σ,Q,
I z(σ, 0, Ax - * 2 ) ( « i ) I +1 z(σ, 0, h2)(Ul)-z(σ, 0, h2){u2) \

^kob sup \h1(u)-h2(u)\ + \z(σ,0,h2)(u1)-z(σ,0,h2){u2)\

ue[σ,σ + t]

for ^ e C ( [ σ , σ + t], Rn) with Aj(σ) = 0 and ute[σ,σ + tl i=l,2. The inequality (2.6)
implies that z(σ, 0, h){u) is continuous in (w, h) e[σ,σ + t]x{he C([σ, σ + /], Rn)\ h(σ) = 0};
hence z(σ, 0, /z)(w) is continuous uniformly in (w, /ι)e [σ, σ +1] x {/zφ: φeB}. Therefore,
the inequality (2.5) implies that {z(σ, 0, hφ)( ): φeB} is uniformly bounded and
equicontinuous in [σ,σ + t]. Then, the compactness of the closure of the set
Γ 0 (σ, t)B = {zσ+t(σ, 0, hφ):φe B] follows from the same argument as in [2, Lemma 2.1].

According to [4], for each σ ^ 0 , there is an nxn matrix of functions

Φσ = {Φu •' ,φn}9φJeX,*nd

(2.7) \\Φj\\<Ml9

for 7 = 1 , * , n, such that A~ (σ)D(σ)Φσ = I, where M1

1
is a positive constant
independent of σ and / is the n x n unit matrix.

THEOREM 2.2. The solution operator T(σ, t) of (2.1) can be written as

Γ(σ, 0 = Tx(σ, t) + T2(σ, t\

Γ^σ, ί): = ΓD(σ, ί)(/ - ΦσA ' \σ)D{σ)).

Γ 2 (σ, t): = ΓD(σ, ί ) ( * σ ^ " \σ)D(σ)) + Γ 0 (σ,

and TD, To are the same as in (2.2). Furthermore, T1 is a linear, bounded operator and
is a contraction for large t; T2 has the property that T2(σ, t)B is precompact for a
bounded BczXif\J 0^s^tT(σ, s)B is bounded.

PROOF. Since
Λσ + ί
D(σ + t)xa + t(σ, φ) = D(σ)φ + f(s, T{σ, s - σ)φ)ds ,
Jσ

we have

T(σ, t)φ = xβ + t(σ, φ) = zσ+t(σ, φ, D(σ)φ + hφ) = zσ+t(σ, φ, D(σ)φ) + zσ+t(σ, 0, hφ)
1
= zσ+t(σ, (I-Φ"A- (σ)D(σ))φ, 0) + za+t(σ, Φ°A-\σ)D(σ)φ, D(σ)φ) + zσ+t(σ, 0, hφ)
a
= TD(σ, / ) ( / - ΦΆ -\σ)D{σ))φ + TD(σ, t){Φ A -\σ)D{σ))φ + T0(σ, t)φ .

Then, from (2.4), we get

||Γp( σ , t){I-Φ°A-\σ)D(σ))φ\\ ^be-°tW-ΦσA-ί{σ)D{σ))φ\\

a
^be- '(l+Mίl+M1lLA)\\φ\\ ,
 112 Z.-X. LI

where a, b are as given in Lemma 1.1, Mx is given in (2.7) and />0 is a constant such
that X" = 1 | Wj|</| w| for all u = (ux, , un)eRn. This means that 7\(σ, t) is a linear,
bounded operator and is a contraction for large t. Since
i)Φσ){A~ \σ)D{σ)φ),
the operator TD(σ, i)(ΦσA~ι{σ)D{σ)) is an operator from X to a subset of X, which is
spanned by {TD(σ, t)φfj= 1, ••*,«} and is bounded on bounded subsets of X. Hence,
σ 1
TD(σ, t)(Φ A~ (σ)D(σ)) takes bounded subsets of X into precompact sets. The last
assertion then follows from Lemma 2.1.

3. Application. In this section, we suppose that D and/in (2.1) are ω-periodic in

t and ω>0 is a constant. In this case, the set {f(t, φ)\ t^0,φeB} is bounded for any
bounded set BaX; hence \Jt^0T(σ, t)B is bounded if B is bounded and the solutions of
(2.1) are uniformly bounded. In the same way as in [4], one can show the following:
LEMMA 3.1. If D() is ω-periodic, then A( ) is ω-periodic and there is an nxn
matrix Φσ = {φί, -- ,φn}, φjSX, \\φj\\<M, Φσ+ω = Φσ,for σ^0,j= 1, ---,n,such that
A~ 1(σ)D(σ)Φσ = /, for σ ^ 0.

