analytic continuation and stability of operator semigroups
TRANSCRIPT
A N A L Y T I C C O N T I N U A T I O N A N D
S T A B I L I T Y O F O P E R A T O R S E M I G R O U P S
By
RALPH CHILL AND YURI TOMILOV*
Abstract. We prove a new criterion for the analytic continuation of functions across a linear boundary. As corollaries, we obtain new conditions for convergence of orbits of operator semigroups on Banach spaces with Fourier type.
1 Introduct ion
The study of analytic extendability of a function across a (in most cases linear)
boundary is a classical subject in analysis. Besides its natural value for function
theory [6, 11 ], it is of importance for the theory of partial differential equations
[24, 15, 16], harmonic analysis (study of ideals in Banach algebras via the Carleman
transforrn) [7, 23, 29], operator theory [35, 39], and mathematical physics [9]. The
first systematic treatment of analytic extension criteria seems to be [10].
One such criterion is provided by the well-known edge-of-the-wedge theorem;
see [38]. It says that two analytic functions f + and f - which are defined in wedges
(open regions) and which coincide in the distributional sense on their common edge
admit a common analytic extension to a neighbourhood of the edge. This principle
is convenient if some a priori information on the behaviour of the integral means
of the two "wedge" functions f • near the edge is available (e.g., local Hardy
space conditions, as for instance in [44, Theorem B] or in [10]). However, if both
functions f • coincide only in the sense of (nontangential) pointwise convergence,
the classical edge-of-the-wedge principle cannot be applied for several reasons.
Nevertheless, it is possible to find an appropriate edge-of-the-wedge theorem even
in this setting if the f • satisfy certain growth restrictions near the edge. See, for
example, [40, Theorem 4.4] where a corrected version of [44, Theorem E] was
proved (and also the comments in [5]).
*This work started during a visit of the first author at the University of Torun. The kind hospitality is gratefully acknowledged. The second author was partially supported by a KBN grant and by the NASA-NSF Twinning Program.
331 JOURNAL D'ANALYSE MATHE/vlATIQL1E, Vol. 93 (2004)
332 R. CHILL AND Y. TOMILOV
In this article, we present a criterion which implies that an analytic function f
defined on the region {z E C �9 -1 _< R e z , I m z <_ 1} \ (-1, 1) extends analytically
through the interval ( -1, 1). If F(z) := f ( z ) - f(2), then we assume that F can be
estimated in the form
(1.1) IF] < IGl. tHI,
where G satisfies an integral condition, and H gatisfies a pointwise convergence
condition. Very formally, this can be considered as a "product" of the two types
of conditions above. While one condition is on nontangential behaviour of H, i.e.,
behaviour in sectors perpendicular to the real axis, the other one is on behaviour of G along the parallels to the real axis. This "orthogonality" of the conditions leads
to difficulties in applying known arguments (a situation which is close in spirit can be found in [I, p. 158]).
Our interest in such problems stems from the study of the asymptotic behaviour
of operator semigroups. In this theory, functions with an estimate of the form (1.1)
appear very naturally when considering weak local resolvents and when using
the resolvent identity. We give several applications of our extension criterion to
stability of bounded operator semigroups on Banach spaces with Fourier type. In
this situation, the main task is to relate the growth of the resolvent of the semigroup
generator near the imaginary axis, the type of the underlying Banach space and the
asymptotic properties of semigroup orbits, thus leading to theorems of tauberian
character. Such a relation is expressed finally in terms of certain properties of
images of the semigroup generator. This part of the paper can be considered as a continuation of the research started in [12]. We discuss also similar results in the
context of discrete operator semigroups.
2 Pre l iminaries
We start by recalling some very basic facts from the theory of Hardy spaces of
analytic and harmonic functions. Throughout, we denote by I~ the open unit disc and by "IF the unit circle.
Def in i t ion 2.1. Forp > 0, HP(D) is the set of all analytic functions f on D for which
(2.1) (f02~ ) l /p
[IflIHp := sup If(reir de < oo. 0_<r<l
Defin i t ion 2.2. For p _> 1, hP(l~) is the set of all harmonic functions f on D
A N A L Y T I C C O N T I N U A T I O N 333
for which
( [2 , , )1/, (2.2) ]]ft]hp := sup ]f(rei•)] p de < oo.
O < : r < l \ J o
It is well-known that i fp > 1, HP(I~), resp. hP(D), are Banach spaces equipped
with the norms []. ]]Hp, resp. []. [Ihp. Moreover, the map
I : f ~ - ~ P f
is an isometric isomorphism from LP(T) onto hP(l~) for every 1 < p < c~, where
P f stands for the Poisson integral of f . For f E LP(~), 1 < p <_ cx~,
lim (Pf)(r~) = f(~), a.e. ~ E V, r---~ 1 -
SO that a function in LP(q~), 1 < p _< c~, can be identified with the boundary value
o f P f E hV(lD); see [3, pp. 103-105, 116].
For f belonging to either of the introduced spaces, we have the estimate
C C (2.3) [f(z)l <- (dist(z,OID))l/P - (1 - Iz l ) l /p ' z e 1~,
for some C > 0 (see, for example, [3, p. 121]). Clearly, similar definitions and
properties can be stated if ID is replaced by any other open disc in C.
Our further reasoning is based on two maximum principles. The first is the
so-called Gabriel inequality, which can be considered as an integral maximum principle for harmonic functions.
Theorem 2.3 (Gabriel's m a x i m u m principle). Let f~ C ~2 be a simply
connected domain. Suppose that F1, F2 C f~ are two closed convex curves such that ['2 is contained in the interior o f the domain f~l bounded by F1. Then for every p > 2, there is a constant Cp (not depending on FI, F2 and ~) such that for every
harmonic function u on ~,
(2.4) ~r2 [u(z)[P ,dz[ < Cp fr l [u(z)[" [dz[.
The inequality (2.4) was first proved by R, M. Gabriel in [19]. A modem
approach to Gabriel-like inequalities can be found in [22] (in particular, see [22, Theorem 1 ] and the comments following it).
Remark 2.4. Our main result (Theorem 3.1 below) can be improved if (2.4) also holds for p > 1. It is known that (2.4) holds for p > 1 if u is analytic on
f~ [19, 22]. Thus, by passing to conjugate functions we may reduce the problem
334 R. CHILL AND Y. TOMILOV
whether the estimate (2.4) holds forp > I to the problem of extending the M. Riesz
inequality (1 < p < 2) for Lp-norms of conjugate functions to general domains
with Lipschitz boundary. This latter problem requires the verification of the
Muckenhoupt condition on the modulus of a conformal mapping of the domain f21
to the unit disc. Unfortunately, we have difficulties with this approach even when
f~l is a convex polygon. In general, the question whether (2.4) holds for 1 < p < 2
seems to be open. Note that a sharper version of the inequality (2.4) appears to be
false even i fp = 2 [18].
