v.

OSCILLATIONS IN NEUTRAL FUNCTIONAL DIFFERENTIAL EQUATIONS"1"

Jack K. Kale

Tills work was partially completed while the author was a guest of
CNR at Florence, Italy, and also was partially supported by the
Air Force Office of Scientific Research, AF-APOSR 71-2078, the
National Aeronautics and Space Administration, NG-L ^0-003-015
United Statec Army - Durham, DA-ARO-D-Jl-12H-71-G12S2.
 OSCILLATIONS IN NEUTRAL FUNCTIONAL DIFFERENTIAL EQUATIONS

Jack K. Hale

Some time ago [1], the author announced a result on the

Fredholm alternative for the existence of periodic solutions of a non-
homogeneous linear neutral functional differential equation (NFDE).
In this paper, we indicate a proof of this result and, at the same time,

use the method of proof to give a brief survey of some recent develop-
ments in the theory of NFDE which have applications far "beyond the
problem of periodic solutions.

Let R = (-00,<»), R = [0,«>), E be any n-dirnensional linear

vector space with norm | -|, C([a,b],E ) the space of continuous func-

tions from [a;b] to E with the topology of uniform convergence.

For a fixed r > 0, let C = C([-r,0],En) with norm |cp| =

n
sup
-r<9<0 |ep(0)| for cp e C. If x e C([a-r,cr+A),E ) for some A > 0,
let x e C. t e [a.o-fA) be defined by x (0) = x(t+0), -r < 0 < 0.
\j T- **• ~—
,n
Let D,L! C -» E be continuous linear operators,

0
/ [dT](0)]ep(0)
-r

D = D + D

DQcp = cp(0) - Z
K= J-
0
D^ = / A(0)cp(0)d0
-r

where r, is an n x n matrix function of bounded variation, A(0) is

an n x n matrix integrable on [-r,0], the A are n x n constant

 -2-

matrices with Z ., | A , | - > 0 as e -» 0 and 0 < T < r. For no-

Tfc<€ K K. -

tational purposes, there is an n x n matrix (i of bounded variation

on [-r,0] such that

0
(2) Up =9(0) - / [dn(0)]cp(e).
-r

A linear homogeneous NFDE is a relation

(3)

A solution of (3) is a continuous function x on some interval [-r,A)7

A > 0, such that Dx. is continuously differentiable and satisfies (3)

on (0,A). For any cp c C, there is a unique solution x = x(cp) of (l)

on [-r, <a) such that x = cp and this solution is continuous in

(ep,t) e C X R+ (see [2]). If T(t,D,L)I C -» C, t e R+, is defined by

(4) T(t,D,L)9 = xt(qp)

then T(t,D,L), t e R } is a strongly continuous semigroup of linear

transformations.

Let C' = {cp e C* DQcp = 0). This is a closed subspace cf the

Eanach space C and T(t,D_,0)! C -> C . Let the spectral radius of

U
^0 0
T(t,DQ,0)| CD be rD (t) and

(5) a,, .
 -3-

The operator D-. is said to be stable if a < 0 (see [8]). Note

that T(t,D_,0)| CL is nothing but the semigroup of linear trans-

formations corresponding to the solutions of the homogeneoxis difference

equation

(6) DQyt = 0.

The operator Dn is stable if the zero solution of (6) is uniformly

asymptotically stable. Also, note that e = -co if D~cp = cp(O);
DQ
that is, Dn is the operator corresponding to the usual retarded

functional differential equations.

For a fixed CD > 0, let 9 = {x e C((-«>,co),En) I x(tfoj) =

x(t),t e R) and & = (E e C((-<»,«>),%n)'. H(0) = 0 and there 5s an
n-vector a and h € P with H(t) + at + h(t)}. For any H e &'

H(t) = at + h(t), h e P^ we let |H] = |a| + supte[()^1 |h(t) | . The

theorem on the Fredholm alternative for periodic solutions can now be

stated as:

______ 0 is stable, H 'e o&,r then the equation

Theorem 1. If D,,

[Dx
(7) -

has a solution in if and onlv if

03

03
(8) . / y(t)dll(t) = 0
0
 -li-

fer all co- periodic row vector solutions y of the "adjoint" equation

0 0
(9) £e [y(t) - / y(t-0)dti(0)] = -/ y(t-0)dri(0).
-r -r

Furthermore, there is a continuous projection operator j: & -»

