Intcgr. equ. oper. theory 32 (1998) 332 - 353 0378-620X/98/030332-22 $1.50+0.20/0 Birkhfiuser Verlag, Basel, 1998 i IntegralEquations and OperatorTheory EXPONENTIAL STABILITY, EXPONENTIAL EXPANSIVENESS, AND EXPONENTIAL DICHOTOMY OF EVOLUTION EQUATIONS ON THE HALF-LINE NGUYEN VAN MINH*, FRANK RABIGER, ROLAND SCHNAUBELT** Let b/ = (U(t, s))t>s>0 be an evolution family on the half-line of bounded lin- ear operators on a Banach space X. We introduce operators Go, Gx and Ix on certain spaces of X-valued continuous functions connected with the integral equation u(t) = U(t, s)u(s) + f', U(t, r162162 and we characterize exponential stability, exponential expansiveness and exponential dichotomy of L/by proper- ties of Go, Gx and Ix, respectively. This extends related results known for finite dimensional spaces and for evolution families on the whole llne, respectively. INTRODUCTION Consider the non-autonomous linear evolution equation -~u(t)d = A(t)u(t), t E J = •+ or R, (NCP) on a Banach space X. If the operators A(t), t E J, are bounded (in particular, if X is finite dimensional) there is an extensive literature initiated by the work of Perron which connects asymptotic properties (of the solutions) of (NCP) with specific properties of the operator L defined by on a space of X-valued functions. For example, the existence of an exponential dichotomy of (NCP) is related to the invertibility of L if J = ]R and to Fredholm properties of L if J = R+ (see [BeG], [BGK], [Cop], [DaK], [MaS], [Pall and the references therein). If the * This work was done while the first author was visiting the Department of Math- ematics of the University of Tfibingen. The support of the Alexander yon Humboldt Foundation is gratefully acknowledged. The author wishes to thank R. Nagel and the Functional Analysis group in T~bingen for their kind hospitality and constant encourage- ment. ** Support by Deutsche Forschungsgemeinschaft DFG is gratefully acknowledged.
22
Embed
Exponential stability, exponential expansiveness, and exponential dichotomy of evolution equations on the half-line
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
E X P O N E N T I A L S T A B I L I T Y , E X P O N E N T I A L E X P A N S I V E N E S S , A N D E X P O N E N T I A L D I C H O T O M Y
OF E V O L U T I O N E Q U A T I O N S O N T H E H A L F - L I N E
NGUYEN VAN MINH*, FRANK RABIGER, ROLAND SCHNAUBELT**
Let b/ = (U(t, s))t>s>0 be an evolution family on the half-line of bounded lin- ear operators on a Banach space X. We introduce operators Go, Gx and Ix on certain spaces of X-valued continuous functions connected with the integral equation u(t) = U(t, s)u(s) + f', U(t, r162162 and we characterize exponential stability, exponential expansiveness and exponential dichotomy of L /by proper- ties of Go, Gx and Ix, respectively. This extends related results known for finite dimensional spaces and for evolution families on the whole llne, respectively.
I N T R O D U C T I O N
Consider the non-autonomous linear evolution equation
-~u(t)d = A(t)u(t), t E J = •+ or R, (NCP)
on a Banach space X. If the operators A(t), t E J , are bounded (in particular, if X is
finite dimensional) there is an extensive literature initiated by the work of Perron which
connects asymptotic properties (of the solutions) of (NCP) with specific properties of the
operator L defined by
on a space of X-valued functions. For example, the existence of an exponential dichotomy
of (NCP) is related to the invertibility of L if J = ]R and to Fredholm properties of L if
J = R+ (see [BeG], [BGK], [Cop], [DaK], [MaS], [Pall and the references therein). If the
* This work was done while the first author was visiting the Department of Math- ematics of the University of Tfibingen. The support of the Alexander yon Humboldt Foundation is gratefully acknowledged. The author wishes to thank R. Nagel and the Functional Analysis group in T~bingen for their kind hospitality and constant encourage- ment.
** Support by Deutsche Forschungsgemeinschaft DFG is gratefully acknowledged.
Van Mirth, Rabigcr and Schnaubelt 333
operators A(t), t E J, are unbounded the situation becomes more delicate and usually one
has to impose rather restrictive conditions in order to obtain similar results referring to
the operator L (e.g. see [Hen], [Zha]).
A less restrictive assumption is the well-posedness of ( N C P ) in the sense that
the solutions of ( N C P ) yield an evolution family 5l = (U(t, s)),>_,, ,,seJ of bounded linear
operators on X (see [Nic], [Paz], [RRS], [RS2], [Sch] and [Tan] for sufficient conditions and
examples). Note that solution does not necessarily mean 'classical' solution. Then instead
of the operator L one investigates the integral equation
u(t) = U( t , s )u(s ) + U(t ,~)f(~)d~, t > s, t , s S J, ( IE)
and its connection with asymptotic properties of the evolution family 7,/(see [Bus], [DaK],
[Dat], [LRS], [LeZ], [Zhi]). For instance, if J = I~ then it is shown in [LRS] that U has
an exponential dichotomy if and only if for every f E Cb(]~, X) there is a unique solution
u e Cb(R, X ) of ( IZ ) .
Another approach uses the so-called evolution semigroup 7" = (T(t))~>_o on a
space of X-valued functions induced by the evolution family/X (see [AuM], [LaM], [LaR],
[LMR1], [LMR2], [Mil], [Mi2], [Nic], [RRS], [RS1], [RS2], [Ran], [Sch] and the references
therein for this concept). Roughly speaking (if J = R) the evolution semigroup T is given
by
T( t ) f (~) = U(~, ( - t ) f ( ( - t), t >_ O,
(see (1.1) for the exact definition in the present context) and in certain situations its
generator G is an extension of - L = _ d +A(-) (see [LMR1, Prop. 2.9], [Sch, Prop. 1.13]).
