Anda di halaman 1dari 78

Chapter 6

Other Applications
In Chapter 4, we applied the energy multiplier method and frequency do-
main criteria to determine the exponential stability of linear dynamic sys-
tems which generate Co-semigroups on Banach or Hilbert spaces. For systems
which satisfy the spectrum-determined growth condition. the exponential sta-
bility can be examined by analyzing the spectrum distribution of the systems.
One advantage of this method over others is that we are able to know not
only whether the systems are exponentially stable. but also the exponential
decay rate.
In this chapter, we shall consider two types of infinite dimensional sys-
tems - wave equations and thermoelastic equations with certain boundary
stabilizers. Certainly. these systems can be viewed as linear Cauchy prob-
lems on appropriate Hilbert spaces. Therefore. the well-posedness of the
problems can be treated in the framework of semigroups discussed in Chap-
ter 2. Because the boundary stabilizers are usually designed to make the
systems dissipative, it is routine to check the weak stability and asymptotic
stability by using the theory developed in Chapter 3. Therefore. our major
concern in this chapter is the exponential stability and the exponential de-
cay rate for these systems. In Section 6.1. we discuss a general hyperbolic
system consisting of a first order partial differential equation in time t and
one-dimensional spatial variable x. This hyperbolic system. which originates
from the work of Neves [124], is often encountered in the counter-flow heat
exchanger process, gas absorber process, tubular reactor process, connected
vibrating strings, and many other applications. In this general system. we
allow only one boundary to possess dynamics. but the analysis method to be
presented also can be applied to treat the cases with two dynamic boundaries.
In order to exploit the solution properties that this general system possesses.
we first consider three reduced systems which are simplified equations either
by ignoring the couplings between state variables. or by replacing the dy-
namic boundary with a static boundary. For the most simple reduced system
Z.-H. Luo et al., Stability and Stabilization of Infinite Dimensional Systems with Applications
Springer-Verlag London Limited 1999
310 Other Applications
among the three. we show that the spectrum-determined growth condition
holds. This, together with a nl11nber of results on the relationship among
semigroups of each reduced system. demonstrates that the growth rate of
the general system is determined completely by its own spectral bound and
that of the most simple nxluccd system. Using this key result. we arc able
to show. in Sections 6.2-6.4. that the spectrum-determined growth condition
is satisfied for wave equations describing the dynamics of serially connected
vibrating strings with point stabilizers, as well as a vibrating cable with a
tip mass. In Sectioll 6.5 and 6.6. we cOllsider a thermoelastic system with
both Dirichlet- Dirichlet and N cnmann- Dirichl<ct boundary cOllditions. It is
shown that the spectrum-determined growth cOlldition holds for these sys-
tems with the aid of a result due to Rellardy [136] on spectrum-determined
growth condition for a class of normal operators with bounded pertllrbatiollS
in Hilbert spaces. Finally. in Section 6.7. we present a counter-example given
by Renardy [136] which shows that the spectrum alone canIlot determine the
growth rate for a two-ciimensiollal hyperbolic system even with lower order
perturbatioIls. This negative result suggests that care lllust be taken when
we treat practical infinite dimensional systems, and also demonstrates the
importance of the theoretical developments presented in this chapter.
6.1 A General linear hyperbolic system
Consider the following olle-dinwllsional linear homogeneous hyperbolic sys-
tem
{
[u(x.t)] +. [1L(X,t)] +G(.X) [H(;E,t)] =O.O<x< 1.
at v(x, t) Ox: o(x. t) o(x. t)
d. (6.1)
-[V(l, t) - Du(l. t)] = Fu(l. t) + CV(l. f).
dt
ufO, t) = Ev(O. t)
for which the following assumptions are always imposed throughout the sec-
tion:
(HI) K(x)=diag(Adx), .... AN(J;).pN+d:r) ..... Pn(:z:)) is a diagonal n x 11
matrix with real entries A;(X) E G
I
[O.l]./lj(x) E GI[O, 1], Ai(.1;) >
O,/lj(x) < 0 for all J: E [0.1] andi = 1. 2 .... . N,j = N +1. N +2 .... . n.
(H2) G(x) is an n x n matrix with entries (:ij(X:). which are continuolls in
x E [0,1].
(H3) if i 'I j and Ai(X) = Aj(x)(or p;(x) = Pj(x)) somewhere in [0,1]. then
Cij(X) == 0 on [0.1].
(H4) 11(X) = (nl(x),u2(X) .... ,UN(X)f is a colulllIl vector in RN (or eN)
and v(x) = (v N+!lX:). 'UN+2 (x) . ... ,vnCz:))T is a column vector in Rn-N
(or e
n
-
N
).
6.1 A General linear hyperbolic system 311
(H5) D. E, F and G are real (or complex) constant matrices of appropriate
SIzes.
We shall find that it is convenient to denote a coluIlln vector u by u =
col[u;] and matrices D.E.F, and G by D = {dij}.E = {eij},F = {fij}, and
G = {gij}, respectively.
For any p, 1 p < 00, let X = (P(O, l))n xRn-N (or (P(O.l))n xC
n
-
N
).
where LP(O, 1) is the usual LP function space with norm II lip' Define a linear
operator A: D(A)(C X) -+ X by
A(u,v,d)
[-K(x) dd
x
-C(x) ] ,Fu(l) + Gv(l)].
D(A) {(u,v,d) E XI (u,v) E (Hl.P(O, l))N x (Hl.P(O,l))n-N,
u(O) = Ev(O), d = v(l) - Du(l)} (6.2)
where H1,P(0, 1) denotes the usual Sobolev space. With the operator A at
hand, we can write (6.1) as an evolution equation on X:
dW'(t) = AW(t)
dt
with W(t) = (u(, t), v(, t), v(l, t) - Du(l, t)).
(6.3)
Lemma 6.1 The operator A defined by (6.2) has compact resolvent and
hence a-(A) consists of only isolated eigenvalues. Furthermore, a-(A) con-
sists of the zeros of an entire function.
Proof. Given (f,g,b) E X, we solve
(,X - A)(u,v,d) = (f,g,b).
that is,
{
[u(x)] = -K-1(x)['x + C(x)] [U(X) ] + K-1(x) [f(X) ] ,
ax v(x) v(x) g(x)
,Xd = b + Fu(l) + GV(l). u(O) = Eu(O). d = u(l) - DU(l).
(6.4)
Denote by M(x, y,'x) the fundamental matrix of the system
dd
x
= -K-1(x)['x + C(x)] .
It follows from (6.4) that
[
U(X)] [E] r -1 [flY)]
v(x) = M(x, O,,X) I v(O) + Jo M(x, y, )")K (y) g(y) dy. (6.5)
312 Other Applications
On the other hand, from the boundary conditions in (6.4), we see that
b =
(-AD-F.A-G) [a.(I)]
1'( 1)
(-AD-F.A-G)M(LO.A) v(O)
+( -AD - F. A - G) 11 M(Ly, A)K-
1
(y) ] dy.
Consequently,
H(A)v(O) = b + 11 (AD + F. G - A)M(l, y, A)K-
1
(y) [ ] dy, (6.6)
where
H(A) = -(AD+F,G-A)M(l;O,A) (6.7)
Defining h(A)=detH(A), we see that A E a(A) if and only if A is a zero of
the entire function h(A). When h(A) =J: 0, A E p(A) and R(A, A)(f,g, b) =
(u,v,d) where (u,v) is given by (6.5) with v(O) determined by (6.6) and
d = v(l) - Da(I). It is seen from (6.5) that R(A, A) is compact for any
A E ptA). 0
Theorem 6.2 The operator A defined by (6.2) generates a Co -semigro'up
T(t) on X.
Proof. We need only to prove the assertion for the case C == 0, because C
is a bounded operator by assumption (H2), and bounded perturbations do
not affect Co-semigroup generations. Assume, for simplicity, that X is real.
The trick is to define an equivalent norm on X by properly choosing
some positive weighting functions j; (x), 1 ::; i ::; Nand gi (x), N + 1 ::; i ::; n.
Namely, define the norm on X as
N 1 n 1 n
II(u,v,d)II
P
= L 1 Ji(x)lui(x)IPdx+ L 1 g;(X)IVi(X)IPdx+ L Id;lp.
i=l 0 j=N+1 0 j=N+1
(6.8)
It is easily verified that X *, the dual space of X, consists of all elements
(a*, v*, dO) with
ui(x) = II(a. v, d)112-Plai(x)IP/qsign(ui(x)), 1::; 'i ::; N,
v;(x) = II(u,v,d)11
2
-
P
lvj(x)IP/qsign(vj(x)), N + 1::; j::; n,
d; = II (u, v, d) 11
2
-
P
ldj IP/qsign( dj ). N + 1 ::; j ::; n.
6.1 A General linear hyperbolic system 313
where q denotes the conjugate number of p. i.e" q satisfies l/p + l/q = l.
For any (u, v, d) E D(A), (u, v, d) 0 and any (u*, v*, d*) E F((u, v. d)) C
X*, where F the duality set, a direct calculation shows that
II(u,v,d)IIP-2((u*,v*,d*), A(u,v,d))
N 11 d n 11 d
L -)"i(X)!i(X) dx lUi (x)IPdx + L -Ilj(x)gj(:r) dx IVj(x)IPdx
i=1 0 j=N+l ()
+(Fu(l) + Gv(l), [v(l) - DU(l)]*)
N n
= - L)"i(l)!i(l)lui(l)IP - L Ilj(l)gj(l)IvJ(l)IP
i=1 j=N+l
N n
+ L )"i(O)J;(O)lui(O)IP + L Ilj(O)gj(O)lvj(O)IP
i=1 j=N+l
N 11 d n 11 d
+ '" IUi{X)IP-[),,;(x)!i(xl]dx + L IVj{x)IP-[llj(x)gj(x)]dx.
L...J 0 dx 0 dx
i=1 j=N+l
+(Fu(l) + GV(l), [v(l) - DU(l)]*) = 11 + 12 + 13 + h
We estimate Ii separately. It is clear from the expression of 13 that
h Co II(u,v,d)II
P
(6.9)
where Co = max i, j max
x
E[O,I] {dd", [)"i(X )J;(x )], dd", [Ilj (x)gj(x)]). Noting that
Ui(O) = L:j=N+l eijVj(O), we see that
N n
12 = L )"i(O)J;(O)lui(O)iP + L Ilj(O)gj(O)lvj(O)iP
i=1 j=N+l
n N n
< L [llj(O)gj(O) + L )"i(O)!i(O)( L leiklq)p/q]IVj(O)iP.
j=N+l i=1
(6.10)
Because )"i(O) > 0 and Ilj(O) < 0 from (HI), we can always find gj(O) > 0
and j; (0) > 0 such that
N n
Ilj(O)gj(O) + L )"i(O)!i(O)( L leiklq)p/q 0, N + 1 j n (6.11)
i=1 k=N+l
holds, which implies that h O.
We now estimate h Using the inequalities (lal + Ibi)P 2
P
(la1
P
+ IW)
and laI
1
/
P
IW/
q
lal/p + Ibl/q which hold for any real a and b, we have
n N n N
14 L I L !jiui(l) + L gjivi(l)IIVj(l) - L djiui(l)IP/q
j=N+l i=1 i=N+l i=1
314 Other Applications
1 n N n
< - L 1 L !ji U ;( 1) + L Yj;v;( l)II'
P j=N+1 ;=1 i=N+1
1 n N
+;- L IVj(1) - l.: djilLi(l)1l'
1 j=N+1 i=l
21' n N n 1
< - L [I L !ji ll i(lW + 1 l.: gji
v
i(lW] + -11(u,u,d)II
1'
P j=N+1 i=l i=N+1 q
21' n N plq N
< L (l.:I!jil
q
) l.:lu;(l)IT'
P j=N+1;=1 ;=1
21' n (n ) piq n . 1
+- l.: l.: Igjil
q
l.: IV;(l)1l' + -11(Il,v,d)ll
p
P j=N+1 i=N+1i=N+1 q
N n. 1
= L CY il ll ;(l)1l'+ L pjlvj(l)IP+-II(n,v,d)II
P
i=l j=N+1 q
with 0i and Pj denoting the obviolls constallts,
Finally, it is verified that
N n
< L[-Ai(l)/;(l) + O'ill ui(l)1l' + l.: [/Jj -l.lj(l)gj(l)llvj(1JIP
i=l j=N+1
N
< l.:[-A;(l)/;(l) + cy;]lll;{l)11'
i=l
n N
+21' l.: Ipj -l1j(l)Yj(l)IIVj(l) - L ri j;lL;(l)IP
j=N+l
n N
+21' l.: l,8j -l.lj(l)gj(1)IIl.: dji nd 1)j1'
j=N+1
;=1
N
< L[-A;{l)/;(l) + CYill ll;(l)11'
;=1
n N
+ L 2
1'
1,8j -llj(l)Yj(1)lloj(l) - Ldj ;ud 1W
j=N+1
i=l
n N N
+21' L l,8j -llj(l)Yj(1)I(L Idjilq)p/q L IIl;{l)11'
j=N+1
;=1 ;=1
6.1 A General linear hyperbolic system 315
N n N
L [ - >"i(l)/i(l) + OOi + 2
P
L Ipj - fi-j(l)Yj(l)I(L Idjilq)P/q] IUi(l)iP
i=1 j=N+l i=1
n N
+ L 2
P
lpj - fi-j(l)gj(l)llvj(l) - L dji lli (l)iP.
i=1
If we choose /i (1) > O. gj (1) > 0 such that
{
->"i(l)f;(l) + OOi + 2
P
2:j'=N+l Ipj -llj(l)gj(l)I(2:f:l Idjilq)p/q :::; o.
2
P
IPj - fi-j(l)gj(l)1 :::; C
fOf any 1 :::; i :::; Nand N + 1 :::; j :::; n, then
II + 14:::; (C II(u.v.d)II
P
.
q
(6.12 )
The estimations of Ii above show that there exists a constant M such that
((u*,v*,d*), A(u,v.d)):::; M 11(Il,v.d)11
2
(6.13)
if we choose the weighting functions fi (x) and Yi (x) such that they satisfy
(6.11) and (6.12) and define the nOfm in X according to (6.8).
Because A - M is dissipative and A has the properties stated in Lemma
6.1. by the standard argument. we conclude that A generates a Co-semigroup
onX. D
We now consider three reduced systems associated with (6.1):
[U(X, t) ] + [ u(x. t) ] + Co(x) [ u(x, t) ] = 0,
at v{x, t) ax v(x. t) v(x, t)
d
dt[v(l, t) - Du(l, t)l = Fu(1. t) + Gv(1. t).
(6.14 )
u(O, t) = Ev(O, t).
[u(x'. t)] + [U(x ..t)] + Co(x) [u(:r,. t)] = 0,
at v(x. t) ax v(x. t) v(x. t)
d (6.15)
dt [v(1, t) - Du(l, t)l = o.
u(O, t) = Ev(O, t).
{
[U(X, t) ] +. K. (x)_.a [U(X, t).] + Co(x) [U(X'. t)] = 0,
at v(x. t) ax v(x. t) v(x. t)
v(l, t) = Du(1. t). u(O. t) = Ev(O. t),
( 6.16)
316 Other Applications
where Co(x) = diag(cidx)).
Each system above can be viewed as all evolution equation W (t) =
AiW(t), -i = 2,3,4 where Ai and D(Ad are defined similarly as A and D(A)
in (6.3). X still serves as the state spaces for (6.14) and (6.15) as for (6.1).
But for (6.16), a closed subspace XO = {(u, 11, 0) E X} of X should be taken
as the state space. XO will be identified with (LP(O.l))n when there is no
cOllfusion from context.
Similar to Lemma 6. L we call show that each Ai generates a Co-semigroup
Ti(t). respectively. on their state spaces.
The main result of this section can be stated as
Theorem 6.3 For any (u, 11, d) EX, denote by TI : X -t Xo; IT( u, 11, d) =
(u, 11, 0) the projection from X to XO. Under assumptions (Hl)-(H5), T(t)-
T4(t)TI is compact on X for each t 0. Therefore, wess(A) = W
ess
(A4)'
The proof of this theorem is completed by using the results of the following
lemmas which show that T(t) - T
2
(t), T
2
(t) - T3(t) and T3(t) - T4(t)TI are
compact for each t 0, respectively. First of all, the compactness of T3 (t) -
T4 (t)TI follows from
where the first equality follows from the fact that the restrictioll of T3 (t)
on XO is equal to T4(t). Since for each t 0 and (u,v.d) E X, T3(t)(J-
TI)( u, 11, d) = dT3 (t )(0,0, 1) is a rank olle bounded operator, T3 (t) - T4 (t)IT is
compact for every t O.
Before proceeding, we state a theorem Oil the compactness of an integral
operator on LP space, which is essential for the proof in the sequel.
Lemma 6.4 Let I c R be a compact interval. If
a( s, x) : I x [a, bj -t [a, bj -is continuous and C
l
in s with (on/os)( s, x) t-
o everywhere.
rilx),r2(x) : [0,1] -t I are continuous, and
K: [a, b] x I -t R( or C) is continuous.
then K : U( a, b) -t U( a, b) defined by
(KJ)(x) = j"2(X) K(:r, s)f(a(s, x))ds
r, (x)
is compact.
6.1 A General linear hyperbolic system 317
Proof. Making change of variable a(s. x) = T and denoting by (3(T, x) its
inverse s = (3( T. x), we have
j
o(r2 (x),x)
(Kf)(x) =
o(rdx),x)
which is obviously a compact operator on V (a, b). o
The case where a(s, x) = x provides an example that condition (aa/as)
:f. 0 cannot be removed in Lemma 6.4.
Lemma 6.5 T2(t) - T3(t) is compact for every t O.
Proof. We first note that if T
2
(t) - T3(t) is compact in t E [0, to] for some
to > 0 then so is it for every t O. Indeed, for any 10 > 0, 10 to,
T2(tO + E) - T3(tO + E)
= T2(E)T2(to) - T3(E)T3(tO)
= T2(E)[T2(to) - T3(tO)] + [T2(E) - T3(E)]T3(tO),
and so T2 (to + E) - T3 (to + E) is compact. Iterating this process, we see that
T
2
(t) - T3(t) is compact for every t O. We show that if we take to = c with
c = min{_l_, 1 I x E [0,1]. 1 N,N + 1 j n}, (6.17)
A;(X) -pAx)
then T2(t) - T3(t) is compact for every t E [0, to]. We do this by directly
solving (6.14) and (6.15) using the method of characteristics.
