Math. Ann.

(2010) 347:455478
DOI 10.1007/s00208-009-0439-0

Mathematische Annalen

Optimal polynomial decay of functions and operator

semigroups
Alexander Borichev Yuri Tomilov

Received: 4 June 2009 / Revised: 7 September 2009 / Published online: 30 October 2009
Springer-Verlag 2009

Abstract We characterize the polynomial decay of orbits of Hilbert space C0 -semigroups in resolvent terms. We also show that results of the same type for general
Banach space semigroups and functions obtained recently in Batty and Duyckaerts
(J Evol Eq 8:765780, 2008) are sharp. This settles a conjecture posed in Batty and
Duyckaerts (2008).
Mathematics Subject Classification (2000)
46B20

Primary 47D06; Secondary 34D05

1 Introduction
One of the main issues in the theory of partial differential equations is to determine
whether the solutions to these equations approach an equilibrium and if yes then

The authors were partially supported by the Marie Curie Transfer of Knowledge programme, project
TODEQ. The first author was also partially supported by the ANR project DYNOP. The second author
was also partially supported by a MNiSzW grant Nr. N201384834.
A. Borichev
Centre de Mathmatiques et Informatique, Universit dAix-Marseille I,
39 rue Frdric Joliot-Curie, 13453 Marseille, France
e-mail: [email protected]
Y. Tomilov (B)
Faculty of Mathematics and Computer Science, Nicolas Copernicus University,
ul. Chopina 12/18, 87-100 Torun, Poland
e-mail: [email protected]
Y. Tomilov
Institute of Mathematics, Polish Academy of Sciences,

Sniadeckich
str. 8, 00-956 Warsaw, Poland

123

456

A. Borichev Y. Tomilov

how fast do the solutions approach it. Recently, an essential progress was achieved
in treating such asymptotic problems by operator-theoretical methods involving
C0 -semigroups. For different accounts of developments, highlights, and techniques
of asymptotic theory of C0 -semigroups see [2,5,10,26].
In particular, the following result was proved in [4, p. 803], see also [5, p. 4142].
(The result is implicitly contained already in [1].)
Theorem 1.1 Let (T (t))t0 be a bounded C0 -semigroup on a Banach space X with
generator A. Suppose that iR is contained in the resolvent set
(A) of A. Then
T (t)A1  0, t .

(1.1)

In other words, all classical solutions to the abstract Cauchy problem

x(t)
= Ax(t), t 0,
x0 X,
x(0) = x0 ,

(1.2)

given by x(t) = T (t)x0 , t 0, x0 D(A), converge uniformly (on the unit ball of
D(A) with the graph norm) to zero at infinity if A satisfies the conditions of Theorem 1.1.
In general, without any additional assumptions, the decay in (1.1) can be arbitrarily slow. However, in a number of special situations involving concrete PDEs, e.g.
damped wave equations, the rate of decay in (1.1) corresponds to the rate of decay
of the energy of the system described by (T (t))t0 , and it is of interest to determine
whether this rate of decay can be achieved.
By rewriting equations in the abstract form (1.2), the rates of the decay of sufficiently smooth orbits for the corresponding semigroup (T (t))t0 (and equivalently of
solutions to (1.2)) can be associated with the size of the resolvent R(, A) = ( A)1
of A on the imaginary axis. This approach was initiated in [19] and later pursued, in
particular, in [7,8,13,20]. However, with a few exceptions, the issue of optimality (or
non-optimality) of the rates of growth has not been studied so far.
The above applications (mainly in the abstract set-up) motivated a thorough study
of the decay rates on T (t)A1  for bounded C0 -semigroups (T (t))t0 on Banach
spaces in [3,21], and, most recently, in [6]. In the latter paper, the authors developed
a unified and simplified approach for estimating the decay rates for T (t)A1  in
terms of the growth of R(is, A), s R, using the contour integrals technique by
NewmanKorevaar. In particular, the following theorem is proved there.
For A as in Theorem 1.1 we define a continuous non-decreasing function
M() = max R(it, A),
t[,]

0,

(1.3)

and the associated function

Mlog () := M() (log(1 + M()) + log(1 + )) , 0.
1
Let Mlog
be the inverse of Mlog defined on [Mlog (0), +).

123

(1.4)

Optimal polynomial decay

457

Theorem 1.2 (Batty, Duyckaerts) Let (T (t))t0 be a bounded C0 -semigroup on a

Banach space X with generator A, such that iR
(A). Let the functions M and
Mlog be defined by (1.3) and (1.4). Then there exist C, B > 0 such that
T (t)A1 

C
1
Mlog
(t/C)

, t B.

(1.5)

Note that in the case > 0, M() C(1 + ), 0, the BattyDuyckaerts

result gives
T (t)A


C

log t
t

1

, t B.

(1.6)

It was conjectured in [6] that Theorem 1.2 can be improved by removing the logarithmic factor in (1.6) in the case when X is a Hilbert space, and that this factor is
necessary if X is merely a Banach space, see [6, Remark 9]. (See also the introduction
in [3] and the comments after Theorem 3.5 therein.)
In our paper, we address the problem of optimality of the rate of decay in (1.6) and
confirm the conjecture from [6] for the case of polynomially growing M. We show that
the logarithmic factor can be dropped in (1.6) if X is a Hilbert space. Thus, various
results on polynomial decay of solutions of PDEs, e.g. in [3,8,21,22] can be improved
to sharp formulations or shown to be sharp. See also [6] and references therein.
On the other hand, we show that Theorem 1.2 is essentially sharp (see Theorems 3.1
and 4.1 below). This is done by a function-theoretical construction which may be interesting for its own sake and may be useful in other instances related to C0 -semigroups
as well. We prove, in particular, that given > 0 there exists a Banach space X and
a bounded C0 -semigroup (T (t))t0 in X with generator A such that
R(is, A) = O(|s| ),

|s| ,

and

lim sup
t

t
log t

1/

T (t)A1  > 0.