For the statement below, some definitions and notation are needed. The
^-neighborhood of a set KaX will be denoted by 0{K, δ) or O(K). Let oc(K) be the
Kuratowski measure of noncompactness of a bounded set KczX. For fixed σ^O, a
family {T(σ, t), t^O} of mappings from X to X is an ω-periodic flow, if Γ(σ, t)x is
continuous in (/, x), T(σ, 0)x = x and Γ(σ, t + ω) = T(σ,t)T(σ, ω). If the system (2.1) is
ω-periodic, then so is the solution operator T(σ, •)• {T(σ, t), t^O} is point (resp.
compact, resp. locally) dissipative if there is a bounded set BaX of attracting each
point x (resp. each compact set //, resp. a neighborhood O(x) of each point x) in X, by
which we mean that for each x (resp. each H, resp. a neighborhood O(x) of each x),
there is an N>0 such that Γ(σ, ήxeB (resp. Γ(σ, ήHaB, resp. T(σ, t)O(x)czB) for
t^N.{ T(σ, t), t^ 0} is said to be conditionally completely continuous, if for any bounded
set BdX with the property that [j 0^s^tT(σ, s)B is bounded, the set T(σ,t)B is
precompact. (Γ(σ,/), ί^O} is a conditional α-contraction if there is a constant A;e[0, 1)
such that (x(T(σ, t)B)^koί(B) for any bounded set BczX with the property that
\J0^s^tT(σ,s)B is bounded. If T(σ, •)(') takes bounded subsets of [0, oo)xX into
bounded sets, then a conditional α-contraction is an α-contraction. The same definitions
can be given for a continuous function T: X->X.
LEMMA 3.2 (cf. [7]). Let {T(σ, t\ t>0} be an ω-periodicflow.IfT(σf t) = S(σ, t) +
U(σ, t), where S(σ, t) is a bounded linear operator such that ιS"(σ, ω) = S(σ, nώ) for any
integer n>0, S(σ,ω) has the spectral radius less than one, and {U(σ, t), t^O} is
conditionally completely continuous, then T(σ,ω) has a fixed point if {T(σ, t),
is compact dissipative.
 REPRESENTATION OF THE SOLUTION OPERATOR 113

THEOREM 3.3. If the solution operator {T(σ, t), t^O} of (2 A) is compact dissipative,
the operator T(σ, ω) has a fixed point.
PROOF. For each σ^O, let S(σ, t)=T1(σ, t). Then
S2(σ, ω)φ = T2(σ, ω)φ = TD(σ, ω)(/- ΦσA " \σ)D(σ))(TD(σ, ω)(I- ΦσA " \σ)D{σ)))φ
= T2D(σ, ω)(I- ΦσA " \σ)D{σ))φ
- TD(σ, ω)(Φ°A ~ \σ)D{σ)){TD{σ, ω)(I- ΦσA " \σ)D(σ))φ)
= TD(σ, 2ω)(/- ΦσA ' \σ)D(σ))φ
-(TD(σ, ώ)Φσ){A~ \σ)D{σ)TD(σ, ω)(I- ΦσA" ^^^(σ))^)
= S(σ, 2ω)φ - (TD(σ, ω)Φσ)(A ~ \σ)D{σ\I- ΦσA ~ \σ)D(σ))φ) = S(σ, 2ω)φ ,
and in general, we have Sn(σ, ω) = S(σ, nω) for any integer n>0. The assertion now
follows from Theorem 2.2 and Lemma 3.2.
COROLLARY 3.4. If the equation (2.1) has a bounded solution x(φ)( ), which is
uniformly stable and asymptotically stable in the large, then (2.1) has an ω-periodic solution.
PROOF. Under the conditions in the corollary, the closure of the set {xt(φ)\ t^O}
is a compact set and it attracts each compact subset of X, since it is uniformly stable
and asymptotically stable in the large (for the details, we refer to [6, pp. 95-98]).
LEMMA 3.5 (cf. [5]). Suppose T: X^X is point dissipative, is a conditional
a-contraction, and satisfies the condition that for any xeX, there is a neighborhood O(x)
such that (J j=1TjO(x) is bounded. Then T is locally dissipative.
THEOREM 3.6. If the solutions of(2λ) are uniformly bounded and ultimately bounded
for a boundb, i.e., for any A>0 there is a β(A)>0 such that \\T(σ, t)φ\\ *ζβ(A),for t^O
and φeX with \\φ\\^A, and for each (σ, φ)e[0, co)xX there is an n(σ, φ)>0 such that
\\T(σ, t)φ\\ ^b for t^n(σ, φ), then (2.1) has an ω-periodic solution.
PROOF. For the ω-periodic flow {T{σ, i), t^O} generated by (2.1), there is an
equivalent norm || || x in X,

