a linear volterra integrodifferential equation for viscoelastic rods and ...

0 downloads 0 Views 666KB Size Report
Abstract. It is proved that the resolvent kernel of a certain Volterra integrodifferential equation in Hilbert space is absolutely integrable on (0, oo).Weaker ...
QUARTERLY OF APPLIED MATHEMATICS VOLUME XLV, NUMBER 3 OCTOBER 1987, PAGES 503-514

A LINEAR VOLTERRAINTEGRODIFFERENTIALEQUATION FOR VISCOELASTICRODS AND PLATES* By RICHARD D. NOREN Old Dominion University, Norfolk, Virginia

Abstract. It is proved that the resolvent kernel of a certain Volterra integrodifferential equation in Hilbert space is absolutely integrable on (0, oo).Weaker assumptions on the convolution kernel appearing in the integral term are used than in existing results. The equation arises in the linear theory of isotropic viscoelastic rods and plates.

1. Introduction. We study the equation

y'(t) =-A* Ly(t) + g(t),

j>(0) = y0, t > 0 (' = ^),

(1.1)

in a Hilbert space H, where y(j and g(t) belong to H, A: R 1-» R is locally absolutely continuous, L denotes a self-adjoint linear operator defined on a dense domain D of H. We assume the spectrum of L is contained in [1, oo) and*denotes the convolution

h\ * h2(t) = (' hx(t - s)h2(s)ds. Jo

Let {Ex} be the spectral family corresponding to L. Define r oo

U(t) = /

«(/, X) dEx

where u{t, A) is the solution of

u'(t) = -\A * u(t), Existence, particular,

u(0) = 1, t > 0.

(1-2)

uniqueness, and representation results for (1.1) work out just as in [6]. In the conclusions of Theorem 1.1 below imply that

U'(t)y

= jt[U(t)y],

if L~l/2y e D~

(1.3)

' Received July 8, 1986. ©1987

503

Brown University

504

RICHARD D. NOREN

moreover, if y0 e D, g: R *-» H is continuous with g(t) e D for all ?, and Lg: R + —>H is locally Bochner-integrable, then the unique solution of (1.1) is given by

)'(

+ 0( r 2),

co,re5),

where S = {t e C: lmr 1, then an easy calculation shows

that (1.15) holds. As already mentioned, if -a'(0) —b'(0) < oo then (1.15) holds. (c) In Sec. 3 we give an example of a piecewise linear function -a'

(with any b) that

satisfies (1.11) and (1.14) but not (1.15). I do not know if (1.19) holds for this example. The condition

a(0) + b(0) < oo is assumed in Theorem 1.1. The following result allows

b( 0) = oo. Theorem

1.2. Assume that (1.10) holds. Let b(t) = t~p, 0 < /? < 1, and assume that a(t)

satisfies (1.11) and (1.14).Then (1.13),(1.18),and (1.19)must hold. Note. Our proof of these two theorems does not extend to a(t) = t~a, 0 < a < 1, even

if -b'(0) < oo. 2. Proof of Theorem 1.1. We first recall some consequences of (1.11). By [5], the functions

a(t)

= 0, a(t) and b(t) lie in {~ir/2 < argw < 0}: if (1.14) also holds, this conclusion remains true when Im t = 0, t > 0. In this paper, -it < arg w < it (w e C). Integration by parts and the Riemann-Lebesgue theorem show that ( T, A)

^T,

(2.21)

A LINEAR VOLTERRA EQUATION

509

Following [6] we show that

\u4(t, X) | < Mr2

(\,t>

1),

(2.22)

\u5(t,A)|


1).

(2.23)

Mr1

Then by (2.21), (2.19), and (1.13), (1.19) follows. Moreover, (1.18) is a consequence of (2.21), (2.22), (2.23), and the Riemann-Lebesgue theorem. For (2.22), integrate by parts in the definition of u4 (using (2.15)) and then use (2.15)

and (2.20). For (2.23), integrate u5 by parts (using (2.16)) and then use (2.11) and (2.16) to see that

x

t2X

1

,

T~3

r-4

+ A"2

+ < M 1 + / —— p \D2(t, X/L)\ |Z)3(t,A/L)|

u5\ t, ^

17

dr,

(2.24)

where L is defined in (2.26) below. Define Wj = coj(A) > 0 by #(coj) = A"1.By (2.6), o>jis unique when it exists. Now let w(A) = cjj(A) when exists and cjj ^ p, and let 0 such that |p,

t2A

\>\).

(2.25)

To proceed we need a lower bound on Im D(t, X/L) = -t0(t) + tX 1. Writing 0(t) explicitly in terms of ,lt6, and 6X,a straightforward calculation shows that

lim 0

LAr) =

piTd^r))2+(mp

+ q)j26{T)ei{T)

;



(p

We then have

e(r)-f

5*L

—i

+ mpT2e2(T)

»(t)4

- »(-r)|i,(T)

- i| - |0(t)

- i,(x)ff(T)

We find a common denominator and use (2.2) and (2.8) to estimate |0(t) The result of this straightforward but tedious calculation is

|$(t)

" L1(t)0(t)|
')(t)

(t > 0).

Therefore,

| Lj(t) - L|< A/|RerV)(r) - Re(^7)(r)| 0. (Similarly for 5fc).

Therefore (2.27), (2.25), (2.29), (2.2), and (2.28) imply that there is a constant M1 > 0 such that L|co — t|(co + t)

e L\a - t|

- b'(s) ds

tL|w — t|

tX

_

^

a(0) coLe

L|co — t|

L|w — t|

tX

2tA

Therefore,

^11/2

L +e)\\D(T,\/L)\2

\D(t,\/L)\3 J 7 M fx

\ + 2pq4>x + q2+ p2(r01)2 + pqT26dl + q2{ T0J2,

and D = p(t>] +(q + mp)x§2+ mq3+ /?(#>(T0,)~ + 2mp(j>T6t6x + mq(T0)2 + (q - mp)l('r6)2.

Therefore, by (2.31), (2.33), and (1.15) we have the estimate 1

+f

T 3

T 4 + A

—--+ /U *L-e

. .» \D(t,. A)|2 + |Z>(r,A)|3

1 /'" + E T 3

r

^L-e$(T)2

2+

T 4 + A 2 ,