' CO CO

such that the set of all H satisfying (8) is (l-J) & and there is
a continuous linear operator _$^: (l-J)i^ -> & such that SjfR is a
- co co
solution of (7) for each H e (l-J) & .
It is part of the conclusion of Theorem 1 that the integral
in (8) is well-defined even though H is only continuous.
It is also possible to explicitly describe the operator J
and in doing this we will also rephrase the entire problem in terms of
the solution of an operator equation in a Banach space. This terminology
makes clear the relationship of the above theorem to the general problem
of solving functional equations .
Let A'. & -» & be defined by
CO CO

t
(10) Ax(t) = Dx, - DK_ - / Lx ds.
t 0 s

It is clear that A is continuous, linear and, furthermore,

that the null space ,%(A) consists of the solutions of the homogeneous
equation (5) in & . It will be shown below that Jfc(A) is finite
dimensional and there is a continuous projection
0 SI &
co -»co
& such
that ^*(A) = S^1 . Theorem 1 implies there is a continuous projection
J'. M -* & such that the range ^>(A) satisfies _^>(A) - (l-j) &
 -5-

and that A has a bounded right inverse 3^. Furthermore, 3^ will be

uniquely specified if we require that J3^ = 0.

If U = (cp.., .. .,cp,) is a basis for the co-periodic solutions
of (3), V = col(\|f ..,,\|; ) is a basis for the co-periodic solutions of
(9), and ' denotes transpose, then S,J can be defined as

—.-1
«••
CO CO
Sh = U / u ' (s)U(s)ds / U' (s)]r
0 0
(11) t
—
•^

~co — CD
JH(t) = / V (s)ds / V(s)V (s)ds V(s)dH(s).
0 0 0

Before proceeding to the proof of the theorem, we remark that

Theorem 1 allows one to immediately apply the usual theory for perturbed

linear systems to equations of the form

.£
(32)
dt

where G(-,cp) e & , F(-,cp) e' & for each 9 e C. In fact, if we de-

fine the operator A as before and define N: & -» & by

* co co

(15) Nx(t) = G(t,x t ) - G(0,x Q ) + / F(s,x s )ds

then equation (12) has a solution x in & if and only if

Ax = NX
 -6-

which by Theorem 1 and the above remarks is equivalent to

x = Sx + _5f (l-j)Nx
(15)
JWx = 0.

One can now apply the theory of [9] to obtain sufficient

conditions for the existence-of co-periodic solutions. In particular^

if G = eG, F = eF, where G(t,cp), F(t^cp) are continuously differ-

entiable in cp and e is a real parameter^ and there is a d-vector

b such that

A(bQ) = 0, det[d&(b 0 )/ab] t 0

CO
A(b) = / V(s)[dG(s,U(s)b) + F(s,U(s)b)ds].
0

then there is an en > 0 such that equation (12) has an co-periodic

solution x(b e), 0 < j e] < e continuous in e and x(b 0) = Ub .

We now proceed to give two methods of proving Theorem 1, the

first method will have general applicability to the discussion of the

local theory of nonlinear equations and the second is applicable to

more general boundary value problems. Basic to both approaches is the

following:

Lemma 1 [3]. There exists a continuous linear map i|rl C -» C and a

. 0
family of maps T, (t,D,L)'. C -» C. completely continuous for each b > 0,

such that
 -7-

(16) T(t,D,L) = T(t,DQ,0)t + T-^t^L).

Frcm a general result on linear operators (see [^,5]), Lemma 1

implies the following: for any fixed a > a^ , all elements n(t) in
D
0.
the spectrum cr(T(t,D,L)) of T(t,D,L) with |n(t)| > eat belong to

the point spectrum, the number of such n(t) is finite, the generalized

eigenspace of each n(t) is_ finite dimensional, and there exists sub-

spaces P ,. », P /,v invariant under T(t,D,L) with P , » finite

dimensional, C = P /.\ © P (t) and the spectrum of T(t,D,L) re-

stricted to P /t\ is a(T(t,D,L))\(|i(t)).