The classical problem of estimating the local energy for solutions to wave equations
leads to the study of decay rates for functions of the form T1 T (t)T2 , where T1 , T2
are bounded operators on X, so that the assumptions are imposed on the cut-off resolvent F() = T1 R(, A)T2 rather than on the resolvent itself, see e.g [6,7,13,27]. Due
to the lack of the Neumann series expansions for F we have to assume that F extends
analytically to the region  of known shape and satisfies certain growth restrictions
there. The domain  and the growth of F in  are not in general related to each other.
Moreover, the operator-theoretical approach can hardly be used to deal with F since
the resolvent identity is not available as well. Thus, it is natural to put the problem into
the framework of decay estimates for L (R+ , X ) functions given that their Laplace

123

458

A. Borichev Y. Tomilov

transforms extend to the specific  with, in our case, polynomial estimates. In this
direction, we obtain a result on the polynomial rates of decay for bounded functions
which partially generalizes [6, Theorem 10]. The result has its version for the rates of
decay of T1 T (t)T2  thus improving [6, Corollary 11].
We use standard notations. Given a closed linear operator A we denote by (A),
(A), D(A), and Im (A) the spectrum of A, the resolvent set of A, the domain of A
and the image of A respectively. By C, C1 etc. we denote generic constants which
may change from line to line.
The plan of the paper is as follows. In the Sect. 2 we characterize the rate of polynomial decay of the semigroup orbits in the Hilbert space via the growth rate of the
resolvent on the imaginary axis. Examples of functions constructed in the Sect. 3 show
that the version of Theorem 1.2 for functions given in [6] is sharp. These examples
are used in the Sect. 4 to show that Theorem 1.2 is itself sharp.
2 Decay of Hilbert space semigroups
We start with recalling two simple and essentially known statements on C0 -semigroups. The first one is a version of the well-known criterion on the generators of
bounded Hilbert space C0 -semigroups, [14,24]. In our case, the proof is particularly
easy and, to make our presentation self-contained, we give a short argument. Let
C+ := {z C : Re z > 0}.
Lemma 2.1 Let (T (t))t0 be a C0 -semigroup on a Hilbert space H with generator
A. Then (T (t))t0 is bounded if and only if
C+
(A), and
 

R( + i, A)x2 + R( + i, A )x2 d <
sup
>0

for every x H.
Remark 2.2 By the closed graph theorem, in the conditions of Lemma 2.1, for x H
we have
 

sup
R( + i, A)x2 + R( + i, A )x2 d Cx2 .
>0

Proof of Lemma 2.1 The necessity is a direct consequence of the Plancherel theorem
applied to the families {e t T (t)x : x H, > 0}, {e t T (t)x : x H, > 0}.
The sufficiency follows from the representation

T (t)x, x
=

1
2i t

1
t +i

1
t i

123

et R 2 (, A)x, x
d,

t > 0,

Optimal polynomial decay

459

where the integral converges absolutely by the Hlder inequality and our assumptions,
see e.g. [14,24,9].


The second statement allows us to cancel the growth of the resolvent by restricting
it to sufficiently smooth elements of H .
Lemma 2.3 Let (T (t))t0 be a bounded C0 -semigroup on a Hilbert space H with
generator A, such that iR
(A). Then for a fixed > 0
R(, A)(A)  C,

Re > 0,

if and only if
R(is, A) = O(|s| ),

s .

(2.1)

Proof The lemma is proved in [18, Lemma 3.2], [15, Lemma 1.1] in a version saying,
in particular, that the condition


R(, A) C 1 + || ,

0 < Re < 1,

is equivalent to
R(, A)(A)  C1 ,

0 < Re < 1.

To get our version of the assertion

to apply the maximum principle to the
 it suffices

2

1 B 2 on the domain { C : Re 0, 1
function F() = R(, A)
|| B} for large B, and to use the estimate
R(, A)

C
, Re > 0,
Re

for with || = B. Then (2.1) implies

R(, A) C(1 + || ), Re > 0.
Since R(, A) is bounded in any halfplane strictly included in C+ , Lemma 2.3 is
reduced to its strip version mentioned above.


The next theorem is one of the main results of the paper. Its proof is based on a
trick: we pass to a matrix semigroup whose boundedness gives the required rate of
decay of T (t)(A)  (and of T (t)A1 ) for free.
Theorem 2.4 Let (T (t))t0 be a bounded C0 -semigroup on a Hilbert space H with
generator A such that iR
(A). Then for a fixed > 0 the following conditions
are equivalent:

123

460

A. Borichev Y. Tomilov

(i)
R(is, A) = O(|s| ),

s .

(2.2)

(ii)
T (t)(A)  = O(t 1 ),

t .

(2.3)

(iii)
T (t)(A) x = o(t 1 ),

t , x H.

(iv)
T (t)A1  = O(t 1/ ),

t .