R e m a r k 2.5. Recall that if the inside curve F2 is convex, then the measure #
defined by #(E) ".'= length (I'2 f3 E) (E C 11~ 2 Borel set) is a Carleson measure; see,
for example, the discussion in [2 1 ]. Hence, if the outside curve F 1 is Lipschitz, then
the inequality (2.4) for 2 < p < ~ can be derived from Dahlberg's characterization
of Carleson measures; see [14, Theorem 1] and the remark following it. If F 1 is
a C 1-curve, then it is possible to extend the range of allowed p in Theorem 2.3 to
p > 1. For C2-curves this result was obtained by HSrmander [26, Theorem 3.1].
However, the reasoning in the proof of our Theorem 3.1 requires the consideration
of a Lipschitz curve F1. Moreover, in order to include p = 2 in Theorem 3.1, we
need to know that the inequality (2.4) holds for some p < 2.
The second maximum principle which is crucial for us gives the estimate of
a subharmonic function on the unit disc in terms of its radial boundary value. In
the following general form, the result is due to Dahlberg [13, Theorem 1]. If u is
harmonic, a proof may be found in Wolf [43, pp. 95-97].
T h e o r e m 2.6 ( D a h l b e r g ' s m a x i m u m principle) . Let u be a nonnegative
subharmonic function on D. Suppose that
1. f o rever ) , ~ E 72 there exists l im~x_ u(r~) =: / (~) , and f E LI(T);
2. we have
sup u ( z ) = o ( ( i - r ) - 2 ) , r i - .
Izl=r
Then
(2.5) u < P f on D,
where, as before, P stands for the Poisson integral.
Observe that if u is a harmonic function on D, if ]u] satisfies the conditions of
Theorem 2.6, and if in addition f E LP(72), then u E hP(/l)). This fact will be used
later.
ANALYTIC CONTINUATION 335
3 Analytic continuation
This section is devoted to the proof of our main result, an analytic extension
criterion.
We define the rectangle
(3.t) R := {z E C : - 1 _< Rez < 1, - 1 < Imz _< 1};
and for 0 E (0, rr/2) we let
Z0 := {z E C : 0 < argz < 7r- 0}.
By C we denote constants which may vary from line to line.
T h e o r e m 3.1. Let f : R\I~ -+ C be analytic, and define F : R\IR -+ C by
F(z) = f ( z ) - f (2) (z E R\~) . Assume that
1. there exists a constant m E [0, 2) such that
sup I : (a +i/3)1 -- O(I/~l-'~), D ~ 0; aE(-1,1)
2. there exist a measurable function G : R \ R ~ C and a continuous function
n : R\Ii~ --+ C such that IFI <_ IGI. Inl and
(3.2) sup lie(" + i/~)ll~:, < ~ f o r s o m e 2 < p <_ oc, ~E(0,1)
and there exists 00 E (0, 7r/2) such that
(3.3) Jim l log(Imz)H(z) l = 0 forevery a E ( -1 ,1) . zEc~+~g O
Then the function f admits an analytic extension to 1L
P r o o f . In order to simplify the notation, we define the function
/ : ( 0 , 1 ) ~
/3 ~ l(f~):--Ilog(/3/2)l.
We define the set S C ( -1 , 1) of singular points by
S := {a E ( -1 , 1) : the function f does not have an analytic
extension near the point a}.
336 R. CHILL AND Y. TOMILOV
If the set S is empty, then the claim is proved. We assume therefore that the set
S is nonempty and we show that this leads to a contradiction.
S tep 1: Define, for every n E N,
S n : = { a E S : sup I I ( Imz)n(z ) l<n} zEa+EOo
= N { a E S : sup II(Imz)H(z)i <_ n}. t3>0 *E'~+~0~
I m z >/~l
By continuity of the functions l and H, the set S,~ is closed for every n E N. By
assumption (3.3),
S - - U s n . nEN
Thus, by Baire's category theorem, there exists no such that Sn0 has nonempty
interior in S, i.e., there exists an interval (a, b) C (-1, 1) such that
SA (a,b) = Sno M (a,b) r 0.
Since the set (a, b)\S is open, it is the countable union of disjoint intervals
(ak, bk), i.e.,
(a, b)\S = U (ak, bk). kEN
We define in the following for every k E N and every 0 E (0, ~r/2) the rectangle
(3.4)
the set
(3.5) T ~ := {z E Rk :
and the disc
Rk := {zE R : a k < R e z <bk},
]arg(z--ak)] <O and ]arg(z-bk) - r I <0},
{ z Ibk~ ak + bk < (3.5) Dk := z E Rk : 2 ~ "
Note that the set Tk ~ is a rhombus for all suffucuently large k.
For every k E N, the function f is analytic in the rectangle R~ (and, in particular,
also in the set Tk ~ and the disc Dk), and therefore the function F is harmonic in Rk. Moreover, there exists a constant M > no such that
(3.7) ]l(Imz)H(z)] <_ M forevery z e U Rk\:F~ ~ kEN
Note that the constant M may be strictly larger than no since the end points of the
interval (a, b) need not belong to Sn0.
ANALYTIC CONTINUATION 337
x
a~,
' ' Rk t
S j /
r
j r
/
Figure 1. The rectangle Rk, the rhombus T ~ and the disc Dk.
Step 2: We define for every k �9 N and every 0 �9 (0, 7r/2) the sets
A ~ fqR and B ~ fqR;
these are, in general, triangles if 0 is close to 7r/2 and k is sufficiently large.
Rk
" , ' - , ' Dk
a'k"
Figure 2. The sets As and Bk.
We show in this step that for every k E N, every s E (2, p) and every 0 �9 (00, 7r/2),
(3.8) F e LS(OAek)
and
(3.9) F e LS(OBg).
338 R. CHILL AND Y. TOMILOV
In fact, we prove only the first relation; the second one can be proved similarly.
Assume first that
A~ = {z E C : z - ak E Eo and lmz < 1},
i.e., the set A ~ is in fact a triangle. k In order to prove F E LS(OA~), we show first that the function F is s-integrable
along the two non-horizontal sides of the triangle OA ~ In fact, we show for every
fixed s E (2,p) and for almost every 0 E (00, 7r - 00) that
f l / s i n 0
(3.10) IF(ak + rei~ dr < oo. ,tO Indeed, by assumption (2), inequality (3.7) and H61der's inequality,
fa~-oo fl/sinO IF(ak + reiO)ls dr dO o Jo
< [ ' [,cotoo IF(ak + + i )1" d,a 30 d - 3 cot 0o
~ ~ folfl_lfl(p_s)/p ( f 'Br176 )s/P IH(a, + a + ifl) a(ak + a + i/~)1 p da d~ \ J - 3 cOt ao
(f?l )s/p < C sup IG(a +ifl)l p da , 3~(od)
where C is a constant depending only on 00, p, s and M.
Since the right-hand side of this inequality is finite by (3.2), the function under
the integral on the left-hand side must be finite almost everywhere. Thus we have
proved that (3.10) holds for almost every 0 E (00, ~r - 80).
Since the integral in (3.10) is the integral of IFI s along one non-horizontal side
of the triangle OAf, and since the integral of lFl s is bounded along the upper side of
that triangle, we deduce from (3.10) that for almost every 0 E (00, 7r/2) the relation
(3.8) holds.