It turns out that in the spirit of Perron's observations asymptotic properties of U like
exponential stability or exponential dichotomy can be described by spectral properties of
the operator G (see [AuM], [LaM], [LaR], [LMR1], [LMR2], [Mil], [Mi2], [RS1], [Rau],
[Sch]). However, we have to point out that most results in this direction are restricted to
the line case J = R.
In the present paper we characterize exponential stability, exponential expan-
siveness and exponential dichotomy of an evolution family U on the half-line J -- R+. Our
approach is based on the use of (generators of) evolution semigroups and their connection
with the integral equation ( IE) (see Section 1). Exponential stability and exponential
expansiveness of U is characterized in Theorem 2.2 and Theorem 2.5, respectively. Sec-
tion 3 deals with exponential stability of individual orbits. Our main interest, however,
is directed to the exponential dichotomy of U which will be characterized in Theorem 4.3
and Theorem 4.5. Concerning the relevance of exponential dichotomy and its far-reaching
applications we refer to [SaS] and the references therein.
334 Van Minh, R~biger and Sctmaubelt
1. P R E L I M I N A R I E S
Recall that a family of operators /2 = (U(t, s))t_>,_>o of bounded linear operators on a
Banach space X is a (strongly continuous, exponentially bounded) evolution family on the
half-line if
1. U(t, t) = Id and U(t, r )U(r , , ) = U(t, ~) for t > r > ~ > O,
2. (,, s) ~+ u ( t , ~)~ is continuous for every �9 �9 X ,
3. there are constants K > 0, ~ �9 R such tha t IIv(*,~)ll < K e ~(*- ') for * > ~ > 0.
Then w(/2) := inf{cr E R : there is K > 0 such that IlU(e,s)]l _< K e ~~ e > s > 0} is
called the growth bound of/2.
Throughout the whole paper the following function spaces (endowed with the sup-norm)
play an important role.
Co := {v: [0, oo) --+ X : v is continuous, v(0) = 0 = lim v(t)}, t - -~oo
c x := { v : [0, o~) -+ x : v is continuous, li_,~ ~(t) = 0},
Cx(to) := {qI,o,oo) : ~ �9 Cx}, to >_ O,
C b := {v : [0, co) ~ X : v is continuous and bounded}.
An evolution family g on X always defines an evolution scmigroup T = (T(t))~___0 of
bounded linear operators on Co and Cx by setting
V ( s , s - t ) v ( s - t ) , s > t , (I.1) [T(t)v](s) = U( s , 0 )~ (0 ) , 0 < ~ < t,
for v in Co and Cx, respectively. It can be easily seen that 7" is strongly continuous. We
denote the infinitesimal generator of 7" on Co and Cx by Go and Gx, respectively. Note
that D(Go) = D ( G x ) N Co and Gay = Gxv for v �9 D(Go), i.e. Go is the part of Gx in
Co.
The aim of this paper is to characterize asymptotic properties of a given evolu-
tion family U by (spectral) properties of the generators G0 and Gx. The following lemma
is the key tool in our strategy. It connects the operators Go and G x with the following
inhomogeneous integral equation
f u(t) = U(t,s)u(s) + U( t , ( ) f ( ( )d ( , t > s > 0. (1.2)
1.1 L E M M A . a) Let u , f E Co. Then u E D(Go) and Gou : - f if and only if
u(t) = U( t , ( ) f ( ( )d ( , t > O. (1.3)
Van Mirth, RAbiger and Schnaubelt 335
b) Let u E Cx and f E Co. Then u e D ( G x ) and G x u = - f if and only if (1.2) holds.
P R O O F . We only prove the first assertion. The proof of the second one follows
the same lines.
Let Gou = - f . The general theory of linear semigroups (see e.g. [Paz, p.4-5]) yields
/0 /0 T( t )u - u = T(~)Goud~ = - T(~) fd~ for t > 0.
Thus,
f0 t u = T( t )u + T(~)fd~.
From the definition of T we easily obtain that u is a solution of the integral equation (1.2)
and has the form (1.3).
Conversely, if u, f �9 Co and u satisfies (1.3)0 then by reversing the above argument we
obtain
f T( t )u - u = - T(~) fd~ for t > 0.
In part icular , this implies u �9 D(Go) and Gou = - f . []
In the following remark we collect some additional propert ies of the operators
Go and G x . For sake of convenience we set U(t, a) = 0 for 0 < t < s.
1.2 R E M A R K S . a) From Lemma 1.1 we immediately obtain tha t Go is injec-
rive and that ker G x = {u 6 C x : u(t) = U(t,O)u(O), t _> 0}.
b) The range R ( G x ) of G x is always contained in Co (in part icular G x is never invertible).
In fact, if u 6 D ( G x ) then
[Cxu](0) = lira T(t)u(O) - u(0) = lira u(0) - u(0) = 0. t,o t t$o t
c) Let to > 0, x �9 X and q0 : [0, oo) ~ ~ be continuously differentiable such tha t
~l[O,to) = O. Set u(t) = T( t )U( t , to )x , t > O, and f ( t ) --- T ' ( t )U(t , to)x, t >_ O. If
u �9 C x and f �9 C0 then an immediate application of L e m m a 1.1 yields u �9 D ( G x )
and G x u = - f .
Next we define an operator I x on C x connected with the integral equation (1.2).
If u, f E C x satisfy (1.2) we set
Ixu := f,
D ( I x ) := {u C C x : there is f C C x such tha t u and f satisfy (1.2)}.
336 Van Mirth, P,/ibiger and Scknaubelt
1.3 LEMMA. ( I x ,D( I z ) ) is a well-defined closed linear operator on Cx and
an e,te.sion of ( - a x , D(ax) ) .