If Dn is stable, then, in particular, there can be only a

finite number of elements of the point spectrum of T(t,D,L) with

moduliL equal to one. Thus, there can be at most a finite number of

o>-periodic solutions of (3) and this implies the existence of the pro-

jection operator P mentioned above with ^£(A) = P ^ .

Furthermore, Dn stable implies only a finite number of ele-

ments of the point spectrum of T(t,D,L) with modulii greater than or

equal to one. The theory in [2] now implies one can decompose C as

C = P ©Q where P,Q are invariant under T(t,D,L), cr(T(t,D,L) |Q) =

a('l(t,D,L)) 0 (X'. \\\ < 1], P finite dimensional and P,Q are de-

termined in the following manner. A number X is said to be a

characteristic value of ( if

0 0
(1?) det A(X) = 0, A(X) = X
-r
- -r/
 -8-

Let D "be stable, A = {XI det A(X) = 0, Re X > 0} and <& = (^

be a basis for the initial values of all solutions of (3) of the form
JLt
E Pk(t)e , X^ e A, PkOO polynomials in t, and ¥ = col(\|r . ..,\|r )

a basis for the corresponding solutions of the adjoint equation (9) of

-X,t
the form E q.k(t)e } X, e A, ^(t) polynomials in t. Let

0 6
(ot/p) = a(0)Dp + / / a(|-e)[dn(e)](p(i)d|
-r-0
o e
- / / a(|-e)[dTi(0)]cp(!)de
-r 0

and (¥,$) = (t.,9k), j,k = 1,2,...,p. Then (Y,$) is nonsingular and

may be taken to be the identity. It follows from [2] that

C = P © Q

(18) P = {9 e C'. 9 = $a for some p-vector a]

Q - (9 e C'. (f,9) = 0 ) .

Thus, any 9 e C can be written as 9 = 9 P +O9 P, 9 = $(Y,9), 90" =

P
9 - 9 e Q.
Since P is invariant under l(t,D,L), there is a p X p

constant matrix E, o(E) = A such that T(t,D,L)$ = Oe . Also, from

[2], $(0) = 0(0)eE0, -r < 0 < 0.

Another fact that is needed is the variation of constants

formula. It is shown in [6] that there is an n x n matrix X(t),

-w < t < oo, of bounded variation, continuous from the right, X(t) = 0,

t < 0, X(0) = I satisfying

 -9-

t
D(Xt) = / L(Xs)ds + 1, t > 0

def
such that, if X. = T(t)X , the initial value problem for (7) is

equivalent to

(19) xt = T(t)cp + / T(t-s)X0dH(s).

If one now defines X^ = *(¥,XQ) = *¥(0), X^ = XQ - X?, then

Oy(t) + xT; implies

"C

ae [yOO - ^(o)11^)]= Ey(t)

(2°) t
x£ = T(t,D,L)cpQ + / T(t-s,D,L)xgdH(s)

where all integrals are to be interpreted as regular integrals in S

for each 6 in [-r,0].

Since the spectrum of T(t,D,L)|Q lies inside the unit

circle, there are positive constants K,a such that |T(t,D,L

Ke~ } t > 0. Therefore, it follows from [6] that one can also suppose

K,a are such that JT(t,D,L)XQ] < Ke"0^, t > 0. Thus, there is a

unique <p = cp (H) continuous and linear in H such that the solution

x. of the second equation in (20) is aperiodic in t. If H(t) =

Ot + h(t), a e En, h e & t and z(t) = y(t) - y(0)h(t) in the first.

equation of (20), then the existence of an Aperiodic solution is'

equivalent to the existence of an Aperiodic solution of the ordinary

 -10-

differential equation

(21) J±i±l = Ez(t) + ET(0)h(t) + Y(0)a.

dt

A necessary and sufficient condition for the existence of an co-periodic

solution of (21) is well-known from ordinary differential equations.

Checking this condition and proceeding as in [10] for retarded functional

differential equations, one completes the proof of Theorem 1.

