f Math. Nachr.

147 (1990) 175-184

Periodic Solutions of Linear Integro-Differential Equations

By T. A. BURTON])
of Carbondale, P. W. ELOEand M. N. ISLAM
of Dayton

(Received August 22, 1988)

Abstract. Using a degree-theoretic result of GRANAS, a homotopy is constructed enabling us to

ahow that if there is an a priori bound on all possible T-periodic solutions of a Volterra equation,
then there is a T-periodic solution. The a priori bound is established by means of a Liapunov
functional. The latter result is unusual in that no bounds on the Liapunov functional are required.
Thus, in addition to the periodic solution, the equation may have both bounded and unbounded
solutions.

1. Introduction
We consider a system of equations
i

(1) ) l)x(t) K
~ ' ( 1= J' C(t - S) X ( S ) ds + p ( t )
-m

in which C is an n x n matrix of continuous functions which are in L1[O,m), D is an

n x n constant matrix, p : (--00, 00) 4 11'0 is continuous and T-periodic, and usually
f p ( 8 ) ds = 0.
0

The object, is to give conditions on D and C to ensure that ( 1 ) has a T-periodic solu-
tion without, placing restrictions on the size of T (other than T > 0 ) or on the size
of P3(1).
The problem has its roots in the theory of ordinary differential equations. A dis-
cussion may be found in [2; pp. 66-68 and pp. 88-89]. Briefly, if x = 0 ' 5 the only
solution of
5' = DX
which is bounded on ( - m , m), thcn
Z' = DX +p(t)
I) Supported in part b y an NSF grant with Number NSF-DMS-8521408.
176 Math. Nachr. 147 (1990)

has a T-periodic solution for any T-periodic p (t ). The result can be broken into th&
illustrative cases.

Case 1. If all characteristic roots of D have negative real parts, then the periodic
solution is
1

z(t) = J eD(f-a)p(s)ds.
-m

Case 2. If all characteristic roots of D have positive real pa*, then the periodic
solution is
m

s ( t )= -J eD('--S'p(g) ds.
1

Case 3. If some roots of D have positive and some roots of D have negative real
parts, then a periodic solution is of a form analogous to a combination of Cases 1 and 2
[2; p. 681.
Now Case 1 was extended to (1) in [2; p. 891 when the resolvent for the unforced
equation (1) is L"0, 00) or when it tends to zero as t + 00. Case 2 was extended to (1)
by WANG[8]. Both extensions involved finding a bounded solution of
1

Y' = DY + 1C(t - 4 Y ( 4 + p ( t )
0
d5

and using a limiting equation argument. Case 1 was also extended to (1) without uae
of the resolvent by HINOand MURAKAMI[6] simply by using a bounded solution of (1).
Our work here focuses on extending Case 3 t o (1) without asking for any bounded
solutions or any restrictions on the size of T or p . LANCIENHOP [7] does partially
extend Case 3 and obtains two very nice results; one restricts T, while the other restricts
the size of p . His work has additional merit in t h a t examination of (1) alone suffices to
conclude that a periodic solution exists. Our work requires more.
We construct a homotopy compatible with differential equations having some
bounded and some unbounded solutions. If it is known t h a t all solutions of the un-
forced equation are bounded (or all are unbounded), then a different homotopy can be
defined [3] which circumvents many of the obstacles encountered here.
The technique me use is that of GRANAS[6] who calls i t the method of continuation of
POINCAR$ edlater refers to i t as a topological transversality method. First, we prove
that if there is an a priori bound on all possible T-periodic solutions of an equation
related to (l),then (1) has a T-periodicsolution. We then give examples using LWKJNOV
functionals and SOBOLEV'S inequality to prove the a prim. bound.
The work of LANQENHOP
[7] involves a generalization of (l),namely,
W

z'@)= J ( [ d m 1 4 - 4)+ PO)

0
Burton/Eloe/Islam, Periodic Solutions 177

where E has bounded total variation. This will include (1) as well as terms like Ds(t - r )
1
and 1U X ( Sds.) The techniques here can be directly applied to such generalizations.
i-r

The next section is a brief synopsis of the work of GRANAS.This particular presen-
tation is mainly that of ELOEand HENDERSON [4].