P R O O F . Let I xu = f and I x u = g for u , f , g E Cx. Then
f * U(t,~)(f(~) -g(~))d~ = 0 for t > s > O.
By the continuity of the integrand we obtain 0 = U(t, t )( f( t) - g(t)) = f ( t ) - g(t), t > O,
i.e. f =g . The c]osedness of I x follows immediately from the definition of I x by the integral equation
(1.2). Finally, Lemma 1.1 shows that I x is an extension o f - G x . []
1.4 R E M A R K S . a) In general, - I x is a proper extension of Gx, in particular,
I x can be surjective (cf. Theorem 4.3). Moreover, the spectra of - I x and Gx are different.
In fact, for A E C with ReA < -w(U) the function uA : t ~+ eAtU(t,O)x, t _> 0, x E X,
belongs to Cx and Ixu~ -- Aux. Thus {A E C : ReA > w(U)} is contained in the (point)
spectrum of - I x whereas the spectrum of Gx is contained in a left half-plane (see [Paz,
Remark 1.5.4]). Note, however, that we always have k e r l x = k e r G x = {u E Cx : u(t) =
v(t, o)~(o), t > o).
b) From Lemma 1.1 it easily follows that Go is the part of - I x in Co, i.e. D(Go) = {u E
D ( I x ) N Co : I x u E Co} and Gou = - I x u for u E D(Go).
2. E X P O N E N T I A L S T A B I L I T Y A N D E X P O N E N T I A L E X P A N S I V E N E S S
In this section we characterize exponentially stable and exponentially expansive evolution
families U = (U(t, s))t>s>0 by means of the operators Go and I x and the integral equation
(1.2), respectively. First we give the definition of exponential stability and exponential
expansiveness of an evolution family.
be
2.1 D E F I N I T I O N . The evolution family U on the Banach space X is said to
a) exponentially stable if there are constants N,r, > 0 such that ]]U(t,s)I I <
Ne -~(t-8) for all t > s > 0 (or, in other words, if w(U) < 0),
b) exponentially expansive if U(t, s) is invertible and there are constants N, v > 0
such tha t lIu(t ,~)zlt _> N~ ' ( t - ' ) I I z l l for x E X a~a t > . > O.
We have the following characterization of exponentially stable evolution families.
Van MirLh, Ri~biger and Sehnaubelt 337
space X .
5) 5O
5ii) 5v)
2.2 T H E O R E M . Let 11 = (U ( t, s))t_>s__O be an evolution family on the Banach
Then the following assertions are equivalent:
N is exponentially stable.
Go is invertible.
For every f ~ Co the function t ~ uf( t) = f t U(t, ~)f(~)d~ belongs to Co.
For every f E Co the function t ~-~ uz(t) = f~ U(t, ~)f(~)d~ belongs to C' .
From the proof of the theorem we separate the following l emma for later use.
2 . 3 L E M M A . Let X : [to, t1) -+ (0, oo) be a continuous function and let e > 0
and K , a k 0 be constants such that x(t) <_ Ke ~(t-t~ and f toX( t )x ( r ~ <_ c for
t E [to,Q). Then x(t) <_ max(cK, K)e~+le-�88176
P R O O F o f L e m m a 2.3. If t E [to, t~) and t _< to + 1, then x(t) <_ Ke ~ < Ke ~+ } e- ~ (t-to).
Now let t E [to,t1) and t > to + 1. Set r := ftSoX(~)-ld~ for s E [to, t l) . By our
assumpt ion r _< cr and hence r > r for t l > sl _> so > to. Thus
1 c -~(t-to-1) < cKe~+}e-}(t-to). [] x(t) = r < r + 1--) e
P R O O F o f T h e o r e m 2.2. The implications (i) ~ (iii) ~ (iv) are obvious.
The equivalence of (ii) and (iii) follows from Lemma 1.1.
(iv) :~ (i): By assumpt ion B : Co -+ C b : f ~ uf is linear and everywhere defined.
Moreover, B is closed, and hence B is bounded. Let c := IIBII.
Now fix to _> 0 and 0 # y E X. Let h := sup{t > to : U(t, to)y # 0). Then U(t, to)y # 0
for t E [t0,tl) and U(t, to)y = 0 for t > tl. Se~ x(t) := [IU(t, to)yll, t E [to,tl). For n E N
sufficiently large choose a real continuous function qo, on [0, c~) such tha t qon has compact
suppor t contained in (to,t1), 0 < qon < 1 and
Let
qOn(t) = 1 for t E [ t o + - , m i n { n , Q - - }]. n
fo(t) = { ~n(t)x(t)-l~(t, to)y
Then f,~ E Co and
x(5) IIU(t, to)yH
for i E [to,t1), for t e [o, to] u It1, ~).
--II u ( t , ~ ) f . ( ~ ) d ~ l l _< cllAll = c
338 Van Minh, R~biger and Schnaubelt
for t E [to,t1). By letting n --4 co we obtain f jox(t)x(~)- 'd~ <_ c for ~ e [to,t1). More-
over , by the exponential boundedness of U we have X(t) < IC ('- ~ on t 0 . An
application of Lamina 2.3 yields
X(t) <_ max{cK, K}e ~+} e -}(t-t~ [lyll
for t E [ t0 , t l ) . Thus
HU(t, t0)yH _< max{cK, K}e •+} e -}(t-t~
for t _ to. Since the constants K, a and c are independent of to and y the exponential
stability of H follows. []
R E M A R K . The proof of (iv) =~ (i) follows ideas used in [DaK, Proof of The-
orem IV.3.3]. By different methods the equivalence of (i) and (iv) has been shown by C.