\\Φ\\<\\Φ\\i<K\\Φ\\, for φeX

such that || ^(σ, ω ) | | 1 < l (cf. [6, p. 92]). The assumptions on the solutions of (2.1)
imply that {T(σ, t), t^0} is point dissipative with B\ = {φeX\\\φ\\1^ίb} attracting points
of X, and for any φeX there is a neighborhood O(φ) such that (J JLίT(σ, t)O{φ) is
bounded. Furthermore, T(σ,ω) is an α-contraction, since Uo^ί< ω ^( σ ' ^ *s a l w a v s
bounded for bounded subsets HczX and {T(σ, t), ί^O} is a conditional α-contraction
with respect to the new norm || 1^. Then by Lemma 3.5, T(σ, ω) is locally dissipative.
Thus the existence of an ω-periodic solution follows from Theorem 4.4 in [6, p. 92].
The following theorem is a generalization of Theorem 6.4 in [6, p. 98].
 114 z.-x. Li

THEOREM 3.7. For an ω-periodic linear nonhomogeneous NFDE with infinite delay,
the existence of a solution boundedfor t^σ implies the existence of an ω-periodic solution.
PROOF. For a linear NFDE

(N) ^-(D(t)xt) = F{t)Xt + h(t),

dt
we have T(σ9 t)φ = L(σ, t)φ + xσ+t(σ, 0, h), where L is the solution operator generated
by the equation

(H) -^-(D(t)xt) = F{t)xt

dt
and x(σ, 0, h) is the solution of (N) with x σ = 0. Since each solution x(σ, φ) of (N) is the
solution of (1.1) with

) = D(σ)φ+ - I F(s)x
F{s)xjs+
sds+ \ h{s)ds,
v σ J

it follows from Lemma 1.1 that

sup D(σ)φ+ \ F[s)xjs+ \ h(s)ds

ί ||^S)|| \\x
\\x\\ds+\
s\\ds+ \\
\h(s)\ds,
s
J σσ
J Jσ

hence

II /
\\xt(σ, <
JMI ^ I I / Λ , T . -Λίί-ffWll 111 .
ίl
I 1 7 / \ 1J \ .1 Λ II ^ ί 5 ) II ^ S

which implies that Γ(σ, ί) takes bounded subsets of [0, o o ) x J into bounded sets in X.
In particular, L(σ, t) has this property. Therefore, Theorem 2.2 implies that
L(σ, ή=Uί(σ, ή+U2(σ, t\ where U2(σ, t) takes bounded subsets of X into precompact
sets, ί/^σ, «ω)= ί/" (σ, ω) and Ux(σ, t) is linear, bounded and || U^σ, t)\\ ^cxe~at. Then,
as in the proof of Theorem 3.6, we can find an equivalent norm ||. || x such that
||ί/i(σ, ω)\\1<l. Since T(σ, ω) is only a translation of L(σ, ω), it follows that Γ(σ, ω)
is an α-contraction. Repeating the same reasoning as in [6, p. 98], one can complete the
proof.

EXAMPLE. Consider the neutral integrodifferential equation

(3.1) — ί x(ί) - Ci(ί, ί + s)x,(s)rfs j = Ax(t) + c2(ί, t + s)xt(s)ds +f(t),

 REPRESENTATION OF THE SOLUTION OPERATOR 115

where xeRn, A is an nxn matrix, and / : R+-+R, ct: R+xR^>R"2 0=1,2) are
continuous. (3.1) has at least one Γ-periodic solution, if the following conditions are
satisfied:
(i) There is a positive constant Γ>0 such that -f(t+T)=f(t) and
φ+T, t+T+s) = Ci(t, t + s), for all ί>0, s^O;
(ii) There are an nxn positive definite symmetric matrix B and positive constant
w( (/=1,2) such that
ATB + BA=-I,

for aΆxeR";
(iii) There are constants y>0 and me(0, 1) such that

Γ
"0

τ
2\A B\mw2 2\B\w2 f°
1- ' 2 \c2(t,t + s)\ds^u>0,
(l-m)wί (l-m)wj_oo
ys
and that J° J cf(ί, ί + s) \e~ ds 0 = 1 , 2) are convergent uniformly for tl
PROOF. Denote
f°
J-
F(ί, 0) = Aφ(0) + c2(ί, ί + s)φ(s)ds
* — oo