Now fix 0 E (0o, ~r/2) such that the relation (3.8) holds. Note that for every
e E (0, �89 the function F is harmonic in a neighbourhood of the cut-offregion
A~ A {z e C: e < I m z < 1 - c} =: A~(e).
Hence, by Gabriel's inequality (2.4), there exists a constant C > 0 such that for
every e E (0, �89 and every 0' e (0, 7r/2),
fo tF(z)l ~ l d z t<C fo IF(z)t s [dzl. (3.11) A,~' (e) - - A~(e)
ANALYTIC CONTINUATION 339
Observe that, by assumption (2), we have
fa~+e cot Oo lim ] IF(o~ + ic)l s dc~ = O.
~.-+0+ J ak-e cot 0o
From this equality, the inequality (3.11) and the continuity of F in R \ (a, b) we
obtain
foAg' IF(z)l" tdzl = limsup f IF(z)l" ldzl e~o+ JOA~'(~)
C limsup / IF(z)l s ldzl < e--~O+ JOAn(e)
~--- C j/0A ~ I F ( z ) l ~ ldzt.
This implies that (3.8) holds for every 0' E (0, 7r/2), if it holds for 0 E (00,7r/2).
Thus we finally obtain that the relation (3.8) holds for every 0 E (00, 7r/2).
If A ~ is not a triangle as we first assumed, then we need only some obvious
technical changes in the proof above in order to see that the relation (3.8) holds also
in this case. This follows from the fact that the function F is uniformly continuous
away from the real axis.
S tep 3: We show that for every k E N and every s E (2,p),
(3.12) F E hS(Dk).
For this, we remark first that for every s E (2,p) and every k E N,
(3.13) [ IF(z)l" Idzl < o0. Jo Dk
Indeed, for every 0 E (00, 7r/2) the integral of the function IFI" over the arcs
ODk M A ~ and ODk M B ~ is finite (A ~ and B ~ are defined as in Step 2). This follows
from (3.8), (3.9) and Gabriel's inequality if one applies again the reasoning with
appropriate cut-off regions as it was done in Step 2. On the remaining bit of OD k in
the upper half plane (if any), the function IFI" is bounded, and thus also integrable.
Since the function F is antisymmetric with respect to the real axis, the relation
(3.13) follows. Second, we observe that the assumption (1) implies that for every 7 E ( �89 1/m)
and every r E (lk/2, lk)
fo2~r [ ( + ) ['Y - -m f2~rSo f ak + re ir de < C I k "~ ] I sin r dr < c%
340 R. CHILL AND Y. TOMILOV
where, as before, lk = (bk - ak)/2 is the radius of the disc De. By the definition of
H "r, this implies that for every 3' E (�89 1/m),
f E H-r(Dk).
Hence, by the definition of F and (2.3), for every 7 E (�89 l /m), there exists a constant C > 0 such that
(3.14) IF(z)[ <~ C(Zk -lzl) -a/-r, z e Dk.
Third, we note that the radial limit
(3.15) lim F ( ak + bk r--+l- \ 2
+ rlke w ) =: F*(e i~ exists for every 0 E [0, 27r).
For 0 E [0,2~r)\{0,Tr}, this follows from the continuity of F in R\IL whereas for
0 = 0 or 0 = 7r, it follows from the fact that the function F is identically 0 on the
interval (ak, bk).
The claim (3.12) now follows from (3.13), (3.14), (3.15) and from Dahlberg's
maximum principle (Theorem 2.6).
S tep 4: Let 01 E (00,7r/2) be such that tan01 = 2tan0o. We prove that for
every s E (2,p) and almost every 0 E (0o,01),
(3.16) E IIFII~,(OTZ) < oo. kEN
Note that for every k large enough, the set T~ is a rhombus, so that its boundary
consists of four line segments, one of which is the curve
bk - ak } p 0 : = z E C : arg(z - ak) = O and ]z -- ak] <_ "2~os-O "
Let lk := (bk - ak)/2 be the radius of the disc Dk. Then we obtain from Fubini's theorem
fa81 ft~/cos 0 dO = E IF(ak + rei~ dr dO
k E N o dO
l t an 01
<- E tF(ak + reW)[Sr-1 r dr dO. k EN o JO
By changing from polar coordinates to cartesian coordinates and using the
ANALYTIC CONTINUATION 341
assumption (2), we obtain
< E fl-x IF(ak + c~ + ifi)l s da dfl kEN J0 J/3 cot 01
f0 /~ tan01 1 [Zc~176176 +a+i f l )H(ak +a+ifl) l(~)l s dad~ -< fi J cot0, kEN
k~EN/lktanO, f f l cot 00 < MS 1 IG(ak + a + ifl)l" da dfl -- . f l l ( f l ) s JecotO,
< MS 1 ]C(a + ifl)l ~ da dfl - f l t ( f l ) s
s _<C" sup IG(a + ifl)[ s da < oo. ~e(O,1)
Since the right-hand side of this inequality is finite, the left-hand side is finite,
too; and therefore the function under the integral must be finite almost everywhere.
This means that for almost every/9 E (/90,/gx),
~ / ~ IF(z)l~ 'dz' < ~176
Proceeding in a similar way for the other three sides of the sets T ~ we obtain
for almost every/9 E (/90,01),
S t e p 5: We show that
L If(z)l s Idzl < kEN T~
(3.17) lira F(. + ifl) = 0 weakly in L s(a, b). ~--+0+
Define 01 as in Step 4, and choose 0 E (0o,01) such that the relation (3.16)
holds.
Fix k E N. For every e E (0,/k/2), we define the cut-off region
Ck~ := {z E TO: Rez E (ak + r -- e)}.
Note that
lim e--+O+ s IF(z)l~ [dzl = ~T: IF(z)l~ Idzl"
342 R. CHILL AND u TOMILOV
In order to see this, it suffices to show that the integral of IF[ ~ along the vertical
lines in the boundary of C o (e) tends to 0 as e -+ 0+. This follows from the fact
that F C hS(Dk) for every s E (2,p) (Step 3) and, for example, from Gabriel's
inequality. Alternatively, one may also use the Poisson integral representation of
F. Note that the length of the vertical lines tends to 0 as e tends to 0.
In the argument using Gabriel's inequality, we consider, for example, the in-
tegral along the vertical line on the left and let 2 < s < s' < p. From HSlder's
inequality and Gabriel's inequality, we obtain for every e > 0 small enough
f etanO I F ( a , + e +i/3)1 ~ d/3 - e tan 0
(f.t.nO ~'1" <- (2etanO)(S'-sl/s' \d-etan0 IF(ak + e + i/3) s d/3)
< ce ("'-~)/"' IIFII~.,(D~),
with a constant C which depends only on O, s and s'. Letting e tend to 0 implies
the above claim.
Hence if/3 E (0, 1), then again by Gabriel's inequality,
IF(a + i/3)1" da < C L IF(z)] ~ Idzl ~<~ - C ~ ( s ) a+i~ET~
for all sufficiently small e. The constant C depends only on s. Letting e tend to O,
we obtain
J'/-<o<~, IF(a + i/3)1" da ___ 6' Jo f IF(z)l' Idzl
for every/3 E (0, 1), where the constant C again depends only on s.