Bu~e [Bus, Theorem 1] (see also [Dat, Theorem 8] for a related result) and the equivalence
of (i), (ii) and (iii) is proved by Y. Latushkin, S. Montgomery-Smith and T. Randolph
[LMR2, Theorem 2.2, Corollary 2.3]. Related results in the autonomous case have been
shown e.g. by J. van Neerven [Nee] and Vfi Qu6c Ph6ng [Vfi]. Note that Theorem 2.2
also holds if we replace Co and C b by LP(R+, X), 1 < p < co, and consider the evolution
semigroup T on LP(IR+, X) induced by H (see [Dat, Theorem 6] and [LMR2, Theorem 2.2,
Corollary 2.3]).
An immediate consequence of Theorem 2.2 is the spectral mapping theorem for
the evolution semigroup on Co. Recall that or(B) denotes the spectrum of a linear operator
B and p(B) := C \ or(B) is the resolvent set. Moreover, s(B) := sup{ReA : A E e(B)} is
the spectral bound and r(B) := sup{]A] : A e a(B)) is the spectral radius of B.
2.4 C O R O L L A R Y . Let gt be an evolution family on the Banach space X.
Then the evolution semigroup T on Co satisfies the spectral mapping theorem
e = \ {0}, t > 0.
Furthermore, a(G0) = {A 6 C: Re), < s(G0)} is a left half-plane and a(T(t)) = {A 6 C:
IA[ < r(T(t))}, t > O, is a disc.
P R O O F . Let A E p(Go) and # E C such that Re# >_ Re),. Note that Go - Aid
is the generator of the evolution semigroup 7~ = (e-~tT(t)),>_o on Co induced by the
evolution family U~ = (e-~(t-s)U(t, s))t>_8>o. Theorem 2.2 implies that U;~ and hence/./~,
Van Minh, Riibiger and Schnaubelt 339
is exponentially stable and, as a consequence, that Go - #Id is invertible, i.e./~ �9 p(Go).
Thus a(G0) -- {), e C : ReA _< s(G0)} is a left half-plane.
If Re,~ > s(Go), then again by Theorem 2.2 the evolution fami ly /~ is exponentially stable,
and hence the induced evolution semigroup Tx is exponentially stable as well. In particular
r(e-a'T(t)) < 1, i.e. r(T(t)) < e ae~'t for t > 0. Thus r(T(t)) < e'(G~ > O. Together
with the spectral inclusion theorem e ' ' (a~ C_ c~(T(t)), t > 0 (see [Paz, 2.2.3]), it follows
that cr(T(t)), t > 0, is a disc and that the spectral mapping theorem holds. []
R E M A R K . The same result (with a different proof) can be found in [LMR2,
Theorem 1.2, Theorem 2.2, Corollary 2.3] and [Sch, Theorem 5.3]. The observation that
on Co the spectrum of T(t), t > O, is always a disc goes back to [Ran, Proposition 2].
We now come to the characterization of exponentially expansive evolution fam-
ilies.
2.5 T H E O R E M . Let l~ = (U ( t, s) )t>s>_o be an evolution family on the Banach
space X . Then the following assertions are equivalent:
(i) Lt is exponentially expansive.
5i) For every to >_ 0 and every f e Cx(to) there is a uniqu~ u: �9 Cx(to) such that
~s t us(t) = v ( t , , ) u : ( ~ ) + u( t ,~) f (Od~, t > ~ > to. (2.1)
R E M A R K . Condition (it) implies that the operator Ix is invertible. However,
Example 4.6 shows that the exponential expansiveness of /J (and hence (it)) is not equiv-
alent to the invertibility of Ix . On the contrary, if U(t, s) is surjective for all t > ~ > 0,
then from the proof of Theorem 2.5 it follows that /4 is exponentially expansive if and only
if Ix is invertible.
As above we separate from the proof of the theorem the following lemma which
is essentially contained in [DaK, Proof of Theorem IV.3.3].
2.6 L E M M A . Let X : [0, oo) --+ (O, oo) be a continuous function and let
c > 0 and K,c~ > 0 be constants such that X(T) <_ Ke~(r-t)x(t) , r > t >_ O, and
f ~ x( t )x(T)- ldT < c, t > O. Then there exists N > 0 only dependent on g , a and c
such that X(t) > ge~(~-~)~(*) for t > ~ > O.
340 Van Minh, 1L~biger and Schnaubelt
P R O O F . Let r f t x ( r ) - l d r , t >_ O. By our assumpt ion r < - c r
Thus r < r -~( t -~) for t > s > 0. On the other hand the exponential es t imate of X
yields o o 1 o o
x(t)(~(t) ~--X(t)~ ~-~dT >_ K-If e-a(~*-t)dT oo
__ : E =:
Moreover, by ou~ assumption, X(~)r -< e, t > ~ > O. Thus
# ~7 x ( t ) > - ~ > - ~ e o " > c X ( ) for t > s > 0. []
Now we are in a position to give the proof of Theorem 2.5.
P R O O F o f T h e o r e m 2.5. (i) =~ (ii): L e t / . / b e exponential ly expansive and
to _> 0. If f E Cx(to) then the function t ~ ul(t ) := - f ~ U ( r 1 6 2 belongs
to Cx(to). It is a s traightforward computat ion to show that u/ satisfies (2.1). If u E
c x ( t 0 ) satls~es (2.1) with y = 0, then ~(t) = U(~,t0)u(t0), t > to. Thus II~(t)ll >__
Ne~(t-to)[[u(to)][ for constants N, t, > 0. This yields u(to) = 0, and hence u = 0.
(ii) ~ (i): A) The assumption implies that Ix is invertible. S i n c e I x is closed its inverse
I x 1 is bounded and c := ][Ixll[ > 0.