For the space Cy of the continuous functions φ: (—oo, 0]-+Rn with the property
that lim^.^e^l^sjl exists, the hypotheses (H 1 )-(H 3 ) are satisfied and K=l,
+
M(ί)->0 (ί-> + oo) (cf. [2]). One can prove that D and F are continuous o n i ? x C r
Γ-periodic in /, and linear bounded in φ, and that D is stable. Moreover, we have

(3.2) \\Φ,Φ)\\^(\\Φ\\+ sup |iϊ( M )|

for the solution z(σ, φ) of (1.1) with zσ = φ and H(σ) = D(σ, φ).
τ
Now, take F(ί, x) = x Bx. We will prove that
(3.3) F(ί, D(/, xt(0, 0)))<M, /^0 ,
2
where M>(2f\B\wj/uwί) J: = suρt>0\f(ή |. Indeed, if (3.3) is not true, then there are a
number to>0 and a sequence {/„}, ίM->/0 + («^oo), such that
116 Z.-X. LI

(3 4 )
V(s, D(s, xs)) < M = V(t0, D(t0, xj) ,
V(tn,D(tn,xtn))>M, n=l,2,

Then, we have

(3.5) Vi3Λ)(to,D(to,xJ)>0,

and

(3.6) ID(s, xs) I < ^ = — {V(t0, D(t0, xt0))}1

w1 wγ
From (3.2) and (3.6), we have

l
(3.7) |xίs)|<||x,K sup l ^ ' ^ l ^ {V(to,D(to,x,o))y<2
o^u^to \—m ( 1 — m)w1

(l-m)wί

It follows that
VOΛ)(t0, D(ί0, xt0)) =
ί Γ° V / Γ° \
= I Ax + c2xtods \ BD +fτBD + DτBl AX + c 2 x ί o ώ j + DτBf
/»0 /»0
= x Γ ^ TBD + xJacτ2dsBD +fτBD + DτBAx + DTB\ c2xtods + DτBf
J -oo J - oo

ΛO ΛO
= DTATBD+ xJoc\dsAτdsAτBD+ xJocτ2dsBD+fτ BD
* — oo J — oo

,4 CiX^ίfc + Z) T 5 c2xtQds + D Γ i?/

J — oo J — oo

V
where D9 ct (i= 1, 2),/, x and xto stand for D(ί 0 , χtQ)9 φθ9 to + s) ( ι = 1, 2),/(ί 0 ), χ(ί 0 )
and xf0(s), respectively. Hence

w 2f|5|
1 2
ί^,-, 4 χ^ Λ D(t~ x \)<C V(t T)(t v Y>4- -f Γ/Y/ Π/> v ^ H /
 REPRESENTATION OF THE SOLUTION OPERATOR 117

= - uM/wl + 2/1B \y/M/w1 < 0 ,

which is contrary to (3.5). Therefore, the solution xt(0, 0) of (N) is bounded. The
assertion now follows from Theorem 3.7.

The author would like to thank Professors Junji Kato and Satoru Murakami for
their valuable suggestions. Thanks are also due to the referee for helpful advice.

REFERENCES

[ 1 ] J. K. HALE, A class of neutral equations with the fixed point property, Proc. Nat. Acad. Sci. U.S.A.
67 (1970), 136-137.
[ 2 ] J. K. HALE AND J. KATO, Phase space for retarded equations with infinite delay, Funk. Ekv. 21 (1978),
11-41.
[ 3 ] Z. C. WANG AND J. H. Wu, Neutral functional differential equations with infinite delay, Funk. Ekv. 28
(2) (1985), 157-170.
[ 4 ] Z. X. Li, The generalized difference equations with infinite delay, to appear.
[ 5 ] A. F. IZE, A. A. FREIRIN AND J. G. Dos REIS, A critical study of stability of neutral functional differential
equations, in Recent Advances in Differential Equations, Academic Press, 1981, 209-230.
[ 6 ] J. K. HALE, Theory of Functional Differential Equations, Appl. Math. Sciences Vol. 3, Springer-Verlag,
1977.
[ 7 ] J. K. HALE AND O. LOPES, Fixed point theorems and dissipative processes, J. Differential Equations
13 (1973), 391-402.
[ 8 ] J. H. Wu, Uniform behaviors of solutions of neutral equations with infinite delay (I), Ann. of Differential
Equations 4 (2) (1988), 189—230.

DEPARTMENT OF MATHEMATICS
HUNAN UNIVERSITY
CHANGSHA, HUNAN
PEOPLE'S REPUBLIC OF CHINA