2. A topological method
The following definitions and Lemmas 2.1 and 2.2 are duc t o GIMNAS [5]. Assume all
topological spaces are HAUSDORFF. Let Y be a topological space, A c X c Y, and A
closed in X.
-
Definitions. (i) A continuous map cp: X --f Y is compact if p(X) is compact.
(ii) h : [ 0 , 13 x X 3 Y is a compact homotopy if h is a homotopy and if for each 2 [O,l],
h I 2 X x’ = hl is compact.
(iii) p: X -+ Y is admissible with respect to A if p is compact and ip I A is fixed point
free. Let M A ( X ,Y) denote the class of admissible maps with respect to A .
(iv) q~ E M A ( X , Y) is inessential if there exists a y~ E M A @ , Y) such that p I A = p I A
and p is fixed point free on X. Otherwise, p E M A ( X , Y) is essential.
(v) A compact homotopy h : [0,1] x X -+ Y is admissible if for each 2 E [0, 11, hi is
admissible. Two mappings p , y ~E M,(x’, 1’) are homotopic in M,(X, Y ) , p “ p , if
there exists an admissible homotopy h : [0, 13 x X 3 Y such that h, = y~ and h, = q.
(vi) F* dcnotcs the class of topological spaces which have the fixed point property
for compact maps. We remark t h a t a convex subset of a BANACH space is a n P* space
by SCHAUDER’S fixed point theorem.
Lemma 2.1. Let Y be a connected space belonging to F*, let X c Y be closed, and let
A = ax. If i p : X -+ Y is a constant mapping, (that i s , ~ ( x=) p for all x E X ) , and
p E X \ A , then p is essential.
Lemma 2.2. Let A = 2 c X c 1’ and # E M , ( X , Y) an adrnzksible function. Then
the following are equivalent:
(a) f : X 3 Y is inessential.
(b) f g an M A @ , Y), where 9 : X -+ Y is without fixed points.
N

3. Notation and periodic solutions

The symbols

(PT7 11*11)
space of continuous T-periodic functions p: R
will denote the BANACH 3 R* with the
supremum norm and

12 Math. Nachr., Bd. 147

 178 Math. Nachr. 147 (1990)

will denote the subspace of PT with q E PTif cp E PT and if the mean value of q is
zero :
T
m(q) = (l/T) J q ( s ) ds = 0.
0

Let 0 1 5 1 and consider the family of equations

t

(11)

To each cp E
~ ' ( t=) 1 Dz(t)

PT
L + J C(t - S)
we associate the function Q E
Z(S) ds + p(t)].
PTdefined by
T
(2) Q ( t ) = q ( t ) - (1P) J v(5) d5
0
:
having mean value zero and the consequential property that J @(s) ds E PT.
0

It is assumed throughout this paper that p E PT.

Note that if p E PT,if p 4 PTand
W

if M =D + J C(u)du is nonsingular, then a translation y =x + d, d constant, can

0
be defined so that (1) becomes an equation with forcing function having mean value
zero.
Now define a n operator h (or h l ) on [0, 11 x PT by the equation
t u t
(3) h(1, (4= 1 @(O)
[
+ D o/F(s) ds +
Lemma 3.1. Let B > 0, Y = P,, X = (9E Y : 11cp,11 5 B ) ,and A = ax.Then h : [0,1]
0 -
J C(u - 8) Q(s) ds du + J p(s)ds .
0 1
xX +Y is a. compact homotopy for h defined by (3).
t
P r o o f . By the remark following (2), if we can show that J C(t - s) +(s) ds E PTfor
--oo I
each q E PT,then it will follow that h : [0, 13 x X --f .'I Clearly, / C(t - s) @(s) ds E PT,
so we must show that it has 0 mean value. In -m

/T -/
0 --oo
C(t - s) Q(s) ds dt

we let Q E P$ and substitute u = t - s to obtain

T I T w
J J C(t - S) Q(s) ds dt = 1J C(U)+(t 7 U ) du dt
0 --oo 0 0

W T
= j Cb)J' CO - ~ ) d t d u = O
0 0

since the inner integral is 0 for each fixed u.