B) Fix 0 # x E X. The unique solvability of (2.1) for f = 0 in Cx yields u(t) := U(t, O)x 0 for t > 0. For each n E N choose a real continuous function ~ , on [0, oo) such that
0 <: ~,, < 1, Tn = 1 on [0,n] and ~o,, = 0 on [ n + 1,oo). Let f n ( t ) : = -~.(t)llu(t)ll-lu(t) for t > 0. An easy computa t ion shows tha t
f , oo ~,(~) d~u(t), t > O, ~ . ( t ) : = II~(r -
and f~ solve (1.2), i.e. Ixun = f,,. Hence, ]]u,,][ < c[]f~]l = c. Lett ing n --+ co we obtain
. 1 L c ilu(Qlfd~ <_ Hu(t)}--~, t >_ O.
The exponential boundedness of U yields II~(~-)ll -< r < ~ ( ~ - 0 II~(~)lh ~ > ~ > 0, for constants K, a > 0 independent of x. By Lemma 2.6 there is a constant N > 0 independent of x
s~ch that Ilu(t)ll _> Ne~('-~)l l~(s)[ l for t > ~ > 0. C) It remains to show that U(t, s) is surjective for t > s > 0. Fix s > 0. Let y E X and
set v(t) := U(t, s)y, t > s. We choose a real continuous function T on [0, oo) with compact
support such that I~(t)l _< 1, supp~ C (s, ~ ) and
f o~ ~(~)d~ --- 1.
Van Minh~ RANger and Schnaubelt 341
Now consider the function w defined by
/ w(t) := v(t) ~,(~)d~, t > ~.
Then w e C x ( s ) a n d w is a solution of equation (2.1) with f ( t ) := -~o(t)v(t), t > s.
Extend f continuously to [0, co) by setting f ( t ) = 0 on [0, s]. Then there is a solution
z E Cx of equation (2.1) on [0, co). By our assumpt ion zl[8,~ ) = w . Thus
a s a consequence [[Pol[ -< 2")'t~ 1. It remains to show that there is a constant c > 0
(independent of to) such that 7to >- c. For this fix xk E Xk, k = 0, 1, with Ilxo II = HXlll = 1.
346 Van Mirth, I~biger and Schnaubek
By the exponential boundedness of U w e have IIU(t, to)(xo + ~1)11 ~ Ke~'(t-t~ + =1 II for t >_ to and constants K, a >__ O. Thus
11=o + ~111 -> K-~e-~(*-*~ to)=o + U(t, to)=lll > K - 1 e-a( t - to) (N-1 eg(t-to) _ Ne-~( t - to) )
= : ct-to~ t ~ to~
and hence 7to -> ct-to. Obviously cm > 0 for m sufficiently large. Thus 0 < C m <~ "[to-
b) Fix t > 0, 0 _< so _< t and z e ke rP( t ) , and let (s,,) be a sequence in [0,t] converging
to so. There is y E ker P(0) such that U(t, O)y = x. By the strong continuity of/,( we have
1 ~ l l u ( ~ . , t ) f z - U(~o,t)lxll = l i p IIu(~., 0)ty - Y(~o, 0)tyll = 0.
c) Note tha t
liP(t) x - P(s)zl i -< liP(t) x - P( t )U(t , s )x l l + l lU(t ,~)P(s)x - P(s)xBt
<_ (sup I[P(r)li)li~ - u ( t , s)~ll + llY(t, ~)P(~)~ - P(Ox[I r>_O
for x E X and t > s > 0. By the strong continuity of U we obtain tha t P( .) is strongly con-
tinuous f rom the right. In order to show strong continuity from the left set Q(.) := I d - P ( . )
and fix t > 0 and x E X . For 0 < s < t we have O(~)~ = U ( ~ , t ) i U ( t , s ) O ( ~ ) ~ =
U(s,t)lQ(t)U(t , s)x. By b) the family (U(s,t)l),e[o.t ] C s P ( t ) , Z ) is strongly continu-
ous and uniformly bounded. The strong continuity of N yields lim~1-t Q(t)U(t, s)x = Q(t)x.
Thus
lira O(s)x = lira U(s, t)l e ( t )U( t , s)x = O(t)=, sit s~t
i.e. Q( . )~ and h e . c e p ( . ) ~ is continuous from the Ie~t.
Assertions d) and e) are immediate consequences of a). []
We come to our first main result. It characterizes evolution families with an
exponential dichotomy by conditions on the operators Go, Gx and Zx, respectively.
4.3 T H E O R E M . Let/d = (U(t, s))t>~>o be an evolution family on the Banach
space X . Then the following assertions are equivalent:
(i) ld has an exponential dichotomy.
(ii) The range R(Gx) of Gx coincides with Co and X0(0) is complemented in X .
(iii) I x is surjective and Xo(0) is complemented in X .
Van Minh, P~biger and Schnaubelt 347
P R O O F . (i) :# (ii): Let P(t), t > O, be the family of projections given by
the exponential dichotomy. Then X0(0) = P(O)X, and hence Xo(0) is complemented. If
f E Co define v : [0, oo) --> X by
/0' v(t) = U(t,~)P(~)f(~)d~ - U(t , ( ) l ( Id - P(())f(~)d~. (4.1)
An easy computa t ion shows tha t v E Cx and v is a solution of equation (1.2). By Lemma
1.1 we have G z v = - f . Together with Remark 1.2 b) this implies R ( G x ) = Co.
(i) ~ (iii): We can use the same arguments as in the proof of (i) =r (ii). Note tha t for
f E Cx the function v defined by (4.1) is also in Cx and satisfies (1.2). Hence by the
definition of I x we have v E D ( I x ) aad I x v = f , i.e. I x is surjective.
(ii) ~ (i): A) Let Z C X be a complement of X0(0) in X , i.e. Z = X0(0) @ Z. Set
X l ( t ) := U(t,O)Z, t > O. Clearly,
u ( t , , ) X o ( , ) c_ Xo(t) , t 2> , , (4.2)
U(t,S)Xl(S) = Xl ( t ) , t 2> s.