The assumption (2), the construction of the set Sn0, the above inequality,
inequality (3.7), and inequality (3.16) from Step 4 imply that for every/3 E (0, 1),
L b ]F(~ + i/3)[ s da
= / ]F(a + i/3)]s da + / IF(a + ifl)]~ da J(a ,b)ns J(a,b)\S
fo rL:' < IG(a + ifl)I s da + IF(a + i/3)1 s da ,b)NS kEN
ANALYTIC CONTINUATION 343
c/. < IG(~ + i~)l ~ d~ + IF((~ + i~)l ' dc~ -- 1 ~_<(,<_b~ kEN ~+i~r~
f , + iZ)I' IF(~ da k e n a + i ~ E T 0
--- 2 \ l o g 2 / j_lm(~+iB)l sac+m IF(z)l~ Idzl - r2
< C,
where C _> 0 is some constant independent of /7. Consequently, the net
(F(. + i~));~e(0,1) is bounded in L~(a, b).
Let now ~ E 79((a, b)\S). It follows from the definition of S and from the fact
that ~ has compact support in (a, b)\S that F(. + i~)lsupp~ converges uniformly to
0 as fl --+ 0+. In particular,
lira F (a + i~)qo(c~) dc~ = O. ~---}0+
Since the net (F(. + i~))ne(o,1) is bounded in LS(a, b) and 7)((a, b)\S) is dense in L s" ((a, b)\S) (s* being the conjugate exponent of s), this implies
lim f F(a + i~)g(a) da = 0 ~-+o+ J(~,b)\s
for every g E L ~" (a, b). On the other hand,
lira [ F(a + iB)g(a) da < lim f IG(a + i$)H(c~ + i~)l Ig(~)l da ~-+0+ J(a,b)nS -- ~-+0+ J(a,b)nS
=0
for every g E L s* (a, b). This follows from the assumption (3.2) and the fact that
lim Iig(a + i ~ ) g ( t ~ ) i l L . * ( ( a , b ) n S ) = 0 f~-~0+
by the dominated convergence theorem. Thus we have proved (3.17).
S t e p 6 : We prove that the function f extends analytically through the interval
(a,b). This follows, in fact, from the classical edge-of-the-wedge theorem. It suffices
to show that for every test function qo E 79(a, b), one has
lim f b L b ~ o + Ja ( f (a + ifl) - f ( a - ifl)) ~(fl) dfl = ~-~o+lim F (a + ifl)qa(fl) d~ -- O.
344 R. CHILL AND Y. TOMILOV
This follows directly from (3.17) in Step 5.
We have finally obtained a contradiction to the assumption that S fq (a, b) is
nonempty. Therefore, the assumption that S is nonempty was not true. The claim
is proved. []
R e m a r k 3.2. Theorem 3.1 would be improved if the assumption (3.3) on the
function H could be replaced by the weaker assumption
(3.18) }im H(z) = O. ZE~o0 +C~
Indeed, the stronger assumption (3.3) on the function H is needed only in Step 4.
It is possible to replace it by (3.18) in other places of the proof by applying the
theory of the Dirichlet problem on Lipschitz domains (and its consequences for
Hardy type spaces on such domains) developed in [14] (see also [31]). However,
we have not been able to weaken it just to (3.18).
This led to the additional term log a in the Theorems 4.2 and 4.9 from the next
section about the stability of operator semigroups. The question whether in general
it is possible to replace (3.3) by (3.18) remains open.
In our next result, we show that if the set of singular points of f is sufficiently
thin, then it is possible to replace the condition (3.3) by the weaker condition
(3.18). For this, we give the definition of a set withfinite 7-entropy.
Defin i t ion 3.3. A closed set E C [a, b] is called a set with finite 7-entropy
('7 > 0) if [a, b]\E = Un%l In for a sequence (In)n>_1 of open intervals such that
oo
(3.19) Z II,~l I l o g l / n l r < oo. r~-~l
R e m a r k 3.4. Note that every closed set E C [a, b] has finite 0-entropy. Sets
with finite 1-entropy are simply called sets with finite entropy in the literature (see
[25] and also [28], where such sets appeared under a different name). Note that
if E is a set with finite entropy, and if E has Lebesgue measure zero, then E is a Carleson set.
Note that if E is such that ~n~__l II, d ~ < oo for some a E (0, 1), then E is a set
with finite "r-entropy for every ~, >_ 0. If, in addition, E has Lebesgue measure 0,
then it is also called a C1-~ set [27]. Such sets have also been studied in [36].
Let R be defined as in (3.1).
ANALYTIC C ONTINUATION 345
T h e o r e m 3.5. Let f : R\I~ ~ C be analytic, and define F : R\I~ --~ C by
F(z) = f ( z ) - f (2) (z �9 R \~ ) . Assume that
1. there exists a constant m �9 [0, 2) such that
sup If(c~ +ifl) I = O(lAI-m), /3 ~ 0; ~e(--1,1)
2. there exist a measurable function G : R \ ~ -~ C and a continuous function
H : R \ R --+ C, such that IFI _ IG[ IHI and
sup Iia(" + i3)llL, < o0 for some 2 < p < oo, /AE(0,1)
(3.20)
and
(3.21) !irn IH(z)l = 0 for every a E (-1, 1) and f ixed 00 �9 (0, 7r/2); zE~+~O 0
3. the set S o f singular points o f f on (-1, 1) is a set with finite 7-entropy for
some 7 > P/(P - 2).
Then the function f admits an analytic extension to R.
Proof . The proof is similar to the proof of Theorem 3.1. Let the set S of
singular points be defined as in the proof of Theorem 3.1, and assume that S is
nonempty. We show that this leads to a contradiction.
S tep 1: In the first step, we let for every n E N,
Sn := {c~ E S : sup In(z)l <_ n}. zE a+I ]%
The sets Sn are closed by the continuity of H, and by assumption (3.21),
S = t J S,~. nEN
Hence, by Baire's category theorem, there exists no E N and (a, b) C (-1, 1) such
that
S M (a,b) = Sno M (a,b) # O.
Let the rectangle Rk, the set Tk ~ and the disc Dk be defined as in (3.4)--(3.6).