 Burton/Eloe/Idam, Periodic Solutions 179

We now show that h is compact. First, bounded sets are mapped into equi-continuous
sets since

where ID1 IM and 1.1 is a compatible matrix norm. Also, if 0 =( t T, then

while

These inequalities show that h maps bounded sets into bounded equi-continuous sets.
Next, we show that h is contiriuous jointly in (1,p). Fix t 2 0 and note t h a t

But

Thus, if 0 5 1 5 1' we have

m

l/'L('J 971) - h(A,P2)11 5 2A 11971 - 91211

[ + IDI + 1Ic(u)l
0
du
1
so h is continuous in y in the supremum norm. Since h is linear in A i t is uniformly con-
tinuous in 1;by a theorem from real variables h is jointly continuous in (A, y ) . Hence,
h is a compact homotopy.
Theorem 1. Assume that there 8 a B1 > 0 such that [lzJ < B, lor every continuous
sohtion x1 E PT of (11) for every ;i E (0,I]. Then ( I ) has a solution in P T .
P r o o f . We note that if 97 E PT is a fixed point of hi, then q'(t)= @ ' ( t ) so that 9 E P$
is a solution of (11) and, by hypothesis, ll$ll < B,. (If A = 0, then from (3), p = 0.)
I2*
180 Math. Nachr. 147 (1990)

Thus, for P(t) = J p ( s ) ds, if for some 1we have h(1,q ) (1) = ~ ( t ) then
,
0

Section 2, hl is admissible. Also, h, h,. -

This defines the B for Lemma 3.1. Hence, hl is fixed point free on A and, by (iii) of

Now, h, = 0 E PT and 0 E X \ A since B > 0. By Lemma 2.1, h, is essential. Also,

ho ,- h,. We let g = h, and = h, (in Lemma 2.2) and we want h, to have a fixed point
in X . If h, is fixed point free, then by (a), h, i s essential, a contradiction. The proof is
complete.
Example 3.1. Consider the two dimensional system

1
f
C(t - s) z(s) ds + p(l)
with

D= ('0 -1
O),

w m W

J J IC(u)l du dt < 00, -1 + J JC(u)ld u < -a for some b: > 0, C continuous, p E PT.
O f 0
Then there is a T-periodic solution for 1 = 1.
P r o o f . Define

so that 17.31 = 1/2 and DTE + ED = -I. If

f m
= zTEz
V ( t ,z(-)) + ( 1 / 2 ) J J ~ C ( U )dus2(s)
-m I-8
( ds

then
V ' ( t , z(*)) EX
= (9')' + zTEd
Burton/Eloe/Islern, Periodic Solutions 181

for appropriate positive constants 8, y , and .M. Hence, if x = x1 = p E PT is a solution,

then V ( t ,p) E PT and an integration of V‘(t, q ) from 0 to T (with 17 = V ( t ) )yields
T
0 = V ( T )- V(0) =( - y R J Is(s)l ds + RMT;
0

if A t (0, 11, then

T
Ix(s)l ds IMT/y.
0

An integration of the equation then yields

T
hi summary, thcre is a J > 0 with J’ [ Ix(s)l + 1z’(s)1]ds < J . By SOBOLEV’Sinequality,
there is a K = K ( T )with 0

T
Ilzll s K J [Ix(s)l + I W I I ds E!B, *
0

This completes the proof.

Remark. In (1 : pp. 127- 133 and 142- 151 J there is extensive discussion on con-
struction of the matrix E from an indefinite matrix D and on the construction of 77
even when D is zero. In [3] w e constructed a homotopy which is effective for non-
linear equations even when C is not of convolution tlype; however, it involves analysis
not of ( l i ) ,but of a related equation (which in the linear case is)
f

(4 1 2.; = i(?d + D)z1 - ndx1 + iJ C ( t , 8 ) 21(s) ds + p(l)

-m

where m is a nonzero constant and I is the n x ?L identity.