B) There are constants N, u > 0 such tha t
llU(t,O)xt[ 2> Ne~('-8)llU(*,0)xll for x E XI(0) and t > s > O. (4.3)
In fact, let Y := {v E D ( G x ) : v(0) E XI(0)} endowed with the graph norm I[vflGx :=
IMI + I laxvll . Then Y is a closed subspace of the Banach space (D(Gx) ,
[l" [lax), and hence Y is complete. By Remark 1.2 a) we have k e r G x = {v E Cx :
v(t) = u ( t , o > fo~ so~e ~ ~ Xo(O)}. Since X = Xo(O) | X~(O) and n ( a x ) = Co we
obtain tha t Gx : Y --+ Co is bijective and hence an isomorphism. Thus there is a constant
> 0 such tha t
]IGxvll 2> ullvllax 2> vllvll for v E Y. (4.4)
Let now 0 5d x E XI(0) and set u(t) := U(t,O)x, t 2> O. By Remark 1.2 a) we have
u(t) ~ 0 for all t 2> 0. For each n E N choose a real continuous function ~o,~ on [0, oo)
such tha t 0 <_ ~o~ _< 1, ~n = 1 o n [1,n] and~o~ = 0 on { 0 } U [ n + l , oo). Set f~(t) :=
-~ , ( t )nu( t ) I i - lu ( t ) , t 2> O, and
u. ( t ) :-- i t - ~ l d ~ ( t ) , t 2> 0.
Then fn E Co and u , E Cx. An easy computat ion shows that fn and u,~ satisfy (1.2). By
L e m m a 1.1 we have u,, E D ( G x ) and Gxu,, =- -f,~. From (4.4) we obtain [Ifnll 2> ~ll~dl.
Lett ing n -+ co this yields
HvJ,t"~'r162 < . - ' II~(t)ll -~, t 2> 0.
348 Van MinlL Ra'biger and Selmaubelt
Now the exponential boundedness of U and Lemma 2.6 imply tha t there is a constant
N > 0 independent o f x such that Ilu(t)]] _> Nd'(t-S)ll~(s)]l, t > s > O.
C) X = Xo(• ) @Xl ( t ) , t > 0:
Let Y C_ Cx be as in B). Then D(Go) = D(Gx) M Co C y and (4.4) yields IiG0vil _> vlivll
for v E n(Go). Thus 0 ~ Aa(Go) n iN and CoroUary 3.3 implies tha t Xo(t) is closed for
4_>0. From (4.2), (4.3) and the closedness of Xl(0),we derive that Xl(t) is closed and Xo(t) n X1(4) = {0} for t _> 0.
Finally, fix g0 > 0 and x E X. Choose a real continuous function T on [0, oo) such tha t ~ has oo d eo compact support contained in [t0, ec) and fro W(r r = 1. Set v(t) := fi ~(~)d~ U(t, to)x
and f( t) := -T(t)U(t , to)x, t >_ to. Then v is a solution of equation (2.1), and v E Cx(to).
Extend f continuously to [0, ~ ) by setting fif0,t0) -- 0. Then f E Co and by assumpt ion
there exists w E Cx such that Gxw = - f . In view of Lemma 1.1 w is a solution of
equation (1.2). In particular, Wl[to,oo ) satisfies (2.1). Thus
v ( t ) - w ( t ) = U( t , to ) (v ( to ) - w( to ) ) = U(t , t o ) ( x - w( to ) ) , t > to.
Since v - wl[to,~ ) E Cx(to) this implies x - w(to) E X0(t0). On the other hand w(0) =
w0 + wl with wk E Xk(0), k = 0, 1. Then w(to) = U(to,O)wo + U(to,O)wl and by (4.2)
we have U(to, O)wk E Xk(to), k = 0, 1. Hence x = x - w(to) + w(to) E Xo(to) + Xl(to).
This proves C).
D) Let P(t) be the projection from X onto Xo(t) with kernel Xl ( t ) , t __ 0. Then (4.2)
implies P(t)U(t,s) = U(t,s)P(s), t >_ s >_ O. From (4.2) and (4.3) we obtain that
U(t,s)l : k e r P ( s ) --+ ker P( t ) , t > s > 0 is an isomorphism. Finally, by (4.3), Theorem 3.2
and our assumpt ion Acy(Go) M iR # iN there exist constmats N, v > 0 such tha t
tlcr(t,,)~H _< Ne-~(*-~)ll~ll for �9 E P ( , ) X , t > ~ > O,
tlU(s,t)lxll <_ Ne-V(t-8)ltxll for x E ke rP ( t ) , t > s > 0.
Thus /2 has an exponential dichotomy.
(iii) ~ (i): We use exactly the same arguments as in (ii) =~ (i). We only have to replace
the opera tor Gx by Ix . []
R E M A R K . Par t C) of the proof shows that for an evolution family/2 with an
exponential dichotomy the generator Go of the evolution semigroup on Co always satisfies
0 r Aa(Go), and hence Az(Go) n iN = 0 (see Section 3, Remark a)).
If X is a Hilbert space we only have to assume the closedness of the stable
subspace.
Van Minh, R/ibiger and Schnaubelt 349
4.4 C O R O L L A R Y . Let l,l = (U(t,s))t>8>o be an evolution family on the
Hilbert space H. Then the following assertions are equivalent:
(i) l~ has an exponential dichotomy.
(ii) R ( G H ) = Co and H0(0) is closed.
5ii) Iu is surjective and Ho(0) is closed.
In the proof of Theorem 4.3, (ii) =~ (i), we saw that the invertibility of Gx
restricted to a certain subspace of Cx plays a crucial role in order to prove the existence of
an exponential dichotomy for a given evolution family/A. In our next result we show that
the existence of an exponential dichotomy can be even characterized by such an invertibility
condition. Let us introduce the following notion. For a closed linear subspace Z of X let
Cz := { f e c x : f(o) e z } .