Let T2(t)(uo,vo,do) = (u,v,d) and write
u(x, t) = col [u;(x, t)],
uo(x) = col [UiO(X)],
v(x, t) = col [Vj(x, t)],
vo(x) = col ho(x)].
Integrating along the characteristic line. we find the general solution of (6.14)
which is given by
{
u(x t) = A.(t - fX -l-dT)e - foX Cii(p)/Ai(p)dp 1 < i < N
" '1'> Jo A'(T) , - - ,
X (6.18)
v(x t) = .I(t+ Jl -l-dT)e- fa Cjj(p)//l-j(p)dp N + 1 < J' < n
J' 'l'J x /l-j(T) , - - ,
where i and 1/Jj, defined on [- fol A;(T)dT,OO) and [JOl /l-)T)dT,OO), respec-
tively, are arbitrary functions to be determined by the initial and boundary
conditions. By the initial conditions, we have
318 Other Applications
or
{
Ao.(e) - .. ( . .. (e)) J'i
IO
) e,,(pl!)..,(pldp
'P' - 1L,0 ,I., f: 0
,I .. (e) - .. ( .. . (e)) ,J")IO) ('Jj(p)/i'J(pjdp
- V)O .1) (. 0 ,
e E [- dT. OJ.
e E p)1')dT.Oj.
( 6.19)
where xii e) and x j (e) an' continuously differentiable functions which are
determined by
(6.20 )
Substituting (6.19) into (6.18). one has
. e,,(p) /).., (pldp -x
Hi(X. tl = Uio(x;(ei(.X. t)).Je ' ..' . t::;.1O Ai(T)dT.
{
.1
",10,1"')) 1
j
". 10 I, III 1 (6.21)
J) (' (1')/1' (p)dp -I
Vj(X. t) = VjO(J;j((lj(X. t)))e-, J)). t::; - J -(-)dT.
,x 11j T
where ei(x. t) = t - and ej(:r. t) = t + ,J: p)r)dT.
Next. note that t - Jo
x
)..,(1')dT + 0 for any 1 ::; i::; Nand
N + 1 ::; j ::; 11 when t ::; c. Frolll the boundary condition /L(O. t) = Ev(O. tl.
which is rewritten as
in tenns of i and 1/)j. we have. for t ::; c. 0::; x ::; 1. and t 2 )..;(1') dT. that
(6.22 )
6.1 A General linear hyperbolic system
t
c
o
I
t=c I

Figure 6.1: Characteristic Lines
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
319
x
(6.21) and (6.22) clearly show that for all x E [0,1]' t E [0, C],Ui(X, t) is
determined completely by the initial conditions and the matrix E; in other
words, it is independent of F and G.
Finally, from
d(t) v(l, t) - DU(l, t)
do + lot [Fu(l, T) + Gv(1, T)]dT
do + Iot[(GD +F)u(l,T) + Gd(T)]dT
do + G 1t d(T)dT + 1t(GD + F)u(LT)dT,
we have
d(t) = v(l, t) - Du(l, t) = eGtd
o
+ lot eG(t-s)[GD + F]u(l, s)ds. (6.23)
320 Other Applications
Therefore,
0(1, t) = Du( 1. t) + e
GI
do + (eG(I-;) [CD + F]u(l. s)ds, t 2: O. (6.24)
ill
It follows that
If t :::; c, then t - :::; 0 for allY 1 :::;; :::; N. Therefore. for c 2: t 2:
J
.I 1 i
- -(-)(T.
X f-LJ T
N + 1:::; j :::; 71,
(6.25)
where L::!..j represents the j-th row of e
GI
do, e
GI
1-8) [CD + F] = {Jij( t - s)},
d = {dij}, and Oij(J. t) = t + I') iT) dT - f01 )',iT) dT.
Up to now, we have found the explicit solution of (6.14) in (t, .r) E [0. c] x
[0,1] which is expressed by (6.21), (6.22) and (6.25). Let T3 (t)(uo,vo,do) =
(it.v,d). Note that (6.15) is a special case of (6.14) for F = C = 0, w(' have,
by (6.21) and (6.22).
u(x, t) --ii(.I:. t) = 0, for 0 :::; t :::; c.O :::; .r :::; l.
and from (6.23)
d(t) - d(t)
(e
Gt
_ l)d
o
+ 11 e
Git
-
B
) (CD + F)u(1, s)ds
(6.26)
6.1 A General linear hyperbolic system 321
t N 1,,10 ,1\,8))
(e
Gt
- l)d
o
+ Jo fij(t - S)UiO(Xi(B;(L s)))e ' c,,(p)/)..'(P)dpds.
(6.27)
Moreover, for 0 :::; t :::; - I: (iT and 0 :::; ;r; :::; L by (6.21) we have
Vj(:1:. t) -Llj(:1:. t) = O.
and for - I: 1') T) dT :::; t :::; c, by (6.25) we have
where ii
j
denotes the j-th row of (e
Gt
-l)d
o
, From (6.20), we already know
that Xi( Bi ( 1, s)) is a continuous function of s, and it satisfies (OXi (Bi (1, s)) /os)
-=J 0 everywhere, According to Lemma 6.4 and (6.26)-(6.28), we see that
T
2
(t) - T3(t) is compact for all 0 :::; t:::; c, 0
In order to show the compactness of T( t) - T2 (t). the following lemma is
needed,
Lemma 6.6 Let S(t) be a Co-semigroup and L a bo'unded linear operator on
a Banach space X, If I: S(t-s)LS(s)ds is compact/or each fixed t E [0, to],
then
{I
t
S(t - s)LS(s)xds. x EX, Ilxll :::; L t E [0. to]}
is precompact. Moreover, I: S(t - s)LS(s)ds is compact in C(O, to; X).
Proof. Given E > 0 and let no = no (E) be an integer such that
1111+
7
> S(t + r - S)LS(s)dsll < f
322 Other Applications
for all t E [0. toJ alld alII' 2: 0 satisfyillg /' :::; to/no. Let tn \)(' defilled by
tn = nto/no. n = O. L ... no - 1. For t E [tn. tn+lJ alld r = t - tn. we hav\'
t S(t - s)LS(s)ds
./0
l
tn+r
S(t" + l' - s)LS(s)ds
0
(tn+r
S(l') ./0 SU" - s)LS(s)ds
1
,/" 11,,+1'
S(t - t,,) S(tn - s)LS(s)ds + S(l' + tn - s)LS(s)ds.
o . til
Since u;n S(tn - S )LS( s ):cds. II xii :::; I} is precompact. so is
K" = {S(t - tn) t" S(t" - s)LS(s)xds. Ilxll :::; L t E [tn. tn+d}
./0
(sec the proof of Theorelll 3.53). Hellce K = Ko U ... U K
no
-
1
is an f net
of U; S(t - s)LS(s)xds. II.fll:::; Lt E [0. toJ. This proves the first part of the
lemma. Secondly. for any II > 0
1111+h S(t + h - s)LS(s)xds -1/ S(t - S)LS(S)XdSII
< 11[I+h S(t + II, - S)LS(S):Z:dsll
+ II[S(h) - IJlt S(t - S)LS(S)XdSII o. a:-; h O.
llniformly for all t E [0. toJ alld 111:11 :::; l. The secolld part thus follows from
t.he Arzela- Ascoli theorem. 0
Lemma 6.7 T(t) - Tz(t) is wmpact for all t 2: o.
Proof. Let to = c be as defined in (6.17). For the same reaSOll as seen ill
the proof of Lemma 6.5. if we call show that T( t) - T
z
(t) is compact. for each
t:::; to. then T(t) - T
2
(t) is compact for every t 2: O. Let
A [H(T)]
B(u.v.d) = (C(x) .0)
6.1 A General linear hyperbolic system 323
where C(x) = {Cij(X)} = -C(x) + diag(cii(x)). Then
T(t) - T
2
(t) = lot T
2
(t - s)BT2(S)ds + lot T2(t - s)B[T(s) - T2(s)]ds.
That is, T(t) - T
2
(t) is the solution of the Volterra integral equation and so
where K : C(O, to; X) --+ C(O, to; X) is a continuous operator. If for any t
0, J; T2(t - s)BT2(s)ds is compact, then by Lemma 6.6,J; T2(t - s)BT2(S)ds
is compact in C(O, to; X). Therefore, T(t) - T2(t) is compact in C(O, to; X).
In particular, T(t) - T2(t) is compact for any t 0. Consequently, in order
to show that T(t) - T2(t) is compact, it is sufficient to show that J; T2(t-
s )BT2(S)ds is compact. This is equivalent to showing that J; T3 (t-s )BT3(S)ds
is compact for each t to by using the following equality
T2(t - s)BT2(S) = T3(t - s)BT3(S) + [T2(t - s) - T3(t - s)]BT3(S)
+T2(t - s)B[T2(S) - T3(S)]
and the property shown in Lemma 6.5. Similarly, since T3(t) - T4(t)II is
compact for any t 0, from
T3(t - s)Bn(s)
= T4(t - s)IIBT4(s)II + [T3(t - s) - T4(t - s)II]BT4(s)II
+T3(t - s)B[T3(S) - T4(s)II],
it suffices to show that
is compact on X for t to. Identifying B with
, [u(x)]
B(u,v) = C(x) v(x)
on (P(O, 1))2N+2 and identifying T4(t) with the Co-semigroup on (P(O, 1) )2N+2,
we need only to show that
is compact on (LP(O, 1))2N+2 for any t to.
324 Other Applications
Let (u(x,s),v(x.s)) = T4(8)(Uo(::c).vo(:r)). Then by letting F G
do = 0 in (6.21). (6.22), and (6.25). we have
.( . ) _ . ( .((J.( ... i)) ,J,,(8,(',"1I c,,(pJ/)..,(p)dp
U, X. S - u'o x, ,x. sex .
j
'X 1
.5:S 0 )..,(T) dT.
. (. ) _ - JX c,,(p)/)..;(pjdp .. ' , (" ,((J. (' .)))
U, x, s - e 0 L.j=N+l e'JoJo.IJ ,j X.8
(6.29 )
and
,( ) _ ,( ,((J.( i)) J"j(o)(',,')1 c))(p)/I,)(pJdp < Jl 1 d
V
J
x,s -VJO X
J
J X,S e, , s_- x 1') IT) T,
N - - F- c,,(p)/)...(p)dp
Vj(X, s) = 2:i=1 dijUio(Xi((Jij(X, s)))e "Ie,)(,'
c 2: 8 2: - J: I')) T) dT, N + 1 :S j :S n.
where Xi((}) and Xj((}) are functions satisfying (6.20) and
{
{X 1 Jl 1
(Ji(X, .8) = s - Jo )..;(T) dT, (Jj(X. '.5) = s + x l'jIT)dT,
- () [1 1 /.1 1 d
(}ij X.8 = S +. x 1')(T)dT -.0 )..,IT) T.
D = {dij }. 1 :S i :S N. N + 1 :S j :S n.
Let
(6.30)
(-ii.(x, t-s),v(x, t-s)) = T4(t-S)C(x)(u(x, s). v(x, s)) = T4(t-s)(ti(:r. s), D(x. s)).
where
Then
Ui(X, s) = 2::=1 cidx)Uk(X. s) + 2:Z=N+1 cik(x)vd:r. 5),
l:Si:SN.
Vj(X, 5) = 2::=1 Cjdx)udx , 5) + cjdx)vd:r, 5),
N + 1 :S j :S n.
j
. .r,(o,(r,t-.;I)) /
_ ,( ) _ A ,( ,((J,(' _)) ), c.,(p) )..,(p)dp
u, X, t - S - u, x, ,X, t s ,s e ' ,
< j
.x 1 d
t - s _ 0 )...(T) T,
(6.31)
- (t ) - ,- J,x c,,(p)/)..,(p)dp .. '.( .((J(" t )) .)' (6.32)
Ui x, ' - 5 - e 0 L.j=N+l e'J OJ X
J
'J:1., - S ,8
[xj(8ij(X,t-s)) c' -(f)) 'I' (f))dp
X eJo -).1 /) .
c 2: t - s 2: Jo
x
1 :S i:S N.
6.1 A General linear hyperbolic system
and
Vj(x, t - s) = Vj(Xj(Oj(x, t _ s)), s)eJxXjI9jlr,,-.)) Cjj(p)//-Lj(p)d
p
,
t - S < - Jl -1-dT
- x /-Lj(T) ,
325
(6.33)
N - - Jl - Ci;(p) / ),,;(p)dp
V
- (x t - s) - d U(x(O (x t - S)) s)e r,(9,jlx . ))
3' - wi=l l.J 1. " 1.J 'I ,
xeJx' Cjj (p)//-Lj (p)dp c> t - s > - Jl -1-dT N + 1 < J' < n.
'- - x /-Lj(T) , --
It is seen that J; T4 (t - S )BT4 (s)( Uo, vo, do )ds, t :::; c, equals the sum of
the terms of the form !(x, t, s)p(x, t, s)ds, where !, rl, r2 are continu-
ous functions of their variables, and p( x, t, s) is one of Ui (Xi( 8
i
(x, t - s)), s),
U;(Xi(Oij(X, t - s)), s), Vj(Xj(Oij(X, t - s)), s) and Vj(Xj(Oj(x, t - s)), s). Let
us consider, for instance, the following term
l
r2 (X,t)
!(x, t, S)Ui(Xi(Oi(X, t - s)), s)ds
rdx,t)
(6.34)
to see what properties it has. The discussions of the other three cases are
similar. Let pi(X,t,S) = Xi(Oi(X,t - s)). Then from (6.31), Ui(Pi,S) are the
sum of the terms Cik(Pi)Uk(Pi, s), 1:::; k :::; Nand cidpi)Vk(Pi, s), N :::; k :::; n,
where Uk(Pi, s) and Vk(Pi, s) can be obtained through initial conditions by
(6.29) and (6.30). By these relations, we can express the integral (6.34) as
the sum of the terms of the form
where rikI, rjk2, gik, gjkCtik. Ctjk are continuous functions of their variables. It
can be checked that all 8Ctik(X, t, s)/8s "I 0, 8Ctjk(X, t, s)/8s "I 0 under the
condition that Cij = 0 if either Ai (x) = Aj (x) or J.ti (x) = J.tj (x) somewhere
for i "I j. Actually, the possibility of 8Cti (x, t, s) / 8s = 0 happens when, for
example, such kind of integrals include the integrand such as
gik(X, t, S)Cik(Pi)UkO(Xk(ih(Pi(X, s), s))),
where Ctik(X, t, s) = xd '19k (Pi( x, s), s)) satisfies 8Ctik( x, t, s)/ 8s = 0 if Ak(Pi) =
Ai (p;). But this is saved by our assumption and Cii = O. The proof is com-
plete. 0
Combining Lemmas 6.5 and 6.7 and noticing the compactness of T3(t) -
T4(t)II, we conclude Theorem 6.3.
326 Other Applications
Example 6.8 It is remarked that assumption (H3) in the beginning of thi$
section cannot be removed for our problem. To see this. consider
{
a [Udx,t)] a [UI(X.t)] [Udx,t)]
at U2(X, t) + K ax U2(X, t) + C U2(X, t) = 0,
v(x, t) v(x, t) v(x, t)
v(l, t) = 0, (UI(O, t), U2(0, t)) = Ev(O, t).
(6.35)
where < x < 1 and
[
1 0 0 ]
K = 0 1 0 ,
o 0 -1
E=(1.0),

000
This is a special case of system (6.16) where both K and C are constant ma-
trices. Let Te(t) be the associated semigroup of (6.35) and T(t) the semigroup
reduced by Te(t) with C = 0 in (6. 35}. Then Te(t) - T(t) is not compact for
t < 2.
Indeed, a simple calculation shows that
{
tuodx-t), O::;x-t::; 1;
{[Te(t) - T(t)](UOI' U02, vo)}(x) = (0, -.XVo(t - x), 0::; t - x::; 1;
O. otherwise.
which is not compact for t < 2
,0).
Before ending this section, we study the system (6.16) in which there are
no dynamics on the boundaries. The purpose for doing this is to show that the
spectrum-determined growth condition holds for T4, i.e., Wess (A4) = S(A4).
If this is true, then we see that
wo(A) = max{S(A), wess(A)}
= max{S(A), Wess (A4)}
max{S(A), S(A
4)}, (6.36)
where the second equality follows from Theorem 6.3. This result clearly
shows that the growth rate of (6.1) is completely determined by the spectral
bound of A and A
4
, which serves as a basic tool for us to prove the spectrul1l-
determined growth condition for various wave equations to be discussed in
the next three sections.
Lemma 6.9 A4 has compact resolvent and o-(A4) = {A I h4().,) = O}, where
h
4
().,) is an exponential polynomial of the form:
m
h
4
().,) = L bke).,wk
k==l
for some integer m, some constants bk and some real Wk, k = 1,2, ... ,rn.
6.1 A General linear hyperbolic system 327
Proof. Given (f,g) E (p(O.l))n, we solve
(A - A4)(U,v) = (f.g).
that is,
{
.!!:... [u(x)] = -K-1(x)[A + Co(x)] [U(X)] + K-
1
(x) [f(X)] ,
dx v(x) v(x) g(x)
u(O) = Ev(O),v(l) = Du(l).
Denote by Y(x, y, A) the fundamental matrix of the system
Due to the special forms of K(x) and Co(x). we are able to obtain the explicit
expression of Y(x, y, A) as follows:
Y( A)=[Ydx,y,A) 0 ]
x, y, 0 Y
2
(x, y, A) ,
. r ds r Cii(S)
Y1 (x, y, A) = dmg(exp( -A i
y
Ai(S) - i
y
Ai(S) ds)),
. l
x
ds l
x
Cjj(s)
Y2 (X,y,A) =dlag(exp(-A -.(-) - -.(-)ds)).
y P,J s y P,J S
Then
= Y(x, 0, A) [ v(O) + 1
x
Y(x, y. A)K-
1
(y) [ dy.