(2.4)

t , x H.

(2.5)

(v)
T (t)A1 x = o(t 1/ ),

Proof The implication (ii) (i) was proved in [6]; the implication (iii) (ii)
is a consequence of the uniform boundedness principle. Moreover, the equivalence
(iv) (ii) was obtained in [3, Proposition 3.1] as a consequence of the moment
inequalities for A. Its o-counterpart (iii) (v) can be obtained by the same
argument. Thus, it remains to prove that (i) (iii).
(i) (iii): Let H = H H be the direct sum of two copies of H . Consider the
operator A on H given by the operator matrix

A=

A (A)
O
A

with the diagonal domain D(A) = D(A) D(A). Then (A) = (A), and the
resolvent R(, A) of A is of the form

R(, A) =

R(, A) R 2 (, A)(A)
O
R(, A)


,

(A).

The operator A is the generator of the C0 -semigroup (T (t))t0 on H defined by


T (t) =

T (t) t T (t)(A)
O
T (t)


,

(2.6)

because the resolvents of A and of the generator of (T (t))t0 coincide. By (2.2) and
Lemma 2.3,
R(, A)(A)  C, Re > 0.

123

Optimal polynomial decay

461

Hence, for every x = (x1 , x2 ) H and C+ ,



R(, A)x2 C R(, A)x1 2 + R(, A)x2 2 ,

(2.7)



R(, A )x2 C R(, A )x1 2 + R(, A )x2 2 .

(2.8)

and similarly

By Lemma 2.1,
sup
>0

 

R( + i, A)x2 + R( + i, A )x2

d <

for every x H , so that

sup
>0

 

R( + i, A)x2 + R( + i, A )x2

d <

for every x H. Then again by Lemma 2.1, (T (t))t0 is bounded on H. Since

(T (t))t0 is bounded, the definition of (T (t))t0 implies that
sup t T (t)(A)  < .
t0

Furthermore, iR (A). Then by Theorem 1.1,

T (t)x 0,

t , for every x H,

(2.9)

since D(A) = Im (A1 ) is dense in H. Again by Theorem 1.1, (2.9) implies that
t T (t)(A) x = o(1),

t , x H.



Remark 2.5 If one is merely interested in the proof of the implication (i) (ii), then
there is no need to invoke Theorem 1.1 as the argument above shows.
There is another argument for (i) (iii), which does not use Theorem 1.1 too.
Let n 1 + be an integer. By the resolvent identity, R(, A)x L 2 (iR, H ), and,
moreover, R(, A)x H 2 (C+ , H ) (the Hardy class in the right half-plane) for every
x from the dense set Im (A)n . Then,

lim

0+

R( + i, A)x2 d = 0,

x Im (A)n ,

(2.10)

123

462

A. Borichev Y. Tomilov

and since (T (t))t0 is bounded, the last relation holds for all x H . By the simple
integral resolvent stability criterion from [25, Theorem 3.1], this means that (T (t))t0
satisfies (2.9). Using the estimates (2.7), (2.8), we conclude that R( +i, A)x, > 0,
satisfies an analog of (2.10) for every x H. Since the semigroup (T (t))t0 is
bounded, by the same stability criterion, (T (t))t0 is stable. Therefore, in particular,
1
T (t)(A) x = o( ),
t

t , x H.

Remark 2.6 Consider the elementary example of a multiplication C0 -semigroup

(T (t))t0 ,
(T (t) f )(z) = et z f (z),

t 0,

on L 2 (S, ), where S := {z C : Re z < 1/(1 + |Im z|) } and is Lebesgue

measure on S.
The operator A f (z) = z f (z) with maximal domain is the generator of (T (t))t0 ,
and for f, g L 2 (S, ) we have

R(, A) f, g
=
S

T (t)(A) f, g
=


S

f ( )g( ) d( )
,

et f ( )g( ) d( )
,
( )

C\S,
t 0.

Straightforward estimates give that

R(is, A) = O(|s| ), |s| , and

lim tT (t)(A)  = 1,

t+

which demonstrates the optimality of the rates of decay assuming (i) in Theorem 2.4.
A similar fact is true for the decay rate of T (t)A1 x, x H , see, for example,
[3, Proposition 4.1].
Remark 2.7 An advantage of the construction in the proof of Theorem 2.4 is that
it reduces the problem of finding optimal polynomial rates in (2.4) under resolvent
growth conditions to proving the boundedness of (T (t))t0 under the same kind of
conditions. While there is a criterion for boundedness of Hilbert space semigroups
in terms of boundary behavior of certain resolvent means, see e.g. [14,24], a similar
criterion for semigroups on Banach spaces is yet to be found. Note that the Banach
space semigroups (T (t))t0 satisfying T (t) w(t), where w : R+ R+ is submultiplicative, can be characterized by Hille-Yosida type conditions, see [12, Theorem
5.1] and the comments following it. However, we do not yet know how to verify such
conditions for (T (t))t0 and for the weight w(t) = C ln(e + t) to conclude that
T (t) C ln(e + t), t 0, recovering thus Theorem 1.2 for polynomially growing
functions M in the Banach space setting.

123

Optimal polynomial decay

463

3 Decay of functions
Let M be a continuous non-decreasing function, and let Mlog be defined by (1.4).
The following result is an analog of Theorem 1.2 for the decay of functions f
L (R+ , X ) (see also [6, Theorem 10]). Given f L (R+ , X ), its Laplace transform
is defined by

f (z) =

ezt f (t) dt.