We now show that Theorem 1 directly extends the results for the O.D.E. problem
stated in three cases in the introduction and that it is independent of the results in
182 Math. Nachr. 147 (1990)

[3] concerning (4). Let

(*I 2' = 1[Dx + p(t)]

where D is n x n, no root of D has 0 real part, and (possibly after a transformation)
p E P",. According to [ l ; p. 1281 there is a C O J l S t a 3 l t matrix E = ET and a positive
+
definite matrix J with DTE ED = - J . Define V ( x )= xTEx so that along any
T-periodic solution of (*) we have V'(x) 5 A[ - x T J s 2 IEJJJpIJ +
lsl] 5 1[ -a 1x12 M] +
for positive constants a and M. The rest of the details of showing 11x11 < B, for some
B, > 0 are identical the those in Example 3.1.We conclude that (*) has a solutionin
P, and that Theorem 1 extends the ODE problem.
However, in [3] when (*) is considered i t takes the form (corresponding to (4))

(**) x' = ?,(ml + D )x - m l x + p ( i ) .

It is required that there is a. bound B on any solution in PT for 0 5 1 5 1. This can
not happen when D is indefinite. Take p(t) = 0 and

The coefficient matrix in (**) is

=J ( I ) .
%(m- 1) - m,
Then

J(0)= (-: ")

-?n and J(1) = (i -B.
Thus, for some 1 c (0, I ) , 0 is a characteristic root of J ( A ) so (**) has arbitrarily large
constant solutions. This shows that [3] and the present paper are independent.
Examplo 3.2. Consider the scaler equation
1

(5 ) y" = 1k(t -
--m
8) ~ ( 8ds
) + ~ ( 1 )

where k : [0, CQ) 3 (-00, 03) is continuous, p E POT, K ( i ) = k(u)du E L'[O, OO),
1 m m f

JR(u)Idu ds exists, K ( 0 ) = 1, and j IK(u)l du = OL < 1. Then (5) has a solution

-m f-8 0
in P,.
Proof. Write ( 6 ) as the system (introducing the parameter 1)
Burton/Eloe/Ishm, Periodic Solutions 183

and note that since K ( 0 ) = 1 we may write

1

(7) y(t) - (d/dt)j/ K ( t - s) y(s) ds

--oo
+ p ( t ) ].
Then (6) is equivalent to the system
t

(8) Ax - (d/dt)J E(t - 5) z(s) ds + p(t)]

--oo

where

For
0 -1
B = ( 4 0)
W

we have ATB + BA = -21. Find y >1 such that (1 + 1') J IK(u)ldu < 2. Then
define 0

V ( t ,z(.))= (Z + 1 j R ( t - S) ~ ( sds) R(t - S) ~ ( s ds

)
--oo

+ 1y J
I W

J IKWl duz2(s)ds
--oo I--a

so that for L > 0 we have

V ( t ,x(-))= 1 [$'AT + pT(t)]B

3
+ 1 [z + i R(t - s) 4 s ) ds
--oo JT B[Az + F(t)J
184 Math. Nachr. 147 (1990)

References

[ 13 BURTON, T. A., Volterrn Integral and Differential Equations, Academic Press, Orlando,
Florida 1983
[2] -, Stability and Periodic Solutions of Ordinary and Functional Differential Equations,
Academic Press, Orlando, Florida 1985
[3] BURTON, T. A., ELOE, P. W., and ISLAM, M. N., Nonlinear integrodifferential equations and
apriori bounds on periodic solutions, Ann. Mat. Pura. Appl. (to appear)
[4] ELOE,P. W., and HENDERSON, J., Nonlinear boundary value problems and a priori bounds
on solutions, SIAM J. Math. Anal. 16 (1984) 642-647
[6] GRANAS,A., Sur la mkthode de continuit4 de PoincarQ, C. R. Acad. Sc. Paris, t. 202 (1976)
983-985
[6] HINO,Y., and MURAKAMI, S., Periodic solutions of a linear Volterra system, Proceedings of
Equadiff S7 (toappear)
[7] LANQENEOP, C. E., Periodic and almost periodic solutions of Volterra integral differential
equations with infinite memory, J. Differential Equations 60 (1985) 391 -403
[8] WANQ,Z., Some remarks on periodic solutions of Volterra equations, Ann. of Differential
Equations 1 (1985) 107-126

Department of Mathematics Departinent oi Mathematics

Southern Illinois Univereity University of Dayton
Ca.rbondale, Illinois 62901-4406 Dayton, Ohio 45469