Denote by Gz the part of Gx in Cz, i.e. D(Gz) = D(Gx) N Cz and Gzv = Gxv for
v E D(Gz) . Then Go is the part of Gz in Co. In the same way let I z be the restriction
of I x to C2, i.e. D(Iz ) = D( I x ) N Cz and I zv = I x v for v E D(Iz) . Notice that the
evolution semigroup 7" l~ves Cz invariant and that Gz is the generator of the restriction
of T to Cz.
With this notat ion we obtain the following characterization of evolution families
with exponential dichotomy. A similar result is shown in [BGK, Theorem 1.1] for the finite
dimensional case (see also [Pal]). We point out that in contrast to Theorem 4.3 we do not
have to assume that X0(0) is complemented.
4.5 T H E O R E M . Let l.f = (U(t, s) ),>8>_o be an evolution family on the Banach
space X and let Z be a closed linear subspace of X . Then the following assertions are
equivalent:
(i) bl has an exponential dichotomy with ker P(0) = Z.
Oi) Gz : D(Gz) C Cz --+ Co is invertibIe.
6ii) I z : D( I z ) :_ Cz --+ Cx is invertible.
P R O O F . (i) ~ (ii) : Let P(t), t > O, be a family of projections given by the
exponential dichotomy such that ker P(0) = Z. Then P ( 0 ) X - X0 (0) and X = X o ( 0 ) � 9 Z.
Fix f E Co. By Theorem 4.3 there is v E D(Gx) such that Gxv = f . On the other hand
u : [0, oo) --+ X : t ~+ U(t, 0)P(0)v(0) belongs to Cx and Cxu = 0 (cf. Remark 1.2 a)).
Then v - u E D(Gz) and Cz(v - u) = Gxv = f . Hence Gz : D(Cz) -+ Co is surjective.
If w E k e r G z then w(t) = U(t,O)w(O), t > 0 (cf. Remark 1.2 a)). Since w 6 Cx we have
w(0) E Z N Xo(0) = {0} and hence w = 0, i.e. Gz is injective.
350 Van Mirth, IEa'biger and Schnaubelt
(i) ~ (iii) can be shown by the same arguments. One only has to use Remark 1.4 a)
instead of Remark 1.2 a).
(ii) :* (i) : Let a z : D ( a z ) -~ Co be invertible. By Remark 1.2 b) we have Co =
R(Gz) = R(Gx) . The closedness of Gz implies that G z 1 is bounded, and hence there
exists v > 0 such that Ilazvll >_ ullvll for all v E D(Gz) . Since Go is the part of Gz
in Co we obtain 0 ~ A~r(Go). Finally we show that X = Xo(0) @ Z. Corollary 3.3
implies that Xo(0) is closed. Now let x E X. If U(t, 0)x = 0 for some t > 0, then
x E Xo(0). Otherwise U(t,O)z 7 ~ 0 for all t >_ 0. Let ~o : [0, oo) --+ IR be continuously
differentiable such that ~O][od I = 1 and TI[2,~) = 0. Set u(t) := T(t)U(t,O)x, t >_ O, and
f ( t ) := T'(t)U(t, O)z, t >_ O. By Remark 1.2 c) we have u E D ( G x ) and Gx u = - f . On
the other hand there exists v E D(Gz) such that Gzv = - f . Thus u - v E ker Gx and
hence
- v)(0 = - = - t > o .
Since u - v E CN this implies x - v(0) E Xo(0). Thus x = x - v(0) + v(0) E Xo(0) + Z.
If y E Z N X0(0), then w defined by w(t) := U(t,O)y, t >_ O, belongs to Cz n k e r G x (see
Remark 1.2 a)). Hence Gzw = 0 and by the invertibility of Gz we have w = 0. Thus
0 = w(0) = y, i.e. Xo(0) M Z = {0}. This shows X = Xo(0) | Z. The assertion now follows
from Theorem 4.3.
(iii) ~ (i) is shown by the same arguments as (ii) ~ (i). []
R E M A R K . As a special case of Theorem 4.5 the invertibility of Gx resp.
I x characterizes evolution families with an exponential dichotomy such that Xo(0) =
P ( 0 ) X = {0}.
We conclude with an example of an evolution family U with non-trivial expo-
nential dichotomy such that I x is invertible. In particular this shows that invertibility of
I x is not equivalent to the exponential expansiveness of/d (el. Theorem 2.5).
4.6 E X A M P L E . Let X = L 1[0,0~). For 0 < s < t < 1 set
and f o r l < s < t s e t
0, 0 < ~ < t - s , Ul( t , s ) f ( ( ) := / ( ~ - t + . s ) , t - s < ~ ,
U2(t,s)f(~) > 1,
Van Mirth, RRbiger and Schnaubclt 351
f E X. Then L/= (U(t, s))t>s>o defined by
{ U~(t,s), 0 < s < t < 1, U(t,s):= U2(t, 1)Ul(1,s), s < l <t,
u 2 ( t , s ) , 1 < s < t,
is an evolution family on X with an exponential dichotomy. The corresponding family of
projections P(t), t >_ 0, is given by
P(t) f = X[o,min{1,t}]f, f E X,
where Xc denotes the characteristic function of a set C. In particular, {0} = P(O)X = X0(0). Theorem 4.5 implies that Ix is invertible. Since ]lU(t, 1)P(1)]] < e -(t-l), t >__ 1,
the evolution family H cannot be exponentially expansive.