Theorem 3.1 (Batty, Duyckaerts) Let f L (R+ , X ) be such that

a)
f extends analytically to the domain
 := {z C : Re z > 1/M(|Im z|)}, and
b)

f (z) M(|Im z|),

z .

Then there exist C, C1 > 0, t0 (depending on  f  and M) such that

t

C1

f (s) ds

M 1 (t/C) ,

f (0)

log

t t0 .

If we are interested only in polynomial rate of growth of

f and in (possibly different) polynomial rate of narrowing of \C+ , we can formulate the following variant
of [6, Theorem 10].
Proposition 3.2 Let , , c1 , c2 > 0,  := {z C : Re z > c1 (1 + |Im z|) }.
f admits an analytic extension to  and
Suppose that f L (R+ , X ) is such that

f (z) c2 (1 + |Im z|) ,

z .

Then



t

log t 1/

f (s) ds
C
,

f (0)
t

t 2,

with C depending only on c1 , c2 ,  f  , , .

Remark 3.3 Thus, given polynomial growth of
f in , the rate of polynomial decay
t
of (
f (0) 0 f (s) ds) is determined only by the shape of . This is in sharp contrast
to the semigroup case where the growth of R(, A) in  and the size of  are related
to each other due to the Neumann series expansions of the resolvent.

123

464

A. Borichev Y. Tomilov

Without loss of generality we assume that

f is continuous up to .
Lemma 3.4 Under the conditions of Proposition 3.2, for any > 0 and for some
C1 , C2 depending only on c1 , c2 ,  f  , , , we have

f (z) C2 (1 + |Im z|)+ ,

z  ,

where  := {z C : Re z > C1 (1 + |Im z|) }.

Proof Suppose that > + (otherwise, there is nothing to prove). We use the fact
that the function log 
f  is subharmonic in . Fix A > 2/ and y > 1 (we deal
with the case y < 1 by symmetry), and set
c1
Q y = {z : c1 (y + 1) < Re z < (y + 1) , y 1 < Im z < y + 1},
c
 A
1

E1 = Q y
(y + 1) + iR ,
A
E 2 = Q y \E 1 .
Next we use that
z E2 ,
log 
f (z) log c2 + log(y + 1),
 f  A
log 
f (z) log
+ log(y + 1), z E 1 ,
c1
and an estimate of harmonic measure in the thin rectangle Q y ,

 c
2
1
(y + 1) + i y, E 2 , Q y < ,
A
A

(3.1)

for large y; here (z, E, U ) is harmonic measure of E U with respect to z U .

By the theorem on two constants (see, for example, [16, VII B1]) we obtain

 c
 c



1
1

f (y + 1) + i y
(y + 1) + i y, E 2 , Q y sup log 
log

f
A
A
E2
 c

1
+ (y + 1) + i y, E 1 , Q y sup log 
f .
A
E1
Hence there exists y0 (depending on c1 , c2 , , , A, and on  f  ) such that

 c


f (y + 1) + i y
( + ) log(y + 1),
log

A
To get (3.1) we could just observe that on Q y we have
(, E 1 , Q y ) (, E 1 , Q) (, E  , Q),

123

y > y0 .

Optimal polynomial decay

465

where Q = {z : c1 (y + 1) < Re z < cA1 (y + 1) } and E  = Q\ Q y , and use

elementary estimates for (, E 1 , Q) and (, E  , Q) obtained via conformal mapping
of the strip Q onto the half-plane.


Proof of Proposition 3.2 By Lemma 3.4, we can assume that 2.
Next, we follow the argument from [6] (which goes back to [17,23]). By the Cauchy
integral formula, for any sufficiently small contour around 0 and for any R > 1 we
have


t
t
 
2
z
dz
1

. (3.2)
1+ 2
f (s) ds =
f (z) ezs f (s) ds e zt
f (0)
2i
R
z

Let 1 = RT C+ , 2 = RT (\C+ ), 3 = RT\C+ , 4 =  RD, where D is

the unit disc of C, and T = D. Then

 


t
t
2

z
zs

e zt dz

f
(0)

1
+
f
(s)
ds
e
f
(s)
ds
2

f
(z)

2
R
z

1
0
0

 

z 2 zt dz

1 + 2 f (z)e
R
z

 

 
2

z
ezs f (s) ds e zt dz

1
+

2
R
z

3
0

 

z 2 zt dz

= I1 + I2 + I3 + I4 .
+

1 + 2 f (z)e
R
z

Now
/2
I1 c
/2


c  f  L (R+ ,X )
es R cos  f (s + t) ds cos d
,
R
0

c1 (R+1)


I2 c
0

/2
I3 c
/2

R
I4 c
0

c2 (R + 1) y dy
cc12 c2 (R + 1)22 ,
R2


c  f  L (R+ ,X )
es R cos  f (t s) ds cos d
,
R
0

c2 (s + 1) ec1 t (s+1)

ds
.
s+1

123

466

A. Borichev Y. Tomilov

Setting R = c(c1 , , )(t/ log t)1/ , we obtain


I1 + I2 + I3 + I4 C(c1 , c2 ,  f  , , )

log t
t

1/
,

t 2.