Moreover, v(l) = Du(l) implies that
Ho(A)v(O) = 11 (D, -I)Y(l, y. A)K-
1
(y) ] dy,
where HO(A) = (-I, D)Y(l, 0, A) [ Therefore, A E (J(A4) if and only if
h4(A) = detHo(A) = O. (6.37)
Because the elements of Ho (A) are the linear combinations of the exponential
functions of the form e-Afo'1/A,(s)ds and e-
A
fo
11
/J1.j(S)dS, h4(A) is therefore
an exponential polynomial. For any A E p(A),
R(A, A4) (x)
[
EV(O)] r -1 [fry)]
Y(x, 0, A) v(O) + io Y(x, y, A)K (y) g(y) dy
(6.38)
328 Other Applications
with
(6.39)
It is readily seen in (6.38) that R(A, A4) is compact for any A E P(A4)' This
proves the lemma. D
Before proving the spectrum-determined growth condition for systcm
(6.16), we need several properties of almost periodic functions.
Recall that a continuous function g(x) (-C)O < x < ex;) is said to be
almost periodic if for every E > O. there is a quantity > 0 such that every
interval of the x-axis of length contains at least one number T such that
Ig(x + T) - g(x)1 < E for all x E (-C)O, C)O).
Consider an exponential polynomial
N
J(z) = 2::>ne
wnZ

n=1
where bn,w
n
are constants. Then g(x)
function for any fixed a E R (see [91]).
(Wn are real) (6.40)
J(n + tx) is an almost pcriodic
Lemma 6.10 (i) Let J(z) be defined by (6.40). Let n = {z I J(z)
o}. Then Jar all z satisJying dist(z, n) 2: 6 > 0, n :S Rez :S P and
IImzl 2: M, there exists a constant m(6, nJ3, M) > 0 such that IJ(z)1 >
m(J,a,{3,M), where a,{3,6, and M are constants.
(ii) For an exponential polynomial J(z) such as (6.40). iJ J(A) i- 0 Jar all
A = (J" + iT with fixed (J", then
with lakle"-'Yk < C)O, where /k are real, and ak are complex con-
stants.
Proof. (ii) is a generalization of the well-known Wiener's theorem on Fourier
series; its proof can be found in [130J. Here, we give a proof of (i). Suppose
otherwise, then there are an infinite number of Zrn with a :S Rezrn :S (3 such
that J(zm) -+ 0 as m -+ 00. Let Zm = nm+i{3m. Then IPml-+ C)O as m -+ oc.
We may assume, without loss of generality, that all Pm are positive. Consider
functions
N
rn(z) = J(z + ipm) = L
n=l
6.1 A General linear hyperbolic system 329
It is obvious that {m (z)} are uniformly bounded in {zla - 6/2 < Rez < (3 +
6/2}. By the Montel's theorem [38], there is a sub series of m (z), still denoted
by m(z), such that m(z) converges uniformly to an analytic function (z)
on {zla - 6/2 < Rez < (3 + 6/2}. Suppose, without loss of generality, that
am -+ ao as m -+ 00, a ao (3. Therefore,
On the other hand, since m (z) does not equal zero for z satisfying a -
6/2 < Rez < (3 + 6/2 and Imz -6 for sufficient large m by assumption,
must be identical to zero by the Hurwitz theorem [38]. It follows that
f(x) = m(x - i(3m) -+ 0 for all a - 6/2 < x < (3 + 6/2 and so f == 0 by the
analyticity of f, which is impossible since the distance between the zeros of
f and z with a Rez (3 and IImzl M is positive by assumption. This
proves the theorem. 0
Theorem 6.11 Let S(A
4
), Wess (A4) and WO(A4) be the spectral bound, es-
sential growth rate and growth rate of the Co -semigroup T4 (t), respectively.
Then
Proof. We first show that S(A4) = WO(A4). Let S(A4) = ao. It suffices to
show that for any a > ao, there is an M > 0 such that
for any W in a dense subset of XO. We shall use the following inverse Laplace
transform
(6.41 )
which holds for some (3 > 0 and all W E (see Theorem 2.40). We
choose those W which are sufficiently smooth such that they have compact
supports in (0,1).
For any a > ao, we claim that the path of integration in (6.41) can be
shifted to Re>.. = a by showing that
where Q(R) = {>.. I a Re>.. (3,IIm>..1 = R}. To this end, we have to
estimate the resolvent which is given in (6.38). Since the resolvent R(>.., A4)
consists of two parts, we estimate only the first part. The estimation of the
330 Other Applications
second part is simple. Write = where L(A) is the matrix whose
elements are composed of the algebraic cofactors of Ho(A). Then the first
part of R(A,A
4
) in (6.38) can be written as
Y(x, 0, A) [ L(A)(D. -l).Io
1
Y(l. y.
( 6.42)
By the expression of H 0 (A). the elements of [ ] L (A)( D, - I) takes the forlll
ae
bA
, where a and b are real. Therefore, the component of (6.42) takes the
form
(6.43 )
where b
1
, b
z
are C
1
functions and a is real. w is a component of W, c(:z:) =
fox p( s )ds, d(y) = I: q( s )ds, p, q are one of 1/ Ai and 1/ Pj. Since dist( A. 0-( A))
> t > 0 for some t > 0 and all A E n(R), it follows from Lemma 6.10 that
there is a positive constant J> 0 such that
(6.44 )
By (6.43) and (6.44), we can write (6.43) in a compact forlll
t
b(x. A) io
where b(x, A) is bounded for 0 :S .r :S 1 and n :S ReA :S (3. Integrating by
parts yields
which tends to zero as I A oc, n :S ReA :S (3. Thus we conclude that
Now, for AO = n + iT, by Lemma 6.10, we can write
1 x
h
4
(AO) = LakeAOl'k,
k=1
(6.45)
(6.46 )
6.1 A General linear hyperbolic system 331
where 'Yk is real and lakle1'k < 00. Then for A = AO, (6.43) is reduced
to
f: ake"'1'kb
1
(x)e-o(a+c(x)) t
k=l 10
00 11
= L ak
e
"'1'k d
1
(x) d2(y)eiTbk-a-c(z)-d(y))w(y)dy,
k=l 0
(6.47)
whered
1
(x) = b
1
(x)e-
o
(a+c(x)),d
2
(y) = b
2
(y)e-
od
(y) areC
1
functions. There-
fore, to estimate the integral of the right-hand side of (6.45), we have to
estimate
lim JR e"'t f: ak
e
1'kd
1
(x) t d2(y)eiT(t+1'k-a-c(x)-d(y))w(y)dy
-R k=l h
00 11 JR
= lim e"'t L ake"'1'kddx) d
2
(y)w(y)dy e
iT
(t+1'k-
a
-
c
(x)-d(y))dT
k=l 0 -R
00
= lim 2e"'t '"' ake"'1'k d
1
(x)
L....
k=l
X 11 d
2
(y)w(y) sin R[t + 'Yk - a - c(x) - d(y)) dy
o t+'Yk-a-c(x)-d(y)
00
- lim 2e",t '"' ak e"'1'k d
1
(x)
L....
k=l
x 1
d
(O) d
3
(z)w(7](z)) sin R[t + 'Yk - a - c(x) - z] dz
o t + 'Yk - a - c( x) - z
00 I
t
+1'k-
a
-
c
(X)-d(O) sinRz
hm 2e",t '"' ake"'1'kdt{x) Lk t x(z). --dz,
L.... " z
00 k=l t+1'k -a-c(x)
(6.48)
where Lk,t,x(z) = d3 (t + 'Yk - a - c(x) - z)w(7](t + 'Yk - a - c(x) - z)) is
a Cl function of z for any fixed (k,t,x) and d
3
(z) = d
2
(7](z))7]'(z) for z =
d(y),y = 7](z) is also C1 with respect to z. Now, if t + 'Yk - a - c(x) and
t + 'Yk - a - c(x) - d(O) have the same sign, then by integrating by parts, we
have
I
t+1'k-a- C(X)-d(O) . R
. sm z
hm Lk,t,x(z)--dz = o.
t+1'k -a-c(z) Z
If t + 'Yk - a - c( x) = 0, then by using the fact that J0
8
sin
z
Rz dz = 11"/2
for any positive 8, we have
l
-
d
(O) sin Rz
lim Lk,t,x(z)--dz
0 Z
332 Other Applications
l
-
d
(O) sin Rz l-
d
(O) sin Rz
lim [Lk t ,,(z) - Lk t ,,(O)J--dz + Lk t ,,(0) --dz
R-+oo 0 ' , , , z ' , 0 Z
. l-
d
(O) sin Rz 7r
= hm Lk t ,,(0) --dz = -Lk t ,,(0).
R-+oo " 0 z 2' ,
If t + 'Yk - a - c(x) and t + 'Yk - a - c(x) - d(O) have the opposite sign, then
. I
t
+1'k -a-c(,,) -d(O) sin Rz
hm Lk,t,x(z)--dz = 7rLk,t,,,(O).
R-+oo t+1'k -a-c(,,) Z
Therefore, the limit of the integral of the right-hand side of (6.48) is bounded
by 7r 1 d
3
(t + 'Yk - a - c(x))w(ry(t + 'Yk - a - c(x))) 1 as R tends to 00. Here
we set w(ry) = 0 for ry outside [0, 1J. Since
Id
l
(x)d
3
(t+ 'Yk - a- c(x) )w(ry(t+'Yk - a - c(x)))1 :S Mlw(ry(t+ 'Yk - a - c(x)))1
for some M which is independent of t and k, we have
IIddx )d3 (t + 'Yk - a - c(x) )w(ry(t + 'Yk - a - c(x)) )IILP(O,l)
< M(l
l
I w(ry(t + 'Yk - a - c(x))) IP dx)l/p
I
t+1'k -a-c(l) 1
= M( 0 -ry'(d(z))c'(x) I w(z) IP dz)l/p
< MllwIILP(O,l)
where M is a constant independent of t and k. Finally, it is easily seen that
the integrals on the right-hand side of (6.48) are uniformly bounded with
respect to k; hence, the limit in R and the summation in k are commutable.
Therefore,
I
R 00 1
lim eDtt L akeDt1'k d
l
(x) 1 d
2
(y)e
iT
(t+1'k -a-c(,,) w(y)dy
R-+oo -R k=l 0
00
= eDtt L ak
eDt
1'k pk ,t(x),
k=l
where Ilpk,tfx)IILP is bounded uniformly by IlwIILP(O,l) for all t and k. Sum-
marizing, we have proven that for any a > ao, there exists a constant M > 0
such that IIT4(t)1I :S Me
Dtt
. Thus, WO(A4) = ao
Furthermore, we claim that there is a sequence An such that h4 (An) =
0, ReA
n
-+ ao, and IAnl -+ 00. Suppose it is not true. Then by the definition
of ao, there exists a Al such that h(Ad = 0, ReAl = ao. On the other hand,
the assumption implies that there is a > 0 such that for all A with IAI
6.2 Stabilization of serially connected vibrating strings 333
sufficiently large on the vertical line ReA = 0:0, I h4(A) by (i) of Lemma
6.10. This, together with h(Ad = 0, contradicts that h4(0: + iT) is almost
periodic with respect to T. Therefore, the assertion is true.
Suppose, without loss of generality, that ReAi f:. ReAj for i f:. j. From the
spectral mapping theorem and Lemma 6.9, it follows that {e
Ant
} C a(T4(t)).
However, e
Ant
are distinct for different n and e
Ant
-+ eo
ot
Hence eo
ot
E
a
ess
(T4(t)). Consequently eO
ot
S res.(T
4
(t)) and so 0:0 S W
ess
(A4)' This
together with 0:0 = completes the proof. 0
6.2 Stabilization of serially connected vibrat-
ing strings
In this section, we consider (N + 1 )-serially connected vibrating strings with
the following dynamic model [27]
Ytt(x,t) - cry",,,,(x,t) = 0, i-I < x < i,i = 1,2, ... ,N,N + 1, (6.49)
where y(x, t) denotes vibration magnitude of the connected string at time t
and position x, and Ci represents the wave speed of the i-th string. Notice
that, for notational simplicity, we have assumed that the length of each seg-
ment of the connected string is 1. At the left end, x = 0, and the right end,
x = N + 1, the following conservative conditions
y(O, t) = 0 or y",(O, t) = 0; y(N + 1, t) = 0 or y",(N + 1, t) = 0 (6.50)
or dissipative boundary conditions
y",(O, t) = t), y",(N + 1, t) = + 1, t), (6.51)
0
are imposed.
At the i-th intermediate node x = i, i = 1,2, ... ,N, either force feedback
dissipative conditions
{
Yt(i-,t) -Yt(i+,t) = -krcry.,(i-,t),
cry.,(i-, t) - CT+Iy.,(i+, t) = 0
or velocity feedback conditions
{
y(i-,t) -y(i+,t) = 0,
(i-, t) - cr+lY" (i+, t) = -krYt (i+, t)
are assumed.
(6.52)
(6.53)
334 Other Applications
The energy of the system is defined hy
1 N+l Ii
E(t) = - 2:= [C;Y;(:I:. t) + y;(x. t)]d:z:
2 i=1 i-I
and formally, it is easy to show that
dE(t)
dt
N
+ 2:= [c;Yx (i- ,t)Yt(i-. t) - c7+1Yx(i+,t)Yd i+. t)] :S O.
i=1
(6.54)
(6.55)
Since we are concerned with the energy dissipation of system (6.49) -
(6.53), we wish to define a sui table state space in which the norm of the state
variable (y, Yt) is equivalent to the system energy E( t) defined in (6.54). De-
note Xo by Xo = span{ o (x)}. where o (x) is the solution of the equilibrium
of system (6.49)-(6.53), i.e., it satisfies
Let
= O. i - 1 < x < i.
o(O) = O. or = 0: o(N + 1) = 0 or + 1) = O.
if (6.50) is imposed,
= O. + 1) = 0 if (6.51) is imposed.
= ri(i+) = 0 if (6.52) is imposed,
x {(,'ljJ)i ((x),'Ij!(x)) E Hl(i - Li) X L2(i - Li).
for x E (i - 1, i). i = 1,2."" N + L
(O) = O. or (N + 1) = O. if (6.50) is assumed.
(i-) = (i+). if (6.53) is assumed.}
In the state space XI XO which denotes the quotient space. system (6.49)-
(6.53) can be written as an evolution equation:
!i[Y]-B[V]
dt Yt - YI'
where
[
(x) ] [1jJ(x)] . .
B 1jJ ( x ) = c; I! ( X ) I - 1 < x < I
6.2 Stabilization of serially connected vibrating strings 335
with the domain D(B) being defined in an obvious way from the boundary
conditions (6.50)-(6.53). Since ::; O. B is dissipative. Moreover. B has
no zero eigenvalues in XI XO. It can be shown easily that B-
1
exists and
hence B generates a Co-semigroup of contractions on XI XO by virtue of the
Liimer-Phillips theorem.
Now make a linear bounded transformation P: XIX
o
-+ (L2(0. 1))2N+2
by
where
and
[
U] = P [Yx ]
v Yt
u(x, t) = COI[U1 (x. t). U2(X. t) . .. uN+1 (x. t)l,
v(x, t) = col[vdx, t). V2(X, t) .... , VN+l(X. t)l,
[
Ui(X, t)] = [ (-l)i
ciYx
(Pi(x). t) + Yt(Pi(X). t) 1
Vi(X, t) 2 -( -l)'CiYx(Pi(x). t) + Yt(Pi(X), t)
(6.56 )
with Pi(X) = (-1)i(1/2 - x) + i -1/2. Then P is a one-to-one mapping and
the norm of [ in H is equal to the energy E(t) of system (6.49)-(6.53).
We shall show that satisfies the evolution equation (6.16), and hence
there exists a Co-semigroup T(t) such that
[ ] = T(t) [ ] = T(t)P [ ] ,
where Yo = Yx(x, 0). Y1 = Yt(x.O) are the initial values of (6.49) - (6.53).
Thus.
E(t) = II [ ] II = IIT(t)P ] II ::; IIT(t)1111P111I [ ] II
which means that the energy E( t) decays asymptotically (or exponentially)
to zero when T( t) is asymptotically (or exponentially) stable. In particular,
when P maps XI XO onto H. then p-
1
is bounded.
] = P-1T(t)P ] = S(t) ]
and S(t) is a Co-semigroup on XI XO. In this case. the stability of T(t)
is equivalent to the stability of S(t). Unfortunately. p-
1
may not always
be bounded. The reason for this is that. in making the transformation. we
336 Other Applications
actually used the boundary condition ydi-,t) = Yt(i+,t), instead of the
original condition y( i-, t) = y( i+, t). Although the latter guarantees the
fonner, the converse is not always true. This causes the trouble that P is
not always onto. In the next section. we shall explain how to deal with the
case where P is not onto. In general, we have to reduce the state space H in
order to assure that P is onto.
We now continue to write (6.49)-(6.53) as the form (6.16) ill the previous
section, by using the transformation (6.56).
First, by using (6.56), (6.49) can be written as the following first order
system on H:
- + - =, <x<
a [U(X', t) ] K a [U(X, t) ] 1
at v(x. t) ax v(x, t)
(6.57)
where K = diag(cl .... ,CN+l,-Cl, ... ,-CN+d which is a special case of
(6.16) in Section 6.1. In order to arrange the boundary conditions in the
form of (6.16), we need to express Ui(O, t) in terms of v;{O, t), and v;{l. t) in
terms of ui(l, t). From condition (6.50), we have
or
{
vdO, t) = udO, t),
VN+l(1,t) = uN+l(I,f),
uN+l (0, t) = vN+dO, f).
and from (6.51) we have
Ul (0, t) = VI (0, t).
(1)
l-k;'tdCNtl (1 f)
vN+l ,t = l+k2! uN+l'"
Ntl CNtl
N + 1 is odd,
N + 1 is even.
N + 1 is odd,
N + 1 is even.
N + 1 is odd,
( )
l-k;'tl!CNtl (0)
UN+l 0, t = l+k2 / VN+l. t ,
N+l CN+l
N + 1 is even.
When i is odd, condition (6.52) implies that
(6.58)
(6.59)
(6.60)
{
[vi(l, t) + ui(l, t)l- [vi+d1, t) + ui+d1, t)l = -k;C;[Vi(l, t) - 'u;(l. t)l.