Then, for every k �9 N, the function f is analytic in the rectangle Rk. Moreover,
there exists a constant M >_ no such that
(3.22) Ig(z)l < M for every z �9 U Rk\T~ ~ kEN
346 R. CHILL AND Y. TOMtLOV
Step 2 and S tep 3 are identical as in the proof of Theorem 3.1. They prove
that for every k E N and every s E (2, p)
F ~ hS(Dk). Step 4: Let 81 E (0o,7r /2) be such that tan01 = 2tan00. We show that for
every s E (2,p) small enough and almost every 0 E (0o, 01),
F 8 (3.23) ~ 1 1 NL,(OT D < 00. kEN
As in Step 4 of Theorem 3.1, we obtain
J;;i _;!o,.,=o, r oo 0, IF(z)l" Idzl dO < 8 -1 IF(ak + a + i~)I: da dE, o ~ J~ cot 01
where F0k and Ik are defined as in the Step 4 of Theorem 3.1. Hence, by HSlder's inequality and the inequality (3.22),
io' (z i ) <: 8 -1 l(o,/~ tan 01) (fl) __ jc~176 IF(ak + ~ + i f l ) l s da 48 \kEN dJ cot 01
~ MsLI~-I( E t~OI)(P-s)ip (Jfa b ) sip IG(a + ifl)l p da dr le Ei~l
fl<l k tan 01
By assumption (3), there exists 7 > P/(/P - 2) such that for every/~ E (0,1),
Ilogni < c IkllogZ i < C < 0 o . tan 01 -
~E~ kEN fl<lle t a n 01
Note that the constant on the right-hand side of this inequality is independent of
B e (0,1). The preceding two inequalities and assumption (3.20) imply that for s e (2,p)
small enough (so that 7 > P/(P - s)),
/o~ fr~ IF(z)l 8 ldzl d O < ex).
Hence, the function under the integral is finite for almost every 0 E (0o,0~).
Proceeding in a similar way for the other four sides of T ~ we have thus proved that inequality (3.23) is true for almost every 0 E (8o, 01).
ANALYTIC CONTINUATION 347
One may now proceed as in S t e p s 5 and 6 of the proof of Theorem 3.1 in
order to obtain that the function f extends analytically through (a, b). This is a
contradiction to the construction of Sn0 and (a, b). Hence, the assumption that S is
nonempty is false, and the claim is proved. [3
Theorem 3.1 has the following analogue for the case of unit disc.
Define R := {z E C : 1/2 < tZl < 2, e iargz E I} ,
where I E T is an open interval. For 0 E (0, lr/2), let
E o : - { z E C : - O < a r g z < O } .
T h e o r e m 3.6. Let f : R \ T --+ C be analytic, and define F : R \ T -+ C by
F(z ) = f ( z ) - f (1 /5) , z E R \ T . Assume that
1. there exists a constant m E [0, 2) such that
sup [f(rei~ = 0(11 - [ r l l - rn ) , r --+ 1; 0e[o,2,r]
2. there exist a measurable function G : R \ T --4 C and a continuous function
H : R \ T ~ C, such that IFI <_ IGI �9 IHI and
sup IlG(rei')llL~(t) < oo for some 2 < p <_ c~, re(l,2)
and f o r somef ixed 0o e (0, lr/2)
lim l l o g [ 1 - 1 z [ t g ( z ) [ = O for every ~ = e i'~ E I. z E ~ O 0
Then the function f admits an analytic extension to R.
P r o o f . Let
/~ := {z E C : - l o g 2 <_ Imz < log2, a < Re z < b},
where a, b E I~ are such that I = {e iz : a < z < b}. Define the function g : /~ ~ C
by
g(z) = f(Hz), z k .
Then g satisfies the assumptions of Theorem 3.1. Hence, g admits an analytic
extension to the interval (a, b). This, however, is equivalent to the fact that f admits
an analytic extension to I. []
348 R. C H I L L A N D Y. T O M I L O V
4 Appl icat ion to stability of operator semigroups on Banach spaces with Fourier type
Let X be a complex Banach space, and (T(t))t>_o be a C0-semigroup on X. We
say that (T(t))t>o is stable if for every x E X,
lira I IT ( t )x l l = 0. t--4oo
Because of many interesting connections with various areas of analysis, the
problem of characterizing the stability of C0-semigroups has attracted considerable
attention in recent years (see the recent accounts [2], [34]). However, the stability
of semigroups is still far from being understood. Necessary and sufficient stability
conditions are available only in exceptional cases.
In this section, we present some new sufficient algebraic stability conditions
which are close to being optimal in Banach spaces with Fourier type; cf. [12].
We start by introducing some notation. For every closed linear operator A on
X with dense domain D(A), we denote by R(A, A) its resolvent defined on the
resolvent set p(A) and by a(A) the spectrum of A. By s we denote the Banach
space of all bounded linear operators on X.
Recall that a Banach space X has Fourier type p E [1, 2] if the Fourier transform
on the vector-valued Schwartz space S(R; X) extends to a bounded linear operator
from LP(II~; X) into Lq(~; X) (q := _x_l~, i.e., if the Hausdorff-Young inequality p - 1 /
holds.
By the Lemma of Riemann-Lebesgue, every Banach space X has the trivial
Fourier type p = 1. Thus, by interpolation, if X has Fourier type p E [1, 2], then
it has Fourier type p' for every p' E [1,p]. If a Banach space X has Fourier type
p E [1, 2], then the dual X* has the same Fourier type p. A Banach space has
Fourier type p = 2, i.e., Plancherel's Theorem holds, if and only if it is isomorphic
to a Hilbert space. The space Lp(f~,d#) has the Fourier type m i n { p , p / ( p - 1)}.
Most of the results relevant to the Fourier type can be found in the survey [20].
Our approach to proving stability of semigroups is based on the following
characterization of stability in terms of complete trajectories. Recall that a function
F : ]I~ --+ X is called a complete trajectory for a semigroup (T(t))t>o if
F(t + s) = T(t)F(s), s E IE, t E ~+.
L e m m a 4.1. Let (T(t) )t>o be a bounded Co-semigroup in a Banach space X.
Then the semigroup (T(t))t>o is stable i f and only i f (T*(t))t>o does not admit a nonzero bounded complete trajectory.
1With the usual convent ion for p = 1 o r p = oo. Note that l i p + 1/q = 1.
ANALYTIC CONTINUATION 349
This statement was proved in [4, Theorem 3.1] (see also [41]).
The following result shows the interplay between the stability of a semigroup,
the growth of the resolvent of the generator and the geometry of the underlying
Banach space. Compare this theorem with [8, Theorem 5] and [40, Theorem 3,4].
T h e o r e m 4.2. Let X be a Banach space having Fourier type p E [1, 2], and
let q = p/(p - 1). Let (T(t))t>o be a bounded Co-semigroup on X with generator
A. Assume that for every fl E 11~ there exists an open interval I 9 /3 and a dense set
M C X such that for every 13' E I and every x E M,
(4.1) lim I logal Ilal/qR(a + il3',A)xll = O. c~---~0+
Then the semigroup (T(t) )t>_o is stable.
P r o o f . For p = 1 and p = 2, the claim actually follows from [5, Theorem 2.4]
and [40, Theorem 3.4], respectively. Hence we can assume that p E (1,2).
Let F : ~ ~ X* be a bounded complete trajectory for the adjoint semigroup
(T*(t))t>o. Replacing F by the weak* convolution F �9 qo for some ~ E LI(I~), we
can without loss of generality assume that F is uniformly continuous. Note that
F �9 ~ is nonzero if F is nonzero and ~ is chosen appropriately.
The Carleman transform fi" of F, defined by
{foe -ZtF(t) dt, Rez > 0,
F ( z ) : = - f ~ e-ZtF(t) dt, R e z < 0 ,
is analytic in C\ilIL We show that it is in fact entire.