Remark 3.5 We can avoid using Lemma 3.4 by replacing the term (1 +
z2 N
)
R2

z2
)
R2

in the

right hand side of (3.2) by (1 +

for a suitable N . However, Lemma 3.4 provides
some additional information on the growth of
f close to the imaginary axis.
The statement of Proposition 3.2 can be sharpened if f belongs to the space of
bounded uniformly continuous X -valued functions BUC(R+ , X ).
Proposition 3.6 Suppose that f BUC(R+ , X ) satisfies the assumptions of Proposition 3.2. Then



t

log t

, t .
f (s) ds
= o

f (0)
t

(3.3)

Proof By Arendt et al. [2, Corollary 4.4.6], our hypothesis on f imply that
 f (t) = o(1), t .

(3.4)

Arguing as in the proof of Proposition 3.2, and using (3.4) we obtain

I1

c
sup  f (s + t) = o
R s0
t

I3 c

sup

/2<</2

cos

 
1
,
R

t ,

es R cos  f (t s) ds = o

 
1
,
R

t ,

and (3.3) follows.

The following result is a version of Proposition 3.2 written in the semigroup language. Let , > 0 and the domain  be as above.
Corollary 3.7 Let (T (t))t0 be a C0 -semigroup on a Banach space X with generator
A, such that supt0 T (t) := M < , and let T1 and T2 be bounded operators on
X . Suppose that F(z) = T1 R(z, A)T2 admits a holomorphic extension to the domain
, and suppose that
F(z) M1 (1 + |Im z|) ,

123

z .

Optimal polynomial decay

467

Then there exist c = c(M, M1 , , ) > 0 such that

T1 T (t)(I A)1 T2  c

log t
t

1/
,

t 2.

(3.5)

Proof As in the proof of [6, Corollary 11] we consider the function


d 
T1 T (t)(I A)1 T2 = T1 T (t)A(I A)1 T2
dt
= T1 T (t)T2 + T1 T (t)(I A)1 T2 .

f (t) =

Its Laplace transform

f extends analytically to  with the same estimates as F (up to
a constant multiple). Since

f (0)

t

f (s) ds = T1 T (t)(I A)1 T2 ,

Proposition 3.2 yields the claim.

Next result shows that Proposition 3.6 is sharp, at least for > /2. It suffices to
consider just scalar-valued functions.
Theorem 3.8 Given > 0, > /2, and a positive function C0 (R+ ), there
exists a function f C0 (R+ ) such that
a)
f admits an analytic extension to the region  := {z C : Re z > 1/(1 +
|Im z|) } and

f (z)(1 + |Im z|) 0, |z| , z ,
and
b)
t
lim sup
t (t) log t





t


 > 0.

f
(0)

f
(s)
ds




0

To prove Theorem 3.8 we imitate the multiplication semigroup example described

in Remark 2.6. However, we choose an infinite charge so that while formal integral expressions for the resolvent, the semigroup and its orbits remain the same as in
Remark 2.6, the corresponding size estimates behave in a different way due to the lack
of absolute convergence of the integrals.
The proof of Theorem 3.8 is based on the following lemma. Denote by A the
characteristic function of a set A.

123

468

A. Borichev Y. Tomilov

Lemma 3.9 Given Q > 0 and > 0, there exist an integer k > Q and a complex
measure with compact support in C\ such that for some B = B(, ), we have


 


d( ) 

z ,

 B(1 + |Im z|) { :| |>Q} (z) + ,

z 
 C\



 





t
e d( ) B{r :|r |>Q} (t) + ,
t 0,



 C\



 





k
1/B 
e d( ) B,


 C\



 



log t


t d( ) 
(Q,2k) (t) +
,
t 0,
e

 B


t
t +1
 C\


(3.6)

(3.7)

(3.8)

(3.9)

and


 



d(
)
log k


.
ek




Bk
 C\

Proof Choose H > Q large enough, and for an integer k, k 2, such that
H k H 3/2 ,
define
A := 2k log k,

:= Ak1 / k,
q := e2i/k ,
w := i H 1.
We define also a finite measure
:=


1sk

where x is the unit mass at x.

123

q s (1 + q s /(Aw))w+q s /A ,

(3.10)

Optimal polynomial decay

469

Observe that

1sk

qs
k
= k
,
s
x q
x 1

1sk

q 2s
kx
= k
.
s
x q
x 1

(3.11)

Indeed, to prove the first identity we use that



qs
P(x)
,
= k
x qs
x 1

1sk

for some polynomial P with deg P < k such that P(x) = P(q x) so that P(x) = const,
and then P(x) = P(0) = k. The second equality can be proved in a similar way by
using that if


q 2s
P(x)
,
= k
x qs
x 1

1sk

then P(q x) = q P(x) which yields P(x) = kx.

Now using (3.11) we get
def

C(z) =

C\

 q s (1 + q s /(Aw))
d( )
=
z
(z w) q s /A
1sk

 
1sk

Aq s
q 2s /w
+
A(z w) q s
A(z w) q s

k A(z w)
+
Ak (z w)k 1 w Ak (z w)k 1
k A
z
.
=
w Ak (z w)k 1

= A

If z , H > 1, then
|z w| 1

1
.
H

Indeed, if Im z < H 1, then |z w| > 1, and otherwise, Re z H .

Now if H2 Im z |z| 2H , z , then we use that A|z w| > 2,

k
|A (z w)k 1| > Ak ek/H /2, to obtain that



z

k A

1k k/H


= c kek/H .
 w Ak (z w)k 1  c k A e
From now on we assume that k satisfies the condition
k/H
ke
H.