C;[Vi( 1, t) - u;( 1, t)l = ci+dui+d 1, t) - vi+d 1, t)l,
or in matrix form
[
-I-k
2
c, 1 ] [ vi(l,t) ] _ [l-k;Ci -1 ] [ u;(l,t) ]
-Ci" -Ci+l Vi+l(l,t) - -Ci -Ci+l ui+d1,t)'
6.2 Stabilization of serially connected vibrating strings 337
Likewise. condition (6.53) implies that
{
v;(l, t) + Ui(1. t) = Vi+1(1, t) + lLi+dl, t)
c;[vi(l, t) - ui(l, t)]- ci+dui+dl, t) - vi+d1, t)] = -kT[vi(l, t) + ui(l, t)],
or in matrix form
[
-1 1] [ Vi (1. t)] [1 -1] [ Ui ( 1, t) ]
-kr-Ci -Ci+1 Vi+1(1,t) = kr-Ci -Ci+1 lLi+1(1,t)'
Similarly, when i is even, conditions ( 6.52) and (6.53) lead, respectively, to
and
Now, introduce matrices Ei and Di as
{
E, = [ 1 -1 [ 1
D, = [c, 1 k; ;+1
1
] -1 [ c,
Then we can rewrite (6.52) and (6.53) in a compact form as
{
[
vi(l,t) ] _ E. [ ui(1,t) ] .. dd
( )
-, ()' t 18 0 ,
Vi+1 1, t Ui+l 1. t
[
Ui(O,t)] E [ Vi(O.t) ] .. . N
ui+dO, t) = i vi+dO. t) ,'t 1S even, 1:::; t:::; ,
and
{
[
vi(1,t) ] -D. [ ui(1.t) ] " dd
( 1 )
-, (). t 1S 0 ,
Vi+1 ,t Ui+1 1, t
Ui(O, t) v;(O, t) . . .
[Ui+1(0, t)] = Di [Vi+l(O, t)] , t 1S even, I :::; t :::; N,
respectively. Thus the boundary conditions can be written as
U(O, t) = Ev(O, t). v(l, t) = Du(1, t),
(6.61)
(6.62)
(6.63)
where E and D are constant matrices constructed, in an obvious way, from
Ei and Di as well as (6.58)-(6.60).
338 Other Applications
Summarizing, we have successfully transformed the equation describing
the dynamics of the serially connected strings and its boundary conditions
into the form of the hyperbolic system (6.16) in Section 6.1. as is evident from
(6.57) and (6.63). Let us put them together for the convenience of referenc('s.
{
0 .[ u(x, t) ] + K. 0 [u(:r, t.l ] = (J,
Ot v(x. t) lfX u(x, t)
u(O, t) = Eu(O, t),
u(I. t) = Du(l. t).
0<:1'<1.
(6.64)
Now, define the operator A ; D(A) c H -+ H by
{
A[U] -_KO [u]
v- lfXv'
(6.65)
D(A) = { E (H2(0, 1))2N+2 I u(O) = Ev(O), v(l) = DU(l)}.
With this setting, system (6.64) can be rewritten as an abstract equation on
the state space H = (2(0, 1))2N+2;
[U(x.t)] _ A [U(X,t)]
Ot u(x, t) - v(x, t) .
(6.66)
The operator A defined above has the following properties.
Theorem 6.12 (i) A is an infinitesimal generator of a Co-semigroup of
contractions on H.
(ii) The resolvent operator R(>", A) is compact (whenever it is well defined),
and is expressed as
R(>.., A) = e-
AK
-
'x
[ v(O) (6.67)
+ [X e-AK-1(x-s) K-
1
[ f(s) ] ds, (6.68)
Jo g(s)
where
v(O) = H-1(>")(D, -I) .1 e-
AK
-
1
(1-S) K-
1
[ ] ds,
and H(>..) = (-DJ)e-
AK
-
1
[ , for any [ ;] E H.
(iii) The spectrum-determined growth condition holds:
wo(A) = we,,,(A) = S(A).
6.2 Stabilization of serially connected vibrating strings 339
(iv) a(A) = {Alh(A) = O}, 'Where h(A) is an exponential polynomial h(A) =
detH(A). Moreover, corresponding to any A E a(A), there exists an
independent eigenfunction
'Where y is any vector satisfying H(A)Y = O. In particular, if 0 E a(A).
then <Po = [ y for any y satisfying (D, -I) [ y = o.
By employing the result in the previous section, we can prove very simply
the following theorem which is obtained in [94] in a different way.
Theorem 6.13 Let h(A) be defined as in Theorem 6.12. Then system (6.66)
is exponentially stable if and only if
inf{lh(iw)l, wE R} > O. (6.69)
Proof. If system (6.66) is exponentially stable. then it is necessary that
S(A) < O. This implies that the distance between the set {iw, wE R} and
the set {.x I h(A) = O} does not equal zero. In view of (i) of Lemma 6.10,
Ih(iw)1 is bounded below from zero, i.e., (6.69) holds. Conversely, if (6.69)
holds, then from (6.68), IIR(iw, A)II is uniformly bounded for w E R. Since
Ile
At
II 1, by the frequency domain criteria of exponential stability discussed
in Chapter 3 (Corollary 3.36), it follows that eAt is exponentially stable. The
theorem is thus proved. 0
So far, we have discussed the exponential stability. We now prove an
asymptotic stability result and, furthermore, we establish a relationship be-
tween the exponential stability and asymptotic stability.
Theorem 6.14 The semigroup eAt generated by A is asymptotically stable
if and only if ReA < 0 for all A E a(A), or equivalently,
(a). 0 a(A);
(b). ReA < 0 for all (A,),A:f. O,:f. 0 satisfying
A2(x) - Cf"(x) = 0, i -1 < x < i, i = 1,2, ... ,N,N + 1,
(O) = 0 or '(O) = 0; (N + 1) = 0 or '(N + 1) = 0,
{
A(i-) - A(i+) = -krc;'(i-), 'f (652)" d (6.70)
2'!.,I( ._) _ 2 ,1..'( .+) _ O. t. zs zmpose ,
ci'l' t ci+l'l' Z -,
{
(i-) - (i+) = 0, 'f (6 53)" d
2,1..'('-) 2 ,1..'('+)_ k2\,I..(+) t . tstmpose.
ci 'I' Z - Ci+ 1 'I' Z - - i "'I' Z ,
340 Other Applications
Proof. Since A has compact resolvent. it follows from Theorem 3.26 that
eAt is asymptotically stable if and only if ReA < 0 for all A E 0-( A). If
(A. ). A f. o. f. 0 solves the eigenvalue problem (6.70). then let
= - '. V.r E [0.1].
[
1t
i
(l:)] 1 [ (-l)i
ci
'(p;(x)) + A(p;(X)) 1
Vi(:D) 2 -(-l)'ci(p'(Pi(:D)) + A(P(Pi(X))
(6.71 )
where p;{x) = (-1)'(1/2 - :1;) +i - 1/2 is defiued before. Since (p;{x)) =
Hlti(X) + Vi(X)]. so ui(:r). vi(.r).1: = 1. 2 ..... N + 1 cannot be identically zero
and
is the eigenvector of A corresponding to A . Conversely, if (u. v) is the eigen-
vector of A corresponding to A. A f. O. then let
1
(pi(X)) = + vi(:r)]. O:SX:Sl.
Since 1ti(X) = u;{OjeA/cix.vi(x) = vi(O)e-A/c,x.i = 1.2, ... N + 1 cannot be
zero identically. ( x). x E [0, N + 1] defined above cannot be zero. and it can
be verified easily that (A. ) solves (6.70). 0
Theorem 6.15 If the semigroup generated by A is asymptotically stable,
and in addition. Ci,1: = 1. 2, ... N + 1 are r:ommensurable, i.e., there is a
constant c such that Ci = c ni with ni being some integers, or equivalently.
cl/ci=rational for 1: = 1,2 .... N + 1. Then the semigroup is exponentially
stable.
Proof. We note from TheoremG.12 that the exponential polynomial h(A) =
L:;;:'1 bke
AWk
where Wk = l/ci. Since ei. i = 1. 2, ... N + 1 are com-
mensurable, so does 1/ Ci; hence, there is a constant c f. 0 such that 1/ Ci = cni
for some integer ni. In this case. Wk = l/ci = eN/<, where Nk is an
integer. Therefore. h(iw) = bkeiwcNk for wE R. h(iw) is then periodic
with respect to w with period 27l' / c. Consequently.
inf{lh(iwll. wE R} = inf{lh(iwll. wE [0, 27l'/c]}.
However. the asymptotic stability just proved implies that the right-hand
side above is positive. thus eAt is exponentially stable by Theorem 6.13. 0
6.3 Two coupled vibrating strings 341
6.3 Two coupled vibrating strings
In this section, we consider two coupled vibrating strings with a force stabi-
lizer in the middle point which is a special case of the problem in the previous
section.
For notational convenience, suppose that each segment of the coupled
string has unit length. Let y(x, t) denote the vibrating magnitude of the
string for x E [0,2]. Then the dynamics of the vibrating string can be written
as
mlYtt(x, t) - T1yxx(x, t) = 0, x E (0,1), t > 0,
m2Ytt(x, t) - T2yxx(x, t) = 0, x E (1,2), t > 0,
y(O, t) = y(2, t) = 0,
y(1-,t) = y(1+,t),
T
1
Yx(1-,t) -T
2
y,,(1+,t) = -kYt(1.t),
y(x,O) = yo(x),Yt(x,O) = ydx),
k > 0,
(6.72)
where Ti and mi, i = 1,2, denote the tension constants and mass densities
per unit length on the i-th string, respectively. The constants Ci = VTdmi
correspond to the wave speeds. Here, we consider the case where both ends
of the string are fixed, i.e., y(O, t) = y(2, t) = 0. We call this symmetric
boundary conditions. The cases such as y(O, t) = Yx(2, t) = will be called
asymmetric boundary conditions.
We show that the solution is asymptotically stable if and only if Cl / C2 is
irrational, but the solution is not exponentially stable regardless of the value
of Cl / C2. We prove these main results by introducing the linear bounded
transformation P, as mentioned before, to transform (6.72) into the gen-
eral linear hyperbolic system in Section 6.1. However, even for this simple
problem, the variable transformation P is not an onto mapping. Thus, the
original system and the transformed system is not equivalent. In fact, we
will see that zero is an eigenvalue of the transformed system. but the origi-
nal system does not contain a zero eigenvalue. By excluding the eigenspace
corresponding to the zero eigenvalue from the state space, then P becomes
a one-to-one mapping.
We first observe that the energy in system (6.72) is dissipative. That is,
if we denote by
1 t[ 2 2
E(t) = "2 io mlYt (x, t) + T1yx(x, t)]dx
1 {2[ 2 2
+"2 il m2Yt (x, t) + T2y",(x, t)]dx
(6.73)
342 Other Applications
the total energy of the system, then formally
dE(t) 2
-- < -I,; (1 t)
dt - Yt '
along the solution of (6.72). Furthermore, we consider (6.72) in the state
space X = HJ (0,2) X L2(0. 2), which is a Hilbert space when equipped with
the inner product induced norm
It is easy to see that
Now define a transformation P : X -+ (LZ(O, 1))4 by
udx, t) = t) + 'Ydx, t)l,
uz(x, t) = tJffi2[cz'Yx(2 - x, t) + 'Yt(2 - x, t)l,
vdx, t) = t) + Yt(x, t)l,
vz(x, t) = - x, t) + Yt(2 - x, t)l.
(6.74)
Using this, we can write (6.72) as the following abstract equation on H =
(L2(0, 1))4:
[U(X,t)] _ A [U(X,t)]
at v(x, t) - v(x, t) ,
(6.75)
where
(6.76)
K = diag( jT;, vr;, - jT;, - vr;),
D(A) = { [ E (Hl(O, 1))4 I u(O) = Ev(O), v(1) = DU(1)}
with
E = [-1 0]
o -1
and
1 [ .jrn1T1 - .jrnZT2 - k 2JrnlT2 1
D = VrnlTl + VrnZT2 + k 2 JrnZT
1
JrnzT
z
- Jrn1T
1
- k .
6.3 Two coupled vibrating strings
343
The energy E(t) can be alternatively expressed as
E(t) = II [ ] I[ .
Lemma 6.16 Let A be defined as -in (6.76). Then E a(A). But 0 is not
an eigenvalue of system (6.72).
Proof. It is seen from Theorem 6.12 that any eigenfunction of A corre-
sponding to is a constant vector <Po, which is found to be
(6.77)
This shows that is an eigenvalue of A. However, if there is a nonzero
satisfying the following eigenproblem of system (6.72):
mlAZ(x) - T1"(x) = 0, x E (0,1),
mZAZ(x) - Tz"(x) = 0, x E (1,2),
(o) = (2) = 0, (6.78)
( 1 -) = ( 1 +),
T
1
'(1-) - T
z
'(l+) = -kA(l), k > 0,
then A satisfies the characteristic equation
(a + b + k)e(1+d)J.' + (b - a - k)eJ.' + (a - b - k)edJ.' - (a + b - k) = 0,
(6.79)
where
Tl Tz Cl A
a= -, b= -, d= -, p,=2-.
Cl Cz Cz Cl
Since A = does not satisfy this equation, is not an eigenvalue of system
(6.72). 0
System (6.72) provides an example that the original system (6.72) is not
equivalent to the transformed system (6.75). The connection between system
(6.72) and (0.75) is, however, obvious in the sense that they have the same
nonzero eigenvalues. This motivates us to consider the following modified
state space
H = {W E (L2(0, 1))4 I (W, <Po) = O}. (6.80)
H is a Hilbert space with the inner product of (LZ (0, 1))4. Denote by AH =
AIH the restriction of A on H. It is observed that when W E D(A), then
(AW, <Po) = 0, i.e., AD (A) c H. Thus,
AHW = AW, 'lfW E D(AH), D(AH) = D(A) nH. (6.81)
344 Other Applications
Hence, A is completely reducible by H ane! span{ po}, and AH is a semigroup
generator on H. It is seen that A and AH have the same nonzero spectrum.
Moreover, it can be readily shown that the transformation (6.74) is a ane-
ta-one transformation from HJ (0. 2) X L2 (0. 2). the original state space for
system (6.72), to H. Indeed, for any (cp,'ljJ) E HJ(O, 2) x L2(0, 2), let
(6.82)
Then it is easy to show that
(6.83)
That is, (Ul,U2,Vl,V2) E H. Conversely, for any (Ul,U2,Vl,V2) E H, there
exists a (cp, 'IjJ) E HJ (0, 2) X L2 (0. 2) such that (6.82) is satisfied.
Based on this observation, we see that the stability of
[u(x, t) ] = AH [U(X, t) ]
at v(x, t) v(x, t)
(6.84)
on H is equivalent to the stability of (6.72) on X.
We now concentrate on analyzing the stability of (6.84).
Theorem 6.17 Wess(AH) = WO(AH) = S(AH) = O. Therefore, system
(6.84) is not exponentially stable.
Proof. For the operator A defined in (6.76) which shares all the properties
stated in Theorem 6.1, since 0 E alA), we have wess(A) = wo(A) = S(A) = O.
By the proof of Theorem 6.11, we know that there are An E alA) such that
ReA
n
-+ S(A) = 0 and IAnl -+ 00 as 11, -+ 00. The conclusion follows. 0
The asymptotic stability of (6.84) depends on the ratio of the two wave
speeds.
Theorem 6.18 All eigenvalues of AH satisfy ReA < 0 if and only if d =
CdC2 is irrational. Consequently, system (6.72) is asymptotically stable if
and only if Cl / C2 is irrational.
Proof. Since the system is dissipative. ReA :::; O. As we have seen in The-
orem (6.14), ReA < 0 if and only if there does not exist a nonzero solution
of (6.78) for any iw, W E R. Suppose there is awE R such that A = iw
satisfies (6.78) with some cp -:j:. O. We show that CdC2 must be rational. Mul-
tiplying the first two equations of (6.78) by 1>( x), where the overbar denotes
complex conjugate, integrating the resultant equations over [0,1] and [1,2].
6.3 Two coupled vibrating strings 345
respectively, adding the two expressions and using the boundary conditions
in (6.78), yield
kAI(1)1
2
+ 11 [T11'(xW + ml
A2
1(xWldx
+ 12 [T21'(xW + m2A21(x)121dx = 0, (6.85)
where A = iw. Letting the imaginary part of (6.85) be equal to zero gives
(1) = o. When A = iw and (1) = 0, we can solve the differential equations
in (6.78) directly to get
{
AI sinwx/cl, x E (0,1),
(x) =
A2 sinw(x - 2)/C2, x E (1,2)
(6.86)
for some constants Al and A
2
Again, by using the fact that (1) = 0, we
see that
(6.87)
We claim that both Al and A2 are not zeros. Indeed, if Al = 0, then (x) = 0
for all x E [0,11. Since we have assumed iw is an eigenvalue, (x) cannot
be identically zero in x E [1, 2], so A2 =I- O. By (6.87), sin W / C2 = 0 and so
COSW/C2 =I- o. However, T
2
'(I+) = w/c2T2A2cosw/c2 = T1'(I-) = 0 leads
to a contradiction that A2 = o. Hence, Al =I- o. Similarly, we can show that
A2 =I- o. Then (6.87) implies that W/CI = q7f and W/C2 = p7f for some integers
p and q. Therefore, d = cI/C2 = p/q is rational.
Conversely, if d = CI/C2 = p/q for some coprime positive integers p and
q, let W = clq = C2p. It can be shown easily that
An = inw7f = iClqn7f, n = 1, 2, ...
satisfy (6.79). The proof is complete.
o
When zero is not an eigenvalue of the operator A, a different method is
required to discuss the stability. The following example demonstrates how
to treat such problems using a general approach. Consider the following
equation with asymmetric conditions:
mlYtt(x, t) - T1y.,.,(x, t) = 0, x E (0,1), t > 0,
m2Ytt(x, t) - T2y",,,, (x, t) = 0, x E (1,2), t > 0,
y(O, t) = y",(2, t) = 0, (6.88)
y(I-,t) = y(I+,t),
T1 y",(I-,t) -T2 y",(I+,t) = -kYt(1,t), k > o.
346 Other Applications
The characteristic equation of (6.88) is
(a + b + /,;)e(l+d)lt + (Il - b + k)ell + (Il - Ii - /';)e
d1t
+ (a + b - k) = 0
(6.89)
where a. band 11 are the same as those defilled in (6.79).
Parallel to Theorem 6.18. we have the following proposition.
Proposition 6.19 System (6.88) is equivalent to its tran$formed system
under the transformation r defined in (6.74). Moreover. system (6.88) is
asymptotically stable if and only if d = (;1/(;2 is irrational.