Indeed, choose/3 E IL and let I 9 /3 and M C X be as in the theorem. Let
x E M, and put f ( t ) := (F(t), x) (t E I~). Then the Carleman transform ] = (F', x)
is analytic in C.\i~, and it satisfies the estimate
(4.2) sup I](a + i/3')1 = O(1/a), a --+ 0 + . /3'ER
We show that ] extends analytically through iI. By [5, Lemma 6.1], for every
a > 0 and every/3' E IL
(4.3) P ( - a + i/3') = R(a + i/3', A*)F(O) + 2 a R ( a + i/3', A * ) F ( - a + i/3').
Moreover, the fact that F is a complete trajectory implies that
R(z,A*)F(O) = -f'(z) w h e n R e z > 0.
350 R. C H I L L A N D Y. T O M I L O V
Hence, for a > 0 and/3' E IL we have
I](a + ii3') - ] ( - a + il3')1 -- 12<al/P/~(-a + i13'), a l / qR (a + i13', A)x) I
_< Ia(a + i~')11g(a + i~3')1,
where
G(a + ir := 112al/PP(-a + il3')11 and H(a + i/3') := Ilal/qR(a + il3',A)xll.
Note that the boundedness of the function F and the Hausdorff-Young
inequality imply
(4.4) sup IIG(a + i')IILq(R) < oo. a > 0
Moreover, from the assumption (4.1), the resolvent identity and the bounded-
ness of the semigroup (T(t))t>o, we obtain for every 190 E (0, 7r/2) and every ~' E I,
lim sup flog a I H(a + i~") a-+O+
l~It'-Btl_<a tan 0 0
< lim sup I log al Ila'/q(R(a + i f l" ,A) - R (a + i~',A))xll ~.--~0+
I ~ " - - B ' l < a tan e o
_< lira sup Ilog al Ila t a n O o R ( a + i l 3 " , A ) a l /qR(a+i l3 ' ,A )x l l c,--.* 0+
l~"-B*l_<~ tan 0 0
< tan00 sup IlZ(t)ll limsup I log al Ilal/qR(a + il3', A)x[I t > 0 a - - ~ 0 +
=0 .
It follows from this inequality and the estimates (4.2) and (4.4) that we can apply
Theorem 3.1 in order to see that the Carleman transform ] extends analytically
through the interval iI.
Hence, by [37, Proposition 0.5 (i)], the support of the distributional Fourier
transform of f = (F, x) does not intersect the interval I for any x E M. Since the
function F is bounded and M is dense in X, this implies that the support of the
distributional Fourier transform of F does not intersect the interval I.
Since/3 E ~ was chosen arbitrarily, the support of the distributional Fourier
transform of F is empty, i.e., F = 0.
Hence there exists no nonzero bounded complete trajectory for (T * (t))t>o. By
Lemma 4.1, the semigroup (T(t))t>_o is stable. []
T h e o r e m 4.3. Let X be a Banach space having Fourier type p E [1, 2], and
let q = p / (p - 1). Let (T(t))t>o be a bounded Co-semigroup on X with generator
ANALYTIC CONTINUATION 351
A. Assume that ia(A) M 1~ has Lebesgue measure 0 and that ia(A) fl [a, b] is a set
with finite 7-entropy for some 7 > q/(q - 2) and every a < b. Assume in addition
that for every 13 E ~, there exists an open interval I 9 13 and a dense set M C X
such that for every 131 C I and every x E M,
(4.5) lim a l/q R(ce + i13', A )x = O. or--+0
Then the semigroup (T(t) )t>_o is stable.
P roo f . The claim for p = 1 and p = 2 was proved without any restriction on
a(A) in [5, Theorem 2.4] and [40, Theorem 3.4], respectively. We may therefore
assume thatp E (1, 2).
Observe first that the nonnegativity and concavity of the function f : (0, +co) -~
N, x ~ x(log x)-r on the interval (0, e 1-'~) imply that f is semiadditive on the same
interval, i.e.,
ak) _< Z S ak) k~N kEN
for every summable sequence (ak) C (0, oo) such that ~k~r~ ak < e 1-'r. Hence, a
closed subset of a set with Lebesgue measure 0 and finite 7-entropy is again a set
with finite 7-entropy. Indeed, let S C S' c I~ be two compact sets, and assume that
S I has Lebesgue measure 0 and that it is a set with finite 7-entropy. Let a < b, and
let [a, b] \ S I = Un~N I~ and [a, b] \ S = U,~N 1, for two sequences (I~) and (In)
of mutually disjoint open intervals. Since S' has Lebesgue measure 0, we find for
every interval I,~ a family of intervals ( I~)m~m. such that In \ S I = Umbra, I~ and
II.I = ErnEM. I t ' l . Since S C S', the index sets M , are mutually disjoint. The claim follows from semiadditivity by noting that only finitely many of the I,~ can
have measure greater than e 1--r.
Let F : II~ --+ X* be a bounded complete trajectory of (T*(t))t>o. As in the
proof of Theorem 4.2, we may assume that F is uniformly continuous. It follows
from [41, Proposition 3.7] that the Carleman spectrum of F,
sp(F) := {13 E IR : i13 is singular point for #}
(~" being the Carleman transform of F), satisfies the relation
(4.6) isp(F) C or(A) M ilk
Hence, by assumption, sp(F) ~ [a, b] is a set with finite 7-entropy for every a < b.
As in the proof of Theorem 4.2, one can show that sp(F) is empty. For this, we
use the assumption (4.5) and the analytic extension criterion from Theorem 3.5.
Since sp(F) is empty, the function F must be 0.
352 R. CHILL AND Y. TOMILOV
Hence there exists no nonzero bounded complete trajectory for (T * (t))t>o. By
Lemma 4.1, the semigroup (T(t))t>_o is stable. E]
From Theorems 4.2 and 4.3, we obtain as corollaries two range conditions
implying stability. In order to state these corollaries, we introduce some more
notation.
Def in i t ion 4.4 ([331, Definhqon 3.1.1, Corollary 5.1.12). Let A be a closed
linear operator with dense domain D(A) such that
(4.7) (0, oo) C p(A) and sup AIIR(A, A)[ [ < co. A>0
For each 7 E (0, 1), define the 7-thfractional power of - A to be the closure of
the operator
sin77rTr fo ~176 dA, x E D(A). (4.8) (-A)Tx . -
For arbitrary 7 > 0, set
( -A) ~ := (-A)'r-['~](-A)[X].
Note that if A generates a bounded C0-semigroup, then for every/3 E ~ the operator A - i t satisfies (4.7). Thus, the fractional powers of i/3- A are well-defined
in this case. Note that for every fl E IR and every v > 'y ___ 0,
(4.9) Irn(i/3 - A) v C Im(ifl - A) ~.
Observe that if A satisfies (4.7), then for every A > 0, the operator R(A, A)
satisfies (4.7), so we may define the fractional powers of R(A, A) by Definition 4.4.