(3.12)

123

470

A. Borichev Y. Tomilov

Then
|C(z)| cH ,

H
Im z |z| 2H, z .
2

(3.13)

If z  and |Im z H | + ||z| H | > H/2, then |z w| > c max(|z|, H ), and

under condition (3.12) we have for large H :

c1 |z| k
c1 |z| k
|C(z)|

.
H (c max(|z|, H ))k
H (c max(|z|, H ))+1
This, together with (3.13), proves (3.6).
Next,
def

L(t) =

et d( ) =

q s (1 + q s /(Aw))et (w+q

s /A)

1sk

C\

and







s
t 
s
s
q t/A 
|L(t)| = e 
q (1 + q /(Aw))e
.
1sk

Furthermore, we have


q s (1 + q s /(Aw))eq

s t/A

1sk

 
1
(q s + q 2s /(Aw))(q s t/A)n
n!
1sk n0

  t km1
t km2
1
1
1
=k
+ km2

Akm1 ( km 1)!
A
( km 2)! Aw

m1


  t k m1 (k 1)! 
km 1
kt k1
1+
,
= k1
A (k 1)!
Ak
( km 1)!
tw
m1

and

|L(t)| =

 m1


km 1
k 3/2 t k1 et  t k
(k 1)!
1
+
.
k!
Ak
( km 1)!
tw
m1

123

(3.14)

Optimal polynomial decay

471

Thus, for some constants c, c1 , c2 , c3 we have



k k(m1) (k 1)! km
k 3/2 t k2 et 
|L(t)| c
k!
HA
( km 1)! H
m1

t 5/2 k2

c1 e k

/(k!H ),

0 t k/H,

and
c2 et k 3/2 t k1 /k! |L(t)| c3 et k 3/2 t k1 /k!,

k/H t A.

If t A, then by (3.14),

|L(t)| 2 et ket/A = 2 k Ak1 et (11/A) .

The function t  et t k2 attains its maximum on [0, k/H ] at t = k/H ; the
function t  et t k1 attains its maximum on [k/H, A] at t = k 1; the function
t  et (11/A) attains its maximum on [A, +) at t = A. Using that A = 2k log k,
by Stirlings formula, we obtain that
0 < c1 |L(k)| c2 |L(k 1)| c3 ,
max |L| = o(1),
H ,
[k/H,k/2]

ek/H k 5/2 (k/H )k2 /(k!H ) 0,

k1 A
k A e 0,
H .

H ,

Hence,
0 < c1 |L(k)| c2 max |L| c3 ,
R+

max

[0,k/2][A,+)

|L| = o(1),

H ,

and (3.7) and (3.8) follow for large H .

Finally,

def

N (t) =

et
C\

et+i H t  s q s t/A
d( )
=
q e
.

w
1sk

123

472

A. Borichev Y. Tomilov

Here we use the formula



q s eq

1sk

s t/A

 

q s (q s t/A)n

1sk n0
 t km1

=k

m1

Akm1

1
n!

1
( km 1)!

  t k m1 (k 1)!
kt k1
= k1
A (k 1)!
Ak
( km 1)!
= (1 + o(1))

m1
kt k1

Ak1 (k 1)!

0 t A, H .

Therefore, by Stirlings formula, we have for large H :

|N (k)| c1

kk
c2
ek
k1
.
H
A (k 1)!
H

(3.15)

Moreover,
t

1/

k
|N (t)| c

tH

 k
t
ekt t 1/
k

+ c1 (k 1/ /H ) (k/2,2k) (t),

0 t A,

(3.16)

and
t

1/

Ak1+1/ keA
c et
1/ t/A
kt e
, t A.
|N (t)|
c1
H
H

(3.17)

Now we fix 0 < < 2 and k = H log H in such a way that k N. Then
(3.12) is satisfied for large H , (3.15) implies (3.10), and (3.16), (3.17) imply (3.9).


Proof of Theorem 3.8 Without loss of generality, we can assume that is non-increasing.
Our function f will be defined by an inductive construction. Set Q 1 = 1. On step
n 1 we use Lemma 3.9 with Q = Q n , = 2n to find n , kn satisfying (3.6)(3.10).
Since
lim

|z|, z

Cn (z) = 0,

lim Ln (t) = 0,

we can find Q n+1 > kn such that

123

Optimal polynomial decay

473

|Cn (z)| ,

|z| > Q n+1 , z ,

(3.18)

|Ln (t)| ,

t > Q n+1 ,

(3.19)

which completes the induction step.

Finally, for some numbers n C, |n | = 1, to be chosen iteratively later on, we
set
f =

n (kn )1/ Ln .

n1

The series above converges absolutely by the choice of n , and therefore f C0 (R+ ).
Similarly, the function

f =

n (kn )1/ Cn

n1

extends analytically to , and

f (z)(1 + |Im z|) 0,

|z| , z .