To show that system (6.88) is not exponentially stable. we need the fol-
lowing lemma.
LeIllIlla 6.20 For any 'irrational d. there is an infinity of fractions Pn/ qn
satisfying
I
]J" I 12
d- - <?'
(j" f};,
where]Jn is odd, f}n is even. and]J" and Cj" lLre coprime.
Proof. We consider only the c:asp when d > O. The proof for d < 0 is
similar. For irrational d > O. 2d also is irrational. From the well-known
results of irrational number approxilllations by rational numbers [73]. we
kllOW that there is an infinity of fractiomi P,J f}" which are coprime and satisfy
I
]in I 1
2d- - < 2'
(j" f}n
If]Jn is odd. then Id - ]In/(2f}n)1 < 2/(21],,)2. If Pn is evell. since Pn and f}n
are coprime, there exist integers fn. g" sl1ch that Ip,J" - f}"gn I = 1. We may
assume. without loss of generality. that 0 < Ifni < If}nl. since if ifni> f}n, we
can write Ifni = a,,(jn + bTL for some integers an' bn with 0 < bTl < (jn- Hellce.
IPnb" - q"c,,1 = 1 for some integer Cn . AS]Jn is even. gn m11st be odd. Noting
that qn + f n :S 2f}n implies 1/ f}n :S 2/( f}" + fn)' we have
12d _ Pn + g" I
<
f}n + In
12d _ ]In I + l]Jn + g"
f}" Ijn + In
Pnl
Ijn
<
+ l]Jnf" - q"g" I
q;, q"(q,, + fn)
1 1 6
< -+ <
(qn + In)2'
q;, (j" (q" + In)
(6.90)
Hence,
+
12
I
d p" g" I < ----,,-
2(q" + In) [2(q" + 1,,)j2'
6.4 A vibration cable with a tip mass 347
Because I(Pn + gn)ln - (qn + In)gnl = 1, Pn + gn and qn + In is coprime.
Since Pn + gn is also odd, Pn + gn and 2(qn + In) are coprime. Because qn
are infinite in number, so are 2(qn + In). Hence, by (6.90) Pn + gn are also
infinite in number. Renaming Pn + gn by Pn and 2(qn + In) by qn in (6.90)
yields the inequality in the Lemma. 0
Using Lemma 6.20, we are in a position to prove the following lemma.
Lemma 6.21 For any irrational d = CI/C2, there exist Wn E R such that
Iwnl -+ 00, e
iwn
-+ 1 and e
idwn
-+ -1 as n -+ 00.
Proof. By Lemma 6.20, d = fu + ft with Icnl :::; 12, qn being even and Pn
qn qn
being odd. Take Wn = 7rqn' Then e
iwn
= 1 and
This is the result required. o
Theorem 6.22 For any irrational d = CI/C2 > 0, there exists an infinite
number of solutions to (6.89) with ReA
n
-+ 0 as n -+ 00. Hence, system
(6.88) is not exponentially stable.
Proof. Suppose this is not true. Then there exist an M > 0 and E > 0
such that II(iw)I > E for alllwi M by Lemma 6.10 (i), where I(p,) denotes
the left-hand side of (6.89). But for Wn in Lemma 6.21, a direct computation
shows that
II(iwnll-+ 0 as n -+ 00,
which is a contradiction. 0
6.4 A vibration cable with a tip mass
In this section, we consider a pinched vibration cable with a tip mass whose
motion is described by the following partial differential equation:
Ytt(x, t) - y",,,,(x, t) = 0, x E (0,1),
y(O, t) = 0,
yx(l, t) + mYtt(l, t) = -u(t),
y(x,O) = Yo(x), ytfx, 0) = Yl (x),
(6.91)
where m > 0 is the tip mass and u(t) is the boundary control force applied
at the free end and (Yo, yd is the initial condition. Our purpose is to find a
348 Other Applications
feedback control law u(t) so that the energy E(t) of the resultillg dosed-loop
system decays uniformly to 2ero. both for the cable deflection and the motion
of the tip mass.
One of the natural control is the liuear feedback control law:
u(t) = cYYt (1 , f). n > n.
by which the closed-loop system will be
{
Ytt(x.tJ-Yxx(:r,t)=O.
y(O, t) = D.
Yx(l, t) + rnYtt(l. t) + aYt(1, t) = O.
(6.92)
(6.93 )
Let V = {u I u E HI(O, 1).u(O) = D}. Consider system (6.93) in the state
Hilbert space H = V X 2(0,1) X R with the inner product
Introduce a linear transformation P:
T 1. I 1 I T
P(u(x). v(x). el) = (2"(o(:r) - u (x)). 2"(v(x) + u (:r)), el) . (6.94)
It is easy to see that P is a unitary mappillg from H to (2(0,1))2 X R. Set
[
u(x, t) 1 [ y(.1:. t) 1
0(:, t) = P t) .
Then system (6.93) is transformed into an equivalent forlll under P:
!
[) [U(X,tl]+K[) [1t(.1;,t)] =0,
(ft u(x, t) ax o(:r, t)
:it [1,'( + u( 1, tll = 1, t) - 1, t),
H(O. t) - -ufO, t).
(6.95 )
which is of the form of (6.1) with N = 1. K(x)=diag(l, -1), C(:rJ = 0, D =
-1. F = (1- aJ/rn. G = -(1 + a)/rn and E = -1. Now the reduced system
(6.16) becomes
{
[) [U(X, t) ] K [) [u(X, t)] - 0
(ft v(x. t) + ax v(x, t) - ,
v(L t) = -1/.(1, t). H.(O. t) = -ufO. t).
(6.96)
6.4 A vibration cable with a tip mass 349
Under p-1 defined by (6.94), system (6.96) is equivalent to the undamped
vibration system
{
Ytt(x, t) - y",,(x, t) = 0, x E (0,1),
yeO, t) = 0, y(l, t) = 0.
(6.97)
It is well known that all the eigenvalues of this system lie on the imaginary
axis; hence, system (6.96) is not asymptotically stable. Therefore, there is
at least one essential spectrum z, with Izl = 1, of the associated semigroup.
From results in Section 6.1, system (6.95) is a compact perturbation of system
(6.96); hence, it is only asymptotically stable (all its eigenvalues locate strictly
on the left-half complex plane), but not exponentially stable. Actually, let
Wess denote the essential growth rate of system (6.95). Then
WeBS = 0.
(6.98)
To achieve exponential stability, we propose another linear feedback control
law:
u(t) = aY"t(1, t) + aYt(l, t), a> 0, a> 0.
Substituting u(t) into (6.91) yields the closed-loop system equation:
{
Ytt(x,t) - y",,(x,t) = o,x E (0,1),
yeO, t) = 0,
y,,(l, t) + mYtt(l, t) + aY"t(1, t) + aYt(1, t) = 0.
Define the energy function E(t) by
(6.99)
(6.100)
111 1 1
E(t) = - t)+y;(x, t)]dx+ - [ay,,(l, t)+mYt(l, t)]2. (6.101)
2 0 2 m + aa
It is easy to verify that
dE(t) <
dt -
along the solutions of (6.100). On the other hand, using the transformation
P defined by (6.94) again, (6.100) is equivalent to the following system
1
8 [U(X, t) ] + K 8 [U(X, t) ] =
at vex, t) Ox vex, t) ,
9t[v(l, t) + t)] = u(l, t) - v(l, t),
u(O, t) = -v(O, t)
(6.102)
350 Other Applications
011 the state space H = (L2(0, 1))2 X R which is a Hilbert space with the
inner product induced 1l0rIll:
Let
W(t) = v(, t)
[
It(.t) ]
Yt(,t)+y",(,t))/2 .
[
(Yt(" t) - Yx(', t))/2 ]
v(l. t) + t) (mYt(1, t) + ay", (1. t))
Then (6.102) can be written as
{
W(t) = AW(t)
W(O) = Wo
(6.103)
with A being defined in an obvious way. Clearly, E(t) = t IIW(t)1I
2
. Hence,
E(t) decays exponentially if and only if IIW(t)II decays exponentially. We
show that IIW(t)1I decays exponentially by using the spectrum-determined
growth condition. For this purpose, we need to consider the following reduced
8ystem of (6.102)
{
a [U(x. t) ] K a [U(x. t)] -
at v(x. t) + ax v(x. t) -
v(l. t) = - t). u(O. t) = -v(O, t).
(6.104)
Let B denote the associated generator of (6.104). Then Theorems 6.3 and
6.11 imply that
wess(A) = wess(B) = wa(B) = S(B), (6.105)
and eAt - e
Bt
is compact for any t O. Our first conclusion is the following
theorem.
Theorem 6.23 The spectrum-determined growth condition holds for system
(6.103), i.e.,
wo(A) = S(A).
Proof. If we can show that S(B) :S S(A), then from (6.36), we see that
SiAl :S wo(A) = max{S(A), S(B)} = S(A);
hence, S(A) = wa(A).
6.4 A vibration cable with a tip mass 351
We know from Section 6.1 that both operators A and B have compact
resolvents, so the spectrum of these two operators contains only isolated
eigenvalues with finite algebraic multiplicities.
Let A E C be an eigenvalue of A and let <I> = (u, v, d) be an associated
eigenvector with A. Then (A-A)<I> = O. This is equivalent to finding a -:j:. 0,
satisfying the following equation:
{
A2(X) - "(x) = 0, x E (0,1),
(O) = 0,
(1 + aA)'(I) + (mA2 + oA)(I) = O.
Solving (6.106), we see that I(A) = 0 in order that (x) -:j:. 0, where
I(A) = ((1- 0) + (a - m)A)e-
2A
+ (1 + 0) + (a + m)A.
Similarly, u(B) consists of roots of g(A) = 0, where
g(A) = (m + a)e
2A
- m + a.
Two cases should be distinguished:
(6.106)
(6.107)
(6.108)
Case 1: m -:j:. a. In this case, all solutions of g(A) = 0 lie on the vertical line:
1 1111-al
ReA = -log -- .
2 m+a
(6.109)
On the other hand, I(A) = 0 can be written asymptotically as
e2A = o-I+(m-a)A = m-a +0(1 A 1-1 ).
l+o+(m+a)A m+a
(6.110)
Using the technique in Chapter 4, we can easily show, by Rouche's theorem,
that there exist an infinite number of solutions to (6.110). Obviously, the
solutions of I (A) = 0 asymptotically satisfy
1 l11l - al
ReA -+ -log -- as IAI -+ 00,
2 m+a
which together with (6.109) shows that S(B) ::; S(A).
Case 2: m = a. In this case, there is no finite solution to g(A) = 0, i.e.,
S(B) = -00. Therefore, S(B) ::; S(A). A special situation in this case is
when m = a and 0 = 1 in which I(A) = 0 has only one solution A = -11m.
o
From the proof of case 1 in Theorem 6.23 and the relationship between
(6.100) and (6.102), we have
352 Other Applications
Corollary 6.24 If m =1= a, then system (6.100) is exponentially stable with
decay rate
1 lIn -al
Wo = -log -- < O.
2 m+a
That is, for any to > 0 such that Wo + f < 0, there ex'ists an M. > 0 such that
E(t) ::; M.e(wo+.)tE(O)
where E(t) is defined in (6.101). When In = (J, and a = 1, then Wo = -11m.
This corollary clearly shows that the sol1ltion of (6.lO3) is exponentially
stable. But this alone does not guarantee that the Illotion of the tip hody
decays exponentially because we have only shown that [ayx(l, t)+mYd1, tW
decays exponentially. For smooth initial conditions. e.g., Wo E D(A), how-
ever. we can further show that
IIW(t)IID(A) lIeAtWoIID(A) = IleAtWol1 + IIAeAtWol1
IleAtWol1 + IleAtAWo11
< M.e(wo+.)t IIWol1 + M.e(wo+.)t IIAWol1
M.c(wo+.)t IIWoIID(A) .
On the other hand. since
IIW(t)IID(A) = IIW(t)IIH + IIAW(t)IIH
(6.111)
and IIAW(t)1I contains the term fol t)dx, which decays exponentially,
it turns out that y;(1. t) decays exponentially with the same decay rate Wo
as E(t) because
Similarly, since
y2(1, t) ::; 11 y;(x, t)dx,
it follows that y2(1, t) also decays exponentially because the right-hand side
in this inequality is a part of E(t), which is already shown to be exponentially
stable in Corollary 6.24. Consequently, we have shown that the motion of
the tip body decays exponentially with the decay rate Woo
Since we have only considered the energy decay for the case m =1= a and
the case m = a, a = 1, we now turn to consider the energy decay for the
remaining case m = a, a =1= 1. In this case, the reduced system (6.lO4)
becomes
{
f) ['U(X,t)] +K f) [U(:E,t)]_O
m v(x, t) ax V(;E. t) - ,
0(1, t) = 0, u(O. t) = -v(O, t).
(6.112)
6.4 A vibration cable with a tip mass 353
By using the method of characteristics, the solution of (6.112) can be
readily found to be
{
Uo (x - t), x 2: t.
u(x,t)= -vo(t-x). x:::; t:::; l+x.
O. t> 1 + x.
{
Vo (x + t). x + t 2: 1.
v(x.t) =
O. x + t > 1.
(6.113)
(6.114)
for any initial condition u(x. 0) = uo(xl. v(x. 0) = vo(x). By the compactness
of eAt - e
Bt
for any t 2: 0, we have the following interesting results.
Theorem 6.25 When m = a, the semigroup eAt with system
(6.102) is compact for t 2: 2. but is not compact for t < 2.
We wish to characterize the decay rate for the case '(n = a, a i= 1 which
is summarized in the following theorem.
Theorem 6.26 Suppose Tn = a,e1' i= 1. Let A E O'(A). Then
(i) There are no other points in O'(A). except A and .x, which lie on the
vertical line {z I Rez = ReA}.
(ii) If a > 1, then there is only one real negative eigenvaZ,ue Ao which is the
dominant eigenvalue of A: that is. for all A E O'(A),A i= Ao. it holds
that ReA < Ao. Furthermore.
1 1 (OO+1+a)
- < Ao < - - log .
a+oo-1 2 oo-1+a
(iii) If a < 1, then there are at most two negati've real eigenvalues {AOI, Aod,
Moreover, (a) if a :::; 1 - oo. then there is no real eigenvalue. and all
eigenvalues satisfy
1 1
ReA < --log --.
2 1 - (t
(b) if there is no real eigenvalue when a > 1 - (t, then all eigenvalues
A E O'(A) satisfy
1 a
ReA < --log --.
2 1- Cl'
(c) if there is at least one real eigenvalue. then a > 1 - a and all
eigenvalues satisfy
1
Re).,<-----
0.-(1-00)
Besides, the maximal real eigenvalue in case (c) is the dominant eigen-
value as in case (ii).
354 Other Applications
Proof. (i). Let A = u + iT E O"(A). Then irA) = 0 reduces to
{
(n - 1)e-
2
0" cos 2T = 1 + (} + 2au,
(1 - n )e-
2
0" sin 2T = 2aT.
(6.115)
Suppose that u + E utA). Then froll! the first equality of (6.115), we
deduce that cos 2T = cos which implies T + = n7r or T - t:. = n7r for
some integer n. By the second equality of (6.115), it follows that 11. = O.
Therefore, = T or = -T. This proves (i). To prove (ii), note that
1'(0") = 2(a - 1)e-
2
0" + 2a > 0 for any real u by a > 1. frO) = 2 aud
f(-oo) = -00. So, f(A) = 0 has ouly one real negative solution AQ. For any
other A = u + iT E utA), by (6.115)
1 + (f + 2au = (et - 1)e-
2
0" cos 2T S (a - 1)e-
2
0",
that is, f(O") SO which implies that 0" S AQ. This, together with (i), implies
the dominance of Ao. Furthermore. by
0= f(AQ) (1 - a)e-
2
,),o + 1 + a + 2aAQ
< (1-a)(1-2AQ)+1+a+ 2aAo,
we have Ao > - a + _ l' On the other hand, since Ao is negative,
0= f(AO) (1 - a)e-
2
,),o + 1 + 0 + 2aAo
> (1-o:)e-
2
,),o+1+(+a(1-e-
2
,),O),
which implies that
1 (0+1+a)
AO < --log .
2 a-1+a
Finally, let a < 1. Then there are three cases:
(a). Note that !,,(u) > 0 for any u E R, the function I'(u) = -2((1 -
0)e-
2
0" -a) has a unique real zero A" = -tlog frO) = 2 and f(-oo) =
00. If a S 1 - 0, f(O") > frO) = 2 for all u < O. Therefore, for any
A = 0" + iT, u < 0, by the first equality of (6.115)
(a - 1)e-
2
0" COS2T = 1 + 0 + 2au 2: 2 - (1- ole-
2
0".
and so
2(1 - a)e-
217
2: (1- n)e-
217
(1- COS2T) 2: 2,
which implies that ReA = u S - t log 1 .
(b). If 1(>') has no real zero with a > 1- a, it is implied from the above that
I(A *) > O. For any complex eigenvalue A = u + iT, it follows from the second
6.5 Dirichlet-Dirichlet boundary conditions
355
quality of (6.115) that sin2T = which implies that 1 and
so ReA = a A'. Therefore, in this case all A E a (A) satisfy
1 a
ReA < --log--.
- 2 I-a
(c). If there is at least one real eigenvalue, it is implied that a > 1 - a
and f(A') O. There is only one real eigenvalue when f(A') = 0 and
two real eigenValues while f(A') < O. It is obvious that in both cases, the
maximal real eigenvalue is greater than or equal to A'. For any complex
eigenvalue A = a + iT, it follows from the second equality of (6.115) that
sin2T = which implies that 1 and so ReA = a A'.
Furthermore, for a real zero A of f(A) = 0, we have
-2aA = (1- a)e-
2A
+ 1 + a> (1- a)(l- 2A) + 1 + a = 2 - 2(1- alA,
which implies A < - a 0)' (c) is thus proved.
o
The spectrum-determined growth condition proved in Theorem 6.23 en-
ables us to determine the decay rate rigorously in most cases by finding the
upper bound of the real parts of eigenvalues of the system. When m =1= a,
Corollary 6.24 shows that the decay rate has nothing to do with the velocity
feedback gain a > O. However, for the case of m = a, things are completely
different. For example, for given m and a > 1, it follows from (ii) of Theorem
6.26 that the decay rate tends to zero as a goes to infinity. On the other
hand, for the case of a > 1, from (ii) of Theorem 6.26, we may make the
decay rate as small as possible by letting a --+ 0 and a --+ 1; for the case of
a < 1, the decay rate also could be made as small as possible by taking a
= E, 1 - a = E2 and E --+ 0 (see (iii) of Theorem 6.26).