A second proof can be obtained from the general theory of two

point boundary-value problems developed in [7]. We summarize the theory

for the autonomous case although the nonautonomous case is also treated

in [?].
For the statement of the principal results on boundary value

problems, some care is needed in the specification of the continuity

properties of the functions r\f\i in (l), (2). Without loss in

generality, one can suppose both r\f[i are continuous from the left

on (-r,0), vanish on [0,«) and are equal to their values at 6 = -r

on (-09, -r]. Let Bft be the space of functions \|r. [-r,0] -> E * (the. "'

space of n-dimensional row vectors) which are of bounded variation on

[-r,0], continuous from the left on (-r,0) and \|;(0) = 0. We identify

BO with the conjugate space of C rfith the pairing

0
(t.cp) = / [dt(e)Me), t eB cp G c.
-r

With p.',T| normalized as above, the adjoint equation (9) as

 -11-

an equation for functions of bounded variation can be written as

(22) y(s) - / [dy(a)]n(s-a) + / y(a)n(s-a)da = constant.

s s

It is then not difficult to prove that for any t e R, \|r e BO, there
is a unique y: R -» E of bounded variation on finite intervals,
continuous from the left such that y = \|r, y vanishes on [t,«) and
*
(22) holds for s < t - r. If this solution is denoted by y(t,i|r) and
y?(0)
s = y(s+0), -r <
••e < 0. y°(0)
s = 0. then y°(t,i|r)
s e B_
u for each
s < t.
Let n: BQ -» B be the quasinilpotent operator defined by

0 0
(2?) fli|r(0) = -/ [di|r(p)]ji(0-p) + / i|r(P)n(0-P)«3p, -r < 0 < 0, $ e B .
6 6

Suppose V is a Banach space, or < T are given real numbers

M,W. C -» V are linear operators with domain dense in C and i" e V
is fixed. Let V* be the conjugate space of V and M*,N* the ad-
joint operators of M,N, respectively. The boundary value problem (l)
is to find a solution of (7) satisfying

(2U)

Theorem 2. For boundary value problem (l) to be solvable, it is

necessary that
 -12-

(25)

for all solutions y, i|; of the adjoint problem: ij; e V*, y satisfies

the adjoint equation (22) on [cr-r, T-r] and z = (l-fft)~ M*\Jr,

If ^(M+WT(T-a)) is closed in V, this condition is both

necessary and sufficient.

To apply this result to the proof of Theorem 1, let V = C,

M = -N = I, f = Of 0 = 0, T = CD. The boundary value problem (l) is

then to find a solution of (7) with x = x . To show j^?(l-T(a>,

is closed observe from Lemma 1, that I - T(co,D0,0)\|r has an inverse

so that

- T(U>,D,L) =

Since T, (o>,D,L) is completely continuous, it follows that this range

is closed. Thus, (8) is necessary and sufficient for the existence of

an co-periodic solution of (7) . This characterizes the range of the operator

A in (10). Since it also is shown in [7] that the dimension of the space of

co-periodic solutions of the adjoint equation is the same as the dimension of

the space of co-periodic solutions of (3) and, thus, is finite, there exists

a continuous projection operator JI £& -», £& such that

It follows that A has a bounded right inverse J€' and Theorem 1 is

proved .
 Incidentally, the above argument applies equally as well to

show that M + NT(to,DQ,0) nonsingular implies ^(M+NT(a>,D,L)) is

closed.

References

1. Hale, J.K., Stability and oscillations in functional differential

equations. Proc. Symp. on Information Theory, Princeton,

2. Hale, J.K. and Meyer, K.R., A class of functional equations of

neutral type. Mem. Amer. Math. Soc., 76(1967) .

3. Hale, J.K., A class of neutral equations with the fixed point

property. Proc. Nat. Acad. Sci. U.S.A., 67(1970), 136-137-
k. Gokhberg^ i.e. and Krein, M.G., Introduction to the Theory of Non-
selfadjoint Operators. Transl. Math. Monographs Amer. Math. Soc.
18, 1969.
5. Ambrosetti, A., Proprieta spettrali di certi operati lineari non
compatti. Rend. Sem. Mat. Padova, ^£(1969), 189-200.

6. Hale, J.K. and Cruz, M.A., Asymptotic behavior of neutral functional

differential equations. Arch. Rat. Mech. Ana., 3^(1969), 331-353.

7. Henry, D., Adjoint theory and boundary value problems for neutral
functional differential equations. J. Differential Equations.
To appear.