Finally, for m 1 we have





km



f (s) ds 
 f (0)


0







km
1/


=

(k
)
N

(k
)
n
n
n m 

(km ) log km 

n1



 


ckm
1/



(k
)
N

(k
)
n
n
n m 

(km ) log km 

1nm







n (kn )1/ N n (km ) .
n>m


km
(km ) log km

On step m 1 we choose m in such a way that




 


1/

n (kn ) N n (km )


1nm




  


 
1/
1/
=  (km ) N m (km ) + 
n (kn ) N n (km ) .
1n<m


123

474

A. Borichev Y. Tomilov

Then
 



 




km

1/
1/
 


(k
)
N

(k
)

(k
)
N

(k
)

n
n
n
m
n
n
n
m

n>m

(km ) log km 

1nm

 



km



n (kn )1/ N n (km )

(km )|N m (km )| 
n>m

(km ) log km



c2
c1
2n/
c3 > 0.
log km n>m


Remark 3.10 A variant of our construction works with = 1 in Theorem 3.8, if one
looks just for f L (R+ ).
4 Decay of Banach space semigroups
Using the construction of Theorem 3.8, we show next that the analogue of Theorem 1.2
for C0 -semigroups on Banach spaces, Theorem 3.1, is also sharp. Estimates for the
local resolvents of group generators similar to the ones used below has been also
employed, in particular, in [2] and [11, Sect. II.4.6].
Theorem 4.1 Given > 0, there exist a Banach space X and a bounded C0 -semigroup (T (t))t0 on X with generator A such that
(a)

R(is, A) = O(|s| ),

|s| ,

and


(b)

t
lim sup
log t
t

1

T (t)A1  > 0.

Proof Let (S(t))t0 be the left shift semigroup on BUC(R+ ), let  := { C :

Re > 1/(1 + |Im |) }, and let 0 :=  { C : |Re | < 1}. Furthermore,
f
let X be the space of functions f BUC(R+ ) such that the Laplace transform
extends to an analytic function in 0 , and
|
f ()|(1 + |Im |) 0,

, 0 .

Then X equipped with the norm

 f  X =:  f  +  f  :=  f  + sup |
f ()|(1 + |Im |) ,
0

123

Optimal polynomial decay

475

is a Banach space. Moreover, S(t)X X , t 0, and the restriction (T (t))t0 of

(S(t))t0 to X is also a C0 -semigroup. To prove this assertion it suffices to observe
that

f () T

(t) f () =


e


f (s) ds

es f (t + s) ds

= (1 e )
f () + e
t

t

es f (s) ds,

Re > 0,

and the same equality holds on 0 . By the definition of X ,

T (t) f f  0,

t 0+,

T (t) f f  X 0,

t 0+.

and then

Let A stand for the generator of (T (t))t0 .

Next we prove that for every f X the local resolvent R(, A) f satisfies the
estimate
R(, A) f  X C (1 + |Im |)  f  X ,

0 < Re < 1.

(4.1)

We will estimate the quantities R(, A) f  and R(, A) f  separately.

Observe first that for every t R+ and every C+ one has

(R(, A) f ) (t) =

f (t + s) ds = e
f ()
t

t

e(ts) f (s) ds.

It follows that for every fixed t R+ the function  (R(, A) f ) (t) extends to an
analytic function on 0 , and moreover
  f 
|(R(, A) f )(t)|

|Re |
 f 
|Re |

if Re > 0,
+|
f ()| if Re < 0.

Applying Levinsons log-log theorem (see, for example, [16, VII D7]) or, rather,
its polynomial growth version [2, Lemma 4.6.6] to (R(, A) f )(t) in the squares
{ : |Re | < (s + 2) , |s Im | < (s + 2) }, we conclude that
R(, A) f  C(1 + |Im |) ( f  +  f  ) ,

0 < Re < 1.

(4.2)

123

476

A. Borichev Y. Tomilov

Fix with Re (0, 1). To estimate R(, A) f  note that

(R(,
A) f )() =


e

es f (t + s) ds dt


=

()t

es f (s) ds dt

1
=


e

1
f (s) ds +

f ()
f ()
=
,

et f (t) dt

Re > 1.

Therefore, R(,
A) f extends analytically to 0 , and


(R(,
A) f )() =

f ()
f ()

,

= , 0 ,



f (),

= .

Now, if | | 1, 0 , then


|(R(,
A) f )()| |
f ()| +
f ()| c(1 + |Im |) (1 + |Im |)  f  .
Furthermore, if
1 > | |

1
, 0 ,
2(1 + |Im |)

then we have




|(R(,
A) f )()| 2(1 + |Im |) |
f ()| + |
f ()|
c(1 + |Im |) (1 + |Im |)  f  .
Finally, if
| |

1
,
2(1 + |Im |)

then, applying Cauchys formula on the circle

C := {z C : |z | =

123

2
(1 + |Im |) },
3

Optimal polynomial decay

477

we obtain that


|(R(,
A) f )()| c(1 + |Im |) (1 + |Im |)  f  .
Thus,
R(, A) f  C(1 + ||)  f  ,

0 < Re < 1.

(4.3)

The estimates (4.2) and (4.3) together give us (4.1). Since

R(, A) X

1
,
dist (, (A))

the estimate (4.1) implies that iR C\ (A), and that

R(, A) X C (1 + |Im |) ,

iR.

(4.4)

Since (A) iR = , by Theorem 1.1 we obtain

f = lim


A

f T (t)A


f


=

T (t) f dt,
0

for every f X .
By Lemma 3.9 there exist f n X , f n = Ln , and kn as n , such that


log kn
|N n (kn )| C()
kn

 f n  X 1,

1/
, n 1.