Moreover, by [32],
(4.10) supA':IIR(A,A)'ql < co, 7 E (0,1). A>0
The following variant of the classical Mean Ergodic Theorem has been proved
in [32] (see also [42] where bounded semigroups were considered).
T h e o r e m 4.5. Let A be a closed linear operator with dense domain such that
(4.7) holds. F/x 7 E (0, 1], x E X. Then x E Im A if and only if
lim a'rR(a, A)'rx = O. a--tO+
I f Im A is dense in X, then for everyfixed 7 E (0, 1],
Im(-A) "r = {x E X : lim R(a, A)'~x exists in X} . a-+O+
ANALYTIC CONTINUATION 353
From this theorem and Theorem 4.2, we obtain the following stability criterion.
C o r o l l a r y 4.6. Let X be a Banach space having Fourier type p E (1, 2].
Let (T(t))t>o be a bounded Co-semigroup on X with generator A. If, for some
E (I/p, 1],
(4.1 1) N Ira(i/3 - A) "r is dense in X, BcR
then the semigroup (T(t) )t>_o is stable.
Proof . The casep = 2 was proved in [12] with 7 = �89 so we may suppose that
p E (1, 2). By Theorem 4.5, for every/3 E II~ and every z E N~ea Ira(i/3 - A) "r,
lira R(a + i/3, A)~z exists in X. a-+0+
Moreover, the assumption (4.1 I) and the equality
Im (i/3 - A) = Im (i/3 - A)~
(see [33, p. 142]) imply that Im (i/3 - A) is dense in X for every/3 E I~. Hence, by
(4.10) and Theorem 4.5,
lim a l - ' rR(a + i~, A)z = 0 a---+O+
Since 7 > 1/p, this implies
lim I logat Ilal/q/~(a + i/3, A)zll -- 0 a-+O+
for every z E n Im(i/3 - A) "r. /~ER
for every z E n Im(i/3 - A) "~, /3ER
where q = p/(p - 1). Since the set Im(i/3 - A) "~ is dense in X by assumption,
the claim follows from Theorem 4.2. []
R e m a r k 4.7. It was proved in [12] that the condition (4.11) with 7 < 1/p
does not, in general, imply the stability of a bounded semigroup (T(t))t>o on a
Banach space X having Fourier type p E [1,2]. The question whether Corollary
4.6 remains true for 7 = 1/p (/9 ~ 1, 2) is open (compare with Remark 3.2).
Under an additional restriction on the spectrum a(A), the condition (4.11) with
"7 = 1/p does imply stability.
Corollary 4.8. Let X be a Banach space having Fourier type p E [1, 2].
Let (T(t))t>_o be a bounded Co-semigroup on X with generator A. Assume that
354 R. CHILL AND Y. TOMILOV
ia ( A ) M ~ has Lebesgue measure 0 and that ia (A) N [a, b] is a set with finite v-entropy
f o r some V > p/ (2 - p) and every a < b. I f
N Im(ifl - A) 1/p is dense in X , tiER
then the semigroup (T(t))t>o is stable.
Proof . The claim follows from Theorem 4.3, using arguments similar to those
in the proof of Corollary 4.6. Note that V > P~ (2 - p ) is equivalent to V > q~ (q - 2)
if q = p/ (p - 1) is the conjugate exponent. []
Statements similar to Theorems 4.2, 4.3 and their Corollaries 4.6, 4.8 are true
for bounded discrete semigroups (Tn)n>0 C s
For example, the following theorem is a discrete analogue of Theorem 4.2.
T h e o r e m 4.9. Let X be a Banach space having Fourier type p E [1, 2], and
let q = p / (p - 1). Let T be a power bounded linear operator on X . Assume that
f o r every ~ E ~, there exists an open interval I C ~, ~ E 1, and a dense set M C X
such that f o r every ~' E I and every x E M,
(4.12) lim I log(r - 1)l II(r - 1 ) l / q R ( r ~ ', T)xll = O. r---r 1 +
Then, f o r every x E X ,
( 4 . 1 3 ) l im IIT"xll = O. n--4oo
Proof . We only sketch the proof, which is similar to the proof of Theorem
4.2, and leave the details to the reader; see also [5].
Recall that (4.13) holds if and only if (T*n)n>0 does not admit a nontrivial
complete bounded trajectory which is now a bounded sequence g : Z ~ X* such
that gn+m = T*ngm for every m E Z and every n E N (see [41]).
So let (gn)nez be any bounded complete trajectory for (T*n),~>0. The Carleman
transform of (gn)nez is defined by
( E~=I zl, g(n), Izl > 1 0(z) := -1 1 n
- ~ n = - ~ z - + ' f f ( ), Izl < 1
and is analytic in C\qr.
In a way similar to that in the proof of [5, Lemma 6.1 ], one obtains the identity
~(z) = R(z*,T*)g(O) + (z - z*)R(z* ,T*)~(z) , z E ~), z* := 1/2.
ANALYTIC CONTINUATION 355
Using this identity, the Hausdorf f -Young inequality and, finally, Theorem 3.6,
we obtain that ~ extends to an entire function.
From limlzl~oo tlO(z)ll = 0 and Liouvil le 's Theorem, we obtain that t~ and thus g
are trivial. Therefore, there does not exist a bounded nonzero complete trajectory
for (T*n)n>O. []
I f T E Z:(X) is power bounded, then for every ~ E "I~ we can define the fractional
powers of the operators ~ - T by using Definition 4.4. Moreover, a discrete analogue
of Theorem 4.5 is true in this context as well as the inequality (4.10).
The proof of the following corollary is similar to the proof of Corollary 4.6 up
to an appropriate change of notation. We leave the details to the reader.
C o r o l l a r y 4 .10 , Let X be a Banach space having Four ier type p E (1, 2]. Let
T E E,(X) be a p o w e r bounde d operator. If, f o r some 7 E (1/p, 1],
(4.14) N Im(~ - T ) "r is dense in X , ~ET
then f o r every x E X
lira I IT~xl l = 0. n--+oo
Remark 4.11. One can also formulate a discrete counterpart of Theorem 4.3
and of Corollary 4.8 if a(T) M qF has Lebesgue measure 0 and is a set with finite
7-entropy. We leave the details to the reader. Note that asymptotic properties of
orbits of such T were studied in [30] and [17].
Acknowledgements. The authors thank A. Aleman, A. Granados and
N. Nikolski for useful discussions, and the referee for a careful reading of the
manuscript and for useful remarks.
REFERENCES
[1] A. Aleman andJ. Cima, An integral operatoron Hp andHardy's inequal#y, J. Analyse Math. 85 (2001), 157-176.
[2] W. Arendt, C. J. K. Batty, M. Hieber and F. Neubrander, Vector-valued Laplace Transforms and Cauchy Problems, Birkh~iuser, Basel, 2001.
[3] S. Axler, P. Bourdon and W. Ramey, Harmonic Function Theory, Springer-Verlag, New York, 2001.
[4] C.J.K. Batty, Z. Brzezniak and D. A. Greenfield, A quantitative asymptotic theorem for contrac- tion semigroups with countable unitary spectrum, Studia Math. 121 (1996), 167-182.