R E F E R E N C E S
[AuM]
[BaV]
[Bee]
[BGK]
[Bus]
[Cop]
[DaK]
[Dat]
[Hen]
[Kat]
[L~M]
Aulbach B., Nguyen Van Minh, Semigroups and exponential stability of nonau- tonomous linear differential equations on ~he half-line, (R.P. Agrawal Ed.), Dy- namical Systems and Aplications, World Scientific, Singapore 1995, pp. 45-61.
Batty C., Vfi Quoc Phdng, Stability of individual elements under one-parameter semigroups, Trans. Amer. Math. Soc. 322 (1990), 805-818.
Ben-Artzi A., Gohberg I., Dichotomies of systems and invertibility of linear ordinary differential operators, Oper. Theory Adv. Appl. 56 (1992), 90-119.
Ben-Artzi A., Gohberg I., Kaashoek M.A., InvertibiIity and dichotomy of differ- ential operators on the half-line, J. Dyn. Differ. Equations 5 (1993), 1-36.
Bu~e C., On the Perron-BelIman theorem for evolutionary processes with expo- nential growth in Banach spaces, preprint.
Coppel W.A., "Dichotomies in Stability Theory", Springer-Verlag, Berlin Hei- delberg, New York, 1978.
Daleckii Ju. L., Krein M.G., "Stability of Solutions of Differential Equations in Banach Spaces", Amer. Math. Soc., Providence RI, 1974.
Datko R., Uniform asymptotic stability of evolutionary processes in a Banach space, SIAM J. Math. Anal. 3 (1972), 428-445.
Henry D., "Geometric Theory of Semilinear Parabolic Equations", Springer- Verlag, Berlin, Heidelberg, New York, 1981.
Kato, T., "Perturbation Theory for Linear Operators", Springer-Verlag, Berlin, Heidelberg, New York, 1966.
Latushkin Y., Montgomery-Smith S., Evolutionary semigroups and Lyapunov theorems in Banach spaces, J. Funct. Anal. 127 (1995), 173-197.
352 Van Mirth, P~'biger and Schnaubelt
[LMm]
[LMR2]
[La~]
[LRS]
[LeZ]
[MaS]
[Mil]
[Mi2]
[Nee]
[Nic]
[Pall
[Paz]
[RRS]
[RS2]
[Rau]
[SaS]
[Sch]
Latushkin Y., Montgomery-Smith S., Randolph T., Evolutionary semigroups and dichotomy of linear skew-product flows on locally compact spaces with Ba- nach fibers, J. Diff. Eq. 125 (1996), 73-116. Latushkin Y., Montgomery-Smith S., Randolph T., Evolution semigroups and robust stability of evolution operators on Banach spaces, preprint.
Latushkin Y., Randolph T., Dichotomy of differential equations on Banach spaces and an algebra of weighted composition operators, Integral Equations Oper. Theory 23 (1995), 472-500. Latushkin Y., Randolph T., Schnaubelt R., Exponential dichotomy and mild so- lutions of nonautonomous equations in Banach spaces, to appear in J. Dynamics Diff. Equations. Levitan B.M., Zhikov V.V., C'Almost Periodic Functions and Differential Equa- tions', Cambridge Univ. Press, 1982. Massera J.J., Schi~ffer J.J., "Linear Differential Equations and Function Spaces', Academic Press, New York, 1966.
Nguyen Van Minh, Semigroups and stability of nonautonomous differential equa- tions in Banach spaces, Trans. Amer. Math. Soc. 345 (1994), 223-242. Nguyen Van Minh, On the proof of characterizations of the exponential di- chotomy, preprint. van Neerven, J.M.A.M., Characterization of exponential stability of a semigroup of operators in terms of its action by convolution on vector-valued function spaces over l{+, J. Diff. Eq. 124 (1996), 324-342. Nickel G., On evolution semigroups and weIl-posedness of non-autonomous Cau- chy problems, PhD thesis, Tiibingen, 1996. Palmer K.J., Exponential dichotomy and Fredholm operators, Proc. Amer. Math. Soc. 104 (1988), 149-156. Pazy A., "Semigroups of Linear Operators and Applications to Partial Differ- ential Equations", Springer-Verlag, Berlin, Heidelberg, New York, 1983. Rs F., Rhandi A., Schnaubelt R., Voigt J., Non-autonomous Miyadera perturbations, preprint. Rs F., Schnaubelt R., The spectral mapping theorem for evolution semi- groups on spaces of vector valued functions, Semigroup Forum 48 (1996), 225- 239. Rs F., Schnaubelt R., Absorption evolution families with applications to non-autonomous diffusion processes, Tiibinger Berichte zur Funktionalanalysis 5 (1995/96), 335-354. Rau R., Hyperbolic evolution semigroups on vector valued function spaces, Semi- group Forum 48 (1994), 107-118. Sacker R., Sell G., Dichotomies for linear evolutionary equations in Banach spaces, J. Diff. Eq. 113 (1994), 17-67. Schnaubelt R., Exponential bounds and hyperbolicity of evolution families, PhD thesis, T~bingen, 1996.
Van Mirth, RAbiger and Sehnaubelt 353
[Tan] Tanabe H., "Equations of Evolution", Pitman, London, 1979. [Vfi] Vfi Qu6c Ph6ng, On the exponential stability and dichotomy of Co-semigroups,
preprint. [Zha] Zhang W., The Fredholm alternative and ezponential dichotomies/or parabolic
equations, J. Math. AnM. Appl. 191 (1995), 180-201. [Zhi] Zhikov V.V., On the theory of the admissibility o/pairs o/function spaces, Soviet
Math. Dokl. 13 (1972), 1108-1111.
Nguyen Van Minh, Department of Mathematics, University of Hanoi, 90 Nguyen Trai, Hanoi, Vietnam. Frank Rs and Roland Schnaubelt, Mathematisches Institut, Unlversitgt Tiibingen, Auf der Morgenstelle 10, D-72076 Tiibingen, Germany email: [email protected] and