Therefore,
T (kn )A1  X

T (kn )A f n  X
f n ( + kn + r ) dr


 

 
 
k
n



 



f n (0)

f n (r ) dr  
f n (s) ds  = |N n (kn )|

 



kn

C()

log kn
kn

1/

Acknowledgments The authors are grateful to the referees for useful comments and suggestions which
led to an improvement of the paper. We would like to thank C. J. K. Batty for sending us a preliminary
version of [6] prior to its publication and A. M. Gomilko for helpful discussions and remarks.

123

478

A. Borichev Y. Tomilov

References
1. Arendt, W., Batty, C.J.K.: Tauberian theorems and stability of one-parameter semigroups. Trans. Amer.
Math. Soc. 306, 837852 (1988)
2. Arendt, W., Batty, C.J.K., Hieber, M., Neubrander, F.: Vector-valued Laplace Transforms and Cauchy
Problems. Monographs in Mathematics, vol. 96. Birkhuser, Basel (2001)
3. Btkai, A., Engel, K.J., Prss, J., Schnaubelt, R.: Polynomial stability of operator semigroups. Math.
Nachr. 279, 14251440 (2006)
4. Batty, C.J.K.: Tauberian theorems for the Laplace-Stieltjes transform. Trans. Amer. Math. Soc. 322,
783804 (1990)
5. Batty, C.J.K.: Asymptotic behaviour of semigroups of operators, In: Functional Analysis and Operator
Theory (Warsaw, 1992), vol. 30. Banach Center Publ. Polish Acad. Sci.,Warsaw, pp. 3552 (1994)
6. Batty, C.J.K., Duyckaerts, T.: Non-uniform stability for bounded semi-groups on Banach spaces.
J. Evol. Eq. 8, 765780 (2008)
7. Burq, N.: Dcroissance de lnergie locale de lquation des ondes pour le problme extrieur et
absence de rsonance au voisinage du rel. Acta Math. 180, 129 (1998)
8. Burq, N., Hitrik, M.: Energy decay for damped wave equations on partially rectangular domains. Math.
Res. Lett. 14, 3547 (2007)
9. Chill, R., Tomilov, Yu.: Stability of C0 -semigroups and geometry of Banach spaces. Math. Proc. Camb.
Philos. Soc. 135, 493511 (2003)
10. Chill, R., Tomilov, Yu.: Stability of operator semigroups: ideas and results. In: Perspectives in operator
theory, vol. 75, pp.71109. Banach Center Publ. Polish Acad. Sci., Warsaw (2007)
11. Chill, R., Tomilov, Yu.: Operators L 1 (R+ ) X and the norm continuity problem for semigroups.
J. Funct. Anal. 256, 352384 (2009)
12. Chojnacki, W.: A generalization of the WidderArendt theorem. Proc. Edinb. Math. Soc. 45, 161
179 (2002)
13. Christianson, H.: Applications of the cutoff resolvent estimates to the wave equation. Math. Res.
Lett. 16, 577590 (2009)
14. Gomilko, A.M.: On conditions for the generating operator of a uniformly bounded C0 -semigroup of
operators. Funktsional. Anal. i Prilozhen. 33, 6669 (1999), transl. in Funct. Anal. Appl. 33, 294296
(1999)
15. Huang, S.-Z., van Neerven, J.M.A.M.: B-convexity, the analytic Radon-Nikodym property and individual stability of C0 -semigroups. J. Math. Anal. Appl. 231, 120 (1999)
16. Koosis, P.: The Logarithmic Integral, vol. I. Cambridge University Press, Cambridge (1988)
17. Korevaar, J.: On Newmans quick way to the prime number theorem. Math. Intell. 4, 108115 (1982)
18. Latushkin, Yu., Shvydkoy, R.: Hyperbolicity of semigroups and Fourier multipliers, In: Systems,
approximation, singular integral operators, and related topics (Bordeaux, 2000), Oper. Theory Adv.
Appl., vol. 129, pp. 341363 Birkhuser, Basel (2001)
19. Lebeau, G.: quation des ondes amorties, In: Algebraic and Geometric Methods in Mathematical
Physics (Kaciveli, 1993), vol. 19, pp. 73109. Math. Phys. Stud, Kluwer, Dordrecht (1996)
20. Lebeau, G., Robbiano, L.: Stabilisation de lquation des ondes par le bord. Duke Math. J. 86, 465
491 (1997)
21. Liu, Z., Rao, B.: Characterization of polynomial decay rate for the solution of linear evolution equation. Z. Angew. Math. Phys. 56, 630644 (2005)
22. Liu, Z., Rao, B.: Frequency domain approach for the polynomial stability of a system of partially
damped wave equations. J. Math. Anal. Appl. 335, 860881 (2007)
23. Newman, D.J.: Simple analytic proof of the prime number theorem. Amer. Math. Monthly 87, 693
696 (1980)
24. Shi, D.-H., Feng, D.-X.: Characteristic conditions of the generation of C0 -semigroups in a Hilbert
space. J. Math. Anal. Appl. 247, 356376 (2000)
25. Tomilov, Y.: Resolvent approach to stability of operator semigroups. J. Oper. Theory 46, 6398 (2001)
26. van Neerven, J.M.A.M.: The asymptotic behaviour of semigroups of linear operators. Operator Theory:
Advances and Applications, vol. 88. Birkhuser, Basel (1996)
27. Vodev, G.: Local energy decay of solutions to the wave equation for nontrapping metrics. Ark. Mat.
42, 379397 (2004)

123