6.5 Thermoelastic system with Dirichlet-
Dirichlet boundary conditions
In this section, we study the stability of a normalized one-dimensional linear
model describing longitudinal vibrations within a thermoelastic rod which
is assumed to be a homogeneous and isotropic cylindrical body positioned
along the x-axis. Let u(x, t) and t?(x, t) denote, respectively, the displacement
and temperature of the rod at position x and time t. Then the normalized
equations, which govern the coupled thermoelastic motions, are given by
{
Utt(x, t) - u",,,,(x, t) + 'Yt?",(x, t) = O. x E (0,1),
t?t(x, t) + 'Yu",t(x, t) - kt?",,,,(x, t) = O.
(6.116)
For the detailed derivation and the physical meaning of this model, the reader
is referred to Day [47] and Carlson [23]. See also Liu and Zheng [95]. In
356 Other Applications
(6.116), the coupling constant, is generally less than 1 and is a measure of
the mechanical-thermal coupling presenting in the system. The constant k
represents the heat conductivity.
Introduce the heat flux q, the stress (j, and the velocity v as
(6.117)
Then the balancing of power flow at each end of the rod leads to the following
boundary conditions
v(i, t)(j(i, t) - q(i, t)1'J(i, t) = 0, i = 0,1, (6.118)
which will be called natural boundary conditions. (6.118) can be alternatively
written as
d
t[
V(i,t) q(i,tl]_o . 01
e 1'J(i, t) (j(i, t) - ./. = , .
That is, the two-row vectors, or the two-column vectors, are linearly depen-
dent, so there exist constants a and {J with a
2
+ (J2 = 1 such that
av(i,t)+{Jq(i,t) =0, a1'J(i,t)+!J(j(i.t) =0, i=O,1 (6.119)
or
av(i, t) + (J1'J(i, t) = 0, aq(i, t) + (J(j(i, t) = 0, i = 0.1. (6.120)
Up to equivalence, (6.119) and (6.120) cover all natural boundary conditions.
For brevity of exposition, we shall consider only a special natural bound-
ary condition: Dirichlet-Dirichlet boundary condition, as specified in the
following
{
Utt(x, t) - u",,(x, t) + ,.1'J,,(x, t) =_0, < x < 1,
1'Jt (x, t) + ,U"t(x, t) - k1'J",,(x, t) - O.
u(i,t) =1'J(i,t) =0. i=O.l.
The system energy associated with (6.121) is given by
111
E(t) = - [u;(x, t) +u;(x, t) + 1'J2(x, t)]dx.
2 0
Formally, it is easily verified that
along the solutions of (6.121). That is, the system is dissipative.
(6.121)
(6.122)
(6.123)
6.5 Dirichlet-Dirichlet boundary conditions 357
To write (6.121) as an abstract equation, let us introduce a Hilbert space
H = HJ(O, 1) x L2(0, 1) X L2(0, 1) with the inner product
Define the operator A : D(A) -+ H by
{
A(u,v,'!?) = (v,u",,,, -'Y'!?""k'!?",,,, -'Yv",),
(6.124)
D(A) = (H2(0, 1) n HJ(O, 1)) x HJ(O, 1) x (H2(0, 1) n HJ(O, 1)).
Then (6.121) can be written as
dw(t) = Aw(t)
dt
(6.125 )
with w(t) = (u(,t),Ut(,t),'!?(,t)). Since the energy is dissipative, the fol-
lowing lemma is readily shown.
Lemma 6.27 The operator A defined by (6.124) is dissipative; A-I exists
and is compact. Therefore, by the Lumer-Phillips theorem, A generates a
Co-semigroup of contractions on H.
We are interested in the problem of whether this semigroup is exponen-
tially stable. Using frequency domain criteria, Liu and Zheng [95] proved
that system (6.125) is exponentially stable. Here, we first show that system
(6.125) satisfies the spectrum-determined growth condition by using a result
due to Renardy [136], and then we show the system is exponentially stable
by analyzing the spectrum distribution of the operator A.
Theorem 6.28 [136J Let A be the infinitesimal generator of a Co-semigroup
on a Hilbert space H. Assume that A is normal, that is, A and A* commute,
and there exist a constant M and an integer n such that the following hold:
(i) For each A E O"(A) with IAI > M - 1, A is an isolated eigenvalue of A
with finite algebraic multiplicity.
(ii) For each z E C,lzl > M, the number of eigenvalues of A in the unit
disk centered at z (counted by algebraic multiplicity) does not exceed n.
Then for any bounded operator B E (H), the spectrum-determined growth
condition holds:
S(A + B) = wo(A + B).
358 Other Applications
Proof. From Corollary 3.40, it suffices to show that
sup I/R(w + iT. A + B)II < 00 for any w > S(A + B). (6.126)
TER
Clearly, for any fixed w > S(A+B), IIR(w+iT. A+ B)II is uniformly bounded
for all T on any compact segment of a real line. so it is sufficient to prove
(6.126) for those A E r = {A I ReA = w.IAI > M + K}. where K is a large
number to be chosen. For any given h E H and A E r. since A E p(A + B).
there is a unique 'u satisfying the following equation
Au - AU + Bu = h. (6.127)
Let P(A) be the orthogonal projection onto the span of all eigenvectors of
A for which the associated eigenvalues lie in a disk of radius K centered at
A. Let Q(A) = 1- PtA). Then H can be decomposed as the direct sum of
X = P(A)H and Y = Q(A)H, i.e.,
H=XffiY.
Set P(A)U = x, Q(A)u = y, P(A)h = f, Q(A)h = 9 and rewrite (6.127) as
Ax - AX + P(A)Bx + P(A)By = f,
Ay - AY + Q(A)By + Q(A)Bx = g.
(6.128)
(6.129)
We see that A - A + Q(A)B, when restricted to y, is invertible for K > liB II
and
II (A - A + Q(A)B) ly111 :S K _IIIBII' A E r.
Indeed, by assumption, A Iy -A is normal, :>0 it follows that
where d(A,I7(A Iy)) denotes the distance between A and I7(A Iy) (see [84.
p.277]). Hence II(A - A)-lQ(A)BII :S IIBII / K < 1 as K > IIBII. It follows
that I + (A - A)-lQ(A)B is invertible in Yand
which holds for all A E r. Let A E r. Solving y from (6.129) and plugging
the result into (6.128), we obtain
= f - P(A)B[A - A + Q(A)Brlg, (6.130)
6.5 Dirichlet-Dirichlet boundary conditions 359
where D.(>') = A - >. + P(>.)B - P(>.)B[A - >. + Q(>.)BJ-1Q(>.)B. Since
>. E p(A+B), D.(>') is invertible and the uniform boundedness of these inverse
will certainly induce the uniform boundedness of IIR(w + iT, A + B)II, since
the operator of the right-hand side of (6.130) is uniformly bounded.
Now in the space X, we can write
m
A - >. = - >')PA.,
i=l
where P
A
is the eigenprojection of A corresponding to the eigenvalue >'i and
m is an integer which depends on >. but has a uniform upper bound N for
all >. E r by assumption. Therefore,
II(A - >.) Ixll :::; NK.
U sing this, it is clear that
IIBII2
IID.(>')II :::; NK + IIBII + K -IIBII'
(6.131)
This indicates that D.(>') is uniformly bounded on X for all >. E r. Since X
is a finite dimensional space, the operator D.{>.) has an n x n matrix repre-
sentation, still denoted by D.(>.), in terms of the orthonormal eigenvectors of
A. (6.131) implies that the elements of this matrix are uniformly bounded
for all >. E r. Define
!A(p,) = det[A - >. - p, + P(>.)B - P(>.)B[A - >. - p, + Q(>.)Br1Q(>.)B].
Then !A(p,) is uniformly bounded for all >. E rand p, in any compact set.
Since Re>. = w is away from IT(A + B) with a positive distance and hence
there is a small 0 > 0 such that>. + p, E p(A + B) for all >. E r and I p, I:::; o.
It is seen from previous arguments that>. + p, E p(A + B) if and only if
!A (p,) ::f. O. Therefore,
IA(P,) ::f. 0 for all >. E r, 1p,1 :::; o.
Because a lower bound on the determinant trivially yields an upper bound
on the norm of the inverse matrix, our proof will be complete if we can show
that I!A(O)I is bounded from below for>. E r.
Suppose that there exists a sequence >'n E r such that IAn (0) ---+ 0 as
n ---+ 00. Since !An (p,) are uniformly bounded in the disk 1p,1 :::; 0, it follows
that there is a subsequence of {!An}' still denoted by !An' and an analytic
function I (p,) such that
!An ---+ I
uniformly in 1p,1 :::; 0/2. This conclusion comes from Mentel's theorem which
says that a sequence of locally uniformly bounded analytic functions contains
360 Other Applications
a subsequence which converges to an analytic function [38]. Since!>.n (J-l) ::j:. 0
in I J-l 8/2 and f(O) = 0, we have f == 0 on {J-l I 1J-l1 < 8} by Hurwitz's
theorem (see, e.g., [38]).
On the other hand, since !>.(J-l) is well defined for>. E rand 1J-l1 < 8, by
analytic extension, f>.'(J-l) are defined on all S6 = {J-l E C I ReJ-l 2: 0, -8 <
IrnJ-l < 8} C piA + B), and they are uniformly bounded for all >. E rand
J-l in any compact subset of S6. Using Mentel's theorem again, we conclude
that !An -t 0 for each J-l E S6. But this is clearly not the case for large It by
the Hille- Yosida theorem, and the theorem follows from this contradiction.
o
The following theorem also is useful.
Theorem 6.29 Let A be the infinitesimal generator of a Co -semigroup on a
Hilbert space H. Assume that there is a Hilbert space K, a normal operator
Ao in K, and a family of bounded operators C(>') on K with the following
properties:
(i) C(>') is defined for Re>. > 710 and its norm is uniformly bO'unded in any
closed half-plane Re>. 2: >'0 > 710'
(ii) Ao satisfies all those hypotheses for the operator A in Theorem 6.28.
(iii) For Re>. > 710, A - >. has a bounded, everywhere defined inverse if and
only if Ao - >. - C(>.) has a bounded. everywhere defined inverse. More-
over, there is a constant Co such that
IIR(>.,A)II Co IIR(>.,Ao + C(>'))II
(6.132)
Then
wo(A) max{S(A)'71o},
Proof. This is done easily by viewing C(>.) as the operator B in Theorem
6.28 and by repeating the proof process of Theorem 6.28. 0
Let us now turn back to analyze the stability of (6.125) by using the
results stated in the above theorems.
Choose K = HJ(O, 1) x 2(0,1) and define
(6.133)
with domain (H2(0, l)nHJ (0,1)) xHJ (0,1). Then An = -Ao is skew-adjoint;
hence, it is normal. It is also clearly seen that Aol is compact. Thus, the
requirements on Ao in Theorem 6.29 are satisfied. Let
C(>.)(u, v) = (0, -"./ D(>. - kD2)-1 Dv) (6.134)
6.5 Dirichlet-Dirichlet boundary conditions 361
where D is used to denote :x and D2 represents the Laplacian with Dirich-
let boundary conditions. Now we verify that the operator C(A) defined by
(6.134) does satisfy the conditions of Theorem 6.29 with fJo = -k1r2. To this
end. we first notice that
-"? D(A - kD2)-1 D -'l(A - kD2)-1 D(D)
2
L(A - kD2)-1( -kD2)
k
2
= L(A - kD2)-1(A - kD2 - A)
k
2
-L(A(A - kD2)-1 - ) (6.135)
k
for any E COO(O, 1). Since D2 is the Laplacian with Dirichlet boundary
conditions, it is well known that it has a set of eigenvalues {-kn
2
7r
2
, n 2:
I} and associated eigenvectors {sin n7rX, n 2: I} which form an orthogonal
basis of L2 (O, 1). In terms of this basis. we can expand ( x) as ( x) =
en sin n7rx. This yields
Therefore, there is a constant M > depending only on AO such that for
all ReA 2: AO > fJo = _k7r
2
.IIA{A - kD
2
)-111 ::; M IIII in L2{0, 1). Since
COO(O, 1) is dense in L2(0.1), we see that
(6.136)
holds uniformly for all ReA 2: AO. This verifies the condition (i) in Theorem
6.29. We show that the condition (iii) in Theorem 6.29 also is satisfied. For
this purpose, taking any (f, g, h) E H and solving the following equation
(A - A)(u, v, '!9) = (AU - v, AV - Uxx + 'Y'!9 x, M - k'!9
xx
+ 'Yvx)
(f. g, h),
we have
(AU - v, AV - Uxx + 'Y'!9
x
) = (f,g),
A'!9 - k'!9 xx + 'YVx = h.
Replacing the spatial derivative by the operator D yields
'!9 = (A - kD2)-1(h - 'YDv),
'!9
x
D(A - kD2)-1(h - 'YDv ),
A - (Ao + C(A))(U,V) = (f,g) + (O, -'YD(A - kD
2
)-lh).
362 Other Applications
This clearly shows that for any ReA > 170. A E ptA) if and only if A E
p(Ao + C(A)). Moreover,
R(A,A)(f,g, h) = (u. v, (A - kD2)-1(h -,Dv)),
(u, v) = R(A, Ao + C(A))[(f,g) + (0, -,D(A - kD
2
)-lh)J.
(6.137)
Consequently, all the conditions of Theorem 6.29 are satisfied and it follows
from the result of Theorem 6.29 that the growth rate of the thermoelastic
system (6.125) satisfies
(6.138)
In view of this relation, it remains to analyze the spectrum distribution of A
in order to prove the exponential stability of (6.125). The rest of this section
is devoted to the spectral analysis for the operator A.
Recall that A-
1
is compact, as stated in Lemma 6.27. Thus, IT(A) consists
only of the eigenvalues of A. It is easily seen that A E IT(A) if and only if
there exists (>, 'IjJ) =I- such that
{
A2>(x) - >"(x) + ,I/J'(x) = 0,
A'lj,,(X) + A,'>'(X) -k1j/'(x) = 0,
>(z) = 1/J(z) = O,l = 0,1.
(6.139)
To eliminate 1/J, we differentiate the first equation of (6.139) and substitute
1/J" into the second equation to obtain
{
A2>(x) - >"(x) + ,'lj/(x) = 0,
,A1/J(X) + A(Ak + ,2)>'(x) - k>"'(x) = 0,
>(i) = A(Ak + ,2)>'(i) - k>"'(i) = 0, i = 0,1.
(6.140)
Differentiating the second equation of (6.140) and substituting 'I/J' into the
first equation of (6.140), we get
with
{
k>""(x) - A(kA +,2 + 1)>"(:z:) + A3>(x) = 0,
>(i) = A(Ak + ,2)>'(i) - k"'(i) = 0, i = 0,1
,A1/J(x) = k"'(x) - ..\(..\k + ,2)'(x).
(6.141)
(6.142)
Thus, solving the eigenvalue problem (6.139) is equivalent to finding a pair
of (A, E C X H4(0, 1) such that > =I- and equation (6.141) holds.
6.5 Dirichlet-Dirichlet boundary conditions
Lemma 6.30 The general solution of (6.141) is
where c;, i = 1,2,3,4 are arbitrary constants and
{
a1 = J ft[k.\ + "(2 + 1 + V(k.\ + "(2 + 1)2 - 4k.\].
a2 = J [k'\ + "(2 + 1 - V(k.\ + "(2 + 1)2 - 4k'\].
363
(6.143)
(6.144)
Proof. The proof is almost obvious by noting that the characteristic equa-
tion of the first equation of (6.141) is
which has four roots specified by
.\
a
2
= -[k.\ + "(2 + 1 V(k'\ + "(2 + 1)2 - 4k'\].
2k
(6.145)
(6.146)
o
We now derive the characteristic equation which an eigenvalue .\ should
satisfy. First, it is remarked that Re.\ < 0 for any .\ E O"(A), because we have
already seen that A is dissipative in Lemma 6.27. Suppose (k.\ + "(2 + 1)2 -
4k'\ = 0, then we have
.\ = 1 - "(2 2"(.i
k k '
(6.147)
which is impossible since "( is a constant less than 1, as we stated in the
beginning of this section. Therefore, (k'\ + "(2 + 1)2 - 4k'\ #- O. The boundary
conditions at x = 0 in (6.141) imply that
(6.148)
where
(6.149)
Similarly, the boundary conditions at x = 1 yield
[
C
C21 ] 1 [exp( a2 - a1 ) (91 + 92 )C3 + exp( -a2 - a1 ) (gl - g2 )C4]
= - 291 exp(a2 + ad(gl - g2)C3 + exp(a1 - a2)(91 + g2)C4 .
(6.150)
364 Other Applications
Thus, from (6.148) and (6.150). the necessary and sufficient condition for
(6.141) to possess a nonzero solution is
which is equivalent to
8gl g
2 - [exp(al - a2) + exp(a2 - ad](gl + g2)2
+[exp( -al - a2) + exp( a2 + ad] (gl - g2)2 = O.
An alternative expression of (6.152) is given below.
(6.151)
(6.152)
Lemma 6.31 The characteristic equation which an eigenvalue A of A should
satisfy is given by
+ a2)
+[exp( 2ad + exp(2a2)](kA +,2 + 1 + _
-[1 + exp(2al + 2a2)](kA +,2 + 1 - + = O.
(6.153)
Proof. From (6.149) and (6.144), it is routine to check that
and so
ala2[A
2
(kA + ,2)2 - Ak(kA + ,2)(ai + +
(al - a2)(-A -
(al
Av>../Vk.
A(kA +,2 + 1)/k,
Consequently, we deduce (6,153) from (6.152).
o
Proposition 6.32 A E O'(A) if and only if A ::j: 0 is a root of equation
(6,153).
6.5 Dirichlet-Dirichlet boundary conditions 365
Proof. This result easily follows from Lemma 6.31. o
In the sequel, we will give an asymptotic analysis for the roots of equa-
tion (6.153). To simplify our presentation, we will use A in replace of kA
throughout our analysis, and restore it when we draw our final conclusions.