[5] C. J. K. Batty, R. Chill and Y. Tomilov, Strong stability of bounded evolution families and semigroups, J. Funct. Anal. 193 (2002), 116-139.
[6] A. Beurling, Analytic continuation across a linear boundary, Acta Math. 128 (1972), 153-182.
[7] A. Borichev and H. Hedenmalm, Completeness of translates in weighted spaces on the half-line, Acta Math. 174 (1995), 1-84.
356 R. CHILL AND Y. TOMILOV
[8] K. N. Boyadzhiev and N. Levan, Strong stability of Hilbert space contraction semigroups, Stud. Sci. Math. Hung. 30 (1995), 165-182.
[9] H. Bremermann, Distributions, Complex Variables, and Fourier Transforms, Addison-Wesley, Reading, Mass., 1965.
[10] T. Carleman, L'int~grale de Fourier et questions que s'y rattachent, Publications Scientifiques de l'Institut Mittag-Leffler, Vol. 1, Uppsala, 1944.
[ 11] E W. Carroll and D. J. Troy, Distributions and analytic continuation, J. Analyse Math. 24 (1971), 87-100.
[12] R. Chill and Y. Tomilov, Stability ofCo-semigroups and geometry of Banach spaces, Math. Proc. Cambridge Philos. Soc. 135 (2003), 493-511.
[13] B. E. J. Dahlberg, On the radial boundary values of subharmonic functions, Math. Scand. 40 (1977), 301-317.
[14] B.E.J . Dahlberg, On the Poisson integral for Lipschitz and C1-domains, Studia Math. 66 (1979), 13-24.
[15] L. Ehrenpreis, Reflection, removable singularities, and approximation for partial differential equations. I, Ann. of Math. 112 (1980), 1-20.
[16] L. Ehrenpreis, Reflection, removable singularities, and approximation for partial differential equations. H, Trans. Amer. Math. Soc. 302 (1987), 1--45.
[17] J. Esterle, M. Zarrabi and M. Rajoelina, On contractions with spectrum contained in the Cantor set, Math. Proc. Cambridge Philos. Soc. 117 (1995), 339-343.
[18] P. C, Fenton, Line integrals of subharmonic functions, J. Math. Anal. Appl. 168 (1992), 108-110.
[19] R.M. Gabriel, Some inequalities concerning integrals of two-dimensional and three-dimensional subharmonicfunctions, J. London Math. Soc. 24 (1949), 313-316.
[20] J. Garcia-Cuerva, K. S. Kazaryan, V. I. Kolyada and Kh. L. Torrea, The Hausdorff-Young inequality with vector-valued coefficients and applications, Uspekhi Mat. Nauk 53 (1998), 3-84; Engl. transl.: Russian Math. Surveys 53 (1998), 435-513.
[21] E Gehring, Variations on a theorem of Fej~r and Riesz, Ann. Univ. Mariae Curie-Sklodowska Sect, A 53 (1999), 57-66.
[22] A. Granados, On a problem raised by Gabriel and Beurling, Michigan Math. J. 46 (1999), 461-487.
[23] V.P. Gurarii, Harmonic analysis in spaces with a weight, Trudy Moskov. Mat. Obsch. 35 (1976), 21-76; Engl. Transl.: Trans. Moscow Math. Soc. 1 (1979), 21-75.
[24] R. Harvey and J. Polking, Removable singularities of solutions of linear partial differential equations, Acta Math. 125 (1970), 39-56.
[25] V. Havin and B. J6ricke, The Uncertainty Principle in Harmonic Analysis, Springer-Verlag, Berlin, 1994.
[26] L. Htirmander, Lp estimates for (pluri-) subharmonicfunctions, Math. Scand. 20 (1967), 65-78.
[27] S. V. H r u ~ v , Sets of uniqueness for the Gevrey classes, Ark. Mat. 15 (1977), 253-304.
[28] S. V. Hrug~v, The problem of simultaneous approximation and of removal of the singularities of Cauchy type integrals, in Spectral Theory of Functions and Operators, Proc. Steklov Inst. Math. 4 (1979), Amer. Math. Soc. Transl., Providence, R.I., pp. 133-203.
[29] Y. Katznelson, An Introduction to Harmonic Analysis, 2nd corrected edition, Dover Publications, New York, 1976.
[30] K. Kelley, Contractions et hyperdistributions ?t spectre de Carleson, J. London Math. Soc. 58 (1998), 185-196.
[31] C. Kenig, Harmonic Analysis Techniques for Second Order Elliptic Boundary Value Problems, CBMS Regional Conf. Ser. in Math., Vol. 83, Amer. Math. Soc., Providence, R.I., 1994.
ANALYTIC CONTINUATION 357
[32] H. Komatsu, Fractional powers of operators IV: potential operators, J. Math. Soc. Japan 21 (i 969), 221-228.
[33] C. Martinez and M. Sanz, The Theory of Fractional Powers of Operators, North-Holland, Amsterdam, 2001.
[34] J. M. A. M. van Neerven, The Asymptotic Behaviour of Semigroups of Linear Operators, Birkhfiuser Verlag, Basel, 1996.
[35] N. K. Nikolski, Treatise on the Shift Operator Spectral Function Theory, Springer-Verlag, Berlin, 1986.
[36] A.V. Noell and T. Wolff, Peak sets for Lip a classes, J. Funct. Anal. 86 (1989), 136--179.
[37] J. Priiss, Evolutionary Integral Equations and Applications, Birkh~iuser Verlag, Basel, 1993.
[38] W. Rudin, Lectures on the Edge-of-the-Wedge Theorem, CBMS Regional Conf. Ser. in Math., Amer. Math. Soc., Providence, R.I., 1971.
[39] B. Sz.-Nagy and C. Foiaw Harmonic Analysis of Operators on Hilbert Space, North-Holland, Amsterdam, 1970.
[40] Y. Tomilov, A resolvent approach to stability of operator semigroups, J. Operator Theory 46 (2001), 63-98.
[41] Qu6c Phong Vfi, On the spectrum, complete trajectories and asymptotic stability of linear semi- dynamical systems, J. Differential Equations 105 (1993), 30-45.
[42] U. Westphal, A generalized version of the abelian mean ergodic theorem with rates for semigroup operators and fractional powers of infinitesimal generators, Results Math. 34 (1998), 381-394.
[43] E Wolf, The Poisson integral A study in the uniqueness of functions, Acta Math. 74 (1941), 65-100.
[44] E Wolf, Extension of analytic functions, Duke Math. J. 14 (1947), 877-887.
Ralph Chill ABTEILUNG ANGEWANDTE ANALYSIS
UNIVERS1T.~T ULM 89069 ULM, GERMANY
email: chillOmathematik, uni-ulm.de
Yuri Tomilov FACULTY OF MATHEMATICS AND COMPUTER SCIENCE
NICOLAS COPERNICUS UNIVERSITY UL. CHOPINA 12/18
87-100 TORUN, POLAND email: tomilov~mat.unl,torun,pl
(Received March 23, 2003 and in revised form August 13, 2003)