Write (6.153) as
where
and
8l.l,\ exp(b
1
+ b2 )
+[exp(2bI) + exp(2b2 )](A + ,.../ + 1 + 2.1,\)(1 - .1,\)2
-[1 + exp(2b
1
+ 2b
2
)](A +,..? + 1 - 2.1,\)(1 + .1,\)2 = 0,
b -
1 - V2k A'
1 I ')'2 + 1 I ( ')'2 + 1) 2 4
= y1+ -A-+ V 1+ -A-
(6.154)
(6.155)
The Taylor expansions of the complex functions f (1/ A) and 1/ f (1/ A) at
A = 00 are given by
(6.156)
(6.157)
Using the polar representation of the complex number, we can write
As the closed right-half plane does not contain any eigenvalues and since A
is symmetric with respect to the real axis, we need only consider those A
where 7r /2 < {) 7r. Furthermore, we divide our analysis into two parts
and formulate the results into two lemmas, and then draw our concluding
theorem from them.
366 Other Applications
Lemma 6.33 Let <5 > 0 be sufficiently small. Then for all 7r - 8 < 19 :::; 7r
where n denotes some positive integer.
Proof. We need to carry out all asymptotic analysis on (6.154). To begin
with. let us note that
exp(b
l
)
exp(-b
2
l
>. ,2
exp( -) + 2k exp( O( 1>'1-
1
))
k '
1>'1 ,2 il>.1
exp( - cos 19 + -k) exp(- sin 19) exp(O(I>'I-
l
))
k 2' k
O(exp( -'11>'1)). for some ,I > O
..;>.
exp( --) exp(0(1>'I-
l
/
2
))
k
IW/
2
19
exp( --k- cos "2) exp(0(1>'I-
l
/
2
)) = 0(1).
Multiplying exp( -2b
2
) on both sides of (6.154) gives
(>. +,2 + 1 + 2";>')(1 - ";>')2 - exp( -2b
2
)(>. +,2 + 1 - 2V,\)(1 + V,\)2
0(1)'1
2
exp( -'11>'1)).
A simple calculation indicates that
(6.158)
which will be our main building block to prove our assertion. Rewrite (6.158)
as
From (6.158), we see that
Hence,
e
ilrn
(2b
2
) = 1 + 0(1)'1-
3
/
2
).
e
Re
(-2b2 ) = 1+0(1)'1-
3
/
2
).
(6.159)
6.5 Dirichlet-Dirichlet boundary conditions 367
On the other hand. the second equality above. together with (6.157), implies
that
1) --+ 7r, as I>" 1--+ cx:;. (6.160)
Indeed, we may even show that 1) = 7r for large enough 1>"1. To see this. we
analyze (6.159) again. Since e
i1m
(2b
2
) = 1 + 0(1)''1-
3
/
2
).
sin(Im(2b
2
)) = 0(1)..1-
3
/
2
)
and
Plugging these into (6.158). we obtain
[1- 4,21>"1-3/2e-i3/2{}]eilm(2h) + 0(1)''1-
5
/
2
)
[1- 4,21>"1-3/2e-i3/2{}][cos(Im(2b2)) + i sin(Im(2b
2
))]
+0(1)''1-
5
/
2
)
_ 31) _
1- 4,21>"1 3/2 cos - + 0(1)''1 5/2)
2 .
Taking logarithm on both sides of the above equation yields
Putting this into (6.157), we obtain
2 ,2 1),2 3,2 31)
(--1>..1
1
/
2
+ _1>"1-
1
/
2
) cos - - -(1- _)1>"1-
3
/
2
cos-
k k 2 2k 4 2
_4,21>"1-
3
/
2
cos 3: + 0(1)''1-
5
/
2
). (6.161)
which implies that cos = 0(1)''1-
2
). If 1) =! 7r for large 1>"1. then by (6.160)
31) 1)
cos 2"1 cos 2" --+ -1. as 1>"1 --+ 00.
But (6.161) then leads to a contradiction that 1>"1
1
/
2
= 0(1). Thus
1) = 7r. for large 1>"1. (6.162)
Finally, from (6.157) and (6.162).
368 Other Applications
Since sin(Im(2b
2
)) = 0(1,\1-
3
/
2
). we have
2 ,2
_1,\1
1
/
2
+ _1,\1-
1
/
2
= 2nlr + 0(1,\1-
3
/
2
)
k k
or
(6.163)
where n is a sufficiently large integer.
o
Lemma 6.34 Let 6 > 0 be sufficiently small. then for 1r /2 < 1) :::; 1r /2 + 6.
2
Re'\ - + 0(1,\1-
3
/
2
). and
1,\1 knlr + 0(1,\1-
1
)
where n is a sufficiently large integer.
Proof. In this case. 1r / 4 < 1)/2 :::; 1r / 4 + 6/2. Thus. there is a ,2 > 0 such
that
and
exp( -b
2
)
v,\
exp( - -) exp( O( 1,\1-
1
/
2
))
k
1,\ 11 /2 19
exp( - --cos -) exp( O( 1,\1-
1
/
2
))
J.,; 2
O( exp( -,21,\1
112
)).
exp(bJ)
,\ ,2
exp(k" + 2k)exp (0(1,\1-
1
)
1,\1 ,2 .1'\1. _
exp( -k cos 19 + 1. ) exp(l- SlIl19) exp( O( 1,\ I 1))
2r.; J.,;
0(1).
The rest of the proof is very similar to that of Lemma 6.33.
Substituting the two equations above into (6.154) we deduce that
and
(,\ + ,2 + 1 + 2/:\)(1 - /:\)2 _ cxp(2bJ)('\ + ,2 + 1 - 2/:\)(1 + /:\)2
0(1,\1
2
exp( _,21,\1
1
/
2
)).
exp( 2bd
= (,\ + ,2 + 1 + 2v,\)(1- v,\)2 + O(ex (_ 21'\11/2))
(,\ + ,2 + 1 - 2 v,\)( 1 + v,\)2 p,
1 + 0(1,\1-
3
/
2
). (6.164)
6.5 Dirichlet-Dirichlet boundary conditions
Write (6.164) as
e
Re
(
2b
d = eilm(-2bd + 0(1)'1-
3
/2).
Comparing this with (6.164) gives
e
Re
(
2b
d = 1 + 0(1)'1-
3
/2),
e
ilm
(-
2b
d = 1 + 0(1)'1-
3
/2).
By noting (6.157), we have
369
(6.165)
2 ,2,2,2
Re(2bd = kl>'1 cos 19 + T + T(I- 4")1>'1-
1
cos 19 + 0(1)'1-2),
Im(2bd = sin 19 + 0(1)'1-
1
).
From (6.165) and (6.166), we immediately conclude that
19 -t 7r /2, as I >. l-t 00.
Furthermore, from (6.165) we have
Putting this into (6.166) yields
hence,
or
2 ,2,2,2 1
-1>.1 cos 19 + - + -(1- -)-cos19 = 0(1)'1-
3
/2).
k k k 4 1>'1 '
Since cos 19 = Re>./I>.I, (6.169) then implies that
2
Re>. = - + 0(1)'1-
3
/2).
Finally, from (6.169) and (6.167) we see that
sin 19 = VI - cos
2
19 = VI + 0(1)'1-2) = 1 + 0(1)'1-2),
Im(2b1 ) = sin 19 + 0(1)'1-
1
) = + 0(1)'1-
1
).
(6.166)
(6.167)
(6.168)
(6.169)
(6.170)
370 Other Applications
These, together with (6.165), yield
(6.171)
where n is a sufficiently large positive integer. o
Combining Lemma 6.33 and Lemma 6.34, we can prove
Theorem 6.35 The eigenvalues of the operator A consist of a real sequence
{An} and a sequence of conjugate pairs {un. un} with
(6.172)
where n is a sufficiently large positive integer.
Proof. We shall establish the existence of the eigenvalues of the operator
A in a similar way as we did in Chapter 4.
Case 1. First, we assume that .\ = 1.\le
i19
.1f - c5 < {) < 1f. with c5 > 0
sufficiently small. Then by (6.158)
exp(-2b
2
) = 1 + 0(1.\1-
3
/
2
). (6.173)
From (6.157), -2b2 = -pfS + f"* + 0(1.\1-
3
/
2
), so (6.173) implies that
J>. ,2 1
exp(-2
T
+ k J>.) = 1 + 0(1.\1-
3
/
2
).
(6.174)
It is easy to see that exp( -2- + f J>:) = 1 has solutions
(6.175)
Let J>. denote the positive branch of the square root of a complex .\, and so
the mapping: .\ -+ J>. maps the negative real axis of the complex plane to
the positive imaginary axis. Instead of (6.174), we let It = J>. and consider
Let
It ,2 1 -3
exp(-2"k + kj;) = 1 + O(lltl ).
It ,2 1
d
2
= -2- + --.
k k It
(6.176)
Let On be a circle with radius Q.\;;-1/2, Q > 0 and centered at iA,. Then
the circumference of this circle can be described by
r
l"/"'I f\ \ -1/2 i19
vn : It = tv An + QAn e,
o :::; {) :::; 21f.
6.5 Dirichlet-Dirichlet boundary conditions 371
(6.177)
In view of (6.176) and (6.175), we conclude that, for all J-l- E rOn,
Red
2
= _ ).;;-1/2 cos'!? + 0().;;-3/2).
Imd2
= _ 2a). -1/2 sin'!? _ 'Y2). -1/2 + O()' -3/2)
k k n k n n
= -2mr - 2: ).;;-1/2 sin'!? + 0().;;-3/2).
Consequently,
11- exp(d
2
W = [1- exp(Red
2
)]2 + 2exp(Red
2
)[1- cos(Imd
2
)]
4a
2
).-1 + 0().-3/2) (6.178)
k2 n n
Since i';>:;" is the unique root of exp(d
2
) = 1 inside On, we can apply
Rouche's theorem to functions exp(d
2
) - 1 and exp(d
2
) - 1- 0(1J-l-1-
3
), and
conclude that there exists a unique zero J-l-n to the equation exp( d
2
) - 1 -
O( 1J-l-1-
3
) = 0 insides On for all n 2: N when N is sufficiently large.
Let >'n = J-l-;'. Since J-l-n is inside On, we know that
(6.179)
Thus, >'n is the unique root of (6.174) and hence of (6.173) in the order of
0(n
2
). Furthermore, by the arbitrariness of a > 0, we see that
(6.180)
Case 2. As for now, we assume that). = 1).le
i19
,7r/2 < '!? :::; 7r/2 + 6, with
6 > 0 sufficiently small. From (6.164)
exp(2bd = 1 + 0(1).1-
3
/
2
). (6.181)
Since b1 = + + 0(1).1-
1
), so (6.181) implies that
). 'Y2
exp(2"k + k) = 1 + 0(1).1-
1
).
(6.182)
Obviously,
'Y2 .
an = -T + tkn7r, n = 0, 1, 2,
(6.183)
372 Other Applications
are the solutions of exp(2f + f) = l.
Let On be the circle with radius la
n
l-
1/2
and centered at an. Then
For any A E rOn, we have
A ...,2
Let Cl = 2].;" + k' Then for all A E rOn.
Re(Cl)
= I a 1- 112 cos 19
k n ,
2mr + sin 19,
and so
11- exp(cIW [1- exp(Re(cr))f + 2exp(Re(cr))[1 - cos(Im(cr))]
= lanl-
1
+ O(lanl-
3/2
) = O(IAI-
1
).
Applying Rouche's theorem to functions exp(2i + f) -1 and exp(2i + f)-
1 - O(IAI-
1
), we conclude that there exists a unique zero an for exp(2f +
f) - 1 - O(IAI-
1
) = 0 inside On for all n 2: N when N is sufficiently large.
Since an lies inside On. by definition.
(6.184)
This together with (6.171) implies that this an is the unique root of (6.181)
with order O(n-
1
) for positive n. 0
The above spectral analysis results clearly indicate that there exist a set
of eigenvalues on the real axis which corresponds to the spectra of the heat
equation, and a set of complex conjugate eigenvalues which corresponds to the
vibration modes of the wave equation. Turning back to (6.138), we see that,
in order for the spectrum determined growth condition to hold for (6.125), a
sufficient condition is that S(A) 2: -k1r2. Physically. this is possible because
-k1r2, the first eigenvalue of the "pure" (without vibration coupling. i.e ..
J = 0 in (6.116)) heat equation, determines the energy decay rate of the
heat equation, and we cannot expect the energy decays more rapidly when
there are coupled vibrations which produce additional heat. To prove this.
it is sufficient to show that (6.153) has a real root in (-k1r-
2
,O). We do this
below.
6.5 Dirichlet-Dirichlet boundary conditions 373
Since we are only concerned with the negative real eigenvalues of A which
are the negative real roots of the characteristic equation (6.153), letting x =
-k>.., we are able to write (6.153) as
8-iivx + [exp(iJ + g) + exp(-iJ - g)](-x + ,2 + 1 + 2iVX)(1- iVX)2
[exp(iJ - g) + exp(g - if)]( -x + ,2 + 1 - 2i VX)( 1 + iVX)2 = 0
(6.185)
where
{
J(x) = *,h(X), g(x) = ififu.
h( x) = J -x + ,2 + 1 + J (-x + ,2 + 1)2 + 4x > 0, for all x > O.
(6.186)
Comparing the real and imaginary parts, we see that (6.185) reduces to
4,2 VX 2,2 vx[exp(g) + exp( -g)] cos(f)
+ [(1 + x)2 + (1- xJl2][exp(g) - exp(-g)]sin(f) = O.
(6.187)
Theorem 6.36 For any 0 < , :::; 1. there is at least one solution x to
equation (6.187) which locates between 0 and k
2
7f2 on the real axis.
Proof. Denote by F(x) the left-hand side of (6.187). By Taylor expansion,
F(x) 2( 2+1)2
we find that limx-+o ,;xx = l' k: 2 This means that F(:c) > 0 for x > 0
sufficiently small. The proof is complete if we can show that F(k
2
7f2) < 0
because F(x) is a continuous function on [0. k
2
7f2] with respect to x. To this
end. let a = k7f. Note that
(x + 1 _,2)2 :::; (-x + ,2 + 1)2 + 4x :::; (x + ,2 + 1)2;
hence. when 0 < , :::; 1. v'2 < h(x) :::; v'2JlH < 2 which gives. in
particular
7f < J(a
2
) < V27f. (6.188)
Taking x = a
2
in (6.186) and rearranging terms we have
,2 = _7f
2
a
2
/ J2 + J2/
7f
2 + a
2
- 1. 9 = 7f
2
a/ f.
(6.189)
Here, and in the following equation. J and 9 denote. respectively. J(a
2
) and
g(a
2
) with an abuse of notation. Substituting (6.189) into F(a
2
) yields
J27f2e7r2ajf F(a
2
)
4a[J4 - 7f4
a
2 + 7f2(a
2
_ 1)J2]e7r2a/f
-2a[J4 - 7f4 a2 + 7f2(a
2
- 1)J2][e27r2u/f + 1] cost!)
+[47f
2
a
2
J2 + (1- a
2
)(f4 - 7f4(2)][e27r2a/ f - 1] sin(!). (6.190)
374 Other Applications
It can be shown that the right-hand side of this equation is less than 0 [65].
Thus, F(a
2
) < O. 0
We have actually proved the following important result.
Corollary 6.37 Suppose 0 < , ::; l.
(i) The spectrum-determined yrowth condition holds for system (6.125).
i.e.,
w(A) = SIAl.
(ii) For any f > 0 sufficiently small. there exists a positive number M > 1
such that
E(t) ::; Me(w(A!+f)t E(O).
(6.191)
where E(t) is the energy function defined by (6.122). Since w(A) < O.
the energy decays exponentially.
6.6 Thermoelastic system with Dirichlet - Neu-
mann boundary conditions
The methods used in the previous section can be applied easily to ther-
moelastic systems with Dirichlet-Neumann and Neumann-Dirichlet natural
boundary conditions. We shall find that the problem with Neumann-Dirichlet
or Dirichlet-Neumann boundary conditions is much simpler than that with
Dirichlet-Dirichlet boundary conditions. because for the present problem the
characteristic equation which an eigenvalue should satisfy is a third order
polynomial. The roots of this polynomial can be directly solved. enabling us
to prove the spectrum-determined growth condition.
Because the analyzes are very similar. we restrict ourselves to the follow-
ing thermoelastic equation with Dirichlet-Neumann boundary condition:
{
Utt(x, t) - uxx(x, t) + ,'dx(x, t) = 0, 0 < x < 1,
'!9t{x, t) + t) - t) = O. 0 < x < 1.
uCt, t) = ()x(t, t) = 0, 'I = 0.1.
(6.192)
To put this in the form of all abstract equation with state space iI
HJ(O,I) x L
2
(0, 1) x L2(0, 1), let us define the operator A : D(A) -+ iI
by
{
A(u,v,'!9) = (v,uxx -,'dx,k1?xx -,v,,),
D(A) = (H2(O, 1) n HJ(O, 1)) x HJ(O, 1) x H;{O, I).
(6.193)
6.6 Dirichlet-Neumann boundary conditions 375
where H;(O, 1) = {'19 E H2(0, 1) I '19'(0) = 1?'(1) = O}. Then (6.192) can be
written as
dw(t) _ A ()
- w t,
dt
(6.194)
with w (t) = (u(, t), Ut (', t), '19(-, t)). Since A is densely defined, we find, after
a simple calculation, that
{
A*(f, g, h) = (-gxJcx + ,hx' ,gx + khxx ),
D(A*) = (H2(0, 1) n HJ(O, 1)) x 1) x H;(O, 1),
(6.195)
and
Re(A*(f,g,h), (f,g,h)) -k l1lhxl2dX::; 0, V(f,g,h) E D(A*),
Re(A(u,v,t9), (u,v,t9)) = -k l11t9xI2dX::; 0, V(U,V,l?) E D(A).
Therefore, both A and A * are dissipative. The closedness of the operator A
is evident by definition. By Corollary 2.28, A generates a Co-semigroup of
contractions on iI.
The operator A has a zero eigenvalue and
<Po = (0, 0, 1) (6.196)
is the unique eigenvector associated with the zero eigenvalue. Physically, the
zero eigenvalue has an easy interpretation in that it corresponds to the heated
equilibrium. From (6.195), we see that <Po is also the unique eigenvector of
the adjoint operator A * corresponding to the zero eigenvalue; hence, R( A) is
orthogonal to N (A). Since we do not wish to consider nonzero equilibrium
solutions, it is natural to define
H = {<p E iI [ (<p, <Po) = O} = HJ(O, 1) x L2(0, 1) x L;(O, 1), (6.197)
where 1) = {f E L2(0.1) I (I, 1) = O}. For notational simplicity, we
still use A to denote A [H. More precisely, by operator A from now on. we
mean the following unbounded operator defined in H:
{
A(u,v,t9) = (v,u
xx
-,'19
x
,k'19
xx
-,v
x
),
D(A) = (H2(0, 1) n HJ(O, 1)) x HJ(O, 1) x H;(O, 1),
(6.198 )
where H;(O, 1) = {f E H2(0, 1) [ /,(0) = /,(1) = 0, (t, 1) = O}.
Lemma 6.38 The operator A defined in (6.198) generates a Co-semigroup
of contractions on H. A -1 is compact and hence 0"( A) consists only of iso-
lated eigenvalues with finite algebraic multiplicities.
376 Other Applications
Proof. For any given (f, g, h) E H, solving
A(u,v,1?) = (v,u",,,, - ,1?",k1?xx - ,v",) = (f,g,h),
we find that
u(x) CIX + 1'" (x - T)g(T)dT +, l
x
19(T)dT,
v(x) = f(x),
1
'" l1x
1?(x) 0 f(T)dT + k 0 (x - T)h(T)dT + C2 X.
where
Therefore, A-I is compact. Since A is dissipative, by the Liimer-Phillips
theorem, A generates a Co-semigroup of contractions on H. 0
Again let K = HJ (0, 1) X L2 (0, 1) and let Ao and C (A) be the operators
defined in (6.133) and (6.134), respectively. Denote by D2 the Laplacian
with Neumann boundary conditions in 1). Following exactly the proof
procedure in Theorem 6.29 yields the following result.
Proposition 6.39 The growth rate of A, defined in (6.198), satisfies the
following relation:
With the aid of this proposition, in order to show the exponential stability
of the semigroup generated by A we need only to analyze the distribution of
eigenvalues of A in the complex plane.
As before, A E O"(A), which must satisfy ReA::; 0 because A is dissipative,
if and only if there exists (cp, 'Ij;) =1= 0 such that
{
A2cp(x) - cp//(x) + ,v/(x) = 0,
A'Ij;(X) + ,A,CP'(X) = 0.
= 'IjJ = 0, 1, = 0,1.
Eliminating 'IjJ from this equation we obtain
{
kcp////(x) - A(kA +,2 + l)cp//(x) + A
3
cp(x) = 0,
cp(i) = cp"(i) = 0, i = 0.1.
(6.199)
(6.200)
6.6 Dirichlet-Neumann boundary conditions 377
Once is obtained, 1/J is determined from
'Y>'1/J(x) = k"'(x) - >.(>.k + 'Y2)'(X). (6.201)
Solving the eigenvalue problem (6.199) is thus equivalent to finding a pair of
(>., ) E C X H4 (0, 1) such that "# 0 and equation (6.200) is satisfied. The
general solution of the first equation of (6.200) is given by
(6.202)
where Ci, i = 1,2,3,4 are arbitrary constants to be determined by the bound-
ary conditions and aI, a2 are the same as defined in (6.144).
We next derive the characteristic equation which an eigenvalue>. should
satisfy. Since (k>' + "12 + 1)2 -4k>. "# 0, we can then use the boundary
condition at x = 0 and (6.202) to obtain
(6.203)
hence, (6.202) becomes
(6.204)
From the boundary conditions at x = 1, we get
(6.205)
Proposition 6.40 >. E O"(A) if and only if >. "# 0 is a root of equation
(6.205).
Theorem 6.41 Asymptotically, the nonzero solutions of (6.205) consist of
a real sequence {>'n} and a sequence of conjugate pairs {O"n,O"n} with
"12 "12 "12 1
O"n = mri- 2k +i
2k2
(1-4")mr +0(n-
2
),
2
>'n = -k(mr)2 + + 0(n-
2
),
where n is a sufficiently large positive integer.
Proof. Equation (6.205) is equivalent to
(6.206)
Things are now much simpler as compared to (6.164) because from e
2a1
= 1
we immediately obtain
. "12 "12 ( "12 ) 1 -2
>. = n7rZ - - - - 1 - - - + 0(1)'1 )
2k 2k2 4 >.
378 Other Applications
for some integer n. This implies that IAI = O(n) which, when substituted
into the right-hand side of the above equation, yields
A =
n1fi _ ,2 _ L (1 _ ,2) 1
21-.; 2k2 4
+O(IAI-
2
)
,2 ,2 ( ,2) 1
1I1fi - - + i- 1- - - + O(n-
2
).
21-.: 2k2 4 n1f
This is the proof of the first part. From the definition of al aml (J,2. it is
3/2
easy to check that a2 = so that we are able to show. in a sil1lilar
manner as was used to solve e
2a
] = 1. that e
2a2
= 1 has solutions '\n
-k(n1f? +,2 + O(n-Z). The proof is completed. 0
Lemma 6.42 ,\ i- 0 is a solution of (6.206) if and only if there is a nonzero
integer n such that
(6.207)
'Where fJn = n
2
1f2. Moreover. for each n > O. (6.207) admits one unique real
solution An and one conjugate pair sol-ution {O"n. an}.
Proof. Obviously, the nonzero solutions A of e
2a
] = 1 or c
2a2
satisfy al = n1fi or az = n1fi for some nonzero integer n, i.e.,
or
).
-(k)' +,2 + 1 - J(k)' +,2 + 1)2 - 4k)') = _n
2
1f2.
2k
Rearranging terms we have
or
A
3
+ k(n
2
1f2)2 + )'(kA +,2 + 1)n
2
1f2 = O.
1 must
This is exactly (6.207), completing the proof of the first part if we note that
the above process is reversible.
When k
2
fJn - 3(,2 + 1) < O. (A) > 0 for all real A, so that there is a
unique real solution An of fn().) = since fn( -x) = -x and fn( oc) = +oc.
When k
2
fJn - 3h
Z
+ 1) 2': 0, f:,().) = 3'\z + 21-.:fJn). + (,Z + l)fJn = has real
roots
(6.208)
6.6 Dirichlet-Neumann boundary conditions 379
Since
and = 3(A-1J2)(A-I71), we see that f:,(A) > 0 for all A> 1/1 or A < 1/2.
Therefore, there is a unique real solution An to fn (A) = 0 with An < 1/2. 0
It is clear from the proof above that
fn(A) < 0 for all A < An. (6.209)
Theorem 6.43 Let fn(A) be defined by (6.207) and {An, Un, an} be the so-
lution of fn(A) = O. Then
A < _ kl11
n - ,2 + l'
(6.210)
for all n > O.
Proof. Let E = 1/(,2+1). It is seen that fn(A) = (A2+ltn)(A+kl1n)+,2I1nA
and
which, together with (6.209), gives
Now, because
and
fn( -kl1n) = -k,2
11
; < 0,
we see that An E (-kl1n' -kl1n/3). In particular, we have
Al > -k7r2.
Solving
fn(A) A
3
+ kl1nA2 + (12 + l)l1nA +
(A - An)(A - Un)(A - (Tn)
(6.211)
A
3
- (un + an + An)A2 + [lunl
2
+ An(Un + an)]A - Anlunl2
o
380
Other Applications
we see that the real parts of the relllaining two complex roots satisfy
Since = 6A + 2kpn < 0 for all A < -kli
n
/3, we can apply Newton's
method to get an approximation of An with an initial guess -kiln, namely
Consequently.
2ReO"n =
This is the desired result. o
Combining (6.211) and Proposition 6.39, we have the following theorem.
Theorem 6.44 The spectrum-determined growth condition holds for system
(6.194) in the state space H, i.e ..
wo(A) = S(A)
and by Theorem 6.4:1.
-br2 "(2 br
2
Wo ( A) ::; max {",2 + 1 ' - - }
, 2 k
2
7r
2
+ "(2 + 1 .
Consequently, the system (6.194) is exponentially stable.
6.7 Renardy's counter-example on spectrum-
determined growth condition
In the previous sections of this chapter. we stated several typical physical
systems in which the spectrum-determined growth condition hold. Unfortu-
nately, it is not true that all physical systems are good enough to possess
this property. We have already seen one counter-example in Chapter 3. but
the example to be given below is more "physical", and the result is more
surprising in the sense that even a lower order derivative perturbation to a
two-dimensional wave equation can destroy the spectrum-determined growth
condition. This counter-example is taken from Renardy [137], but we provide
a detailed proof.
6.7 Renardy's counter-example ... 381
Consider the system
{
Utt = u"'''' + Uyy + eiyu"" -00 < x, y < 00,
u(x + 271", y) = u(x, y), u(x, y + 271") = u(x, y).
(6.212)
We proceed as usual by rewriting it as
(6.213)
on the Hilbert space H = Hl X L2 = {(f, g) E Hl(C
2
) X L2(C
2
)1/,g are 271"-
periodic in both directions} with the inner product
((ul,vIl, (U2,V2)) = i: i:[UlU2 + U1:x U2", + Ul yU2y + VlV2]dxdy.
In (6.213), the operator A is defined by
{
A(u,v) = (v,u",,,, +uyy +eiyu",),
D(A) = ((u,v) E H I (u,v) E H2(C2) X Hl(C
2
)}.
(6.214)
Lemma 6.45 The operator A defined by (6.214) generates a Co-semigroup
on H and A has compact resolvent.
Proof. First, since for any (u, v) E D(A),
Re(A(u, v), (u,v))
= Re f:'It i:[vu + v"'u'" + 'OyUy + (u",,,, + Uyy + eiYu",)vldxdy
= Re i: [vu + eiYu",vldxdy
1j'lt j'lt
< 2" -'It -'It [21v1
2
+ lul
2
+ lu", 1
2
1dxdy
< Il(u, 'Olil
2
we see that A - I is dissipative. We further show that (A - I)-1 exists and
is compact. To this end, take any (f,g) E H and solve (A- I)(u,v) = (f,g)
which is rewritten as
{
V = u + I,
u"'''' + Uyy + eiy,u", - u = I + g.
(6.215)
Let I(x, y)+g(x, y) = wn(y)e
in
"" u(x, y) = un(y)e
in
",. Then
the second equation above leads to
(ine
iy
- n
2
- l)u
n
(y) + = wn(y), n = 0, 1, ....
382 Other Applications
Let 'lln(Y) = = L:;,';=_oobnmeimy. Then the above
equation reduces to
(_n
2
- rn
2
- l)anm + inan(rn-I) = bnm, rn = 0, 1, .... (6.216)
Therefore, for m 2 1. we have
in 1
a
nm
n2 + rn2 + 1 an(m-l) - n2 + m 2 + 1 b
nm
(in)m
= [n
2
+ m
2
+ l][n
2
+ (m - 1)2 + IJ [n
2
+ 12 + 1Jan0
(in)m-I
.,.--,--::----:-::--:,-----'----'--::---::-----:--::----::-----:-b I - ...
[n
2
+ m
2
+ l][n
2
+ (m - 1)2 + IJ .. [n
2
+ 12 + IJ n
n2 + + 1 b
nm
(6.217)
and
It follows that
(in)m
[n
2
+ (m - 1)2 + l][n
2
+ (m - 2)2 + IJ .. [n
2
+ IJ an(-m)
1 in
anO + n2 + 1 b
no
+ [11,2 + 1][11,2 + 12 + IJ
bn
(-I) + ...
(in)m-I
+[11,2 + (m -1)2 + 1][11,2 + (m - 2)2 + IJ [11,2 + 1J
bn
(-m+I)'
Let m -7 00. The left-hand side above tends to zero for any fixed n because
the requirement of'll E HI implies that an(-m) -70 as m -700. Hence,
00 (in)m-l
ano = - [n2 + (m -1)2 + l][n2 + (m - 2)2 + IJ ... [n2 + 1jbn(-m+I)'
(6.219)
6.7 Renardy's counter-example ... 383
Substituting this into (6.218) yields
00 (in)k-m-l
an(-m)= L [n2+(k_l)2+1J ... [n2+m,2+1Jbn(-k+ll (6.220)
k=m+l
Thus, a
nm
is uniquely determined from (6.217), (6.219) and (6.220). We
now show that
n==-oo m=-oo
Indeed,
n=-oom=l
00
x( L Ibn(-k+1lI
2
)
k=m+l
00 00 k 00 i-m
(; Ib
n
(_klI
2
fl + (i - 1)2 : IJ'" [n
2
+ m,2 + IJ
J2
00 00
< L L
n=-oo k=l
where
Hence, :E:=l n2Ian(_m)12 < 00. Similarly, by (6.217) and (6.219),
00 00 00
L n
2
1anol
2
< 00, L L n
2
1anml
2
< 00
n=-oo n=-(X! m=!
384 Other Applications
which yields n
2
1anml
2
< 00. Using the relation (n
2
+m
2
+
l)anm = -bnm + inan(m-l)' it is not difficult to verify that
x x
L L (71.
2
+ 7//
2
+ 1)21anm1
2
< 00
n=-x, m==-tX,.
which shows that u(:c.y) = E H2(C
2
) and so
v E Hl(C
2
) in view of (6.215). We have actually proved that (A_I)-1 exists
and is compact by the Sobolev imbedding theorem because (A - I)-I H C
H2(C2) X Hl(C
2
). These results, together with the dissipativity of A - I.
guarantee the Co-semigroup generation of A on H by the Liimer-Phillips
0
Now, we state the main result of this section.
Theorem 6.46 Let A be defined a8 in (6.214). Then S(A)
1/2.
O.wo(A) >
Proof. Since the resolvent of A is compact, the spectrum of A consists
only of its eigenvalues. First we prove that S(A) = 0 by showing that all
eigenvalues of A locate on the imaginary axis. Let A be an eigenvalue of A and
u(x, y) be the eigenvector associated with A. Because u(x, y) is 27f-periodic,
u(x, y) can be expanded as a Fourier series u(x, y) = E::"=-oo wn(y)e
in
'" and
each wn(y)e
in
'" is the eigenvector corresponding to A by the orthogonality
of ein",. Therefore, we may set. without loss of generality, that u(x,y) =
w(y )e
inx
. The resulting spectral problem is
(6.221)
where W is required to be 27f-periodic. The problem is trivial for n = 0 and
we shall therefore assume n f. O. By Fourier series expansion
00
w(y) = L wme
imy
,
m=-oo
equation (6.221) is transformed into
(6.222)
We can now start at some value m = M such that WM f. 0 (not all Wm
are identically zero), and then recursively compute W M -1, W M -2, .. , from
(6.222). IfWM-k f. 0 for all k = 1,2, ... , then Iwml-t 00 as m -t -00,
contradicting the hypothesis that W is square integrable. Hence, there is an
m ::; M such that W
m
-l = 0 and Wk f. 0, m ::; k ::; M. Applying this to
6.7 Renardy's counter-example ... 385
(6.222) yields A2 + n
2
+ m
2
= 0, or A = iJn2 + m
2
, which shows that the
spectrum is on the imaginary axis, that is, S(A) = O.
Next, we demonstrate that wo(A) is at least 1/2. If this is not true, then.
by the Hille-Yosida theorem, the resolvent of A must be bounded on the line
ReA = 1/2. Consider now the sequence
(6.223)
where En > 0 is such that En 0 and 00 as n 00. We note that
An = 1/2 + in + 0(1) as n 00, the real part of An approaches 1/2, and
IAnl is proportional to n as n 00. The function cp(y, En) is constructed as
follows. Let cp be a nonzero COO-function with compact support contained in
(-1,1). Consider function En) = cp(Y/En). Without loss of generality, we
may assume En < 71", SO that the support of En) is contained in (-71",71").
We can then take cp(y, En) to be the 271"-periodic continuation of En). For
such chosen un(x, y), it is easy to verify that
i: i: [lun(x, y)12 + Ium,(x, y)1
2
+ Iuny(x, yW]dxdy
= 271" i:[(l + n
2
)1CP(Y/En W + E;;-2Icp'(Y/En)1
2
]dy
= ill En[(2 + n
2
)lcp(zW + E;;-2Icp'(zW]dz,
which indicates that the H
1
-norm of Un is proportional to nFn and 2-norm
of Un is proportional to Fn as n 00. On the other hand, the following
equality holds
\2 _ {)2un _ {)2un _ iy{)2Un _ . ( _ iy) _ in", A." ( )
"nUn {)X
2
{)y2 e ()x - zn 1 e Un e 'I' y, En .
(6.224)
By using the fact that 11-e
z
l Izleizi which is true for any complex number
z, we see that
n
2
i: i: 11- e
iy
12Iun (x, y) 12 dxdy
= 271"n2En f1 11- eienZI2Icp(z)12dz
-1
< ill Icp(zWdz,
which shows that the 2-norm of the first term on the right-hand side of
(6.224) is proportional to Similarly, it is shown that the 2-norm of
386 Other Applications
the second term on the right-hand side of (6.224) is proportional to
Let 'Un = An'Un. Then
hence, II(A - An)(Un. {In)II
H
is proportional to (/,1:;/2 +1';;-3/2. Since lI(un,l'n)IIH
is proportional to .jEn' we have
This means that the resolvent of A is not uniformly bounded on ReA =
1/2, contradicting the assumption. Consequently, the growth rate of the
semigroup generated by A is at least 1/2. that is. wo(A) 2: 1/2. 0
6.8 Notes and references
The study on the general linear hyperbolic system in Section 6.1 originates
from Neves [124). Section 6.2 is studied by Chen et a!.[27). Our interest
here is to show that it can be recast into the form of the general hyperbolic
system and thus the results obtained in the previous section can be applied to
derive the exponential stability conditions [24]. The results in Section 6.3 are
documented in Guo and Zhu [71 J. The stabilization control problem for the
vibration cable with a tip mass in Section 6.4 was first considered in Morgiil
et al. [118], bllt here we proved the exponential stability of the boundary
controlled closed-loop system by verifying the spectrum-determined growth
condition [69J. Theorems 6.28 and 6.29 appeared in Renardy [136] and the
spectral analysis results in Section 6.5 are obtained by Guo and Yung [70] and
[65]. The results in Section 6.6 have been given by Hansen [72J but OUl' proof is
new. The counter-example in Section 6.7 is taken from Renardy [137] to show
that not every physically plausible system satisfies the spectrum-determined
growth condition.

Anda mungkin juga menyukai